2 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
3 // vim: ts=8 sw=2 smarttab
5 #include "include/compat.h"
9 #include <boost/algorithm/string.hpp>
11 #include <boost/format.hpp>
12 #include <boost/optional.hpp>
13 #include <boost/utility/in_place_factory.hpp>
15 #include "common/ceph_json.h"
16 #include "common/utf8.h"
18 #include "common/errno.h"
19 #include "common/Formatter.h"
20 #include "common/Throttle.h"
21 #include "common/Finisher.h"
23 #include "rgw_rados.h"
24 #include "rgw_cache.h"
26 #include "rgw_acl_s3.h" /* for dumping s3policy in debug log */
27 #include "rgw_metadata.h"
28 #include "rgw_bucket.h"
29 #include "rgw_rest_conn.h"
30 #include "rgw_cr_rados.h"
31 #include "rgw_cr_rest.h"
33 #include "cls/rgw/cls_rgw_ops.h"
34 #include "cls/rgw/cls_rgw_types.h"
35 #include "cls/rgw/cls_rgw_client.h"
36 #include "cls/rgw/cls_rgw_const.h"
37 #include "cls/refcount/cls_refcount_client.h"
38 #include "cls/version/cls_version_client.h"
39 #include "cls/log/cls_log_client.h"
40 #include "cls/statelog/cls_statelog_client.h"
41 #include "cls/timeindex/cls_timeindex_client.h"
42 #include "cls/lock/cls_lock_client.h"
43 #include "cls/user/cls_user_client.h"
44 #include "osd/osd_types.h"
46 #include "rgw_tools.h"
47 #include "rgw_coroutine.h"
48 #include "rgw_compression.h"
50 #undef fork // fails to compile RGWPeriod::fork() below
52 #include "common/Clock.h"
54 #include "include/rados/librados.hpp"
55 using namespace librados
;
63 #include "auth/Crypto.h" // get_random_bytes()
70 #include "rgw_object_expirer_core.h"
72 #include "rgw_data_sync.h"
73 #include "rgw_realm_watcher.h"
74 #include "rgw_reshard.h"
76 #include "compressor/Compressor.h"
78 #define dout_context g_ceph_context
79 #define dout_subsys ceph_subsys_rgw
83 static string notify_oid_prefix
= "notify";
84 static string
*notify_oids
= NULL
;
85 static string shadow_ns
= "shadow";
86 static string dir_oid_prefix
= ".dir.";
87 static string default_storage_pool_suffix
= "rgw.buckets.data";
88 static string default_bucket_index_pool_suffix
= "rgw.buckets.index";
89 static string default_storage_extra_pool_suffix
= "rgw.buckets.non-ec";
90 static string avail_pools
= ".pools.avail";
92 static string zone_info_oid_prefix
= "zone_info.";
93 static string zone_names_oid_prefix
= "zone_names.";
94 static string region_info_oid_prefix
= "region_info.";
95 static string zone_group_info_oid_prefix
= "zonegroup_info.";
96 static string realm_names_oid_prefix
= "realms_names.";
97 static string realm_info_oid_prefix
= "realms.";
98 static string default_region_info_oid
= "default.region";
99 static string default_zone_group_info_oid
= "default.zonegroup";
100 static string period_info_oid_prefix
= "periods.";
101 static string period_latest_epoch_info_oid
= ".latest_epoch";
102 static string region_map_oid
= "region_map";
103 static string zonegroup_map_oid
= "zonegroup_map";
104 static string log_lock_name
= "rgw_log_lock";
105 static string default_realm_info_oid
= "default.realm";
106 const string default_zonegroup_name
= "default";
107 const string default_zone_name
= "default";
108 static string zonegroup_names_oid_prefix
= "zonegroups_names.";
109 static RGWObjCategory main_category
= RGW_OBJ_CATEGORY_MAIN
;
110 #define RGW_USAGE_OBJ_PREFIX "usage."
111 #define FIRST_EPOCH 1
112 static string RGW_DEFAULT_ZONE_ROOT_POOL
= "rgw.root";
113 static string RGW_DEFAULT_ZONEGROUP_ROOT_POOL
= "rgw.root";
114 static string RGW_DEFAULT_REALM_ROOT_POOL
= "rgw.root";
115 static string RGW_DEFAULT_PERIOD_ROOT_POOL
= "rgw.root";
117 #define RGW_STATELOG_OBJ_PREFIX "statelog."
119 #define dout_subsys ceph_subsys_rgw
122 static bool rgw_get_obj_data_pool(const RGWZoneGroup
& zonegroup
, const RGWZoneParams
& zone_params
,
123 const string
& placement_id
, const rgw_obj
& obj
, rgw_pool
*pool
)
125 if (!zone_params
.get_head_data_pool(placement_id
, obj
, pool
)) {
126 RGWZonePlacementInfo placement
;
127 if (!zone_params
.get_placement(zonegroup
.default_placement
, &placement
)) {
131 if (!obj
.in_extra_data
) {
132 *pool
= placement
.data_pool
;
134 *pool
= placement
.get_data_extra_pool();
141 static bool rgw_obj_to_raw(const RGWZoneGroup
& zonegroup
, const RGWZoneParams
& zone_params
,
142 const string
& placement_id
, const rgw_obj
& obj
, rgw_raw_obj
*raw_obj
)
144 get_obj_bucket_and_oid_loc(obj
, raw_obj
->oid
, raw_obj
->loc
);
146 return rgw_get_obj_data_pool(zonegroup
, zone_params
, placement_id
, obj
, &raw_obj
->pool
);
149 rgw_raw_obj
rgw_obj_select::get_raw_obj(const RGWZoneGroup
& zonegroup
, const RGWZoneParams
& zone_params
) const
153 rgw_obj_to_raw(zonegroup
, zone_params
, placement_rule
, obj
, &r
);
159 rgw_raw_obj
rgw_obj_select::get_raw_obj(RGWRados
*store
) const
163 store
->obj_to_raw(placement_rule
, obj
, &r
);
169 int rgw_init_ioctx(librados::Rados
*rados
, const rgw_pool
& pool
, IoCtx
& ioctx
, bool create
)
171 int r
= rados
->ioctx_create(pool
.name
.c_str(), ioctx
);
172 if (r
== -ENOENT
&& create
) {
173 r
= rados
->pool_create(pool
.name
.c_str());
174 if (r
< 0 && r
!= -EEXIST
) {
178 r
= rados
->ioctx_create(pool
.name
.c_str(), ioctx
);
183 r
= ioctx
.application_enable(pg_pool_t::APPLICATION_NAME_RGW
, false);
184 if (r
< 0 && r
!= -EOPNOTSUPP
) {
190 if (!pool
.ns
.empty()) {
191 ioctx
.set_namespace(pool
.ns
);
197 void RGWObjectCtxImpl
<rgw_obj
, RGWObjState
>::invalidate(rgw_obj
& obj
) {
198 RWLock::WLocker
wl(lock
);
199 auto iter
= objs_state
.find(obj
);
200 if (iter
== objs_state
.end()) {
203 bool is_atomic
= iter
->second
.is_atomic
;
204 bool prefetch_data
= iter
->second
.prefetch_data
;
206 objs_state
.erase(iter
);
208 if (is_atomic
|| prefetch_data
) {
209 auto& s
= objs_state
[obj
];
210 s
.is_atomic
= is_atomic
;
211 s
.prefetch_data
= prefetch_data
;
216 void RGWObjectCtxImpl
<rgw_raw_obj
, RGWRawObjState
>::invalidate(rgw_raw_obj
& obj
) {
217 RWLock::WLocker
wl(lock
);
218 auto iter
= objs_state
.find(obj
);
219 if (iter
== objs_state
.end()) {
223 objs_state
.erase(iter
);
226 void RGWDefaultZoneGroupInfo::dump(Formatter
*f
) const {
227 encode_json("default_zonegroup", default_zonegroup
, f
);
230 void RGWDefaultZoneGroupInfo::decode_json(JSONObj
*obj
) {
232 JSONDecoder::decode_json("default_zonegroup", default_zonegroup
, obj
);
233 /* backward compatability with region */
234 if (default_zonegroup
.empty()) {
235 JSONDecoder::decode_json("default_region", default_zonegroup
, obj
);
239 rgw_pool
RGWZoneGroup::get_pool(CephContext
*cct_
)
241 if (cct_
->_conf
->rgw_zonegroup_root_pool
.empty()) {
242 return rgw_pool(RGW_DEFAULT_ZONEGROUP_ROOT_POOL
);
245 return rgw_pool(cct_
->_conf
->rgw_zonegroup_root_pool
);
248 int RGWZoneGroup::create_default(bool old_format
)
250 name
= default_zonegroup_name
;
253 RGWZoneGroupPlacementTarget placement_target
;
254 placement_target
.name
= "default-placement";
255 placement_targets
[placement_target
.name
] = placement_target
;
256 default_placement
= "default-placement";
258 RGWZoneParams
zone_params(default_zone_name
);
260 int r
= zone_params
.init(cct
, store
, false);
262 ldout(cct
, 0) << "create_default: error initializing zone params: " << cpp_strerror(-r
) << dendl
;
266 r
= zone_params
.create_default();
267 if (r
< 0 && r
!= -EEXIST
) {
268 ldout(cct
, 0) << "create_default: error in create_default zone params: " << cpp_strerror(-r
) << dendl
;
270 } else if (r
== -EEXIST
) {
271 ldout(cct
, 10) << "zone_params::create_default() returned -EEXIST, we raced with another default zone_params creation" << dendl
;
272 zone_params
.clear_id();
273 r
= zone_params
.init(cct
, store
);
275 ldout(cct
, 0) << "create_default: error in init existing zone params: " << cpp_strerror(-r
) << dendl
;
278 ldout(cct
, 20) << "zone_params::create_default() " << zone_params
.get_name() << " id " << zone_params
.get_id()
282 RGWZone
& default_zone
= zones
[zone_params
.get_id()];
283 default_zone
.name
= zone_params
.get_name();
284 default_zone
.id
= zone_params
.get_id();
285 master_zone
= default_zone
.id
;
288 if (r
< 0 && r
!= -EEXIST
) {
289 ldout(cct
, 0) << "error storing zone group info: " << cpp_strerror(-r
) << dendl
;
294 ldout(cct
, 10) << "create_default() returned -EEXIST, we raced with another zonegroup creation" << dendl
;
296 r
= init(cct
, store
);
306 post_process_params();
311 const string
RGWZoneGroup::get_default_oid(bool old_region_format
)
313 if (old_region_format
) {
314 if (cct
->_conf
->rgw_default_region_info_oid
.empty()) {
315 return default_region_info_oid
;
317 return cct
->_conf
->rgw_default_region_info_oid
;
320 string default_oid
= cct
->_conf
->rgw_default_zonegroup_info_oid
;
322 if (cct
->_conf
->rgw_default_zonegroup_info_oid
.empty()) {
323 default_oid
= default_zone_group_info_oid
;
326 default_oid
+= "." + realm_id
;
331 const string
& RGWZoneGroup::get_info_oid_prefix(bool old_region_format
)
333 if (old_region_format
) {
334 return region_info_oid_prefix
;
336 return zone_group_info_oid_prefix
;
339 const string
& RGWZoneGroup::get_names_oid_prefix()
341 return zonegroup_names_oid_prefix
;
344 const string
& RGWZoneGroup::get_predefined_name(CephContext
*cct
) {
345 return cct
->_conf
->rgw_zonegroup
;
348 int RGWZoneGroup::equals(const string
& other_zonegroup
) const
350 if (is_master
&& other_zonegroup
.empty())
353 return (id
== other_zonegroup
);
356 int RGWZoneGroup::add_zone(const RGWZoneParams
& zone_params
, bool *is_master
, bool *read_only
,
357 const list
<string
>& endpoints
, const string
*ptier_type
,
358 bool *psync_from_all
, list
<string
>& sync_from
, list
<string
>& sync_from_rm
)
360 auto& zone_id
= zone_params
.get_id();
361 auto& zone_name
= zone_params
.get_name();
363 // check for duplicate zone name on insert
364 if (!zones
.count(zone_id
)) {
365 for (const auto& zone
: zones
) {
366 if (zone
.second
.name
== zone_name
) {
367 ldout(cct
, 0) << "ERROR: found existing zone name " << zone_name
368 << " (" << zone
.first
<< ") in zonegroup " << get_name() << dendl
;
376 if (!master_zone
.empty() && master_zone
!= zone_params
.get_id()) {
377 ldout(cct
, 0) << "NOTICE: overriding master zone: " << master_zone
<< dendl
;
379 master_zone
= zone_params
.get_id();
380 } else if (master_zone
== zone_params
.get_id()) {
385 RGWZone
& zone
= zones
[zone_params
.get_id()];
386 zone
.name
= zone_params
.get_name();
387 zone
.id
= zone_params
.get_id();
388 if (!endpoints
.empty()) {
389 zone
.endpoints
= endpoints
;
392 zone
.read_only
= *read_only
;
395 zone
.tier_type
= *ptier_type
;
398 if (psync_from_all
) {
399 zone
.sync_from_all
= *psync_from_all
;
402 for (auto add
: sync_from
) {
403 zone
.sync_from
.insert(add
);
406 for (auto rm
: sync_from_rm
) {
407 zone
.sync_from
.erase(rm
);
410 post_process_params();
416 int RGWZoneGroup::rename_zone(const RGWZoneParams
& zone_params
)
418 RGWZone
& zone
= zones
[zone_params
.get_id()];
419 zone
.name
= zone_params
.get_name();
424 void RGWZoneGroup::post_process_params()
426 bool log_data
= zones
.size() > 1;
428 if (master_zone
.empty()) {
429 map
<string
, RGWZone
>::iterator iter
= zones
.begin();
430 if (iter
!= zones
.end()) {
431 master_zone
= iter
->first
;
435 for (map
<string
, RGWZone
>::iterator iter
= zones
.begin(); iter
!= zones
.end(); ++iter
) {
436 RGWZone
& zone
= iter
->second
;
437 zone
.log_data
= log_data
;
439 RGWZoneParams
zone_params(zone
.id
, zone
.name
);
440 int ret
= zone_params
.init(cct
, store
);
442 ldout(cct
, 0) << "WARNING: could not read zone params for zone id=" << zone
.id
<< " name=" << zone
.name
<< dendl
;
446 for (map
<string
, RGWZonePlacementInfo
>::iterator iter
= zone_params
.placement_pools
.begin();
447 iter
!= zone_params
.placement_pools
.end(); ++iter
) {
448 const string
& placement_name
= iter
->first
;
449 if (placement_targets
.find(placement_name
) == placement_targets
.end()) {
450 RGWZoneGroupPlacementTarget placement_target
;
451 placement_target
.name
= placement_name
;
452 placement_targets
[placement_name
] = placement_target
;
457 if (default_placement
.empty() && !placement_targets
.empty()) {
458 default_placement
= placement_targets
.begin()->first
;
462 int RGWZoneGroup::remove_zone(const std::string
& zone_id
)
464 map
<string
, RGWZone
>::iterator iter
= zones
.find(zone_id
);
465 if (iter
== zones
.end()) {
466 ldout(cct
, 0) << "zone id " << zone_id
<< " is not a part of zonegroup "
473 post_process_params();
478 int RGWZoneGroup::read_default_id(string
& default_id
, bool old_format
)
480 if (realm_id
.empty()) {
481 /* try using default realm */
483 int ret
= realm
.init(cct
, store
);
484 // no default realm exist
486 return read_id(default_zonegroup_name
, default_id
);
488 realm_id
= realm
.get_id();
491 return RGWSystemMetaObj::read_default_id(default_id
, old_format
);
494 int RGWZoneGroup::set_as_default(bool exclusive
)
496 if (realm_id
.empty()) {
497 /* try using default realm */
499 int ret
= realm
.init(cct
, store
);
501 ldout(cct
, 10) << "could not read realm id: " << cpp_strerror(-ret
) << dendl
;
504 realm_id
= realm
.get_id();
507 return RGWSystemMetaObj::set_as_default(exclusive
);
510 int RGWSystemMetaObj::init(CephContext
*_cct
, RGWRados
*_store
, bool setup_obj
, bool old_format
)
518 if (old_format
&& id
.empty()) {
525 name
= get_predefined_name(cct
);
528 r
= use_default(old_format
);
532 } else if (!old_format
) {
533 r
= read_id(name
, id
);
536 ldout(cct
, 0) << "error in read_id for object name: " << name
<< " : " << cpp_strerror(-r
) << dendl
;
543 return read_info(id
, old_format
);
546 int RGWSystemMetaObj::read_default(RGWDefaultSystemMetaObjInfo
& default_info
, const string
& oid
)
548 auto pool
= get_pool(cct
);
550 RGWObjectCtx
obj_ctx(store
);
551 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, oid
, bl
, NULL
, NULL
);
556 bufferlist::iterator iter
= bl
.begin();
557 ::decode(default_info
, iter
);
558 } catch (buffer::error
& err
) {
559 ldout(cct
, 0) << "error decoding data from " << pool
<< ":" << oid
<< dendl
;
566 int RGWSystemMetaObj::read_default_id(string
& default_id
, bool old_format
)
568 RGWDefaultSystemMetaObjInfo default_info
;
570 int ret
= read_default(default_info
, get_default_oid(old_format
));
575 default_id
= default_info
.default_id
;
580 int RGWSystemMetaObj::use_default(bool old_format
)
582 return read_default_id(id
, old_format
);
585 int RGWSystemMetaObj::set_as_default(bool exclusive
)
587 string oid
= get_default_oid();
589 rgw_pool
pool(get_pool(cct
));
592 RGWDefaultSystemMetaObjInfo default_info
;
593 default_info
.default_id
= id
;
595 ::encode(default_info
, bl
);
597 int ret
= rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(),
598 exclusive
, NULL
, real_time(), NULL
);
605 int RGWSystemMetaObj::read_id(const string
& obj_name
, string
& object_id
)
607 rgw_pool
pool(get_pool(cct
));
610 string oid
= get_names_oid_prefix() + obj_name
;
612 RGWObjectCtx
obj_ctx(store
);
613 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, oid
, bl
, NULL
, NULL
);
618 RGWNameToId nameToId
;
620 bufferlist::iterator iter
= bl
.begin();
621 ::decode(nameToId
, iter
);
622 } catch (buffer::error
& err
) {
623 ldout(cct
, 0) << "ERROR: failed to decode obj from " << pool
<< ":" << oid
<< dendl
;
626 object_id
= nameToId
.obj_id
;
630 int RGWSystemMetaObj::delete_obj(bool old_format
)
632 rgw_pool
pool(get_pool(cct
));
634 /* check to see if obj is the default */
635 RGWDefaultSystemMetaObjInfo default_info
;
636 int ret
= read_default(default_info
, get_default_oid(old_format
));
637 if (ret
< 0 && ret
!= -ENOENT
)
639 if (default_info
.default_id
== id
|| (old_format
&& default_info
.default_id
== name
)) {
640 string oid
= get_default_oid(old_format
);
641 rgw_raw_obj
default_named_obj(pool
, oid
);
642 ret
= store
->delete_system_obj(default_named_obj
);
644 ldout(cct
, 0) << "Error delete default obj name " << name
<< ": " << cpp_strerror(-ret
) << dendl
;
649 string oid
= get_names_oid_prefix() + name
;
650 rgw_raw_obj
object_name(pool
, oid
);
651 ret
= store
->delete_system_obj(object_name
);
653 ldout(cct
, 0) << "Error delete obj name " << name
<< ": " << cpp_strerror(-ret
) << dendl
;
658 string oid
= get_info_oid_prefix(old_format
);
665 rgw_raw_obj
object_id(pool
, oid
);
666 ret
= store
->delete_system_obj(object_id
);
668 ldout(cct
, 0) << "Error delete object id " << id
<< ": " << cpp_strerror(-ret
) << dendl
;
674 int RGWSystemMetaObj::store_name(bool exclusive
)
676 rgw_pool
pool(get_pool(cct
));
677 string oid
= get_names_oid_prefix() + name
;
679 RGWNameToId nameToId
;
680 nameToId
.obj_id
= id
;
683 ::encode(nameToId
, bl
);
684 return rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(), exclusive
, NULL
, real_time(), NULL
);
687 int RGWSystemMetaObj::rename(const string
& new_name
)
690 int ret
= read_id(new_name
, new_id
);
694 if (ret
< 0 && ret
!= -ENOENT
) {
695 ldout(cct
, 0) << "Error read_id " << new_name
<< ": " << cpp_strerror(-ret
) << dendl
;
698 string old_name
= name
;
702 ldout(cct
, 0) << "Error storing new obj info " << new_name
<< ": " << cpp_strerror(-ret
) << dendl
;
705 ret
= store_name(true);
707 ldout(cct
, 0) << "Error storing new name " << new_name
<< ": " << cpp_strerror(-ret
) << dendl
;
710 /* delete old name */
711 rgw_pool
pool(get_pool(cct
));
712 string oid
= get_names_oid_prefix() + old_name
;
713 rgw_raw_obj
old_name_obj(pool
, oid
);
714 ret
= store
->delete_system_obj(old_name_obj
);
716 ldout(cct
, 0) << "Error delete old obj name " << old_name
<< ": " << cpp_strerror(-ret
) << dendl
;
723 int RGWSystemMetaObj::read_info(const string
& obj_id
, bool old_format
)
725 rgw_pool
pool(get_pool(cct
));
729 string oid
= get_info_oid_prefix(old_format
) + obj_id
;
731 RGWObjectCtx
obj_ctx(store
);
732 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, oid
, bl
, NULL
, NULL
);
734 ldout(cct
, 0) << "failed reading obj info from " << pool
<< ":" << oid
<< ": " << cpp_strerror(-ret
) << dendl
;
739 bufferlist::iterator iter
= bl
.begin();
740 ::decode(*this, iter
);
741 } catch (buffer::error
& err
) {
742 ldout(cct
, 0) << "ERROR: failed to decode obj from " << pool
<< ":" << oid
<< dendl
;
749 int RGWSystemMetaObj::read()
751 int ret
= read_id(name
, id
);
756 return read_info(id
);
759 int RGWSystemMetaObj::create(bool exclusive
)
763 /* check to see the name is not used */
764 ret
= read_id(name
, id
);
765 if (exclusive
&& ret
== 0) {
766 ldout(cct
, 10) << "ERROR: name " << name
<< " already in use for obj id " << id
<< dendl
;
768 } else if ( ret
< 0 && ret
!= -ENOENT
) {
769 ldout(cct
, 0) << "failed reading obj id " << id
<< ": " << cpp_strerror(-ret
) << dendl
;
774 /* create unique id */
777 new_uuid
.generate_random();
778 new_uuid
.print(uuid_str
);
782 ret
= store_info(exclusive
);
784 ldout(cct
, 0) << "ERROR: storing info for " << id
<< ": " << cpp_strerror(-ret
) << dendl
;
788 return store_name(exclusive
);
791 int RGWSystemMetaObj::store_info(bool exclusive
)
793 rgw_pool
pool(get_pool(cct
));
795 string oid
= get_info_oid_prefix() + id
;
799 return rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(), exclusive
, NULL
, real_time(), NULL
);
802 int RGWSystemMetaObj::write(bool exclusive
)
804 int ret
= store_info(exclusive
);
806 ldout(cct
, 20) << __func__
<< "(): store_info() returned ret=" << ret
<< dendl
;
809 ret
= store_name(exclusive
);
811 ldout(cct
, 20) << __func__
<< "(): store_name() returned ret=" << ret
<< dendl
;
818 const string
& RGWRealm::get_predefined_name(CephContext
*cct
) {
819 return cct
->_conf
->rgw_realm
;
822 int RGWRealm::create(bool exclusive
)
824 int ret
= RGWSystemMetaObj::create(exclusive
);
826 ldout(cct
, 0) << "ERROR creating new realm object " << name
<< ": " << cpp_strerror(-ret
) << dendl
;
829 // create the control object for watch/notify
830 ret
= create_control(exclusive
);
832 ldout(cct
, 0) << "ERROR creating control for new realm " << name
<< ": " << cpp_strerror(-ret
) << dendl
;
836 if (current_period
.empty()) {
837 /* create new period for the realm */
838 ret
= period
.init(cct
, store
, id
, name
, false);
842 ret
= period
.create(true);
844 ldout(cct
, 0) << "ERROR: creating new period for realm " << name
<< ": " << cpp_strerror(-ret
) << dendl
;
848 period
= RGWPeriod(current_period
, 0);
849 int ret
= period
.init(cct
, store
, id
, name
);
851 ldout(cct
, 0) << "ERROR: failed to init period " << current_period
<< dendl
;
855 ret
= set_current_period(period
);
857 ldout(cct
, 0) << "ERROR: failed set current period " << current_period
<< dendl
;
860 // try to set as default. may race with another create, so pass exclusive=true
861 // so we don't override an existing default
862 ret
= set_as_default(true);
863 if (ret
< 0 && ret
!= -EEXIST
) {
864 ldout(cct
, 0) << "WARNING: failed to set realm as default realm, ret=" << ret
<< dendl
;
870 int RGWRealm::delete_obj()
872 int ret
= RGWSystemMetaObj::delete_obj();
876 return delete_control();
879 int RGWRealm::create_control(bool exclusive
)
881 auto pool
= rgw_pool
{get_pool(cct
)};
882 auto oid
= get_control_oid();
883 return rgw_put_system_obj(store
, pool
, oid
, nullptr, 0, exclusive
,
884 nullptr, real_time(), nullptr);
887 int RGWRealm::delete_control()
889 auto pool
= rgw_pool
{get_pool(cct
)};
890 auto obj
= rgw_raw_obj
{pool
, get_control_oid()};
891 return store
->delete_system_obj(obj
);
894 rgw_pool
RGWRealm::get_pool(CephContext
*cct
)
896 if (cct
->_conf
->rgw_realm_root_pool
.empty()) {
897 return rgw_pool(RGW_DEFAULT_REALM_ROOT_POOL
);
899 return rgw_pool(cct
->_conf
->rgw_realm_root_pool
);
902 const string
RGWRealm::get_default_oid(bool old_format
)
904 if (cct
->_conf
->rgw_default_realm_info_oid
.empty()) {
905 return default_realm_info_oid
;
907 return cct
->_conf
->rgw_default_realm_info_oid
;
910 const string
& RGWRealm::get_names_oid_prefix()
912 return realm_names_oid_prefix
;
915 const string
& RGWRealm::get_info_oid_prefix(bool old_format
)
917 return realm_info_oid_prefix
;
920 int RGWRealm::set_current_period(RGWPeriod
& period
)
922 // update realm epoch to match the period's
923 if (epoch
> period
.get_realm_epoch()) {
924 ldout(cct
, 0) << "ERROR: set_current_period with old realm epoch "
925 << period
.get_realm_epoch() << ", current epoch=" << epoch
<< dendl
;
928 if (epoch
== period
.get_realm_epoch() && current_period
!= period
.get_id()) {
929 ldout(cct
, 0) << "ERROR: set_current_period with same realm epoch "
930 << period
.get_realm_epoch() << ", but different period id "
931 << period
.get_id() << " != " << current_period
<< dendl
;
935 epoch
= period
.get_realm_epoch();
936 current_period
= period
.get_id();
940 ldout(cct
, 0) << "ERROR: period update: " << cpp_strerror(-ret
) << dendl
;
944 ret
= period
.reflect();
946 ldout(cct
, 0) << "ERROR: period.reflect(): " << cpp_strerror(-ret
) << dendl
;
953 string
RGWRealm::get_control_oid()
955 return get_info_oid_prefix() + id
+ ".control";
958 int RGWRealm::notify_zone(bufferlist
& bl
)
960 // open a context on the realm's pool
961 rgw_pool pool
{get_pool(cct
)};
963 int r
= rgw_init_ioctx(store
->get_rados_handle(), pool
, ctx
);
965 ldout(cct
, 0) << "Failed to open pool " << pool
<< dendl
;
968 // send a notify on the realm object
969 r
= ctx
.notify2(get_control_oid(), bl
, 0, nullptr);
971 ldout(cct
, 0) << "Realm notify failed with " << r
<< dendl
;
977 int RGWRealm::notify_new_period(const RGWPeriod
& period
)
980 // push the period to dependent zonegroups/zones
981 ::encode(RGWRealmNotify::ZonesNeedPeriod
, bl
);
982 ::encode(period
, bl
);
983 // reload the gateway with the new period
984 ::encode(RGWRealmNotify::Reload
, bl
);
986 return notify_zone(bl
);
989 std::string
RGWPeriodConfig::get_oid(const std::string
& realm_id
)
991 if (realm_id
.empty()) {
992 return "period_config.default";
994 return "period_config." + realm_id
;
997 rgw_pool
RGWPeriodConfig::get_pool(CephContext
*cct
)
999 const auto& pool_name
= cct
->_conf
->rgw_period_root_pool
;
1000 if (pool_name
.empty()) {
1001 return {RGW_DEFAULT_PERIOD_ROOT_POOL
};
1006 int RGWPeriodConfig::read(RGWRados
*store
, const std::string
& realm_id
)
1008 RGWObjectCtx
obj_ctx(store
);
1009 const auto& pool
= get_pool(store
->ctx());
1010 const auto& oid
= get_oid(realm_id
);
1013 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, oid
, bl
, nullptr, nullptr);
1018 bufferlist::iterator iter
= bl
.begin();
1019 ::decode(*this, iter
);
1020 } catch (buffer::error
& err
) {
1026 int RGWPeriodConfig::write(RGWRados
*store
, const std::string
& realm_id
)
1028 const auto& pool
= get_pool(store
->ctx());
1029 const auto& oid
= get_oid(realm_id
);
1031 ::encode(*this, bl
);
1032 return rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(),
1033 false, nullptr, real_time(), nullptr);
1036 int RGWPeriod::init(CephContext
*_cct
, RGWRados
*_store
, const string
& period_realm_id
,
1037 const string
& period_realm_name
, bool setup_obj
)
1041 realm_id
= period_realm_id
;
1042 realm_name
= period_realm_name
;
1047 return init(_cct
, _store
, setup_obj
);
1051 int RGWPeriod::init(CephContext
*_cct
, RGWRados
*_store
, bool setup_obj
)
1060 RGWRealm
realm(realm_id
, realm_name
);
1061 int ret
= realm
.init(cct
, store
);
1063 ldout(cct
, 0) << "RGWPeriod::init failed to init realm " << realm_name
<< " id " << realm_id
<< " : " <<
1064 cpp_strerror(-ret
) << dendl
;
1067 id
= realm
.get_current_period();
1068 realm_id
= realm
.get_id();
1072 int ret
= use_latest_epoch();
1074 ldout(cct
, 0) << "failed to use_latest_epoch period id " << id
<< " realm " << realm_name
<< " id " << realm_id
1075 << " : " << cpp_strerror(-ret
) << dendl
;
1084 int RGWPeriod::get_zonegroup(RGWZoneGroup
& zonegroup
, const string
& zonegroup_id
) {
1085 map
<string
, RGWZoneGroup
>::const_iterator iter
;
1086 if (!zonegroup_id
.empty()) {
1087 iter
= period_map
.zonegroups
.find(zonegroup_id
);
1089 iter
= period_map
.zonegroups
.find("default");
1091 if (iter
!= period_map
.zonegroups
.end()) {
1092 zonegroup
= iter
->second
;
1099 const string
& RGWPeriod::get_latest_epoch_oid()
1101 if (cct
->_conf
->rgw_period_latest_epoch_info_oid
.empty()) {
1102 return period_latest_epoch_info_oid
;
1104 return cct
->_conf
->rgw_period_latest_epoch_info_oid
;
1107 const string
& RGWPeriod::get_info_oid_prefix()
1109 return period_info_oid_prefix
;
1112 const string
RGWPeriod::get_period_oid_prefix()
1114 return get_info_oid_prefix() + id
;
1117 const string
RGWPeriod::get_period_oid()
1119 std::ostringstream oss
;
1120 oss
<< get_period_oid_prefix();
1121 // skip the epoch for the staging period
1122 if (id
!= get_staging_id(realm_id
))
1123 oss
<< "." << epoch
;
1127 int RGWPeriod::read_latest_epoch(RGWPeriodLatestEpochInfo
& info
,
1128 RGWObjVersionTracker
*objv
)
1130 string oid
= get_period_oid_prefix() + get_latest_epoch_oid();
1132 rgw_pool
pool(get_pool(cct
));
1134 RGWObjectCtx
obj_ctx(store
);
1135 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, oid
, bl
, objv
, nullptr);
1137 ldout(cct
, 1) << "error read_lastest_epoch " << pool
<< ":" << oid
<< dendl
;
1141 bufferlist::iterator iter
= bl
.begin();
1142 ::decode(info
, iter
);
1143 } catch (buffer::error
& err
) {
1144 ldout(cct
, 0) << "error decoding data from " << pool
<< ":" << oid
<< dendl
;
1151 int RGWPeriod::get_latest_epoch(epoch_t
& latest_epoch
)
1153 RGWPeriodLatestEpochInfo info
;
1155 int ret
= read_latest_epoch(info
);
1160 latest_epoch
= info
.epoch
;
1165 int RGWPeriod::use_latest_epoch()
1167 RGWPeriodLatestEpochInfo info
;
1168 int ret
= read_latest_epoch(info
);
1178 int RGWPeriod::set_latest_epoch(epoch_t epoch
, bool exclusive
,
1179 RGWObjVersionTracker
*objv
)
1181 string oid
= get_period_oid_prefix() + get_latest_epoch_oid();
1183 rgw_pool
pool(get_pool(cct
));
1186 RGWPeriodLatestEpochInfo info
;
1191 return rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(),
1192 exclusive
, objv
, real_time(), nullptr);
1195 int RGWPeriod::update_latest_epoch(epoch_t epoch
)
1197 static constexpr int MAX_RETRIES
= 20;
1199 for (int i
= 0; i
< MAX_RETRIES
; i
++) {
1200 RGWPeriodLatestEpochInfo info
;
1201 RGWObjVersionTracker objv
;
1202 bool exclusive
= false;
1204 // read existing epoch
1205 int r
= read_latest_epoch(info
, &objv
);
1207 // use an exclusive create to set the epoch atomically
1209 ldout(cct
, 20) << "creating initial latest_epoch=" << epoch
1210 << " for period=" << id
<< dendl
;
1212 ldout(cct
, 0) << "ERROR: failed to read latest_epoch" << dendl
;
1214 } else if (epoch
<= info
.epoch
) {
1215 r
= -EEXIST
; // fail with EEXIST if epoch is not newer
1216 ldout(cct
, 1) << "found existing latest_epoch " << info
.epoch
1217 << " >= given epoch " << epoch
<< ", returning r=" << r
<< dendl
;
1220 ldout(cct
, 20) << "updating latest_epoch from " << info
.epoch
1221 << " -> " << epoch
<< " on period=" << id
<< dendl
;
1224 r
= set_latest_epoch(epoch
, exclusive
, &objv
);
1226 continue; // exclusive create raced with another update, retry
1227 } else if (r
== -ECANCELED
) {
1228 continue; // write raced with a conflicting version, retry
1231 ldout(cct
, 0) << "ERROR: failed to write latest_epoch" << dendl
;
1234 return 0; // return success
1237 return -ECANCELED
; // fail after max retries
1240 int RGWPeriod::delete_obj()
1242 rgw_pool
pool(get_pool(cct
));
1244 // delete the object for each period epoch
1245 for (epoch_t e
= 1; e
<= epoch
; e
++) {
1246 RGWPeriod p
{get_id(), e
};
1247 rgw_raw_obj oid
{pool
, p
.get_period_oid()};
1248 int ret
= store
->delete_system_obj(oid
);
1250 ldout(cct
, 0) << "WARNING: failed to delete period object " << oid
1251 << ": " << cpp_strerror(-ret
) << dendl
;
1255 // delete the .latest_epoch object
1256 rgw_raw_obj oid
{pool
, get_period_oid_prefix() + get_latest_epoch_oid()};
1257 int ret
= store
->delete_system_obj(oid
);
1259 ldout(cct
, 0) << "WARNING: failed to delete period object " << oid
1260 << ": " << cpp_strerror(-ret
) << dendl
;
1265 int RGWPeriod::read_info()
1267 rgw_pool
pool(get_pool(cct
));
1271 RGWObjectCtx
obj_ctx(store
);
1272 int ret
= rgw_get_system_obj(store
, obj_ctx
, pool
, get_period_oid(), bl
, NULL
, NULL
);
1274 ldout(cct
, 0) << "failed reading obj info from " << pool
<< ":" << get_period_oid() << ": " << cpp_strerror(-ret
) << dendl
;
1279 bufferlist::iterator iter
= bl
.begin();
1280 ::decode(*this, iter
);
1281 } catch (buffer::error
& err
) {
1282 ldout(cct
, 0) << "ERROR: failed to decode obj from " << pool
<< ":" << get_period_oid() << dendl
;
1289 int RGWPeriod::create(bool exclusive
)
1293 /* create unique id */
1296 new_uuid
.generate_random();
1297 new_uuid
.print(uuid_str
);
1300 epoch
= FIRST_EPOCH
;
1304 ret
= store_info(exclusive
);
1306 ldout(cct
, 0) << "ERROR: storing info for " << id
<< ": " << cpp_strerror(-ret
) << dendl
;
1310 ret
= set_latest_epoch(epoch
);
1312 ldout(cct
, 0) << "ERROR: setting latest epoch " << id
<< ": " << cpp_strerror(-ret
) << dendl
;
1318 int RGWPeriod::store_info(bool exclusive
)
1320 rgw_pool
pool(get_pool(cct
));
1322 string oid
= get_period_oid();
1324 ::encode(*this, bl
);
1326 return rgw_put_system_obj(store
, pool
, oid
, bl
.c_str(), bl
.length(),
1327 exclusive
, NULL
, real_time(), NULL
);
1330 rgw_pool
RGWPeriod::get_pool(CephContext
*cct
)
1332 if (cct
->_conf
->rgw_period_root_pool
.empty()) {
1333 return rgw_pool(RGW_DEFAULT_PERIOD_ROOT_POOL
);
1335 return rgw_pool(cct
->_conf
->rgw_period_root_pool
);
1338 int RGWPeriod::add_zonegroup(const RGWZoneGroup
& zonegroup
)
1340 if (zonegroup
.realm_id
!= realm_id
) {
1343 int ret
= period_map
.update(zonegroup
, cct
);
1345 ldout(cct
, 0) << "ERROR: updating period map: " << cpp_strerror(-ret
) << dendl
;
1349 return store_info(false);
1352 int RGWPeriod::update()
1354 ldout(cct
, 20) << __func__
<< " realm " << realm_id
<< " period " << get_id() << dendl
;
1355 list
<string
> zonegroups
;
1356 int ret
= store
->list_zonegroups(zonegroups
);
1358 ldout(cct
, 0) << "ERROR: failed to list zonegroups: " << cpp_strerror(-ret
) << dendl
;
1362 // clear zone short ids of removed zones. period_map.update() will add the
1363 // remaining zones back
1364 period_map
.short_zone_ids
.clear();
1366 for (auto& iter
: zonegroups
) {
1367 RGWZoneGroup
zg(string(), iter
);
1368 ret
= zg
.init(cct
, store
);
1370 ldout(cct
, 0) << "WARNING: zg.init() failed: " << cpp_strerror(-ret
) << dendl
;
1374 if (zg
.realm_id
!= realm_id
) {
1375 ldout(cct
, 20) << "skipping zonegroup " << zg
.get_name() << " zone realm id " << zg
.realm_id
<< ", not on our realm " << realm_id
<< dendl
;
1379 if (zg
.master_zone
.empty()) {
1380 ldout(cct
, 0) << "ERROR: zonegroup " << zg
.get_name() << " should have a master zone " << dendl
;
1384 if (zg
.is_master_zonegroup()) {
1385 master_zonegroup
= zg
.get_id();
1386 master_zone
= zg
.master_zone
;
1389 int ret
= period_map
.update(zg
, cct
);
1395 ret
= period_config
.read(store
, realm_id
);
1396 if (ret
< 0 && ret
!= -ENOENT
) {
1397 ldout(cct
, 0) << "ERROR: failed to read period config: "
1398 << cpp_strerror(ret
) << dendl
;
1404 int RGWPeriod::reflect()
1406 for (auto& iter
: period_map
.zonegroups
) {
1407 RGWZoneGroup
& zg
= iter
.second
;
1408 zg
.reinit_instance(cct
, store
);
1409 int r
= zg
.write(false);
1411 ldout(cct
, 0) << "ERROR: failed to store zonegroup info for zonegroup=" << iter
.first
<< ": " << cpp_strerror(-r
) << dendl
;
1414 if (zg
.is_master_zonegroup()) {
1415 // set master as default if no default exists
1416 r
= zg
.set_as_default(true);
1418 ldout(cct
, 1) << "Set the period's master zonegroup " << zg
.get_id()
1419 << " as the default" << dendl
;
1424 int r
= period_config
.write(store
, realm_id
);
1426 ldout(cct
, 0) << "ERROR: failed to store period config: "
1427 << cpp_strerror(-r
) << dendl
;
1433 void RGWPeriod::fork()
1435 ldout(cct
, 20) << __func__
<< " realm " << realm_id
<< " period " << id
<< dendl
;
1436 predecessor_uuid
= id
;
1437 id
= get_staging_id(realm_id
);
1442 static int read_sync_status(RGWRados
*store
, rgw_meta_sync_status
*sync_status
)
1444 // initialize a sync status manager to read the status
1445 RGWMetaSyncStatusManager
mgr(store
, store
->get_async_rados());
1450 r
= mgr
.read_sync_status(sync_status
);
1455 int RGWPeriod::update_sync_status(const RGWPeriod
¤t_period
,
1456 std::ostream
& error_stream
,
1457 bool force_if_stale
)
1459 rgw_meta_sync_status status
;
1460 int r
= read_sync_status(store
, &status
);
1462 ldout(cct
, 0) << "period failed to read sync status: "
1463 << cpp_strerror(-r
) << dendl
;
1467 std::vector
<std::string
> markers
;
1469 const auto current_epoch
= current_period
.get_realm_epoch();
1470 if (current_epoch
!= status
.sync_info
.realm_epoch
) {
1471 // no sync status markers for the current period
1472 assert(current_epoch
> status
.sync_info
.realm_epoch
);
1473 const int behind
= current_epoch
- status
.sync_info
.realm_epoch
;
1474 if (!force_if_stale
&& current_epoch
> 1) {
1475 error_stream
<< "ERROR: This zone is " << behind
<< " period(s) behind "
1476 "the current master zone in metadata sync. If this zone is promoted "
1477 "to master, any metadata changes during that time are likely to "
1479 "Waiting for this zone to catch up on metadata sync (see "
1480 "'radosgw-admin sync status') is recommended.\n"
1481 "To promote this zone to master anyway, add the flag "
1482 "--yes-i-really-mean-it." << std::endl
;
1485 // empty sync status markers - other zones will skip this period during
1486 // incremental metadata sync
1487 markers
.resize(status
.sync_info
.num_shards
);
1489 markers
.reserve(status
.sync_info
.num_shards
);
1490 for (auto& i
: status
.sync_markers
) {
1491 auto& marker
= i
.second
;
1492 // filter out markers from other periods
1493 if (marker
.realm_epoch
!= current_epoch
) {
1494 marker
.marker
.clear();
1496 markers
.emplace_back(std::move(marker
.marker
));
1500 std::swap(sync_status
, markers
);
1504 int RGWPeriod::commit(RGWRealm
& realm
, const RGWPeriod
& current_period
,
1505 std::ostream
& error_stream
, bool force_if_stale
)
1507 ldout(cct
, 20) << __func__
<< " realm " << realm
.get_id() << " period " << current_period
.get_id() << dendl
;
1508 // gateway must be in the master zone to commit
1509 if (master_zone
!= store
->get_zone_params().get_id()) {
1510 error_stream
<< "Cannot commit period on zone "
1511 << store
->get_zone_params().get_id() << ", it must be sent to "
1512 "the period's master zone " << master_zone
<< '.' << std::endl
;
1515 // period predecessor must match current period
1516 if (predecessor_uuid
!= current_period
.get_id()) {
1517 error_stream
<< "Period predecessor " << predecessor_uuid
1518 << " does not match current period " << current_period
.get_id()
1519 << ". Use 'period pull' to get the latest period from the master, "
1520 "reapply your changes, and try again." << std::endl
;
1523 // realm epoch must be 1 greater than current period
1524 if (realm_epoch
!= current_period
.get_realm_epoch() + 1) {
1525 error_stream
<< "Period's realm epoch " << realm_epoch
1526 << " does not come directly after current realm epoch "
1527 << current_period
.get_realm_epoch() << ". Use 'realm pull' to get the "
1528 "latest realm and period from the master zone, reapply your changes, "
1529 "and try again." << std::endl
;
1532 // did the master zone change?
1533 if (master_zone
!= current_period
.get_master_zone()) {
1534 // store the current metadata sync status in the period
1535 int r
= update_sync_status(current_period
, error_stream
, force_if_stale
);
1537 ldout(cct
, 0) << "failed to update metadata sync status: "
1538 << cpp_strerror(-r
) << dendl
;
1541 // create an object with a new period id
1544 ldout(cct
, 0) << "failed to create new period: " << cpp_strerror(-r
) << dendl
;
1547 // set as current period
1548 r
= realm
.set_current_period(*this);
1550 ldout(cct
, 0) << "failed to update realm's current period: "
1551 << cpp_strerror(-r
) << dendl
;
1554 ldout(cct
, 4) << "Promoted to master zone and committed new period "
1556 realm
.notify_new_period(*this);
1559 // period must be based on current epoch
1560 if (epoch
!= current_period
.get_epoch()) {
1561 error_stream
<< "Period epoch " << epoch
<< " does not match "
1562 "predecessor epoch " << current_period
.get_epoch()
1563 << ". Use 'period pull' to get the latest epoch from the master zone, "
1564 "reapply your changes, and try again." << std::endl
;
1567 // set period as next epoch
1568 set_id(current_period
.get_id());
1569 set_epoch(current_period
.get_epoch() + 1);
1570 set_predecessor(current_period
.get_predecessor());
1571 realm_epoch
= current_period
.get_realm_epoch();
1572 // write the period to rados
1573 int r
= store_info(false);
1575 ldout(cct
, 0) << "failed to store period: " << cpp_strerror(-r
) << dendl
;
1578 // set as latest epoch
1579 r
= update_latest_epoch(epoch
);
1581 // already have this epoch (or a more recent one)
1585 ldout(cct
, 0) << "failed to set latest epoch: " << cpp_strerror(-r
) << dendl
;
1590 ldout(cct
, 0) << "failed to update local objects: " << cpp_strerror(-r
) << dendl
;
1593 ldout(cct
, 4) << "Committed new epoch " << epoch
1594 << " for period " << id
<< dendl
;
1595 realm
.notify_new_period(*this);
1599 int RGWZoneParams::create_default(bool old_format
)
1601 name
= default_zone_name
;
1616 int get_zones_pool_set(CephContext
* cct
,
1618 const list
<string
>& zones
,
1619 const string
& my_zone_id
,
1620 set
<rgw_pool
>& pool_names
)
1622 for(auto const& iter
: zones
) {
1623 RGWZoneParams
zone(iter
);
1624 int r
= zone
.init(cct
, store
);
1626 ldout(cct
, 0) << "Error: init zone " << iter
<< ":" << cpp_strerror(-r
) << dendl
;
1629 if (zone
.get_id() != my_zone_id
) {
1630 pool_names
.insert(zone
.domain_root
);
1631 pool_names
.insert(zone
.metadata_heap
);
1632 pool_names
.insert(zone
.control_pool
);
1633 pool_names
.insert(zone
.gc_pool
);
1634 pool_names
.insert(zone
.log_pool
);
1635 pool_names
.insert(zone
.intent_log_pool
);
1636 pool_names
.insert(zone
.usage_log_pool
);
1637 pool_names
.insert(zone
.user_keys_pool
);
1638 pool_names
.insert(zone
.user_email_pool
);
1639 pool_names
.insert(zone
.user_swift_pool
);
1640 pool_names
.insert(zone
.user_uid_pool
);
1641 pool_names
.insert(zone
.roles_pool
);
1642 pool_names
.insert(zone
.reshard_pool
);
1643 for(auto& iter
: zone
.placement_pools
) {
1644 pool_names
.insert(iter
.second
.index_pool
);
1645 pool_names
.insert(iter
.second
.data_pool
);
1646 pool_names
.insert(iter
.second
.data_extra_pool
);
1653 rgw_pool
fix_zone_pool_dup(set
<rgw_pool
> pools
,
1654 const string
& default_prefix
,
1655 const string
& default_suffix
,
1656 const rgw_pool
& suggested_pool
)
1658 string suggested_name
= suggested_pool
.to_str();
1660 string prefix
= default_prefix
;
1661 string suffix
= default_suffix
;
1663 if (!suggested_pool
.empty()) {
1664 prefix
= suggested_name
.substr(0, suggested_name
.find("."));
1665 suffix
= suggested_name
.substr(prefix
.length());
1668 rgw_pool
pool(prefix
+ suffix
);
1670 if (pools
.find(pool
) == pools
.end()) {
1674 pool
= prefix
+ "_" + std::to_string(std::rand()) + suffix
;
1675 if (pools
.find(pool
) == pools
.end()) {
1682 int RGWZoneParams::fix_pool_names()
1686 int r
= store
->list_zones(zones
);
1688 ldout(cct
, 10) << "WARNING: store->list_zones() returned r=" << r
<< dendl
;
1691 set
<rgw_pool
> pools
;
1692 r
= get_zones_pool_set(cct
, store
, zones
, id
, pools
);
1694 ldout(cct
, 0) << "Error: get_zones_pool_names" << r
<< dendl
;
1698 domain_root
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:root", domain_root
);
1699 if (!metadata_heap
.name
.empty()) {
1700 metadata_heap
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:heap", metadata_heap
);
1702 control_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.control", control_pool
);
1703 gc_pool
= fix_zone_pool_dup(pools
, name
,".rgw.log:gc", gc_pool
);
1704 lc_pool
= fix_zone_pool_dup(pools
, name
,".rgw.log:lc", lc_pool
);
1705 log_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.log", log_pool
);
1706 intent_log_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.log:intent", intent_log_pool
);
1707 usage_log_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.log:usage", usage_log_pool
);
1708 user_keys_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:users.keys", user_keys_pool
);
1709 user_email_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:users.email", user_email_pool
);
1710 user_swift_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:users.swift", user_swift_pool
);
1711 user_uid_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:users.uid", user_uid_pool
);
1712 roles_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.meta:roles", roles_pool
);
1713 reshard_pool
= fix_zone_pool_dup(pools
, name
, ".rgw.log:reshard", reshard_pool
);
1715 for(auto& iter
: placement_pools
) {
1716 iter
.second
.index_pool
= fix_zone_pool_dup(pools
, name
, "." + default_bucket_index_pool_suffix
,
1717 iter
.second
.index_pool
);
1718 iter
.second
.data_pool
= fix_zone_pool_dup(pools
, name
, "." + default_storage_pool_suffix
,
1719 iter
.second
.data_pool
);
1720 iter
.second
.data_extra_pool
= fix_zone_pool_dup(pools
, name
, "." + default_storage_extra_pool_suffix
,
1721 iter
.second
.data_extra_pool
);
1727 int RGWZoneParams::create(bool exclusive
)
1729 /* check for old pools config */
1730 rgw_raw_obj
obj(domain_root
, avail_pools
);
1731 int r
= store
->raw_obj_stat(obj
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
);
1733 ldout(store
->ctx(), 10) << "couldn't find old data placement pools config, setting up new ones for the zone" << dendl
;
1734 /* a new system, let's set new placement info */
1735 RGWZonePlacementInfo default_placement
;
1736 default_placement
.index_pool
= name
+ "." + default_bucket_index_pool_suffix
;
1737 default_placement
.data_pool
= name
+ "." + default_storage_pool_suffix
;
1738 default_placement
.data_extra_pool
= name
+ "." + default_storage_extra_pool_suffix
;
1739 placement_pools
["default-placement"] = default_placement
;
1742 r
= fix_pool_names();
1744 ldout(cct
, 0) << "ERROR: fix_pool_names returned r=" << r
<< dendl
;
1748 r
= RGWSystemMetaObj::create(exclusive
);
1753 // try to set as default. may race with another create, so pass exclusive=true
1754 // so we don't override an existing default
1755 r
= set_as_default(true);
1756 if (r
< 0 && r
!= -EEXIST
) {
1757 ldout(cct
, 10) << "WARNING: failed to set zone as default, r=" << r
<< dendl
;
1763 rgw_pool
RGWZoneParams::get_pool(CephContext
*cct
)
1765 if (cct
->_conf
->rgw_zone_root_pool
.empty()) {
1766 return rgw_pool(RGW_DEFAULT_ZONE_ROOT_POOL
);
1769 return rgw_pool(cct
->_conf
->rgw_zone_root_pool
);
1772 const string
RGWZoneParams::get_default_oid(bool old_format
)
1775 return cct
->_conf
->rgw_default_zone_info_oid
;
1778 return cct
->_conf
->rgw_default_zone_info_oid
+ "." + realm_id
;
1781 const string
& RGWZoneParams::get_names_oid_prefix()
1783 return zone_names_oid_prefix
;
1786 const string
& RGWZoneParams::get_info_oid_prefix(bool old_format
)
1788 return zone_info_oid_prefix
;
1791 const string
& RGWZoneParams::get_predefined_name(CephContext
*cct
) {
1792 return cct
->_conf
->rgw_zone
;
1795 int RGWZoneParams::init(CephContext
*cct
, RGWRados
*store
, bool setup_obj
, bool old_format
)
1798 name
= cct
->_conf
->rgw_zone
;
1801 return RGWSystemMetaObj::init(cct
, store
, setup_obj
, old_format
);
1804 int RGWZoneParams::read_default_id(string
& default_id
, bool old_format
)
1806 if (realm_id
.empty()) {
1807 /* try using default realm */
1809 int ret
= realm
.init(cct
, store
);
1810 //no default realm exist
1812 return read_id(default_zone_name
, default_id
);
1814 realm_id
= realm
.get_id();
1817 return RGWSystemMetaObj::read_default_id(default_id
, old_format
);
1821 int RGWZoneParams::set_as_default(bool exclusive
)
1823 if (realm_id
.empty()) {
1824 /* try using default realm */
1826 int ret
= realm
.init(cct
, store
);
1828 ldout(cct
, 10) << "could not read realm id: " << cpp_strerror(-ret
) << dendl
;
1831 realm_id
= realm
.get_id();
1834 return RGWSystemMetaObj::set_as_default(exclusive
);
1837 const string
& RGWZoneParams::get_compression_type(const string
& placement_rule
) const
1839 static const std::string NONE
{"none"};
1840 auto p
= placement_pools
.find(placement_rule
);
1841 if (p
== placement_pools
.end()) {
1844 const auto& type
= p
->second
.compression_type
;
1845 return !type
.empty() ? type
: NONE
;
1848 void RGWPeriodMap::encode(bufferlist
& bl
) const {
1849 ENCODE_START(2, 1, bl
);
1851 ::encode(zonegroups
, bl
);
1852 ::encode(master_zonegroup
, bl
);
1853 ::encode(short_zone_ids
, bl
);
1857 void RGWPeriodMap::decode(bufferlist::iterator
& bl
) {
1858 DECODE_START(2, bl
);
1860 ::decode(zonegroups
, bl
);
1861 ::decode(master_zonegroup
, bl
);
1862 if (struct_v
>= 2) {
1863 ::decode(short_zone_ids
, bl
);
1867 zonegroups_by_api
.clear();
1868 for (map
<string
, RGWZoneGroup
>::iterator iter
= zonegroups
.begin();
1869 iter
!= zonegroups
.end(); ++iter
) {
1870 RGWZoneGroup
& zonegroup
= iter
->second
;
1871 zonegroups_by_api
[zonegroup
.api_name
] = zonegroup
;
1872 if (zonegroup
.is_master_zonegroup()) {
1873 master_zonegroup
= zonegroup
.get_id();
1878 // run an MD5 hash on the zone_id and return the first 32 bits
1879 static uint32_t gen_short_zone_id(const std::string zone_id
)
1881 unsigned char md5
[CEPH_CRYPTO_MD5_DIGESTSIZE
];
1883 hash
.Update((const byte
*)zone_id
.c_str(), zone_id
.size());
1887 memcpy((char *)&short_id
, md5
, sizeof(short_id
));
1888 return std::max(short_id
, 1u);
1891 int RGWPeriodMap::update(const RGWZoneGroup
& zonegroup
, CephContext
*cct
)
1893 if (zonegroup
.is_master_zonegroup() && (!master_zonegroup
.empty() && zonegroup
.get_id() != master_zonegroup
)) {
1894 ldout(cct
,0) << "Error updating periodmap, multiple master zonegroups configured "<< dendl
;
1895 ldout(cct
,0) << "master zonegroup: " << master_zonegroup
<< " and " << zonegroup
.get_id() <<dendl
;
1898 map
<string
, RGWZoneGroup
>::iterator iter
= zonegroups
.find(zonegroup
.get_id());
1899 if (iter
!= zonegroups
.end()) {
1900 RGWZoneGroup
& old_zonegroup
= iter
->second
;
1901 if (!old_zonegroup
.api_name
.empty()) {
1902 zonegroups_by_api
.erase(old_zonegroup
.api_name
);
1905 zonegroups
[zonegroup
.get_id()] = zonegroup
;
1907 if (!zonegroup
.api_name
.empty()) {
1908 zonegroups_by_api
[zonegroup
.api_name
] = zonegroup
;
1911 if (zonegroup
.is_master_zonegroup()) {
1912 master_zonegroup
= zonegroup
.get_id();
1913 } else if (master_zonegroup
== zonegroup
.get_id()) {
1914 master_zonegroup
= "";
1917 for (auto& i
: zonegroup
.zones
) {
1918 auto& zone
= i
.second
;
1919 if (short_zone_ids
.find(zone
.id
) != short_zone_ids
.end()) {
1922 // calculate the zone's short id
1923 uint32_t short_id
= gen_short_zone_id(zone
.id
);
1925 // search for an existing zone with the same short id
1926 for (auto& s
: short_zone_ids
) {
1927 if (s
.second
== short_id
) {
1928 ldout(cct
, 0) << "New zone '" << zone
.name
<< "' (" << zone
.id
1929 << ") generates the same short_zone_id " << short_id
1930 << " as existing zone id " << s
.first
<< dendl
;
1935 short_zone_ids
[zone
.id
] = short_id
;
1941 uint32_t RGWPeriodMap::get_zone_short_id(const string
& zone_id
) const
1943 auto i
= short_zone_ids
.find(zone_id
);
1944 if (i
== short_zone_ids
.end()) {
1950 int RGWZoneGroupMap::read(CephContext
*cct
, RGWRados
*store
)
1954 int ret
= period
.init(cct
, store
);
1956 cerr
<< "failed to read current period info: " << cpp_strerror(ret
);
1960 bucket_quota
= period
.get_config().bucket_quota
;
1961 user_quota
= period
.get_config().user_quota
;
1962 zonegroups
= period
.get_map().zonegroups
;
1963 zonegroups_by_api
= period
.get_map().zonegroups_by_api
;
1964 master_zonegroup
= period
.get_map().master_zonegroup
;
1969 void RGWRegionMap::encode(bufferlist
& bl
) const {
1970 ENCODE_START( 3, 1, bl
);
1971 ::encode(regions
, bl
);
1972 ::encode(master_region
, bl
);
1973 ::encode(bucket_quota
, bl
);
1974 ::encode(user_quota
, bl
);
1978 void RGWRegionMap::decode(bufferlist::iterator
& bl
) {
1979 DECODE_START(3, bl
);
1980 ::decode(regions
, bl
);
1981 ::decode(master_region
, bl
);
1983 ::decode(bucket_quota
, bl
);
1985 ::decode(user_quota
, bl
);
1989 void RGWZoneGroupMap::encode(bufferlist
& bl
) const {
1990 ENCODE_START( 3, 1, bl
);
1991 ::encode(zonegroups
, bl
);
1992 ::encode(master_zonegroup
, bl
);
1993 ::encode(bucket_quota
, bl
);
1994 ::encode(user_quota
, bl
);
1998 void RGWZoneGroupMap::decode(bufferlist::iterator
& bl
) {
1999 DECODE_START(3, bl
);
2000 ::decode(zonegroups
, bl
);
2001 ::decode(master_zonegroup
, bl
);
2003 ::decode(bucket_quota
, bl
);
2005 ::decode(user_quota
, bl
);
2008 zonegroups_by_api
.clear();
2009 for (map
<string
, RGWZoneGroup
>::iterator iter
= zonegroups
.begin();
2010 iter
!= zonegroups
.end(); ++iter
) {
2011 RGWZoneGroup
& zonegroup
= iter
->second
;
2012 zonegroups_by_api
[zonegroup
.api_name
] = zonegroup
;
2013 if (zonegroup
.is_master_zonegroup()) {
2014 master_zonegroup
= zonegroup
.get_name();
2019 void RGWObjVersionTracker::prepare_op_for_read(ObjectReadOperation
*op
)
2021 obj_version
*check_objv
= version_for_check();
2024 cls_version_check(*op
, *check_objv
, VER_COND_EQ
);
2027 cls_version_read(*op
, &read_version
);
2030 void RGWObjVersionTracker::prepare_op_for_write(ObjectWriteOperation
*op
)
2032 obj_version
*check_objv
= version_for_check();
2033 obj_version
*modify_version
= version_for_write();
2036 cls_version_check(*op
, *check_objv
, VER_COND_EQ
);
2039 if (modify_version
) {
2040 cls_version_set(*op
, *modify_version
);
2042 cls_version_inc(*op
);
2046 void RGWObjManifest::obj_iterator::operator++()
2048 if (manifest
->explicit_objs
) {
2051 if (explicit_iter
== manifest
->objs
.end()) {
2052 ofs
= manifest
->obj_size
;
2056 update_explicit_pos();
2062 uint64_t obj_size
= manifest
->get_obj_size();
2063 uint64_t head_size
= manifest
->get_head_size();
2065 if (ofs
== obj_size
) {
2069 if (manifest
->rules
.empty()) {
2073 /* are we still pointing at the head? */
2074 if (ofs
< head_size
) {
2075 rule_iter
= manifest
->rules
.begin();
2076 RGWObjManifestRule
*rule
= &rule_iter
->second
;
2077 ofs
= MIN(head_size
, obj_size
);
2080 stripe_size
= MIN(obj_size
- ofs
, rule
->stripe_max_size
);
2081 if (rule
->part_size
> 0) {
2082 stripe_size
= MIN(stripe_size
, rule
->part_size
);
2088 RGWObjManifestRule
*rule
= &rule_iter
->second
;
2090 stripe_ofs
+= rule
->stripe_max_size
;
2092 dout(20) << "RGWObjManifest::operator++(): rule->part_size=" << rule
->part_size
<< " rules.size()=" << manifest
->rules
.size() << dendl
;
2094 if (rule
->part_size
> 0) {
2095 /* multi part, multi stripes object */
2097 dout(20) << "RGWObjManifest::operator++(): stripe_ofs=" << stripe_ofs
<< " part_ofs=" << part_ofs
<< " rule->part_size=" << rule
->part_size
<< dendl
;
2099 if (stripe_ofs
>= part_ofs
+ rule
->part_size
) {
2100 /* moved to the next part */
2102 part_ofs
+= rule
->part_size
;
2103 stripe_ofs
= part_ofs
;
2105 bool last_rule
= (next_rule_iter
== manifest
->rules
.end());
2106 /* move to the next rule? */
2107 if (!last_rule
&& stripe_ofs
>= next_rule_iter
->second
.start_ofs
) {
2108 rule_iter
= next_rule_iter
;
2109 last_rule
= (next_rule_iter
== manifest
->rules
.end());
2113 cur_part_id
= rule_iter
->second
.start_part_num
;
2118 rule
= &rule_iter
->second
;
2121 stripe_size
= MIN(rule
->part_size
- (stripe_ofs
- part_ofs
), rule
->stripe_max_size
);
2124 cur_override_prefix
= rule
->override_prefix
;
2127 if (ofs
> obj_size
) {
2133 dout(20) << "RGWObjManifest::operator++(): result: ofs=" << ofs
<< " stripe_ofs=" << stripe_ofs
<< " part_ofs=" << part_ofs
<< " rule->part_size=" << rule
->part_size
<< dendl
;
2137 int RGWObjManifest::generator::create_begin(CephContext
*cct
, RGWObjManifest
*_m
, const string
& placement_rule
, rgw_bucket
& _b
, rgw_obj
& _obj
)
2141 manifest
->set_tail_placement(placement_rule
, _b
);
2142 manifest
->set_head(placement_rule
, _obj
, 0);
2145 if (manifest
->get_prefix().empty()) {
2147 gen_rand_alphanumeric(cct
, buf
, sizeof(buf
) - 1);
2149 string oid_prefix
= ".";
2150 oid_prefix
.append(buf
);
2151 oid_prefix
.append("_");
2153 manifest
->set_prefix(oid_prefix
);
2156 bool found
= manifest
->get_rule(0, &rule
);
2158 derr
<< "ERROR: manifest->get_rule() could not find rule" << dendl
;
2162 uint64_t head_size
= manifest
->get_head_size();
2164 if (head_size
> 0) {
2165 cur_stripe_size
= head_size
;
2167 cur_stripe_size
= rule
.stripe_max_size
;
2170 cur_part_id
= rule
.start_part_num
;
2172 manifest
->get_implicit_location(cur_part_id
, cur_stripe
, 0, NULL
, &cur_obj
);
2174 // Normal object which not generated through copy operation
2175 manifest
->set_tail_instance(_obj
.key
.instance
);
2177 manifest
->update_iterators();
2182 int RGWObjManifest::generator::create_next(uint64_t ofs
)
2184 if (ofs
< last_ofs
) /* only going forward */
2187 uint64_t max_head_size
= manifest
->get_max_head_size();
2189 if (ofs
< max_head_size
) {
2190 manifest
->set_head_size(ofs
);
2193 if (ofs
>= max_head_size
) {
2194 manifest
->set_head_size(max_head_size
);
2195 cur_stripe
= (ofs
- max_head_size
) / rule
.stripe_max_size
;
2196 cur_stripe_size
= rule
.stripe_max_size
;
2198 if (cur_part_id
== 0 && max_head_size
> 0) {
2204 manifest
->set_obj_size(ofs
);
2206 manifest
->get_implicit_location(cur_part_id
, cur_stripe
, ofs
, NULL
, &cur_obj
);
2208 manifest
->update_iterators();
2213 const RGWObjManifest::obj_iterator
& RGWObjManifest::obj_begin()
2218 const RGWObjManifest::obj_iterator
& RGWObjManifest::obj_end()
2223 RGWObjManifest::obj_iterator
RGWObjManifest::obj_find(uint64_t ofs
)
2225 if (ofs
> obj_size
) {
2228 RGWObjManifest::obj_iterator
iter(this);
2233 int RGWObjManifest::append(RGWObjManifest
& m
, RGWZoneGroup
& zonegroup
, RGWZoneParams
& zone_params
)
2235 if (explicit_objs
|| m
.explicit_objs
) {
2236 return append_explicit(m
, zonegroup
, zone_params
);
2239 if (rules
.empty()) {
2244 string override_prefix
;
2246 if (prefix
.empty()) {
2250 if (prefix
!= m
.prefix
) {
2251 override_prefix
= m
.prefix
;
2254 map
<uint64_t, RGWObjManifestRule
>::iterator miter
= m
.rules
.begin();
2255 if (miter
== m
.rules
.end()) {
2256 return append_explicit(m
, zonegroup
, zone_params
);
2259 for (; miter
!= m
.rules
.end(); ++miter
) {
2260 map
<uint64_t, RGWObjManifestRule
>::reverse_iterator last_rule
= rules
.rbegin();
2262 RGWObjManifestRule
& rule
= last_rule
->second
;
2264 if (rule
.part_size
== 0) {
2265 rule
.part_size
= obj_size
- rule
.start_ofs
;
2268 RGWObjManifestRule
& next_rule
= miter
->second
;
2269 if (!next_rule
.part_size
) {
2270 next_rule
.part_size
= m
.obj_size
- next_rule
.start_ofs
;
2273 string rule_prefix
= prefix
;
2274 if (!rule
.override_prefix
.empty()) {
2275 rule_prefix
= rule
.override_prefix
;
2278 string next_rule_prefix
= m
.prefix
;
2279 if (!next_rule
.override_prefix
.empty()) {
2280 next_rule_prefix
= next_rule
.override_prefix
;
2283 if (rule
.part_size
!= next_rule
.part_size
||
2284 rule
.stripe_max_size
!= next_rule
.stripe_max_size
||
2285 rule_prefix
!= next_rule_prefix
) {
2286 if (next_rule_prefix
!= prefix
) {
2287 append_rules(m
, miter
, &next_rule_prefix
);
2289 append_rules(m
, miter
, NULL
);
2294 uint64_t expected_part_num
= rule
.start_part_num
+ 1;
2295 if (rule
.part_size
> 0) {
2296 expected_part_num
= rule
.start_part_num
+ (obj_size
+ next_rule
.start_ofs
- rule
.start_ofs
) / rule
.part_size
;
2299 if (expected_part_num
!= next_rule
.start_part_num
) {
2300 append_rules(m
, miter
, NULL
);
2305 set_obj_size(obj_size
+ m
.obj_size
);
2310 int RGWObjManifest::append(RGWObjManifest
& m
, RGWRados
*store
)
2312 return append(m
, store
->get_zonegroup(), store
->get_zone_params());
2315 void RGWObjManifest::append_rules(RGWObjManifest
& m
, map
<uint64_t, RGWObjManifestRule
>::iterator
& miter
,
2316 string
*override_prefix
)
2318 for (; miter
!= m
.rules
.end(); ++miter
) {
2319 RGWObjManifestRule rule
= miter
->second
;
2320 rule
.start_ofs
+= obj_size
;
2321 if (override_prefix
)
2322 rule
.override_prefix
= *override_prefix
;
2323 rules
[rule
.start_ofs
] = rule
;
2327 void RGWObjManifest::convert_to_explicit(const RGWZoneGroup
& zonegroup
, const RGWZoneParams
& zone_params
)
2329 if (explicit_objs
) {
2332 obj_iterator iter
= obj_begin();
2334 while (iter
!= obj_end()) {
2335 RGWObjManifestPart
& part
= objs
[iter
.get_stripe_ofs()];
2336 const rgw_obj_select
& os
= iter
.get_location();
2337 const rgw_raw_obj
& raw_loc
= os
.get_raw_obj(zonegroup
, zone_params
);
2340 uint64_t ofs
= iter
.get_stripe_ofs();
2345 rgw_raw_obj_to_obj(tail_placement
.bucket
, raw_loc
, &part
.loc
);
2348 uint64_t next_ofs
= iter
.get_stripe_ofs();
2350 part
.size
= next_ofs
- ofs
;
2353 explicit_objs
= true;
2358 int RGWObjManifest::append_explicit(RGWObjManifest
& m
, const RGWZoneGroup
& zonegroup
, const RGWZoneParams
& zone_params
)
2360 if (!explicit_objs
) {
2361 convert_to_explicit(zonegroup
, zone_params
);
2363 if (!m
.explicit_objs
) {
2364 m
.convert_to_explicit(zonegroup
, zone_params
);
2366 map
<uint64_t, RGWObjManifestPart
>::iterator iter
;
2367 uint64_t base
= obj_size
;
2368 for (iter
= m
.objs
.begin(); iter
!= m
.objs
.end(); ++iter
) {
2369 RGWObjManifestPart
& part
= iter
->second
;
2370 objs
[base
+ iter
->first
] = part
;
2372 obj_size
+= m
.obj_size
;
2377 bool RGWObjManifest::get_rule(uint64_t ofs
, RGWObjManifestRule
*rule
)
2379 if (rules
.empty()) {
2383 map
<uint64_t, RGWObjManifestRule
>::iterator iter
= rules
.upper_bound(ofs
);
2384 if (iter
!= rules
.begin()) {
2388 *rule
= iter
->second
;
2393 void RGWObjVersionTracker::generate_new_write_ver(CephContext
*cct
)
2395 write_version
.ver
= 1;
2398 write_version
.tag
.clear();
2399 append_rand_alpha(cct
, write_version
.tag
, write_version
.tag
, TAG_LEN
);
2402 int RGWPutObjProcessor::complete(size_t accounted_size
, const string
& etag
,
2403 real_time
*mtime
, real_time set_mtime
,
2404 map
<string
, bufferlist
>& attrs
, real_time delete_at
,
2405 const char *if_match
, const char *if_nomatch
, const string
*user_data
,
2406 rgw_zone_set
*zones_trace
)
2408 int r
= do_complete(accounted_size
, etag
, mtime
, set_mtime
, attrs
, delete_at
, if_match
, if_nomatch
, user_data
, zones_trace
);
2412 is_complete
= !canceled
;
2416 CephContext
*RGWPutObjProcessor::ctx()
2418 return store
->ctx();
2421 RGWPutObjProcessor_Aio::~RGWPutObjProcessor_Aio()
2428 set
<rgw_raw_obj
>::iterator iter
;
2429 bool need_to_remove_head
= false;
2430 rgw_raw_obj raw_head
;
2432 if (!head_obj
.empty()) {
2433 store
->obj_to_raw(bucket_info
.placement_rule
, head_obj
, &raw_head
);
2437 * We should delete the object in the "multipart" namespace to avoid race condition.
2438 * Such race condition is caused by the fact that the multipart object is the gatekeeper of a multipart
2439 * upload, when it is deleted, a second upload would start with the same suffix("2/"), therefore, objects
2440 * written by the second upload may be deleted by the first upload.
2441 * details is describled on #11749
2443 * The above comment still stands, but instead of searching for a specific object in the multipart
2444 * namespace, we just make sure that we remove the object that is marked as the head object after
2445 * we remove all the other raw objects. Note that we use different call to remove the head object,
2446 * as this one needs to go via the bucket index prepare/complete 2-phase commit scheme.
2448 for (iter
= written_objs
.begin(); iter
!= written_objs
.end(); ++iter
) {
2449 const rgw_raw_obj
& obj
= *iter
;
2450 if (!head_obj
.empty() && obj
== raw_head
) {
2451 ldout(store
->ctx(), 5) << "NOTE: we should not process the head object (" << obj
<< ") here" << dendl
;
2452 need_to_remove_head
= true;
2456 int r
= store
->delete_raw_obj(obj
);
2457 if (r
< 0 && r
!= -ENOENT
) {
2458 ldout(store
->ctx(), 5) << "WARNING: failed to remove obj (" << obj
<< "), leaked" << dendl
;
2462 if (need_to_remove_head
) {
2463 ldout(store
->ctx(), 5) << "NOTE: we are going to process the head obj (" << raw_head
<< ")" << dendl
;
2464 int r
= store
->delete_obj(obj_ctx
, bucket_info
, head_obj
, 0, 0);
2465 if (r
< 0 && r
!= -ENOENT
) {
2466 ldout(store
->ctx(), 0) << "WARNING: failed to remove obj (" << raw_head
<< "), leaked" << dendl
;
2471 int RGWPutObjProcessor_Aio::handle_obj_data(rgw_raw_obj
& obj
, bufferlist
& bl
, off_t ofs
, off_t abs_ofs
, void **phandle
, bool exclusive
)
2473 if ((uint64_t)abs_ofs
+ bl
.length() > obj_len
)
2474 obj_len
= abs_ofs
+ bl
.length();
2476 if (!(obj
== last_written_obj
)) {
2477 last_written_obj
= obj
;
2480 // For the first call pass -1 as the offset to
2482 return store
->aio_put_obj_data(NULL
, obj
, bl
, ((ofs
!= 0) ? ofs
: -1), exclusive
, phandle
);
2485 struct put_obj_aio_info
RGWPutObjProcessor_Aio::pop_pending()
2487 struct put_obj_aio_info info
;
2488 info
= pending
.front();
2489 pending
.pop_front();
2490 pending_size
-= info
.size
;
2494 int RGWPutObjProcessor_Aio::wait_pending_front()
2496 if (pending
.empty()) {
2499 struct put_obj_aio_info info
= pop_pending();
2500 int ret
= store
->aio_wait(info
.handle
);
2503 add_written_obj(info
.obj
);
2509 bool RGWPutObjProcessor_Aio::pending_has_completed()
2511 if (pending
.empty())
2514 struct put_obj_aio_info
& info
= pending
.front();
2515 return store
->aio_completed(info
.handle
);
2518 int RGWPutObjProcessor_Aio::drain_pending()
2521 while (!pending
.empty()) {
2522 int r
= wait_pending_front();
2529 int RGWPutObjProcessor_Aio::throttle_data(void *handle
, const rgw_raw_obj
& obj
, uint64_t size
, bool need_to_wait
)
2531 bool _wait
= need_to_wait
;
2534 struct put_obj_aio_info info
;
2535 info
.handle
= handle
;
2538 pending_size
+= size
;
2539 pending
.push_back(info
);
2541 size_t orig_size
= pending_size
;
2543 /* first drain complete IOs */
2544 while (pending_has_completed()) {
2545 int r
= wait_pending_front();
2552 /* resize window in case messages are draining too fast */
2553 if (orig_size
- pending_size
>= window_size
) {
2554 window_size
+= store
->ctx()->_conf
->rgw_max_chunk_size
;
2555 uint64_t max_window_size
= store
->ctx()->_conf
->rgw_put_obj_max_window_size
;
2556 if (window_size
> max_window_size
) {
2557 window_size
= max_window_size
;
2561 /* now throttle. Note that need_to_wait should only affect the first IO operation */
2562 if (pending_size
> window_size
|| _wait
) {
2563 int r
= wait_pending_front();
2570 int RGWPutObjProcessor_Atomic::write_data(bufferlist
& bl
, off_t ofs
, void **phandle
, rgw_raw_obj
*pobj
, bool exclusive
)
2572 if (ofs
>= next_part_ofs
) {
2573 int r
= prepare_next_part(ofs
);
2586 return RGWPutObjProcessor_Aio::handle_obj_data(cur_obj
, bl
, ofs
- cur_part_ofs
, ofs
, phandle
, exclusive
);
2589 int RGWPutObjProcessor_Aio::prepare(RGWRados
*store
, string
*oid_rand
)
2591 RGWPutObjProcessor::prepare(store
, oid_rand
);
2593 window_size
= store
->ctx()->_conf
->rgw_put_obj_min_window_size
;
2598 int RGWPutObjProcessor_Atomic::handle_data(bufferlist
& bl
, off_t ofs
, void **phandle
, rgw_raw_obj
*pobj
, bool *again
)
2601 uint64_t max_write_size
= MIN(max_chunk_size
, (uint64_t)next_part_ofs
- data_ofs
);
2603 pending_data_bl
.claim_append(bl
);
2604 if (pending_data_bl
.length() < max_write_size
) {
2609 pending_data_bl
.splice(0, max_write_size
, &bl
);
2611 /* do we have enough data pending accumulated that needs to be written? */
2612 *again
= (pending_data_bl
.length() >= max_chunk_size
);
2614 if (!data_ofs
&& !immutable_head()) {
2615 first_chunk
.claim(bl
);
2616 obj_len
= (uint64_t)first_chunk
.length();
2617 int r
= prepare_next_part(obj_len
);
2624 off_t write_ofs
= data_ofs
;
2625 data_ofs
= write_ofs
+ bl
.length();
2626 bool exclusive
= (!write_ofs
&& immutable_head()); /* immutable head object, need to verify nothing exists there
2627 we could be racing with another upload, to the same
2628 object and cleanup can be messy */
2629 int ret
= write_data(bl
, write_ofs
, phandle
, pobj
, exclusive
);
2630 if (ret
>= 0) { /* we might return, need to clear bl as it was already sent */
2637 int RGWPutObjProcessor_Atomic::prepare_init(RGWRados
*store
, string
*oid_rand
)
2639 RGWPutObjProcessor_Aio::prepare(store
, oid_rand
);
2641 int r
= store
->get_max_chunk_size(bucket_info
.placement_rule
, head_obj
, &max_chunk_size
);
2649 int RGWPutObjProcessor_Atomic::prepare(RGWRados
*store
, string
*oid_rand
)
2651 head_obj
.init(bucket
, obj_str
);
2653 int r
= prepare_init(store
, oid_rand
);
2658 if (!version_id
.empty()) {
2659 head_obj
.key
.set_instance(version_id
);
2660 } else if (versioned_object
) {
2661 store
->gen_rand_obj_instance_name(&head_obj
);
2664 manifest
.set_trivial_rule(max_chunk_size
, store
->ctx()->_conf
->rgw_obj_stripe_size
);
2666 r
= manifest_gen
.create_begin(store
->ctx(), &manifest
, bucket_info
.placement_rule
, head_obj
.bucket
, head_obj
);
2674 int RGWPutObjProcessor_Atomic::prepare_next_part(off_t ofs
) {
2676 int ret
= manifest_gen
.create_next(ofs
);
2678 lderr(store
->ctx()) << "ERROR: manifest_gen.create_next() returned ret=" << ret
<< dendl
;
2682 next_part_ofs
= ofs
+ manifest_gen
.cur_stripe_max_size();
2683 cur_obj
= manifest_gen
.get_cur_obj(store
);
2688 int RGWPutObjProcessor_Atomic::complete_parts()
2690 if (obj_len
> (uint64_t)cur_part_ofs
) {
2691 return prepare_next_part(obj_len
);
2696 int RGWPutObjProcessor_Atomic::complete_writing_data()
2698 if (!data_ofs
&& !immutable_head()) {
2699 /* only claim if pending_data_bl() is not empty. This is needed because we might be called twice
2700 * (e.g., when a retry due to race happens). So a second call to first_chunk.claim() would
2701 * clobber first_chunk
2703 if (pending_data_bl
.length() > 0) {
2704 first_chunk
.claim(pending_data_bl
);
2706 obj_len
= (uint64_t)first_chunk
.length();
2708 while (pending_data_bl
.length()) {
2709 void *handle
= nullptr;
2711 uint64_t max_write_size
= MIN(max_chunk_size
, (uint64_t)next_part_ofs
- data_ofs
);
2712 if (max_write_size
> pending_data_bl
.length()) {
2713 max_write_size
= pending_data_bl
.length();
2716 pending_data_bl
.splice(0, max_write_size
, &bl
);
2717 uint64_t write_len
= bl
.length();
2718 int r
= write_data(bl
, data_ofs
, &handle
, &obj
, false);
2720 ldout(store
->ctx(), 0) << "ERROR: write_data() returned " << r
<< dendl
;
2723 data_ofs
+= write_len
;
2724 r
= throttle_data(handle
, obj
, write_len
, false);
2726 ldout(store
->ctx(), 0) << "ERROR: throttle_data() returned " << r
<< dendl
;
2730 if (data_ofs
>= next_part_ofs
) {
2731 r
= prepare_next_part(data_ofs
);
2733 ldout(store
->ctx(), 0) << "ERROR: prepare_next_part() returned " << r
<< dendl
;
2738 int r
= complete_parts();
2743 r
= drain_pending();
2750 int RGWPutObjProcessor_Atomic::do_complete(size_t accounted_size
, const string
& etag
,
2751 real_time
*mtime
, real_time set_mtime
,
2752 map
<string
, bufferlist
>& attrs
,
2753 real_time delete_at
,
2754 const char *if_match
,
2755 const char *if_nomatch
, const string
*user_data
,
2756 rgw_zone_set
*zones_trace
) {
2757 int r
= complete_writing_data();
2761 obj_ctx
.obj
.set_atomic(head_obj
);
2763 RGWRados::Object
op_target(store
, bucket_info
, obj_ctx
, head_obj
);
2765 /* some object types shouldn't be versioned, e.g., multipart parts */
2766 op_target
.set_versioning_disabled(!versioned_object
);
2768 RGWRados::Object::Write
obj_op(&op_target
);
2770 obj_op
.meta
.data
= &first_chunk
;
2771 obj_op
.meta
.manifest
= &manifest
;
2772 obj_op
.meta
.ptag
= &unique_tag
; /* use req_id as operation tag */
2773 obj_op
.meta
.if_match
= if_match
;
2774 obj_op
.meta
.if_nomatch
= if_nomatch
;
2775 obj_op
.meta
.mtime
= mtime
;
2776 obj_op
.meta
.set_mtime
= set_mtime
;
2777 obj_op
.meta
.owner
= bucket_info
.owner
;
2778 obj_op
.meta
.flags
= PUT_OBJ_CREATE
;
2779 obj_op
.meta
.olh_epoch
= olh_epoch
;
2780 obj_op
.meta
.delete_at
= delete_at
;
2781 obj_op
.meta
.user_data
= user_data
;
2782 obj_op
.meta
.zones_trace
= zones_trace
;
2783 obj_op
.meta
.modify_tail
= true;
2785 r
= obj_op
.write_meta(obj_len
, accounted_size
, attrs
);
2790 canceled
= obj_op
.meta
.canceled
;
2795 int RGWRados::watch(const string
& oid
, uint64_t *watch_handle
, librados::WatchCtx2
*ctx
) {
2796 int r
= control_pool_ctx
.watch2(oid
, watch_handle
, ctx
);
2802 int RGWRados::unwatch(uint64_t watch_handle
)
2804 int r
= control_pool_ctx
.unwatch2(watch_handle
);
2806 ldout(cct
, 0) << "ERROR: rados->unwatch2() returned r=" << r
<< dendl
;
2809 r
= rados
[0].watch_flush();
2811 ldout(cct
, 0) << "ERROR: rados->watch_flush() returned r=" << r
<< dendl
;
2817 void RGWRados::add_watcher(int i
)
2819 ldout(cct
, 20) << "add_watcher() i=" << i
<< dendl
;
2820 Mutex::Locker
l(watchers_lock
);
2821 watchers_set
.insert(i
);
2822 if (watchers_set
.size() == (size_t)num_watchers
) {
2823 ldout(cct
, 2) << "all " << num_watchers
<< " watchers are set, enabling cache" << dendl
;
2824 set_cache_enabled(true);
2828 void RGWRados::remove_watcher(int i
)
2830 ldout(cct
, 20) << "remove_watcher() i=" << i
<< dendl
;
2831 Mutex::Locker
l(watchers_lock
);
2832 size_t orig_size
= watchers_set
.size();
2833 watchers_set
.erase(i
);
2834 if (orig_size
== (size_t)num_watchers
&&
2835 watchers_set
.size() < orig_size
) { /* actually removed */
2836 ldout(cct
, 2) << "removed watcher, disabling cache" << dendl
;
2837 set_cache_enabled(false);
2841 class RGWWatcher
: public librados::WatchCtx2
{
2845 uint64_t watch_handle
;
2847 class C_ReinitWatch
: public Context
{
2848 RGWWatcher
*watcher
;
2850 explicit C_ReinitWatch(RGWWatcher
*_watcher
) : watcher(_watcher
) {}
2851 void finish(int r
) override
{
2856 RGWWatcher(RGWRados
*r
, int i
, const string
& o
) : rados(r
), index(i
), oid(o
), watch_handle(0) {}
2857 void handle_notify(uint64_t notify_id
,
2859 uint64_t notifier_id
,
2860 bufferlist
& bl
) override
{
2861 ldout(rados
->ctx(), 10) << "RGWWatcher::handle_notify() "
2862 << " notify_id " << notify_id
2863 << " cookie " << cookie
2864 << " notifier " << notifier_id
2865 << " bl.length()=" << bl
.length() << dendl
;
2866 rados
->watch_cb(notify_id
, cookie
, notifier_id
, bl
);
2868 bufferlist reply_bl
; // empty reply payload
2869 rados
->control_pool_ctx
.notify_ack(oid
, notify_id
, cookie
, reply_bl
);
2871 void handle_error(uint64_t cookie
, int err
) override
{
2872 lderr(rados
->ctx()) << "RGWWatcher::handle_error cookie " << cookie
2873 << " err " << cpp_strerror(err
) << dendl
;
2874 rados
->remove_watcher(index
);
2875 rados
->schedule_context(new C_ReinitWatch(this));
2879 int ret
= unregister_watch();
2881 ldout(rados
->ctx(), 0) << "ERROR: unregister_watch() returned ret=" << ret
<< dendl
;
2884 ret
= register_watch();
2886 ldout(rados
->ctx(), 0) << "ERROR: register_watch() returned ret=" << ret
<< dendl
;
2891 int unregister_watch() {
2892 int r
= rados
->unwatch(watch_handle
);
2896 rados
->remove_watcher(index
);
2900 int register_watch() {
2901 int r
= rados
->watch(oid
, &watch_handle
, this);
2905 rados
->add_watcher(index
);
2910 class RGWMetaNotifierManager
: public RGWCoroutinesManager
{
2912 RGWHTTPManager http_manager
;
2915 RGWMetaNotifierManager(RGWRados
*_store
) : RGWCoroutinesManager(_store
->ctx(), _store
->get_cr_registry()), store(_store
),
2916 http_manager(store
->ctx(), completion_mgr
) {
2917 http_manager
.set_threaded();
2920 int notify_all(map
<string
, RGWRESTConn
*>& conn_map
, set
<int>& shards
) {
2921 rgw_http_param_pair pairs
[] = { { "type", "metadata" },
2925 list
<RGWCoroutinesStack
*> stacks
;
2926 for (map
<string
, RGWRESTConn
*>::iterator iter
= conn_map
.begin(); iter
!= conn_map
.end(); ++iter
) {
2927 RGWRESTConn
*conn
= iter
->second
;
2928 RGWCoroutinesStack
*stack
= new RGWCoroutinesStack(store
->ctx(), this);
2929 stack
->call(new RGWPostRESTResourceCR
<set
<int>, int>(store
->ctx(), conn
, &http_manager
, "/admin/log", pairs
, shards
, NULL
));
2931 stacks
.push_back(stack
);
2937 class RGWDataNotifierManager
: public RGWCoroutinesManager
{
2939 RGWHTTPManager http_manager
;
2942 RGWDataNotifierManager(RGWRados
*_store
) : RGWCoroutinesManager(_store
->ctx(), _store
->get_cr_registry()), store(_store
),
2943 http_manager(store
->ctx(), completion_mgr
) {
2944 http_manager
.set_threaded();
2947 int notify_all(map
<string
, RGWRESTConn
*>& conn_map
, map
<int, set
<string
> >& shards
) {
2948 rgw_http_param_pair pairs
[] = { { "type", "data" },
2950 { "source-zone", store
->get_zone_params().get_id().c_str() },
2953 list
<RGWCoroutinesStack
*> stacks
;
2954 for (map
<string
, RGWRESTConn
*>::iterator iter
= conn_map
.begin(); iter
!= conn_map
.end(); ++iter
) {
2955 RGWRESTConn
*conn
= iter
->second
;
2956 RGWCoroutinesStack
*stack
= new RGWCoroutinesStack(store
->ctx(), this);
2957 stack
->call(new RGWPostRESTResourceCR
<map
<int, set
<string
> >, int>(store
->ctx(), conn
, &http_manager
, "/admin/log", pairs
, shards
, NULL
));
2959 stacks
.push_back(stack
);
2965 class RGWRadosThread
{
2966 class Worker
: public Thread
{
2968 RGWRadosThread
*processor
;
2973 Mutex::Locker
l(lock
);
2977 void wait_interval(const utime_t
& wait_time
) {
2978 Mutex::Locker
l(lock
);
2979 cond
.WaitInterval(lock
, wait_time
);
2983 Worker(CephContext
*_cct
, RGWRadosThread
*_p
) : cct(_cct
), processor(_p
), lock("RGWRadosThread::Worker") {}
2984 void *entry() override
;
2986 Mutex::Locker
l(lock
);
2997 std::atomic
<bool> down_flag
= { false };
3001 virtual uint64_t interval_msec() = 0;
3002 virtual void stop_process() {}
3004 RGWRadosThread(RGWRados
*_store
, const string
& thread_name
= "radosgw")
3005 : worker(NULL
), cct(_store
->ctx()), store(_store
), thread_name(thread_name
) {}
3006 virtual ~RGWRadosThread() {
3010 virtual int init() { return 0; }
3011 virtual int process() = 0;
3013 bool going_down() { return down_flag
; }
3025 void RGWRadosThread::start()
3027 worker
= new Worker(cct
, this);
3028 worker
->create(thread_name
.c_str());
3031 void RGWRadosThread::stop()
3043 void *RGWRadosThread::Worker::entry() {
3044 uint64_t msec
= processor
->interval_msec();
3045 utime_t interval
= utime_t(msec
/ 1000, (msec
% 1000) * 1000000);
3048 utime_t start
= ceph_clock_now();
3049 int r
= processor
->process();
3051 dout(0) << "ERROR: processor->process() returned error r=" << r
<< dendl
;
3054 if (processor
->going_down())
3057 utime_t end
= ceph_clock_now();
3060 uint64_t cur_msec
= processor
->interval_msec();
3061 if (cur_msec
!= msec
) { /* was it reconfigured? */
3063 interval
= utime_t(msec
/ 1000, (msec
% 1000) * 1000000);
3067 if (interval
<= end
)
3068 continue; // next round
3070 utime_t wait_time
= interval
;
3073 wait_interval(wait_time
);
3077 } while (!processor
->going_down());
3082 class RGWMetaNotifier
: public RGWRadosThread
{
3083 RGWMetaNotifierManager notify_mgr
;
3084 RGWMetadataLog
*const log
;
3086 uint64_t interval_msec() override
{
3087 return cct
->_conf
->rgw_md_notify_interval_msec
;
3090 RGWMetaNotifier(RGWRados
*_store
, RGWMetadataLog
* log
)
3091 : RGWRadosThread(_store
, "meta-notifier"), notify_mgr(_store
), log(log
) {}
3093 int process() override
;
3096 int RGWMetaNotifier::process()
3100 log
->read_clear_modified(shards
);
3102 if (shards
.empty()) {
3106 for (set
<int>::iterator iter
= shards
.begin(); iter
!= shards
.end(); ++iter
) {
3107 ldout(cct
, 20) << __func__
<< "(): notifying mdlog change, shard_id=" << *iter
<< dendl
;
3110 notify_mgr
.notify_all(store
->zone_conn_map
, shards
);
3115 class RGWDataNotifier
: public RGWRadosThread
{
3116 RGWDataNotifierManager notify_mgr
;
3118 uint64_t interval_msec() override
{
3119 return cct
->_conf
->get_val
<int64_t>("rgw_data_notify_interval_msec");
3122 RGWDataNotifier(RGWRados
*_store
) : RGWRadosThread(_store
, "data-notifier"), notify_mgr(_store
) {}
3124 int process() override
;
3127 int RGWDataNotifier::process()
3129 if (!store
->data_log
) {
3133 map
<int, set
<string
> > shards
;
3135 store
->data_log
->read_clear_modified(shards
);
3137 if (shards
.empty()) {
3141 for (map
<int, set
<string
> >::iterator iter
= shards
.begin(); iter
!= shards
.end(); ++iter
) {
3142 ldout(cct
, 20) << __func__
<< "(): notifying datalog change, shard_id=" << iter
->first
<< ": " << iter
->second
<< dendl
;
3145 notify_mgr
.notify_all(store
->zone_data_notify_to_map
, shards
);
3150 class RGWSyncProcessorThread
: public RGWRadosThread
{
3152 RGWSyncProcessorThread(RGWRados
*_store
, const string
& thread_name
= "radosgw") : RGWRadosThread(_store
, thread_name
) {}
3153 RGWSyncProcessorThread(RGWRados
*_store
) : RGWRadosThread(_store
) {}
3154 ~RGWSyncProcessorThread() override
{}
3155 int init() override
= 0 ;
3156 int process() override
= 0;
3159 class RGWMetaSyncProcessorThread
: public RGWSyncProcessorThread
3161 RGWMetaSyncStatusManager sync
;
3163 uint64_t interval_msec() override
{
3164 return 0; /* no interval associated, it'll run once until stopped */
3166 void stop_process() override
{
3170 RGWMetaSyncProcessorThread(RGWRados
*_store
, RGWAsyncRadosProcessor
*async_rados
)
3171 : RGWSyncProcessorThread(_store
, "meta-sync"), sync(_store
, async_rados
) {}
3173 void wakeup_sync_shards(set
<int>& shard_ids
) {
3174 for (set
<int>::iterator iter
= shard_ids
.begin(); iter
!= shard_ids
.end(); ++iter
) {
3178 RGWMetaSyncStatusManager
* get_manager() { return &sync
; }
3180 int init() override
{
3181 int ret
= sync
.init();
3183 ldout(store
->ctx(), 0) << "ERROR: sync.init() returned " << ret
<< dendl
;
3189 int process() override
{
3195 class RGWDataSyncProcessorThread
: public RGWSyncProcessorThread
3197 RGWDataSyncStatusManager sync
;
3200 uint64_t interval_msec() override
{
3202 return 0; /* no interval associated, it'll run once until stopped */
3204 #define DATA_SYNC_INIT_WAIT_SEC 20
3205 return DATA_SYNC_INIT_WAIT_SEC
* 1000;
3208 void stop_process() override
{
3212 RGWDataSyncProcessorThread(RGWRados
*_store
, RGWAsyncRadosProcessor
*async_rados
,
3213 const string
& _source_zone
,
3214 rgw::BucketChangeObserver
*observer
)
3215 : RGWSyncProcessorThread(_store
, "data-sync"),
3216 sync(_store
, async_rados
, _source_zone
, observer
),
3217 initialized(false) {}
3219 void wakeup_sync_shards(map
<int, set
<string
> >& shard_ids
) {
3220 for (map
<int, set
<string
> >::iterator iter
= shard_ids
.begin(); iter
!= shard_ids
.end(); ++iter
) {
3221 sync
.wakeup(iter
->first
, iter
->second
);
3224 RGWDataSyncStatusManager
* get_manager() { return &sync
; }
3226 int init() override
{
3230 int process() override
{
3231 while (!initialized
) {
3235 int ret
= sync
.init();
3240 /* we'll be back! */
3248 class RGWSyncLogTrimThread
: public RGWSyncProcessorThread
3250 RGWCoroutinesManager crs
;
3252 rgw::BucketTrimManager
*bucket_trim
;
3253 RGWHTTPManager http
;
3254 const utime_t trim_interval
;
3256 uint64_t interval_msec() override
{ return 0; }
3257 void stop_process() override
{ crs
.stop(); }
3259 RGWSyncLogTrimThread(RGWRados
*store
, rgw::BucketTrimManager
*bucket_trim
,
3261 : RGWSyncProcessorThread(store
, "sync-log-trim"),
3262 crs(store
->ctx(), store
->get_cr_registry()), store(store
),
3263 bucket_trim(bucket_trim
),
3264 http(store
->ctx(), crs
.get_completion_mgr()),
3265 trim_interval(interval
, 0)
3268 int init() override
{
3269 return http
.set_threaded();
3271 int process() override
{
3272 list
<RGWCoroutinesStack
*> stacks
;
3273 auto meta
= new RGWCoroutinesStack(store
->ctx(), &crs
);
3274 meta
->call(create_meta_log_trim_cr(store
, &http
,
3275 cct
->_conf
->rgw_md_log_max_shards
,
3277 stacks
.push_back(meta
);
3279 auto data
= new RGWCoroutinesStack(store
->ctx(), &crs
);
3280 data
->call(create_data_log_trim_cr(store
, &http
,
3281 cct
->_conf
->rgw_data_log_num_shards
,
3283 stacks
.push_back(data
);
3285 auto bucket
= new RGWCoroutinesStack(store
->ctx(), &crs
);
3286 bucket
->call(bucket_trim
->create_bucket_trim_cr(&http
));
3287 stacks
.push_back(bucket
);
3294 void RGWRados::wakeup_meta_sync_shards(set
<int>& shard_ids
)
3296 Mutex::Locker
l(meta_sync_thread_lock
);
3297 if (meta_sync_processor_thread
) {
3298 meta_sync_processor_thread
->wakeup_sync_shards(shard_ids
);
3302 void RGWRados::wakeup_data_sync_shards(const string
& source_zone
, map
<int, set
<string
> >& shard_ids
)
3304 ldout(ctx(), 20) << __func__
<< ": source_zone=" << source_zone
<< ", shard_ids=" << shard_ids
<< dendl
;
3305 Mutex::Locker
l(data_sync_thread_lock
);
3306 map
<string
, RGWDataSyncProcessorThread
*>::iterator iter
= data_sync_processor_threads
.find(source_zone
);
3307 if (iter
== data_sync_processor_threads
.end()) {
3308 ldout(ctx(), 10) << __func__
<< ": couldn't find sync thread for zone " << source_zone
<< ", skipping async data sync processing" << dendl
;
3312 RGWDataSyncProcessorThread
*thread
= iter
->second
;
3314 thread
->wakeup_sync_shards(shard_ids
);
3317 RGWMetaSyncStatusManager
* RGWRados::get_meta_sync_manager()
3319 Mutex::Locker
l(meta_sync_thread_lock
);
3320 if (meta_sync_processor_thread
) {
3321 return meta_sync_processor_thread
->get_manager();
3326 RGWDataSyncStatusManager
* RGWRados::get_data_sync_manager(const std::string
& source_zone
)
3328 Mutex::Locker
l(data_sync_thread_lock
);
3329 auto thread
= data_sync_processor_threads
.find(source_zone
);
3330 if (thread
== data_sync_processor_threads
.end()) {
3333 return thread
->second
->get_manager();
3336 int RGWRados::get_required_alignment(const rgw_pool
& pool
, uint64_t *alignment
)
3339 int r
= open_pool_ctx(pool
, ioctx
);
3341 ldout(cct
, 0) << "ERROR: open_pool_ctx() returned " << r
<< dendl
;
3346 r
= ioctx
.pool_requires_alignment2(&requires
);
3348 ldout(cct
, 0) << "ERROR: ioctx.pool_requires_alignment2() returned "
3359 r
= ioctx
.pool_required_alignment2(&align
);
3361 ldout(cct
, 0) << "ERROR: ioctx.pool_required_alignment2() returned "
3366 ldout(cct
, 20) << "required alignment=" << align
<< dendl
;
3372 int RGWRados::get_max_chunk_size(const rgw_pool
& pool
, uint64_t *max_chunk_size
)
3374 uint64_t alignment
= 0;
3375 int r
= get_required_alignment(pool
, &alignment
);
3380 uint64_t config_chunk_size
= cct
->_conf
->rgw_max_chunk_size
;
3382 if (alignment
== 0) {
3383 *max_chunk_size
= config_chunk_size
;
3387 if (config_chunk_size
<= alignment
) {
3388 *max_chunk_size
= alignment
;
3392 *max_chunk_size
= config_chunk_size
- (config_chunk_size
% alignment
);
3394 ldout(cct
, 20) << "max_chunk_size=" << *max_chunk_size
<< dendl
;
3399 int RGWRados::get_max_chunk_size(const string
& placement_rule
, const rgw_obj
& obj
, uint64_t *max_chunk_size
)
3402 if (!get_obj_data_pool(placement_rule
, obj
, &pool
)) {
3403 ldout(cct
, 0) << "ERROR: failed to get data pool for object " << obj
<< dendl
;
3406 return get_max_chunk_size(pool
, max_chunk_size
);
3409 class RGWIndexCompletionManager
;
3411 struct complete_op_data
{
3412 Mutex lock
{"complete_op_data"};
3413 AioCompletion
*rados_completion
{nullptr};
3414 int manager_shard_id
{-1};
3415 RGWIndexCompletionManager
*manager
{nullptr};
3419 rgw_bucket_entry_ver ver
;
3420 cls_rgw_obj_key key
;
3421 rgw_bucket_dir_entry_meta dir_meta
;
3422 list
<cls_rgw_obj_key
> remove_objs
;
3425 rgw_zone_set zones_trace
;
3427 bool stopped
{false};
3430 Mutex::Locker
l(lock
);
3435 class RGWIndexCompletionThread
: public RGWRadosThread
{
3438 uint64_t interval_msec() override
{
3442 list
<complete_op_data
*> completions
;
3444 Mutex completions_lock
;
3446 RGWIndexCompletionThread(RGWRados
*_store
)
3447 : RGWRadosThread(_store
, "index-complete"), store(_store
), completions_lock("RGWIndexCompletionThread::completions_lock") {}
3449 int process() override
;
3451 void add_completion(complete_op_data
*completion
) {
3453 Mutex::Locker
l(completions_lock
);
3454 completions
.push_back(completion
);
3461 int RGWIndexCompletionThread::process()
3463 list
<complete_op_data
*> comps
;
3466 Mutex::Locker
l(completions_lock
);
3467 completions
.swap(comps
);
3470 for (auto c
: comps
) {
3471 std::unique_ptr
<complete_op_data
> up
{c
};
3476 ldout(store
->ctx(), 20) << __func__
<< "(): handling completion for key=" << c
->key
<< dendl
;
3478 RGWRados::BucketShard
bs(store
);
3480 int r
= bs
.init(c
->obj
.bucket
, c
->obj
);
3482 ldout(cct
, 0) << "ERROR: " << __func__
<< "(): failed to initialize BucketShard, obj=" << c
->obj
<< " r=" << r
<< dendl
;
3483 /* not much to do */
3487 r
= store
->guard_reshard(&bs
, c
->obj
, [&](RGWRados::BucketShard
*bs
) -> int {
3488 librados::ObjectWriteOperation o
;
3489 cls_rgw_guard_bucket_resharding(o
, -ERR_BUSY_RESHARDING
);
3490 cls_rgw_bucket_complete_op(o
, c
->op
, c
->tag
, c
->ver
, c
->key
, c
->dir_meta
, &c
->remove_objs
,
3491 c
->log_op
, c
->bilog_op
, &c
->zones_trace
);
3493 return bs
->index_ctx
.operate(bs
->bucket_obj
, &o
);
3496 ldout(cct
, 0) << "ERROR: " << __func__
<< "(): bucket index completion failed, obj=" << c
->obj
<< " r=" << r
<< dendl
;
3497 /* ignoring error, can't do anything about it */
3500 r
= store
->data_log
->add_entry(bs
.bucket
, bs
.shard_id
);
3502 lderr(store
->ctx()) << "ERROR: failed writing data log" << dendl
;
3509 class RGWIndexCompletionManager
{
3510 RGWRados
*store
{nullptr};
3511 vector
<Mutex
*> locks
;
3512 vector
<set
<complete_op_data
*> > completions
;
3514 RGWIndexCompletionThread
*completion_thread
{nullptr};
3518 std::atomic
<int> cur_shard
{0};
3522 RGWIndexCompletionManager(RGWRados
*_store
) : store(_store
) {
3523 num_shards
= store
->ctx()->_conf
->rgw_thread_pool_size
;
3525 for (int i
= 0; i
< num_shards
; i
++) {
3527 snprintf(buf
, sizeof(buf
), "RGWIndexCompletionManager::lock::%d", i
);
3528 locks
.push_back(new Mutex(buf
));
3531 completions
.resize(num_shards
);
3533 ~RGWIndexCompletionManager() {
3536 for (auto l
: locks
) {
3542 int result
= cur_shard
% num_shards
;
3547 void create_completion(const rgw_obj
& obj
,
3548 RGWModifyOp op
, string
& tag
,
3549 rgw_bucket_entry_ver
& ver
,
3550 const cls_rgw_obj_key
& key
,
3551 rgw_bucket_dir_entry_meta
& dir_meta
,
3552 list
<cls_rgw_obj_key
> *remove_objs
, bool log_op
,
3554 rgw_zone_set
*zones_trace
,
3555 complete_op_data
**result
);
3556 bool handle_completion(completion_t cb
, complete_op_data
*arg
);
3559 completion_thread
= new RGWIndexCompletionThread(store
);
3560 int ret
= completion_thread
->init();
3564 completion_thread
->start();
3568 if (completion_thread
) {
3569 completion_thread
->stop();
3570 delete completion_thread
;
3573 for (int i
= 0; i
< num_shards
; ++i
) {
3574 Mutex::Locker
l(*locks
[i
]);
3575 for (auto c
: completions
[i
]) {
3576 Mutex::Locker
cl(c
->lock
);
3580 completions
.clear();
3584 static void obj_complete_cb(completion_t cb
, void *arg
)
3586 complete_op_data
*completion
= (complete_op_data
*)arg
;
3587 completion
->lock
.Lock();
3588 if (completion
->stopped
) {
3589 completion
->lock
.Unlock(); /* can drop lock, no one else is referencing us */
3593 bool need_delete
= completion
->manager
->handle_completion(cb
, completion
);
3594 completion
->lock
.Unlock();
3601 void RGWIndexCompletionManager::create_completion(const rgw_obj
& obj
,
3602 RGWModifyOp op
, string
& tag
,
3603 rgw_bucket_entry_ver
& ver
,
3604 const cls_rgw_obj_key
& key
,
3605 rgw_bucket_dir_entry_meta
& dir_meta
,
3606 list
<cls_rgw_obj_key
> *remove_objs
, bool log_op
,
3608 rgw_zone_set
*zones_trace
,
3609 complete_op_data
**result
)
3611 complete_op_data
*entry
= new complete_op_data
;
3613 int shard_id
= next_shard();
3615 entry
->manager_shard_id
= shard_id
;
3616 entry
->manager
= this;
3622 entry
->dir_meta
= dir_meta
;
3623 entry
->log_op
= log_op
;
3624 entry
->bilog_op
= bilog_op
;
3627 for (auto iter
= remove_objs
->begin(); iter
!= remove_objs
->end(); ++iter
) {
3628 entry
->remove_objs
.push_back(*iter
);
3633 entry
->zones_trace
= *zones_trace
;
3635 entry
->zones_trace
.insert(store
->get_zone().id
);
3640 entry
->rados_completion
= librados::Rados::aio_create_completion(entry
, NULL
, obj_complete_cb
);
3642 Mutex::Locker
l(*locks
[shard_id
]);
3643 completions
[shard_id
].insert(entry
);
3646 bool RGWIndexCompletionManager::handle_completion(completion_t cb
, complete_op_data
*arg
)
3648 int shard_id
= arg
->manager_shard_id
;
3650 Mutex::Locker
l(*locks
[shard_id
]);
3652 auto& comps
= completions
[shard_id
];
3654 auto iter
= comps
.find(arg
);
3655 if (iter
== comps
.end()) {
3662 int r
= rados_aio_get_return_value(cb
);
3663 if (r
!= -ERR_BUSY_RESHARDING
) {
3666 completion_thread
->add_completion(arg
);
3670 void RGWRados::finalize()
3672 if (run_sync_thread
) {
3673 Mutex::Locker
l(meta_sync_thread_lock
);
3674 meta_sync_processor_thread
->stop();
3676 Mutex::Locker
dl(data_sync_thread_lock
);
3677 for (auto iter
: data_sync_processor_threads
) {
3678 RGWDataSyncProcessorThread
*thread
= iter
.second
;
3681 if (sync_log_trimmer
) {
3682 sync_log_trimmer
->stop();
3686 async_rados
->stop();
3688 if (run_sync_thread
) {
3689 delete meta_sync_processor_thread
;
3690 meta_sync_processor_thread
= NULL
;
3691 Mutex::Locker
dl(data_sync_thread_lock
);
3692 for (auto iter
: data_sync_processor_threads
) {
3693 RGWDataSyncProcessorThread
*thread
= iter
.second
;
3696 data_sync_processor_threads
.clear();
3697 delete sync_log_trimmer
;
3698 sync_log_trimmer
= nullptr;
3699 bucket_trim
= boost::none
;
3704 if (need_watch_notify()) {
3708 /* delete finisher only after cleaning up watches, as watch error path might call
3709 * into finisher. We stop finisher before finalizing watch to make sure we don't
3710 * actually handle any racing work
3714 if (meta_notifier
) {
3715 meta_notifier
->stop();
3716 delete meta_notifier
;
3718 if (data_notifier
) {
3719 data_notifier
->stop();
3720 delete data_notifier
;
3736 delete rest_master_conn
;
3738 map
<string
, RGWRESTConn
*>::iterator iter
;
3739 for (iter
= zone_conn_map
.begin(); iter
!= zone_conn_map
.end(); ++iter
) {
3740 RGWRESTConn
*conn
= iter
->second
;
3744 for (iter
= zonegroup_conn_map
.begin(); iter
!= zonegroup_conn_map
.end(); ++iter
) {
3745 RGWRESTConn
*conn
= iter
->second
;
3748 RGWQuotaHandler::free_handler(quota_handler
);
3754 delete obj_tombstone_cache
;
3755 delete sync_modules_manager
;
3757 if (reshard_wait
.get()) {
3758 reshard_wait
->stop();
3759 reshard_wait
.reset();
3762 if (run_reshard_thread
) {
3763 reshard
->stop_processor();
3766 delete index_completion_manager
;
3770 * Initialize the RADOS instance and prepare to do other ops
3771 * Returns 0 on success, -ERR# on failure.
3773 int RGWRados::init_rados()
3776 auto handles
= std::vector
<librados::Rados
>{cct
->_conf
->rgw_num_rados_handles
};
3778 for (auto& r
: handles
) {
3779 ret
= r
.init_with_context(cct
);
3789 sync_modules_manager
= new RGWSyncModulesManager();
3791 rgw_register_sync_modules(sync_modules_manager
);
3793 auto crs
= std::unique_ptr
<RGWCoroutinesManagerRegistry
>{
3794 new RGWCoroutinesManagerRegistry(cct
)};
3795 ret
= crs
->hook_to_admin_command("cr dump");
3800 meta_mgr
= new RGWMetadataManager(cct
, this);
3801 data_log
= new RGWDataChangesLog(cct
, this);
3802 cr_registry
= crs
.release();
3804 std::swap(handles
, rados
);
3809 int RGWRados::register_to_service_map(const string
& daemon_type
, const map
<string
, string
>& meta
)
3811 map
<string
,string
> metadata
= meta
;
3812 metadata
["num_handles"] = stringify(rados
.size());
3813 metadata
["zonegroup_id"] = zonegroup
.get_id();
3814 metadata
["zonegroup_name"] = zonegroup
.get_name();
3815 metadata
["zone_name"] = zone_name();
3816 metadata
["zone_id"] = zone_id();;
3817 string name
= cct
->_conf
->name
.get_id();
3818 if (name
.find("rgw.") == 0) {
3819 name
= name
.substr(4);
3821 int ret
= rados
[0].service_daemon_register(daemon_type
, name
, metadata
);
3823 ldout(cct
, 0) << "ERROR: service_daemon_register() returned ret=" << ret
<< ": " << cpp_strerror(-ret
) << dendl
;
3831 * Add new connection to connections map
3832 * @param zonegroup_conn_map map which new connection will be added to
3833 * @param zonegroup zonegroup which new connection will connect to
3834 * @param new_connection pointer to new connection instance
3836 static void add_new_connection_to_map(map
<string
, RGWRESTConn
*> &zonegroup_conn_map
,
3837 const RGWZoneGroup
&zonegroup
, RGWRESTConn
*new_connection
)
3839 // Delete if connection is already exists
3840 map
<string
, RGWRESTConn
*>::iterator iterZoneGroup
= zonegroup_conn_map
.find(zonegroup
.get_id());
3841 if (iterZoneGroup
!= zonegroup_conn_map
.end()) {
3842 delete iterZoneGroup
->second
;
3845 // Add new connection to connections map
3846 zonegroup_conn_map
[zonegroup
.get_id()] = new_connection
;
3849 int RGWRados::convert_regionmap()
3851 RGWZoneGroupMap zonegroupmap
;
3853 string pool_name
= cct
->_conf
->rgw_zone_root_pool
;
3854 if (pool_name
.empty()) {
3855 pool_name
= RGW_DEFAULT_ZONE_ROOT_POOL
;
3857 string oid
= region_map_oid
;
3859 rgw_pool
pool(pool_name
);
3861 RGWObjectCtx
obj_ctx(this);
3862 int ret
= rgw_get_system_obj(this, obj_ctx
, pool
, oid
, bl
, NULL
, NULL
);
3863 if (ret
< 0 && ret
!= -ENOENT
) {
3865 } else if (ret
== -ENOENT
) {
3870 bufferlist::iterator iter
= bl
.begin();
3871 ::decode(zonegroupmap
, iter
);
3872 } catch (buffer::error
& err
) {
3873 ldout(cct
, 0) << "error decoding regionmap from " << pool
<< ":" << oid
<< dendl
;
3877 for (map
<string
, RGWZoneGroup
>::iterator iter
= zonegroupmap
.zonegroups
.begin();
3878 iter
!= zonegroupmap
.zonegroups
.end(); ++iter
) {
3879 RGWZoneGroup
& zonegroup
= iter
->second
;
3880 ret
= zonegroup
.init(cct
, this, false);
3881 ret
= zonegroup
.update();
3882 if (ret
< 0 && ret
!= -ENOENT
) {
3883 ldout(cct
, 0) << "Error could not update zonegroup " << zonegroup
.get_name() << ": " <<
3884 cpp_strerror(-ret
) << dendl
;
3886 } else if (ret
== -ENOENT
) {
3887 ret
= zonegroup
.create();
3889 ldout(cct
, 0) << "Error could not create " << zonegroup
.get_name() << ": " <<
3890 cpp_strerror(-ret
) << dendl
;
3896 current_period
.set_user_quota(zonegroupmap
.user_quota
);
3897 current_period
.set_bucket_quota(zonegroupmap
.bucket_quota
);
3899 // remove the region_map so we don't try to convert again
3900 rgw_raw_obj
obj(pool
, oid
);
3901 ret
= delete_system_obj(obj
);
3903 ldout(cct
, 0) << "Error could not remove " << obj
3904 << " after upgrading to zonegroup map: " << cpp_strerror(ret
) << dendl
;
3912 * Replace all region configuration with zonegroup for
3913 * backward compatability
3914 * Returns 0 on success, -ERR# on failure.
3916 int RGWRados::replace_region_with_zonegroup()
3918 /* copy default region */
3919 /* convert default region to default zonegroup */
3920 string default_oid
= cct
->_conf
->rgw_default_region_info_oid
;
3921 if (default_oid
.empty()) {
3922 default_oid
= default_region_info_oid
;
3926 RGWZoneGroup default_zonegroup
;
3927 rgw_pool pool
{default_zonegroup
.get_pool(cct
)};
3928 string oid
= "converted";
3930 RGWObjectCtx
obj_ctx(this);
3932 int ret
= rgw_get_system_obj(this, obj_ctx
, pool
,oid
, bl
, NULL
, NULL
);
3933 if (ret
< 0 && ret
!= -ENOENT
) {
3934 ldout(cct
, 0) << __func__
<< " failed to read converted: ret "<< ret
<< " " << cpp_strerror(-ret
)
3937 } else if (ret
!= -ENOENT
) {
3938 ldout(cct
, 20) << "System already converted " << dendl
;
3942 string default_region
;
3943 ret
= default_zonegroup
.init(cct
, this, false, true);
3945 ldout(cct
, 0) << __func__
<< " failed init default region: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
3948 ret
= default_zonegroup
.read_default_id(default_region
, true);
3949 if (ret
< 0 && ret
!= -ENOENT
) {
3950 ldout(cct
, 0) << __func__
<< " failed reading old default region: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
3954 /* convert regions to zonegroups */
3955 list
<string
> regions
;
3956 ret
= list_regions(regions
);
3957 if (ret
< 0 && ret
!= -ENOENT
) {
3958 ldout(cct
, 0) << __func__
<< " failed to list regions: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
3960 } else if (ret
== -ENOENT
|| regions
.empty()) {
3961 RGWZoneParams
zoneparams(default_zone_name
);
3962 int ret
= zoneparams
.init(cct
, this);
3963 if (ret
< 0 && ret
!= -ENOENT
) {
3964 ldout(cct
, 0) << __func__
<< ": error initializing default zone params: " << cpp_strerror(-ret
) << dendl
;
3967 /* update master zone */
3968 RGWZoneGroup
default_zg(default_zonegroup_name
);
3969 ret
= default_zg
.init(cct
, this);
3970 if (ret
< 0 && ret
!= -ENOENT
) {
3971 ldout(cct
, 0) << __func__
<< ": error in initializing default zonegroup: " << cpp_strerror(-ret
) << dendl
;
3974 if (ret
!= -ENOENT
&& default_zg
.master_zone
.empty()) {
3975 default_zg
.master_zone
= zoneparams
.get_id();
3976 return default_zg
.update();
3981 string master_region
, master_zone
;
3982 for (list
<string
>::iterator iter
= regions
.begin(); iter
!= regions
.end(); ++iter
) {
3983 if (*iter
!= default_zonegroup_name
){
3984 RGWZoneGroup
region(*iter
);
3985 int ret
= region
.init(cct
, this, true, true);
3987 ldout(cct
, 0) << __func__
<< " failed init region "<< *iter
<< ": " << cpp_strerror(-ret
) << dendl
;
3990 if (region
.is_master_zonegroup()) {
3991 master_region
= region
.get_id();
3992 master_zone
= region
.master_zone
;
3997 /* create realm if there is none.
3998 The realm name will be the region and zone concatenated
3999 realm id will be mds of its name */
4000 if (realm
.get_id().empty() && !master_region
.empty() && !master_zone
.empty()) {
4001 string new_realm_name
= master_region
+ "." + master_zone
;
4002 unsigned char md5
[CEPH_CRYPTO_MD5_DIGESTSIZE
];
4003 char md5_str
[CEPH_CRYPTO_MD5_DIGESTSIZE
* 2 + 1];
4005 hash
.Update((const byte
*)new_realm_name
.c_str(), new_realm_name
.length());
4007 buf_to_hex(md5
, CEPH_CRYPTO_MD5_DIGESTSIZE
, md5_str
);
4008 string
new_realm_id(md5_str
);
4009 RGWRealm
new_realm(new_realm_id
,new_realm_name
);
4010 ret
= new_realm
.init(cct
, this, false);
4012 ldout(cct
, 0) << __func__
<< " Error initing new realm: " << cpp_strerror(-ret
) << dendl
;
4015 ret
= new_realm
.create();
4016 if (ret
< 0 && ret
!= -EEXIST
) {
4017 ldout(cct
, 0) << __func__
<< " Error creating new realm: " << cpp_strerror(-ret
) << dendl
;
4020 ret
= new_realm
.set_as_default();
4022 ldout(cct
, 0) << __func__
<< " Error setting realm as default: " << cpp_strerror(-ret
) << dendl
;
4025 ret
= realm
.init(cct
, this);
4027 ldout(cct
, 0) << __func__
<< " Error initing realm: " << cpp_strerror(-ret
) << dendl
;
4030 ret
= current_period
.init(cct
, this, realm
.get_id(), realm
.get_name());
4032 ldout(cct
, 0) << __func__
<< " Error initing current period: " << cpp_strerror(-ret
) << dendl
;
4037 list
<string
>::iterator iter
;
4038 /* create zonegroups */
4039 for (iter
= regions
.begin(); iter
!= regions
.end(); ++iter
)
4041 ldout(cct
, 0) << __func__
<< "Converting " << *iter
<< dendl
;
4042 /* check to see if we don't have already a zonegroup with this name */
4043 RGWZoneGroup
new_zonegroup(*iter
);
4044 ret
= new_zonegroup
.init(cct
, this);
4045 if (ret
== 0 && new_zonegroup
.get_id() != *iter
) {
4046 ldout(cct
, 0) << __func__
<< " zonegroup "<< *iter
<< " already exists id " << new_zonegroup
.get_id () <<
4047 " skipping conversion " << dendl
;
4050 RGWZoneGroup
zonegroup(*iter
);
4051 zonegroup
.set_id(*iter
);
4052 int ret
= zonegroup
.init(cct
, this, true, true);
4054 ldout(cct
, 0) << __func__
<< " failed init zonegroup: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4057 zonegroup
.realm_id
= realm
.get_id();
4058 /* fix default region master zone */
4059 if (*iter
== default_zonegroup_name
&& zonegroup
.master_zone
.empty()) {
4060 ldout(cct
, 0) << __func__
<< " Setting default zone as master for default region" << dendl
;
4061 zonegroup
.master_zone
= default_zone_name
;
4063 ret
= zonegroup
.update();
4064 if (ret
< 0 && ret
!= -EEXIST
) {
4065 ldout(cct
, 0) << __func__
<< " failed to update zonegroup " << *iter
<< ": ret "<< ret
<< " " << cpp_strerror(-ret
)
4069 ret
= zonegroup
.update_name();
4070 if (ret
< 0 && ret
!= -EEXIST
) {
4071 ldout(cct
, 0) << __func__
<< " failed to update_name for zonegroup " << *iter
<< ": ret "<< ret
<< " " << cpp_strerror(-ret
)
4075 if (zonegroup
.get_name() == default_region
) {
4076 ret
= zonegroup
.set_as_default();
4078 ldout(cct
, 0) << __func__
<< " failed to set_as_default " << *iter
<< ": ret "<< ret
<< " " << cpp_strerror(-ret
)
4083 for (map
<string
, RGWZone
>::const_iterator iter
= zonegroup
.zones
.begin(); iter
!= zonegroup
.zones
.end();
4085 ldout(cct
, 0) << __func__
<< " Converting zone" << iter
->first
<< dendl
;
4086 RGWZoneParams
zoneparams(iter
->first
, iter
->first
);
4087 zoneparams
.set_id(iter
->first
);
4088 zoneparams
.realm_id
= realm
.get_id();
4089 ret
= zoneparams
.init(cct
, this);
4090 if (ret
< 0 && ret
!= -ENOENT
) {
4091 ldout(cct
, 0) << __func__
<< " failed to init zoneparams " << iter
->first
<< ": " << cpp_strerror(-ret
) << dendl
;
4093 } else if (ret
== -ENOENT
) {
4094 ldout(cct
, 0) << __func__
<< " zone is part of another cluster " << iter
->first
<< " skipping " << dendl
;
4097 zonegroup
.realm_id
= realm
.get_id();
4098 ret
= zoneparams
.update();
4099 if (ret
< 0 && ret
!= -EEXIST
) {
4100 ldout(cct
, 0) << __func__
<< " failed to update zoneparams " << iter
->first
<< ": " << cpp_strerror(-ret
) << dendl
;
4103 ret
= zoneparams
.update_name();
4104 if (ret
< 0 && ret
!= -EEXIST
) {
4105 ldout(cct
, 0) << __func__
<< " failed to init zoneparams " << iter
->first
<< ": " << cpp_strerror(-ret
) << dendl
;
4110 if (!current_period
.get_id().empty()) {
4111 ret
= current_period
.add_zonegroup(zonegroup
);
4113 ldout(cct
, 0) << __func__
<< " failed to add zonegroup to current_period: " << cpp_strerror(-ret
) << dendl
;
4119 if (!current_period
.get_id().empty()) {
4120 ret
= current_period
.update();
4122 ldout(cct
, 0) << __func__
<< " failed to update new period: " << cpp_strerror(-ret
) << dendl
;
4125 ret
= current_period
.store_info(false);
4127 ldout(cct
, 0) << __func__
<< " failed to store new period: " << cpp_strerror(-ret
) << dendl
;
4130 ret
= current_period
.reflect();
4132 ldout(cct
, 0) << __func__
<< " failed to update local objects: " << cpp_strerror(-ret
) << dendl
;
4137 for (auto const& iter
: regions
) {
4138 RGWZoneGroup
zonegroup(iter
);
4139 int ret
= zonegroup
.init(cct
, this, true, true);
4141 ldout(cct
, 0) << __func__
<< " failed init zonegroup" << iter
<< ": ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4144 ret
= zonegroup
.delete_obj(true);
4145 if (ret
< 0 && ret
!= -ENOENT
) {
4146 ldout(cct
, 0) << __func__
<< " failed to delete region " << iter
<< ": ret "<< ret
<< " " << cpp_strerror(-ret
)
4152 /* mark as converted */
4153 ret
= rgw_put_system_obj(this, pool
, oid
, bl
.c_str(), bl
.length(),
4154 true, NULL
, real_time(), NULL
);
4156 ldout(cct
, 0) << __func__
<< " failed to mark cluster as converted: ret "<< ret
<< " " << cpp_strerror(-ret
)
4164 int RGWRados::init_zg_from_period(bool *initialized
)
4166 *initialized
= false;
4168 if (current_period
.get_id().empty()) {
4172 int ret
= zonegroup
.init(cct
, this);
4173 ldout(cct
, 20) << "period zonegroup init ret " << ret
<< dendl
;
4174 if (ret
== -ENOENT
) {
4178 ldout(cct
, 0) << "failed reading zonegroup info: " << cpp_strerror(-ret
) << dendl
;
4181 ldout(cct
, 20) << "period zonegroup name " << zonegroup
.get_name() << dendl
;
4183 map
<string
, RGWZoneGroup
>::const_iterator iter
=
4184 current_period
.get_map().zonegroups
.find(zonegroup
.get_id());
4186 if (iter
!= current_period
.get_map().zonegroups
.end()) {
4187 ldout(cct
, 20) << "using current period zonegroup " << zonegroup
.get_name() << dendl
;
4188 zonegroup
= iter
->second
;
4189 ret
= zonegroup
.init(cct
, this, false);
4191 ldout(cct
, 0) << "failed init zonegroup: " << " " << cpp_strerror(-ret
) << dendl
;
4194 ret
= zone_params
.init(cct
, this);
4195 if (ret
< 0 && ret
!= -ENOENT
) {
4196 ldout(cct
, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret
) << dendl
;
4198 } if (ret
==-ENOENT
&& zonegroup
.get_name() == default_zonegroup_name
) {
4199 ldout(cct
, 10) << " Using default name "<< default_zone_name
<< dendl
;
4200 zone_params
.set_name(default_zone_name
);
4201 ret
= zone_params
.init(cct
, this);
4202 if (ret
< 0 && ret
!= -ENOENT
) {
4203 ldout(cct
, 0) << "failed reading zone params info: " << " " << cpp_strerror(-ret
) << dendl
;
4208 for (iter
= current_period
.get_map().zonegroups
.begin();
4209 iter
!= current_period
.get_map().zonegroups
.end(); ++iter
){
4210 const RGWZoneGroup
& zg
= iter
->second
;
4211 // use endpoints from the zonegroup's master zone
4212 auto master
= zg
.zones
.find(zg
.master_zone
);
4213 if (master
== zg
.zones
.end()) {
4214 // fix missing master zone for a single zone zonegroup
4215 if (zg
.master_zone
.empty() && zg
.zones
.size() == 1) {
4216 master
= zg
.zones
.begin();
4217 ldout(cct
, 0) << "zonegroup " << zg
.get_name() << " missing master_zone, setting zone " <<
4218 master
->second
.name
<< " id:" << master
->second
.id
<< " as master" << dendl
;
4219 if (zonegroup
.get_id() == zg
.get_id()) {
4220 zonegroup
.master_zone
= master
->second
.id
;
4221 ret
= zonegroup
.update();
4223 ldout(cct
, 0) << "error updating zonegroup : " << cpp_strerror(-ret
) << dendl
;
4227 RGWZoneGroup
fixed_zg(zg
.get_id(),zg
.get_name());
4228 ret
= fixed_zg
.init(cct
, this);
4230 ldout(cct
, 0) << "error initializing zonegroup : " << cpp_strerror(-ret
) << dendl
;
4233 fixed_zg
.master_zone
= master
->second
.id
;
4234 ret
= fixed_zg
.update();
4236 ldout(cct
, 0) << "error initializing zonegroup : " << cpp_strerror(-ret
) << dendl
;
4241 ldout(cct
, 0) << "zonegroup " << zg
.get_name() << " missing zone for master_zone=" <<
4242 zg
.master_zone
<< dendl
;
4246 const auto& endpoints
= master
->second
.endpoints
;
4247 add_new_connection_to_map(zonegroup_conn_map
, zg
, new RGWRESTConn(cct
, this, zg
.get_id(), endpoints
));
4248 if (!current_period
.get_master_zonegroup().empty() &&
4249 zg
.get_id() == current_period
.get_master_zonegroup()) {
4250 rest_master_conn
= new RGWRESTConn(cct
, this, zg
.get_id(), endpoints
);
4254 *initialized
= true;
4259 int RGWRados::init_zg_from_local(bool *creating_defaults
)
4261 int ret
= zonegroup
.init(cct
, this);
4262 if ( (ret
< 0 && ret
!= -ENOENT
) || (ret
== -ENOENT
&& !cct
->_conf
->rgw_zonegroup
.empty())) {
4263 ldout(cct
, 0) << "failed reading zonegroup info: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4265 } else if (ret
== -ENOENT
) {
4266 *creating_defaults
= true;
4267 ldout(cct
, 10) << "Creating default zonegroup " << dendl
;
4268 ret
= zonegroup
.create_default();
4270 ldout(cct
, 0) << "failure in zonegroup create_default: ret "<< ret
<< " " << cpp_strerror(-ret
)
4274 ret
= zonegroup
.init(cct
, this);
4276 ldout(cct
, 0) << "failure in zonegroup create_default: ret "<< ret
<< " " << cpp_strerror(-ret
)
4281 ldout(cct
, 20) << "zonegroup " << zonegroup
.get_name() << dendl
;
4282 if (zonegroup
.is_master_zonegroup()) {
4283 // use endpoints from the zonegroup's master zone
4284 auto master
= zonegroup
.zones
.find(zonegroup
.master_zone
);
4285 if (master
== zonegroup
.zones
.end()) {
4286 // fix missing master zone for a single zone zonegroup
4287 if (zonegroup
.master_zone
.empty() && zonegroup
.zones
.size() == 1) {
4288 master
= zonegroup
.zones
.begin();
4289 ldout(cct
, 0) << "zonegroup " << zonegroup
.get_name() << " missing master_zone, setting zone " <<
4290 master
->second
.name
<< " id:" << master
->second
.id
<< " as master" << dendl
;
4291 zonegroup
.master_zone
= master
->second
.id
;
4292 ret
= zonegroup
.update();
4294 ldout(cct
, 0) << "error initializing zonegroup : " << cpp_strerror(-ret
) << dendl
;
4298 ldout(cct
, 0) << "zonegroup " << zonegroup
.get_name() << " missing zone for "
4299 "master_zone=" << zonegroup
.master_zone
<< dendl
;
4303 const auto& endpoints
= master
->second
.endpoints
;
4304 rest_master_conn
= new RGWRESTConn(cct
, this, zonegroup
.get_id(), endpoints
);
4311 bool RGWRados::zone_syncs_from(RGWZone
& target_zone
, RGWZone
& source_zone
)
4313 return target_zone
.syncs_from(source_zone
.name
) &&
4314 sync_modules_manager
->supports_data_export(source_zone
.tier_type
);
4318 * Initialize the RADOS instance and prepare to do other ops
4319 * Returns 0 on success, -ERR# on failure.
4321 int RGWRados::init_complete()
4323 int ret
= realm
.init(cct
, this);
4324 if (ret
< 0 && ret
!= -ENOENT
) {
4325 ldout(cct
, 0) << "failed reading realm info: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4327 } else if (ret
!= -ENOENT
) {
4328 ldout(cct
, 20) << "realm " << realm
.get_name() << " " << realm
.get_id() << dendl
;
4329 ret
= current_period
.init(cct
, this, realm
.get_id(), realm
.get_name());
4330 if (ret
< 0 && ret
!= -ENOENT
) {
4331 ldout(cct
, 0) << "failed reading current period info: " << " " << cpp_strerror(-ret
) << dendl
;
4334 ldout(cct
, 20) << "current period " << current_period
.get_id() << dendl
;
4337 ret
= replace_region_with_zonegroup();
4339 lderr(cct
) << "failed converting region to zonegroup : ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4343 ret
= convert_regionmap();
4345 lderr(cct
) << "failed converting regionmap: " << cpp_strerror(-ret
) << dendl
;
4349 bool zg_initialized
= false;
4351 if (!current_period
.get_id().empty()) {
4352 ret
= init_zg_from_period(&zg_initialized
);
4358 bool creating_defaults
= false;
4359 bool using_local
= (!zg_initialized
);
4361 ldout(cct
, 10) << " cannot find current period zonegroup using local zonegroup" << dendl
;
4362 ret
= init_zg_from_local(&creating_defaults
);
4366 // read period_config into current_period
4367 auto& period_config
= current_period
.get_config();
4368 ret
= period_config
.read(this, zonegroup
.realm_id
);
4369 if (ret
< 0 && ret
!= -ENOENT
) {
4370 ldout(cct
, 0) << "ERROR: failed to read period config: "
4371 << cpp_strerror(ret
) << dendl
;
4376 ldout(cct
, 10) << "Cannot find current period zone using local zone" << dendl
;
4377 if (creating_defaults
&& cct
->_conf
->rgw_zone
.empty()) {
4378 ldout(cct
, 10) << " Using default name "<< default_zone_name
<< dendl
;
4379 zone_params
.set_name(default_zone_name
);
4382 ret
= zone_params
.init(cct
, this);
4383 if (ret
< 0 && ret
!= -ENOENT
) {
4384 lderr(cct
) << "failed reading zone info: ret "<< ret
<< " " << cpp_strerror(-ret
) << dendl
;
4387 map
<string
, RGWZone
>::iterator zone_iter
= get_zonegroup().zones
.find(zone_params
.get_id());
4388 if (zone_iter
== get_zonegroup().zones
.end()) {
4390 lderr(cct
) << "Cannot find zone id=" << zone_params
.get_id() << " (name=" << zone_params
.get_name() << ")" << dendl
;
4393 ldout(cct
, 1) << "Cannot find zone id=" << zone_params
.get_id() << " (name=" << zone_params
.get_name() << "), switching to local zonegroup configuration" << dendl
;
4394 ret
= init_zg_from_local(&creating_defaults
);
4398 zone_iter
= get_zonegroup().zones
.find(zone_params
.get_id());
4400 if (zone_iter
!= get_zonegroup().zones
.end()) {
4401 zone_public_config
= zone_iter
->second
;
4402 ldout(cct
, 20) << "zone " << zone_params
.get_name() << dendl
;
4404 lderr(cct
) << "Cannot find zone id=" << zone_params
.get_id() << " (name=" << zone_params
.get_name() << ")" << dendl
;
4408 zone_short_id
= current_period
.get_map().get_zone_short_id(zone_params
.get_id());
4410 if (run_sync_thread
) {
4411 ret
= sync_modules_manager
->create_instance(cct
, zone_public_config
.tier_type
, zone_params
.tier_config
, &sync_module
);
4413 lderr(cct
) << "ERROR: failed to init sync module instance, ret=" << ret
<< dendl
;
4418 writeable_zone
= (zone_public_config
.tier_type
.empty() || zone_public_config
.tier_type
== "rgw");
4420 init_unique_trans_id_deps();
4422 finisher
= new Finisher(cct
);
4425 period_puller
.reset(new RGWPeriodPuller(this));
4426 period_history
.reset(new RGWPeriodHistory(cct
, period_puller
.get(),
4429 if (need_watch_notify()) {
4432 lderr(cct
) << "ERROR: failed to initialize watch: " << cpp_strerror(-ret
) << dendl
;
4437 /* first build all zones index */
4438 for (auto ziter
: get_zonegroup().zones
) {
4439 const string
& id
= ziter
.first
;
4440 RGWZone
& z
= ziter
.second
;
4441 zone_id_by_name
[z
.name
] = id
;
4445 if (zone_by_id
.find(zone_id()) == zone_by_id
.end()) {
4446 ldout(cct
, 0) << "WARNING: could not find zone config in zonegroup for local zone (" << zone_id() << "), will use defaults" << dendl
;
4448 zone_public_config
= zone_by_id
[zone_id()];
4449 for (auto ziter
: get_zonegroup().zones
) {
4450 const string
& id
= ziter
.first
;
4451 RGWZone
& z
= ziter
.second
;
4452 if (id
== zone_id()) {
4455 if (z
.endpoints
.empty()) {
4456 ldout(cct
, 0) << "WARNING: can't generate connection for zone " << z
.id
<< " id " << z
.name
<< ": no endpoints defined" << dendl
;
4459 ldout(cct
, 20) << "generating connection object for zone " << z
.name
<< " id " << z
.id
<< dendl
;
4460 RGWRESTConn
*conn
= new RGWRESTConn(cct
, this, z
.id
, z
.endpoints
);
4461 zone_conn_map
[id
] = conn
;
4462 if (zone_syncs_from(zone_public_config
, z
) ||
4463 zone_syncs_from(z
, zone_public_config
)) {
4464 if (zone_syncs_from(zone_public_config
, z
)) {
4465 zone_data_sync_from_map
[id
] = conn
;
4467 if (zone_syncs_from(z
, zone_public_config
)) {
4468 zone_data_notify_to_map
[id
] = conn
;
4471 ldout(cct
, 20) << "NOTICE: not syncing to/from zone " << z
.name
<< " id " << z
.id
<< dendl
;
4475 ret
= open_root_pool_ctx();
4479 ret
= open_gc_pool_ctx();
4483 ret
= open_lc_pool_ctx();
4487 ret
= open_objexp_pool_ctx();
4491 ret
= open_reshard_pool_ctx();
4495 pools_initialized
= true;
4498 gc
->initialize(cct
, this);
4500 obj_expirer
= new RGWObjectExpirer(this);
4502 if (use_gc_thread
) {
4503 gc
->start_processor();
4504 obj_expirer
->start_processor();
4507 /* no point of running sync thread if we don't have a master zone configured
4508 or there is no rest_master_conn */
4509 if (get_zonegroup().master_zone
.empty() || !rest_master_conn
4510 || current_period
.get_id().empty()) {
4511 run_sync_thread
= false;
4514 if (run_sync_thread
) {
4515 // initialize the log period history
4516 meta_mgr
->init_oldest_log_period();
4519 async_rados
= new RGWAsyncRadosProcessor(this, cct
->_conf
->rgw_num_async_rados_threads
);
4520 async_rados
->start();
4522 ret
= meta_mgr
->init(current_period
.get_id());
4524 lderr(cct
) << "ERROR: failed to initialize metadata log: "
4525 << cpp_strerror(-ret
) << dendl
;
4529 if (is_meta_master()) {
4530 auto md_log
= meta_mgr
->get_log(current_period
.get_id());
4531 meta_notifier
= new RGWMetaNotifier(this, md_log
);
4532 meta_notifier
->start();
4535 if (run_sync_thread
) {
4536 Mutex::Locker
l(meta_sync_thread_lock
);
4537 meta_sync_processor_thread
= new RGWMetaSyncProcessorThread(this, async_rados
);
4538 ret
= meta_sync_processor_thread
->init();
4540 ldout(cct
, 0) << "ERROR: failed to initialize meta sync thread" << dendl
;
4543 meta_sync_processor_thread
->start();
4545 // configure the bucket trim manager
4546 rgw::BucketTrimConfig config
;
4547 rgw::configure_bucket_trim(cct
, config
);
4549 bucket_trim
.emplace(this, config
);
4550 ret
= bucket_trim
->init();
4552 ldout(cct
, 0) << "ERROR: failed to start bucket trim manager" << dendl
;
4556 Mutex::Locker
dl(data_sync_thread_lock
);
4557 for (auto iter
: zone_data_sync_from_map
) {
4558 ldout(cct
, 5) << "starting data sync thread for zone " << iter
.first
<< dendl
;
4559 auto *thread
= new RGWDataSyncProcessorThread(this, async_rados
, iter
.first
,
4561 ret
= thread
->init();
4563 ldout(cct
, 0) << "ERROR: failed to initialize data sync thread" << dendl
;
4567 data_sync_processor_threads
[iter
.first
] = thread
;
4569 auto interval
= cct
->_conf
->rgw_sync_log_trim_interval
;
4571 sync_log_trimmer
= new RGWSyncLogTrimThread(this, &*bucket_trim
, interval
);
4572 ret
= sync_log_trimmer
->init();
4574 ldout(cct
, 0) << "ERROR: failed to initialize sync log trim thread" << dendl
;
4577 sync_log_trimmer
->start();
4580 data_notifier
= new RGWDataNotifier(this);
4581 data_notifier
->start();
4584 lc
->initialize(cct
, this);
4587 lc
->start_processor();
4589 quota_handler
= RGWQuotaHandler::generate_handler(this, quota_threads
);
4591 bucket_index_max_shards
= (cct
->_conf
->rgw_override_bucket_index_max_shards
? cct
->_conf
->rgw_override_bucket_index_max_shards
:
4592 get_zone().bucket_index_max_shards
);
4593 if (bucket_index_max_shards
> get_max_bucket_shards()) {
4594 bucket_index_max_shards
= get_max_bucket_shards();
4595 ldout(cct
, 1) << __func__
<< " bucket index max shards is too large, reset to value: "
4596 << get_max_bucket_shards() << dendl
;
4598 ldout(cct
, 20) << __func__
<< " bucket index max shards: " << bucket_index_max_shards
<< dendl
;
4600 binfo_cache
= new RGWChainedCacheImpl
<bucket_info_entry
>;
4601 binfo_cache
->init(this);
4603 bool need_tombstone_cache
= !zone_data_notify_to_map
.empty(); /* have zones syncing from us */
4605 if (need_tombstone_cache
) {
4606 obj_tombstone_cache
= new tombstone_cache_t(cct
->_conf
->rgw_obj_tombstone_cache_size
);
4609 reshard_wait
= std::make_shared
<RGWReshardWait
>(this);
4611 reshard
= new RGWReshard(this);
4613 /* only the master zone in the zonegroup reshards buckets */
4614 run_reshard_thread
= run_reshard_thread
&& (get_zonegroup().master_zone
== zone_public_config
.id
);
4615 if (run_reshard_thread
) {
4616 reshard
->start_processor();
4619 index_completion_manager
= new RGWIndexCompletionManager(this);
4620 ret
= index_completion_manager
->start();
4626 * Initialize the RADOS instance and prepare to do other ops
4627 * Returns 0 on success, -ERR# on failure.
4629 int RGWRados::initialize()
4637 return init_complete();
4640 void RGWRados::finalize_watch()
4642 for (int i
= 0; i
< num_watchers
; i
++) {
4643 RGWWatcher
*watcher
= watchers
[i
];
4644 watcher
->unregister_watch();
4648 delete[] notify_oids
;
4652 void RGWRados::schedule_context(Context
*c
) {
4656 int RGWRados::list_raw_prefixed_objs(const rgw_pool
& pool
, const string
& prefix
, list
<string
>& result
)
4659 RGWListRawObjsCtx ctx
;
4662 int r
= list_raw_objects(pool
, prefix
, 1000,
4663 ctx
, oids
, &is_truncated
);
4667 list
<string
>::iterator iter
;
4668 for (iter
= oids
.begin(); iter
!= oids
.end(); ++iter
) {
4669 string
& val
= *iter
;
4670 if (val
.size() > prefix
.size())
4671 result
.push_back(val
.substr(prefix
.size()));
4673 } while (is_truncated
);
4678 int RGWRados::list_regions(list
<string
>& regions
)
4680 RGWZoneGroup zonegroup
;
4682 return list_raw_prefixed_objs(zonegroup
.get_pool(cct
), region_info_oid_prefix
, regions
);
4685 int RGWRados::list_zonegroups(list
<string
>& zonegroups
)
4687 RGWZoneGroup zonegroup
;
4689 return list_raw_prefixed_objs(zonegroup
.get_pool(cct
), zonegroup_names_oid_prefix
, zonegroups
);
4692 int RGWRados::list_zones(list
<string
>& zones
)
4694 RGWZoneParams zoneparams
;
4696 return list_raw_prefixed_objs(zoneparams
.get_pool(cct
), zone_names_oid_prefix
, zones
);
4699 int RGWRados::list_realms(list
<string
>& realms
)
4701 RGWRealm
realm(cct
, this);
4702 return list_raw_prefixed_objs(realm
.get_pool(cct
), realm_names_oid_prefix
, realms
);
4705 int RGWRados::list_periods(list
<string
>& periods
)
4708 list
<string
> raw_periods
;
4709 int ret
= list_raw_prefixed_objs(period
.get_pool(cct
), period
.get_info_oid_prefix(), raw_periods
);
4713 for (const auto& oid
: raw_periods
) {
4714 size_t pos
= oid
.find(".");
4715 if (pos
!= std::string::npos
) {
4716 periods
.push_back(oid
.substr(0, pos
));
4718 periods
.push_back(oid
);
4721 periods
.sort(); // unique() only detects duplicates if they're adjacent
4727 int RGWRados::list_periods(const string
& current_period
, list
<string
>& periods
)
4730 string period_id
= current_period
;
4731 while(!period_id
.empty()) {
4732 RGWPeriod
period(period_id
);
4733 ret
= period
.init(cct
, this);
4737 periods
.push_back(period
.get_id());
4738 period_id
= period
.get_predecessor();
4745 * Open the pool used as root for this gateway
4746 * Returns: 0 on success, -ERR# otherwise.
4748 int RGWRados::open_root_pool_ctx()
4750 return rgw_init_ioctx(get_rados_handle(), get_zone_params().domain_root
, root_pool_ctx
, true);
4753 int RGWRados::open_gc_pool_ctx()
4755 return rgw_init_ioctx(get_rados_handle(), get_zone_params().gc_pool
, gc_pool_ctx
, true);
4758 int RGWRados::open_lc_pool_ctx()
4760 return rgw_init_ioctx(get_rados_handle(), get_zone_params().lc_pool
, lc_pool_ctx
, true);
4763 int RGWRados::open_objexp_pool_ctx()
4765 return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, objexp_pool_ctx
, true);
4768 int RGWRados::open_reshard_pool_ctx()
4770 return rgw_init_ioctx(get_rados_handle(), get_zone_params().reshard_pool
, reshard_pool_ctx
, true);
4773 int RGWRados::init_watch()
4775 int r
= rgw_init_ioctx(&rados
[0], get_zone_params().control_pool
, control_pool_ctx
, true);
4780 num_watchers
= cct
->_conf
->rgw_num_control_oids
;
4782 bool compat_oid
= (num_watchers
== 0);
4784 if (num_watchers
<= 0)
4787 notify_oids
= new string
[num_watchers
];
4788 watchers
= new RGWWatcher
*[num_watchers
];
4790 for (int i
=0; i
< num_watchers
; i
++) {
4791 string
& notify_oid
= notify_oids
[i
];
4792 notify_oid
= notify_oid_prefix
;
4795 snprintf(buf
, sizeof(buf
), ".%d", i
);
4796 notify_oid
.append(buf
);
4798 r
= control_pool_ctx
.create(notify_oid
, false);
4799 if (r
< 0 && r
!= -EEXIST
)
4802 RGWWatcher
*watcher
= new RGWWatcher(this, i
, notify_oid
);
4803 watchers
[i
] = watcher
;
4805 r
= watcher
->register_watch();
4810 watch_initialized
= true;
4812 set_cache_enabled(true);
4817 void RGWRados::pick_control_oid(const string
& key
, string
& notify_oid
)
4819 uint32_t r
= ceph_str_hash_linux(key
.c_str(), key
.size());
4821 int i
= r
% num_watchers
;
4823 snprintf(buf
, sizeof(buf
), ".%d", i
);
4825 notify_oid
= notify_oid_prefix
;
4826 notify_oid
.append(buf
);
4829 int RGWRados::open_pool_ctx(const rgw_pool
& pool
, librados::IoCtx
& io_ctx
)
4831 librados::Rados
*rad
= get_rados_handle();
4832 int r
= rgw_init_ioctx(rad
, pool
, io_ctx
);
4836 if (!pools_initialized
)
4839 r
= rad
->pool_create(pool
.name
.c_str());
4840 if (r
< 0 && r
!= -EEXIST
)
4843 r
= rgw_init_ioctx(rad
, pool
, io_ctx
);
4847 r
= io_ctx
.application_enable(pg_pool_t::APPLICATION_NAME_RGW
, false);
4848 if (r
< 0 && r
!= -EOPNOTSUPP
)
4853 void RGWRados::build_bucket_index_marker(const string
& shard_id_str
, const string
& shard_marker
,
4856 *marker
= shard_id_str
;
4857 marker
->append(BucketIndexShardsManager::KEY_VALUE_SEPARATOR
);
4858 marker
->append(shard_marker
);
4862 int RGWRados::open_bucket_index_ctx(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
)
4864 const string
*rule
= &bucket_info
.placement_rule
;
4865 if (rule
->empty()) {
4866 rule
= &zonegroup
.default_placement
;
4868 auto iter
= zone_params
.placement_pools
.find(*rule
);
4869 if (iter
== zone_params
.placement_pools
.end()) {
4870 ldout(cct
, 0) << "could not find placement rule " << *rule
<< " within zonegroup " << dendl
;
4874 int r
= open_pool_ctx(iter
->second
.index_pool
, index_ctx
);
4882 * set up a bucket listing.
4883 * handle is filled in.
4884 * Returns 0 on success, -ERR# otherwise.
4886 int RGWRados::list_buckets_init(RGWAccessHandle
*handle
)
4888 librados::NObjectIterator
*state
= new librados::NObjectIterator(root_pool_ctx
.nobjects_begin());
4889 *handle
= (RGWAccessHandle
)state
;
4894 * get the next bucket in the listing.
4896 * handle is updated.
4897 * returns 0 on success, -ERR# otherwise.
4899 int RGWRados::list_buckets_next(rgw_bucket_dir_entry
& obj
, RGWAccessHandle
*handle
)
4901 librados::NObjectIterator
*state
= (librados::NObjectIterator
*)*handle
;
4904 if (*state
== root_pool_ctx
.nobjects_end()) {
4909 obj
.key
.name
= (*state
)->get_oid();
4910 if (obj
.key
.name
[0] == '_') {
4911 obj
.key
.name
= obj
.key
.name
.substr(1);
4915 } while (obj
.key
.name
[0] == '.'); /* skip all entries starting with '.' */
4923 struct log_list_state
{
4925 librados::IoCtx io_ctx
;
4926 librados::NObjectIterator obit
;
4929 int RGWRados::log_list_init(const string
& prefix
, RGWAccessHandle
*handle
)
4931 log_list_state
*state
= new log_list_state
;
4932 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, state
->io_ctx
);
4937 state
->prefix
= prefix
;
4938 state
->obit
= state
->io_ctx
.nobjects_begin();
4939 *handle
= (RGWAccessHandle
)state
;
4943 int RGWRados::log_list_next(RGWAccessHandle handle
, string
*name
)
4945 log_list_state
*state
= static_cast<log_list_state
*>(handle
);
4947 if (state
->obit
== state
->io_ctx
.nobjects_end()) {
4951 if (state
->prefix
.length() &&
4952 state
->obit
->get_oid().find(state
->prefix
) != 0) {
4956 *name
= state
->obit
->get_oid();
4963 int RGWRados::log_remove(const string
& name
)
4965 librados::IoCtx io_ctx
;
4966 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
);
4969 return io_ctx
.remove(name
);
4972 struct log_show_state
{
4973 librados::IoCtx io_ctx
;
4975 bufferlist::iterator p
;
4979 log_show_state() : pos(0), eof(false) {}
4982 int RGWRados::log_show_init(const string
& name
, RGWAccessHandle
*handle
)
4984 log_show_state
*state
= new log_show_state
;
4985 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, state
->io_ctx
);
4991 *handle
= (RGWAccessHandle
)state
;
4995 int RGWRados::log_show_next(RGWAccessHandle handle
, rgw_log_entry
*entry
)
4997 log_show_state
*state
= static_cast<log_show_state
*>(handle
);
4998 off_t off
= state
->p
.get_off();
5000 ldout(cct
, 10) << "log_show_next pos " << state
->pos
<< " bl " << state
->bl
.length()
5002 << " eof " << (int)state
->eof
5005 unsigned chunk
= 1024*1024;
5006 if ((state
->bl
.length() - off
) < chunk
/2 && !state
->eof
) {
5008 int r
= state
->io_ctx
.read(state
->name
, more
, chunk
, state
->pos
);
5014 old
.substr_of(state
->bl
, off
, state
->bl
.length() - off
);
5015 } catch (buffer::error
& err
) {
5019 state
->bl
.claim(old
);
5020 state
->bl
.claim_append(more
);
5021 state
->p
= state
->bl
.begin();
5022 if ((unsigned)r
< chunk
)
5024 ldout(cct
, 10) << " read " << r
<< dendl
;
5028 return 0; // end of file
5030 ::decode(*entry
, state
->p
);
5032 catch (const buffer::error
&e
) {
5039 * usage_log_hash: get usage log key hash, based on name and index
5041 * Get the usage object name. Since a user may have more than 1
5042 * object holding that info (multiple shards), we use index to
5043 * specify that shard number. Once index exceeds max shards it
5045 * If name is not being set, results for all users will be returned
5046 * and index will wrap only after total shards number.
5048 * @param cct [in] ceph context
5049 * @param name [in] user name
5050 * @param hash [out] hash value
5051 * @param index [in] shard index number
5053 static void usage_log_hash(CephContext
*cct
, const string
& name
, string
& hash
, uint32_t index
)
5055 uint32_t val
= index
;
5057 if (!name
.empty()) {
5058 int max_user_shards
= cct
->_conf
->rgw_usage_max_user_shards
;
5059 val
%= max_user_shards
;
5060 val
+= ceph_str_hash_linux(name
.c_str(), name
.size());
5063 int max_shards
= cct
->_conf
->rgw_usage_max_shards
;
5064 snprintf(buf
, sizeof(buf
), RGW_USAGE_OBJ_PREFIX
"%u", (unsigned)(val
% max_shards
));
5068 int RGWRados::log_usage(map
<rgw_user_bucket
, RGWUsageBatch
>& usage_info
)
5072 map
<string
, rgw_usage_log_info
> log_objs
;
5077 /* restructure usage map, zone by object hash */
5078 map
<rgw_user_bucket
, RGWUsageBatch
>::iterator iter
;
5079 for (iter
= usage_info
.begin(); iter
!= usage_info
.end(); ++iter
) {
5080 const rgw_user_bucket
& ub
= iter
->first
;
5081 RGWUsageBatch
& info
= iter
->second
;
5083 if (ub
.user
.empty()) {
5084 ldout(cct
, 0) << "WARNING: RGWRados::log_usage(): user name empty (bucket=" << ub
.bucket
<< "), skipping" << dendl
;
5088 if (ub
.user
!= last_user
) {
5089 /* index *should* be random, but why waste extra cycles
5090 in most cases max user shards is not going to exceed 1,
5091 so just incrementing it */
5092 usage_log_hash(cct
, ub
.user
, hash
, index
++);
5094 last_user
= ub
.user
;
5095 vector
<rgw_usage_log_entry
>& v
= log_objs
[hash
].entries
;
5097 for (auto miter
= info
.m
.begin(); miter
!= info
.m
.end(); ++miter
) {
5098 v
.push_back(miter
->second
);
5102 map
<string
, rgw_usage_log_info
>::iterator liter
;
5104 for (liter
= log_objs
.begin(); liter
!= log_objs
.end(); ++liter
) {
5105 int r
= cls_obj_usage_log_add(liter
->first
, liter
->second
);
5112 int RGWRados::read_usage(const rgw_user
& user
, uint64_t start_epoch
, uint64_t end_epoch
, uint32_t max_entries
,
5113 bool *is_truncated
, RGWUsageIter
& usage_iter
, map
<rgw_user_bucket
, rgw_usage_log_entry
>& usage
)
5115 uint32_t num
= max_entries
;
5116 string hash
, first_hash
;
5117 string user_str
= user
.to_str();
5118 usage_log_hash(cct
, user_str
, first_hash
, 0);
5120 if (usage_iter
.index
) {
5121 usage_log_hash(cct
, user_str
, hash
, usage_iter
.index
);
5129 map
<rgw_user_bucket
, rgw_usage_log_entry
> ret_usage
;
5130 map
<rgw_user_bucket
, rgw_usage_log_entry
>::iterator iter
;
5132 int ret
= cls_obj_usage_log_read(hash
, user_str
, start_epoch
, end_epoch
, num
,
5133 usage_iter
.read_iter
, ret_usage
, is_truncated
);
5140 num
-= ret_usage
.size();
5142 for (iter
= ret_usage
.begin(); iter
!= ret_usage
.end(); ++iter
) {
5143 usage
[iter
->first
].aggregate(iter
->second
);
5147 if (!*is_truncated
) {
5148 usage_iter
.read_iter
.clear();
5149 usage_log_hash(cct
, user_str
, hash
, ++usage_iter
.index
);
5151 } while (num
&& !*is_truncated
&& hash
!= first_hash
);
5155 int RGWRados::trim_usage(rgw_user
& user
, uint64_t start_epoch
, uint64_t end_epoch
)
5158 string hash
, first_hash
;
5159 string user_str
= user
.to_str();
5160 usage_log_hash(cct
, user_str
, first_hash
, index
);
5164 int ret
= cls_obj_usage_log_trim(hash
, user_str
, start_epoch
, end_epoch
);
5166 if (ret
< 0 && ret
!= -ENOENT
)
5169 usage_log_hash(cct
, user_str
, hash
, ++index
);
5170 } while (hash
!= first_hash
);
5175 int RGWRados::key_to_shard_id(const string
& key
, int max_shards
)
5177 return rgw_shards_hash(key
, max_shards
);
5180 void RGWRados::shard_name(const string
& prefix
, unsigned max_shards
, const string
& key
, string
& name
, int *shard_id
)
5182 uint32_t val
= ceph_str_hash_linux(key
.c_str(), key
.size());
5185 *shard_id
= val
% max_shards
;
5187 snprintf(buf
, sizeof(buf
), "%u", (unsigned)(val
% max_shards
));
5188 name
= prefix
+ buf
;
5191 void RGWRados::shard_name(const string
& prefix
, unsigned max_shards
, const string
& section
, const string
& key
, string
& name
)
5193 uint32_t val
= ceph_str_hash_linux(key
.c_str(), key
.size());
5194 val
^= ceph_str_hash_linux(section
.c_str(), section
.size());
5196 snprintf(buf
, sizeof(buf
), "%u", (unsigned)(val
% max_shards
));
5197 name
= prefix
+ buf
;
5200 void RGWRados::shard_name(const string
& prefix
, unsigned shard_id
, string
& name
)
5203 snprintf(buf
, sizeof(buf
), "%u", shard_id
);
5204 name
= prefix
+ buf
;
5208 void RGWRados::time_log_prepare_entry(cls_log_entry
& entry
, const real_time
& ut
, const string
& section
, const string
& key
, bufferlist
& bl
)
5210 cls_log_add_prepare_entry(entry
, utime_t(ut
), section
, key
, bl
);
5213 int RGWRados::time_log_add_init(librados::IoCtx
& io_ctx
)
5215 return rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
, true);
5219 int RGWRados::time_log_add(const string
& oid
, const real_time
& ut
, const string
& section
, const string
& key
, bufferlist
& bl
)
5221 librados::IoCtx io_ctx
;
5223 int r
= time_log_add_init(io_ctx
);
5228 ObjectWriteOperation op
;
5230 cls_log_add(op
, t
, section
, key
, bl
);
5232 return io_ctx
.operate(oid
, &op
);
5235 int RGWRados::time_log_add(const string
& oid
, list
<cls_log_entry
>& entries
,
5236 librados::AioCompletion
*completion
, bool monotonic_inc
)
5238 librados::IoCtx io_ctx
;
5240 int r
= time_log_add_init(io_ctx
);
5245 ObjectWriteOperation op
;
5246 cls_log_add(op
, entries
, monotonic_inc
);
5249 r
= io_ctx
.operate(oid
, &op
);
5251 r
= io_ctx
.aio_operate(oid
, completion
, &op
);
5256 int RGWRados::time_log_list(const string
& oid
, const real_time
& start_time
, const real_time
& end_time
,
5257 int max_entries
, list
<cls_log_entry
>& entries
,
5258 const string
& marker
,
5262 librados::IoCtx io_ctx
;
5264 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
);
5267 librados::ObjectReadOperation op
;
5269 utime_t
st(start_time
);
5270 utime_t
et(end_time
);
5272 cls_log_list(op
, st
, et
, marker
, max_entries
, entries
,
5273 out_marker
, truncated
);
5277 int ret
= io_ctx
.operate(oid
, &op
, &obl
);
5284 int RGWRados::time_log_info(const string
& oid
, cls_log_header
*header
)
5286 librados::IoCtx io_ctx
;
5288 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
);
5291 librados::ObjectReadOperation op
;
5293 cls_log_info(op
, header
);
5297 int ret
= io_ctx
.operate(oid
, &op
, &obl
);
5304 int RGWRados::time_log_info_async(librados::IoCtx
& io_ctx
, const string
& oid
, cls_log_header
*header
, librados::AioCompletion
*completion
)
5306 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
);
5310 librados::ObjectReadOperation op
;
5312 cls_log_info(op
, header
);
5314 int ret
= io_ctx
.aio_operate(oid
, completion
, &op
, NULL
);
5321 int RGWRados::time_log_trim(const string
& oid
, const real_time
& start_time
, const real_time
& end_time
,
5322 const string
& from_marker
, const string
& to_marker
,
5323 librados::AioCompletion
*completion
)
5325 librados::IoCtx io_ctx
;
5327 int r
= rgw_init_ioctx(get_rados_handle(), get_zone_params().log_pool
, io_ctx
);
5331 utime_t
st(start_time
);
5332 utime_t
et(end_time
);
5334 ObjectWriteOperation op
;
5335 cls_log_trim(op
, st
, et
, from_marker
, to_marker
);
5338 r
= io_ctx
.operate(oid
, &op
);
5340 r
= io_ctx
.aio_operate(oid
, completion
, &op
);
5345 string
RGWRados::objexp_hint_get_shardname(int shard_num
)
5348 snprintf(buf
, sizeof(buf
), "%010u", (unsigned)shard_num
);
5350 string
objname("obj_delete_at_hint.");
5351 return objname
+ buf
;
5354 int RGWRados::objexp_key_shard(const rgw_obj_index_key
& key
)
5356 string obj_key
= key
.name
+ key
.instance
;
5357 int num_shards
= cct
->_conf
->rgw_objexp_hints_num_shards
;
5358 uint32_t sid
= ceph_str_hash_linux(obj_key
.c_str(), obj_key
.size());
5359 uint32_t sid2
= sid
^ ((sid
& 0xFF) << 24);
5360 sid
= rgw_shards_mod(sid2
, num_shards
);
5364 static string
objexp_hint_get_keyext(const string
& tenant_name
,
5365 const string
& bucket_name
,
5366 const string
& bucket_id
,
5367 const rgw_obj_key
& obj_key
)
5369 return tenant_name
+ (tenant_name
.empty() ? "" : ":") + bucket_name
+ ":" + bucket_id
+
5370 ":" + obj_key
.name
+ ":" + obj_key
.instance
;
5373 int RGWRados::objexp_hint_add(const ceph::real_time
& delete_at
,
5374 const string
& tenant_name
,
5375 const string
& bucket_name
,
5376 const string
& bucket_id
,
5377 const rgw_obj_index_key
& obj_key
)
5379 const string keyext
= objexp_hint_get_keyext(tenant_name
, bucket_name
,
5380 bucket_id
, obj_key
);
5381 objexp_hint_entry he
= {
5382 .tenant
= tenant_name
,
5383 .bucket_name
= bucket_name
,
5384 .bucket_id
= bucket_id
,
5386 .exp_time
= delete_at
};
5389 ObjectWriteOperation op
;
5390 cls_timeindex_add(op
, utime_t(delete_at
), keyext
, hebl
);
5392 string shard_name
= objexp_hint_get_shardname(objexp_key_shard(obj_key
));
5393 return objexp_pool_ctx
.operate(shard_name
, &op
);
5396 void RGWRados::objexp_get_shard(int shard_num
,
5397 string
& shard
) /* out */
5399 shard
= objexp_hint_get_shardname(shard_num
);
5402 int RGWRados::objexp_hint_list(const string
& oid
,
5403 const ceph::real_time
& start_time
,
5404 const ceph::real_time
& end_time
,
5405 const int max_entries
,
5406 const string
& marker
,
5407 list
<cls_timeindex_entry
>& entries
, /* out */
5408 string
*out_marker
, /* out */
5409 bool *truncated
) /* out */
5411 librados::ObjectReadOperation op
;
5412 cls_timeindex_list(op
, utime_t(start_time
), utime_t(end_time
), marker
, max_entries
, entries
,
5413 out_marker
, truncated
);
5416 int ret
= objexp_pool_ctx
.operate(oid
, &op
, &obl
);
5418 if ((ret
< 0 ) && (ret
!= -ENOENT
)) {
5422 if ((ret
== -ENOENT
) && truncated
) {
5429 int RGWRados::objexp_hint_parse(cls_timeindex_entry
&ti_entry
, /* in */
5430 objexp_hint_entry
& hint_entry
) /* out */
5433 bufferlist::iterator iter
= ti_entry
.value
.begin();
5434 ::decode(hint_entry
, iter
);
5435 } catch (buffer::error
& err
) {
5436 ldout(cct
, 0) << "ERROR: couldn't decode avail_pools" << dendl
;
5442 int RGWRados::objexp_hint_trim(const string
& oid
,
5443 const ceph::real_time
& start_time
,
5444 const ceph::real_time
& end_time
,
5445 const string
& from_marker
,
5446 const string
& to_marker
)
5448 int ret
= cls_timeindex_trim(objexp_pool_ctx
, oid
, utime_t(start_time
), utime_t(end_time
),
5449 from_marker
, to_marker
);
5450 if ((ret
< 0 ) && (ret
!= -ENOENT
)) {
5457 int RGWRados::lock_exclusive(rgw_pool
& pool
, const string
& oid
, timespan
& duration
,
5458 string
& zone_id
, string
& owner_id
) {
5459 librados::IoCtx io_ctx
;
5461 int r
= rgw_init_ioctx(get_rados_handle(), pool
, io_ctx
);
5465 uint64_t msec
= std::chrono::duration_cast
<std::chrono::milliseconds
>(duration
).count();
5466 utime_t
ut(msec
/ 1000, msec
% 1000);
5468 rados::cls::lock::Lock
l(log_lock_name
);
5470 l
.set_cookie(owner_id
);
5474 return l
.lock_exclusive(&io_ctx
, oid
);
5477 int RGWRados::unlock(rgw_pool
& pool
, const string
& oid
, string
& zone_id
, string
& owner_id
) {
5478 librados::IoCtx io_ctx
;
5480 int r
= rgw_init_ioctx(get_rados_handle(), pool
, io_ctx
);
5485 rados::cls::lock::Lock
l(log_lock_name
);
5487 l
.set_cookie(owner_id
);
5489 return l
.unlock(&io_ctx
, oid
);
5492 int RGWRados::decode_policy(bufferlist
& bl
, ACLOwner
*owner
)
5494 bufferlist::iterator i
= bl
.begin();
5495 RGWAccessControlPolicy
policy(cct
);
5497 policy
.decode_owner(i
);
5498 } catch (buffer::error
& err
) {
5499 ldout(cct
, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl
;
5502 *owner
= policy
.get_owner();
5506 int rgw_policy_from_attrset(CephContext
*cct
, map
<string
, bufferlist
>& attrset
, RGWAccessControlPolicy
*policy
)
5508 map
<string
, bufferlist
>::iterator aiter
= attrset
.find(RGW_ATTR_ACL
);
5509 if (aiter
== attrset
.end())
5512 bufferlist
& bl
= aiter
->second
;
5513 bufferlist::iterator iter
= bl
.begin();
5515 policy
->decode(iter
);
5516 } catch (buffer::error
& err
) {
5517 ldout(cct
, 0) << "ERROR: could not decode policy, caught buffer::error" << dendl
;
5520 if (cct
->_conf
->subsys
.should_gather(ceph_subsys_rgw
, 15)) {
5521 RGWAccessControlPolicy_S3
*s3policy
= static_cast<RGWAccessControlPolicy_S3
*>(policy
);
5522 ldout(cct
, 15) << __func__
<< " Read AccessControlPolicy";
5523 s3policy
->to_xml(*_dout
);
5530 int RGWRados::Bucket::update_bucket_id(const string
& new_bucket_id
)
5532 rgw_bucket bucket
= bucket_info
.bucket
;
5533 bucket
.update_bucket_id(new_bucket_id
);
5535 RGWObjectCtx
obj_ctx(store
);
5537 int ret
= store
->get_bucket_instance_info(obj_ctx
, bucket
, bucket_info
, nullptr, nullptr);
5546 * get listing of the objects in a bucket.
5548 * max: maximum number of results to return
5549 * bucket: bucket to list contents of
5550 * prefix: only return results that match this prefix
5551 * delim: do not include results that match this string.
5552 * Any skipped results will have the matching portion of their name
5553 * inserted in common_prefixes with a "true" mark.
5554 * marker: if filled in, begin the listing with this object.
5555 * end_marker: if filled in, end the listing with this object.
5556 * result: the objects are put in here.
5557 * common_prefixes: if delim is filled in, any matching prefixes are placed here.
5558 * is_truncated: if number of objects in the bucket is bigger than max, then truncated.
5560 int RGWRados::Bucket::List::list_objects(int64_t max
,
5561 vector
<rgw_bucket_dir_entry
> *result
,
5562 map
<string
, bool> *common_prefixes
,
5565 RGWRados
*store
= target
->get_store();
5566 CephContext
*cct
= store
->ctx();
5567 int shard_id
= target
->get_shard_id();
5570 bool truncated
= true;
5571 int read_ahead
= std::max(cct
->_conf
->rgw_list_bucket_min_readahead
,max
);
5575 rgw_obj_key
marker_obj(params
.marker
.name
, params
.marker
.instance
, params
.ns
);
5576 rgw_obj_index_key cur_marker
;
5577 marker_obj
.get_index_key(&cur_marker
);
5579 rgw_obj_key
end_marker_obj(params
.end_marker
.name
, params
.end_marker
.instance
,
5581 rgw_obj_index_key cur_end_marker
;
5582 end_marker_obj
.get_index_key(&cur_end_marker
);
5583 const bool cur_end_marker_valid
= !params
.end_marker
.empty();
5585 rgw_obj_key
prefix_obj(params
.prefix
);
5586 prefix_obj
.ns
= params
.ns
;
5587 string cur_prefix
= prefix_obj
.get_index_key_name();
5589 string bigger_than_delim
;
5591 if (!params
.delim
.empty()) {
5592 unsigned long val
= decode_utf8((unsigned char *)params
.delim
.c_str(), params
.delim
.size());
5593 char buf
[params
.delim
.size() + 16];
5594 int r
= encode_utf8(val
+ 1, (unsigned char *)buf
);
5596 ldout(cct
,0) << "ERROR: encode_utf8() failed" << dendl
;
5601 bigger_than_delim
= buf
;
5603 /* if marker points at a common prefix, fast forward it into its upperbound string */
5604 int delim_pos
= cur_marker
.name
.find(params
.delim
, cur_prefix
.size());
5605 if (delim_pos
>= 0) {
5606 string s
= cur_marker
.name
.substr(0, delim_pos
);
5607 s
.append(bigger_than_delim
);
5612 string skip_after_delim
;
5613 while (truncated
&& count
<= max
) {
5614 if (skip_after_delim
> cur_marker
.name
) {
5615 cur_marker
= skip_after_delim
;
5616 ldout(cct
, 20) << "setting cur_marker=" << cur_marker
.name
<< "[" << cur_marker
.instance
<< "]" << dendl
;
5618 std::map
<string
, rgw_bucket_dir_entry
> ent_map
;
5619 int r
= store
->cls_bucket_list(target
->get_bucket_info(), shard_id
, cur_marker
, cur_prefix
,
5620 read_ahead
+ 1 - count
, params
.list_versions
, ent_map
,
5621 &truncated
, &cur_marker
);
5625 std::map
<string
, rgw_bucket_dir_entry
>::iterator eiter
;
5626 for (eiter
= ent_map
.begin(); eiter
!= ent_map
.end(); ++eiter
) {
5627 rgw_bucket_dir_entry
& entry
= eiter
->second
;
5628 rgw_obj_index_key index_key
= entry
.key
;
5630 rgw_obj_key
obj(index_key
);
5632 /* note that parse_raw_oid() here will not set the correct object's instance, as
5633 * rgw_obj_index_key encodes that separately. We don't need to set the instance because it's
5634 * not needed for the checks here and we end up using the raw entry for the return vector
5636 bool valid
= rgw_obj_key::parse_raw_oid(index_key
.name
, &obj
);
5638 ldout(cct
, 0) << "ERROR: could not parse object name: " << obj
.name
<< dendl
;
5641 bool check_ns
= (obj
.ns
== params
.ns
);
5642 if (!params
.list_versions
&& !entry
.is_visible()) {
5646 if (params
.enforce_ns
&& !check_ns
) {
5647 if (!params
.ns
.empty()) {
5648 /* we've iterated past the namespace we're searching -- done now */
5653 /* we're not looking at the namespace this object is in, next! */
5657 if (cur_end_marker_valid
&& cur_end_marker
<= index_key
) {
5663 params
.marker
= index_key
;
5664 next_marker
= index_key
;
5667 if (params
.filter
&& !params
.filter
->filter(obj
.name
, index_key
.name
))
5670 if (params
.prefix
.size() && (obj
.name
.compare(0, params
.prefix
.size(), params
.prefix
) != 0))
5673 if (!params
.delim
.empty()) {
5674 int delim_pos
= obj
.name
.find(params
.delim
, params
.prefix
.size());
5676 if (delim_pos
>= 0) {
5677 string prefix_key
= obj
.name
.substr(0, delim_pos
+ 1);
5679 if (common_prefixes
&&
5680 common_prefixes
->find(prefix_key
) == common_prefixes
->end()) {
5685 next_marker
= prefix_key
;
5686 (*common_prefixes
)[prefix_key
] = true;
5688 int marker_delim_pos
= cur_marker
.name
.find(params
.delim
, cur_prefix
.size());
5690 skip_after_delim
= cur_marker
.name
.substr(0, marker_delim_pos
);
5691 skip_after_delim
.append(bigger_than_delim
);
5693 ldout(cct
, 20) << "skip_after_delim=" << skip_after_delim
<< dendl
;
5707 result
->emplace_back(std::move(entry
));
5711 // Either the back-end telling us truncated, or we don't consume all
5712 // items returned per the amount caller request
5713 truncated
= (truncated
|| eiter
!= ent_map
.end());
5718 *is_truncated
= truncated
;
5724 * create a rados pool, associated meta info
5725 * returns 0 on success, -ERR# otherwise.
5727 int RGWRados::create_pool(const rgw_pool
& pool
)
5731 librados::Rados
*rad
= get_rados_handle();
5732 ret
= rad
->pool_create(pool
.name
.c_str(), 0);
5735 else if (ret
== -ERANGE
) {
5738 << " ERROR: librados::Rados::pool_create returned " << cpp_strerror(-ret
)
5739 << " (this can be due to a pool or placement group misconfiguration, e.g., pg_num < pgp_num)"
5745 librados::IoCtx io_ctx
;
5746 ret
= rad
->ioctx_create(pool
.name
.c_str(), io_ctx
);
5750 ret
= io_ctx
.application_enable(pg_pool_t::APPLICATION_NAME_RGW
, false);
5751 if (ret
< 0 && ret
!= -EOPNOTSUPP
)
5756 int RGWRados::init_bucket_index(RGWBucketInfo
& bucket_info
, int num_shards
)
5758 librados::IoCtx index_ctx
; // context for new bucket
5760 string dir_oid
= dir_oid_prefix
;
5761 int r
= open_bucket_index_ctx(bucket_info
, index_ctx
);
5766 dir_oid
.append(bucket_info
.bucket
.bucket_id
);
5768 map
<int, string
> bucket_objs
;
5769 get_bucket_index_objects(dir_oid
, num_shards
, bucket_objs
);
5771 return CLSRGWIssueBucketIndexInit(index_ctx
, bucket_objs
, cct
->_conf
->rgw_bucket_index_max_aio
)();
5774 void RGWRados::create_bucket_id(string
*bucket_id
)
5776 uint64_t iid
= instance_id();
5777 uint64_t bid
= next_bucket_id();
5778 char buf
[get_zone_params().get_id().size() + 48];
5779 snprintf(buf
, sizeof(buf
), "%s.%llu.%llu", get_zone_params().get_id().c_str(), (long long)iid
, (long long)bid
);
5783 int RGWRados::create_bucket(RGWUserInfo
& owner
, rgw_bucket
& bucket
,
5784 const string
& zonegroup_id
,
5785 const string
& placement_rule
,
5786 const string
& swift_ver_location
,
5787 const RGWQuotaInfo
* pquota_info
,
5788 map
<std::string
, bufferlist
>& attrs
,
5789 RGWBucketInfo
& info
,
5791 obj_version
*pep_objv
,
5792 real_time creation_time
,
5793 rgw_bucket
*pmaster_bucket
,
5794 uint32_t *pmaster_num_shards
,
5797 #define MAX_CREATE_RETRIES 20 /* need to bound retries */
5798 string selected_placement_rule_name
;
5799 RGWZonePlacementInfo rule_info
;
5801 for (int i
= 0; i
< MAX_CREATE_RETRIES
; i
++) {
5803 ret
= select_bucket_placement(owner
, zonegroup_id
, placement_rule
,
5804 &selected_placement_rule_name
, &rule_info
);
5808 if (!pmaster_bucket
) {
5809 create_bucket_id(&bucket
.marker
);
5810 bucket
.bucket_id
= bucket
.marker
;
5812 bucket
.marker
= pmaster_bucket
->marker
;
5813 bucket
.bucket_id
= pmaster_bucket
->bucket_id
;
5816 RGWObjVersionTracker
& objv_tracker
= info
.objv_tracker
;
5819 objv_tracker
.write_version
= *pobjv
;
5821 objv_tracker
.generate_new_write_ver(cct
);
5824 info
.bucket
= bucket
;
5825 info
.owner
= owner
.user_id
;
5826 info
.zonegroup
= zonegroup_id
;
5827 info
.placement_rule
= selected_placement_rule_name
;
5828 info
.index_type
= rule_info
.index_type
;
5829 info
.swift_ver_location
= swift_ver_location
;
5830 info
.swift_versioning
= (!swift_ver_location
.empty());
5831 if (pmaster_num_shards
) {
5832 info
.num_shards
= *pmaster_num_shards
;
5834 info
.num_shards
= bucket_index_max_shards
;
5836 info
.bucket_index_shard_hash_type
= RGWBucketInfo::MOD
;
5837 info
.requester_pays
= false;
5838 if (real_clock::is_zero(creation_time
)) {
5839 info
.creation_time
= ceph::real_clock::now();
5841 info
.creation_time
= creation_time
;
5844 info
.quota
= *pquota_info
;
5847 int r
= init_bucket_index(info
, info
.num_shards
);
5852 ret
= put_linked_bucket_info(info
, exclusive
, ceph::real_time(), pep_objv
, &attrs
, true);
5853 if (ret
== -EEXIST
) {
5854 librados::IoCtx index_ctx
;
5855 map
<int, string
> bucket_objs
;
5856 int r
= open_bucket_index(info
, index_ctx
, bucket_objs
);
5860 /* we need to reread the info and return it, caller will have a use for it */
5861 RGWObjVersionTracker instance_ver
= info
.objv_tracker
;
5862 info
.objv_tracker
.clear();
5863 RGWObjectCtx
obj_ctx(this);
5864 r
= get_bucket_info(obj_ctx
, bucket
.tenant
, bucket
.name
, info
, NULL
, NULL
);
5869 ldout(cct
, 0) << "get_bucket_info returned " << r
<< dendl
;
5873 /* only remove it if it's a different bucket instance */
5874 if (info
.bucket
.bucket_id
!= bucket
.bucket_id
) {
5875 /* remove bucket meta instance */
5876 string entry
= bucket
.get_key();
5877 r
= rgw_bucket_instance_remove_entry(this, entry
, &instance_ver
);
5881 map
<int, string
>::const_iterator biter
;
5882 for (biter
= bucket_objs
.begin(); biter
!= bucket_objs
.end(); ++biter
) {
5883 // Do best effort removal
5884 index_ctx
.remove(biter
->second
);
5887 /* ret == -ENOENT here */
5892 /* this is highly unlikely */
5893 ldout(cct
, 0) << "ERROR: could not create bucket, continuously raced with bucket creation and removal" << dendl
;
5897 int RGWRados::select_new_bucket_location(RGWUserInfo
& user_info
, const string
& zonegroup_id
, const string
& request_rule
,
5898 string
*pselected_rule_name
, RGWZonePlacementInfo
*rule_info
)
5901 /* first check that zonegroup exists within current period. */
5902 RGWZoneGroup zonegroup
;
5903 int ret
= get_zonegroup(zonegroup_id
, zonegroup
);
5905 ldout(cct
, 0) << "could not find zonegroup " << zonegroup_id
<< " in current period" << dendl
;
5909 /* find placement rule. Hierarchy: request rule > user default rule > zonegroup default rule */
5910 std::map
<std::string
, RGWZoneGroupPlacementTarget
>::const_iterator titer
;
5912 if (!request_rule
.empty()) {
5913 titer
= zonegroup
.placement_targets
.find(request_rule
);
5914 if (titer
== zonegroup
.placement_targets
.end()) {
5915 ldout(cct
, 0) << "could not find requested placement id " << request_rule
5916 << " within zonegroup " << dendl
;
5917 return -ERR_INVALID_LOCATION_CONSTRAINT
;
5919 } else if (!user_info
.default_placement
.empty()) {
5920 titer
= zonegroup
.placement_targets
.find(user_info
.default_placement
);
5921 if (titer
== zonegroup
.placement_targets
.end()) {
5922 ldout(cct
, 0) << "could not find user default placement id " << user_info
.default_placement
5923 << " within zonegroup " << dendl
;
5924 return -ERR_INVALID_LOCATION_CONSTRAINT
;
5927 if (zonegroup
.default_placement
.empty()) { // zonegroup default rule as fallback, it should not be empty.
5928 ldout(cct
, 0) << "misconfiguration, zonegroup default placement id should not be empty." << dendl
;
5929 return -ERR_ZONEGROUP_DEFAULT_PLACEMENT_MISCONFIGURATION
;
5931 titer
= zonegroup
.placement_targets
.find(zonegroup
.default_placement
);
5932 if (titer
== zonegroup
.placement_targets
.end()) {
5933 ldout(cct
, 0) << "could not find zonegroup default placement id " << zonegroup
.default_placement
5934 << " within zonegroup " << dendl
;
5935 return -ERR_INVALID_LOCATION_CONSTRAINT
;
5940 /* now check tag for the rule, whether user is permitted to use rule */
5941 const auto& target_rule
= titer
->second
;
5942 if (!target_rule
.user_permitted(user_info
.placement_tags
)) {
5943 ldout(cct
, 0) << "user not permitted to use placement rule " << titer
->first
<< dendl
;
5947 if (pselected_rule_name
)
5948 *pselected_rule_name
= titer
->first
;
5950 return select_bucket_location_by_rule(titer
->first
, rule_info
);
5953 int RGWRados::select_bucket_location_by_rule(const string
& location_rule
, RGWZonePlacementInfo
*rule_info
)
5955 if (location_rule
.empty()) {
5956 /* we can only reach here if we're trying to set a bucket location from a bucket
5957 * created on a different zone, using a legacy / default pool configuration
5959 return select_legacy_bucket_placement(rule_info
);
5963 * make sure that zone has this rule configured. We're
5964 * checking it for the local zone, because that's where this bucket object is going to
5967 map
<string
, RGWZonePlacementInfo
>::iterator piter
= get_zone_params().placement_pools
.find(location_rule
);
5968 if (piter
== get_zone_params().placement_pools
.end()) {
5969 /* couldn't find, means we cannot really place data for this bucket in this zone */
5970 if (get_zonegroup().equals(zonegroup
.get_id())) {
5971 /* that's a configuration error, zone should have that rule, as we're within the requested
5975 /* oh, well, data is not going to be placed here, bucket object is just a placeholder */
5980 RGWZonePlacementInfo
& placement_info
= piter
->second
;
5983 *rule_info
= placement_info
;
5989 int RGWRados::select_bucket_placement(RGWUserInfo
& user_info
, const string
& zonegroup_id
, const string
& placement_rule
,
5990 string
*pselected_rule_name
, RGWZonePlacementInfo
*rule_info
)
5992 if (!get_zone_params().placement_pools
.empty()) {
5993 return select_new_bucket_location(user_info
, zonegroup_id
, placement_rule
,
5994 pselected_rule_name
, rule_info
);
5997 if (pselected_rule_name
) {
5998 pselected_rule_name
->clear();
6001 return select_legacy_bucket_placement(rule_info
);
6004 int RGWRados::select_legacy_bucket_placement(RGWZonePlacementInfo
*rule_info
)
6007 map
<string
, bufferlist
> m
;
6009 bool write_map
= false;
6011 rgw_raw_obj
obj(get_zone_params().domain_root
, avail_pools
);
6013 RGWObjectCtx
obj_ctx(this);
6014 int ret
= rgw_get_system_obj(this, obj_ctx
, get_zone_params().domain_root
, avail_pools
, map_bl
, NULL
, NULL
);
6020 bufferlist::iterator iter
= map_bl
.begin();
6022 } catch (buffer::error
& err
) {
6023 ldout(cct
, 0) << "ERROR: couldn't decode avail_pools" << dendl
;
6029 ret
= omap_get_all(obj
, header
, m
);
6034 if (ret
< 0 || m
.empty()) {
6035 vector
<rgw_pool
> pools
;
6036 string s
= string("default.") + default_storage_pool_suffix
;
6037 pools
.push_back(rgw_pool(s
));
6038 vector
<int> retcodes
;
6040 ret
= create_pools(pools
, retcodes
);
6043 ret
= omap_set(obj
, s
, bl
);
6051 ::encode(m
, new_bl
);
6052 ret
= put_system_obj_data(NULL
, obj
, new_bl
, -1, false);
6054 ldout(cct
, 0) << "WARNING: could not save avail pools map info ret=" << ret
<< dendl
;
6058 map
<string
, bufferlist
>::iterator miter
;
6061 for (miter
= m
.begin(); miter
!= m
.end(); ++miter
) {
6062 v
.push_back(miter
->first
);
6066 ret
= get_random_bytes((char *)&r
, sizeof(r
));
6070 int i
= r
% v
.size();
6074 pool_name
= miter
->first
;
6077 rule_info
->data_pool
= pool_name
;
6078 rule_info
->data_extra_pool
= pool_name
;
6079 rule_info
->index_pool
= pool_name
;
6080 rule_info
->index_type
= RGWBIType_Normal
;
6085 bool RGWRados::get_obj_data_pool(const string
& placement_rule
, const rgw_obj
& obj
, rgw_pool
*pool
)
6087 return rgw_get_obj_data_pool(zonegroup
, zone_params
, placement_rule
, obj
, pool
);
6090 bool RGWRados::obj_to_raw(const string
& placement_rule
, const rgw_obj
& obj
, rgw_raw_obj
*raw_obj
)
6092 get_obj_bucket_and_oid_loc(obj
, raw_obj
->oid
, raw_obj
->loc
);
6094 return get_obj_data_pool(placement_rule
, obj
, &raw_obj
->pool
);
6097 int RGWRados::update_placement_map()
6100 map
<string
, bufferlist
> m
;
6101 rgw_raw_obj
obj(get_zone_params().domain_root
, avail_pools
);
6102 int ret
= omap_get_all(obj
, header
, m
);
6107 ::encode(m
, new_bl
);
6108 ret
= put_system_obj_data(NULL
, obj
, new_bl
, -1, false);
6110 ldout(cct
, 0) << "WARNING: could not save avail pools map info ret=" << ret
<< dendl
;
6116 int RGWRados::add_bucket_placement(const rgw_pool
& new_pool
)
6118 librados::Rados
*rad
= get_rados_handle();
6119 int ret
= rad
->pool_lookup(new_pool
.name
.c_str());
6120 if (ret
< 0) // DNE, or something
6123 rgw_raw_obj
obj(get_zone_params().domain_root
, avail_pools
);
6124 bufferlist empty_bl
;
6125 ret
= omap_set(obj
, new_pool
.to_str(), empty_bl
);
6127 // don't care about return value
6128 update_placement_map();
6133 int RGWRados::remove_bucket_placement(const rgw_pool
& old_pool
)
6135 rgw_raw_obj
obj(get_zone_params().domain_root
, avail_pools
);
6136 int ret
= omap_del(obj
, old_pool
.to_str());
6138 // don't care about return value
6139 update_placement_map();
6144 int RGWRados::list_placement_set(set
<rgw_pool
>& names
)
6147 map
<string
, bufferlist
> m
;
6149 rgw_raw_obj
obj(get_zone_params().domain_root
, avail_pools
);
6150 int ret
= omap_get_all(obj
, header
, m
);
6155 map
<string
, bufferlist
>::iterator miter
;
6156 for (miter
= m
.begin(); miter
!= m
.end(); ++miter
) {
6157 names
.insert(rgw_pool(miter
->first
));
6160 return names
.size();
6163 int RGWRados::create_pools(vector
<rgw_pool
>& pools
, vector
<int>& retcodes
)
6165 vector
<librados::PoolAsyncCompletion
*> completions
;
6168 librados::Rados
*rad
= get_rados_handle();
6169 for (auto iter
= pools
.begin(); iter
!= pools
.end(); ++iter
) {
6170 librados::PoolAsyncCompletion
*c
= librados::Rados::pool_async_create_completion();
6171 completions
.push_back(c
);
6172 rgw_pool
& pool
= *iter
;
6173 int ret
= rad
->pool_create_async(pool
.name
.c_str(), c
);
6174 rets
.push_back(ret
);
6177 vector
<int>::iterator riter
;
6178 vector
<librados::PoolAsyncCompletion
*>::iterator citer
;
6181 assert(rets
.size() == completions
.size());
6182 for (riter
= rets
.begin(), citer
= completions
.begin(); riter
!= rets
.end(); ++riter
, ++citer
) {
6184 PoolAsyncCompletion
*c
= *citer
;
6187 r
= c
->get_return_value();
6189 ldout(cct
, 0) << "WARNING: async pool_create returned " << r
<< dendl
;
6194 retcodes
.push_back(r
);
6200 std::vector
<librados::IoCtx
> io_ctxs
;
6202 for (auto pool
: pools
) {
6203 io_ctxs
.emplace_back();
6204 int ret
= rad
->ioctx_create(pool
.name
.c_str(), io_ctxs
.back());
6206 ldout(cct
, 0) << "WARNING: ioctx_create returned " << ret
<< dendl
;
6209 retcodes
.push_back(ret
);
6215 completions
.clear();
6216 for (auto &io_ctx
: io_ctxs
) {
6217 librados::PoolAsyncCompletion
*c
=
6218 librados::Rados::pool_async_create_completion();
6219 completions
.push_back(c
);
6220 int ret
= io_ctx
.application_enable_async(pg_pool_t::APPLICATION_NAME_RGW
,
6226 for (auto c
: completions
) {
6228 int ret
= c
->get_return_value();
6229 if (ret
== -EOPNOTSUPP
) {
6231 } else if (ret
< 0) {
6232 ldout(cct
, 0) << "WARNING: async application_enable returned " << ret
6237 retcodes
.push_back(ret
);
6242 int RGWRados::get_obj_head_ioctx(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, librados::IoCtx
*ioctx
)
6245 get_obj_bucket_and_oid_loc(obj
, oid
, key
);
6248 if (!get_obj_data_pool(bucket_info
.placement_rule
, obj
, &pool
)) {
6249 ldout(cct
, 0) << "ERROR: cannot get data pool for obj=" << obj
<< ", probably misconfiguration" << dendl
;
6253 int r
= open_pool_ctx(pool
, *ioctx
);
6258 ioctx
->locator_set_key(key
);
6263 int RGWRados::get_obj_head_ref(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, rgw_rados_ref
*ref
)
6265 get_obj_bucket_and_oid_loc(obj
, ref
->oid
, ref
->key
);
6268 if (!get_obj_data_pool(bucket_info
.placement_rule
, obj
, &pool
)) {
6269 ldout(cct
, 0) << "ERROR: cannot get data pool for obj=" << obj
<< ", probably misconfiguration" << dendl
;
6273 int r
= open_pool_ctx(pool
, ref
->ioctx
);
6278 ref
->ioctx
.locator_set_key(ref
->key
);
6283 int RGWRados::get_raw_obj_ref(const rgw_raw_obj
& obj
, rgw_rados_ref
*ref
)
6290 if (ref
->oid
.empty()) {
6291 ref
->oid
= obj
.pool
.to_str();
6292 ref
->pool
= get_zone_params().domain_root
;
6294 ref
->pool
= obj
.pool
;
6296 r
= open_pool_ctx(ref
->pool
, ref
->ioctx
);
6300 ref
->ioctx
.locator_set_key(ref
->key
);
6305 int RGWRados::get_system_obj_ref(const rgw_raw_obj
& obj
, rgw_rados_ref
*ref
)
6307 return get_raw_obj_ref(obj
, ref
);
6311 * fixes an issue where head objects were supposed to have a locator created, but ended
6314 int RGWRados::fix_head_obj_locator(const RGWBucketInfo
& bucket_info
, bool copy_obj
, bool remove_bad
, rgw_obj_key
& key
)
6316 const rgw_bucket
& bucket
= bucket_info
.bucket
;
6320 rgw_obj
obj(bucket
, key
);
6322 get_obj_bucket_and_oid_loc(obj
, oid
, locator
);
6324 if (locator
.empty()) {
6325 ldout(cct
, 20) << "object does not have a locator, nothing to fix" << dendl
;
6329 librados::IoCtx ioctx
;
6331 int ret
= get_obj_head_ioctx(bucket_info
, obj
, &ioctx
);
6333 cerr
<< "ERROR: get_obj_head_ioctx() returned ret=" << ret
<< std::endl
;
6336 ioctx
.locator_set_key(string()); /* override locator for this object, use empty locator */
6341 struct timespec mtime_ts
;
6342 map
<string
, bufferlist
> attrs
;
6343 librados::ObjectReadOperation op
;
6344 op
.getxattrs(&attrs
, NULL
);
6345 op
.stat2(&size
, &mtime_ts
, NULL
);
6346 #define HEAD_SIZE 512 * 1024
6347 op
.read(0, HEAD_SIZE
, &data
, NULL
);
6349 ret
= ioctx
.operate(oid
, &op
, NULL
);
6351 lderr(cct
) << "ERROR: ioctx.operate(oid=" << oid
<< ") returned ret=" << ret
<< dendl
;
6355 if (size
> HEAD_SIZE
) {
6356 lderr(cct
) << "ERROR: returned object size (" << size
<< ") > HEAD_SIZE (" << HEAD_SIZE
<< ")" << dendl
;
6360 if (size
!= data
.length()) {
6361 lderr(cct
) << "ERROR: returned object size (" << size
<< ") != data.length() (" << data
.length() << ")" << dendl
;
6366 librados::ObjectWriteOperation wop
;
6368 wop
.mtime2(&mtime_ts
);
6370 map
<string
, bufferlist
>::iterator iter
;
6371 for (iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
6372 wop
.setxattr(iter
->first
.c_str(), iter
->second
);
6377 ioctx
.locator_set_key(locator
);
6378 ioctx
.operate(oid
, &wop
);
6382 ioctx
.locator_set_key(string());
6384 ret
= ioctx
.remove(oid
);
6386 lderr(cct
) << "ERROR: failed to remove original bad object" << dendl
;
6394 int RGWRados::move_rados_obj(librados::IoCtx
& src_ioctx
,
6395 const string
& src_oid
, const string
& src_locator
,
6396 librados::IoCtx
& dst_ioctx
,
6397 const string
& dst_oid
, const string
& dst_locator
)
6400 #define COPY_BUF_SIZE (4 * 1024 * 1024)
6402 uint64_t chunk_size
= COPY_BUF_SIZE
;
6406 struct timespec mtime_ts
;
6409 if (src_oid
== dst_oid
&& src_locator
== dst_locator
) {
6413 src_ioctx
.locator_set_key(src_locator
);
6414 dst_ioctx
.locator_set_key(dst_locator
);
6418 ObjectReadOperation rop
;
6419 ObjectWriteOperation wop
;
6422 rop
.stat2(&size
, &mtime_ts
, NULL
);
6423 mtime
= real_clock::from_timespec(mtime_ts
);
6425 rop
.read(ofs
, chunk_size
, &data
, NULL
);
6426 ret
= src_ioctx
.operate(src_oid
, &rop
, NULL
);
6431 if (data
.length() == 0) {
6436 wop
.create(true); /* make it exclusive */
6437 wop
.mtime2(&mtime_ts
);
6438 mtime
= real_clock::from_timespec(mtime_ts
);
6440 wop
.write(ofs
, data
);
6441 ret
= dst_ioctx
.operate(dst_oid
, &wop
);
6442 ofs
+= data
.length();
6443 done
= data
.length() != chunk_size
;
6447 lderr(cct
) << "ERROR: " << __func__
<< ": copying " << src_oid
<< " -> " << dst_oid
6448 << ": expected " << size
<< " bytes to copy, ended up with " << ofs
<< dendl
;
6453 src_ioctx
.remove(src_oid
);
6458 lderr(cct
) << "ERROR: failed to copy " << src_oid
<< " -> " << dst_oid
<< dendl
;
6463 * fixes an issue where head objects were supposed to have a locator created, but ended
6466 int RGWRados::fix_tail_obj_locator(const RGWBucketInfo
& bucket_info
, rgw_obj_key
& key
, bool fix
, bool *need_fix
)
6468 const rgw_bucket
& bucket
= bucket_info
.bucket
;
6469 rgw_obj
obj(bucket
, key
);
6476 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
6481 RGWObjState
*astate
= NULL
;
6482 RGWObjectCtx
rctx(this);
6483 r
= get_obj_state(&rctx
, bucket_info
, obj
, &astate
, false);
6487 if (astate
->has_manifest
) {
6488 RGWObjManifest::obj_iterator miter
;
6489 RGWObjManifest
& manifest
= astate
->manifest
;
6490 for (miter
= manifest
.obj_begin(); miter
!= manifest
.obj_end(); ++miter
) {
6491 rgw_raw_obj raw_loc
= miter
.get_location().get_raw_obj(this);
6496 rgw_raw_obj_to_obj(manifest
.get_tail_placement().bucket
, raw_loc
, &loc
);
6498 if (loc
.key
.ns
.empty()) {
6499 /* continue, we're only interested in tail objects */
6503 get_obj_bucket_and_oid_loc(loc
, oid
, locator
);
6504 ref
.ioctx
.locator_set_key(locator
);
6506 ldout(cct
, 20) << __func__
<< ": key=" << key
<< " oid=" << oid
<< " locator=" << locator
<< dendl
;
6508 r
= ref
.ioctx
.stat(oid
, NULL
, NULL
);
6514 prepend_bucket_marker(bucket
, loc
.key
.name
, bad_loc
);
6516 /* create a new ioctx with the bad locator */
6517 librados::IoCtx src_ioctx
;
6518 src_ioctx
.dup(ref
.ioctx
);
6519 src_ioctx
.locator_set_key(bad_loc
);
6521 r
= src_ioctx
.stat(oid
, NULL
, NULL
);
6523 /* cannot find a broken part */
6526 ldout(cct
, 20) << __func__
<< ": found bad object part: " << loc
<< dendl
;
6531 r
= move_rados_obj(src_ioctx
, oid
, bad_loc
, ref
.ioctx
, oid
, locator
);
6533 lderr(cct
) << "ERROR: copy_rados_obj() on oid=" << oid
<< " returned r=" << r
<< dendl
;
6542 int RGWRados::BucketShard::init(const rgw_bucket
& _bucket
, const rgw_obj
& obj
)
6546 RGWObjectCtx
obj_ctx(store
);
6548 RGWBucketInfo bucket_info
;
6549 int ret
= store
->get_bucket_instance_info(obj_ctx
, bucket
, bucket_info
, NULL
, NULL
);
6554 ret
= store
->open_bucket_index_shard(bucket_info
, index_ctx
, obj
.get_hash_object(), &bucket_obj
, &shard_id
);
6556 ldout(store
->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret
<< dendl
;
6559 ldout(store
->ctx(), 20) << " bucket index object: " << bucket_obj
<< dendl
;
6564 int RGWRados::BucketShard::init(const rgw_bucket
& _bucket
, int sid
)
6569 RGWObjectCtx
obj_ctx(store
);
6571 RGWBucketInfo bucket_info
;
6572 int ret
= store
->get_bucket_instance_info(obj_ctx
, bucket
, bucket_info
, NULL
, NULL
);
6577 ret
= store
->open_bucket_index_shard(bucket_info
, index_ctx
, shard_id
, &bucket_obj
);
6579 ldout(store
->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret
<< dendl
;
6582 ldout(store
->ctx(), 20) << " bucket index object: " << bucket_obj
<< dendl
;
6587 int RGWRados::BucketShard::init(const RGWBucketInfo
& bucket_info
, int sid
)
6589 bucket
= bucket_info
.bucket
;
6592 int ret
= store
->open_bucket_index_shard(bucket_info
, index_ctx
, shard_id
, &bucket_obj
);
6594 ldout(store
->ctx(), 0) << "ERROR: open_bucket_index_shard() returned ret=" << ret
<< dendl
;
6597 ldout(store
->ctx(), 20) << " bucket index object: " << bucket_obj
<< dendl
;
6603 /* Execute @handler on last item in bucket listing for bucket specified
6604 * in @bucket_info. @obj_prefix and @obj_delim narrow down the listing
6605 * to objects matching these criterias. */
6606 int RGWRados::on_last_entry_in_listing(RGWBucketInfo
& bucket_info
,
6607 const std::string
& obj_prefix
,
6608 const std::string
& obj_delim
,
6609 std::function
<int(const rgw_bucket_dir_entry
&)> handler
)
6611 RGWRados::Bucket
target(this, bucket_info
);
6612 RGWRados::Bucket::List
list_op(&target
);
6614 list_op
.params
.prefix
= obj_prefix
;
6615 list_op
.params
.delim
= obj_delim
;
6617 ldout(cct
, 20) << "iterating listing for bucket=" << bucket_info
.bucket
.name
6618 << ", obj_prefix=" << obj_prefix
6619 << ", obj_delim=" << obj_delim
6622 bool is_truncated
= false;
6624 boost::optional
<rgw_bucket_dir_entry
> last_entry
;
6625 /* We need to rewind to the last object in a listing. */
6627 /* List bucket entries in chunks. */
6628 static constexpr int MAX_LIST_OBJS
= 100;
6629 std::vector
<rgw_bucket_dir_entry
> entries(MAX_LIST_OBJS
);
6631 int ret
= list_op
.list_objects(MAX_LIST_OBJS
, &entries
, nullptr,
6635 } else if (!entries
.empty()) {
6636 last_entry
= entries
.back();
6638 } while (is_truncated
);
6641 return handler(*last_entry
);
6644 /* Empty listing - no items we can run handler on. */
6649 int RGWRados::swift_versioning_copy(RGWObjectCtx
& obj_ctx
,
6650 const rgw_user
& user
,
6651 RGWBucketInfo
& bucket_info
,
6654 if (! swift_versioning_enabled(bucket_info
)) {
6658 obj_ctx
.obj
.set_atomic(obj
);
6660 RGWObjState
* state
= nullptr;
6661 int r
= get_obj_state(&obj_ctx
, bucket_info
, obj
, &state
, false);
6666 if (!state
->exists
) {
6673 const string
& src_name
= obj
.get_oid();
6674 char buf
[src_name
.size() + 32];
6675 struct timespec ts
= ceph::real_clock::to_timespec(state
->mtime
);
6676 snprintf(buf
, sizeof(buf
), "%03x%s/%lld.%06ld", (int)src_name
.size(),
6677 src_name
.c_str(), (long long)ts
.tv_sec
, ts
.tv_nsec
/ 1000);
6679 RGWBucketInfo dest_bucket_info
;
6681 r
= get_bucket_info(obj_ctx
, bucket_info
.bucket
.tenant
, bucket_info
.swift_ver_location
, dest_bucket_info
, NULL
, NULL
);
6683 ldout(cct
, 10) << "failed to read dest bucket info: r=" << r
<< dendl
;
6685 return -ERR_PRECONDITION_FAILED
;
6690 if (dest_bucket_info
.owner
!= bucket_info
.owner
) {
6691 return -ERR_PRECONDITION_FAILED
;
6694 rgw_obj
dest_obj(dest_bucket_info
.bucket
, buf
);
6695 obj_ctx
.obj
.set_atomic(dest_obj
);
6699 r
= copy_obj(obj_ctx
,
6703 NULL
, /* req_info *info */
6709 NULL
, /* time_t *src_mtime */
6710 NULL
, /* time_t *mtime */
6711 NULL
, /* const time_t *mod_ptr */
6712 NULL
, /* const time_t *unmod_ptr */
6713 false, /* bool high_precision_time */
6714 NULL
, /* const char *if_match */
6715 NULL
, /* const char *if_nomatch */
6716 RGWRados::ATTRSMOD_NONE
,
6717 true, /* bool copy_if_newer */
6719 RGW_OBJ_CATEGORY_MAIN
,
6720 0, /* uint64_t olh_epoch */
6721 real_time(), /* time_t delete_at */
6722 NULL
, /* string *version_id */
6723 NULL
, /* string *ptag */
6724 NULL
, /* string *petag */
6725 NULL
, /* void (*progress_cb)(off_t, void *) */
6726 NULL
); /* void *progress_data */
6727 if (r
== -ECANCELED
|| r
== -ENOENT
) {
6728 /* Has already been overwritten, meaning another rgw process already
6736 int RGWRados::swift_versioning_restore(RGWObjectCtx
& obj_ctx
,
6737 const rgw_user
& user
,
6738 RGWBucketInfo
& bucket_info
,
6740 bool& restored
) /* out */
6742 if (! swift_versioning_enabled(bucket_info
)) {
6746 /* Bucket info of the bucket that stores previous versions of our object. */
6747 RGWBucketInfo archive_binfo
;
6749 int ret
= get_bucket_info(obj_ctx
, bucket_info
.bucket
.tenant
,
6750 bucket_info
.swift_ver_location
, archive_binfo
,
6756 /* Abort the operation if the bucket storing our archive belongs to someone
6757 * else. This is a limitation in comparison to Swift as we aren't taking ACLs
6758 * into consideration. For we can live with that.
6760 * TODO: delegate this check to un upper layer and compare with ACLs. */
6761 if (bucket_info
.owner
!= archive_binfo
.owner
) {
6765 /* This code will be executed on latest version of the object. */
6766 const auto handler
= [&](const rgw_bucket_dir_entry
& entry
) -> int {
6767 std::string no_client_id
;
6768 std::string no_op_id
;
6769 std::string no_zone
;
6771 /* We don't support object versioning of Swift API on those buckets that
6772 * are already versioned using the S3 mechanism. This affects also bucket
6773 * storing archived objects. Otherwise the delete operation would create
6774 * a deletion marker. */
6775 if (archive_binfo
.versioned()) {
6777 return -ERR_PRECONDITION_FAILED
;
6780 /* We are requesting ATTRSMOD_NONE so the attr attribute is perfectly
6781 * irrelevant and may be safely skipped. */
6782 std::map
<std::string
, ceph::bufferlist
> no_attrs
;
6784 rgw_obj
archive_obj(archive_binfo
.bucket
, entry
.key
);
6785 obj_ctx
.obj
.set_atomic(archive_obj
);
6786 obj_ctx
.obj
.set_atomic(obj
);
6788 int ret
= copy_obj(obj_ctx
,
6792 nullptr, /* req_info *info */
6795 archive_obj
, /* src obj */
6796 bucket_info
, /* dest bucket info */
6797 archive_binfo
, /* src bucket info */
6798 nullptr, /* time_t *src_mtime */
6799 nullptr, /* time_t *mtime */
6800 nullptr, /* const time_t *mod_ptr */
6801 nullptr, /* const time_t *unmod_ptr */
6802 false, /* bool high_precision_time */
6803 nullptr, /* const char *if_match */
6804 nullptr, /* const char *if_nomatch */
6805 RGWRados::ATTRSMOD_NONE
,
6806 true, /* bool copy_if_newer */
6808 RGW_OBJ_CATEGORY_MAIN
,
6809 0, /* uint64_t olh_epoch */
6810 real_time(), /* time_t delete_at */
6811 nullptr, /* string *version_id */
6812 nullptr, /* string *ptag */
6813 nullptr, /* string *petag */
6814 nullptr, /* void (*progress_cb)(off_t, void *) */
6815 nullptr); /* void *progress_data */
6816 if (ret
== -ECANCELED
|| ret
== -ENOENT
) {
6817 /* Has already been overwritten, meaning another rgw process already
6820 } else if (ret
< 0) {
6826 /* Need to remove the archived copy. */
6827 ret
= delete_obj(obj_ctx
, archive_binfo
, archive_obj
,
6828 archive_binfo
.versioning_status());
6833 const std::string
& obj_name
= obj
.get_oid();
6834 const auto prefix
= boost::str(boost::format("%03x%s") % obj_name
.size()
6837 return on_last_entry_in_listing(archive_binfo
, prefix
, std::string(),
6842 * Write/overwrite an object to the bucket storage.
6843 * bucket: the bucket to store the object in
6844 * obj: the object name/key
6845 * data: the object contents/value
6846 * size: the amount of data to write (data must be this long)
6847 * accounted_size: original size of data before compression, encryption
6848 * mtime: if non-NULL, writes the given mtime to the bucket storage
6849 * attrs: all the given attrs are written to bucket storage for the given object
6850 * exclusive: create object exclusively
6851 * Returns: 0 on success, -ERR# otherwise.
6853 int RGWRados::Object::Write::_do_write_meta(uint64_t size
, uint64_t accounted_size
,
6854 map
<string
, bufferlist
>& attrs
,
6855 bool assume_noent
, bool modify_tail
,
6858 RGWRados::Bucket::UpdateIndex
*index_op
= static_cast<RGWRados::Bucket::UpdateIndex
*>(_index_op
);
6859 RGWRados
*store
= target
->get_store();
6861 ObjectWriteOperation op
;
6864 int r
= target
->get_state(&state
, false, assume_noent
);
6868 rgw_obj
& obj
= target
->get_obj();
6870 if (obj
.get_oid().empty()) {
6871 ldout(store
->ctx(), 0) << "ERROR: " << __func__
<< "(): cannot write object with empty name" << dendl
;
6876 r
= store
->get_obj_head_ref(target
->get_bucket_info(), obj
, &ref
);
6880 bool is_olh
= state
->is_olh
;
6882 bool reset_obj
= (meta
.flags
& PUT_OBJ_CREATE
) != 0;
6884 const string
*ptag
= meta
.ptag
;
6885 if (!ptag
&& !index_op
->get_optag()->empty()) {
6886 ptag
= index_op
->get_optag();
6888 r
= target
->prepare_atomic_modification(op
, reset_obj
, ptag
, meta
.if_match
, meta
.if_nomatch
, false, modify_tail
);
6892 if (real_clock::is_zero(meta
.set_mtime
)) {
6893 meta
.set_mtime
= real_clock::now();
6896 if (state
->is_olh
) {
6897 op
.setxattr(RGW_ATTR_OLH_ID_TAG
, state
->olh_tag
);
6900 struct timespec mtime_ts
= real_clock::to_timespec(meta
.set_mtime
);
6901 op
.mtime2(&mtime_ts
);
6904 /* if we want to overwrite the data, we also want to overwrite the
6905 xattrs, so just remove the object */
6906 op
.write_full(*meta
.data
);
6910 string content_type
;
6913 map
<string
, bufferlist
>::iterator iter
;
6915 for (iter
= meta
.rmattrs
->begin(); iter
!= meta
.rmattrs
->end(); ++iter
) {
6916 const string
& name
= iter
->first
;
6917 op
.rmxattr(name
.c_str());
6921 if (meta
.manifest
) {
6922 /* remove existing manifest attr */
6923 iter
= attrs
.find(RGW_ATTR_MANIFEST
);
6924 if (iter
!= attrs
.end())
6928 ::encode(*meta
.manifest
, bl
);
6929 op
.setxattr(RGW_ATTR_MANIFEST
, bl
);
6932 for (iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
6933 const string
& name
= iter
->first
;
6934 bufferlist
& bl
= iter
->second
;
6939 op
.setxattr(name
.c_str(), bl
);
6941 if (name
.compare(RGW_ATTR_ETAG
) == 0) {
6943 } else if (name
.compare(RGW_ATTR_CONTENT_TYPE
) == 0) {
6944 content_type
= bl
.c_str();
6945 } else if (name
.compare(RGW_ATTR_ACL
) == 0) {
6949 if (attrs
.find(RGW_ATTR_PG_VER
) == attrs
.end()) {
6950 cls_rgw_obj_store_pg_ver(op
, RGW_ATTR_PG_VER
);
6953 if (attrs
.find(RGW_ATTR_SOURCE_ZONE
) == attrs
.end()) {
6955 ::encode(store
->get_zone_short_id(), bl
);
6956 op
.setxattr(RGW_ATTR_SOURCE_ZONE
, bl
);
6967 if (!reset_obj
) { //Multipart upload, it has immutable head.
6968 orig_exists
= false;
6971 orig_exists
= state
->exists
;
6972 orig_size
= state
->accounted_size
;
6975 bool versioned_target
= (meta
.olh_epoch
> 0 || !obj
.key
.instance
.empty());
6977 bool versioned_op
= (target
->versioning_enabled() || is_olh
|| versioned_target
);
6980 index_op
->set_bilog_flags(RGW_BILOG_FLAG_VERSIONED_OP
);
6983 if (!index_op
->is_prepared()) {
6984 r
= index_op
->prepare(CLS_RGW_OP_ADD
, &state
->write_tag
);
6989 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
6990 if (r
< 0) { /* we can expect to get -ECANCELED if object was replaced under,
6991 or -ENOENT if was removed, or -EEXIST if it did not exist
6992 before and now it does */
6993 if (r
== -EEXIST
&& assume_noent
) {
6994 target
->invalidate_state();
7000 epoch
= ref
.ioctx
.get_last_version();
7001 poolid
= ref
.ioctx
.get_id();
7003 r
= target
->complete_atomic_modification();
7005 ldout(store
->ctx(), 0) << "ERROR: complete_atomic_modification returned r=" << r
<< dendl
;
7008 r
= index_op
->complete(poolid
, epoch
, size
, accounted_size
,
7009 meta
.set_mtime
, etag
, content_type
, &acl_bl
,
7010 meta
.category
, meta
.remove_objs
, meta
.user_data
);
7015 *meta
.mtime
= meta
.set_mtime
;
7018 /* note that index_op was using state so we couldn't invalidate it earlier */
7019 target
->invalidate_state();
7023 r
= store
->set_olh(target
->get_ctx(), target
->get_bucket_info(), obj
, false, NULL
, meta
.olh_epoch
, real_time(), false, meta
.zones_trace
);
7029 if (!real_clock::is_zero(meta
.delete_at
)) {
7030 rgw_obj_index_key obj_key
;
7031 obj
.key
.get_index_key(&obj_key
);
7033 r
= store
->objexp_hint_add(meta
.delete_at
,
7034 obj
.bucket
.tenant
, obj
.bucket
.name
, obj
.bucket
.bucket_id
, obj_key
);
7036 ldout(store
->ctx(), 0) << "ERROR: objexp_hint_add() returned r=" << r
<< ", object will not get removed" << dendl
;
7037 /* ignoring error, nothing we can do at this point */
7040 meta
.canceled
= false;
7042 /* update quota cache */
7043 if (meta
.completeMultipart
){
7044 store
->quota_handler
->update_stats(meta
.owner
, obj
.bucket
, (orig_exists
? 0 : 1),
7048 store
->quota_handler
->update_stats(meta
.owner
, obj
.bucket
, (orig_exists
? 0 : 1),
7049 accounted_size
, orig_size
);
7054 int ret
= index_op
->cancel();
7056 ldout(store
->ctx(), 0) << "ERROR: index_op.cancel()() returned ret=" << ret
<< dendl
;
7059 meta
.canceled
= true;
7061 /* we lost in a race. There are a few options:
7062 * - existing object was rewritten (ECANCELED)
7063 * - non existing object was created (EEXIST)
7064 * - object was removed (ENOENT)
7065 * should treat it as a success
7067 if (meta
.if_match
== NULL
&& meta
.if_nomatch
== NULL
) {
7068 if (r
== -ECANCELED
|| r
== -ENOENT
|| r
== -EEXIST
) {
7072 if (meta
.if_match
!= NULL
) {
7073 // only overwrite existing object
7074 if (strcmp(meta
.if_match
, "*") == 0) {
7076 r
= -ERR_PRECONDITION_FAILED
;
7077 } else if (r
== -ECANCELED
) {
7083 if (meta
.if_nomatch
!= NULL
) {
7084 // only create a new object
7085 if (strcmp(meta
.if_nomatch
, "*") == 0) {
7087 r
= -ERR_PRECONDITION_FAILED
;
7088 } else if (r
== -ENOENT
) {
7098 int RGWRados::Object::Write::write_meta(uint64_t size
, uint64_t accounted_size
,
7099 map
<string
, bufferlist
>& attrs
)
7101 RGWBucketInfo
& bucket_info
= target
->get_bucket_info();
7103 RGWRados::Bucket
bop(target
->get_store(), bucket_info
);
7104 RGWRados::Bucket::UpdateIndex
index_op(&bop
, target
->get_obj());
7105 index_op
.set_zones_trace(meta
.zones_trace
);
7107 bool assume_noent
= (meta
.if_match
== NULL
&& meta
.if_nomatch
== NULL
);
7110 r
= _do_write_meta(size
, accounted_size
, attrs
, assume_noent
, meta
.modify_tail
, (void *)&index_op
);
7112 assume_noent
= false;
7115 if (!assume_noent
) {
7116 r
= _do_write_meta(size
, accounted_size
, attrs
, assume_noent
, meta
.modify_tail
, (void *)&index_op
);
7121 /** Write/overwrite a system object. */
7122 int RGWRados::put_system_obj_impl(rgw_raw_obj
& obj
, uint64_t size
, real_time
*mtime
,
7123 map
<std::string
, bufferlist
>& attrs
, int flags
,
7125 RGWObjVersionTracker
*objv_tracker
,
7126 real_time set_mtime
/* 0 for don't set */)
7129 int r
= get_system_obj_ref(obj
, &ref
);
7133 ObjectWriteOperation op
;
7135 if (flags
& PUT_OBJ_EXCL
) {
7136 if (!(flags
& PUT_OBJ_CREATE
))
7138 op
.create(true); // exclusive create
7141 op
.set_op_flags2(LIBRADOS_OP_FLAG_FAILOK
);
7146 objv_tracker
->prepare_op_for_write(&op
);
7149 if (real_clock::is_zero(set_mtime
)) {
7150 set_mtime
= real_clock::now();
7153 struct timespec mtime_ts
= real_clock::to_timespec(set_mtime
);
7154 op
.mtime2(&mtime_ts
);
7155 op
.write_full(data
);
7159 for (map
<string
, bufferlist
>::iterator iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
7160 const string
& name
= iter
->first
;
7161 bufferlist
& bl
= iter
->second
;
7166 op
.setxattr(name
.c_str(), bl
);
7169 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
7175 objv_tracker
->apply_write();
7185 int RGWRados::put_system_obj_data(void *ctx
, rgw_raw_obj
& obj
, bufferlist
& bl
,
7186 off_t ofs
, bool exclusive
,
7187 RGWObjVersionTracker
*objv_tracker
)
7190 int r
= get_system_obj_ref(obj
, &ref
);
7195 ObjectWriteOperation op
;
7201 objv_tracker
->prepare_op_for_write(&op
);
7208 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
7213 objv_tracker
->apply_write();
7219 * Write/overwrite an object to the bucket storage.
7220 * bucket: the bucket to store the object in
7221 * obj: the object name/key
7222 * data: the object contents/value
7223 * offset: the offet to write to in the object
7224 * If this is -1, we will overwrite the whole object.
7225 * size: the amount of data to write (data must be this long)
7226 * attrs: all the given attrs are written to bucket storage for the given object
7227 * Returns: 0 on success, -ERR# otherwise.
7230 int RGWRados::aio_put_obj_data(void *ctx
, rgw_raw_obj
& obj
, bufferlist
& bl
,
7231 off_t ofs
, bool exclusive
,
7235 int r
= get_raw_obj_ref(obj
, &ref
);
7240 AioCompletion
*c
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
7243 ObjectWriteOperation op
;
7253 r
= ref
.ioctx
.aio_operate(ref
.oid
, c
, &op
);
7260 int RGWRados::aio_wait(void *handle
)
7262 AioCompletion
*c
= (AioCompletion
*)handle
;
7264 int ret
= c
->get_return_value();
7269 bool RGWRados::aio_completed(void *handle
)
7271 AioCompletion
*c
= (AioCompletion
*)handle
;
7272 return c
->is_safe();
7275 class RGWRadosPutObj
: public RGWGetDataCB
7279 RGWPutObjDataProcessor
*filter
;
7280 boost::optional
<RGWPutObj_Compress
>& compressor
;
7281 CompressorRef
& plugin
;
7282 RGWPutObjProcessor_Atomic
*processor
;
7283 RGWOpStateSingleOp
*opstate
;
7284 void (*progress_cb
)(off_t
, void *);
7285 void *progress_data
;
7286 bufferlist extra_data_bl
;
7287 uint64_t extra_data_left
;
7289 map
<string
, bufferlist
> src_attrs
;
7291 RGWRadosPutObj(CephContext
* cct
,
7292 CompressorRef
& plugin
,
7293 boost::optional
<RGWPutObj_Compress
>& compressor
,
7294 RGWPutObjProcessor_Atomic
*p
,
7295 RGWOpStateSingleOp
*_ops
,
7296 void (*_progress_cb
)(off_t
, void *),
7297 void *_progress_data
) :
7300 compressor(compressor
),
7304 progress_cb(_progress_cb
),
7305 progress_data(_progress_data
),
7309 int process_attrs(void) {
7310 if (extra_data_bl
.length()) {
7312 if (!jp
.parse(extra_data_bl
.c_str(), extra_data_bl
.length())) {
7313 ldout(cct
, 0) << "failed to parse response extra data. len=" << extra_data_bl
.length() << " data=" << extra_data_bl
.c_str() << dendl
;
7317 JSONDecoder::decode_json("attrs", src_attrs
, &jp
);
7319 src_attrs
.erase(RGW_ATTR_COMPRESSION
);
7320 src_attrs
.erase(RGW_ATTR_MANIFEST
); // not interested in original object layout
7323 if (plugin
&& src_attrs
.find(RGW_ATTR_CRYPT_MODE
) == src_attrs
.end()) {
7324 //do not compress if object is encrypted
7325 compressor
= boost::in_place(cct
, plugin
, filter
);
7326 filter
= &*compressor
;
7331 int handle_data(bufferlist
& bl
, off_t ofs
, off_t len
) override
{
7333 progress_cb(ofs
, progress_data
);
7335 if (extra_data_left
) {
7336 size_t extra_len
= bl
.length();
7337 if (extra_len
> extra_data_left
)
7338 extra_len
= extra_data_left
;
7341 bl
.splice(0, extra_len
, &extra
);
7342 extra_data_bl
.append(extra
);
7344 extra_data_left
-= extra_len
;
7345 if (extra_data_left
== 0) {
7346 int res
= process_attrs();
7350 if (bl
.length() == 0) {
7355 // adjust ofs based on extra_data_len, so the result is a logical offset
7356 // into the object data
7357 assert(uint64_t(ofs
) >= extra_data_len
);
7358 ofs
-= extra_data_len
;
7360 data_len
+= bl
.length();
7363 bool need_opstate
= true;
7366 void *handle
= NULL
;
7368 uint64_t size
= bl
.length();
7369 int ret
= filter
->handle_data(bl
, ofs
, &handle
, &obj
, &again
);
7373 if (need_opstate
&& opstate
) {
7374 /* need to update opstate repository with new state. This is ratelimited, so we're not
7375 * really doing it every time
7377 ret
= opstate
->renew_state();
7379 ldout(cct
, 0) << "ERROR: RGWRadosPutObj::handle_data(): failed to renew op state ret=" << ret
<< dendl
;
7380 int r
= filter
->throttle_data(handle
, obj
, size
, false);
7382 ldout(cct
, 0) << "ERROR: RGWRadosPutObj::handle_data(): processor->throttle_data() returned " << r
<< dendl
;
7384 /* could not renew state! might have been marked as cancelled */
7387 need_opstate
= false;
7390 ret
= filter
->throttle_data(handle
, obj
, size
, false);
7398 bufferlist
& get_extra_data() { return extra_data_bl
; }
7400 map
<string
, bufferlist
>& get_attrs() { return src_attrs
; }
7402 void set_extra_data_len(uint64_t len
) override
{
7403 extra_data_left
= len
;
7404 RGWGetDataCB::set_extra_data_len(len
);
7407 uint64_t get_data_len() {
7411 int complete(const string
& etag
, real_time
*mtime
, real_time set_mtime
,
7412 map
<string
, bufferlist
>& attrs
, real_time delete_at
, rgw_zone_set
*zones_trace
) {
7413 return processor
->complete(data_len
, etag
, mtime
, set_mtime
, attrs
, delete_at
, NULL
, NULL
, NULL
, zones_trace
);
7416 bool is_canceled() {
7417 return processor
->is_canceled();
7422 * prepare attrset depending on attrs_mod.
7424 static void set_copy_attrs(map
<string
, bufferlist
>& src_attrs
,
7425 map
<string
, bufferlist
>& attrs
,
7426 RGWRados::AttrsMod attrs_mod
)
7428 switch (attrs_mod
) {
7429 case RGWRados::ATTRSMOD_NONE
:
7432 case RGWRados::ATTRSMOD_REPLACE
:
7433 if (!attrs
[RGW_ATTR_ETAG
].length()) {
7434 attrs
[RGW_ATTR_ETAG
] = src_attrs
[RGW_ATTR_ETAG
];
7436 if (!attrs
[RGW_ATTR_TAIL_TAG
].length()) {
7437 auto ttiter
= src_attrs
.find(RGW_ATTR_TAIL_TAG
);
7438 if (ttiter
!= src_attrs
.end()) {
7439 attrs
[RGW_ATTR_TAIL_TAG
] = src_attrs
[RGW_ATTR_TAIL_TAG
];
7443 case RGWRados::ATTRSMOD_MERGE
:
7444 for (map
<string
, bufferlist
>::iterator it
= src_attrs
.begin(); it
!= src_attrs
.end(); ++it
) {
7445 if (attrs
.find(it
->first
) == attrs
.end()) {
7446 attrs
[it
->first
] = it
->second
;
7453 int RGWRados::rewrite_obj(RGWBucketInfo
& dest_bucket_info
, rgw_obj
& obj
)
7455 map
<string
, bufferlist
> attrset
;
7459 RGWObjectCtx
rctx(this);
7461 RGWRados::Object
op_target(this, dest_bucket_info
, rctx
, obj
);
7462 RGWRados::Object::Read
read_op(&op_target
);
7464 read_op
.params
.attrs
= &attrset
;
7465 read_op
.params
.lastmod
= &mtime
;
7466 read_op
.params
.obj_size
= &obj_size
;
7468 int ret
= read_op
.prepare();
7472 attrset
.erase(RGW_ATTR_ID_TAG
);
7473 attrset
.erase(RGW_ATTR_TAIL_TAG
);
7475 uint64_t max_chunk_size
;
7477 ret
= get_max_chunk_size(dest_bucket_info
.placement_rule
, obj
, &max_chunk_size
);
7479 ldout(cct
, 0) << "ERROR: failed to get max_chunk_size() for bucket " << obj
.bucket
<< dendl
;
7483 return copy_obj_data(rctx
, dest_bucket_info
, read_op
, obj_size
- 1, obj
, obj
,
7484 max_chunk_size
, NULL
, mtime
, attrset
,
7485 RGW_OBJ_CATEGORY_MAIN
, 0, real_time(),
7486 (obj
.key
.instance
.empty() ? NULL
: &(obj
.key
.instance
)),
7490 struct obj_time_weight
{
7492 uint32_t zone_short_id
;
7494 bool high_precision
;
7496 obj_time_weight() : zone_short_id(0), pg_ver(0), high_precision(false) {}
7498 bool compare_low_precision(const obj_time_weight
& rhs
) {
7499 struct timespec l
= ceph::real_clock::to_timespec(mtime
);
7500 struct timespec r
= ceph::real_clock::to_timespec(rhs
.mtime
);
7509 if (zone_short_id
!= rhs
.zone_short_id
) {
7510 return (zone_short_id
< rhs
.zone_short_id
);
7512 return (pg_ver
< rhs
.pg_ver
);
7516 bool operator<(const obj_time_weight
& rhs
) {
7517 if (!high_precision
|| !rhs
.high_precision
) {
7518 return compare_low_precision(rhs
);
7520 if (mtime
> rhs
.mtime
) {
7523 if (mtime
< rhs
.mtime
) {
7526 if (zone_short_id
!= rhs
.zone_short_id
) {
7527 return (zone_short_id
< rhs
.zone_short_id
);
7529 return (pg_ver
< rhs
.pg_ver
);
7532 void init(const real_time
& _mtime
, uint32_t _short_id
, uint64_t _pg_ver
) {
7534 zone_short_id
= _short_id
;
7538 void init(RGWObjState
*state
) {
7539 mtime
= state
->mtime
;
7540 zone_short_id
= state
->zone_short_id
;
7541 pg_ver
= state
->pg_ver
;
7545 inline ostream
& operator<<(ostream
& out
, const obj_time_weight
&o
) {
7548 if (o
.zone_short_id
!= 0 || o
.pg_ver
!= 0) {
7549 out
<< "[zid=" << o
.zone_short_id
<< ", pgv=" << o
.pg_ver
<< "]";
7555 class RGWGetExtraDataCB
: public RGWGetDataCB
{
7556 bufferlist extra_data
;
7558 RGWGetExtraDataCB() {}
7559 int handle_data(bufferlist
& bl
, off_t bl_ofs
, off_t bl_len
) override
{
7560 if (extra_data
.length() < extra_data_len
) {
7561 off_t max
= extra_data_len
- extra_data
.length();
7565 bl
.splice(0, max
, &extra_data
);
7570 bufferlist
& get_extra_data() {
7575 int RGWRados::stat_remote_obj(RGWObjectCtx
& obj_ctx
,
7576 const rgw_user
& user_id
,
7577 const string
& client_id
,
7579 const string
& source_zone
,
7581 RGWBucketInfo
& src_bucket_info
,
7582 real_time
*src_mtime
,
7584 const real_time
*mod_ptr
,
7585 const real_time
*unmod_ptr
,
7586 bool high_precision_time
,
7587 const char *if_match
,
7588 const char *if_nomatch
,
7589 map
<string
, bufferlist
> *pattrs
,
7594 /* source is in a different zonegroup, copy from there */
7596 RGWRESTStreamRWRequest
*in_stream_req
;
7598 map
<string
, bufferlist
> src_attrs
;
7599 append_rand_alpha(cct
, tag
, tag
, 32);
7600 obj_time_weight set_mtime_weight
;
7601 set_mtime_weight
.high_precision
= high_precision_time
;
7604 if (source_zone
.empty()) {
7605 if (src_bucket_info
.zonegroup
.empty()) {
7606 /* source is in the master zonegroup */
7607 conn
= rest_master_conn
;
7609 map
<string
, RGWRESTConn
*>::iterator iter
= zonegroup_conn_map
.find(src_bucket_info
.zonegroup
);
7610 if (iter
== zonegroup_conn_map
.end()) {
7611 ldout(cct
, 0) << "could not find zonegroup connection to zonegroup: " << source_zone
<< dendl
;
7614 conn
= iter
->second
;
7617 map
<string
, RGWRESTConn
*>::iterator iter
= zone_conn_map
.find(source_zone
);
7618 if (iter
== zone_conn_map
.end()) {
7619 ldout(cct
, 0) << "could not find zone connection to zone: " << source_zone
<< dendl
;
7622 conn
= iter
->second
;
7625 RGWGetExtraDataCB cb
;
7627 map
<string
, string
> req_headers
;
7628 real_time set_mtime
;
7630 const real_time
*pmod
= mod_ptr
;
7632 obj_time_weight dest_mtime_weight
;
7634 constexpr bool prepend_meta
= true;
7635 constexpr bool get_op
= true;
7636 constexpr bool rgwx_stat
= true;
7637 constexpr bool sync_manifest
= true;
7638 constexpr bool skip_decrypt
= true;
7639 int ret
= conn
->get_obj(user_id
, info
, src_obj
, pmod
, unmod_ptr
,
7640 dest_mtime_weight
.zone_short_id
, dest_mtime_weight
.pg_ver
,
7641 prepend_meta
, get_op
, rgwx_stat
,
7642 sync_manifest
, skip_decrypt
, &cb
, &in_stream_req
);
7647 ret
= conn
->complete_request(in_stream_req
, etag
, &set_mtime
, psize
, req_headers
);
7652 bufferlist
& extra_data_bl
= cb
.get_extra_data();
7653 if (extra_data_bl
.length()) {
7655 if (!jp
.parse(extra_data_bl
.c_str(), extra_data_bl
.length())) {
7656 ldout(cct
, 0) << "failed to parse response extra data. len=" << extra_data_bl
.length() << " data=" << extra_data_bl
.c_str() << dendl
;
7660 JSONDecoder::decode_json("attrs", src_attrs
, &jp
);
7662 src_attrs
.erase(RGW_ATTR_MANIFEST
); // not interested in original object layout
7666 *src_mtime
= set_mtime
;
7670 map
<string
, bufferlist
>::iterator iter
= src_attrs
.find(RGW_ATTR_ETAG
);
7671 if (iter
!= src_attrs
.end()) {
7672 bufferlist
& etagbl
= iter
->second
;
7673 *petag
= etagbl
.to_str();
7678 *pattrs
= src_attrs
;
7684 int RGWRados::fetch_remote_obj(RGWObjectCtx
& obj_ctx
,
7685 const rgw_user
& user_id
,
7686 const string
& client_id
,
7687 const string
& op_id
,
7688 bool record_op_state
,
7690 const string
& source_zone
,
7693 RGWBucketInfo
& dest_bucket_info
,
7694 RGWBucketInfo
& src_bucket_info
,
7695 real_time
*src_mtime
,
7697 const real_time
*mod_ptr
,
7698 const real_time
*unmod_ptr
,
7699 bool high_precision_time
,
7700 const char *if_match
,
7701 const char *if_nomatch
,
7704 map
<string
, bufferlist
>& attrs
,
7705 RGWObjCategory category
,
7707 real_time delete_at
,
7710 ceph::buffer::list
*petag
,
7711 void (*progress_cb
)(off_t
, void *),
7712 void *progress_data
,
7713 rgw_zone_set
*zones_trace
)
7715 /* source is in a different zonegroup, copy from there */
7717 RGWRESTStreamRWRequest
*in_stream_req
;
7720 append_rand_alpha(cct
, tag
, tag
, 32);
7721 obj_time_weight set_mtime_weight
;
7722 set_mtime_weight
.high_precision
= high_precision_time
;
7724 RGWPutObjProcessor_Atomic
processor(obj_ctx
,
7725 dest_bucket_info
, dest_obj
.bucket
, dest_obj
.key
.name
,
7726 cct
->_conf
->rgw_obj_stripe_size
, tag
, dest_bucket_info
.versioning_enabled());
7727 if (version_id
&& *version_id
!= "null") {
7728 processor
.set_version_id(*version_id
);
7730 processor
.set_olh_epoch(olh_epoch
);
7731 int ret
= processor
.prepare(this, NULL
);
7737 if (source_zone
.empty()) {
7738 if (dest_bucket_info
.zonegroup
.empty()) {
7739 /* source is in the master zonegroup */
7740 conn
= rest_master_conn
;
7742 map
<string
, RGWRESTConn
*>::iterator iter
= zonegroup_conn_map
.find(src_bucket_info
.zonegroup
);
7743 if (iter
== zonegroup_conn_map
.end()) {
7744 ldout(cct
, 0) << "could not find zonegroup connection to zonegroup: " << source_zone
<< dendl
;
7747 conn
= iter
->second
;
7750 map
<string
, RGWRESTConn
*>::iterator iter
= zone_conn_map
.find(source_zone
);
7751 if (iter
== zone_conn_map
.end()) {
7752 ldout(cct
, 0) << "could not find zone connection to zone: " << source_zone
<< dendl
;
7755 conn
= iter
->second
;
7758 string obj_name
= dest_obj
.bucket
.name
+ "/" + dest_obj
.get_oid();
7760 RGWOpStateSingleOp
*opstate
= NULL
;
7762 if (record_op_state
) {
7763 opstate
= new RGWOpStateSingleOp(this, client_id
, op_id
, obj_name
);
7765 ret
= opstate
->set_state(RGWOpState::OPSTATE_IN_PROGRESS
);
7767 ldout(cct
, 0) << "ERROR: failed to set opstate ret=" << ret
<< dendl
;
7773 boost::optional
<RGWPutObj_Compress
> compressor
;
7774 CompressorRef plugin
;
7776 const auto& compression_type
= zone_params
.get_compression_type(
7777 dest_bucket_info
.placement_rule
);
7778 if (compression_type
!= "none") {
7779 plugin
= Compressor::create(cct
, compression_type
);
7781 ldout(cct
, 1) << "Cannot load plugin for compression type "
7782 << compression_type
<< dendl
;
7786 RGWRadosPutObj
cb(cct
, plugin
, compressor
, &processor
, opstate
, progress_cb
, progress_data
);
7789 map
<string
, string
> req_headers
;
7790 real_time set_mtime
;
7792 RGWObjState
*dest_state
= NULL
;
7794 const real_time
*pmod
= mod_ptr
;
7796 obj_time_weight dest_mtime_weight
;
7798 if (copy_if_newer
) {
7799 /* need to get mtime for destination */
7800 ret
= get_obj_state(&obj_ctx
, dest_bucket_info
, dest_obj
, &dest_state
, false);
7804 if (!real_clock::is_zero(dest_state
->mtime
)) {
7805 dest_mtime_weight
.init(dest_state
);
7806 pmod
= &dest_mtime_weight
.mtime
;
7810 static constexpr bool prepend_meta
= true;
7811 static constexpr bool get_op
= true;
7812 static constexpr bool rgwx_stat
= false;
7813 static constexpr bool sync_manifest
= true;
7814 static constexpr bool skip_decrypt
= true;
7815 ret
= conn
->get_obj(user_id
, info
, src_obj
, pmod
, unmod_ptr
,
7816 dest_mtime_weight
.zone_short_id
, dest_mtime_weight
.pg_ver
,
7817 prepend_meta
, get_op
, rgwx_stat
,
7818 sync_manifest
, skip_decrypt
, &cb
, &in_stream_req
);
7823 ret
= conn
->complete_request(in_stream_req
, etag
, &set_mtime
, nullptr, req_headers
);
7827 if (compressor
&& compressor
->is_compressed()) {
7829 RGWCompressionInfo cs_info
;
7830 cs_info
.compression_type
= plugin
->get_type_name();
7831 cs_info
.orig_size
= cb
.get_data_len();
7832 cs_info
.blocks
= move(compressor
->get_compression_blocks());
7833 ::encode(cs_info
, tmp
);
7834 cb
.get_attrs()[RGW_ATTR_COMPRESSION
] = tmp
;
7837 if (source_zone
.empty()) { /* need to preserve expiration if copy in the same zonegroup */
7838 cb
.get_attrs().erase(RGW_ATTR_DELETE_AT
);
7840 map
<string
, bufferlist
>::iterator iter
= cb
.get_attrs().find(RGW_ATTR_DELETE_AT
);
7841 if (iter
!= cb
.get_attrs().end()) {
7843 ::decode(delete_at
, iter
->second
);
7844 } catch (buffer::error
& err
) {
7845 ldout(cct
, 0) << "ERROR: failed to decode delete_at field in intra zone copy" << dendl
;
7851 *src_mtime
= set_mtime
;
7855 const auto iter
= cb
.get_attrs().find(RGW_ATTR_ETAG
);
7856 if (iter
!= cb
.get_attrs().end()) {
7857 *petag
= iter
->second
;
7861 if (source_zone
.empty()) {
7862 set_copy_attrs(cb
.get_attrs(), attrs
, attrs_mod
);
7864 attrs
= cb
.get_attrs();
7867 if (copy_if_newer
) {
7868 uint64_t pg_ver
= 0;
7869 auto i
= attrs
.find(RGW_ATTR_PG_VER
);
7870 if (i
!= attrs
.end() && i
->second
.length() > 0) {
7871 bufferlist::iterator iter
= i
->second
.begin();
7873 ::decode(pg_ver
, iter
);
7874 } catch (buffer::error
& err
) {
7875 ldout(ctx(), 0) << "ERROR: failed to decode pg ver attribute, ignoring" << dendl
;
7876 /* non critical error */
7879 set_mtime_weight
.init(set_mtime
, get_zone_short_id(), pg_ver
);
7882 #define MAX_COMPLETE_RETRY 100
7883 for (i
= 0; i
< MAX_COMPLETE_RETRY
; i
++) {
7884 ret
= cb
.complete(etag
, mtime
, set_mtime
, attrs
, delete_at
, zones_trace
);
7888 if (copy_if_newer
&& cb
.is_canceled()) {
7889 ldout(cct
, 20) << "raced with another write of obj: " << dest_obj
<< dendl
;
7890 obj_ctx
.obj
.invalidate(dest_obj
); /* object was overwritten */
7891 ret
= get_obj_state(&obj_ctx
, dest_bucket_info
, dest_obj
, &dest_state
, false);
7893 ldout(cct
, 0) << "ERROR: " << __func__
<< ": get_err_state() returned ret=" << ret
<< dendl
;
7896 dest_mtime_weight
.init(dest_state
);
7897 dest_mtime_weight
.high_precision
= high_precision_time
;
7898 if (!dest_state
->exists
||
7899 dest_mtime_weight
< set_mtime_weight
) {
7900 ldout(cct
, 20) << "retrying writing object mtime=" << set_mtime
<< " dest_state->mtime=" << dest_state
->mtime
<< " dest_state->exists=" << dest_state
->exists
<< dendl
;
7903 ldout(cct
, 20) << "not retrying writing object mtime=" << set_mtime
<< " dest_state->mtime=" << dest_state
->mtime
<< " dest_state->exists=" << dest_state
->exists
<< dendl
;
7909 if (i
== MAX_COMPLETE_RETRY
) {
7910 ldout(cct
, 0) << "ERROR: retried object completion too many times, something is wrong!" << dendl
;
7916 ret
= opstate
->set_state(RGWOpState::OPSTATE_COMPLETE
);
7918 ldout(cct
, 0) << "ERROR: failed to set opstate ret=" << ret
<< dendl
;
7925 if (copy_if_newer
&& ret
== -ERR_NOT_MODIFIED
) {
7929 RGWOpState::OpState state
;
7931 state
= RGWOpState::OPSTATE_ERROR
;
7933 state
= RGWOpState::OPSTATE_COMPLETE
;
7935 int r
= opstate
->set_state(state
);
7937 ldout(cct
, 0) << "ERROR: failed to set opstate r=" << ret
<< dendl
;
7945 int RGWRados::copy_obj_to_remote_dest(RGWObjState
*astate
,
7946 map
<string
, bufferlist
>& src_attrs
,
7947 RGWRados::Object::Read
& read_op
,
7948 const rgw_user
& user_id
,
7954 RGWRESTStreamWriteRequest
*out_stream_req
;
7956 int ret
= rest_master_conn
->put_obj_init(user_id
, dest_obj
, astate
->size
, src_attrs
, &out_stream_req
);
7961 ret
= read_op
.iterate(0, astate
->size
- 1, out_stream_req
->get_out_cb());
7963 delete out_stream_req
;
7967 ret
= rest_master_conn
->complete_request(out_stream_req
, etag
, mtime
);
7976 * dest_obj: the object to copy into
7977 * src_obj: the object to copy from
7978 * attrs: usage depends on attrs_mod parameter
7979 * attrs_mod: the modification mode of the attrs, may have the following values:
7980 * ATTRSMOD_NONE - the attributes of the source object will be
7981 * copied without modifications, attrs parameter is ignored;
7982 * ATTRSMOD_REPLACE - new object will have the attributes provided by attrs
7983 * parameter, source object attributes are not copied;
7984 * ATTRSMOD_MERGE - any conflicting meta keys on the source object's attributes
7985 * are overwritten by values contained in attrs parameter.
7986 * err: stores any errors resulting from the get of the original object
7987 * Returns: 0 on success, -ERR# otherwise.
7989 int RGWRados::copy_obj(RGWObjectCtx
& obj_ctx
,
7990 const rgw_user
& user_id
,
7991 const string
& client_id
,
7992 const string
& op_id
,
7994 const string
& source_zone
,
7997 RGWBucketInfo
& dest_bucket_info
,
7998 RGWBucketInfo
& src_bucket_info
,
7999 real_time
*src_mtime
,
8001 const real_time
*mod_ptr
,
8002 const real_time
*unmod_ptr
,
8003 bool high_precision_time
,
8004 const char *if_match
,
8005 const char *if_nomatch
,
8008 map
<string
, bufferlist
>& attrs
,
8009 RGWObjCategory category
,
8011 real_time delete_at
,
8014 ceph::buffer::list
*petag
,
8015 void (*progress_cb
)(off_t
, void *),
8016 void *progress_data
)
8020 rgw_obj shadow_obj
= dest_obj
;
8026 append_rand_alpha(cct
, dest_obj
.get_oid(), shadow_oid
, 32);
8027 shadow_obj
.init_ns(dest_obj
.bucket
, shadow_oid
, shadow_ns
);
8029 remote_dest
= !get_zonegroup().equals(dest_bucket_info
.zonegroup
);
8030 remote_src
= !get_zonegroup().equals(src_bucket_info
.zonegroup
);
8032 if (remote_src
&& remote_dest
) {
8033 ldout(cct
, 0) << "ERROR: can't copy object when both src and dest buckets are remote" << dendl
;
8037 ldout(cct
, 5) << "Copy object " << src_obj
.bucket
<< ":" << src_obj
.get_oid() << " => " << dest_obj
.bucket
<< ":" << dest_obj
.get_oid() << dendl
;
8039 if (remote_src
|| !source_zone
.empty()) {
8040 return fetch_remote_obj(obj_ctx
, user_id
, client_id
, op_id
, true, info
, source_zone
,
8041 dest_obj
, src_obj
, dest_bucket_info
, src_bucket_info
, src_mtime
, mtime
, mod_ptr
,
8042 unmod_ptr
, high_precision_time
,
8043 if_match
, if_nomatch
, attrs_mod
, copy_if_newer
, attrs
, category
,
8044 olh_epoch
, delete_at
, version_id
, ptag
, petag
, progress_cb
, progress_data
);
8047 map
<string
, bufferlist
> src_attrs
;
8048 RGWRados::Object
src_op_target(this, src_bucket_info
, obj_ctx
, src_obj
);
8049 RGWRados::Object::Read
read_op(&src_op_target
);
8051 read_op
.conds
.mod_ptr
= mod_ptr
;
8052 read_op
.conds
.unmod_ptr
= unmod_ptr
;
8053 read_op
.conds
.high_precision_time
= high_precision_time
;
8054 read_op
.conds
.if_match
= if_match
;
8055 read_op
.conds
.if_nomatch
= if_nomatch
;
8056 read_op
.params
.attrs
= &src_attrs
;
8057 read_op
.params
.lastmod
= src_mtime
;
8058 read_op
.params
.obj_size
= &obj_size
;
8060 ret
= read_op
.prepare();
8065 src_attrs
[RGW_ATTR_ACL
] = attrs
[RGW_ATTR_ACL
];
8066 src_attrs
.erase(RGW_ATTR_DELETE_AT
);
8068 set_copy_attrs(src_attrs
, attrs
, attrs_mod
);
8069 attrs
.erase(RGW_ATTR_ID_TAG
);
8070 attrs
.erase(RGW_ATTR_PG_VER
);
8071 attrs
.erase(RGW_ATTR_SOURCE_ZONE
);
8072 map
<string
, bufferlist
>::iterator cmp
= src_attrs
.find(RGW_ATTR_COMPRESSION
);
8073 if (cmp
!= src_attrs
.end())
8074 attrs
[RGW_ATTR_COMPRESSION
] = cmp
->second
;
8076 RGWObjManifest manifest
;
8077 RGWObjState
*astate
= NULL
;
8079 ret
= get_obj_state(&obj_ctx
, src_bucket_info
, src_obj
, &astate
);
8084 vector
<rgw_raw_obj
> ref_objs
;
8087 /* dest is in a different zonegroup, copy it there */
8088 return copy_obj_to_remote_dest(astate
, attrs
, read_op
, user_id
, dest_obj
, mtime
);
8090 uint64_t max_chunk_size
;
8092 ret
= get_max_chunk_size(dest_bucket_info
.placement_rule
, dest_obj
, &max_chunk_size
);
8094 ldout(cct
, 0) << "ERROR: failed to get max_chunk_size() for bucket " << dest_obj
.bucket
<< dendl
;
8100 if (!get_obj_data_pool(src_bucket_info
.placement_rule
, src_obj
, &src_pool
)) {
8101 ldout(cct
, 0) << "ERROR: failed to locate data pool for " << src_obj
<< dendl
;
8104 if (!get_obj_data_pool(dest_bucket_info
.placement_rule
, dest_obj
, &dest_pool
)) {
8105 ldout(cct
, 0) << "ERROR: failed to locate data pool for " << dest_obj
<< dendl
;
8110 bool copy_data
= !astate
->has_manifest
|| (src_pool
!= dest_pool
);
8111 bool copy_first
= false;
8112 if (astate
->has_manifest
) {
8113 if (!astate
->manifest
.has_tail()) {
8116 uint64_t head_size
= astate
->manifest
.get_head_size();
8118 if (head_size
> 0) {
8119 if (head_size
> max_chunk_size
) {
8129 const auto iter
= attrs
.find(RGW_ATTR_ETAG
);
8130 if (iter
!= attrs
.end()) {
8131 *petag
= iter
->second
;
8135 if (copy_data
) { /* refcounting tail wouldn't work here, just copy the data */
8136 return copy_obj_data(obj_ctx
, dest_bucket_info
, read_op
, obj_size
- 1, dest_obj
, src_obj
,
8137 max_chunk_size
, mtime
, real_time(), attrs
, category
, olh_epoch
, delete_at
,
8138 version_id
, ptag
, petag
);
8141 RGWObjManifest::obj_iterator miter
= astate
->manifest
.obj_begin();
8143 if (copy_first
) { // we need to copy first chunk, not increase refcount
8148 ret
= get_raw_obj_ref(miter
.get_location().get_raw_obj(this), &ref
);
8153 bool versioned_dest
= dest_bucket_info
.versioning_enabled();
8155 if (version_id
&& !version_id
->empty()) {
8156 versioned_dest
= true;
8157 dest_obj
.key
.set_instance(*version_id
);
8158 } else if (versioned_dest
) {
8159 gen_rand_obj_instance_name(&dest_obj
);
8162 bufferlist first_chunk
;
8164 bool copy_itself
= (dest_obj
== src_obj
);
8165 RGWObjManifest
*pmanifest
;
8166 ldout(cct
, 20) << "dest_obj=" << dest_obj
<< " src_obj=" << src_obj
<< " copy_itself=" << (int)copy_itself
<< dendl
;
8168 RGWRados::Object
dest_op_target(this, dest_bucket_info
, obj_ctx
, dest_obj
);
8169 RGWRados::Object::Write
write_op(&dest_op_target
);
8178 append_rand_alpha(cct
, tag
, tag
, 32);
8182 attrs
.erase(RGW_ATTR_TAIL_TAG
);
8183 manifest
= astate
->manifest
;
8184 const rgw_bucket_placement
& tail_placement
= manifest
.get_tail_placement();
8185 if (tail_placement
.bucket
.name
.empty()) {
8186 manifest
.set_tail_placement(tail_placement
.placement_rule
, src_obj
.bucket
);
8189 for (; miter
!= astate
->manifest
.obj_end(); ++miter
) {
8190 ObjectWriteOperation op
;
8191 ref_tag
= tag
+ '\0';
8192 cls_refcount_get(op
, ref_tag
, true);
8193 const rgw_raw_obj
& loc
= miter
.get_location().get_raw_obj(this);
8194 ref
.ioctx
.locator_set_key(loc
.loc
);
8196 ret
= ref
.ioctx
.operate(loc
.oid
, &op
);
8201 ref_objs
.push_back(loc
);
8204 pmanifest
= &manifest
;
8206 pmanifest
= &astate
->manifest
;
8207 /* don't send the object's tail for garbage collection */
8208 astate
->keep_tail
= true;
8212 ret
= read_op
.read(0, max_chunk_size
, first_chunk
);
8217 pmanifest
->set_head(dest_bucket_info
.placement_rule
, dest_obj
, first_chunk
.length());
8219 pmanifest
->set_head(dest_bucket_info
.placement_rule
, dest_obj
, 0);
8222 write_op
.meta
.data
= &first_chunk
;
8223 write_op
.meta
.manifest
= pmanifest
;
8224 write_op
.meta
.ptag
= &tag
;
8225 write_op
.meta
.owner
= dest_bucket_info
.owner
;
8226 write_op
.meta
.mtime
= mtime
;
8227 write_op
.meta
.flags
= PUT_OBJ_CREATE
;
8228 write_op
.meta
.category
= category
;
8229 write_op
.meta
.olh_epoch
= olh_epoch
;
8230 write_op
.meta
.delete_at
= delete_at
;
8231 write_op
.meta
.modify_tail
= !copy_itself
;
8233 ret
= write_op
.write_meta(obj_size
, astate
->accounted_size
, attrs
);
8242 vector
<rgw_raw_obj
>::iterator riter
;
8244 /* rollback reference */
8245 for (riter
= ref_objs
.begin(); riter
!= ref_objs
.end(); ++riter
) {
8246 ObjectWriteOperation op
;
8247 cls_refcount_put(op
, tag
, true);
8249 ref
.ioctx
.locator_set_key(riter
->loc
);
8251 int r
= ref
.ioctx
.operate(riter
->oid
, &op
);
8253 ldout(cct
, 0) << "ERROR: cleanup after error failed to drop reference on obj=" << *riter
<< dendl
;
8261 int RGWRados::copy_obj_data(RGWObjectCtx
& obj_ctx
,
8262 RGWBucketInfo
& dest_bucket_info
,
8263 RGWRados::Object::Read
& read_op
, off_t end
,
8266 uint64_t max_chunk_size
,
8268 real_time set_mtime
,
8269 map
<string
, bufferlist
>& attrs
,
8270 RGWObjCategory category
,
8272 real_time delete_at
,
8275 ceph::buffer::list
*petag
)
8277 bufferlist first_chunk
;
8278 RGWObjManifest manifest
;
8281 append_rand_alpha(cct
, tag
, tag
, 32);
8283 RGWPutObjProcessor_Atomic
processor(obj_ctx
,
8284 dest_bucket_info
, dest_obj
.bucket
, dest_obj
.key
.name
,
8285 cct
->_conf
->rgw_obj_stripe_size
, tag
, dest_bucket_info
.versioning_enabled());
8287 processor
.set_version_id(*version_id
);
8289 processor
.set_olh_epoch(olh_epoch
);
8290 int ret
= processor
.prepare(this, NULL
);
8298 ret
= read_op
.read(ofs
, end
, bl
);
8300 uint64_t read_len
= ret
;
8307 ret
= processor
.handle_data(bl
, ofs
, &handle
, &obj
, &again
);
8311 ret
= processor
.throttle_data(handle
, obj
, read_len
, false);
8317 } while (ofs
<= end
);
8320 auto iter
= attrs
.find(RGW_ATTR_ETAG
);
8321 if (iter
!= attrs
.end()) {
8322 bufferlist
& bl
= iter
->second
;
8323 etag
= string(bl
.c_str(), bl
.length());
8329 uint64_t accounted_size
;
8331 bool compressed
{false};
8332 RGWCompressionInfo cs_info
;
8333 ret
= rgw_compression_info_from_attrset(attrs
, compressed
, cs_info
);
8335 ldout(cct
, 0) << "ERROR: failed to read compression info" << dendl
;
8338 // pass original size if compressed
8339 accounted_size
= compressed
? cs_info
.orig_size
: ofs
;
8342 return processor
.complete(accounted_size
, etag
, mtime
, set_mtime
, attrs
, delete_at
);
8345 bool RGWRados::is_meta_master()
8347 if (!get_zonegroup().is_master_zonegroup()) {
8351 return (get_zonegroup().master_zone
== zone_public_config
.id
);
8355 * Check to see if the bucket metadata could be synced
8356 * bucket: the bucket to check
8357 * Returns false is the bucket is not synced
8359 bool RGWRados::is_syncing_bucket_meta(const rgw_bucket
& bucket
)
8362 /* no current period */
8363 if (current_period
.get_id().empty()) {
8367 /* zonegroup is not master zonegroup */
8368 if (!get_zonegroup().is_master_zonegroup()) {
8372 /* single zonegroup and a single zone */
8373 if (current_period
.is_single_zonegroup() && get_zonegroup().zones
.size() == 1) {
8377 /* zone is not master */
8378 if (get_zonegroup().master_zone
.compare(zone_public_config
.id
) != 0) {
8385 int RGWRados::check_bucket_empty(RGWBucketInfo
& bucket_info
)
8387 std::map
<string
, rgw_bucket_dir_entry
> ent_map
;
8388 rgw_obj_index_key marker
;
8393 #define NUM_ENTRIES 1000
8394 int r
= cls_bucket_list(bucket_info
, RGW_NO_SHARD
, marker
, prefix
, NUM_ENTRIES
, true, ent_map
,
8395 &is_truncated
, &marker
);
8400 std::map
<string
, rgw_bucket_dir_entry
>::iterator eiter
;
8401 for (eiter
= ent_map
.begin(); eiter
!= ent_map
.end(); ++eiter
) {
8404 if (rgw_obj_key::oid_to_key_in_ns(eiter
->second
.key
.name
, &obj
, ns
))
8407 } while (is_truncated
);
8413 * bucket: the name of the bucket to delete
8414 * Returns 0 on success, -ERR# otherwise.
8416 int RGWRados::delete_bucket(RGWBucketInfo
& bucket_info
, RGWObjVersionTracker
& objv_tracker
, bool check_empty
)
8418 const rgw_bucket
& bucket
= bucket_info
.bucket
;
8419 librados::IoCtx index_ctx
;
8420 map
<int, string
> bucket_objs
;
8421 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
);
8426 r
= check_bucket_empty(bucket_info
);
8432 r
= rgw_bucket_delete_bucket_obj(this, bucket
.tenant
, bucket
.name
, objv_tracker
);
8436 /* if the bucket is not synced we can remove the meta file */
8437 if (!is_syncing_bucket_meta(bucket
)) {
8438 RGWObjVersionTracker objv_tracker
;
8439 string entry
= bucket
.get_key();
8440 r
= rgw_bucket_instance_remove_entry(this, entry
, &objv_tracker
);
8444 /* remove bucket index objects*/
8445 map
<int, string
>::const_iterator biter
;
8446 for (biter
= bucket_objs
.begin(); biter
!= bucket_objs
.end(); ++biter
) {
8447 index_ctx
.remove(biter
->second
);
8453 int RGWRados::set_bucket_owner(rgw_bucket
& bucket
, ACLOwner
& owner
)
8456 map
<string
, bufferlist
> attrs
;
8457 RGWObjectCtx
obj_ctx(this);
8459 if (bucket
.bucket_id
.empty()) {
8460 r
= get_bucket_info(obj_ctx
, bucket
.tenant
, bucket
.name
, info
, NULL
, &attrs
);
8462 r
= get_bucket_instance_info(obj_ctx
, bucket
, info
, nullptr, &attrs
);
8465 ldout(cct
, 0) << "NOTICE: get_bucket_info on bucket=" << bucket
.name
<< " returned err=" << r
<< dendl
;
8469 info
.owner
= owner
.get_id();
8471 r
= put_bucket_instance_info(info
, false, real_time(), &attrs
);
8473 ldout(cct
, 0) << "NOTICE: put_bucket_info on bucket=" << bucket
.name
<< " returned err=" << r
<< dendl
;
8481 int RGWRados::set_buckets_enabled(vector
<rgw_bucket
>& buckets
, bool enabled
)
8485 vector
<rgw_bucket
>::iterator iter
;
8487 for (iter
= buckets
.begin(); iter
!= buckets
.end(); ++iter
) {
8488 rgw_bucket
& bucket
= *iter
;
8490 ldout(cct
, 20) << "enabling bucket name=" << bucket
.name
<< dendl
;
8492 ldout(cct
, 20) << "disabling bucket name=" << bucket
.name
<< dendl
;
8495 map
<string
, bufferlist
> attrs
;
8496 RGWObjectCtx
obj_ctx(this);
8497 int r
= get_bucket_info(obj_ctx
, bucket
.tenant
, bucket
.name
, info
, NULL
, &attrs
);
8499 ldout(cct
, 0) << "NOTICE: get_bucket_info on bucket=" << bucket
.name
<< " returned err=" << r
<< ", skipping bucket" << dendl
;
8504 info
.flags
&= ~BUCKET_SUSPENDED
;
8506 info
.flags
|= BUCKET_SUSPENDED
;
8509 r
= put_bucket_instance_info(info
, false, real_time(), &attrs
);
8511 ldout(cct
, 0) << "NOTICE: put_bucket_info on bucket=" << bucket
.name
<< " returned err=" << r
<< ", skipping bucket" << dendl
;
8519 int RGWRados::bucket_suspended(rgw_bucket
& bucket
, bool *suspended
)
8521 RGWBucketInfo bucket_info
;
8522 RGWObjectCtx
obj_ctx(this);
8523 int ret
= get_bucket_info(obj_ctx
, bucket
.tenant
, bucket
.name
, bucket_info
, NULL
);
8528 *suspended
= ((bucket_info
.flags
& BUCKET_SUSPENDED
) != 0);
8532 int RGWRados::Object::complete_atomic_modification()
8534 if (!state
->has_manifest
|| state
->keep_tail
)
8537 cls_rgw_obj_chain chain
;
8538 store
->update_gc_chain(obj
, state
->manifest
, &chain
);
8540 if (chain
.empty()) {
8544 string tag
= (state
->tail_tag
.length() > 0 ? state
->tail_tag
.to_str() : state
->obj_tag
.to_str());
8545 return store
->gc
->send_chain(chain
, tag
, false); // do it async
8548 void RGWRados::update_gc_chain(rgw_obj
& head_obj
, RGWObjManifest
& manifest
, cls_rgw_obj_chain
*chain
)
8550 RGWObjManifest::obj_iterator iter
;
8551 rgw_raw_obj raw_head
;
8552 obj_to_raw(manifest
.get_head_placement_rule(), head_obj
, &raw_head
);
8553 for (iter
= manifest
.obj_begin(); iter
!= manifest
.obj_end(); ++iter
) {
8554 const rgw_raw_obj
& mobj
= iter
.get_location().get_raw_obj(this);
8555 if (mobj
== raw_head
)
8557 cls_rgw_obj_key
key(mobj
.oid
);
8558 chain
->push_obj(mobj
.pool
.to_str(), key
, mobj
.loc
);
8562 int RGWRados::send_chain_to_gc(cls_rgw_obj_chain
& chain
, const string
& tag
, bool sync
)
8564 return gc
->send_chain(chain
, tag
, sync
);
8567 int RGWRados::open_bucket_index(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
, string
& bucket_oid
)
8569 const rgw_bucket
& bucket
= bucket_info
.bucket
;
8570 int r
= open_bucket_index_ctx(bucket_info
, index_ctx
);
8574 if (bucket
.bucket_id
.empty()) {
8575 ldout(cct
, 0) << "ERROR: empty bucket id for bucket operation" << dendl
;
8579 bucket_oid
= dir_oid_prefix
;
8580 bucket_oid
.append(bucket
.bucket_id
);
8585 int RGWRados::open_bucket_index_base(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
,
8586 string
& bucket_oid_base
) {
8587 const rgw_bucket
& bucket
= bucket_info
.bucket
;
8588 int r
= open_bucket_index_ctx(bucket_info
, index_ctx
);
8592 if (bucket
.bucket_id
.empty()) {
8593 ldout(cct
, 0) << "ERROR: empty bucket_id for bucket operation" << dendl
;
8597 bucket_oid_base
= dir_oid_prefix
;
8598 bucket_oid_base
.append(bucket
.bucket_id
);
8604 int RGWRados::open_bucket_index(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
,
8605 map
<int, string
>& bucket_objs
, int shard_id
, map
<int, string
> *bucket_instance_ids
) {
8606 string bucket_oid_base
;
8607 int ret
= open_bucket_index_base(bucket_info
, index_ctx
, bucket_oid_base
);
8612 get_bucket_index_objects(bucket_oid_base
, bucket_info
.num_shards
, bucket_objs
, shard_id
);
8613 if (bucket_instance_ids
) {
8614 get_bucket_instance_ids(bucket_info
, shard_id
, bucket_instance_ids
);
8619 template<typename T
>
8620 int RGWRados::open_bucket_index(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
,
8621 map
<int, string
>& oids
, map
<int, T
>& bucket_objs
,
8622 int shard_id
, map
<int, string
> *bucket_instance_ids
)
8624 int ret
= open_bucket_index(bucket_info
, index_ctx
, oids
, shard_id
, bucket_instance_ids
);
8628 map
<int, string
>::const_iterator iter
= oids
.begin();
8629 for (; iter
!= oids
.end(); ++iter
) {
8630 bucket_objs
[iter
->first
] = T();
8635 int RGWRados::open_bucket_index_shard(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
,
8636 const string
& obj_key
, string
*bucket_obj
, int *shard_id
)
8638 string bucket_oid_base
;
8639 int ret
= open_bucket_index_base(bucket_info
, index_ctx
, bucket_oid_base
);
8643 RGWObjectCtx
obj_ctx(this);
8645 ret
= get_bucket_index_object(bucket_oid_base
, obj_key
, bucket_info
.num_shards
,
8646 (RGWBucketInfo::BIShardsHashType
)bucket_info
.bucket_index_shard_hash_type
, bucket_obj
, shard_id
);
8648 ldout(cct
, 10) << "get_bucket_index_object() returned ret=" << ret
<< dendl
;
8654 int RGWRados::open_bucket_index_shard(const RGWBucketInfo
& bucket_info
, librados::IoCtx
& index_ctx
,
8655 int shard_id
, string
*bucket_obj
)
8657 string bucket_oid_base
;
8658 int ret
= open_bucket_index_base(bucket_info
, index_ctx
, bucket_oid_base
);
8662 RGWObjectCtx
obj_ctx(this);
8664 get_bucket_index_object(bucket_oid_base
, bucket_info
.num_shards
,
8665 shard_id
, bucket_obj
);
8669 static void accumulate_raw_stats(const rgw_bucket_dir_header
& header
,
8670 map
<RGWObjCategory
, RGWStorageStats
>& stats
)
8672 for (const auto& pair
: header
.stats
) {
8673 const RGWObjCategory category
= static_cast<RGWObjCategory
>(pair
.first
);
8674 const rgw_bucket_category_stats
& header_stats
= pair
.second
;
8676 RGWStorageStats
& s
= stats
[category
];
8678 s
.category
= category
;
8679 s
.size
+= header_stats
.total_size
;
8680 s
.size_rounded
+= header_stats
.total_size_rounded
;
8681 s
.size_utilized
+= header_stats
.actual_size
;
8682 s
.num_objects
+= header_stats
.num_entries
;
8686 int RGWRados::bucket_check_index(RGWBucketInfo
& bucket_info
,
8687 map
<RGWObjCategory
, RGWStorageStats
> *existing_stats
,
8688 map
<RGWObjCategory
, RGWStorageStats
> *calculated_stats
)
8690 librados::IoCtx index_ctx
;
8691 // key - bucket index object id
8692 // value - bucket index check OP returned result with the given bucket index object (shard)
8693 map
<int, string
> oids
;
8694 map
<int, struct rgw_cls_check_index_ret
> bucket_objs_ret
;
8696 int ret
= open_bucket_index(bucket_info
, index_ctx
, oids
, bucket_objs_ret
);
8701 ret
= CLSRGWIssueBucketCheck(index_ctx
, oids
, bucket_objs_ret
, cct
->_conf
->rgw_bucket_index_max_aio
)();
8706 // Aggregate results (from different shards if there is any)
8707 map
<int, struct rgw_cls_check_index_ret
>::iterator iter
;
8708 for (iter
= bucket_objs_ret
.begin(); iter
!= bucket_objs_ret
.end(); ++iter
) {
8709 accumulate_raw_stats(iter
->second
.existing_header
, *existing_stats
);
8710 accumulate_raw_stats(iter
->second
.calculated_header
, *calculated_stats
);
8716 int RGWRados::bucket_rebuild_index(RGWBucketInfo
& bucket_info
)
8718 librados::IoCtx index_ctx
;
8719 map
<int, string
> bucket_objs
;
8721 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
);
8726 return CLSRGWIssueBucketRebuild(index_ctx
, bucket_objs
, cct
->_conf
->rgw_bucket_index_max_aio
)();
8729 int RGWRados::bucket_set_reshard(RGWBucketInfo
& bucket_info
, const cls_rgw_bucket_instance_entry
& entry
)
8731 librados::IoCtx index_ctx
;
8732 map
<int, string
> bucket_objs
;
8734 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
);
8739 return CLSRGWIssueSetBucketResharding(index_ctx
, bucket_objs
, entry
, cct
->_conf
->rgw_bucket_index_max_aio
)();
8742 int RGWRados::defer_gc(void *ctx
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
)
8744 RGWObjectCtx
*rctx
= static_cast<RGWObjectCtx
*>(ctx
);
8745 std::string oid
, key
;
8746 get_obj_bucket_and_oid_loc(obj
, oid
, key
);
8750 RGWObjState
*state
= NULL
;
8752 int r
= get_obj_state(rctx
, bucket_info
, obj
, &state
, false);
8756 if (!state
->is_atomic
) {
8757 ldout(cct
, 20) << "state for obj=" << obj
<< " is not atomic, not deferring gc operation" << dendl
;
8763 if (state
->tail_tag
.length() > 0) {
8764 tag
= state
->tail_tag
.c_str();
8765 } else if (state
->obj_tag
.length() > 0) {
8766 tag
= state
->obj_tag
.c_str();
8768 ldout(cct
, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl
;
8772 ldout(cct
, 0) << "defer chain tag=" << tag
<< dendl
;
8774 return gc
->defer_chain(tag
, false);
8777 void RGWRados::remove_rgw_head_obj(ObjectWriteOperation
& op
)
8779 list
<string
> prefixes
;
8780 prefixes
.push_back(RGW_ATTR_OLH_PREFIX
);
8781 cls_rgw_remove_obj(op
, prefixes
);
8784 void RGWRados::cls_obj_check_prefix_exist(ObjectOperation
& op
, const string
& prefix
, bool fail_if_exist
)
8786 cls_rgw_obj_check_attrs_prefix(op
, prefix
, fail_if_exist
);
8789 void RGWRados::cls_obj_check_mtime(ObjectOperation
& op
, const real_time
& mtime
, bool high_precision_time
, RGWCheckMTimeType type
)
8791 cls_rgw_obj_check_mtime(op
, mtime
, high_precision_time
, type
);
8797 * bucket: name of the bucket storing the object
8798 * obj: name of the object to delete
8799 * Returns: 0 on success, -ERR# otherwise.
8801 int RGWRados::Object::Delete::delete_obj()
8803 RGWRados
*store
= target
->get_store();
8804 rgw_obj
& src_obj
= target
->get_obj();
8805 const string
& instance
= src_obj
.key
.instance
;
8806 rgw_obj obj
= src_obj
;
8808 if (instance
== "null") {
8809 obj
.key
.instance
.clear();
8812 bool explicit_marker_version
= (!params
.marker_version_id
.empty());
8814 if (params
.versioning_status
& BUCKET_VERSIONED
|| explicit_marker_version
) {
8815 if (instance
.empty() || explicit_marker_version
) {
8816 rgw_obj marker
= obj
;
8818 if (!params
.marker_version_id
.empty()) {
8819 if (params
.marker_version_id
!= "null") {
8820 marker
.key
.set_instance(params
.marker_version_id
);
8822 } else if ((params
.versioning_status
& BUCKET_VERSIONS_SUSPENDED
) == 0) {
8823 store
->gen_rand_obj_instance_name(&marker
);
8826 result
.version_id
= marker
.key
.instance
;
8827 result
.delete_marker
= true;
8829 struct rgw_bucket_dir_entry_meta meta
;
8831 meta
.owner
= params
.obj_owner
.get_id().to_str();
8832 meta
.owner_display_name
= params
.obj_owner
.get_display_name();
8834 if (real_clock::is_zero(params
.mtime
)) {
8835 meta
.mtime
= real_clock::now();
8837 meta
.mtime
= params
.mtime
;
8840 int r
= store
->set_olh(target
->get_ctx(), target
->get_bucket_info(), marker
, true, &meta
, params
.olh_epoch
, params
.unmod_since
, params
.high_precision_time
, params
.zones_trace
);
8845 rgw_bucket_dir_entry dirent
;
8847 int r
= store
->bi_get_instance(target
->get_bucket_info(), obj
, &dirent
);
8851 result
.delete_marker
= dirent
.is_delete_marker();
8852 r
= store
->unlink_obj_instance(target
->get_ctx(), target
->get_bucket_info(), obj
, params
.olh_epoch
, params
.zones_trace
);
8856 result
.version_id
= instance
;
8860 int r
= target
->get_bucket_shard(&bs
);
8862 ldout(store
->ctx(), 5) << "failed to get BucketShard object: r=" << r
<< dendl
;
8866 if (target
->bucket_info
.datasync_flag_enabled()) {
8867 r
= store
->data_log
->add_entry(bs
->bucket
, bs
->shard_id
);
8869 lderr(store
->ctx()) << "ERROR: failed writing data log" << dendl
;
8878 int r
= store
->get_obj_head_ref(target
->get_bucket_info(), obj
, &ref
);
8884 r
= target
->get_state(&state
, false);
8888 ObjectWriteOperation op
;
8890 if (!real_clock::is_zero(params
.unmod_since
)) {
8891 struct timespec ctime
= ceph::real_clock::to_timespec(state
->mtime
);
8892 struct timespec unmod
= ceph::real_clock::to_timespec(params
.unmod_since
);
8893 if (!params
.high_precision_time
) {
8898 ldout(store
->ctx(), 10) << "If-UnModified-Since: " << params
.unmod_since
<< " Last-Modified: " << ctime
<< dendl
;
8899 if (ctime
> unmod
) {
8900 return -ERR_PRECONDITION_FAILED
;
8903 /* only delete object if mtime is less than or equal to params.unmod_since */
8904 store
->cls_obj_check_mtime(op
, params
.unmod_since
, params
.high_precision_time
, CLS_RGW_CHECK_TIME_MTIME_LE
);
8906 uint64_t obj_size
= state
->size
;
8908 if (!real_clock::is_zero(params
.expiration_time
)) {
8910 real_time delete_at
;
8912 if (state
->get_attr(RGW_ATTR_DELETE_AT
, bl
)) {
8914 bufferlist::iterator iter
= bl
.begin();
8915 ::decode(delete_at
, iter
);
8916 } catch (buffer::error
& err
) {
8917 ldout(store
->ctx(), 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl
;
8921 if (params
.expiration_time
!= delete_at
) {
8922 return -ERR_PRECONDITION_FAILED
;
8925 return -ERR_PRECONDITION_FAILED
;
8929 if (!state
->exists
) {
8930 target
->invalidate_state();
8934 r
= target
->prepare_atomic_modification(op
, false, NULL
, NULL
, NULL
, true, false);
8938 RGWBucketInfo
& bucket_info
= target
->get_bucket_info();
8940 RGWRados::Bucket
bop(store
, bucket_info
);
8941 RGWRados::Bucket::UpdateIndex
index_op(&bop
, obj
);
8943 index_op
.set_zones_trace(params
.zones_trace
);
8944 index_op
.set_bilog_flags(params
.bilog_flags
);
8946 r
= index_op
.prepare(CLS_RGW_OP_DEL
, &state
->write_tag
);
8950 store
->remove_rgw_head_obj(op
);
8951 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
8952 bool need_invalidate
= false;
8953 if (r
== -ECANCELED
) {
8954 /* raced with another operation, we can regard it as removed */
8955 need_invalidate
= true;
8959 int64_t poolid
= ref
.ioctx
.get_id();
8961 tombstone_cache_t
*obj_tombstone_cache
= store
->get_tombstone_cache();
8962 if (obj_tombstone_cache
) {
8963 tombstone_entry entry
{*state
};
8964 obj_tombstone_cache
->add(obj
, entry
);
8966 r
= index_op
.complete_del(poolid
, ref
.ioctx
.get_last_version(), state
->mtime
, params
.remove_objs
);
8968 int ret
= target
->complete_atomic_modification();
8970 ldout(store
->ctx(), 0) << "ERROR: complete_atomic_modification returned ret=" << ret
<< dendl
;
8972 /* other than that, no need to propagate error */
8974 int ret
= index_op
.cancel();
8976 ldout(store
->ctx(), 0) << "ERROR: index_op.cancel() returned ret=" << ret
<< dendl
;
8980 if (need_invalidate
) {
8981 target
->invalidate_state();
8987 /* update quota cache */
8988 store
->quota_handler
->update_stats(params
.bucket_owner
, obj
.bucket
, -1, 0, obj_size
);
8993 int RGWRados::delete_obj(RGWObjectCtx
& obj_ctx
,
8994 const RGWBucketInfo
& bucket_info
,
8996 int versioning_status
,
8997 uint16_t bilog_flags
,
8998 const real_time
& expiration_time
,
8999 rgw_zone_set
*zones_trace
)
9001 RGWRados::Object
del_target(this, bucket_info
, obj_ctx
, obj
);
9002 RGWRados::Object::Delete
del_op(&del_target
);
9004 del_op
.params
.bucket_owner
= bucket_info
.owner
;
9005 del_op
.params
.versioning_status
= versioning_status
;
9006 del_op
.params
.bilog_flags
= bilog_flags
;
9007 del_op
.params
.expiration_time
= expiration_time
;
9008 del_op
.params
.zones_trace
= zones_trace
;
9010 return del_op
.delete_obj();
9013 int RGWRados::delete_raw_obj(const rgw_raw_obj
& obj
)
9016 int r
= get_raw_obj_ref(obj
, &ref
);
9021 ObjectWriteOperation op
;
9024 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
9031 int RGWRados::delete_system_obj(rgw_raw_obj
& obj
, RGWObjVersionTracker
*objv_tracker
)
9034 ldout(cct
, 1) << "delete_system_obj got empty object name "
9035 << obj
<< ", returning EINVAL" << dendl
;
9039 int r
= get_raw_obj_ref(obj
, &ref
);
9044 ObjectWriteOperation op
;
9047 objv_tracker
->prepare_op_for_write(&op
);
9051 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
9058 int RGWRados::delete_obj_index(const rgw_obj
& obj
)
9060 std::string oid
, key
;
9061 get_obj_bucket_and_oid_loc(obj
, oid
, key
);
9063 RGWObjectCtx
obj_ctx(this);
9065 RGWBucketInfo bucket_info
;
9066 int ret
= get_bucket_instance_info(obj_ctx
, obj
.bucket
, bucket_info
, NULL
, NULL
);
9068 ldout(cct
, 0) << "ERROR: " << __func__
<< "() get_bucket_instance_info(bucket=" << obj
.bucket
<< ") returned ret=" << ret
<< dendl
;
9072 RGWRados::Bucket
bop(this, bucket_info
);
9073 RGWRados::Bucket::UpdateIndex
index_op(&bop
, obj
);
9075 real_time removed_mtime
;
9076 int r
= index_op
.complete_del(-1 /* pool */, 0, removed_mtime
, NULL
);
9081 static void generate_fake_tag(RGWRados
*store
, map
<string
, bufferlist
>& attrset
, RGWObjManifest
& manifest
, bufferlist
& manifest_bl
, bufferlist
& tag_bl
)
9085 RGWObjManifest::obj_iterator mi
= manifest
.obj_begin();
9086 if (mi
!= manifest
.obj_end()) {
9087 if (manifest
.has_tail()) // first object usually points at the head, let's skip to a more unique part
9089 tag
= mi
.get_location().get_raw_obj(store
).oid
;
9093 unsigned char md5
[CEPH_CRYPTO_MD5_DIGESTSIZE
];
9094 char md5_str
[CEPH_CRYPTO_MD5_DIGESTSIZE
* 2 + 1];
9096 hash
.Update((const byte
*)manifest_bl
.c_str(), manifest_bl
.length());
9098 map
<string
, bufferlist
>::iterator iter
= attrset
.find(RGW_ATTR_ETAG
);
9099 if (iter
!= attrset
.end()) {
9100 bufferlist
& bl
= iter
->second
;
9101 hash
.Update((const byte
*)bl
.c_str(), bl
.length());
9105 buf_to_hex(md5
, CEPH_CRYPTO_MD5_DIGESTSIZE
, md5_str
);
9106 tag
.append(md5_str
);
9108 ldout(store
->ctx(), 10) << "generate_fake_tag new tag=" << tag
<< dendl
;
9110 tag_bl
.append(tag
.c_str(), tag
.size() + 1);
9113 static bool is_olh(map
<string
, bufferlist
>& attrs
)
9115 map
<string
, bufferlist
>::iterator iter
= attrs
.find(RGW_ATTR_OLH_INFO
);
9116 return (iter
!= attrs
.end());
9119 static bool has_olh_tag(map
<string
, bufferlist
>& attrs
)
9121 map
<string
, bufferlist
>::iterator iter
= attrs
.find(RGW_ATTR_OLH_ID_TAG
);
9122 return (iter
!= attrs
.end());
9125 int RGWRados::get_olh_target_state(RGWObjectCtx
& obj_ctx
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
,
9126 RGWObjState
*olh_state
, RGWObjState
**target_state
)
9128 assert(olh_state
->is_olh
);
9131 int r
= RGWRados::follow_olh(bucket_info
, obj_ctx
, olh_state
, obj
, &target
); /* might return -EAGAIN */
9135 r
= get_obj_state(&obj_ctx
, bucket_info
, target
, target_state
, false);
9143 int RGWRados::get_system_obj_state_impl(RGWObjectCtx
*rctx
, rgw_raw_obj
& obj
, RGWRawObjState
**state
, RGWObjVersionTracker
*objv_tracker
)
9149 RGWRawObjState
*s
= rctx
->raw
.get_state(obj
);
9150 ldout(cct
, 20) << "get_system_obj_state: rctx=" << (void *)rctx
<< " obj=" << obj
<< " state=" << (void *)s
<< " s->prefetch_data=" << s
->prefetch_data
<< dendl
;
9158 int r
= raw_obj_stat(obj
, &s
->size
, &s
->mtime
, &s
->epoch
, &s
->attrset
, (s
->prefetch_data
? &s
->data
: NULL
), objv_tracker
);
9161 s
->has_attrs
= true;
9162 s
->mtime
= real_time();
9169 s
->has_attrs
= true;
9170 s
->obj_tag
= s
->attrset
[RGW_ATTR_ID_TAG
];
9172 if (s
->obj_tag
.length())
9173 ldout(cct
, 20) << "get_system_obj_state: setting s->obj_tag to "
9174 << s
->obj_tag
.c_str() << dendl
;
9176 ldout(cct
, 20) << "get_system_obj_state: s->obj_tag was set empty" << dendl
;
9181 int RGWRados::get_system_obj_state(RGWObjectCtx
*rctx
, rgw_raw_obj
& obj
, RGWRawObjState
**state
, RGWObjVersionTracker
*objv_tracker
)
9186 ret
= get_system_obj_state_impl(rctx
, obj
, state
, objv_tracker
);
9187 } while (ret
== -EAGAIN
);
9192 int RGWRados::get_obj_state_impl(RGWObjectCtx
*rctx
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
,
9193 RGWObjState
**state
, bool follow_olh
, bool assume_noent
)
9199 bool need_follow_olh
= follow_olh
&& obj
.key
.instance
.empty();
9201 RGWObjState
*s
= rctx
->obj
.get_state(obj
);
9202 ldout(cct
, 20) << "get_obj_state: rctx=" << (void *)rctx
<< " obj=" << obj
<< " state=" << (void *)s
<< " s->prefetch_data=" << s
->prefetch_data
<< dendl
;
9205 if (s
->is_olh
&& need_follow_olh
) {
9206 return get_olh_target_state(*rctx
, bucket_info
, obj
, s
, state
);
9213 rgw_raw_obj raw_obj
;
9214 obj_to_raw(bucket_info
.placement_rule
, obj
, &raw_obj
);
9218 if (!assume_noent
) {
9219 r
= RGWRados::raw_obj_stat(raw_obj
, &s
->size
, &s
->mtime
, &s
->epoch
, &s
->attrset
, (s
->prefetch_data
? &s
->data
: NULL
), NULL
);
9224 s
->has_attrs
= true;
9225 tombstone_entry entry
;
9226 if (obj_tombstone_cache
&& obj_tombstone_cache
->find(obj
, entry
)) {
9227 s
->mtime
= entry
.mtime
;
9228 s
->zone_short_id
= entry
.zone_short_id
;
9229 s
->pg_ver
= entry
.pg_ver
;
9230 ldout(cct
, 20) << __func__
<< "(): found obj in tombstone cache: obj=" << obj
9231 << " mtime=" << s
->mtime
<< " pgv=" << s
->pg_ver
<< dendl
;
9233 s
->mtime
= real_time();
9241 s
->has_attrs
= true;
9242 s
->accounted_size
= s
->size
;
9244 auto iter
= s
->attrset
.find(RGW_ATTR_COMPRESSION
);
9245 const bool compressed
= (iter
!= s
->attrset
.end());
9247 // use uncompressed size for accounted_size
9249 RGWCompressionInfo info
;
9250 auto p
= iter
->second
.begin();
9252 s
->accounted_size
= info
.orig_size
;
9253 } catch (buffer::error
&) {
9254 dout(0) << "ERROR: could not decode compression info for object: " << obj
<< dendl
;
9259 iter
= s
->attrset
.find(RGW_ATTR_SHADOW_OBJ
);
9260 if (iter
!= s
->attrset
.end()) {
9261 bufferlist bl
= iter
->second
;
9262 bufferlist::iterator it
= bl
.begin();
9263 it
.copy(bl
.length(), s
->shadow_obj
);
9264 s
->shadow_obj
[bl
.length()] = '\0';
9266 s
->obj_tag
= s
->attrset
[RGW_ATTR_ID_TAG
];
9267 auto ttiter
= s
->attrset
.find(RGW_ATTR_TAIL_TAG
);
9268 if (ttiter
!= s
->attrset
.end()) {
9269 s
->tail_tag
= s
->attrset
[RGW_ATTR_TAIL_TAG
];
9272 bufferlist manifest_bl
= s
->attrset
[RGW_ATTR_MANIFEST
];
9273 if (manifest_bl
.length()) {
9274 bufferlist::iterator miter
= manifest_bl
.begin();
9276 ::decode(s
->manifest
, miter
);
9277 s
->has_manifest
= true;
9278 s
->manifest
.set_head(bucket_info
.placement_rule
, obj
, s
->size
); /* patch manifest to reflect the head we just read, some manifests might be
9279 broken due to old bugs */
9280 s
->size
= s
->manifest
.get_obj_size();
9282 s
->accounted_size
= s
->size
;
9283 } catch (buffer::error
& err
) {
9284 ldout(cct
, 0) << "ERROR: couldn't decode manifest" << dendl
;
9287 ldout(cct
, 10) << "manifest: total_size = " << s
->manifest
.get_obj_size() << dendl
;
9288 if (cct
->_conf
->subsys
.should_gather(ceph_subsys_rgw
, 20) && s
->manifest
.has_explicit_objs()) {
9289 RGWObjManifest::obj_iterator mi
;
9290 for (mi
= s
->manifest
.obj_begin(); mi
!= s
->manifest
.obj_end(); ++mi
) {
9291 ldout(cct
, 20) << "manifest: ofs=" << mi
.get_ofs() << " loc=" << mi
.get_location().get_raw_obj(this) << dendl
;
9295 if (!s
->obj_tag
.length()) {
9297 * Uh oh, something's wrong, object with manifest should have tag. Let's
9298 * create one out of the manifest, would be unique
9300 generate_fake_tag(this, s
->attrset
, s
->manifest
, manifest_bl
, s
->obj_tag
);
9304 map
<string
, bufferlist
>::iterator aiter
= s
->attrset
.find(RGW_ATTR_PG_VER
);
9305 if (aiter
!= s
->attrset
.end()) {
9306 bufferlist
& pg_ver_bl
= aiter
->second
;
9307 if (pg_ver_bl
.length()) {
9308 bufferlist::iterator pgbl
= pg_ver_bl
.begin();
9310 ::decode(s
->pg_ver
, pgbl
);
9311 } catch (buffer::error
& err
) {
9312 ldout(cct
, 0) << "ERROR: couldn't decode pg ver attr for object " << s
->obj
<< ", non-critical error, ignoring" << dendl
;
9316 aiter
= s
->attrset
.find(RGW_ATTR_SOURCE_ZONE
);
9317 if (aiter
!= s
->attrset
.end()) {
9318 bufferlist
& zone_short_id_bl
= aiter
->second
;
9319 if (zone_short_id_bl
.length()) {
9320 bufferlist::iterator zbl
= zone_short_id_bl
.begin();
9322 ::decode(s
->zone_short_id
, zbl
);
9323 } catch (buffer::error
& err
) {
9324 ldout(cct
, 0) << "ERROR: couldn't decode zone short id attr for object " << s
->obj
<< ", non-critical error, ignoring" << dendl
;
9328 if (s
->obj_tag
.length())
9329 ldout(cct
, 20) << "get_obj_state: setting s->obj_tag to " << s
->obj_tag
.c_str() << dendl
;
9331 ldout(cct
, 20) << "get_obj_state: s->obj_tag was set empty" << dendl
;
9333 /* an object might not be olh yet, but could have olh id tag, so we should set it anyway if
9334 * it exist, and not only if is_olh() returns true
9336 iter
= s
->attrset
.find(RGW_ATTR_OLH_ID_TAG
);
9337 if (iter
!= s
->attrset
.end()) {
9338 s
->olh_tag
= iter
->second
;
9341 if (is_olh(s
->attrset
)) {
9344 ldout(cct
, 20) << __func__
<< ": setting s->olh_tag to " << string(s
->olh_tag
.c_str(), s
->olh_tag
.length()) << dendl
;
9346 if (need_follow_olh
) {
9347 return get_olh_target_state(*rctx
, bucket_info
, obj
, s
, state
);
9354 int RGWRados::get_obj_state(RGWObjectCtx
*rctx
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, RGWObjState
**state
,
9355 bool follow_olh
, bool assume_noent
)
9360 ret
= get_obj_state_impl(rctx
, bucket_info
, obj
, state
, follow_olh
, assume_noent
);
9361 } while (ret
== -EAGAIN
);
9366 int RGWRados::Object::get_manifest(RGWObjManifest
**pmanifest
)
9368 RGWObjState
*astate
;
9369 int r
= get_state(&astate
, true);
9374 *pmanifest
= &astate
->manifest
;
9379 int RGWRados::Object::Read::get_attr(const char *name
, bufferlist
& dest
)
9382 int r
= source
->get_state(&state
, true);
9387 if (!state
->get_attr(name
, dest
))
9394 int RGWRados::Object::Stat::stat_async()
9396 RGWObjectCtx
& ctx
= source
->get_ctx();
9397 rgw_obj
& obj
= source
->get_obj();
9398 RGWRados
*store
= source
->get_store();
9400 RGWObjState
*s
= ctx
.obj
.get_state(obj
); /* calling this one directly because otherwise a sync request will be sent */
9404 result
.size
= s
->size
;
9405 result
.mtime
= ceph::real_clock::to_timespec(s
->mtime
);
9406 result
.attrs
= s
->attrset
;
9407 result
.has_manifest
= s
->has_manifest
;
9408 result
.manifest
= s
->manifest
;
9414 get_obj_bucket_and_oid_loc(obj
, oid
, loc
);
9416 int r
= store
->get_obj_head_ioctx(source
->get_bucket_info(), obj
, &state
.io_ctx
);
9421 librados::ObjectReadOperation op
;
9422 op
.stat2(&result
.size
, &result
.mtime
, NULL
);
9423 op
.getxattrs(&result
.attrs
, NULL
);
9424 state
.completion
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
9425 state
.io_ctx
.locator_set_key(loc
);
9426 r
= state
.io_ctx
.aio_operate(oid
, state
.completion
, &op
, NULL
);
9428 ldout(store
->ctx(), 5) << __func__
9429 << ": ERROR: aio_operate() returned ret=" << r
9438 int RGWRados::Object::Stat::wait()
9440 if (!state
.completion
) {
9444 state
.completion
->wait_for_safe();
9445 state
.ret
= state
.completion
->get_return_value();
9446 state
.completion
->release();
9448 if (state
.ret
!= 0) {
9455 int RGWRados::Object::Stat::finish()
9457 map
<string
, bufferlist
>::iterator iter
= result
.attrs
.find(RGW_ATTR_MANIFEST
);
9458 if (iter
!= result
.attrs
.end()) {
9459 bufferlist
& bl
= iter
->second
;
9460 bufferlist::iterator biter
= bl
.begin();
9462 ::decode(result
.manifest
, biter
);
9463 } catch (buffer::error
& err
) {
9464 RGWRados
*store
= source
->get_store();
9465 ldout(store
->ctx(), 0) << "ERROR: " << __func__
<< ": failed to decode manifest" << dendl
;
9468 result
.has_manifest
= true;
9475 * Get an attribute for a system object.
9476 * obj: the object to get attr
9477 * name: name of the attr to retrieve
9478 * dest: bufferlist to store the result in
9479 * Returns: 0 on success, -ERR# otherwise.
9481 int RGWRados::system_obj_get_attr(rgw_raw_obj
& obj
, const char *name
, bufferlist
& dest
)
9484 int r
= get_system_obj_ref(obj
, &ref
);
9489 ObjectReadOperation op
;
9492 op
.getxattr(name
, &dest
, &rval
);
9494 r
= ref
.ioctx
.operate(ref
.oid
, &op
, NULL
);
9501 int RGWRados::append_atomic_test(RGWObjectCtx
*rctx
,
9502 const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
,
9503 ObjectOperation
& op
, RGWObjState
**pstate
)
9508 int r
= get_obj_state(rctx
, bucket_info
, obj
, pstate
, false);
9512 RGWObjState
*state
= *pstate
;
9514 if (!state
->is_atomic
) {
9515 ldout(cct
, 20) << "state for obj=" << obj
<< " is not atomic, not appending atomic test" << dendl
;
9519 if (state
->obj_tag
.length() > 0 && !state
->fake_tag
) {// check for backward compatibility
9520 op
.cmpxattr(RGW_ATTR_ID_TAG
, LIBRADOS_CMPXATTR_OP_EQ
, state
->obj_tag
);
9522 ldout(cct
, 20) << "state->obj_tag is empty, not appending atomic test" << dendl
;
9527 int RGWRados::Object::get_state(RGWObjState
**pstate
, bool follow_olh
, bool assume_noent
)
9529 return store
->get_obj_state(&ctx
, bucket_info
, obj
, pstate
, follow_olh
, assume_noent
);
9532 void RGWRados::Object::invalidate_state()
9534 ctx
.obj
.invalidate(obj
);
9537 void RGWRados::SystemObject::invalidate_state()
9539 ctx
.raw
.invalidate(obj
);
9542 int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation
& op
, bool reset_obj
, const string
*ptag
,
9543 const char *if_match
, const char *if_nomatch
, bool removal_op
,
9546 int r
= get_state(&state
, false);
9550 bool need_guard
= (state
->has_manifest
|| (state
->obj_tag
.length() != 0) ||
9551 if_match
!= NULL
|| if_nomatch
!= NULL
) &&
9554 if (!state
->is_atomic
) {
9555 ldout(store
->ctx(), 20) << "prepare_atomic_modification: state is not atomic. state=" << (void *)state
<< dendl
;
9559 store
->remove_rgw_head_obj(op
); // we're not dropping reference here, actually removing object
9566 /* first verify that the object wasn't replaced under */
9567 if (if_nomatch
== NULL
|| strcmp(if_nomatch
, "*") != 0) {
9568 op
.cmpxattr(RGW_ATTR_ID_TAG
, LIBRADOS_CMPXATTR_OP_EQ
, state
->obj_tag
);
9569 // FIXME: need to add FAIL_NOTEXIST_OK for racing deletion
9573 if (strcmp(if_match
, "*") == 0) {
9574 // test the object is existing
9575 if (!state
->exists
) {
9576 return -ERR_PRECONDITION_FAILED
;
9580 if (!state
->get_attr(RGW_ATTR_ETAG
, bl
) ||
9581 strncmp(if_match
, bl
.c_str(), bl
.length()) != 0) {
9582 return -ERR_PRECONDITION_FAILED
;
9588 if (strcmp(if_nomatch
, "*") == 0) {
9589 // test the object is NOT existing
9590 if (state
->exists
) {
9591 return -ERR_PRECONDITION_FAILED
;
9595 if (!state
->get_attr(RGW_ATTR_ETAG
, bl
) ||
9596 strncmp(if_nomatch
, bl
.c_str(), bl
.length()) == 0) {
9597 return -ERR_PRECONDITION_FAILED
;
9604 if (state
->exists
) {
9606 store
->remove_rgw_head_obj(op
);
9613 /* the object is being removed, no need to update its tag */
9618 state
->write_tag
= *ptag
;
9620 append_rand_alpha(store
->ctx(), state
->write_tag
, state
->write_tag
, 32);
9623 bl
.append(state
->write_tag
.c_str(), state
->write_tag
.size() + 1);
9625 ldout(store
->ctx(), 10) << "setting object write_tag=" << state
->write_tag
<< dendl
;
9627 op
.setxattr(RGW_ATTR_ID_TAG
, bl
);
9629 op
.setxattr(RGW_ATTR_TAIL_TAG
, bl
);
9635 int RGWRados::system_obj_set_attr(void *ctx
, rgw_raw_obj
& obj
, const char *name
, bufferlist
& bl
,
9636 RGWObjVersionTracker
*objv_tracker
)
9638 map
<string
, bufferlist
> attrs
;
9640 return system_obj_set_attrs(ctx
, obj
, attrs
, NULL
, objv_tracker
);
9643 int RGWRados::system_obj_set_attrs(void *ctx
, rgw_raw_obj
& obj
,
9644 map
<string
, bufferlist
>& attrs
,
9645 map
<string
, bufferlist
>* rmattrs
,
9646 RGWObjVersionTracker
*objv_tracker
)
9649 int r
= get_system_obj_ref(obj
, &ref
);
9653 ObjectWriteOperation op
;
9656 objv_tracker
->prepare_op_for_write(&op
);
9659 map
<string
, bufferlist
>::iterator iter
;
9661 for (iter
= rmattrs
->begin(); iter
!= rmattrs
->end(); ++iter
) {
9662 const string
& name
= iter
->first
;
9663 op
.rmxattr(name
.c_str());
9667 for (iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
9668 const string
& name
= iter
->first
;
9669 bufferlist
& bl
= iter
->second
;
9674 op
.setxattr(name
.c_str(), bl
);
9682 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
9690 * Set an attr on an object.
9691 * bucket: name of the bucket holding the object
9692 * obj: name of the object to set the attr on
9693 * name: the attr to set
9694 * bl: the contents of the attr
9695 * Returns: 0 on success, -ERR# otherwise.
9697 int RGWRados::set_attr(void *ctx
, const RGWBucketInfo
& bucket_info
, rgw_obj
& obj
, const char *name
, bufferlist
& bl
)
9699 map
<string
, bufferlist
> attrs
;
9701 return set_attrs(ctx
, bucket_info
, obj
, attrs
, NULL
);
9704 int RGWRados::set_attrs(void *ctx
, const RGWBucketInfo
& bucket_info
, rgw_obj
& obj
,
9705 map
<string
, bufferlist
>& attrs
,
9706 map
<string
, bufferlist
>* rmattrs
)
9709 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
9713 RGWObjectCtx
*rctx
= static_cast<RGWObjectCtx
*>(ctx
);
9715 ObjectWriteOperation op
;
9716 RGWObjState
*state
= NULL
;
9718 r
= append_atomic_test(rctx
, bucket_info
, obj
, op
, &state
);
9722 map
<string
, bufferlist
>::iterator iter
;
9724 for (iter
= rmattrs
->begin(); iter
!= rmattrs
->end(); ++iter
) {
9725 const string
& name
= iter
->first
;
9726 op
.rmxattr(name
.c_str());
9730 const rgw_bucket
& bucket
= obj
.bucket
;
9732 for (iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
9733 const string
& name
= iter
->first
;
9734 bufferlist
& bl
= iter
->second
;
9739 op
.setxattr(name
.c_str(), bl
);
9741 if (name
.compare(RGW_ATTR_DELETE_AT
) == 0) {
9746 rgw_obj_index_key obj_key
;
9747 obj
.key
.get_index_key(&obj_key
);
9749 objexp_hint_add(ts
, bucket
.tenant
, bucket
.name
, bucket
.bucket_id
, obj_key
);
9750 } catch (buffer::error
& err
) {
9751 ldout(cct
, 0) << "ERROR: failed to decode " RGW_ATTR_DELETE_AT
<< " attr" << dendl
;
9759 RGWObjectCtx
obj_ctx(this);
9762 RGWRados::Bucket
bop(this, bucket_info
);
9763 RGWRados::Bucket::UpdateIndex
index_op(&bop
, obj
);
9767 append_rand_alpha(cct
, tag
, tag
, 32);
9768 state
->write_tag
= tag
;
9769 r
= index_op
.prepare(CLS_RGW_OP_ADD
, &state
->write_tag
);
9774 bl
.append(tag
.c_str(), tag
.size() + 1);
9775 op
.setxattr(RGW_ATTR_ID_TAG
, bl
);
9779 real_time mtime
= real_clock::now();
9780 struct timespec mtime_ts
= real_clock::to_timespec(mtime
);
9781 op
.mtime2(&mtime_ts
);
9782 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
9785 bufferlist acl_bl
= attrs
[RGW_ATTR_ACL
];
9786 bufferlist etag_bl
= attrs
[RGW_ATTR_ETAG
];
9787 bufferlist content_type_bl
= attrs
[RGW_ATTR_CONTENT_TYPE
];
9788 string
etag(etag_bl
.c_str(), etag_bl
.length());
9789 string
content_type(content_type_bl
.c_str(), content_type_bl
.length());
9790 uint64_t epoch
= ref
.ioctx
.get_last_version();
9791 int64_t poolid
= ref
.ioctx
.get_id();
9792 r
= index_op
.complete(poolid
, epoch
, state
->size
, state
->accounted_size
,
9793 mtime
, etag
, content_type
, &acl_bl
,
9794 RGW_OBJ_CATEGORY_MAIN
, NULL
);
9796 int ret
= index_op
.cancel();
9798 ldout(cct
, 0) << "ERROR: complete_update_index_cancel() returned ret=" << ret
<< dendl
;
9806 state
->obj_tag
.swap(bl
);
9808 for (iter
= rmattrs
->begin(); iter
!= rmattrs
->end(); ++iter
) {
9809 state
->attrset
.erase(iter
->first
);
9812 for (iter
= attrs
.begin(); iter
!= attrs
.end(); ++iter
) {
9813 state
->attrset
[iter
->first
] = iter
->second
;
9820 int RGWRados::Object::Read::prepare()
9822 RGWRados
*store
= source
->get_store();
9823 CephContext
*cct
= store
->ctx();
9827 map
<string
, bufferlist
>::iterator iter
;
9829 RGWObjState
*astate
;
9830 int r
= source
->get_state(&astate
, true);
9834 if (!astate
->exists
) {
9838 const RGWBucketInfo
& bucket_info
= source
->get_bucket_info();
9840 state
.obj
= astate
->obj
;
9841 store
->obj_to_raw(bucket_info
.placement_rule
, state
.obj
, &state
.head_obj
);
9843 r
= store
->get_obj_head_ioctx(bucket_info
, state
.obj
, &state
.io_ctx
);
9848 *params
.attrs
= astate
->attrset
;
9849 if (cct
->_conf
->subsys
.should_gather(ceph_subsys_rgw
, 20)) {
9850 for (iter
= params
.attrs
->begin(); iter
!= params
.attrs
->end(); ++iter
) {
9851 ldout(cct
, 20) << "Read xattr: " << iter
->first
<< dendl
;
9856 /* Convert all times go GMT to make them compatible */
9857 if (conds
.mod_ptr
|| conds
.unmod_ptr
) {
9858 obj_time_weight src_weight
;
9859 src_weight
.init(astate
);
9860 src_weight
.high_precision
= conds
.high_precision_time
;
9862 obj_time_weight dest_weight
;
9863 dest_weight
.high_precision
= conds
.high_precision_time
;
9865 if (conds
.mod_ptr
) {
9866 dest_weight
.init(*conds
.mod_ptr
, conds
.mod_zone_id
, conds
.mod_pg_ver
);
9867 ldout(cct
, 10) << "If-Modified-Since: " << dest_weight
<< " Last-Modified: " << src_weight
<< dendl
;
9868 if (!(dest_weight
< src_weight
)) {
9869 return -ERR_NOT_MODIFIED
;
9873 if (conds
.unmod_ptr
) {
9874 dest_weight
.init(*conds
.unmod_ptr
, conds
.mod_zone_id
, conds
.mod_pg_ver
);
9875 ldout(cct
, 10) << "If-UnModified-Since: " << dest_weight
<< " Last-Modified: " << src_weight
<< dendl
;
9876 if (dest_weight
< src_weight
) {
9877 return -ERR_PRECONDITION_FAILED
;
9881 if (conds
.if_match
|| conds
.if_nomatch
) {
9882 r
= get_attr(RGW_ATTR_ETAG
, etag
);
9886 if (conds
.if_match
) {
9887 string if_match_str
= rgw_string_unquote(conds
.if_match
);
9888 ldout(cct
, 10) << "ETag: " << etag
.c_str() << " " << " If-Match: " << if_match_str
<< dendl
;
9889 if (if_match_str
.compare(etag
.c_str()) != 0) {
9890 return -ERR_PRECONDITION_FAILED
;
9894 if (conds
.if_nomatch
) {
9895 string if_nomatch_str
= rgw_string_unquote(conds
.if_nomatch
);
9896 ldout(cct
, 10) << "ETag: " << etag
.c_str() << " " << " If-NoMatch: " << if_nomatch_str
<< dendl
;
9897 if (if_nomatch_str
.compare(etag
.c_str()) == 0) {
9898 return -ERR_NOT_MODIFIED
;
9903 if (params
.obj_size
)
9904 *params
.obj_size
= astate
->size
;
9906 *params
.lastmod
= astate
->mtime
;
9911 int RGWRados::Object::Read::range_to_ofs(uint64_t obj_size
, int64_t &ofs
, int64_t &end
)
9918 } else if (end
< 0) {
9923 if (ofs
>= (off_t
)obj_size
) {
9926 if (end
>= (off_t
)obj_size
) {
9933 int RGWRados::SystemObject::get_state(RGWRawObjState
**pstate
, RGWObjVersionTracker
*objv_tracker
)
9935 return store
->get_system_obj_state(&ctx
, obj
, pstate
, objv_tracker
);
9938 int RGWRados::stat_system_obj(RGWObjectCtx
& obj_ctx
,
9939 RGWRados::SystemObject::Read::GetObjState
& state
,
9941 map
<string
, bufferlist
> *attrs
,
9944 RGWObjVersionTracker
*objv_tracker
)
9946 RGWRawObjState
*astate
= NULL
;
9948 int r
= get_system_obj_state(&obj_ctx
, obj
, &astate
, objv_tracker
);
9952 if (!astate
->exists
) {
9957 *attrs
= astate
->attrset
;
9958 if (cct
->_conf
->subsys
.should_gather(ceph_subsys_rgw
, 20)) {
9959 map
<string
, bufferlist
>::iterator iter
;
9960 for (iter
= attrs
->begin(); iter
!= attrs
->end(); ++iter
) {
9961 ldout(cct
, 20) << "Read xattr: " << iter
->first
<< dendl
;
9967 *obj_size
= astate
->size
;
9969 *lastmod
= astate
->mtime
;
9975 int RGWRados::Bucket::UpdateIndex::guard_reshard(BucketShard
**pbs
, std::function
<int(BucketShard
*)> call
)
9977 RGWRados
*store
= target
->get_store();
9981 #define NUM_RESHARD_RETRIES 10
9982 for (int i
= 0; i
< NUM_RESHARD_RETRIES
; ++i
) {
9983 int ret
= get_bucket_shard(&bs
);
9985 ldout(store
->ctx(), 5) << "failed to get BucketShard object: ret=" << ret
<< dendl
;
9989 if (r
!= -ERR_BUSY_RESHARDING
) {
9992 ldout(store
->ctx(), 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl
;
9993 string new_bucket_id
;
9994 r
= store
->block_while_resharding(bs
, &new_bucket_id
);
9995 if (r
== -ERR_BUSY_RESHARDING
) {
10001 ldout(store
->ctx(), 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id
<< dendl
;
10002 i
= 0; /* resharding is finished, make sure we can retry */
10003 r
= target
->update_bucket_id(new_bucket_id
);
10005 ldout(store
->ctx(), 0) << "ERROR: update_bucket_id() new_bucket_id=" << new_bucket_id
<< " returned r=" << r
<< dendl
;
10022 int RGWRados::SystemObject::Read::stat(RGWObjVersionTracker
*objv_tracker
)
10024 RGWRados
*store
= source
->get_store();
10025 rgw_raw_obj
& obj
= source
->get_obj();
10027 return store
->stat_system_obj(source
->get_ctx(), state
, obj
, stat_params
.attrs
,
10028 stat_params
.lastmod
, stat_params
.obj_size
, objv_tracker
);
10031 int RGWRados::Bucket::UpdateIndex::prepare(RGWModifyOp op
, const string
*write_tag
)
10036 RGWRados
*store
= target
->get_store();
10038 if (write_tag
&& write_tag
->length()) {
10039 optag
= string(write_tag
->c_str(), write_tag
->length());
10041 if (optag
.empty()) {
10042 append_rand_alpha(store
->ctx(), optag
, optag
, 32);
10046 int r
= guard_reshard(nullptr, [&](BucketShard
*bs
) -> int {
10047 return store
->cls_obj_prepare_op(*bs
, op
, optag
, obj
, bilog_flags
, zones_trace
);
10058 int RGWRados::Bucket::UpdateIndex::complete(int64_t poolid
, uint64_t epoch
,
10059 uint64_t size
, uint64_t accounted_size
,
10060 ceph::real_time
& ut
, const string
& etag
,
10061 const string
& content_type
,
10062 bufferlist
*acl_bl
,
10063 RGWObjCategory category
,
10064 list
<rgw_obj_index_key
> *remove_objs
, const string
*user_data
)
10069 RGWRados
*store
= target
->get_store();
10072 int ret
= get_bucket_shard(&bs
);
10074 ldout(store
->ctx(), 5) << "failed to get BucketShard object: ret=" << ret
<< dendl
;
10078 rgw_bucket_dir_entry ent
;
10079 obj
.key
.get_index_key(&ent
.key
);
10080 ent
.meta
.size
= size
;
10081 ent
.meta
.accounted_size
= accounted_size
;
10082 ent
.meta
.mtime
= ut
;
10083 ent
.meta
.etag
= etag
;
10085 ent
.meta
.user_data
= *user_data
;
10088 if (acl_bl
&& acl_bl
->length()) {
10089 int ret
= store
->decode_policy(*acl_bl
, &owner
);
10091 ldout(store
->ctx(), 0) << "WARNING: could not decode policy ret=" << ret
<< dendl
;
10094 ent
.meta
.owner
= owner
.get_id().to_str();
10095 ent
.meta
.owner_display_name
= owner
.get_display_name();
10096 ent
.meta
.content_type
= content_type
;
10098 ret
= store
->cls_obj_complete_add(*bs
, obj
, optag
, poolid
, epoch
, ent
, category
, remove_objs
, bilog_flags
, zones_trace
);
10100 if (target
->bucket_info
.datasync_flag_enabled()) {
10101 int r
= store
->data_log
->add_entry(bs
->bucket
, bs
->shard_id
);
10103 lderr(store
->ctx()) << "ERROR: failed writing data log" << dendl
;
10110 int RGWRados::Bucket::UpdateIndex::complete_del(int64_t poolid
, uint64_t epoch
,
10111 real_time
& removed_mtime
,
10112 list
<rgw_obj_index_key
> *remove_objs
)
10117 RGWRados
*store
= target
->get_store();
10120 int ret
= get_bucket_shard(&bs
);
10122 ldout(store
->ctx(), 5) << "failed to get BucketShard object: ret=" << ret
<< dendl
;
10126 ret
= store
->cls_obj_complete_del(*bs
, optag
, poolid
, epoch
, obj
, removed_mtime
, remove_objs
, bilog_flags
, zones_trace
);
10128 if (target
->bucket_info
.datasync_flag_enabled()) {
10129 int r
= store
->data_log
->add_entry(bs
->bucket
, bs
->shard_id
);
10131 lderr(store
->ctx()) << "ERROR: failed writing data log" << dendl
;
10139 int RGWRados::Bucket::UpdateIndex::cancel()
10144 RGWRados
*store
= target
->get_store();
10147 int ret
= guard_reshard(&bs
, [&](BucketShard
*bs
) -> int {
10148 return store
->cls_obj_complete_cancel(*bs
, optag
, obj
, bilog_flags
, zones_trace
);
10152 * need to update data log anyhow, so that whoever follows needs to update its internal markers
10153 * for following the specific bucket shard log. Otherwise they end up staying behind, and users
10154 * have no way to tell that they're all caught up
10156 if (target
->bucket_info
.datasync_flag_enabled()) {
10157 int r
= store
->data_log
->add_entry(bs
->bucket
, bs
->shard_id
);
10159 lderr(store
->ctx()) << "ERROR: failed writing data log" << dendl
;
10166 int RGWRados::Object::Read::read(int64_t ofs
, int64_t end
, bufferlist
& bl
)
10168 RGWRados
*store
= source
->get_store();
10169 CephContext
*cct
= store
->ctx();
10171 rgw_raw_obj read_obj
;
10172 uint64_t read_ofs
= ofs
;
10173 uint64_t len
, read_len
;
10174 bool reading_from_head
= true;
10175 ObjectReadOperation op
;
10177 bool merge_bl
= false;
10178 bufferlist
*pbl
= &bl
;
10179 bufferlist read_bl
;
10180 uint64_t max_chunk_size
;
10182 RGWObjState
*astate
;
10183 int r
= source
->get_state(&astate
, true);
10190 len
= end
- ofs
+ 1;
10192 if (astate
->has_manifest
&& astate
->manifest
.has_tail()) {
10193 /* now get the relevant object part */
10194 RGWObjManifest::obj_iterator iter
= astate
->manifest
.obj_find(ofs
);
10196 uint64_t stripe_ofs
= iter
.get_stripe_ofs();
10197 read_obj
= iter
.get_location().get_raw_obj(store
);
10198 len
= min(len
, iter
.get_stripe_size() - (ofs
- stripe_ofs
));
10199 read_ofs
= iter
.location_ofs() + (ofs
- stripe_ofs
);
10200 reading_from_head
= (read_obj
== state
.head_obj
);
10202 read_obj
= state
.head_obj
;
10205 r
= store
->get_max_chunk_size(read_obj
.pool
, &max_chunk_size
);
10207 ldout(cct
, 0) << "ERROR: failed to get max_chunk_size() for pool " << read_obj
.pool
<< dendl
;
10211 if (len
> max_chunk_size
)
10212 len
= max_chunk_size
;
10215 state
.io_ctx
.locator_set_key(read_obj
.loc
);
10219 if (reading_from_head
) {
10220 /* only when reading from the head object do we need to do the atomic test */
10221 r
= store
->append_atomic_test(&source
->get_ctx(), source
->get_bucket_info(), state
.obj
, op
, &astate
);
10225 if (astate
&& astate
->prefetch_data
) {
10226 if (!ofs
&& astate
->data
.length() >= len
) {
10228 return bl
.length();
10231 if (ofs
< astate
->data
.length()) {
10232 unsigned copy_len
= min((uint64_t)astate
->data
.length() - ofs
, len
);
10233 astate
->data
.copy(ofs
, copy_len
, bl
);
10234 read_len
-= copy_len
;
10235 read_ofs
+= copy_len
;
10237 return bl
.length();
10245 ldout(cct
, 20) << "rados->read obj-ofs=" << ofs
<< " read_ofs=" << read_ofs
<< " read_len=" << read_len
<< dendl
;
10246 op
.read(read_ofs
, read_len
, pbl
, NULL
);
10248 r
= state
.io_ctx
.operate(read_obj
.oid
, &op
, NULL
);
10249 ldout(cct
, 20) << "rados->read r=" << r
<< " bl.length=" << bl
.length() << dendl
;
10256 bl
.append(read_bl
);
10259 return bl
.length();
10262 int RGWRados::SystemObject::Read::GetObjState::get_ref(RGWRados
*store
, rgw_raw_obj
& obj
, rgw_rados_ref
**pref
)
10265 int r
= store
->get_raw_obj_ref(obj
, &ref
);
10276 int RGWRados::get_system_obj(RGWObjectCtx
& obj_ctx
, RGWRados::SystemObject::Read::GetObjState
& read_state
,
10277 RGWObjVersionTracker
*objv_tracker
, rgw_raw_obj
& obj
,
10278 bufferlist
& bl
, off_t ofs
, off_t end
,
10279 map
<string
, bufferlist
> *attrs
,
10280 rgw_cache_entry_info
*cache_info
,
10281 boost::optional
<obj_version
>)
10284 ObjectReadOperation op
;
10289 len
= end
- ofs
+ 1;
10291 if (objv_tracker
) {
10292 objv_tracker
->prepare_op_for_read(&op
);
10295 ldout(cct
, 20) << "rados->read ofs=" << ofs
<< " len=" << len
<< dendl
;
10296 op
.read(ofs
, len
, &bl
, NULL
);
10299 op
.getxattrs(attrs
, NULL
);
10302 rgw_rados_ref
*ref
;
10303 int r
= read_state
.get_ref(this, obj
, &ref
);
10305 ldout(cct
, 20) << "read_state.get_ref() on obj=" << obj
<< " returned " << r
<< dendl
;
10308 r
= ref
->ioctx
.operate(ref
->oid
, &op
, NULL
);
10310 ldout(cct
, 20) << "rados->read r=" << r
<< " bl.length=" << bl
.length() << dendl
;
10313 ldout(cct
, 20) << "rados->read r=" << r
<< " bl.length=" << bl
.length() << dendl
;
10315 uint64_t op_ver
= ref
->ioctx
.get_last_version();
10317 if (read_state
.last_ver
> 0 &&
10318 read_state
.last_ver
!= op_ver
) {
10319 ldout(cct
, 5) << "raced with an object write, abort" << dendl
;
10323 read_state
.last_ver
= op_ver
;
10325 return bl
.length();
10328 int RGWRados::SystemObject::Read::read(int64_t ofs
, int64_t end
, bufferlist
& bl
,
10329 RGWObjVersionTracker
*objv_tracker
,
10330 boost::optional
<obj_version
> refresh_version
)
10332 RGWRados
*store
= source
->get_store();
10333 rgw_raw_obj
& obj
= source
->get_obj();
10335 return store
->get_system_obj(source
->get_ctx(), state
, objv_tracker
, obj
, bl
,
10336 ofs
, end
, read_params
.attrs
,
10337 read_params
.cache_info
, refresh_version
);
10340 int RGWRados::SystemObject::Read::get_attr(const char *name
, bufferlist
& dest
)
10342 RGWRados
*store
= source
->get_store();
10343 rgw_raw_obj
& obj
= source
->get_obj();
10345 return store
->system_obj_get_attr(obj
, name
, dest
);
10348 struct get_obj_data
;
10350 struct get_obj_aio_data
{
10351 struct get_obj_data
*op_data
;
10356 struct get_obj_io
{
10361 static void _get_obj_aio_completion_cb(completion_t cb
, void *arg
);
10363 struct get_obj_data
: public RefCountedObject
{
10368 map
<off_t
, get_obj_io
> io_map
;
10369 map
<off_t
, librados::AioCompletion
*> completion_map
;
10370 uint64_t total_read
;
10373 list
<get_obj_aio_data
> aio_data
;
10374 RGWGetDataCB
*client_cb
;
10375 std::atomic
<bool> cancelled
= { false };
10376 std::atomic
<int64_t> err_code
= { 0 };
10378 list
<bufferlist
> read_list
;
10380 explicit get_obj_data(CephContext
*_cct
)
10382 rados(NULL
), ctx(NULL
),
10383 total_read(0), lock("get_obj_data"), data_lock("get_obj_data::data_lock"),
10385 throttle(cct
, "get_obj_data", cct
->_conf
->rgw_get_obj_window_size
, false) {}
10386 ~get_obj_data() override
{ }
10387 void set_cancelled(int r
) {
10392 bool is_cancelled() {
10396 int get_err_code() {
10400 int wait_next_io(bool *done
) {
10402 map
<off_t
, librados::AioCompletion
*>::iterator iter
= completion_map
.begin();
10403 if (iter
== completion_map
.end()) {
10408 off_t cur_ofs
= iter
->first
;
10409 librados::AioCompletion
*c
= iter
->second
;
10412 c
->wait_for_safe_and_cb();
10413 int r
= c
->get_return_value();
10416 completion_map
.erase(cur_ofs
);
10418 if (completion_map
.empty()) {
10428 void add_io(off_t ofs
, off_t len
, bufferlist
**pbl
, AioCompletion
**pc
) {
10429 Mutex::Locker
l(lock
);
10431 const auto& io_iter
= io_map
.insert(
10432 map
<off_t
, get_obj_io
>::value_type(ofs
, get_obj_io()));
10434 assert(io_iter
.second
); // assert new insertion
10436 get_obj_io
& io
= (io_iter
.first
)->second
;
10439 struct get_obj_aio_data aio
;
10442 aio
.op_data
= this;
10444 aio_data
.push_back(aio
);
10446 struct get_obj_aio_data
*paio_data
= &aio_data
.back(); /* last element */
10448 librados::AioCompletion
*c
= librados::Rados::aio_create_completion((void *)paio_data
, NULL
, _get_obj_aio_completion_cb
);
10449 completion_map
[ofs
] = c
;
10453 /* we have a reference per IO, plus one reference for the calling function.
10454 * reference is dropped for each callback, plus when we're done iterating
10455 * over the parts */
10459 void cancel_io(off_t ofs
) {
10460 ldout(cct
, 20) << "get_obj_data::cancel_io() ofs=" << ofs
<< dendl
;
10462 map
<off_t
, AioCompletion
*>::iterator iter
= completion_map
.find(ofs
);
10463 if (iter
!= completion_map
.end()) {
10464 AioCompletion
*c
= iter
->second
;
10466 completion_map
.erase(ofs
);
10471 /* we don't drop a reference here -- e.g., not calling d->put(), because we still
10472 * need IoCtx to live, as io callback may still be called
10476 void cancel_all_io() {
10477 ldout(cct
, 20) << "get_obj_data::cancel_all_io()" << dendl
;
10478 Mutex::Locker
l(lock
);
10479 for (map
<off_t
, librados::AioCompletion
*>::iterator iter
= completion_map
.begin();
10480 iter
!= completion_map
.end(); ++iter
) {
10481 librados::AioCompletion
*c
= iter
->second
;
10486 int get_complete_ios(off_t ofs
, list
<bufferlist
>& bl_list
) {
10487 Mutex::Locker
l(lock
);
10489 map
<off_t
, get_obj_io
>::iterator liter
= io_map
.begin();
10491 if (liter
== io_map
.end() ||
10492 liter
->first
!= ofs
) {
10496 map
<off_t
, librados::AioCompletion
*>::iterator aiter
;
10497 aiter
= completion_map
.find(ofs
);
10498 if (aiter
== completion_map
.end()) {
10499 /* completion map does not hold this io, it was cancelled */
10503 AioCompletion
*completion
= aiter
->second
;
10504 int r
= completion
->get_return_value();
10508 for (; aiter
!= completion_map
.end(); ++aiter
) {
10509 completion
= aiter
->second
;
10510 if (!completion
->is_safe()) {
10511 /* reached a request that is not yet complete, stop */
10515 r
= completion
->get_return_value();
10517 set_cancelled(r
); /* mark it as cancelled, so that we don't continue processing next operations */
10523 map
<off_t
, get_obj_io
>::iterator old_liter
= liter
++;
10524 bl_list
.push_back(old_liter
->second
.bl
);
10525 io_map
.erase(old_liter
);
10532 static int _get_obj_iterate_cb(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, const rgw_raw_obj
& read_obj
, off_t obj_ofs
, off_t read_ofs
, off_t len
, bool is_head_obj
, RGWObjState
*astate
, void *arg
)
10534 struct get_obj_data
*d
= (struct get_obj_data
*)arg
;
10536 return d
->rados
->get_obj_iterate_cb(d
->ctx
, astate
, bucket_info
, obj
, read_obj
, obj_ofs
, read_ofs
, len
, is_head_obj
, arg
);
10539 static void _get_obj_aio_completion_cb(completion_t cb
, void *arg
)
10541 struct get_obj_aio_data
*aio_data
= (struct get_obj_aio_data
*)arg
;
10542 struct get_obj_data
*d
= aio_data
->op_data
;
10544 d
->rados
->get_obj_aio_completion_cb(cb
, arg
);
10548 void RGWRados::get_obj_aio_completion_cb(completion_t c
, void *arg
)
10550 struct get_obj_aio_data
*aio_data
= (struct get_obj_aio_data
*)arg
;
10551 struct get_obj_data
*d
= aio_data
->op_data
;
10552 off_t ofs
= aio_data
->ofs
;
10553 off_t len
= aio_data
->len
;
10555 list
<bufferlist
> bl_list
;
10556 list
<bufferlist
>::iterator iter
;
10559 ldout(cct
, 20) << "get_obj_aio_completion_cb: io completion ofs=" << ofs
<< " len=" << len
<< dendl
;
10560 d
->throttle
.put(len
);
10562 r
= rados_aio_get_return_value(c
);
10564 ldout(cct
, 0) << "ERROR: got unexpected error when trying to read object: " << r
<< dendl
;
10565 d
->set_cancelled(r
);
10569 if (d
->is_cancelled()) {
10573 d
->data_lock
.Lock();
10575 r
= d
->get_complete_ios(ofs
, bl_list
);
10580 d
->read_list
.splice(d
->read_list
.end(), bl_list
);
10583 d
->data_lock
.Unlock();
10589 int RGWRados::flush_read_list(struct get_obj_data
*d
)
10591 d
->data_lock
.Lock();
10592 list
<bufferlist
> l
;
10593 l
.swap(d
->read_list
);
10595 d
->read_list
.clear();
10597 d
->data_lock
.Unlock();
10601 list
<bufferlist
>::iterator iter
;
10602 for (iter
= l
.begin(); iter
!= l
.end(); ++iter
) {
10603 bufferlist
& bl
= *iter
;
10604 r
= d
->client_cb
->handle_data(bl
, 0, bl
.length());
10606 dout(0) << "ERROR: flush_read_list(): d->client_cb->handle_data() returned " << r
<< dendl
;
10611 d
->data_lock
.Lock();
10614 d
->set_cancelled(r
);
10616 d
->data_lock
.Unlock();
10620 int RGWRados::get_obj_iterate_cb(RGWObjectCtx
*ctx
, RGWObjState
*astate
,
10621 const RGWBucketInfo
& bucket_info
,
10622 const rgw_obj
& obj
,
10623 const rgw_raw_obj
& read_obj
,
10625 off_t read_ofs
, off_t len
,
10626 bool is_head_obj
, void *arg
)
10628 RGWObjectCtx
*rctx
= static_cast<RGWObjectCtx
*>(ctx
);
10629 ObjectReadOperation op
;
10630 struct get_obj_data
*d
= (struct get_obj_data
*)arg
;
10638 /* only when reading from the head object do we need to do the atomic test */
10639 r
= append_atomic_test(rctx
, bucket_info
, obj
, op
, &astate
);
10644 obj_ofs
< astate
->data
.length()) {
10645 unsigned chunk_len
= min((uint64_t)astate
->data
.length() - obj_ofs
, (uint64_t)len
);
10647 d
->data_lock
.Lock();
10648 r
= d
->client_cb
->handle_data(astate
->data
, obj_ofs
, chunk_len
);
10649 d
->data_lock
.Unlock();
10654 d
->total_read
+= chunk_len
;
10658 read_ofs
+= chunk_len
;
10659 obj_ofs
+= chunk_len
;
10665 d
->throttle
.get(len
);
10666 if (d
->is_cancelled()) {
10667 return d
->get_err_code();
10670 /* add io after we check that we're not cancelled, otherwise we're going to have trouble
10673 d
->add_io(obj_ofs
, len
, &pbl
, &c
);
10675 ldout(cct
, 20) << "rados->get_obj_iterate_cb oid=" << read_obj
.oid
<< " obj-ofs=" << obj_ofs
<< " read_ofs=" << read_ofs
<< " len=" << len
<< dendl
;
10676 op
.read(read_ofs
, len
, pbl
, NULL
);
10678 librados::IoCtx
io_ctx(d
->io_ctx
);
10679 io_ctx
.locator_set_key(read_obj
.loc
);
10681 r
= io_ctx
.aio_operate(read_obj
.oid
, c
, &op
, NULL
);
10683 ldout(cct
, 0) << "rados->aio_operate r=" << r
<< dendl
;
10687 // Flush data to client if there is any
10688 r
= flush_read_list(d
);
10695 ldout(cct
, 20) << "cancelling io r=" << r
<< " obj_ofs=" << obj_ofs
<< dendl
;
10696 d
->set_cancelled(r
);
10697 d
->cancel_io(obj_ofs
);
10702 int RGWRados::Object::Read::iterate(int64_t ofs
, int64_t end
, RGWGetDataCB
*cb
)
10704 RGWRados
*store
= source
->get_store();
10705 CephContext
*cct
= store
->ctx();
10707 struct get_obj_data
*data
= new get_obj_data(cct
);
10710 RGWObjectCtx
& obj_ctx
= source
->get_ctx();
10712 data
->rados
= store
;
10713 data
->io_ctx
.dup(state
.io_ctx
);
10714 data
->client_cb
= cb
;
10716 int r
= store
->iterate_obj(obj_ctx
, source
->get_bucket_info(), state
.obj
, ofs
, end
, cct
->_conf
->rgw_get_obj_max_req_size
, _get_obj_iterate_cb
, (void *)data
);
10718 data
->cancel_all_io();
10723 r
= data
->wait_next_io(&done
);
10725 dout(10) << "get_obj_iterate() r=" << r
<< ", canceling all io" << dendl
;
10726 data
->cancel_all_io();
10729 r
= store
->flush_read_list(data
);
10731 dout(10) << "get_obj_iterate() r=" << r
<< ", canceling all io" << dendl
;
10732 data
->cancel_all_io();
10742 int RGWRados::iterate_obj(RGWObjectCtx
& obj_ctx
,
10743 const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
,
10744 off_t ofs
, off_t end
,
10745 uint64_t max_chunk_size
,
10746 int (*iterate_obj_cb
)(const RGWBucketInfo
&, const rgw_obj
& obj
,
10747 const rgw_raw_obj
&, off_t
, off_t
, off_t
, bool,
10748 RGWObjState
*, void *),
10751 rgw_raw_obj head_obj
;
10752 rgw_raw_obj read_obj
;
10753 uint64_t read_ofs
= ofs
;
10755 bool reading_from_head
= true;
10756 RGWObjState
*astate
= NULL
;
10758 obj_to_raw(bucket_info
.placement_rule
, obj
, &head_obj
);
10760 int r
= get_obj_state(&obj_ctx
, bucket_info
, obj
, &astate
, false);
10768 len
= end
- ofs
+ 1;
10770 if (astate
->has_manifest
) {
10771 /* now get the relevant object stripe */
10772 RGWObjManifest::obj_iterator iter
= astate
->manifest
.obj_find(ofs
);
10774 RGWObjManifest::obj_iterator obj_end
= astate
->manifest
.obj_end();
10776 for (; iter
!= obj_end
&& ofs
<= end
; ++iter
) {
10777 off_t stripe_ofs
= iter
.get_stripe_ofs();
10778 off_t next_stripe_ofs
= stripe_ofs
+ iter
.get_stripe_size();
10780 while (ofs
< next_stripe_ofs
&& ofs
<= end
) {
10781 read_obj
= iter
.get_location().get_raw_obj(this);
10782 uint64_t read_len
= min(len
, iter
.get_stripe_size() - (ofs
- stripe_ofs
));
10783 read_ofs
= iter
.location_ofs() + (ofs
- stripe_ofs
);
10785 if (read_len
> max_chunk_size
) {
10786 read_len
= max_chunk_size
;
10789 reading_from_head
= (read_obj
== head_obj
);
10790 r
= iterate_obj_cb(bucket_info
, obj
, read_obj
, ofs
, read_ofs
, read_len
, reading_from_head
, astate
, arg
);
10800 while (ofs
<= end
) {
10801 read_obj
= head_obj
;
10802 uint64_t read_len
= min(len
, max_chunk_size
);
10804 r
= iterate_obj_cb(bucket_info
, obj
, read_obj
, ofs
, ofs
, read_len
, reading_from_head
, astate
, arg
);
10817 int RGWRados::obj_operate(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, ObjectWriteOperation
*op
)
10820 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
10825 return ref
.ioctx
.operate(ref
.oid
, op
);
10828 int RGWRados::obj_operate(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, ObjectReadOperation
*op
)
10831 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
10838 return ref
.ioctx
.operate(ref
.oid
, op
, &outbl
);
10841 int RGWRados::olh_init_modification_impl(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
, const rgw_obj
& olh_obj
, string
*op_tag
)
10843 ObjectWriteOperation op
;
10845 assert(olh_obj
.key
.instance
.empty());
10847 bool has_tag
= (state
.exists
&& has_olh_tag(state
.attrset
));
10849 if (!state
.exists
) {
10852 op
.assert_exists();
10853 struct timespec mtime_ts
= real_clock::to_timespec(state
.mtime
);
10854 op
.mtime2(&mtime_ts
);
10858 * 3 possible cases: olh object doesn't exist, it exists as an olh, it exists as a regular object.
10859 * If it exists as a regular object we'll need to transform it into an olh. We'll do it in two
10860 * steps, first change its tag and set the olh pending attrs. Once write is done we'll need to
10861 * truncate it, remove extra attrs, and send it to the garbage collection. The bucket index olh
10862 * log will reflect that.
10864 * Need to generate separate olh and obj tags, as olh can be colocated with object data. obj_tag
10865 * is used for object data instance, olh_tag for olh instance.
10868 /* guard against racing writes */
10869 bucket_index_guard_olh_op(state
, op
);
10875 int ret
= gen_rand_alphanumeric_lower(cct
, &obj_tag
, 32);
10877 ldout(cct
, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret
<< dendl
;
10881 bl
.append(obj_tag
.c_str(), obj_tag
.size());
10882 op
.setxattr(RGW_ATTR_ID_TAG
, bl
);
10884 state
.attrset
[RGW_ATTR_ID_TAG
] = bl
;
10885 state
.obj_tag
= bl
;
10889 ret
= gen_rand_alphanumeric_lower(cct
, &olh_tag
, 32);
10891 ldout(cct
, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret
<< dendl
;
10895 olh_bl
.append(olh_tag
.c_str(), olh_tag
.size());
10896 op
.setxattr(RGW_ATTR_OLH_ID_TAG
, olh_bl
);
10898 state
.attrset
[RGW_ATTR_OLH_ID_TAG
] = olh_bl
;
10899 state
.olh_tag
= olh_bl
;
10900 state
.is_olh
= true;
10903 op
.setxattr(RGW_ATTR_OLH_VER
, verbl
);
10907 RGWOLHPendingInfo pending_info
;
10908 pending_info
.time
= real_clock::now();
10909 ::encode(pending_info
, bl
);
10911 #define OLH_PENDING_TAG_LEN 32
10912 /* tag will start with current time epoch, this so that entries are sorted by time */
10914 utime_t
ut(pending_info
.time
);
10915 snprintf(buf
, sizeof(buf
), "%016llx", (unsigned long long)ut
.sec());
10919 int ret
= gen_rand_alphanumeric_lower(cct
, &s
, OLH_PENDING_TAG_LEN
- op_tag
->size());
10921 ldout(cct
, 0) << "ERROR: gen_rand_alphanumeric_lower() returned ret=" << ret
<< dendl
;
10926 string attr_name
= RGW_ATTR_OLH_PENDING_PREFIX
;
10927 attr_name
.append(*op_tag
);
10929 op
.setxattr(attr_name
.c_str(), bl
);
10931 ret
= obj_operate(bucket_info
, olh_obj
, &op
);
10936 state
.exists
= true;
10937 state
.attrset
[attr_name
] = bl
;
10942 int RGWRados::olh_init_modification(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
, const rgw_obj
& obj
, string
*op_tag
)
10946 ret
= olh_init_modification_impl(bucket_info
, state
, obj
, op_tag
);
10947 if (ret
== -EEXIST
) {
10954 int RGWRados::guard_reshard(BucketShard
*bs
, const rgw_obj
& obj_instance
, std::function
<int(BucketShard
*)> call
)
10957 const rgw_obj
*pobj
= &obj_instance
;
10960 for (int i
= 0; i
< NUM_RESHARD_RETRIES
; ++i
) {
10961 r
= bs
->init(pobj
->bucket
, *pobj
);
10963 ldout(cct
, 5) << "bs.init() returned ret=" << r
<< dendl
;
10967 if (r
!= -ERR_BUSY_RESHARDING
) {
10970 ldout(cct
, 0) << "NOTICE: resharding operation on bucket index detected, blocking" << dendl
;
10971 string new_bucket_id
;
10972 r
= block_while_resharding(bs
, &new_bucket_id
);
10973 if (r
== -ERR_BUSY_RESHARDING
) {
10979 ldout(cct
, 20) << "reshard completion identified, new_bucket_id=" << new_bucket_id
<< dendl
;
10980 i
= 0; /* resharding is finished, make sure we can retry */
10983 obj
.bucket
.update_bucket_id(new_bucket_id
);
10994 int RGWRados::block_while_resharding(RGWRados::BucketShard
*bs
, string
*new_bucket_id
)
10996 std::shared_ptr
<RGWReshardWait
> waiter
= reshard_wait
;
10998 return waiter
->block_while_resharding(bs
, new_bucket_id
);
11001 int RGWRados::bucket_index_link_olh(const RGWBucketInfo
& bucket_info
, RGWObjState
& olh_state
, const rgw_obj
& obj_instance
,
11002 bool delete_marker
,
11003 const string
& op_tag
,
11004 struct rgw_bucket_dir_entry_meta
*meta
,
11005 uint64_t olh_epoch
,
11006 real_time unmod_since
, bool high_precision_time
, rgw_zone_set
*_zones_trace
)
11009 int r
= get_obj_head_ref(bucket_info
, obj_instance
, &ref
);
11014 rgw_zone_set zones_trace
;
11015 if (_zones_trace
) {
11016 zones_trace
= *_zones_trace
;
11018 zones_trace
.insert(get_zone().id
);
11021 BucketShard
bs(this);
11023 cls_rgw_obj_key
key(obj_instance
.key
.get_index_key_name(), obj_instance
.key
.instance
);
11024 r
= guard_reshard(&bs
, obj_instance
, [&](BucketShard
*bs
) -> int {
11025 librados::ObjectWriteOperation op
;
11026 cls_rgw_guard_bucket_resharding(op
, -ERR_BUSY_RESHARDING
);
11027 return cls_rgw_bucket_link_olh(bs
->index_ctx
, op
,
11028 bs
->bucket_obj
, key
, olh_state
.olh_tag
, delete_marker
, op_tag
, meta
, olh_epoch
,
11029 unmod_since
, high_precision_time
,
11030 get_zone().log_data
, zones_trace
);
11033 ldout(cct
, 20) << "cls_rgw_bucket_link_olh() returned r=" << r
<< dendl
;
11040 void RGWRados::bucket_index_guard_olh_op(RGWObjState
& olh_state
, ObjectOperation
& op
)
11042 ldout(cct
, 20) << __func__
<< "(): olh_state.olh_tag=" << string(olh_state
.olh_tag
.c_str(), olh_state
.olh_tag
.length()) << dendl
;
11043 op
.cmpxattr(RGW_ATTR_OLH_ID_TAG
, CEPH_OSD_CMPXATTR_OP_EQ
, olh_state
.olh_tag
);
11046 int RGWRados::bucket_index_unlink_instance(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj_instance
,
11047 const string
& op_tag
, const string
& olh_tag
, uint64_t olh_epoch
, rgw_zone_set
*_zones_trace
)
11050 int r
= get_obj_head_ref(bucket_info
, obj_instance
, &ref
);
11055 rgw_zone_set zones_trace
;
11056 if (_zones_trace
) {
11057 zones_trace
= *_zones_trace
;
11059 zones_trace
.insert(get_zone().id
);
11061 BucketShard
bs(this);
11063 cls_rgw_obj_key
key(obj_instance
.key
.get_index_key_name(), obj_instance
.key
.instance
);
11064 r
= guard_reshard(&bs
, obj_instance
, [&](BucketShard
*bs
) -> int {
11065 librados::ObjectWriteOperation op
;
11066 cls_rgw_guard_bucket_resharding(op
, -ERR_BUSY_RESHARDING
);
11067 return cls_rgw_bucket_unlink_instance(bs
->index_ctx
, op
, bs
->bucket_obj
, key
, op_tag
,
11068 olh_tag
, olh_epoch
, get_zone().log_data
, zones_trace
);
11071 ldout(cct
, 20) << "cls_rgw_bucket_link_olh() returned r=" << r
<< dendl
;
11078 int RGWRados::bucket_index_read_olh_log(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
,
11079 const rgw_obj
& obj_instance
, uint64_t ver_marker
,
11080 map
<uint64_t, vector
<rgw_bucket_olh_log_entry
> > *log
,
11081 bool *is_truncated
)
11084 int r
= get_obj_head_ref(bucket_info
, obj_instance
, &ref
);
11089 BucketShard
bs(this);
11090 int ret
= bs
.init(obj_instance
.bucket
, obj_instance
);
11092 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
11096 string
olh_tag(state
.olh_tag
.c_str(), state
.olh_tag
.length());
11098 cls_rgw_obj_key
key(obj_instance
.key
.get_index_key_name(), string());
11100 ret
= guard_reshard(&bs
, obj_instance
, [&](BucketShard
*bs
) -> int {
11101 ObjectReadOperation op
;
11102 cls_rgw_guard_bucket_resharding(op
, -ERR_BUSY_RESHARDING
);
11103 return cls_rgw_get_olh_log(bs
->index_ctx
, bs
->bucket_obj
, op
,
11104 key
, ver_marker
, olh_tag
, log
, is_truncated
);
11107 ldout(cct
, 20) << "cls_rgw_get_olh_log() returned r=" << r
<< dendl
;
11114 int RGWRados::bucket_index_trim_olh_log(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
, const rgw_obj
& obj_instance
, uint64_t ver
)
11117 int r
= get_obj_head_ref(bucket_info
, obj_instance
, &ref
);
11122 BucketShard
bs(this);
11123 int ret
= bs
.init(obj_instance
.bucket
, obj_instance
);
11125 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
11129 string
olh_tag(state
.olh_tag
.c_str(), state
.olh_tag
.length());
11131 cls_rgw_obj_key
key(obj_instance
.key
.get_index_key_name(), string());
11133 ret
= guard_reshard(&bs
, obj_instance
, [&](BucketShard
*pbs
) -> int {
11134 ObjectWriteOperation op
;
11135 cls_rgw_guard_bucket_resharding(op
, -ERR_BUSY_RESHARDING
);
11136 cls_rgw_trim_olh_log(op
, key
, ver
, olh_tag
);
11137 return pbs
->index_ctx
.operate(pbs
->bucket_obj
, &op
);
11140 ldout(cct
, 20) << "cls_rgw_trim_olh_log() returned r=" << ret
<< dendl
;
11147 int RGWRados::bucket_index_clear_olh(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
, const rgw_obj
& obj_instance
)
11150 int r
= get_obj_head_ref(bucket_info
, obj_instance
, &ref
);
11155 BucketShard
bs(this);
11157 string
olh_tag(state
.olh_tag
.c_str(), state
.olh_tag
.length());
11159 cls_rgw_obj_key
key(obj_instance
.key
.get_index_key_name(), string());
11161 int ret
= guard_reshard(&bs
, obj_instance
, [&](BucketShard
*pbs
) -> int {
11162 ObjectWriteOperation op
;
11163 cls_rgw_guard_bucket_resharding(op
, -ERR_BUSY_RESHARDING
);
11164 return cls_rgw_clear_olh(pbs
->index_ctx
, op
, pbs
->bucket_obj
, key
, olh_tag
);
11167 ldout(cct
, 5) << "cls_rgw_clear_olh() returned ret=" << ret
<< dendl
;
11174 int RGWRados::apply_olh_log(RGWObjectCtx
& obj_ctx
, RGWObjState
& state
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
,
11175 bufferlist
& olh_tag
, map
<uint64_t, vector
<rgw_bucket_olh_log_entry
> >& log
,
11176 uint64_t *plast_ver
, rgw_zone_set
* zones_trace
)
11182 librados::ObjectWriteOperation op
;
11184 uint64_t last_ver
= log
.rbegin()->first
;
11185 *plast_ver
= last_ver
;
11187 map
<uint64_t, vector
<rgw_bucket_olh_log_entry
> >::iterator iter
= log
.begin();
11189 op
.cmpxattr(RGW_ATTR_OLH_ID_TAG
, CEPH_OSD_CMPXATTR_OP_EQ
, olh_tag
);
11190 op
.cmpxattr(RGW_ATTR_OLH_VER
, CEPH_OSD_CMPXATTR_OP_GT
, last_ver
);
11192 struct timespec mtime_ts
= real_clock::to_timespec(state
.mtime
);
11193 op
.mtime2(&mtime_ts
);
11195 bool need_to_link
= false;
11196 cls_rgw_obj_key key
;
11197 bool delete_marker
= false;
11198 list
<cls_rgw_obj_key
> remove_instances
;
11199 bool need_to_remove
= false;
11201 for (iter
= log
.begin(); iter
!= log
.end(); ++iter
) {
11202 vector
<rgw_bucket_olh_log_entry
>::iterator viter
= iter
->second
.begin();
11203 for (; viter
!= iter
->second
.end(); ++viter
) {
11204 rgw_bucket_olh_log_entry
& entry
= *viter
;
11206 ldout(cct
, 20) << "olh_log_entry: op=" << (int)entry
.op
11207 << " key=" << entry
.key
.name
<< "[" << entry
.key
.instance
<< "] "
11208 << (entry
.delete_marker
? "(delete)" : "") << dendl
;
11209 switch (entry
.op
) {
11210 case CLS_RGW_OLH_OP_REMOVE_INSTANCE
:
11211 remove_instances
.push_back(entry
.key
);
11213 case CLS_RGW_OLH_OP_LINK_OLH
:
11214 need_to_link
= true;
11215 need_to_remove
= false;
11217 delete_marker
= entry
.delete_marker
;
11219 case CLS_RGW_OLH_OP_UNLINK_OLH
:
11220 need_to_remove
= true;
11221 need_to_link
= false;
11224 ldout(cct
, 0) << "ERROR: apply_olh_log: invalid op: " << (int)entry
.op
<< dendl
;
11227 string attr_name
= RGW_ATTR_OLH_PENDING_PREFIX
;
11228 attr_name
.append(entry
.op_tag
);
11229 op
.rmxattr(attr_name
.c_str());
11234 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
11239 const rgw_bucket
& bucket
= obj
.bucket
;
11241 if (need_to_link
) {
11242 rgw_obj
target(bucket
, key
);
11244 info
.target
= target
;
11245 info
.removed
= delete_marker
;
11247 ::encode(info
, bl
);
11248 op
.setxattr(RGW_ATTR_OLH_INFO
, bl
);
11251 /* first remove object instances */
11252 for (list
<cls_rgw_obj_key
>::iterator liter
= remove_instances
.begin();
11253 liter
!= remove_instances
.end(); ++liter
) {
11254 cls_rgw_obj_key
& key
= *liter
;
11255 rgw_obj
obj_instance(bucket
, key
);
11256 int ret
= delete_obj(obj_ctx
, bucket_info
, obj_instance
, 0, RGW_BILOG_FLAG_VERSIONED_OP
, ceph::real_time(), zones_trace
);
11257 if (ret
< 0 && ret
!= -ENOENT
) {
11258 ldout(cct
, 0) << "ERROR: delete_obj() returned " << ret
<< " obj_instance=" << obj_instance
<< dendl
;
11263 /* update olh object */
11264 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
11265 if (r
== -ECANCELED
) {
11269 ldout(cct
, 0) << "ERROR: could not apply olh update, r=" << r
<< dendl
;
11273 r
= bucket_index_trim_olh_log(bucket_info
, state
, obj
, last_ver
);
11275 ldout(cct
, 0) << "ERROR: could not trim olh log, r=" << r
<< dendl
;
11279 if (need_to_remove
) {
11280 ObjectWriteOperation rm_op
;
11282 rm_op
.cmpxattr(RGW_ATTR_OLH_ID_TAG
, CEPH_OSD_CMPXATTR_OP_EQ
, olh_tag
);
11283 rm_op
.cmpxattr(RGW_ATTR_OLH_VER
, CEPH_OSD_CMPXATTR_OP_GT
, last_ver
);
11284 cls_obj_check_prefix_exist(rm_op
, RGW_ATTR_OLH_PENDING_PREFIX
, true); /* fail if found one of these, pending modification */
11287 r
= ref
.ioctx
.operate(ref
.oid
, &rm_op
);
11288 if (r
== -ECANCELED
) {
11289 return 0; /* someone else won this race */
11292 * only clear if was successful, otherwise we might clobber pending operations on this object
11294 r
= bucket_index_clear_olh(bucket_info
, state
, obj
);
11296 ldout(cct
, 0) << "ERROR: could not clear bucket index olh entries r=" << r
<< dendl
;
11306 * read olh log and apply it
11308 int RGWRados::update_olh(RGWObjectCtx
& obj_ctx
, RGWObjState
*state
, const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, rgw_zone_set
*zones_trace
)
11310 map
<uint64_t, vector
<rgw_bucket_olh_log_entry
> > log
;
11312 uint64_t ver_marker
= 0;
11315 int ret
= bucket_index_read_olh_log(bucket_info
, *state
, obj
, ver_marker
, &log
, &is_truncated
);
11319 ret
= apply_olh_log(obj_ctx
, *state
, bucket_info
, obj
, state
->olh_tag
, log
, &ver_marker
, zones_trace
);
11323 } while (is_truncated
);
11328 int RGWRados::set_olh(RGWObjectCtx
& obj_ctx
, RGWBucketInfo
& bucket_info
, const rgw_obj
& target_obj
, bool delete_marker
, rgw_bucket_dir_entry_meta
*meta
,
11329 uint64_t olh_epoch
, real_time unmod_since
, bool high_precision_time
, rgw_zone_set
*zones_trace
)
11333 rgw_obj olh_obj
= target_obj
;
11334 olh_obj
.key
.instance
.clear();
11336 RGWObjState
*state
= NULL
;
11341 #define MAX_ECANCELED_RETRY 100
11342 for (i
= 0; i
< MAX_ECANCELED_RETRY
; i
++) {
11343 if (ret
== -ECANCELED
) {
11344 obj_ctx
.obj
.invalidate(olh_obj
);
11347 ret
= get_obj_state(&obj_ctx
, bucket_info
, olh_obj
, &state
, false); /* don't follow olh */
11352 ret
= olh_init_modification(bucket_info
, *state
, olh_obj
, &op_tag
);
11354 ldout(cct
, 20) << "olh_init_modification() target_obj=" << target_obj
<< " delete_marker=" << (int)delete_marker
<< " returned " << ret
<< dendl
;
11355 if (ret
== -ECANCELED
) {
11360 ret
= bucket_index_link_olh(bucket_info
, *state
, target_obj
, delete_marker
, op_tag
, meta
, olh_epoch
, unmod_since
, high_precision_time
, zones_trace
);
11362 ldout(cct
, 20) << "bucket_index_link_olh() target_obj=" << target_obj
<< " delete_marker=" << (int)delete_marker
<< " returned " << ret
<< dendl
;
11363 if (ret
== -ECANCELED
) {
11371 if (i
== MAX_ECANCELED_RETRY
) {
11372 ldout(cct
, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl
;
11376 ret
= update_olh(obj_ctx
, state
, bucket_info
, olh_obj
);
11377 if (ret
== -ECANCELED
) { /* already did what we needed, no need to retry, raced with another user */
11381 ldout(cct
, 20) << "update_olh() target_obj=" << target_obj
<< " returned " << ret
<< dendl
;
11388 int RGWRados::unlink_obj_instance(RGWObjectCtx
& obj_ctx
, RGWBucketInfo
& bucket_info
, const rgw_obj
& target_obj
,
11389 uint64_t olh_epoch
, rgw_zone_set
*zones_trace
)
11393 rgw_obj olh_obj
= target_obj
;
11394 olh_obj
.key
.instance
.clear();
11396 RGWObjState
*state
= NULL
;
11401 for (i
= 0; i
< MAX_ECANCELED_RETRY
; i
++) {
11402 if (ret
== -ECANCELED
) {
11403 obj_ctx
.obj
.invalidate(olh_obj
);
11406 ret
= get_obj_state(&obj_ctx
, bucket_info
, olh_obj
, &state
, false); /* don't follow olh */
11410 ret
= olh_init_modification(bucket_info
, *state
, olh_obj
, &op_tag
);
11412 ldout(cct
, 20) << "olh_init_modification() target_obj=" << target_obj
<< " returned " << ret
<< dendl
;
11413 if (ret
== -ECANCELED
) {
11419 string
olh_tag(state
->olh_tag
.c_str(), state
->olh_tag
.length());
11421 ret
= bucket_index_unlink_instance(bucket_info
, target_obj
, op_tag
, olh_tag
, olh_epoch
, zones_trace
);
11423 ldout(cct
, 20) << "bucket_index_unlink_instance() target_obj=" << target_obj
<< " returned " << ret
<< dendl
;
11424 if (ret
== -ECANCELED
) {
11432 if (i
== MAX_ECANCELED_RETRY
) {
11433 ldout(cct
, 0) << "ERROR: exceeded max ECANCELED retries, aborting (EIO)" << dendl
;
11437 ret
= update_olh(obj_ctx
, state
, bucket_info
, olh_obj
, zones_trace
);
11438 if (ret
== -ECANCELED
) { /* already did what we needed, no need to retry, raced with another user */
11442 ldout(cct
, 20) << "update_olh() target_obj=" << target_obj
<< " returned " << ret
<< dendl
;
11449 void RGWRados::gen_rand_obj_instance_name(rgw_obj
*target_obj
)
11451 #define OBJ_INSTANCE_LEN 32
11452 char buf
[OBJ_INSTANCE_LEN
+ 1];
11454 gen_rand_alphanumeric_no_underscore(cct
, buf
, OBJ_INSTANCE_LEN
); /* don't want it to get url escaped,
11455 no underscore for instance name due to the way we encode the raw keys */
11457 target_obj
->key
.set_instance(buf
);
11460 static void filter_attrset(map
<string
, bufferlist
>& unfiltered_attrset
, const string
& check_prefix
,
11461 map
<string
, bufferlist
> *attrset
)
11464 map
<string
, bufferlist
>::iterator iter
;
11465 for (iter
= unfiltered_attrset
.lower_bound(check_prefix
);
11466 iter
!= unfiltered_attrset
.end(); ++iter
) {
11467 if (!boost::algorithm::starts_with(iter
->first
, check_prefix
))
11469 (*attrset
)[iter
->first
] = iter
->second
;
11473 int RGWRados::get_olh(const RGWBucketInfo
& bucket_info
, const rgw_obj
& obj
, RGWOLHInfo
*olh
)
11475 map
<string
, bufferlist
> unfiltered_attrset
;
11477 ObjectReadOperation op
;
11478 op
.getxattrs(&unfiltered_attrset
, NULL
);
11481 int r
= obj_operate(bucket_info
, obj
, &op
);
11486 map
<string
, bufferlist
> attrset
;
11488 filter_attrset(unfiltered_attrset
, RGW_ATTR_OLH_PREFIX
, &attrset
);
11490 map
<string
, bufferlist
>::iterator iter
= attrset
.find(RGW_ATTR_OLH_INFO
);
11491 if (iter
== attrset
.end()) { /* not an olh */
11496 bufferlist::iterator biter
= iter
->second
.begin();
11497 ::decode(*olh
, biter
);
11498 } catch (buffer::error
& err
) {
11499 ldout(cct
, 0) << "ERROR: failed to decode olh info" << dendl
;
11506 void RGWRados::check_pending_olh_entries(map
<string
, bufferlist
>& pending_entries
,
11507 map
<string
, bufferlist
> *rm_pending_entries
)
11509 map
<string
, bufferlist
>::iterator iter
= pending_entries
.begin();
11511 real_time now
= real_clock::now();
11513 while (iter
!= pending_entries
.end()) {
11514 bufferlist::iterator biter
= iter
->second
.begin();
11515 RGWOLHPendingInfo pending_info
;
11517 ::decode(pending_info
, biter
);
11518 } catch (buffer::error
& err
) {
11519 /* skipping bad entry, we could remove it but it might hide a bug */
11520 ldout(cct
, 0) << "ERROR: failed to decode pending entry " << iter
->first
<< dendl
;
11525 map
<string
, bufferlist
>::iterator cur_iter
= iter
;
11527 if (now
- pending_info
.time
>= make_timespan(cct
->_conf
->rgw_olh_pending_timeout_sec
)) {
11528 (*rm_pending_entries
)[cur_iter
->first
] = cur_iter
->second
;
11529 pending_entries
.erase(cur_iter
);
11531 /* entries names are sorted by time (rounded to a second) */
11537 int RGWRados::remove_olh_pending_entries(const RGWBucketInfo
& bucket_info
, RGWObjState
& state
, const rgw_obj
& olh_obj
, map
<string
, bufferlist
>& pending_attrs
)
11539 ObjectWriteOperation op
;
11541 bucket_index_guard_olh_op(state
, op
);
11543 for (map
<string
, bufferlist
>::iterator iter
= pending_attrs
.begin(); iter
!= pending_attrs
.end(); ++iter
) {
11544 op
.rmxattr(iter
->first
.c_str());
11548 int r
= get_obj_head_ref(bucket_info
, olh_obj
, &ref
);
11553 /* update olh object */
11554 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
11555 if (r
== -ENOENT
|| r
== -ECANCELED
) {
11556 /* raced with some other change, shouldn't sweat about it */
11560 ldout(cct
, 0) << "ERROR: could not apply olh update, r=" << r
<< dendl
;
11567 int RGWRados::follow_olh(const RGWBucketInfo
& bucket_info
, RGWObjectCtx
& obj_ctx
, RGWObjState
*state
, const rgw_obj
& olh_obj
, rgw_obj
*target
)
11569 map
<string
, bufferlist
> pending_entries
;
11570 filter_attrset(state
->attrset
, RGW_ATTR_OLH_PENDING_PREFIX
, &pending_entries
);
11572 map
<string
, bufferlist
> rm_pending_entries
;
11573 check_pending_olh_entries(pending_entries
, &rm_pending_entries
);
11575 if (!rm_pending_entries
.empty()) {
11576 int ret
= remove_olh_pending_entries(bucket_info
, *state
, olh_obj
, rm_pending_entries
);
11578 ldout(cct
, 20) << "ERROR: rm_pending_entries returned ret=" << ret
<< dendl
;
11582 if (!pending_entries
.empty()) {
11583 ldout(cct
, 20) << __func__
<< "(): found pending entries, need to update_olh() on bucket=" << olh_obj
.bucket
<< dendl
;
11585 int ret
= update_olh(obj_ctx
, state
, bucket_info
, olh_obj
);
11591 map
<string
, bufferlist
>::iterator iter
= state
->attrset
.find(RGW_ATTR_OLH_INFO
);
11592 assert(iter
!= state
->attrset
.end());
11595 bufferlist::iterator biter
= iter
->second
.begin();
11596 ::decode(olh
, biter
);
11597 } catch (buffer::error
& err
) {
11598 ldout(cct
, 0) << "ERROR: failed to decode olh info" << dendl
;
11606 *target
= olh
.target
;
11611 int RGWRados::raw_obj_stat(rgw_raw_obj
& obj
, uint64_t *psize
, real_time
*pmtime
, uint64_t *epoch
,
11612 map
<string
, bufferlist
> *attrs
, bufferlist
*first_chunk
,
11613 RGWObjVersionTracker
*objv_tracker
)
11616 int r
= get_raw_obj_ref(obj
, &ref
);
11621 map
<string
, bufferlist
> unfiltered_attrset
;
11623 struct timespec mtime_ts
;
11625 ObjectReadOperation op
;
11626 if (objv_tracker
) {
11627 objv_tracker
->prepare_op_for_read(&op
);
11630 op
.getxattrs(&unfiltered_attrset
, NULL
);
11632 if (psize
|| pmtime
) {
11633 op
.stat2(&size
, &mtime_ts
, NULL
);
11636 op
.read(0, cct
->_conf
->rgw_max_chunk_size
, first_chunk
, NULL
);
11639 r
= ref
.ioctx
.operate(ref
.oid
, &op
, &outbl
);
11642 *epoch
= ref
.ioctx
.get_last_version();
11651 *pmtime
= ceph::real_clock::from_timespec(mtime_ts
);
11653 filter_attrset(unfiltered_attrset
, RGW_ATTR_PREFIX
, attrs
);
11659 int RGWRados::get_bucket_stats(RGWBucketInfo
& bucket_info
, int shard_id
, string
*bucket_ver
, string
*master_ver
,
11660 map
<RGWObjCategory
, RGWStorageStats
>& stats
, string
*max_marker
, bool *syncstopped
)
11662 map
<string
, rgw_bucket_dir_header
> headers
;
11663 map
<int, string
> bucket_instance_ids
;
11664 int r
= cls_bucket_head(bucket_info
, shard_id
, headers
, &bucket_instance_ids
);
11669 assert(headers
.size() == bucket_instance_ids
.size());
11671 map
<string
, rgw_bucket_dir_header
>::iterator iter
= headers
.begin();
11672 map
<int, string
>::iterator viter
= bucket_instance_ids
.begin();
11673 BucketIndexShardsManager ver_mgr
;
11674 BucketIndexShardsManager master_ver_mgr
;
11675 BucketIndexShardsManager marker_mgr
;
11677 for(; iter
!= headers
.end(); ++iter
, ++viter
) {
11678 accumulate_raw_stats(iter
->second
, stats
);
11679 snprintf(buf
, sizeof(buf
), "%lu", (unsigned long)iter
->second
.ver
);
11680 ver_mgr
.add(viter
->first
, string(buf
));
11681 snprintf(buf
, sizeof(buf
), "%lu", (unsigned long)iter
->second
.master_ver
);
11682 master_ver_mgr
.add(viter
->first
, string(buf
));
11683 if (shard_id
>= 0) {
11684 *max_marker
= iter
->second
.max_marker
;
11686 marker_mgr
.add(viter
->first
, iter
->second
.max_marker
);
11688 if (syncstopped
!= NULL
)
11689 *syncstopped
= iter
->second
.syncstopped
;
11691 ver_mgr
.to_string(bucket_ver
);
11692 master_ver_mgr
.to_string(master_ver
);
11693 if (shard_id
< 0) {
11694 marker_mgr
.to_string(max_marker
);
11699 int RGWRados::get_bi_log_status(RGWBucketInfo
& bucket_info
, int shard_id
,
11700 map
<int, string
>& markers
)
11702 map
<string
, rgw_bucket_dir_header
> headers
;
11703 map
<int, string
> bucket_instance_ids
;
11704 int r
= cls_bucket_head(bucket_info
, shard_id
, headers
, &bucket_instance_ids
);
11708 assert(headers
.size() == bucket_instance_ids
.size());
11710 map
<string
, rgw_bucket_dir_header
>::iterator iter
= headers
.begin();
11711 map
<int, string
>::iterator viter
= bucket_instance_ids
.begin();
11713 for(; iter
!= headers
.end(); ++iter
, ++viter
) {
11714 if (shard_id
>= 0) {
11715 markers
[shard_id
] = iter
->second
.max_marker
;
11717 markers
[viter
->first
] = iter
->second
.max_marker
;
11723 class RGWGetBucketStatsContext
: public RGWGetDirHeader_CB
{
11724 RGWGetBucketStats_CB
*cb
;
11726 map
<RGWObjCategory
, RGWStorageStats
> stats
;
11732 RGWGetBucketStatsContext(RGWGetBucketStats_CB
*_cb
, uint32_t _pendings
)
11733 : cb(_cb
), pendings(_pendings
), stats(), ret_code(0), should_cb(true),
11734 lock("RGWGetBucketStatsContext") {}
11736 void handle_response(int r
, rgw_bucket_dir_header
& header
) override
{
11737 Mutex::Locker
l(lock
);
11740 accumulate_raw_stats(header
, stats
);
11745 // Are we all done?
11746 if (--pendings
== 0) {
11748 cb
->set_response(&stats
);
11750 cb
->handle_response(ret_code
);
11757 Mutex::Locker
l(lock
);
11762 int RGWRados::get_bucket_stats_async(RGWBucketInfo
& bucket_info
, int shard_id
, RGWGetBucketStats_CB
*ctx
)
11765 RGWGetBucketStatsContext
*get_ctx
= new RGWGetBucketStatsContext(ctx
, bucket_info
.num_shards
? : 1);
11767 int r
= cls_bucket_head_async(bucket_info
, shard_id
, get_ctx
, &num_aio
);
11771 get_ctx
->unset_cb();
11778 class RGWGetUserStatsContext
: public RGWGetUserHeader_CB
{
11779 RGWGetUserStats_CB
*cb
;
11782 explicit RGWGetUserStatsContext(RGWGetUserStats_CB
* const cb
)
11785 void handle_response(int r
, cls_user_header
& header
) override
{
11786 const cls_user_stats
& hs
= header
.stats
;
11788 RGWStorageStats stats
;
11790 stats
.size
= hs
.total_bytes
;
11791 stats
.size_rounded
= hs
.total_bytes_rounded
;
11792 stats
.num_objects
= hs
.total_entries
;
11794 cb
->set_response(stats
);
11797 cb
->handle_response(r
);
11803 int RGWRados::get_user_stats(const rgw_user
& user
, RGWStorageStats
& stats
)
11805 string user_str
= user
.to_str();
11807 cls_user_header header
;
11808 int r
= cls_user_get_header(user_str
, &header
);
11812 const cls_user_stats
& hs
= header
.stats
;
11814 stats
.size
= hs
.total_bytes
;
11815 stats
.size_rounded
= hs
.total_bytes_rounded
;
11816 stats
.num_objects
= hs
.total_entries
;
11821 int RGWRados::get_user_stats_async(const rgw_user
& user
, RGWGetUserStats_CB
*ctx
)
11823 string user_str
= user
.to_str();
11825 RGWGetUserStatsContext
*get_ctx
= new RGWGetUserStatsContext(ctx
);
11826 int r
= cls_user_get_header_async(user_str
, get_ctx
);
11836 void RGWRados::get_bucket_meta_oid(const rgw_bucket
& bucket
, string
& oid
)
11838 oid
= RGW_BUCKET_INSTANCE_MD_PREFIX
+ bucket
.get_key(':');
11841 void RGWRados::get_bucket_instance_obj(const rgw_bucket
& bucket
, rgw_raw_obj
& obj
)
11843 if (!bucket
.oid
.empty()) {
11844 obj
.init(get_zone_params().domain_root
, bucket
.oid
);
11847 get_bucket_meta_oid(bucket
, oid
);
11848 obj
.init(get_zone_params().domain_root
, oid
);
11852 int RGWRados::get_bucket_instance_info(RGWObjectCtx
& obj_ctx
, const string
& meta_key
, RGWBucketInfo
& info
,
11853 real_time
*pmtime
, map
<string
, bufferlist
> *pattrs
)
11855 size_t pos
= meta_key
.find(':');
11856 if (pos
== string::npos
) {
11859 string oid
= RGW_BUCKET_INSTANCE_MD_PREFIX
+ meta_key
;
11860 rgw_bucket_instance_key_to_oid(oid
);
11862 return get_bucket_instance_from_oid(obj_ctx
, oid
, info
, pmtime
, pattrs
);
11865 int RGWRados::get_bucket_instance_info(RGWObjectCtx
& obj_ctx
, const rgw_bucket
& bucket
, RGWBucketInfo
& info
,
11866 real_time
*pmtime
, map
<string
, bufferlist
> *pattrs
)
11869 if (bucket
.oid
.empty()) {
11870 get_bucket_meta_oid(bucket
, oid
);
11875 return get_bucket_instance_from_oid(obj_ctx
, oid
, info
, pmtime
, pattrs
);
11878 int RGWRados::get_bucket_instance_from_oid(RGWObjectCtx
& obj_ctx
, const string
& oid
, RGWBucketInfo
& info
,
11879 real_time
*pmtime
, map
<string
, bufferlist
> *pattrs
,
11880 rgw_cache_entry_info
*cache_info
,
11881 boost::optional
<obj_version
> refresh_version
)
11883 ldout(cct
, 20) << "reading from " << get_zone_params().domain_root
<< ":" << oid
<< dendl
;
11887 int ret
= rgw_get_system_obj(this, obj_ctx
, get_zone_params().domain_root
,
11888 oid
, epbl
, &info
.objv_tracker
, pmtime
, pattrs
,
11889 cache_info
, refresh_version
);
11894 bufferlist::iterator iter
= epbl
.begin();
11896 ::decode(info
, iter
);
11897 } catch (buffer::error
& err
) {
11898 ldout(cct
, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl
;
11901 info
.bucket
.oid
= oid
;
11905 int RGWRados::get_bucket_entrypoint_info(RGWObjectCtx
& obj_ctx
,
11906 const string
& tenant_name
,
11907 const string
& bucket_name
,
11908 RGWBucketEntryPoint
& entry_point
,
11909 RGWObjVersionTracker
*objv_tracker
,
11911 map
<string
, bufferlist
> *pattrs
,
11912 rgw_cache_entry_info
*cache_info
,
11913 boost::optional
<obj_version
> refresh_version
)
11916 string bucket_entry
;
11918 rgw_make_bucket_entry_name(tenant_name
, bucket_name
, bucket_entry
);
11919 int ret
= rgw_get_system_obj(this, obj_ctx
, get_zone_params().domain_root
,
11920 bucket_entry
, bl
, objv_tracker
, pmtime
, pattrs
,
11921 cache_info
, refresh_version
);
11926 bufferlist::iterator iter
= bl
.begin();
11928 ::decode(entry_point
, iter
);
11929 } catch (buffer::error
& err
) {
11930 ldout(cct
, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl
;
11936 int RGWRados::convert_old_bucket_info(RGWObjectCtx
& obj_ctx
,
11937 const string
& tenant_name
,
11938 const string
& bucket_name
)
11940 RGWBucketEntryPoint entry_point
;
11941 real_time ep_mtime
;
11942 RGWObjVersionTracker ot
;
11943 map
<string
, bufferlist
> attrs
;
11944 RGWBucketInfo info
;
11946 ldout(cct
, 10) << "RGWRados::convert_old_bucket_info(): bucket=" << bucket_name
<< dendl
;
11948 int ret
= get_bucket_entrypoint_info(obj_ctx
, tenant_name
, bucket_name
, entry_point
, &ot
, &ep_mtime
, &attrs
);
11950 ldout(cct
, 0) << "ERROR: get_bucket_entrypoint_info() returned " << ret
<< " bucket=" << bucket_name
<< dendl
;
11954 if (!entry_point
.has_bucket_info
) {
11955 /* already converted! */
11959 info
= entry_point
.old_bucket_info
;
11960 info
.bucket
.oid
= bucket_name
;
11961 info
.ep_objv
= ot
.read_version
;
11963 ot
.generate_new_write_ver(cct
);
11965 ret
= put_linked_bucket_info(info
, false, ep_mtime
, &ot
.write_version
, &attrs
, true);
11967 ldout(cct
, 0) << "ERROR: failed to put_linked_bucket_info(): " << ret
<< dendl
;
11974 int RGWRados::_get_bucket_info(RGWObjectCtx
& obj_ctx
,
11975 const string
& tenant
,
11976 const string
& bucket_name
,
11977 RGWBucketInfo
& info
,
11979 map
<string
, bufferlist
> *pattrs
,
11980 boost::optional
<obj_version
> refresh_version
)
11982 bucket_info_entry e
;
11983 string bucket_entry
;
11984 rgw_make_bucket_entry_name(tenant
, bucket_name
, bucket_entry
);
11987 if (binfo_cache
->find(bucket_entry
, &e
)) {
11988 if (refresh_version
&&
11989 e
.info
.objv_tracker
.read_version
.compare(&(*refresh_version
))) {
11990 lderr(cct
) << "WARNING: The bucket info cache is inconsistent. This is "
11991 << "a failure that should be debugged. I am a nice machine, "
11992 << "so I will try to recover." << dendl
;
11993 binfo_cache
->invalidate(bucket_entry
);
12003 RGWBucketEntryPoint entry_point
;
12004 real_time ep_mtime
;
12005 RGWObjVersionTracker ot
;
12006 rgw_cache_entry_info entry_cache_info
;
12007 int ret
= get_bucket_entrypoint_info(obj_ctx
, tenant
, bucket_name
,
12008 entry_point
, &ot
, &ep_mtime
, pattrs
,
12009 &entry_cache_info
, refresh_version
);
12011 /* only init these fields */
12012 info
.bucket
.tenant
= tenant
;
12013 info
.bucket
.name
= bucket_name
;
12017 if (entry_point
.has_bucket_info
) {
12018 info
= entry_point
.old_bucket_info
;
12019 info
.bucket
.oid
= bucket_name
;
12020 info
.bucket
.tenant
= tenant
;
12021 info
.ep_objv
= ot
.read_version
;
12022 ldout(cct
, 20) << "rgw_get_bucket_info: old bucket info, bucket=" << info
.bucket
<< " owner " << info
.owner
<< dendl
;
12026 /* data is in the bucket instance object, we need to get attributes from there, clear everything
12033 ldout(cct
, 20) << "rgw_get_bucket_info: bucket instance: " << entry_point
.bucket
<< dendl
;
12036 /* read bucket instance info */
12039 get_bucket_meta_oid(entry_point
.bucket
, oid
);
12041 rgw_cache_entry_info cache_info
;
12043 ret
= get_bucket_instance_from_oid(obj_ctx
, oid
, e
.info
, &e
.mtime
, &e
.attrs
,
12044 &cache_info
, refresh_version
);
12045 e
.info
.ep_objv
= ot
.read_version
;
12048 lderr(cct
) << "ERROR: get_bucket_instance_from_oid failed: " << ret
<< dendl
;
12049 info
.bucket
.tenant
= tenant
;
12050 info
.bucket
.name
= bucket_name
;
12051 // XXX and why return anything in case of an error anyway?
12060 list
<rgw_cache_entry_info
*> cache_info_entries
;
12061 cache_info_entries
.push_back(&entry_cache_info
);
12062 cache_info_entries
.push_back(&cache_info
);
12065 /* chain to both bucket entry point and bucket instance */
12066 if (!binfo_cache
->put(this, bucket_entry
, &e
, cache_info_entries
)) {
12067 ldout(cct
, 20) << "couldn't put binfo cache entry, might have raced with data changes" << dendl
;
12070 if (refresh_version
&&
12071 refresh_version
->compare(&info
.objv_tracker
.read_version
)) {
12072 lderr(cct
) << "WARNING: The OSD has the same version I have. Something may "
12073 << "have gone squirrelly. An administrator may have forced a "
12074 << "change; otherwise there is a problem somewhere." << dendl
;
12080 int RGWRados::get_bucket_info(RGWObjectCtx
& obj_ctx
,
12081 const string
& tenant
, const string
& bucket_name
,
12082 RGWBucketInfo
& info
,
12083 real_time
*pmtime
, map
<string
, bufferlist
> *pattrs
)
12085 return _get_bucket_info(obj_ctx
, tenant
, bucket_name
, info
, pmtime
,
12086 pattrs
, boost::none
);
12089 int RGWRados::try_refresh_bucket_info(RGWBucketInfo
& info
,
12090 ceph::real_time
*pmtime
,
12091 map
<string
, bufferlist
> *pattrs
)
12093 RGWObjectCtx
obj_ctx(this);
12095 return _get_bucket_info(obj_ctx
, info
.bucket
.tenant
, info
.bucket
.name
,
12096 info
, pmtime
, pattrs
, info
.objv_tracker
.read_version
);
12099 int RGWRados::put_bucket_entrypoint_info(const string
& tenant_name
, const string
& bucket_name
, RGWBucketEntryPoint
& entry_point
,
12100 bool exclusive
, RGWObjVersionTracker
& objv_tracker
, real_time mtime
,
12101 map
<string
, bufferlist
> *pattrs
)
12104 ::encode(entry_point
, epbl
);
12105 string bucket_entry
;
12106 rgw_make_bucket_entry_name(tenant_name
, bucket_name
, bucket_entry
);
12107 return rgw_bucket_store_info(this, bucket_entry
, epbl
, exclusive
, pattrs
, &objv_tracker
, mtime
);
12110 int RGWRados::put_bucket_instance_info(RGWBucketInfo
& info
, bool exclusive
,
12111 real_time mtime
, map
<string
, bufferlist
> *pattrs
)
12113 info
.has_instance_obj
= true;
12116 ::encode(info
, bl
);
12118 string key
= info
.bucket
.get_key(); /* when we go through meta api, we don't use oid directly */
12119 int ret
= rgw_bucket_instance_store_info(this, key
, bl
, exclusive
, pattrs
, &info
.objv_tracker
, mtime
);
12120 if (ret
== -EEXIST
) {
12121 /* well, if it's exclusive we shouldn't overwrite it, because we might race with another
12122 * bucket operation on this specific bucket (e.g., being synced from the master), but
12123 * since bucket instace meta object is unique for this specific bucket instace, we don't
12124 * need to return an error.
12125 * A scenario where we'd get -EEXIST here, is in a multi-zone config, we're not on the
12126 * master, creating a bucket, sending bucket creation to the master, we create the bucket
12127 * locally, while in the sync thread we sync the new bucket.
12134 int RGWRados::put_linked_bucket_info(RGWBucketInfo
& info
, bool exclusive
, real_time mtime
, obj_version
*pep_objv
,
12135 map
<string
, bufferlist
> *pattrs
, bool create_entry_point
)
12137 bool create_head
= !info
.has_instance_obj
|| create_entry_point
;
12139 int ret
= put_bucket_instance_info(info
, exclusive
, mtime
, pattrs
);
12145 return 0; /* done! */
12147 RGWBucketEntryPoint entry_point
;
12148 entry_point
.bucket
= info
.bucket
;
12149 entry_point
.owner
= info
.owner
;
12150 entry_point
.creation_time
= info
.creation_time
;
12151 entry_point
.linked
= true;
12152 RGWObjVersionTracker ot
;
12153 if (pep_objv
&& !pep_objv
->tag
.empty()) {
12154 ot
.write_version
= *pep_objv
;
12156 ot
.generate_new_write_ver(cct
);
12158 *pep_objv
= ot
.write_version
;
12161 ret
= put_bucket_entrypoint_info(info
.bucket
.tenant
, info
.bucket
.name
, entry_point
, exclusive
, ot
, mtime
, NULL
);
12168 int RGWRados::omap_get_vals(rgw_raw_obj
& obj
, bufferlist
& header
, const string
& marker
, uint64_t count
, std::map
<string
, bufferlist
>& m
)
12171 int r
= get_raw_obj_ref(obj
, &ref
);
12176 r
= ref
.ioctx
.omap_get_vals(ref
.oid
, marker
, count
, &m
);
12184 int RGWRados::omap_get_all(rgw_raw_obj
& obj
, bufferlist
& header
,
12185 std::map
<string
, bufferlist
>& m
)
12188 int r
= get_raw_obj_ref(obj
, &ref
);
12193 #define MAX_OMAP_GET_ENTRIES 1024
12194 const int count
= MAX_OMAP_GET_ENTRIES
;
12195 string start_after
;
12198 std::map
<string
, bufferlist
> t
;
12199 r
= ref
.ioctx
.omap_get_vals(ref
.oid
, start_after
, count
, &t
);
12206 start_after
= t
.rbegin()->first
;
12207 m
.insert(t
.begin(), t
.end());
12212 int RGWRados::omap_set(rgw_raw_obj
& obj
, const std::string
& key
, bufferlist
& bl
)
12215 int r
= get_raw_obj_ref(obj
, &ref
);
12219 ldout(cct
, 15) << "omap_set obj=" << obj
<< " key=" << key
<< dendl
;
12221 map
<string
, bufferlist
> m
;
12224 r
= ref
.ioctx
.omap_set(ref
.oid
, m
);
12229 int RGWRados::omap_set(rgw_raw_obj
& obj
, std::map
<std::string
, bufferlist
>& m
)
12232 int r
= get_raw_obj_ref(obj
, &ref
);
12237 r
= ref
.ioctx
.omap_set(ref
.oid
, m
);
12242 int RGWRados::omap_del(rgw_raw_obj
& obj
, const std::string
& key
)
12245 int r
= get_raw_obj_ref(obj
, &ref
);
12253 r
= ref
.ioctx
.omap_rm_keys(ref
.oid
, k
);
12257 int RGWRados::update_containers_stats(map
<string
, RGWBucketEnt
>& m
)
12259 RGWObjectCtx
obj_ctx(this);
12261 map
<string
, RGWBucketEnt
>::iterator iter
;
12262 for (iter
= m
.begin(); iter
!= m
.end(); ++iter
) {
12263 RGWBucketEnt
& ent
= iter
->second
;
12264 rgw_bucket
& bucket
= ent
.bucket
;
12267 ent
.size_rounded
= 0;
12269 map
<string
, rgw_bucket_dir_header
> headers
;
12271 RGWBucketInfo bucket_info
;
12272 int ret
= get_bucket_instance_info(obj_ctx
, bucket
, bucket_info
, NULL
, NULL
);
12277 int r
= cls_bucket_head(bucket_info
, RGW_NO_SHARD
, headers
);
12281 map
<string
, rgw_bucket_dir_header
>::iterator hiter
= headers
.begin();
12282 for (; hiter
!= headers
.end(); ++hiter
) {
12283 RGWObjCategory category
= main_category
;
12284 map
<uint8_t, struct rgw_bucket_category_stats
>::iterator iter
= (hiter
->second
.stats
).find((uint8_t)category
);
12285 if (iter
!= hiter
->second
.stats
.end()) {
12286 struct rgw_bucket_category_stats
& stats
= iter
->second
;
12287 ent
.count
+= stats
.num_entries
;
12288 ent
.size
+= stats
.total_size
;
12289 ent
.size_rounded
+= stats
.total_size_rounded
;
12293 // fill in placement_rule from the bucket instance for use in swift's
12294 // per-storage policy statistics
12295 ent
.placement_rule
= std::move(bucket_info
.placement_rule
);
12301 int RGWRados::append_async(rgw_raw_obj
& obj
, size_t size
, bufferlist
& bl
)
12304 int r
= get_raw_obj_ref(obj
, &ref
);
12308 librados::Rados
*rad
= get_rados_handle();
12309 librados::AioCompletion
*completion
= rad
->aio_create_completion(NULL
, NULL
, NULL
);
12311 r
= ref
.ioctx
.aio_append(ref
.oid
, completion
, bl
, size
);
12312 completion
->release();
12316 int RGWRados::distribute(const string
& key
, bufferlist
& bl
)
12319 * we were called before watch was initialized. This can only happen if we're updating some system
12320 * config object (e.g., zone info) during init. Don't try to distribute the cache info for these
12321 * objects, they're currently only read on startup anyway.
12323 if (!watch_initialized
)
12327 pick_control_oid(key
, notify_oid
);
12329 ldout(cct
, 10) << "distributing notification oid=" << notify_oid
<< " bl.length()=" << bl
.length() << dendl
;
12330 return control_pool_ctx
.notify2(notify_oid
, bl
, 0, NULL
);
12333 int RGWRados::pool_iterate_begin(const rgw_pool
& pool
, RGWPoolIterCtx
& ctx
)
12335 librados::IoCtx
& io_ctx
= ctx
.io_ctx
;
12336 librados::NObjectIterator
& iter
= ctx
.iter
;
12338 int r
= open_pool_ctx(pool
, io_ctx
);
12342 iter
= io_ctx
.nobjects_begin();
12347 int RGWRados::pool_iterate_begin(const rgw_pool
& pool
, const string
& cursor
, RGWPoolIterCtx
& ctx
)
12349 librados::IoCtx
& io_ctx
= ctx
.io_ctx
;
12350 librados::NObjectIterator
& iter
= ctx
.iter
;
12352 int r
= open_pool_ctx(pool
, io_ctx
);
12356 librados::ObjectCursor oc
;
12357 if (!oc
.from_str(cursor
)) {
12358 ldout(cct
, 10) << "failed to parse cursor: " << cursor
<< dendl
;
12362 iter
= io_ctx
.nobjects_begin(oc
);
12367 string
RGWRados::pool_iterate_get_cursor(RGWPoolIterCtx
& ctx
)
12369 return ctx
.iter
.get_cursor().to_str();
12372 int RGWRados::pool_iterate(RGWPoolIterCtx
& ctx
, uint32_t num
, vector
<rgw_bucket_dir_entry
>& objs
,
12373 bool *is_truncated
, RGWAccessListFilter
*filter
)
12375 librados::IoCtx
& io_ctx
= ctx
.io_ctx
;
12376 librados::NObjectIterator
& iter
= ctx
.iter
;
12378 if (iter
== io_ctx
.nobjects_end())
12383 for (i
= 0; i
< num
&& iter
!= io_ctx
.nobjects_end(); ++i
, ++iter
) {
12384 rgw_bucket_dir_entry e
;
12386 string oid
= iter
->get_oid();
12387 ldout(cct
, 20) << "RGWRados::pool_iterate: got " << oid
<< dendl
;
12389 // fill it in with initial values; we may correct later
12390 if (filter
&& !filter
->filter(oid
, oid
))
12398 *is_truncated
= (iter
!= io_ctx
.nobjects_end());
12400 return objs
.size();
12402 struct RGWAccessListFilterPrefix
: public RGWAccessListFilter
{
12405 explicit RGWAccessListFilterPrefix(const string
& _prefix
) : prefix(_prefix
) {}
12406 bool filter(string
& name
, string
& key
) override
{
12407 return (prefix
.compare(key
.substr(0, prefix
.size())) == 0);
12411 int RGWRados::list_raw_objects_init(const rgw_pool
& pool
, const string
& marker
, RGWListRawObjsCtx
*ctx
)
12413 if (!ctx
->initialized
) {
12414 int r
= pool_iterate_begin(pool
, marker
, ctx
->iter_ctx
);
12416 ldout(cct
, 10) << "failed to list objects pool_iterate_begin() returned r=" << r
<< dendl
;
12419 ctx
->initialized
= true;
12424 int RGWRados::list_raw_objects_next(const string
& prefix_filter
, int max
,
12425 RGWListRawObjsCtx
& ctx
, list
<string
>& oids
,
12426 bool *is_truncated
)
12428 if (!ctx
.initialized
) {
12431 RGWAccessListFilterPrefix
filter(prefix_filter
);
12432 vector
<rgw_bucket_dir_entry
> objs
;
12433 int r
= pool_iterate(ctx
.iter_ctx
, max
, objs
, is_truncated
, &filter
);
12436 ldout(cct
, 10) << "failed to list objects pool_iterate returned r=" << r
<< dendl
;
12440 vector
<rgw_bucket_dir_entry
>::iterator iter
;
12441 for (iter
= objs
.begin(); iter
!= objs
.end(); ++iter
) {
12442 oids
.push_back(iter
->key
.name
);
12445 return oids
.size();
12448 int RGWRados::list_raw_objects(const rgw_pool
& pool
, const string
& prefix_filter
,
12449 int max
, RGWListRawObjsCtx
& ctx
, list
<string
>& oids
,
12450 bool *is_truncated
)
12452 if (!ctx
.initialized
) {
12453 int r
= list_raw_objects_init(pool
, string(), &ctx
);
12459 return list_raw_objects_next(prefix_filter
, max
, ctx
, oids
, is_truncated
);
12462 string
RGWRados::list_raw_objs_get_cursor(RGWListRawObjsCtx
& ctx
)
12464 return pool_iterate_get_cursor(ctx
.iter_ctx
);
12467 int RGWRados::list_bi_log_entries(RGWBucketInfo
& bucket_info
, int shard_id
, string
& marker
, uint32_t max
,
12468 std::list
<rgw_bi_log_entry
>& result
, bool *truncated
)
12470 ldout(cct
, 20) << __func__
<< ": " << bucket_info
.bucket
<< " marker " << marker
<< " shard_id=" << shard_id
<< " max " << max
<< dendl
;
12473 librados::IoCtx index_ctx
;
12474 map
<int, string
> oids
;
12475 map
<int, cls_rgw_bi_log_list_ret
> bi_log_lists
;
12476 map
<int, string
> bucket_instance_ids
;
12477 int r
= open_bucket_index(bucket_info
, index_ctx
, oids
, shard_id
, &bucket_instance_ids
);
12481 BucketIndexShardsManager marker_mgr
;
12482 bool has_shards
= (oids
.size() > 1 || shard_id
>= 0);
12483 // If there are multiple shards for the bucket index object, the marker
12484 // should have the pattern '{shard_id_1}#{shard_marker_1},{shard_id_2}#
12485 // {shard_marker_2}...', if there is no sharding, the bi_log_list should
12486 // only contain one record, and the key is the bucket instance id.
12487 r
= marker_mgr
.from_string(marker
, shard_id
);
12491 r
= CLSRGWIssueBILogList(index_ctx
, marker_mgr
, max
, oids
, bi_log_lists
, cct
->_conf
->rgw_bucket_index_max_aio
)();
12495 map
<int, list
<rgw_bi_log_entry
>::iterator
> vcurrents
;
12496 map
<int, list
<rgw_bi_log_entry
>::iterator
> vends
;
12498 *truncated
= false;
12500 map
<int, cls_rgw_bi_log_list_ret
>::iterator miter
= bi_log_lists
.begin();
12501 for (; miter
!= bi_log_lists
.end(); ++miter
) {
12502 int shard_id
= miter
->first
;
12503 vcurrents
[shard_id
] = miter
->second
.entries
.begin();
12504 vends
[shard_id
] = miter
->second
.entries
.end();
12506 *truncated
= (*truncated
|| miter
->second
.truncated
);
12511 bool has_more
= true;
12512 map
<int, list
<rgw_bi_log_entry
>::iterator
>::iterator viter
;
12513 map
<int, list
<rgw_bi_log_entry
>::iterator
>::iterator eiter
;
12514 while (total
< max
&& has_more
) {
12517 viter
= vcurrents
.begin();
12518 eiter
= vends
.begin();
12520 for (; total
< max
&& viter
!= vcurrents
.end(); ++viter
, ++eiter
) {
12521 assert (eiter
!= vends
.end());
12523 int shard_id
= viter
->first
;
12524 list
<rgw_bi_log_entry
>::iterator
& liter
= viter
->second
;
12526 if (liter
== eiter
->second
){
12529 rgw_bi_log_entry
& entry
= *(liter
);
12532 snprintf(buf
, sizeof(buf
), "%d", shard_id
);
12534 build_bucket_index_marker(buf
, entry
.id
, &tmp_id
);
12535 entry
.id
.swap(tmp_id
);
12537 marker_mgr
.add(shard_id
, entry
.id
);
12538 result
.push_back(entry
);
12546 for (viter
= vcurrents
.begin(), eiter
= vends
.begin(); viter
!= vcurrents
.end(); ++viter
, ++eiter
) {
12547 assert (eiter
!= vends
.end());
12548 *truncated
= (*truncated
|| (viter
->second
!= eiter
->second
));
12552 // Refresh marker, if there are multiple shards, the output will look like
12553 // '{shard_oid_1}#{shard_marker_1},{shard_oid_2}#{shard_marker_2}...',
12554 // if there is no sharding, the simply marker (without oid) is returned
12556 marker_mgr
.to_string(&marker
);
12558 if (!result
.empty()) {
12559 marker
= result
.rbegin()->id
;
12566 int RGWRados::trim_bi_log_entries(RGWBucketInfo
& bucket_info
, int shard_id
, string
& start_marker
, string
& end_marker
)
12568 librados::IoCtx index_ctx
;
12569 map
<int, string
> bucket_objs
;
12571 BucketIndexShardsManager start_marker_mgr
;
12572 BucketIndexShardsManager end_marker_mgr
;
12574 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
, shard_id
);
12579 r
= start_marker_mgr
.from_string(start_marker
, shard_id
);
12584 r
= end_marker_mgr
.from_string(end_marker
, shard_id
);
12589 return CLSRGWIssueBILogTrim(index_ctx
, start_marker_mgr
, end_marker_mgr
, bucket_objs
,
12590 cct
->_conf
->rgw_bucket_index_max_aio
)();
12595 int RGWRados::resync_bi_log_entries(RGWBucketInfo
& bucket_info
, int shard_id
)
12597 librados::IoCtx index_ctx
;
12598 map
<int, string
> bucket_objs
;
12599 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
, shard_id
);
12603 return CLSRGWIssueResyncBucketBILog(index_ctx
, bucket_objs
, cct
->_conf
->rgw_bucket_index_max_aio
)();
12606 int RGWRados::stop_bi_log_entries(RGWBucketInfo
& bucket_info
, int shard_id
)
12608 librados::IoCtx index_ctx
;
12609 map
<int, string
> bucket_objs
;
12610 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
, shard_id
);
12614 return CLSRGWIssueBucketBILogStop(index_ctx
, bucket_objs
, cct
->_conf
->rgw_bucket_index_max_aio
)();
12617 int RGWRados::bi_get_instance(const RGWBucketInfo
& bucket_info
, rgw_obj
& obj
, rgw_bucket_dir_entry
*dirent
)
12620 int r
= get_obj_head_ref(bucket_info
, obj
, &ref
);
12625 rgw_cls_bi_entry bi_entry
;
12626 r
= bi_get(obj
.bucket
, obj
, InstanceIdx
, &bi_entry
);
12627 if (r
< 0 && r
!= -ENOENT
) {
12628 ldout(cct
, 0) << "ERROR: bi_get() returned r=" << r
<< dendl
;
12633 bufferlist::iterator iter
= bi_entry
.data
.begin();
12635 ::decode(*dirent
, iter
);
12636 } catch (buffer::error
& err
) {
12637 ldout(cct
, 0) << "ERROR: failed to decode bi_entry()" << dendl
;
12644 int RGWRados::bi_get(rgw_bucket
& bucket
, rgw_obj
& obj
, BIIndexType index_type
, rgw_cls_bi_entry
*entry
)
12646 BucketShard
bs(this);
12647 int ret
= bs
.init(bucket
, obj
);
12649 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
12653 cls_rgw_obj_key
key(obj
.key
.get_index_key_name(), obj
.key
.instance
);
12655 ret
= cls_rgw_bi_get(bs
.index_ctx
, bs
.bucket_obj
, index_type
, key
, entry
);
12662 void RGWRados::bi_put(ObjectWriteOperation
& op
, BucketShard
& bs
, rgw_cls_bi_entry
& entry
)
12664 cls_rgw_bi_put(op
, bs
.bucket_obj
, entry
);
12667 int RGWRados::bi_put(BucketShard
& bs
, rgw_cls_bi_entry
& entry
)
12669 int ret
= cls_rgw_bi_put(bs
.index_ctx
, bs
.bucket_obj
, entry
);
12676 int RGWRados::bi_put(rgw_bucket
& bucket
, rgw_obj
& obj
, rgw_cls_bi_entry
& entry
)
12678 BucketShard
bs(this);
12679 int ret
= bs
.init(bucket
, obj
);
12681 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
12685 return bi_put(bs
, entry
);
12688 int RGWRados::bi_list(rgw_bucket
& bucket
, const string
& obj_name
, const string
& marker
, uint32_t max
, list
<rgw_cls_bi_entry
> *entries
, bool *is_truncated
)
12690 rgw_obj
obj(bucket
, obj_name
);
12691 BucketShard
bs(this);
12692 int ret
= bs
.init(bucket
, obj
);
12694 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
12698 ret
= cls_rgw_bi_list(bs
.index_ctx
, bs
.bucket_obj
, obj_name
, marker
, max
, entries
, is_truncated
);
12699 if (ret
== -ENOENT
) {
12700 *is_truncated
= false;
12708 int RGWRados::bi_list(BucketShard
& bs
, const string
& filter_obj
, const string
& marker
, uint32_t max
, list
<rgw_cls_bi_entry
> *entries
, bool *is_truncated
)
12710 int ret
= cls_rgw_bi_list(bs
.index_ctx
, bs
.bucket_obj
, filter_obj
, marker
, max
, entries
, is_truncated
);
12717 int RGWRados::bi_remove(BucketShard
& bs
)
12719 int ret
= bs
.index_ctx
.remove(bs
.bucket_obj
);
12720 if (ret
== -ENOENT
) {
12724 ldout(cct
, 5) << "bs.index_ctx.remove(" << bs
.bucket_obj
<< ") returned ret=" << ret
<< dendl
;
12731 int RGWRados::bi_list(rgw_bucket
& bucket
, int shard_id
, const string
& filter_obj
, const string
& marker
, uint32_t max
, list
<rgw_cls_bi_entry
> *entries
, bool *is_truncated
)
12733 BucketShard
bs(this);
12734 int ret
= bs
.init(bucket
, shard_id
);
12736 ldout(cct
, 5) << "bs.init() returned ret=" << ret
<< dendl
;
12740 return bi_list(bs
, filter_obj
, marker
, max
, entries
, is_truncated
);
12743 int RGWRados::gc_operate(string
& oid
, librados::ObjectWriteOperation
*op
)
12745 return gc_pool_ctx
.operate(oid
, op
);
12748 int RGWRados::gc_aio_operate(string
& oid
, librados::ObjectWriteOperation
*op
)
12750 AioCompletion
*c
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
12751 int r
= gc_pool_ctx
.aio_operate(oid
, c
, op
);
12756 int RGWRados::gc_operate(string
& oid
, librados::ObjectReadOperation
*op
, bufferlist
*pbl
)
12758 return gc_pool_ctx
.operate(oid
, op
, pbl
);
12761 int RGWRados::list_gc_objs(int *index
, string
& marker
, uint32_t max
, bool expired_only
, std::list
<cls_rgw_gc_obj_info
>& result
, bool *truncated
)
12763 return gc
->list(index
, marker
, max
, expired_only
, result
, truncated
);
12766 int RGWRados::process_gc()
12768 return gc
->process();
12771 int RGWRados::list_lc_progress(const string
& marker
, uint32_t max_entries
, map
<string
, int> *progress_map
)
12773 return lc
->list_lc_progress(marker
, max_entries
, progress_map
);
12776 int RGWRados::process_lc()
12778 return lc
->process();
12781 int RGWRados::process_expire_objects()
12783 obj_expirer
->inspect_all_shards(utime_t(), ceph_clock_now());
12787 int RGWRados::cls_rgw_init_index(librados::IoCtx
& index_ctx
, librados::ObjectWriteOperation
& op
, string
& oid
)
12790 cls_rgw_bucket_init(op
);
12791 return index_ctx
.operate(oid
, &op
);
12794 int RGWRados::cls_obj_prepare_op(BucketShard
& bs
, RGWModifyOp op
, string
& tag
,
12795 rgw_obj
& obj
, uint16_t bilog_flags
, rgw_zone_set
*_zones_trace
)
12797 rgw_zone_set zones_trace
;
12798 if (_zones_trace
) {
12799 zones_trace
= *_zones_trace
;
12802 zones_trace
.insert(get_zone().id
);
12805 ObjectWriteOperation o
;
12806 cls_rgw_obj_key
key(obj
.key
.get_index_key_name(), obj
.key
.instance
);
12807 cls_rgw_guard_bucket_resharding(o
, -ERR_BUSY_RESHARDING
);
12808 cls_rgw_bucket_prepare_op(o
, op
, tag
, key
, obj
.key
.get_loc(), get_zone().log_data
, bilog_flags
, zones_trace
);
12809 return bs
.index_ctx
.operate(bs
.bucket_obj
, &o
);
12812 int RGWRados::cls_obj_complete_op(BucketShard
& bs
, const rgw_obj
& obj
, RGWModifyOp op
, string
& tag
,
12813 int64_t pool
, uint64_t epoch
,
12814 rgw_bucket_dir_entry
& ent
, RGWObjCategory category
,
12815 list
<rgw_obj_index_key
> *remove_objs
, uint16_t bilog_flags
, rgw_zone_set
*_zones_trace
)
12817 ObjectWriteOperation o
;
12818 rgw_bucket_dir_entry_meta dir_meta
;
12819 dir_meta
= ent
.meta
;
12820 dir_meta
.category
= category
;
12822 rgw_bucket_entry_ver ver
;
12825 cls_rgw_obj_key
key(ent
.key
.name
, ent
.key
.instance
);
12826 cls_rgw_guard_bucket_resharding(o
, -ERR_BUSY_RESHARDING
);
12827 cls_rgw_bucket_complete_op(o
, op
, tag
, ver
, key
, dir_meta
, remove_objs
,
12828 get_zone().log_data
, bilog_flags
, _zones_trace
);
12829 complete_op_data
*arg
;
12830 index_completion_manager
->create_completion(obj
, op
, tag
, ver
, key
, dir_meta
, remove_objs
,
12831 get_zone().log_data
, bilog_flags
, _zones_trace
, &arg
);
12832 librados::AioCompletion
*completion
= arg
->rados_completion
;
12833 int ret
= bs
.index_ctx
.aio_operate(bs
.bucket_obj
, arg
->rados_completion
, &o
);
12834 completion
->release(); /* can't reference arg here, as it might have already been released */
12838 int RGWRados::cls_obj_complete_add(BucketShard
& bs
, const rgw_obj
& obj
, string
& tag
,
12839 int64_t pool
, uint64_t epoch
,
12840 rgw_bucket_dir_entry
& ent
, RGWObjCategory category
,
12841 list
<rgw_obj_index_key
> *remove_objs
, uint16_t bilog_flags
, rgw_zone_set
*zones_trace
)
12843 return cls_obj_complete_op(bs
, obj
, CLS_RGW_OP_ADD
, tag
, pool
, epoch
, ent
, category
, remove_objs
, bilog_flags
, zones_trace
);
12846 int RGWRados::cls_obj_complete_del(BucketShard
& bs
, string
& tag
,
12847 int64_t pool
, uint64_t epoch
,
12849 real_time
& removed_mtime
,
12850 list
<rgw_obj_index_key
> *remove_objs
,
12851 uint16_t bilog_flags
,
12852 rgw_zone_set
*zones_trace
)
12854 rgw_bucket_dir_entry ent
;
12855 ent
.meta
.mtime
= removed_mtime
;
12856 obj
.key
.get_index_key(&ent
.key
);
12857 return cls_obj_complete_op(bs
, obj
, CLS_RGW_OP_DEL
, tag
, pool
, epoch
, ent
, RGW_OBJ_CATEGORY_NONE
, remove_objs
, bilog_flags
, zones_trace
);
12860 int RGWRados::cls_obj_complete_cancel(BucketShard
& bs
, string
& tag
, rgw_obj
& obj
, uint16_t bilog_flags
, rgw_zone_set
*zones_trace
)
12862 rgw_bucket_dir_entry ent
;
12863 obj
.key
.get_index_key(&ent
.key
);
12864 return cls_obj_complete_op(bs
, obj
, CLS_RGW_OP_CANCEL
, tag
, -1 /* pool id */, 0, ent
, RGW_OBJ_CATEGORY_NONE
, NULL
, bilog_flags
, zones_trace
);
12867 int RGWRados::cls_obj_set_bucket_tag_timeout(RGWBucketInfo
& bucket_info
, uint64_t timeout
)
12869 librados::IoCtx index_ctx
;
12870 map
<int, string
> bucket_objs
;
12871 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
);
12875 return CLSRGWIssueSetTagTimeout(index_ctx
, bucket_objs
, cct
->_conf
->rgw_bucket_index_max_aio
, timeout
)();
12878 int RGWRados::cls_bucket_list(RGWBucketInfo
& bucket_info
, int shard_id
, rgw_obj_index_key
& start
, const string
& prefix
,
12879 uint32_t num_entries
, bool list_versions
, map
<string
, rgw_bucket_dir_entry
>& m
,
12880 bool *is_truncated
, rgw_obj_index_key
*last_entry
,
12881 bool (*force_check_filter
)(const string
& name
))
12883 ldout(cct
, 10) << "cls_bucket_list " << bucket_info
.bucket
<< " start " << start
.name
<< "[" << start
.instance
<< "] num_entries " << num_entries
<< dendl
;
12885 librados::IoCtx index_ctx
;
12886 // key - oid (for different shards if there is any)
12887 // value - list result for the corresponding oid (shard), it is filled by the AIO callback
12888 map
<int, string
> oids
;
12889 map
<int, struct rgw_cls_list_ret
> list_results
;
12890 int r
= open_bucket_index(bucket_info
, index_ctx
, oids
, shard_id
);
12894 cls_rgw_obj_key
start_key(start
.name
, start
.instance
);
12895 r
= CLSRGWIssueBucketList(index_ctx
, start_key
, prefix
, num_entries
, list_versions
,
12896 oids
, list_results
, cct
->_conf
->rgw_bucket_index_max_aio
)();
12900 // Create a list of iterators that are used to iterate each shard
12901 vector
<map
<string
, struct rgw_bucket_dir_entry
>::iterator
> vcurrents(list_results
.size());
12902 vector
<map
<string
, struct rgw_bucket_dir_entry
>::iterator
> vends(list_results
.size());
12903 vector
<string
> vnames(list_results
.size());
12904 map
<int, struct rgw_cls_list_ret
>::iterator iter
= list_results
.begin();
12905 *is_truncated
= false;
12906 for (; iter
!= list_results
.end(); ++iter
) {
12907 vcurrents
.push_back(iter
->second
.dir
.m
.begin());
12908 vends
.push_back(iter
->second
.dir
.m
.end());
12909 vnames
.push_back(oids
[iter
->first
]);
12910 *is_truncated
= (*is_truncated
|| iter
->second
.is_truncated
);
12913 // Create a map to track the next candidate entry from each shard, if the entry
12914 // from a specified shard is selected/erased, the next entry from that shard will
12915 // be inserted for next round selection
12916 map
<string
, size_t> candidates
;
12917 for (size_t i
= 0; i
< vcurrents
.size(); ++i
) {
12918 if (vcurrents
[i
] != vends
[i
]) {
12919 candidates
[vcurrents
[i
]->first
] = i
;
12923 map
<string
, bufferlist
> updates
;
12924 uint32_t count
= 0;
12925 while (count
< num_entries
&& !candidates
.empty()) {
12927 // Select the next one
12928 int pos
= candidates
.begin()->second
;
12929 const string
& name
= vcurrents
[pos
]->first
;
12930 struct rgw_bucket_dir_entry
& dirent
= vcurrents
[pos
]->second
;
12932 bool force_check
= force_check_filter
&&
12933 force_check_filter(dirent
.key
.name
);
12934 if ((!dirent
.exists
&& !dirent
.is_delete_marker()) ||
12935 !dirent
.pending_map
.empty() ||
12937 /* there are uncommitted ops. We need to check the current state,
12938 * and if the tags are old we need to do cleanup as well. */
12939 librados::IoCtx sub_ctx
;
12940 sub_ctx
.dup(index_ctx
);
12941 r
= check_disk_state(sub_ctx
, bucket_info
, dirent
, dirent
, updates
[vnames
[pos
]]);
12942 if (r
< 0 && r
!= -ENOENT
) {
12947 ldout(cct
, 10) << "RGWRados::cls_bucket_list: got " << dirent
.key
.name
<< "[" << dirent
.key
.instance
<< "]" << dendl
;
12948 m
[name
] = std::move(dirent
);
12952 // Refresh the candidates map
12953 candidates
.erase(candidates
.begin());
12955 if (vcurrents
[pos
] != vends
[pos
]) {
12956 candidates
[vcurrents
[pos
]->first
] = pos
;
12960 // Suggest updates if there is any
12961 map
<string
, bufferlist
>::iterator miter
= updates
.begin();
12962 for (; miter
!= updates
.end(); ++miter
) {
12963 if (miter
->second
.length()) {
12964 ObjectWriteOperation o
;
12965 cls_rgw_suggest_changes(o
, miter
->second
);
12966 // we don't care if we lose suggested updates, send them off blindly
12967 AioCompletion
*c
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
12968 index_ctx
.aio_operate(miter
->first
, c
, &o
);
12973 // Check if all the returned entries are consumed or not
12974 for (size_t i
= 0; i
< vcurrents
.size(); ++i
) {
12975 if (vcurrents
[i
] != vends
[i
])
12976 *is_truncated
= true;
12979 *last_entry
= m
.rbegin()->first
;
12984 int RGWRados::cls_obj_usage_log_add(const string
& oid
, rgw_usage_log_info
& info
)
12986 rgw_raw_obj
obj(get_zone_params().usage_log_pool
, oid
);
12989 int r
= get_raw_obj_ref(obj
, &ref
);
12994 ObjectWriteOperation op
;
12995 cls_rgw_usage_log_add(op
, info
);
12997 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
13001 int RGWRados::cls_obj_usage_log_read(string
& oid
, string
& user
, uint64_t start_epoch
, uint64_t end_epoch
, uint32_t max_entries
,
13002 string
& read_iter
, map
<rgw_user_bucket
, rgw_usage_log_entry
>& usage
, bool *is_truncated
)
13004 rgw_raw_obj
obj(get_zone_params().usage_log_pool
, oid
);
13007 int r
= get_raw_obj_ref(obj
, &ref
);
13012 *is_truncated
= false;
13014 r
= cls_rgw_usage_log_read(ref
.ioctx
, ref
.oid
, user
, start_epoch
, end_epoch
,
13015 max_entries
, read_iter
, usage
, is_truncated
);
13020 int RGWRados::cls_obj_usage_log_trim(string
& oid
, string
& user
, uint64_t start_epoch
, uint64_t end_epoch
)
13022 rgw_raw_obj
obj(get_zone_params().usage_log_pool
, oid
);
13025 int r
= get_raw_obj_ref(obj
, &ref
);
13030 r
= cls_rgw_usage_log_trim(ref
.ioctx
, ref
.oid
, user
, start_epoch
, end_epoch
);
13034 int RGWRados::remove_objs_from_index(RGWBucketInfo
& bucket_info
, list
<rgw_obj_index_key
>& oid_list
)
13036 librados::IoCtx index_ctx
;
13039 uint8_t suggest_flag
= (get_zone().log_data
? CEPH_RGW_DIR_SUGGEST_LOG_OP
: 0);
13041 int r
= open_bucket_index(bucket_info
, index_ctx
, dir_oid
);
13045 bufferlist updates
;
13047 for (auto iter
= oid_list
.begin(); iter
!= oid_list
.end(); ++iter
) {
13048 rgw_bucket_dir_entry entry
;
13050 dout(2) << "RGWRados::remove_objs_from_index bucket=" << bucket_info
.bucket
<< " obj=" << entry
.key
.name
<< ":" << entry
.key
.instance
<< dendl
;
13051 entry
.ver
.epoch
= (uint64_t)-1; // ULLONG_MAX, needed to that objclass doesn't skip out request
13052 updates
.append(CEPH_RGW_REMOVE
| suggest_flag
);
13053 ::encode(entry
, updates
);
13058 r
= index_ctx
.exec(dir_oid
, RGW_CLASS
, RGW_DIR_SUGGEST_CHANGES
, updates
, out
);
13063 int RGWRados::check_disk_state(librados::IoCtx io_ctx
,
13064 const RGWBucketInfo
& bucket_info
,
13065 rgw_bucket_dir_entry
& list_state
,
13066 rgw_bucket_dir_entry
& object
,
13067 bufferlist
& suggested_updates
)
13069 const rgw_bucket
& bucket
= bucket_info
.bucket
;
13070 uint8_t suggest_flag
= (get_zone().log_data
? CEPH_RGW_DIR_SUGGEST_LOG_OP
: 0);
13074 rgw_obj
obj(bucket
, list_state
.key
);
13077 get_obj_bucket_and_oid_loc(obj
, oid
, loc
);
13079 if (loc
!= list_state
.locator
) {
13080 ldout(cct
, 0) << "WARNING: generated locator (" << loc
<< ") is different from listed locator (" << list_state
.locator
<< ")" << dendl
;
13083 io_ctx
.locator_set_key(list_state
.locator
);
13085 RGWObjState
*astate
= NULL
;
13086 RGWObjectCtx
rctx(this);
13087 int r
= get_obj_state(&rctx
, bucket_info
, obj
, &astate
, false);
13091 list_state
.pending_map
.clear(); // we don't need this and it inflates size
13092 if (!astate
->exists
) {
13093 /* object doesn't exist right now -- hopefully because it's
13094 * marked as !exists and got deleted */
13095 if (list_state
.exists
) {
13096 /* FIXME: what should happen now? Work out if there are any
13097 * non-bad ways this could happen (there probably are, but annoying
13100 // encode a suggested removal of that key
13101 list_state
.ver
.epoch
= io_ctx
.get_last_version();
13102 list_state
.ver
.pool
= io_ctx
.get_id();
13103 cls_rgw_encode_suggestion(CEPH_RGW_REMOVE
, list_state
, suggested_updates
);
13108 string content_type
;
13111 object
.meta
.size
= astate
->size
;
13112 object
.meta
.accounted_size
= astate
->accounted_size
;
13113 object
.meta
.mtime
= astate
->mtime
;
13115 map
<string
, bufferlist
>::iterator iter
= astate
->attrset
.find(RGW_ATTR_ETAG
);
13116 if (iter
!= astate
->attrset
.end()) {
13117 etag
= iter
->second
.c_str();
13119 iter
= astate
->attrset
.find(RGW_ATTR_CONTENT_TYPE
);
13120 if (iter
!= astate
->attrset
.end()) {
13121 content_type
= iter
->second
.c_str();
13123 iter
= astate
->attrset
.find(RGW_ATTR_ACL
);
13124 if (iter
!= astate
->attrset
.end()) {
13125 r
= decode_policy(iter
->second
, &owner
);
13127 dout(0) << "WARNING: could not decode policy for object: " << obj
<< dendl
;
13131 if (astate
->has_manifest
) {
13132 RGWObjManifest::obj_iterator miter
;
13133 RGWObjManifest
& manifest
= astate
->manifest
;
13134 for (miter
= manifest
.obj_begin(); miter
!= manifest
.obj_end(); ++miter
) {
13135 const rgw_raw_obj
& raw_loc
= miter
.get_location().get_raw_obj(this);
13137 rgw_raw_obj_to_obj(manifest
.get_obj().bucket
, raw_loc
, &loc
);
13139 if (loc
.key
.ns
== RGW_OBJ_NS_MULTIPART
) {
13140 dout(10) << "check_disk_state(): removing manifest part from index: " << loc
<< dendl
;
13141 r
= delete_obj_index(loc
);
13143 dout(0) << "WARNING: delete_obj_index() returned r=" << r
<< dendl
;
13149 object
.meta
.etag
= etag
;
13150 object
.meta
.content_type
= content_type
;
13151 object
.meta
.owner
= owner
.get_id().to_str();
13152 object
.meta
.owner_display_name
= owner
.get_display_name();
13154 // encode suggested updates
13155 list_state
.ver
.pool
= io_ctx
.get_id();
13156 list_state
.ver
.epoch
= astate
->epoch
;
13157 list_state
.meta
.size
= object
.meta
.size
;
13158 list_state
.meta
.accounted_size
= object
.meta
.accounted_size
;
13159 list_state
.meta
.mtime
= object
.meta
.mtime
;
13160 list_state
.meta
.category
= main_category
;
13161 list_state
.meta
.etag
= etag
;
13162 list_state
.meta
.content_type
= content_type
;
13163 if (astate
->obj_tag
.length() > 0)
13164 list_state
.tag
= astate
->obj_tag
.c_str();
13165 list_state
.meta
.owner
= owner
.get_id().to_str();
13166 list_state
.meta
.owner_display_name
= owner
.get_display_name();
13168 list_state
.exists
= true;
13169 cls_rgw_encode_suggestion(CEPH_RGW_UPDATE
| suggest_flag
, list_state
, suggested_updates
);
13173 int RGWRados::cls_bucket_head(const RGWBucketInfo
& bucket_info
, int shard_id
, map
<string
, struct rgw_bucket_dir_header
>& headers
, map
<int, string
> *bucket_instance_ids
)
13175 librados::IoCtx index_ctx
;
13176 map
<int, string
> oids
;
13177 map
<int, struct rgw_cls_list_ret
> list_results
;
13178 int r
= open_bucket_index(bucket_info
, index_ctx
, oids
, list_results
, shard_id
, bucket_instance_ids
);
13182 r
= CLSRGWIssueGetDirHeader(index_ctx
, oids
, list_results
, cct
->_conf
->rgw_bucket_index_max_aio
)();
13186 map
<int, struct rgw_cls_list_ret
>::iterator iter
= list_results
.begin();
13187 for(; iter
!= list_results
.end(); ++iter
) {
13188 headers
[oids
[iter
->first
]] = iter
->second
.dir
.header
;
13193 int RGWRados::cls_bucket_head_async(const RGWBucketInfo
& bucket_info
, int shard_id
, RGWGetDirHeader_CB
*ctx
, int *num_aio
)
13195 librados::IoCtx index_ctx
;
13196 map
<int, string
> bucket_objs
;
13197 int r
= open_bucket_index(bucket_info
, index_ctx
, bucket_objs
, shard_id
);
13201 map
<int, string
>::iterator iter
= bucket_objs
.begin();
13202 for (; iter
!= bucket_objs
.end(); ++iter
) {
13203 r
= cls_rgw_get_dir_header_async(index_ctx
, iter
->second
, static_cast<RGWGetDirHeader_CB
*>(ctx
->get()));
13214 int RGWRados::cls_user_get_header(const string
& user_id
, cls_user_header
*header
)
13216 string buckets_obj_id
;
13217 rgw_get_buckets_obj(user_id
, buckets_obj_id
);
13218 rgw_raw_obj
obj(get_zone_params().user_uid_pool
, buckets_obj_id
);
13221 int r
= get_raw_obj_ref(obj
, &ref
);
13226 librados::ObjectReadOperation op
;
13228 ::cls_user_get_header(op
, header
, &rc
);
13230 r
= ref
.ioctx
.operate(ref
.oid
, &op
, &ibl
);
13239 int RGWRados::cls_user_get_header_async(const string
& user_id
, RGWGetUserHeader_CB
*ctx
)
13241 string buckets_obj_id
;
13242 rgw_get_buckets_obj(user_id
, buckets_obj_id
);
13243 rgw_raw_obj
obj(get_zone_params().user_uid_pool
, buckets_obj_id
);
13246 int r
= get_raw_obj_ref(obj
, &ref
);
13251 r
= ::cls_user_get_header_async(ref
.ioctx
, ref
.oid
, ctx
);
13258 int RGWRados::cls_user_sync_bucket_stats(rgw_raw_obj
& user_obj
, const RGWBucketInfo
& bucket_info
)
13260 map
<string
, struct rgw_bucket_dir_header
> headers
;
13261 int r
= cls_bucket_head(bucket_info
, RGW_NO_SHARD
, headers
);
13263 ldout(cct
, 20) << "cls_bucket_header() returned " << r
<< dendl
;
13267 cls_user_bucket_entry entry
;
13269 bucket_info
.bucket
.convert(&entry
.bucket
);
13271 for (const auto& hiter
: headers
) {
13272 for (const auto& iter
: hiter
.second
.stats
) {
13273 const struct rgw_bucket_category_stats
& header_stats
= iter
.second
;
13274 entry
.size
+= header_stats
.total_size
;
13275 entry
.size_rounded
+= header_stats
.total_size_rounded
;
13276 entry
.count
+= header_stats
.num_entries
;
13280 list
<cls_user_bucket_entry
> entries
;
13281 entries
.push_back(entry
);
13283 r
= cls_user_update_buckets(user_obj
, entries
, false);
13285 ldout(cct
, 20) << "cls_user_update_buckets() returned " << r
<< dendl
;
13292 int RGWRados::cls_user_get_bucket_stats(const rgw_bucket
& bucket
, cls_user_bucket_entry
& entry
)
13294 map
<string
, struct rgw_bucket_dir_header
> headers
;
13295 RGWBucketInfo bucket_info
;
13296 RGWObjectCtx
obj_ctx(this);
13297 int ret
= get_bucket_instance_info(obj_ctx
, bucket
, bucket_info
, NULL
, NULL
);
13302 ret
= cls_bucket_head(bucket_info
, RGW_NO_SHARD
, headers
);
13304 ldout(cct
, 20) << "cls_bucket_header() returned " << ret
<< dendl
;
13308 bucket
.convert(&entry
.bucket
);
13310 for (const auto& hiter
: headers
) {
13311 for (const auto& iter
: hiter
.second
.stats
) {
13312 const struct rgw_bucket_category_stats
& header_stats
= iter
.second
;
13313 entry
.size
+= header_stats
.total_size
;
13314 entry
.size_rounded
+= header_stats
.total_size_rounded
;
13315 entry
.count
+= header_stats
.num_entries
;
13322 int RGWRados::cls_user_list_buckets(rgw_raw_obj
& obj
,
13323 const string
& in_marker
,
13324 const string
& end_marker
,
13325 const int max_entries
,
13326 list
<cls_user_bucket_entry
>& entries
,
13327 string
* const out_marker
,
13328 bool * const truncated
)
13331 int r
= get_raw_obj_ref(obj
, &ref
);
13336 librados::ObjectReadOperation op
;
13339 cls_user_bucket_list(op
, in_marker
, end_marker
, max_entries
, entries
, out_marker
, truncated
, &rc
);
13341 r
= ref
.ioctx
.operate(ref
.oid
, &op
, &ibl
);
13350 int RGWRados::cls_user_update_buckets(rgw_raw_obj
& obj
, list
<cls_user_bucket_entry
>& entries
, bool add
)
13353 int r
= get_raw_obj_ref(obj
, &ref
);
13358 librados::ObjectWriteOperation op
;
13359 cls_user_set_buckets(op
, entries
, add
);
13360 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
13367 int RGWRados::complete_sync_user_stats(const rgw_user
& user_id
)
13369 string buckets_obj_id
;
13370 rgw_get_buckets_obj(user_id
, buckets_obj_id
);
13371 rgw_raw_obj
obj(get_zone_params().user_uid_pool
, buckets_obj_id
);
13372 return cls_user_complete_stats_sync(obj
);
13375 int RGWRados::cls_user_complete_stats_sync(rgw_raw_obj
& obj
)
13378 int r
= get_raw_obj_ref(obj
, &ref
);
13383 librados::ObjectWriteOperation op
;
13384 ::cls_user_complete_stats_sync(op
);
13385 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
13392 int RGWRados::cls_user_add_bucket(rgw_raw_obj
& obj
, const cls_user_bucket_entry
& entry
)
13394 list
<cls_user_bucket_entry
> l
;
13395 l
.push_back(entry
);
13397 return cls_user_update_buckets(obj
, l
, true);
13400 int RGWRados::cls_user_remove_bucket(rgw_raw_obj
& obj
, const cls_user_bucket
& bucket
)
13403 int r
= get_system_obj_ref(obj
, &ref
);
13408 librados::ObjectWriteOperation op
;
13409 ::cls_user_remove_bucket(op
, bucket
);
13410 r
= ref
.ioctx
.operate(ref
.oid
, &op
);
13417 int RGWRados::check_bucket_shards(const RGWBucketInfo
& bucket_info
, const rgw_bucket
& bucket
,
13418 RGWQuotaInfo
& bucket_quota
)
13420 if (!cct
->_conf
->rgw_dynamic_resharding
) {
13424 bool need_resharding
= false;
13425 int num_source_shards
= (bucket_info
.num_shards
> 0 ? bucket_info
.num_shards
: 1);
13426 uint32_t suggested_num_shards
;
13428 int ret
= quota_handler
->check_bucket_shards((uint64_t)cct
->_conf
->rgw_max_objs_per_shard
,
13429 num_source_shards
, bucket_info
.owner
, bucket
, bucket_quota
,
13430 1, need_resharding
, &suggested_num_shards
);
13435 if (need_resharding
) {
13436 ldout(cct
, 20) << __func__
<< " bucket " << bucket
.name
<< " need resharding " <<
13437 " old num shards " << bucket_info
.num_shards
<< " new num shards " << suggested_num_shards
<<
13439 return add_bucket_to_reshard(bucket_info
, suggested_num_shards
);
13445 int RGWRados::add_bucket_to_reshard(const RGWBucketInfo
& bucket_info
, uint32_t new_num_shards
)
13447 RGWReshard
reshard(this);
13449 uint32_t num_source_shards
= (bucket_info
.num_shards
> 0 ? bucket_info
.num_shards
: 1);
13451 new_num_shards
= min(new_num_shards
, get_max_bucket_shards());
13452 if (new_num_shards
<= num_source_shards
) {
13453 ldout(cct
, 20) << "not resharding bucket name=" << bucket_info
.bucket
.name
<< ", orig_num=" << num_source_shards
<< ", new_num_shards=" << new_num_shards
<< dendl
;
13457 cls_rgw_reshard_entry entry
;
13458 entry
.time
= real_clock::now();
13459 entry
.tenant
= bucket_info
.owner
.tenant
;
13460 entry
.bucket_name
= bucket_info
.bucket
.name
;
13461 entry
.bucket_id
= bucket_info
.bucket
.bucket_id
;
13462 entry
.old_num_shards
= num_source_shards
;
13463 entry
.new_num_shards
= new_num_shards
;
13465 return reshard
.add(entry
);
13468 int RGWRados::check_quota(const rgw_user
& bucket_owner
, rgw_bucket
& bucket
,
13469 RGWQuotaInfo
& user_quota
, RGWQuotaInfo
& bucket_quota
, uint64_t obj_size
)
13471 return quota_handler
->check_quota(bucket_owner
, bucket
, user_quota
, bucket_quota
, 1, obj_size
);
13474 void RGWRados::get_bucket_index_objects(const string
& bucket_oid_base
,
13475 uint32_t num_shards
, map
<int, string
>& bucket_objects
, int shard_id
)
13478 bucket_objects
[0] = bucket_oid_base
;
13480 char buf
[bucket_oid_base
.size() + 32];
13481 if (shard_id
< 0) {
13482 for (uint32_t i
= 0; i
< num_shards
; ++i
) {
13483 snprintf(buf
, sizeof(buf
), "%s.%d", bucket_oid_base
.c_str(), i
);
13484 bucket_objects
[i
] = buf
;
13487 if ((uint32_t)shard_id
> num_shards
) {
13490 snprintf(buf
, sizeof(buf
), "%s.%d", bucket_oid_base
.c_str(), shard_id
);
13491 bucket_objects
[shard_id
] = buf
;
13496 void RGWRados::get_bucket_instance_ids(const RGWBucketInfo
& bucket_info
, int shard_id
, map
<int, string
> *result
)
13498 const rgw_bucket
& bucket
= bucket_info
.bucket
;
13499 string plain_id
= bucket
.name
+ ":" + bucket
.bucket_id
;
13500 if (!bucket_info
.num_shards
) {
13501 (*result
)[0] = plain_id
;
13504 if (shard_id
< 0) {
13505 for (uint32_t i
= 0; i
< bucket_info
.num_shards
; ++i
) {
13506 snprintf(buf
, sizeof(buf
), ":%d", i
);
13507 (*result
)[i
] = plain_id
+ buf
;
13510 if ((uint32_t)shard_id
> bucket_info
.num_shards
) {
13513 snprintf(buf
, sizeof(buf
), ":%d", shard_id
);
13514 (*result
)[shard_id
] = plain_id
+ buf
;
13519 int RGWRados::get_target_shard_id(const RGWBucketInfo
& bucket_info
, const string
& obj_key
,
13523 switch (bucket_info
.bucket_index_shard_hash_type
) {
13524 case RGWBucketInfo::MOD
:
13525 if (!bucket_info
.num_shards
) {
13530 uint32_t sid
= ceph_str_hash_linux(obj_key
.c_str(), obj_key
.size());
13531 uint32_t sid2
= sid
^ ((sid
& 0xFF) << 24);
13532 sid
= rgw_shards_mod(sid2
, bucket_info
.num_shards
);
13534 *shard_id
= (int)sid
;
13544 void RGWRados::get_bucket_index_object(const string
& bucket_oid_base
, uint32_t num_shards
,
13545 int shard_id
, string
*bucket_obj
)
13548 // By default with no sharding, we use the bucket oid as itself
13549 (*bucket_obj
) = bucket_oid_base
;
13551 char buf
[bucket_oid_base
.size() + 32];
13552 snprintf(buf
, sizeof(buf
), "%s.%d", bucket_oid_base
.c_str(), shard_id
);
13553 (*bucket_obj
) = buf
;
13557 int RGWRados::get_bucket_index_object(const string
& bucket_oid_base
, const string
& obj_key
,
13558 uint32_t num_shards
, RGWBucketInfo::BIShardsHashType hash_type
, string
*bucket_obj
, int *shard_id
)
13561 switch (hash_type
) {
13562 case RGWBucketInfo::MOD
:
13564 // By default with no sharding, we use the bucket oid as itself
13565 (*bucket_obj
) = bucket_oid_base
;
13570 uint32_t sid
= ceph_str_hash_linux(obj_key
.c_str(), obj_key
.size());
13571 uint32_t sid2
= sid
^ ((sid
& 0xFF) << 24);
13572 sid
= rgw_shards_mod(sid2
, num_shards
);
13573 char buf
[bucket_oid_base
.size() + 32];
13574 snprintf(buf
, sizeof(buf
), "%s.%d", bucket_oid_base
.c_str(), sid
);
13575 (*bucket_obj
) = buf
;
13577 *shard_id
= (int)sid
;
13587 void RGWStateLog::oid_str(int shard
, string
& oid
) {
13588 oid
= RGW_STATELOG_OBJ_PREFIX
+ module_name
+ ".";
13590 snprintf(buf
, sizeof(buf
), "%d", shard
);
13594 int RGWStateLog::get_shard_num(const string
& object
) {
13595 uint32_t val
= ceph_str_hash_linux(object
.c_str(), object
.length());
13596 return val
% num_shards
;
13599 string
RGWStateLog::get_oid(const string
& object
) {
13600 int shard
= get_shard_num(object
);
13602 oid_str(shard
, oid
);
13606 int RGWStateLog::open_ioctx(librados::IoCtx
& ioctx
) {
13608 store
->get_log_pool(pool
);
13609 int r
= rgw_init_ioctx(store
->get_rados_handle(), pool
, ioctx
);
13611 lderr(store
->ctx()) << "ERROR: could not open rados pool" << dendl
;
13617 int RGWStateLog::store_entry(const string
& client_id
, const string
& op_id
, const string
& object
,
13618 uint32_t state
, bufferlist
*bl
, uint32_t *check_state
)
13620 if (client_id
.empty() ||
13623 ldout(store
->ctx(), 0) << "client_id / op_id / object is empty" << dendl
;
13626 librados::IoCtx ioctx
;
13627 int r
= open_ioctx(ioctx
);
13631 string oid
= get_oid(object
);
13633 librados::ObjectWriteOperation op
;
13635 cls_statelog_check_state(op
, client_id
, op_id
, object
, *check_state
);
13637 utime_t ts
= ceph_clock_now();
13639 cls_statelog_add(op
, client_id
, op_id
, object
, ts
, state
, (bl
? *bl
: nobl
));
13640 r
= ioctx
.operate(oid
, &op
);
13648 int RGWStateLog::remove_entry(const string
& client_id
, const string
& op_id
, const string
& object
)
13650 if (client_id
.empty() ||
13653 ldout(store
->ctx(), 0) << "client_id / op_id / object is empty" << dendl
;
13656 librados::IoCtx ioctx
;
13657 int r
= open_ioctx(ioctx
);
13661 string oid
= get_oid(object
);
13663 librados::ObjectWriteOperation op
;
13664 cls_statelog_remove_by_object(op
, object
, op_id
);
13665 r
= ioctx
.operate(oid
, &op
);
13673 void RGWStateLog::init_list_entries(const string
& client_id
, const string
& op_id
, const string
& object
,
13676 list_state
*state
= new list_state
;
13677 state
->client_id
= client_id
;
13678 state
->op_id
= op_id
;
13679 state
->object
= object
;
13680 if (object
.empty()) {
13681 state
->cur_shard
= 0;
13682 state
->max_shard
= num_shards
- 1;
13684 state
->cur_shard
= state
->max_shard
= get_shard_num(object
);
13686 *handle
= (void *)state
;
13689 int RGWStateLog::list_entries(void *handle
, int max_entries
,
13690 list
<cls_statelog_entry
>& entries
,
13693 list_state
*state
= static_cast<list_state
*>(handle
);
13695 librados::IoCtx ioctx
;
13696 int r
= open_ioctx(ioctx
);
13702 for (; state
->cur_shard
<= state
->max_shard
&& max_entries
> 0; ++state
->cur_shard
) {
13704 oid_str(state
->cur_shard
, oid
);
13706 librados::ObjectReadOperation op
;
13707 list
<cls_statelog_entry
> ents
;
13709 cls_statelog_list(op
, state
->client_id
, state
->op_id
, state
->object
, state
->marker
,
13710 max_entries
, ents
, &state
->marker
, &truncated
);
13712 r
= ioctx
.operate(oid
, &op
, &ibl
);
13713 if (r
== -ENOENT
) {
13718 ldout(store
->ctx(), 0) << "cls_statelog_list returned " << r
<< dendl
;
13723 state
->marker
.clear();
13726 max_entries
-= ents
.size();
13728 entries
.splice(entries
.end(), ents
);
13734 *done
= (state
->cur_shard
> state
->max_shard
);
13739 void RGWStateLog::finish_list_entries(void *handle
)
13741 list_state
*state
= static_cast<list_state
*>(handle
);
13745 void RGWStateLog::dump_entry(const cls_statelog_entry
& entry
, Formatter
*f
)
13747 f
->open_object_section("statelog_entry");
13748 f
->dump_string("client_id", entry
.client_id
);
13749 f
->dump_string("op_id", entry
.op_id
);
13750 f
->dump_string("object", entry
.object
);
13751 entry
.timestamp
.gmtime_nsec(f
->dump_stream("timestamp"));
13752 if (!dump_entry_internal(entry
, f
)) {
13753 f
->dump_int("state", entry
.state
);
13755 f
->close_section();
13758 RGWOpState::RGWOpState(RGWRados
*_store
) : RGWStateLog(_store
, _store
->ctx()->_conf
->rgw_num_zone_opstate_shards
, string("obj_opstate"))
13762 bool RGWOpState::dump_entry_internal(const cls_statelog_entry
& entry
, Formatter
*f
)
13765 switch ((OpState
)entry
.state
) {
13766 case OPSTATE_UNKNOWN
:
13769 case OPSTATE_IN_PROGRESS
:
13772 case OPSTATE_COMPLETE
:
13775 case OPSTATE_ERROR
:
13778 case OPSTATE_ABORT
:
13781 case OPSTATE_CANCELLED
:
13787 f
->dump_string("state", s
);
13791 int RGWOpState::state_from_str(const string
& s
, OpState
*state
)
13793 if (s
== "unknown") {
13794 *state
= OPSTATE_UNKNOWN
;
13795 } else if (s
== "in-progress") {
13796 *state
= OPSTATE_IN_PROGRESS
;
13797 } else if (s
== "complete") {
13798 *state
= OPSTATE_COMPLETE
;
13799 } else if (s
== "error") {
13800 *state
= OPSTATE_ERROR
;
13801 } else if (s
== "abort") {
13802 *state
= OPSTATE_ABORT
;
13803 } else if (s
== "cancelled") {
13804 *state
= OPSTATE_CANCELLED
;
13812 int RGWOpState::set_state(const string
& client_id
, const string
& op_id
, const string
& object
, OpState state
)
13814 uint32_t s
= (uint32_t)state
;
13815 return store_entry(client_id
, op_id
, object
, s
, NULL
, NULL
);
13818 int RGWOpState::renew_state(const string
& client_id
, const string
& op_id
, const string
& object
, OpState state
)
13820 uint32_t s
= (uint32_t)state
;
13821 return store_entry(client_id
, op_id
, object
, s
, NULL
, &s
);
13824 RGWOpStateSingleOp::RGWOpStateSingleOp(RGWRados
*store
, const string
& cid
, const string
& oid
,
13825 const string
& obj
) : os(store
), client_id(cid
), op_id(oid
), object(obj
)
13827 cct
= store
->ctx();
13828 cur_state
= RGWOpState::OPSTATE_UNKNOWN
;
13831 int RGWOpStateSingleOp::set_state(RGWOpState::OpState state
) {
13832 last_update
= real_clock::now();
13834 return os
.set_state(client_id
, op_id
, object
, state
);
13837 int RGWOpStateSingleOp::renew_state() {
13838 real_time now
= real_clock::now();
13840 int rate_limit_sec
= cct
->_conf
->rgw_opstate_ratelimit_sec
;
13842 if (rate_limit_sec
&& now
- last_update
< make_timespan(rate_limit_sec
)) {
13847 return os
.renew_state(client_id
, op_id
, object
, cur_state
);
13851 uint64_t RGWRados::instance_id()
13853 return get_rados_handle()->get_instance_id();
13856 uint64_t RGWRados::next_bucket_id()
13858 Mutex::Locker
l(bucket_id_lock
);
13859 return ++max_bucket_id
;
13862 RGWRados
*RGWStoreManager::init_storage_provider(CephContext
*cct
, bool use_gc_thread
, bool use_lc_thread
, bool quota_threads
, bool run_sync_thread
, bool run_reshard_thread
)
13864 int use_cache
= cct
->_conf
->rgw_cache_enabled
;
13865 RGWRados
*store
= NULL
;
13867 store
= new RGWRados
;
13869 store
= new RGWCache
<RGWRados
>;
13872 if (store
->initialize(cct
, use_gc_thread
, use_lc_thread
, quota_threads
, run_sync_thread
, run_reshard_thread
) < 0) {
13880 RGWRados
*RGWStoreManager::init_raw_storage_provider(CephContext
*cct
)
13882 RGWRados
*store
= NULL
;
13883 store
= new RGWRados
;
13885 store
->set_context(cct
);
13887 if (store
->init_rados() < 0) {
13895 void RGWStoreManager::close_storage(RGWRados
*store
)
13905 librados::Rados
* RGWRados::get_rados_handle()
13907 if (rados
.size() == 1) {
13910 handle_lock
.get_read();
13911 pthread_t id
= pthread_self();
13912 std::map
<pthread_t
, int>:: iterator it
= rados_map
.find(id
);
13914 if (it
!= rados_map
.end()) {
13915 handle_lock
.put_read();
13916 return &rados
[it
->second
];
13918 handle_lock
.put_read();
13919 handle_lock
.get_write();
13920 const uint32_t handle
= next_rados_handle
;
13921 rados_map
[id
] = handle
;
13922 if (++next_rados_handle
== rados
.size()) {
13923 next_rados_handle
= 0;
13925 handle_lock
.put_write();
13926 return &rados
[handle
];
13931 int RGWRados::delete_raw_obj_aio(const rgw_raw_obj
& obj
, list
<librados::AioCompletion
*>& handles
)
13934 int ret
= get_raw_obj_ref(obj
, &ref
);
13936 lderr(cct
) << "ERROR: failed to get obj ref with ret=" << ret
<< dendl
;
13940 ObjectWriteOperation op
;
13941 list
<string
> prefixes
;
13942 cls_rgw_remove_obj(op
, prefixes
);
13944 AioCompletion
*c
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
13945 ret
= ref
.ioctx
.aio_operate(ref
.oid
, c
, &op
);
13947 lderr(cct
) << "ERROR: AioOperate failed with ret=" << ret
<< dendl
;
13952 handles
.push_back(c
);
13957 int RGWRados::delete_obj_aio(const rgw_obj
& obj
,
13958 RGWBucketInfo
& bucket_info
, RGWObjState
*astate
,
13959 list
<librados::AioCompletion
*>& handles
, bool keep_index_consistent
)
13962 int ret
= get_obj_head_ref(bucket_info
, obj
, &ref
);
13964 lderr(cct
) << "ERROR: failed to get obj ref with ret=" << ret
<< dendl
;
13968 if (keep_index_consistent
) {
13969 RGWRados::Bucket
bop(this, bucket_info
);
13970 RGWRados::Bucket::UpdateIndex
index_op(&bop
, obj
);
13972 ret
= index_op
.prepare(CLS_RGW_OP_DEL
, &astate
->write_tag
);
13974 lderr(cct
) << "ERROR: failed to prepare index op with ret=" << ret
<< dendl
;
13979 ObjectWriteOperation op
;
13980 list
<string
> prefixes
;
13981 cls_rgw_remove_obj(op
, prefixes
);
13983 AioCompletion
*c
= librados::Rados::aio_create_completion(NULL
, NULL
, NULL
);
13984 ret
= ref
.ioctx
.aio_operate(ref
.oid
, c
, &op
);
13986 lderr(cct
) << "ERROR: AioOperate failed with ret=" << ret
<< dendl
;
13991 handles
.push_back(c
);
13993 if (keep_index_consistent
) {
13994 ret
= delete_obj_index(obj
);
13996 lderr(cct
) << "ERROR: failed to delete obj index with ret=" << ret
<< dendl
;
14003 int rgw_compression_info_from_attrset(map
<string
, bufferlist
>& attrs
, bool& need_decompress
, RGWCompressionInfo
& cs_info
) {
14004 map
<string
, bufferlist
>::iterator value
= attrs
.find(RGW_ATTR_COMPRESSION
);
14005 if (value
!= attrs
.end()) {
14006 bufferlist::iterator bliter
= value
->second
.begin();
14008 ::decode(cs_info
, bliter
);
14009 } catch (buffer::error
& err
) {
14012 if (cs_info
.blocks
.size() == 0) {
14015 if (cs_info
.compression_type
!= "none")
14016 need_decompress
= true;
14018 need_decompress
= false;
14021 need_decompress
= false;