1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
7 * Copyright (c) 2015 Hewlett-Packard Development Company, L.P.
9 * This is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License version 2.1, as published by the Free Software
12 * Foundation. See file COPYING.
15 #include "include/compat.h"
16 #include "include/int_types.h"
17 #include "boost/tuple/tuple.hpp"
21 #include <sys/types.h>
27 #include <sys/ioctl.h>
29 #if defined(__linux__)
36 #include "include/linux_fiemap.h"
38 #include "common/xattr.h"
39 #include "chain_xattr.h"
41 #if defined(DARWIN) || defined(__FreeBSD__)
42 #include <sys/param.h>
43 #include <sys/mount.h>
50 #include "FileStore.h"
51 #include "GenericFileStoreBackend.h"
52 #include "BtrfsFileStoreBackend.h"
53 #include "XfsFileStoreBackend.h"
54 #include "ZFSFileStoreBackend.h"
55 #include "common/BackTrace.h"
56 #include "include/types.h"
57 #include "FileJournal.h"
59 #include "osd/osd_types.h"
60 #include "include/color.h"
61 #include "include/buffer.h"
63 #include "common/Timer.h"
64 #include "common/debug.h"
65 #include "common/errno.h"
66 #include "common/run_cmd.h"
67 #include "common/safe_io.h"
68 #include "common/perf_counters.h"
69 #include "common/sync_filesystem.h"
70 #include "common/fd.h"
71 #include "HashIndex.h"
72 #include "DBObjectMap.h"
73 #include "kv/KeyValueDB.h"
75 #include "common/ceph_crypto.h"
76 using ceph::crypto::SHA1
;
78 #include "include/assert.h"
80 #include "common/config.h"
81 #include "common/blkdev.h"
84 #define TRACEPOINT_DEFINE
85 #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE
86 #include "tracing/objectstore.h"
87 #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE
88 #undef TRACEPOINT_DEFINE
90 #define tracepoint(...)
93 #define dout_context cct
94 #define dout_subsys ceph_subsys_filestore
96 #define dout_prefix *_dout << "filestore(" << basedir << ") "
98 #define COMMIT_SNAP_ITEM "snap_%llu"
99 #define CLUSTER_SNAP_ITEM "clustersnap_%s"
101 #define REPLAY_GUARD_XATTR "user.cephos.seq"
102 #define GLOBAL_REPLAY_GUARD_XATTR "user.cephos.gseq"
104 // XATTR_SPILL_OUT_NAME as a xattr is used to maintain that indicates whether
105 // xattrs spill over into DBObjectMap, if XATTR_SPILL_OUT_NAME exists in file
106 // xattrs and the value is "no", it indicates no xattrs in DBObjectMap
107 #define XATTR_SPILL_OUT_NAME "user.cephos.spill_out"
108 #define XATTR_NO_SPILL_OUT "0"
109 #define XATTR_SPILL_OUT "1"
110 #define __FUNC__ __func__ << "(" << __LINE__ << ")"
112 //Initial features in new superblock.
113 static CompatSet
get_fs_initial_compat_set() {
114 CompatSet::FeatureSet ceph_osd_feature_compat
;
115 CompatSet::FeatureSet ceph_osd_feature_ro_compat
;
116 CompatSet::FeatureSet ceph_osd_feature_incompat
;
117 return CompatSet(ceph_osd_feature_compat
, ceph_osd_feature_ro_compat
,
118 ceph_osd_feature_incompat
);
121 //Features are added here that this FileStore supports.
122 static CompatSet
get_fs_supported_compat_set() {
123 CompatSet compat
= get_fs_initial_compat_set();
124 //Any features here can be set in code, but not in initial superblock
125 compat
.incompat
.insert(CEPH_FS_FEATURE_INCOMPAT_SHARDS
);
129 int FileStore::validate_hobject_key(const hobject_t
&obj
) const
131 unsigned len
= LFNIndex::get_max_escaped_name_len(obj
);
132 return len
> m_filestore_max_xattr_value_size
? -ENAMETOOLONG
: 0;
135 int FileStore::get_block_device_fsid(CephContext
* cct
, const string
& path
,
138 // make sure we don't try to use aio or direct_io (and get annoying
139 // error messages from failing to do so); performance implications
140 // should be irrelevant for this use
141 FileJournal
j(cct
, *fsid
, 0, 0, path
.c_str(), false, false);
142 return j
.peek_fsid(*fsid
);
145 void FileStore::FSPerfTracker::update_from_perfcounters(
146 PerfCounters
&logger
)
148 os_commit_latency
.consume_next(
150 l_filestore_journal_latency
));
151 os_apply_latency
.consume_next(
153 l_filestore_apply_latency
));
157 ostream
& operator<<(ostream
& out
, const FileStore::OpSequencer
& s
)
159 return out
<< *s
.parent
;
162 int FileStore::get_cdir(const coll_t
& cid
, char *s
, int len
)
164 const string
&cid_str(cid
.to_str());
165 return snprintf(s
, len
, "%s/current/%s", basedir
.c_str(), cid_str
.c_str());
168 int FileStore::get_index(const coll_t
& cid
, Index
*index
)
170 int r
= index_manager
.get_index(cid
, basedir
, index
);
171 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
175 int FileStore::init_index(const coll_t
& cid
)
178 get_cdir(cid
, path
, sizeof(path
));
179 int r
= index_manager
.init_index(cid
, path
, target_version
);
180 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
184 int FileStore::lfn_find(const ghobject_t
& oid
, const Index
& index
, IndexedPath
*path
)
190 assert(NULL
!= index
.index
);
191 r
= (index
.index
)->lookup(oid
, path
, &exist
);
193 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
201 int FileStore::lfn_truncate(const coll_t
& cid
, const ghobject_t
& oid
, off_t length
)
204 int r
= lfn_open(cid
, oid
, false, &fd
);
207 r
= ::ftruncate(**fd
, length
);
210 if (r
>= 0 && m_filestore_sloppy_crc
) {
211 int rc
= backend
->_crc_update_truncate(**fd
, length
);
215 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
219 int FileStore::lfn_stat(const coll_t
& cid
, const ghobject_t
& oid
, struct stat
*buf
)
223 int r
= get_index(cid
, &index
);
227 assert(NULL
!= index
.index
);
228 RWLock::RLocker
l((index
.index
)->access_lock
);
230 r
= lfn_find(oid
, index
, &path
);
233 r
= ::stat(path
->path(), buf
);
239 int FileStore::lfn_open(const coll_t
& cid
,
240 const ghobject_t
& oid
,
247 bool need_lock
= true;
252 if (cct
->_conf
->filestore_odsync_write
) {
260 if (!((*index
).index
)) {
261 r
= get_index(cid
, index
);
263 dout(10) << __FUNC__
<< ": could not get index r = " << r
<< dendl
;
271 assert(NULL
!= (*index
).index
);
273 ((*index
).index
)->access_lock
.get_write();
276 *outfd
= fdcache
.lookup(oid
);
279 ((*index
).index
)->access_lock
.put_write();
287 IndexedPath
*path
= &path2
;
289 r
= (*index
)->lookup(oid
, path
, &exist
);
291 derr
<< "could not find " << oid
<< " in index: "
292 << cpp_strerror(-r
) << dendl
;
296 r
= ::open((*path
)->path(), flags
, 0644);
299 dout(10) << "error opening file " << (*path
)->path() << " with flags="
300 << flags
<< ": " << cpp_strerror(-r
) << dendl
;
304 if (create
&& (!exist
)) {
305 r
= (*index
)->created(oid
, (*path
)->path());
307 VOID_TEMP_FAILURE_RETRY(::close(fd
));
308 derr
<< "error creating " << oid
<< " (" << (*path
)->path()
309 << ") in index: " << cpp_strerror(-r
) << dendl
;
312 r
= chain_fsetxattr
<true, true>(
313 fd
, XATTR_SPILL_OUT_NAME
,
314 XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
));
316 VOID_TEMP_FAILURE_RETRY(::close(fd
));
317 derr
<< "error setting spillout xattr for oid " << oid
<< " (" << (*path
)->path()
318 << "):" << cpp_strerror(-r
) << dendl
;
325 *outfd
= fdcache
.add(oid
, fd
, &existed
);
327 TEMP_FAILURE_RETRY(::close(fd
));
330 *outfd
= std::make_shared
<FDCache::FD
>(fd
);
334 ((*index
).index
)->access_lock
.put_write();
342 ((*index
).index
)->access_lock
.put_write();
345 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
349 void FileStore::lfn_close(FDRef fd
)
353 int FileStore::lfn_link(const coll_t
& c
, const coll_t
& newcid
, const ghobject_t
& o
, const ghobject_t
& newoid
)
355 Index index_new
, index_old
;
356 IndexedPath path_new
, path_old
;
359 bool index_same
= false;
361 r
= get_index(newcid
, &index_new
);
364 r
= get_index(c
, &index_old
);
367 } else if (c
== newcid
) {
368 r
= get_index(c
, &index_old
);
371 index_new
= index_old
;
374 r
= get_index(c
, &index_old
);
377 r
= get_index(newcid
, &index_new
);
382 assert(NULL
!= index_old
.index
);
383 assert(NULL
!= index_new
.index
);
387 RWLock::RLocker
l1((index_old
.index
)->access_lock
);
389 r
= index_old
->lookup(o
, &path_old
, &exist
);
391 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
397 RWLock::WLocker
l2((index_new
.index
)->access_lock
);
399 r
= index_new
->lookup(newoid
, &path_new
, &exist
);
401 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
407 dout(25) << __FUNC__
<< ": path_old: " << path_old
<< dendl
;
408 dout(25) << __FUNC__
<< ": path_new: " << path_new
<< dendl
;
409 r
= ::link(path_old
->path(), path_new
->path());
413 r
= index_new
->created(newoid
, path_new
->path());
415 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
419 RWLock::WLocker
l1((index_old
.index
)->access_lock
);
421 r
= index_old
->lookup(o
, &path_old
, &exist
);
423 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
429 r
= index_new
->lookup(newoid
, &path_new
, &exist
);
431 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
437 dout(25) << __FUNC__
<< ": path_old: " << path_old
<< dendl
;
438 dout(25) << __FUNC__
<< ": path_new: " << path_new
<< dendl
;
439 r
= ::link(path_old
->path(), path_new
->path());
443 // make sure old fd for unlinked/overwritten file is gone
444 fdcache
.clear(newoid
);
446 r
= index_new
->created(newoid
, path_new
->path());
448 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
455 int FileStore::lfn_unlink(const coll_t
& cid
, const ghobject_t
& o
,
456 const SequencerPosition
&spos
,
457 bool force_clear_omap
)
460 int r
= get_index(cid
, &index
);
462 dout(25) << __FUNC__
<< ": get_index failed " << cpp_strerror(r
) << dendl
;
466 assert(NULL
!= index
.index
);
467 RWLock::WLocker
l((index
.index
)->access_lock
);
472 r
= index
->lookup(o
, &path
, &hardlink
);
474 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
478 if (!force_clear_omap
) {
479 if (hardlink
== 0 || hardlink
== 1) {
480 force_clear_omap
= true;
483 if (force_clear_omap
) {
484 dout(20) << __FUNC__
<< ": clearing omap on " << o
485 << " in cid " << cid
<< dendl
;
486 r
= object_map
->clear(o
, &spos
);
487 if (r
< 0 && r
!= -ENOENT
) {
488 dout(25) << __FUNC__
<< ": omap clear failed " << cpp_strerror(r
) << dendl
;
489 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
492 if (cct
->_conf
->filestore_debug_inject_read_err
) {
493 debug_obj_on_delete(o
);
495 if (!m_disable_wbthrottle
) {
496 wbthrottle
.clear_object(o
); // should be only non-cache ref
500 /* Ensure that replay of this op doesn't result in the object_map
503 if (!backend
->can_checkpoint())
504 object_map
->sync(&o
, &spos
);
507 if (!m_disable_wbthrottle
) {
508 wbthrottle
.clear_object(o
); // should be only non-cache ref
513 r
= index
->unlink(o
);
515 dout(25) << __FUNC__
<< ": index unlink failed " << cpp_strerror(r
) << dendl
;
521 FileStore::FileStore(CephContext
* cct
, const std::string
&base
,
522 const std::string
&jdev
, osflagbits_t flags
,
523 const char *name
, bool do_update
) :
524 JournalingObjectStore(cct
, base
),
526 basedir(base
), journalpath(jdev
),
527 generic_flags(flags
),
529 fsid_fd(-1), op_fd(-1),
530 basedir_fd(-1), current_fd(-1),
532 index_manager(cct
, do_update
),
533 lock("FileStore::lock"),
535 sync_entry_timeo_lock("FileStore::sync_entry_timeo_lock"),
536 timer(cct
, sync_entry_timeo_lock
),
537 stop(false), sync_thread(this),
541 m_disable_wbthrottle(cct
->_conf
->filestore_odsync_write
||
542 !cct
->_conf
->filestore_wbthrottle_enable
),
543 throttle_ops(cct
, "filestore_ops", cct
->_conf
->filestore_caller_concurrency
),
544 throttle_bytes(cct
, "filestore_bytes", cct
->_conf
->filestore_caller_concurrency
),
545 m_ondisk_finisher_num(cct
->_conf
->filestore_ondisk_finisher_threads
),
546 m_apply_finisher_num(cct
->_conf
->filestore_apply_finisher_threads
),
547 op_tp(cct
, "FileStore::op_tp", "tp_fstore_op", cct
->_conf
->filestore_op_threads
, "filestore_op_threads"),
548 op_wq(this, cct
->_conf
->filestore_op_thread_timeout
,
549 cct
->_conf
->filestore_op_thread_suicide_timeout
, &op_tp
),
551 trace_endpoint("0.0.0.0", 0, "FileStore"),
552 read_error_lock("FileStore::read_error_lock"),
553 m_filestore_commit_timeout(cct
->_conf
->filestore_commit_timeout
),
554 m_filestore_journal_parallel(cct
->_conf
->filestore_journal_parallel
),
555 m_filestore_journal_trailing(cct
->_conf
->filestore_journal_trailing
),
556 m_filestore_journal_writeahead(cct
->_conf
->filestore_journal_writeahead
),
557 m_filestore_fiemap_threshold(cct
->_conf
->filestore_fiemap_threshold
),
558 m_filestore_max_sync_interval(cct
->_conf
->filestore_max_sync_interval
),
559 m_filestore_min_sync_interval(cct
->_conf
->filestore_min_sync_interval
),
560 m_filestore_fail_eio(cct
->_conf
->filestore_fail_eio
),
561 m_filestore_fadvise(cct
->_conf
->filestore_fadvise
),
562 do_update(do_update
),
563 m_journal_dio(cct
->_conf
->journal_dio
),
564 m_journal_aio(cct
->_conf
->journal_aio
),
565 m_journal_force_aio(cct
->_conf
->journal_force_aio
),
566 m_osd_rollback_to_cluster_snap(cct
->_conf
->osd_rollback_to_cluster_snap
),
567 m_osd_use_stale_snap(cct
->_conf
->osd_use_stale_snap
),
568 m_filestore_do_dump(false),
569 m_filestore_dump_fmt(true),
570 m_filestore_sloppy_crc(cct
->_conf
->filestore_sloppy_crc
),
571 m_filestore_sloppy_crc_block_size(cct
->_conf
->filestore_sloppy_crc_block_size
),
572 m_filestore_max_alloc_hint_size(cct
->_conf
->filestore_max_alloc_hint_size
),
574 m_filestore_max_inline_xattr_size(0),
575 m_filestore_max_inline_xattrs(0),
576 m_filestore_max_xattr_value_size(0)
578 m_filestore_kill_at
= cct
->_conf
->filestore_kill_at
;
579 for (int i
= 0; i
< m_ondisk_finisher_num
; ++i
) {
581 oss
<< "filestore-ondisk-" << i
;
582 Finisher
*f
= new Finisher(cct
, oss
.str(), "fn_odsk_fstore");
583 ondisk_finishers
.push_back(f
);
585 for (int i
= 0; i
< m_apply_finisher_num
; ++i
) {
587 oss
<< "filestore-apply-" << i
;
588 Finisher
*f
= new Finisher(cct
, oss
.str(), "fn_appl_fstore");
589 apply_finishers
.push_back(f
);
593 oss
<< basedir
<< "/current";
594 current_fn
= oss
.str();
597 sss
<< basedir
<< "/current/commit_op_seq";
598 current_op_seq_fn
= sss
.str();
601 if (cct
->_conf
->filestore_omap_backend_path
!= "") {
602 omap_dir
= cct
->_conf
->filestore_omap_backend_path
;
604 omss
<< basedir
<< "/current/omap";
605 omap_dir
= omss
.str();
609 PerfCountersBuilder
plb(cct
, internal_name
, l_filestore_first
, l_filestore_last
);
611 plb
.add_u64(l_filestore_journal_queue_ops
, "journal_queue_ops", "Operations in journal queue");
612 plb
.add_u64(l_filestore_journal_ops
, "journal_ops", "Active journal entries to be applied");
613 plb
.add_u64(l_filestore_journal_queue_bytes
, "journal_queue_bytes", "Size of journal queue");
614 plb
.add_u64(l_filestore_journal_bytes
, "journal_bytes", "Active journal operation size to be applied");
615 plb
.add_time_avg(l_filestore_journal_latency
, "journal_latency", "Average journal queue completing latency");
616 plb
.add_u64_counter(l_filestore_journal_wr
, "journal_wr", "Journal write IOs");
617 plb
.add_u64_avg(l_filestore_journal_wr_bytes
, "journal_wr_bytes", "Journal data written");
618 plb
.add_u64(l_filestore_op_queue_max_ops
, "op_queue_max_ops", "Max operations in writing to FS queue");
619 plb
.add_u64(l_filestore_op_queue_ops
, "op_queue_ops", "Operations in writing to FS queue");
620 plb
.add_u64_counter(l_filestore_ops
, "ops", "Operations written to store");
621 plb
.add_u64(l_filestore_op_queue_max_bytes
, "op_queue_max_bytes", "Max data in writing to FS queue");
622 plb
.add_u64(l_filestore_op_queue_bytes
, "op_queue_bytes", "Size of writing to FS queue");
623 plb
.add_u64_counter(l_filestore_bytes
, "bytes", "Data written to store");
624 plb
.add_time_avg(l_filestore_apply_latency
, "apply_latency", "Apply latency");
625 plb
.add_u64(l_filestore_committing
, "committing", "Is currently committing");
627 plb
.add_u64_counter(l_filestore_commitcycle
, "commitcycle", "Commit cycles");
628 plb
.add_time_avg(l_filestore_commitcycle_interval
, "commitcycle_interval", "Average interval between commits");
629 plb
.add_time_avg(l_filestore_commitcycle_latency
, "commitcycle_latency", "Average latency of commit");
630 plb
.add_u64_counter(l_filestore_journal_full
, "journal_full", "Journal writes while full");
631 plb
.add_time_avg(l_filestore_queue_transaction_latency_avg
, "queue_transaction_latency_avg", "Store operation queue latency");
632 plb
.add_time(l_filestore_sync_pause_max_lat
, "sync_pause_max_latency", "Max latency of op_wq pause before syncfs");
634 logger
= plb
.create_perf_counters();
636 cct
->get_perfcounters_collection()->add(logger
);
637 cct
->_conf
->add_observer(this);
639 superblock
.compat_features
= get_fs_initial_compat_set();
642 FileStore::~FileStore()
644 for (vector
<Finisher
*>::iterator it
= ondisk_finishers
.begin(); it
!= ondisk_finishers
.end(); ++it
) {
648 for (vector
<Finisher
*>::iterator it
= apply_finishers
.begin(); it
!= apply_finishers
.end(); ++it
) {
652 cct
->_conf
->remove_observer(this);
653 cct
->get_perfcounters_collection()->remove(logger
);
656 journal
->logger
= NULL
;
659 if (m_filestore_do_dump
) {
664 static void get_attrname(const char *name
, char *buf
, int len
)
666 snprintf(buf
, len
, "user.ceph.%s", name
);
669 bool parse_attrname(char **name
)
671 if (strncmp(*name
, "user.ceph.", 10) == 0) {
678 void FileStore::collect_metadata(map
<string
,string
> *pm
)
680 char partition_path
[PATH_MAX
];
681 char dev_node
[PATH_MAX
];
684 (*pm
)["filestore_backend"] = backend
->get_name();
686 ss
<< "0x" << std::hex
<< m_fs_type
<< std::dec
;
687 (*pm
)["filestore_f_type"] = ss
.str();
689 if (cct
->_conf
->filestore_collect_device_partition_information
) {
690 rc
= get_device_by_fd(fsid_fd
, partition_path
, dev_node
, PATH_MAX
);
698 (*pm
)["backend_filestore_partition_path"] = "unknown";
699 (*pm
)["backend_filestore_dev_node"] = "unknown";
702 (*pm
)["backend_filestore_partition_path"] = string(partition_path
);
703 (*pm
)["backend_filestore_dev_node"] = "unknown";
706 (*pm
)["backend_filestore_partition_path"] = string(partition_path
);
707 (*pm
)["backend_filestore_dev_node"] = string(dev_node
);
711 int FileStore::statfs(struct store_statfs_t
*buf0
)
715 if (::statfs(basedir
.c_str(), &buf
) < 0) {
717 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
718 assert(r
!= -ENOENT
);
721 buf0
->total
= buf
.f_blocks
* buf
.f_bsize
;
722 buf0
->available
= buf
.f_bavail
* buf
.f_bsize
;
723 // Adjust for writes pending in the journal
725 uint64_t estimate
= journal
->get_journal_size_estimate();
726 if (buf0
->available
> estimate
)
727 buf0
->available
-= estimate
;
735 void FileStore::new_journal()
737 if (journalpath
.length()) {
738 dout(10) << "open_journal at " << journalpath
<< dendl
;
739 journal
= new FileJournal(cct
, fsid
, &finisher
, &sync_cond
,
741 m_journal_dio
, m_journal_aio
,
742 m_journal_force_aio
);
744 journal
->logger
= logger
;
749 int FileStore::dump_journal(ostream
& out
)
753 if (!journalpath
.length())
756 FileJournal
*journal
= new FileJournal(cct
, fsid
, &finisher
, &sync_cond
, journalpath
.c_str(), m_journal_dio
);
757 r
= journal
->dump(out
);
762 FileStoreBackend
*FileStoreBackend::create(long f_type
, FileStore
*fs
)
765 #if defined(__linux__)
766 case BTRFS_SUPER_MAGIC
:
767 return new BtrfsFileStoreBackend(fs
);
769 case XFS_SUPER_MAGIC
:
770 return new XfsFileStoreBackend(fs
);
774 case ZFS_SUPER_MAGIC
:
775 return new ZFSFileStoreBackend(fs
);
778 return new GenericFileStoreBackend(fs
);
782 void FileStore::create_backend(long f_type
)
786 assert(backend
== NULL
);
787 backend
= FileStoreBackend::create(f_type
, this);
789 dout(0) << "backend " << backend
->get_name()
790 << " (magic 0x" << std::hex
<< f_type
<< std::dec
<< ")"
794 #if defined(__linux__)
795 case BTRFS_SUPER_MAGIC
:
796 if (!m_disable_wbthrottle
){
797 wbthrottle
.set_fs(WBThrottle::BTRFS
);
801 case XFS_SUPER_MAGIC
:
802 // wbthrottle is constructed with fs(WBThrottle::XFS)
807 set_xattr_limits_via_conf();
810 int FileStore::mkfs()
813 char fsid_fn
[PATH_MAX
];
816 uuid_d old_omap_fsid
;
818 dout(1) << "mkfs in " << basedir
<< dendl
;
819 basedir_fd
= ::open(basedir
.c_str(), O_RDONLY
);
820 if (basedir_fd
< 0) {
822 derr
<< __FUNC__
<< ": failed to open base dir " << basedir
<< ": " << cpp_strerror(ret
) << dendl
;
827 snprintf(fsid_fn
, sizeof(fsid_fn
), "%s/fsid", basedir
.c_str());
828 fsid_fd
= ::open(fsid_fn
, O_RDWR
|O_CREAT
, 0644);
831 derr
<< __FUNC__
<< ": failed to open " << fsid_fn
<< ": " << cpp_strerror(ret
) << dendl
;
832 goto close_basedir_fd
;
835 if (lock_fsid() < 0) {
840 if (read_fsid(fsid_fd
, &old_fsid
) < 0 || old_fsid
.is_zero()) {
841 if (fsid
.is_zero()) {
842 fsid
.generate_random();
843 dout(1) << __FUNC__
<< ": generated fsid " << fsid
<< dendl
;
845 dout(1) << __FUNC__
<< ": using provided fsid " << fsid
<< dendl
;
848 fsid
.print(fsid_str
);
849 strcat(fsid_str
, "\n");
850 ret
= ::ftruncate(fsid_fd
, 0);
853 derr
<< __FUNC__
<< ": failed to truncate fsid: "
854 << cpp_strerror(ret
) << dendl
;
857 ret
= safe_write(fsid_fd
, fsid_str
, strlen(fsid_str
));
859 derr
<< __FUNC__
<< ": failed to write fsid: "
860 << cpp_strerror(ret
) << dendl
;
863 if (::fsync(fsid_fd
) < 0) {
865 derr
<< __FUNC__
<< ": close failed: can't write fsid: "
866 << cpp_strerror(ret
) << dendl
;
869 dout(10) << __FUNC__
<< ": fsid is " << fsid
<< dendl
;
871 if (!fsid
.is_zero() && fsid
!= old_fsid
) {
872 derr
<< __FUNC__
<< ": on-disk fsid " << old_fsid
<< " != provided " << fsid
<< dendl
;
877 dout(1) << __FUNC__
<< ": fsid is already set to " << fsid
<< dendl
;
881 ret
= write_version_stamp();
883 derr
<< __FUNC__
<< ": write_version_stamp() failed: "
884 << cpp_strerror(ret
) << dendl
;
889 superblock
.omap_backend
= cct
->_conf
->filestore_omap_backend
;
890 ret
= write_superblock();
892 derr
<< __FUNC__
<< ": write_superblock() failed: "
893 << cpp_strerror(ret
) << dendl
;
897 struct statfs basefs
;
898 ret
= ::fstatfs(basedir_fd
, &basefs
);
901 derr
<< __FUNC__
<< ": cannot fstatfs basedir "
902 << cpp_strerror(ret
) << dendl
;
906 #if defined(__linux__)
907 if (basefs
.f_type
== BTRFS_SUPER_MAGIC
&&
908 !g_ceph_context
->check_experimental_feature_enabled("btrfs")) {
909 derr
<< __FUNC__
<< ": deprecated btrfs support is not enabled" << dendl
;
914 create_backend(basefs
.f_type
);
916 ret
= backend
->create_current();
918 derr
<< __FUNC__
<< ": failed to create current/ " << cpp_strerror(ret
) << dendl
;
922 // write initial op_seq
924 uint64_t initial_seq
= 0;
925 int fd
= read_op_seq(&initial_seq
);
928 derr
<< __FUNC__
<< ": failed to create " << current_op_seq_fn
<< ": "
929 << cpp_strerror(ret
) << dendl
;
932 if (initial_seq
== 0) {
933 ret
= write_op_seq(fd
, 1);
935 VOID_TEMP_FAILURE_RETRY(::close(fd
));
936 derr
<< __FUNC__
<< ": failed to write to " << current_op_seq_fn
<< ": "
937 << cpp_strerror(ret
) << dendl
;
941 if (backend
->can_checkpoint()) {
943 current_fd
= ::open(current_fn
.c_str(), O_RDONLY
);
944 assert(current_fd
>= 0);
946 snprintf(s
, sizeof(s
), COMMIT_SNAP_ITEM
, 1ull);
947 ret
= backend
->create_checkpoint(s
, NULL
);
948 VOID_TEMP_FAILURE_RETRY(::close(current_fd
));
949 if (ret
< 0 && ret
!= -EEXIST
) {
950 VOID_TEMP_FAILURE_RETRY(::close(fd
));
951 derr
<< __FUNC__
<< ": failed to create snap_1: " << cpp_strerror(ret
) << dendl
;
956 VOID_TEMP_FAILURE_RETRY(::close(fd
));
958 ret
= KeyValueDB::test_init(superblock
.omap_backend
, omap_dir
);
960 derr
<< __FUNC__
<< ": failed to create " << cct
->_conf
->filestore_omap_backend
<< dendl
;
963 // create fsid under omap
966 char omap_fsid_fn
[PATH_MAX
];
967 snprintf(omap_fsid_fn
, sizeof(omap_fsid_fn
), "%s/osd_uuid", omap_dir
.c_str());
968 omap_fsid_fd
= ::open(omap_fsid_fn
, O_RDWR
|O_CREAT
, 0644);
969 if (omap_fsid_fd
< 0) {
971 derr
<< __FUNC__
<< ": failed to open " << omap_fsid_fn
<< ": " << cpp_strerror(ret
) << dendl
;
975 if (read_fsid(omap_fsid_fd
, &old_omap_fsid
) < 0 || old_omap_fsid
.is_zero()) {
976 assert(!fsid
.is_zero());
977 fsid
.print(fsid_str
);
978 strcat(fsid_str
, "\n");
979 ret
= ::ftruncate(omap_fsid_fd
, 0);
982 derr
<< __FUNC__
<< ": failed to truncate fsid: "
983 << cpp_strerror(ret
) << dendl
;
984 goto close_omap_fsid_fd
;
986 ret
= safe_write(omap_fsid_fd
, fsid_str
, strlen(fsid_str
));
988 derr
<< __FUNC__
<< ": failed to write fsid: "
989 << cpp_strerror(ret
) << dendl
;
990 goto close_omap_fsid_fd
;
992 dout(10) << __FUNC__
<< ": write success, fsid:" << fsid_str
<< ", ret:" << ret
<< dendl
;
993 if (::fsync(omap_fsid_fd
) < 0) {
995 derr
<< __FUNC__
<< ": close failed: can't write fsid: "
996 << cpp_strerror(ret
) << dendl
;
997 goto close_omap_fsid_fd
;
999 dout(10) << "mkfs omap fsid is " << fsid
<< dendl
;
1001 if (fsid
!= old_omap_fsid
) {
1002 derr
<< __FUNC__
<< ": " << omap_fsid_fn
1003 << " has existed omap fsid " << old_omap_fsid
1004 << " != expected osd fsid " << fsid
1007 goto close_omap_fsid_fd
;
1009 dout(1) << __FUNC__
<< ": omap fsid is already set to " << fsid
<< dendl
;
1012 dout(1) << cct
->_conf
->filestore_omap_backend
<< " db exists/created" << dendl
;
1017 goto close_omap_fsid_fd
;
1019 ret
= write_meta("type", "filestore");
1021 goto close_omap_fsid_fd
;
1023 dout(1) << "mkfs done in " << basedir
<< dendl
;
1027 VOID_TEMP_FAILURE_RETRY(::close(omap_fsid_fd
));
1029 VOID_TEMP_FAILURE_RETRY(::close(fsid_fd
));
1032 VOID_TEMP_FAILURE_RETRY(::close(basedir_fd
));
1038 int FileStore::mkjournal()
1043 snprintf(fn
, sizeof(fn
), "%s/fsid", basedir
.c_str());
1044 int fd
= ::open(fn
, O_RDONLY
, 0644);
1047 derr
<< __FUNC__
<< ": open error: " << cpp_strerror(err
) << dendl
;
1050 ret
= read_fsid(fd
, &fsid
);
1052 derr
<< __FUNC__
<< ": read error: " << cpp_strerror(ret
) << dendl
;
1053 VOID_TEMP_FAILURE_RETRY(::close(fd
));
1056 VOID_TEMP_FAILURE_RETRY(::close(fd
));
1062 ret
= journal
->check();
1064 ret
= journal
->create();
1066 derr
<< __FUNC__
<< ": error creating journal on " << journalpath
1067 << ": " << cpp_strerror(ret
) << dendl
;
1069 dout(0) << __FUNC__
<< ": created journal on " << journalpath
<< dendl
;
1077 int FileStore::read_fsid(int fd
, uuid_d
*uuid
)
1080 memset(fsid_str
, 0, sizeof(fsid_str
));
1081 int ret
= safe_read(fd
, fsid_str
, sizeof(fsid_str
));
1085 // old 64-bit fsid... mirror it.
1086 *(uint64_t*)&uuid
->bytes()[0] = *(uint64_t*)fsid_str
;
1087 *(uint64_t*)&uuid
->bytes()[8] = *(uint64_t*)fsid_str
;
1095 if (!uuid
->parse(fsid_str
))
1100 int FileStore::lock_fsid()
1103 memset(&l
, 0, sizeof(l
));
1105 l
.l_whence
= SEEK_SET
;
1108 int r
= ::fcntl(fsid_fd
, F_SETLK
, &l
);
1111 dout(0) << __FUNC__
<< ": failed to lock " << basedir
<< "/fsid, is another ceph-osd still running? "
1112 << cpp_strerror(err
) << dendl
;
1118 bool FileStore::test_mount_in_use()
1120 dout(5) << __FUNC__
<< ": basedir " << basedir
<< " journal " << journalpath
<< dendl
;
1122 snprintf(fn
, sizeof(fn
), "%s/fsid", basedir
.c_str());
1124 // verify fs isn't in use
1126 fsid_fd
= ::open(fn
, O_RDWR
, 0644);
1128 return 0; // no fsid, ok.
1129 bool inuse
= lock_fsid() < 0;
1130 VOID_TEMP_FAILURE_RETRY(::close(fsid_fd
));
1135 bool FileStore::is_rotational()
1139 rotational
= backend
->is_rotational();
1141 int fd
= ::open(basedir
.c_str(), O_RDONLY
);
1145 int r
= ::fstatfs(fd
, &st
);
1150 create_backend(st
.f_type
);
1151 rotational
= backend
->is_rotational();
1155 dout(10) << __func__
<< " " << (int)rotational
<< dendl
;
1159 bool FileStore::is_journal_rotational()
1161 bool journal_rotational
;
1163 journal_rotational
= backend
->is_journal_rotational();
1165 int fd
= ::open(journalpath
.c_str(), O_RDONLY
);
1169 int r
= ::fstatfs(fd
, &st
);
1174 create_backend(st
.f_type
);
1175 journal_rotational
= backend
->is_journal_rotational();
1179 dout(10) << __func__
<< " " << (int)journal_rotational
<< dendl
;
1180 return journal_rotational
;
1183 int FileStore::_detect_fs()
1186 int r
= ::fstatfs(basedir_fd
, &st
);
1190 blk_size
= st
.f_bsize
;
1192 #if defined(__linux__)
1193 if (st
.f_type
== BTRFS_SUPER_MAGIC
&&
1194 !g_ceph_context
->check_experimental_feature_enabled("btrfs")) {
1195 derr
<<__FUNC__
<< ": deprecated btrfs support is not enabled" << dendl
;
1200 create_backend(st
.f_type
);
1202 r
= backend
->detect_features();
1204 derr
<< __FUNC__
<< ": detect_features error: " << cpp_strerror(r
) << dendl
;
1212 snprintf(fn
, sizeof(fn
), "%s/xattr_test", basedir
.c_str());
1213 int tmpfd
= ::open(fn
, O_CREAT
|O_WRONLY
|O_TRUNC
, 0700);
1216 derr
<< __FUNC__
<< ": unable to create " << fn
<< ": " << cpp_strerror(ret
) << dendl
;
1220 int ret
= chain_fsetxattr(tmpfd
, "user.test", &x
, sizeof(x
));
1222 ret
= chain_fgetxattr(tmpfd
, "user.test", &y
, sizeof(y
));
1223 if ((ret
< 0) || (x
!= y
)) {
1224 derr
<< "Extended attributes don't appear to work. ";
1226 *_dout
<< "Got error " + cpp_strerror(ret
) + ". ";
1227 *_dout
<< "If you are using ext3 or ext4, be sure to mount the underlying "
1228 << "file system with the 'user_xattr' option." << dendl
;
1230 VOID_TEMP_FAILURE_RETRY(::close(tmpfd
));
1235 memset(buf
, 0, sizeof(buf
)); // shut up valgrind
1236 chain_fsetxattr(tmpfd
, "user.test", &buf
, sizeof(buf
));
1237 chain_fsetxattr(tmpfd
, "user.test2", &buf
, sizeof(buf
));
1238 chain_fsetxattr(tmpfd
, "user.test3", &buf
, sizeof(buf
));
1239 chain_fsetxattr(tmpfd
, "user.test4", &buf
, sizeof(buf
));
1240 ret
= chain_fsetxattr(tmpfd
, "user.test5", &buf
, sizeof(buf
));
1241 if (ret
== -ENOSPC
) {
1242 dout(0) << "limited size xattrs" << dendl
;
1244 chain_fremovexattr(tmpfd
, "user.test");
1245 chain_fremovexattr(tmpfd
, "user.test2");
1246 chain_fremovexattr(tmpfd
, "user.test3");
1247 chain_fremovexattr(tmpfd
, "user.test4");
1248 chain_fremovexattr(tmpfd
, "user.test5");
1251 VOID_TEMP_FAILURE_RETRY(::close(tmpfd
));
1256 int FileStore::_sanity_check_fs()
1260 if (((int)m_filestore_journal_writeahead
+
1261 (int)m_filestore_journal_parallel
+
1262 (int)m_filestore_journal_trailing
) > 1) {
1263 dout(0) << "mount ERROR: more than one of filestore journal {writeahead,parallel,trailing} enabled" << dendl
;
1265 << " ** WARNING: more than one of 'filestore journal {writeahead,parallel,trailing}'\n"
1266 << " is enabled in ceph.conf. You must choose a single journal mode."
1267 << TEXT_NORMAL
<< std::endl
;
1271 if (!backend
->can_checkpoint()) {
1272 if (!journal
|| !m_filestore_journal_writeahead
) {
1273 dout(0) << "mount WARNING: no btrfs, and no journal in writeahead mode; data may be lost" << dendl
;
1275 << " ** WARNING: no btrfs AND (no journal OR journal not in writeahead mode)\n"
1276 << " For non-btrfs volumes, a writeahead journal is required to\n"
1277 << " maintain on-disk consistency in the event of a crash. Your conf\n"
1278 << " should include something like:\n"
1279 << " osd journal = /path/to/journal_device_or_file\n"
1280 << " filestore journal writeahead = true\n"
1286 dout(0) << "mount WARNING: no journal" << dendl
;
1288 << " ** WARNING: No osd journal is configured: write latency may be high.\n"
1289 << " If you will not be using an osd journal, write latency may be\n"
1290 << " relatively high. It can be reduced somewhat by lowering\n"
1291 << " filestore_max_sync_interval, but lower values mean lower write\n"
1292 << " throughput, especially with spinning disks.\n"
1299 int FileStore::write_superblock()
1302 ::encode(superblock
, bl
);
1303 return safe_write_file(basedir
.c_str(), "superblock",
1304 bl
.c_str(), bl
.length());
1307 int FileStore::read_superblock()
1309 bufferptr
bp(PATH_MAX
);
1310 int ret
= safe_read_file(basedir
.c_str(), "superblock",
1311 bp
.c_str(), bp
.length());
1313 if (ret
== -ENOENT
) {
1314 // If the file doesn't exist write initial CompatSet
1315 return write_superblock();
1321 bl
.push_back(std::move(bp
));
1322 bufferlist::iterator i
= bl
.begin();
1323 ::decode(superblock
, i
);
1327 int FileStore::update_version_stamp()
1329 return write_version_stamp();
1332 int FileStore::version_stamp_is_valid(uint32_t *version
)
1334 bufferptr
bp(PATH_MAX
);
1335 int ret
= safe_read_file(basedir
.c_str(), "store_version",
1336 bp
.c_str(), bp
.length());
1341 bl
.push_back(std::move(bp
));
1342 bufferlist::iterator i
= bl
.begin();
1343 ::decode(*version
, i
);
1344 dout(10) << __FUNC__
<< ": was " << *version
<< " vs target "
1345 << target_version
<< dendl
;
1346 if (*version
== target_version
)
1352 int FileStore::write_version_stamp()
1354 dout(1) << __FUNC__
<< ": " << target_version
<< dendl
;
1356 ::encode(target_version
, bl
);
1358 return safe_write_file(basedir
.c_str(), "store_version",
1359 bl
.c_str(), bl
.length());
1362 int FileStore::upgrade()
1364 dout(1) << __FUNC__
<< dendl
;
1366 int r
= version_stamp_is_valid(&version
);
1369 derr
<< "The store_version file doesn't exist." << dendl
;
1378 derr
<< "ObjectStore is old at version " << version
<< ". Please upgrade to firefly v0.80.x, convert your store, and then upgrade." << dendl
;
1382 // nothing necessary in FileStore for v3 -> v4 upgrade; we just need to
1383 // open up DBObjectMap with the do_upgrade flag, which we already did.
1384 update_version_stamp();
1388 int FileStore::read_op_seq(uint64_t *seq
)
1390 int op_fd
= ::open(current_op_seq_fn
.c_str(), O_CREAT
|O_RDWR
, 0644);
1393 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
1397 memset(s
, 0, sizeof(s
));
1398 int ret
= safe_read(op_fd
, s
, sizeof(s
) - 1);
1400 derr
<< __FUNC__
<< ": error reading " << current_op_seq_fn
<< ": " << cpp_strerror(ret
) << dendl
;
1401 VOID_TEMP_FAILURE_RETRY(::close(op_fd
));
1402 assert(!m_filestore_fail_eio
|| ret
!= -EIO
);
1409 int FileStore::write_op_seq(int fd
, uint64_t seq
)
1412 snprintf(s
, sizeof(s
), "%" PRId64
"\n", seq
);
1413 int ret
= TEMP_FAILURE_RETRY(::pwrite(fd
, s
, strlen(s
), 0));
1416 assert(!m_filestore_fail_eio
|| ret
!= -EIO
);
1421 int FileStore::mount()
1425 uint64_t initial_op_seq
;
1427 set
<string
> cluster_snaps
;
1428 CompatSet supported_compat_set
= get_fs_supported_compat_set();
1430 dout(5) << "basedir " << basedir
<< " journal " << journalpath
<< dendl
;
1432 ret
= set_throttle_params();
1436 // make sure global base dir exists
1437 if (::access(basedir
.c_str(), R_OK
| W_OK
)) {
1439 derr
<< __FUNC__
<< ": unable to access basedir '" << basedir
<< "': "
1440 << cpp_strerror(ret
) << dendl
;
1445 snprintf(buf
, sizeof(buf
), "%s/fsid", basedir
.c_str());
1446 fsid_fd
= ::open(buf
, O_RDWR
, 0644);
1449 derr
<< __FUNC__
<< ": error opening '" << buf
<< "': "
1450 << cpp_strerror(ret
) << dendl
;
1454 ret
= read_fsid(fsid_fd
, &fsid
);
1456 derr
<< __FUNC__
<< ": error reading fsid_fd: " << cpp_strerror(ret
)
1461 if (lock_fsid() < 0) {
1462 derr
<< __FUNC__
<< ": lock_fsid failed" << dendl
;
1467 dout(10) << "mount fsid is " << fsid
<< dendl
;
1470 uint32_t version_stamp
;
1471 ret
= version_stamp_is_valid(&version_stamp
);
1473 derr
<< __FUNC__
<< ": error in version_stamp_is_valid: "
1474 << cpp_strerror(ret
) << dendl
;
1476 } else if (ret
== 0) {
1477 if (do_update
|| (int)version_stamp
< cct
->_conf
->filestore_update_to
) {
1478 derr
<< __FUNC__
<< ": stale version stamp detected: "
1480 << ". Proceeding, do_update "
1481 << "is set, performing disk format upgrade."
1486 derr
<< __FUNC__
<< ": stale version stamp " << version_stamp
1487 << ". Please run the FileStore update script before starting the "
1488 << "OSD, or set filestore_update_to to " << target_version
1489 << " (currently " << cct
->_conf
->filestore_update_to
<< ")"
1495 ret
= read_superblock();
1500 // Check if this FileStore supports all the necessary features to mount
1501 if (supported_compat_set
.compare(superblock
.compat_features
) == -1) {
1502 derr
<< __FUNC__
<< ": Incompatible features set "
1503 << superblock
.compat_features
<< dendl
;
1508 // open some dir handles
1509 basedir_fd
= ::open(basedir
.c_str(), O_RDONLY
);
1510 if (basedir_fd
< 0) {
1512 derr
<< __FUNC__
<< ": failed to open " << basedir
<< ": "
1513 << cpp_strerror(ret
) << dendl
;
1518 // test for btrfs, xattrs, etc.
1521 derr
<< __FUNC__
<< ": error in _detect_fs: "
1522 << cpp_strerror(ret
) << dendl
;
1523 goto close_basedir_fd
;
1528 ret
= backend
->list_checkpoints(ls
);
1530 derr
<< __FUNC__
<< ": error in _list_snaps: "<< cpp_strerror(ret
) << dendl
;
1531 goto close_basedir_fd
;
1534 long long unsigned c
, prev
= 0;
1535 char clustersnap
[NAME_MAX
];
1536 for (list
<string
>::iterator it
= ls
.begin(); it
!= ls
.end(); ++it
) {
1537 if (sscanf(it
->c_str(), COMMIT_SNAP_ITEM
, &c
) == 1) {
1541 } else if (sscanf(it
->c_str(), CLUSTER_SNAP_ITEM
, clustersnap
) == 1)
1542 cluster_snaps
.insert(*it
);
1546 if (m_osd_rollback_to_cluster_snap
.length() &&
1547 cluster_snaps
.count(m_osd_rollback_to_cluster_snap
) == 0) {
1548 derr
<< "rollback to cluster snapshot '" << m_osd_rollback_to_cluster_snap
<< "': not found" << dendl
;
1550 goto close_basedir_fd
;
1554 snprintf(nosnapfn
, sizeof(nosnapfn
), "%s/nosnap", current_fn
.c_str());
1556 if (backend
->can_checkpoint()) {
1557 if (snaps
.empty()) {
1558 dout(0) << __FUNC__
<< ": WARNING: no consistent snaps found, store may be in inconsistent state" << dendl
;
1561 uint64_t curr_seq
= 0;
1563 if (m_osd_rollback_to_cluster_snap
.length()) {
1565 << " ** NOTE: rolling back to cluster snapshot " << m_osd_rollback_to_cluster_snap
<< " **"
1568 assert(cluster_snaps
.count(m_osd_rollback_to_cluster_snap
));
1569 snprintf(s
, sizeof(s
), CLUSTER_SNAP_ITEM
, m_osd_rollback_to_cluster_snap
.c_str());
1572 int fd
= read_op_seq(&curr_seq
);
1574 VOID_TEMP_FAILURE_RETRY(::close(fd
));
1578 dout(10) << " current/ seq was " << curr_seq
<< dendl
;
1580 dout(10) << " current/ missing entirely (unusual, but okay)" << dendl
;
1582 uint64_t cp
= snaps
.back();
1583 dout(10) << " most recent snap from " << snaps
<< " is " << cp
<< dendl
;
1585 // if current/ is marked as non-snapshotted, refuse to roll
1586 // back (without clear direction) to avoid throwing out new
1589 if (::stat(nosnapfn
, &st
) == 0) {
1590 if (!m_osd_use_stale_snap
) {
1591 derr
<< "ERROR: " << nosnapfn
<< " exists, not rolling back to avoid losing new data" << dendl
;
1592 derr
<< "Force rollback to old snapshotted version with 'osd use stale snap = true'" << dendl
;
1593 derr
<< "config option for --osd-use-stale-snap startup argument." << dendl
;
1595 goto close_basedir_fd
;
1597 derr
<< "WARNING: user forced start with data sequence mismatch: current was " << curr_seq
1598 << ", newest snap is " << cp
<< dendl
;
1600 << " ** WARNING: forcing the use of stale snapshot data **"
1601 << TEXT_NORMAL
<< std::endl
;
1604 dout(10) << __FUNC__
<< ": rolling back to consistent snap " << cp
<< dendl
;
1605 snprintf(s
, sizeof(s
), COMMIT_SNAP_ITEM
, (long long unsigned)cp
);
1609 ret
= backend
->rollback_to(s
);
1611 derr
<< __FUNC__
<< ": error rolling back to " << s
<< ": "
1612 << cpp_strerror(ret
) << dendl
;
1613 goto close_basedir_fd
;
1619 current_fd
= ::open(current_fn
.c_str(), O_RDONLY
);
1620 if (current_fd
< 0) {
1622 derr
<< __FUNC__
<< ": error opening: " << current_fn
<< ": " << cpp_strerror(ret
) << dendl
;
1623 goto close_basedir_fd
;
1626 assert(current_fd
>= 0);
1628 op_fd
= read_op_seq(&initial_op_seq
);
1631 derr
<< __FUNC__
<< ": read_op_seq failed" << dendl
;
1632 goto close_current_fd
;
1635 dout(5) << "mount op_seq is " << initial_op_seq
<< dendl
;
1636 if (initial_op_seq
== 0) {
1637 derr
<< "mount initial op seq is 0; something is wrong" << dendl
;
1639 goto close_current_fd
;
1642 if (!backend
->can_checkpoint()) {
1643 // mark current/ as non-snapshotted so that we don't rollback away
1645 int r
= ::creat(nosnapfn
, 0644);
1648 derr
<< __FUNC__
<< ": failed to create current/nosnap" << dendl
;
1649 goto close_current_fd
;
1651 VOID_TEMP_FAILURE_RETRY(::close(r
));
1653 // clear nosnap marker, if present.
1657 // check fsid with omap
1660 char omap_fsid_buf
[PATH_MAX
];
1661 struct ::stat omap_fsid_stat
;
1662 snprintf(omap_fsid_buf
, sizeof(omap_fsid_buf
), "%s/osd_uuid", omap_dir
.c_str());
1663 // if osd_uuid not exists, assume as this omap matchs corresponding osd
1664 if (::stat(omap_fsid_buf
, &omap_fsid_stat
) != 0){
1665 dout(10) << __FUNC__
<< ": osd_uuid not found under omap, "
1666 << "assume as matched."
1669 // if osd_uuid exists, compares osd_uuid with fsid
1670 omap_fsid_fd
= ::open(omap_fsid_buf
, O_RDONLY
, 0644);
1671 if (omap_fsid_fd
< 0) {
1673 derr
<< __FUNC__
<< ": error opening '" << omap_fsid_buf
<< "': "
1674 << cpp_strerror(ret
)
1676 goto close_current_fd
;
1678 ret
= read_fsid(omap_fsid_fd
, &omap_fsid
);
1679 VOID_TEMP_FAILURE_RETRY(::close(omap_fsid_fd
));
1680 omap_fsid_fd
= -1; // defensive
1682 derr
<< __FUNC__
<< ": error reading omap_fsid_fd"
1683 << ", omap_fsid = " << omap_fsid
1684 << cpp_strerror(ret
)
1686 goto close_current_fd
;
1688 if (fsid
!= omap_fsid
) {
1689 derr
<< __FUNC__
<< ": " << omap_fsid_buf
1690 << " has existed omap fsid " << omap_fsid
1691 << " != expected osd fsid " << fsid
1694 goto close_current_fd
;
1698 dout(0) << "start omap initiation" << dendl
;
1699 if (!(generic_flags
& SKIP_MOUNT_OMAP
)) {
1700 KeyValueDB
* omap_store
= KeyValueDB::create(cct
,
1701 superblock
.omap_backend
,
1703 if (omap_store
== NULL
)
1705 derr
<< __FUNC__
<< ": Error creating " << superblock
.omap_backend
<< dendl
;
1707 goto close_current_fd
;
1710 if (superblock
.omap_backend
== "rocksdb")
1711 ret
= omap_store
->init(cct
->_conf
->filestore_rocksdb_options
);
1713 ret
= omap_store
->init();
1716 derr
<< __FUNC__
<< ": Error initializing omap_store: " << cpp_strerror(ret
) << dendl
;
1717 goto close_current_fd
;
1721 if (omap_store
->create_and_open(err
)) {
1723 derr
<< __FUNC__
<< ": Error initializing " << superblock
.omap_backend
1724 << " : " << err
.str() << dendl
;
1726 goto close_current_fd
;
1729 DBObjectMap
*dbomap
= new DBObjectMap(cct
, omap_store
);
1730 ret
= dbomap
->init(do_update
);
1733 derr
<< __FUNC__
<< ": Error initializing DBObjectMap: " << ret
<< dendl
;
1734 goto close_current_fd
;
1738 if (cct
->_conf
->filestore_debug_omap_check
&& !dbomap
->check(err2
)) {
1739 derr
<< err2
.str() << dendl
;
1742 goto close_current_fd
;
1744 object_map
.reset(dbomap
);
1750 // select journal mode?
1752 if (!m_filestore_journal_writeahead
&&
1753 !m_filestore_journal_parallel
&&
1754 !m_filestore_journal_trailing
) {
1755 if (!backend
->can_checkpoint()) {
1756 m_filestore_journal_writeahead
= true;
1757 dout(0) << __FUNC__
<< ": enabling WRITEAHEAD journal mode: checkpoint is not enabled" << dendl
;
1759 m_filestore_journal_parallel
= true;
1760 dout(0) << __FUNC__
<< ": enabling PARALLEL journal mode: fs, checkpoint is enabled" << dendl
;
1763 if (m_filestore_journal_writeahead
)
1764 dout(0) << __FUNC__
<< ": WRITEAHEAD journal mode explicitly enabled in conf" << dendl
;
1765 if (m_filestore_journal_parallel
)
1766 dout(0) << __FUNC__
<< ": PARALLEL journal mode explicitly enabled in conf" << dendl
;
1767 if (m_filestore_journal_trailing
)
1768 dout(0) << __FUNC__
<< ": TRAILING journal mode explicitly enabled in conf" << dendl
;
1770 if (m_filestore_journal_writeahead
)
1771 journal
->set_wait_on_full(true);
1773 dout(0) << __FUNC__
<< ": no journal" << dendl
;
1776 ret
= _sanity_check_fs();
1778 derr
<< __FUNC__
<< ": _sanity_check_fs failed with error "
1780 goto close_current_fd
;
1783 // Cleanup possibly invalid collections
1785 vector
<coll_t
> collections
;
1786 ret
= list_collections(collections
, true);
1788 derr
<< "Error " << ret
<< " while listing collections" << dendl
;
1789 goto close_current_fd
;
1791 for (vector
<coll_t
>::iterator i
= collections
.begin();
1792 i
!= collections
.end();
1795 ret
= get_index(*i
, &index
);
1797 derr
<< "Unable to mount index " << *i
1798 << " with error: " << ret
<< dendl
;
1799 goto close_current_fd
;
1801 assert(NULL
!= index
.index
);
1802 RWLock::WLocker
l((index
.index
)->access_lock
);
1807 if (!m_disable_wbthrottle
) {
1810 dout(0) << __FUNC__
<< ": INFO: WbThrottle is disabled" << dendl
;
1811 if (cct
->_conf
->filestore_odsync_write
) {
1812 dout(0) << __FUNC__
<< ": INFO: O_DSYNC write is enabled" << dendl
;
1815 sync_thread
.create("filestore_sync");
1817 if (!(generic_flags
& SKIP_JOURNAL_REPLAY
)) {
1818 ret
= journal_replay(initial_op_seq
);
1820 derr
<< __FUNC__
<< ": failed to open journal " << journalpath
<< ": " << cpp_strerror(ret
) << dendl
;
1821 if (ret
== -ENOTTY
) {
1822 derr
<< "maybe journal is not pointing to a block device and its size "
1823 << "wasn't configured?" << dendl
;
1832 if (cct
->_conf
->filestore_debug_omap_check
&& !object_map
->check(err2
)) {
1833 derr
<< err2
.str() << dendl
;
1839 init_temp_collections();
1844 for (vector
<Finisher
*>::iterator it
= ondisk_finishers
.begin(); it
!= ondisk_finishers
.end(); ++it
) {
1847 for (vector
<Finisher
*>::iterator it
= apply_finishers
.begin(); it
!= apply_finishers
.end(); ++it
) {
1854 if (cct
->_conf
->filestore_update_to
>= (int)get_target_version()) {
1855 int err
= upgrade();
1857 derr
<< "error converting store" << dendl
;
1873 if (!m_disable_wbthrottle
) {
1877 VOID_TEMP_FAILURE_RETRY(::close(current_fd
));
1880 VOID_TEMP_FAILURE_RETRY(::close(basedir_fd
));
1883 VOID_TEMP_FAILURE_RETRY(::close(fsid_fd
));
1886 assert(!m_filestore_fail_eio
|| ret
!= -EIO
);
1893 void FileStore::init_temp_collections()
1895 dout(10) << __FUNC__
<< dendl
;
1897 int r
= list_collections(ls
, true);
1900 dout(20) << " ls " << ls
<< dendl
;
1902 SequencerPosition spos
;
1905 for (vector
<coll_t
>::iterator p
= ls
.begin(); p
!= ls
.end(); ++p
)
1908 dout(20) << " temps " << temps
<< dendl
;
1910 for (vector
<coll_t
>::iterator p
= ls
.begin(); p
!= ls
.end(); ++p
) {
1915 coll_t temp
= p
->get_temp();
1916 if (temps
.count(temp
)) {
1919 dout(10) << __FUNC__
<< ": creating " << temp
<< dendl
;
1920 r
= _create_collection(temp
, 0, spos
);
1925 for (set
<coll_t
>::iterator p
= temps
.begin(); p
!= temps
.end(); ++p
) {
1926 dout(10) << __FUNC__
<< ": removing stray " << *p
<< dendl
;
1927 r
= _collection_remove_recursive(*p
, spos
);
1932 int FileStore::umount()
1934 dout(5) << __FUNC__
<< ": " << basedir
<< dendl
;
1945 if (!m_disable_wbthrottle
){
1951 if (!(generic_flags
& SKIP_JOURNAL_REPLAY
))
1952 journal_write_close();
1954 for (vector
<Finisher
*>::iterator it
= ondisk_finishers
.begin(); it
!= ondisk_finishers
.end(); ++it
) {
1957 for (vector
<Finisher
*>::iterator it
= apply_finishers
.begin(); it
!= apply_finishers
.end(); ++it
) {
1962 VOID_TEMP_FAILURE_RETRY(::close(fsid_fd
));
1966 VOID_TEMP_FAILURE_RETRY(::close(op_fd
));
1969 if (current_fd
>= 0) {
1970 VOID_TEMP_FAILURE_RETRY(::close(current_fd
));
1973 if (basedir_fd
>= 0) {
1974 VOID_TEMP_FAILURE_RETRY(::close(basedir_fd
));
1986 Mutex::Locker
l(sync_entry_timeo_lock
);
1997 /// -----------------------------
1999 FileStore::Op
*FileStore::build_op(vector
<Transaction
>& tls
,
2000 Context
*onreadable
,
2001 Context
*onreadable_sync
,
2002 TrackedOpRef osd_op
)
2004 uint64_t bytes
= 0, ops
= 0;
2005 for (vector
<Transaction
>::iterator p
= tls
.begin();
2008 bytes
+= (*p
).get_num_bytes();
2009 ops
+= (*p
).get_num_ops();
2013 o
->start
= ceph_clock_now();
2014 o
->tls
= std::move(tls
);
2015 o
->onreadable
= onreadable
;
2016 o
->onreadable_sync
= onreadable_sync
;
2025 void FileStore::queue_op(OpSequencer
*osr
, Op
*o
)
2027 // queue op on sequencer, then queue sequencer for the threadpool,
2028 // so that regardless of which order the threads pick up the
2029 // sequencer, the op order will be preserved.
2032 o
->trace
.event("queued");
2034 logger
->inc(l_filestore_ops
);
2035 logger
->inc(l_filestore_bytes
, o
->bytes
);
2037 dout(5) << __FUNC__
<< ": " << o
<< " seq " << o
->op
2039 << " " << o
->bytes
<< " bytes"
2040 << " (queue has " << throttle_ops
.get_current() << " ops and " << throttle_bytes
.get_current() << " bytes)"
2045 void FileStore::op_queue_reserve_throttle(Op
*o
)
2048 throttle_bytes
.get(o
->bytes
);
2050 logger
->set(l_filestore_op_queue_ops
, throttle_ops
.get_current());
2051 logger
->set(l_filestore_op_queue_bytes
, throttle_bytes
.get_current());
2054 void FileStore::op_queue_release_throttle(Op
*o
)
2057 throttle_bytes
.put(o
->bytes
);
2058 logger
->set(l_filestore_op_queue_ops
, throttle_ops
.get_current());
2059 logger
->set(l_filestore_op_queue_bytes
, throttle_bytes
.get_current());
2062 void FileStore::_do_op(OpSequencer
*osr
, ThreadPool::TPHandle
&handle
)
2064 if (!m_disable_wbthrottle
) {
2065 wbthrottle
.throttle();
2068 if (cct
->_conf
->filestore_inject_stall
) {
2069 int orig
= cct
->_conf
->filestore_inject_stall
;
2070 dout(5) << __FUNC__
<< ": filestore_inject_stall " << orig
<< ", sleeping" << dendl
;
2072 cct
->_conf
->set_val("filestore_inject_stall", "0");
2073 dout(5) << __FUNC__
<< ": done stalling" << dendl
;
2076 osr
->apply_lock
.Lock();
2077 Op
*o
= osr
->peek_queue();
2078 o
->trace
.event("op_apply_start");
2079 apply_manager
.op_apply_start(o
->op
);
2080 dout(5) << __FUNC__
<< ": " << o
<< " seq " << o
->op
<< " " << *osr
<< "/" << osr
->parent
<< " start" << dendl
;
2081 o
->trace
.event("_do_transactions start");
2082 int r
= _do_transactions(o
->tls
, o
->op
, &handle
);
2083 o
->trace
.event("op_apply_finish");
2084 apply_manager
.op_apply_finish(o
->op
);
2085 dout(10) << __FUNC__
<< ": " << o
<< " seq " << o
->op
<< " r = " << r
2086 << ", finisher " << o
->onreadable
<< " " << o
->onreadable_sync
<< dendl
;
2092 void FileStore::_finish_op(OpSequencer
*osr
)
2094 list
<Context
*> to_queue
;
2095 Op
*o
= osr
->dequeue(&to_queue
);
2097 utime_t lat
= ceph_clock_now();
2100 dout(10) << __FUNC__
<< ": " << o
<< " seq " << o
->op
<< " " << *osr
<< "/" << osr
->parent
<< " lat " << lat
<< dendl
;
2101 osr
->apply_lock
.Unlock(); // locked in _do_op
2102 o
->trace
.event("_finish_op");
2104 // called with tp lock held
2105 op_queue_release_throttle(o
);
2107 logger
->tinc(l_filestore_apply_latency
, lat
);
2109 if (o
->onreadable_sync
) {
2110 o
->onreadable_sync
->complete(0);
2112 if (o
->onreadable
) {
2113 apply_finishers
[osr
->id
% m_apply_finisher_num
]->queue(o
->onreadable
);
2115 if (!to_queue
.empty()) {
2116 apply_finishers
[osr
->id
% m_apply_finisher_num
]->queue(to_queue
);
2122 struct C_JournaledAhead
: public Context
{
2124 FileStore::OpSequencer
*osr
;
2128 C_JournaledAhead(FileStore
*f
, FileStore::OpSequencer
*os
, FileStore::Op
*o
, Context
*ondisk
):
2129 fs(f
), osr(os
), o(o
), ondisk(ondisk
) { }
2130 void finish(int r
) override
{
2131 fs
->_journaled_ahead(osr
, o
, ondisk
);
2135 int FileStore::queue_transactions(Sequencer
*posr
, vector
<Transaction
>& tls
,
2136 TrackedOpRef osd_op
,
2137 ThreadPool::TPHandle
*handle
)
2139 Context
*onreadable
;
2141 Context
*onreadable_sync
;
2142 ObjectStore::Transaction::collect_contexts(
2143 tls
, &onreadable
, &ondisk
, &onreadable_sync
);
2145 if (cct
->_conf
->objectstore_blackhole
) {
2146 dout(0) << __FUNC__
<< ": objectstore_blackhole = TRUE, dropping transaction"
2150 delete onreadable_sync
;
2154 utime_t start
= ceph_clock_now();
2155 // set up the sequencer
2159 osr
= static_cast<OpSequencer
*>(posr
->p
.get());
2160 dout(5) << __FUNC__
<< ": existing " << osr
<< " " << *osr
<< dendl
;
2162 osr
= new OpSequencer(cct
, ++next_osr_id
);
2166 dout(5) << __FUNC__
<< ": new " << osr
<< " " << *osr
<< dendl
;
2169 // used to include osr information in tracepoints during transaction apply
2170 for (vector
<Transaction
>::iterator i
= tls
.begin(); i
!= tls
.end(); ++i
) {
2174 ZTracer::Trace trace
;
2175 if (osd_op
&& osd_op
->pg_trace
) {
2176 osd_op
->store_trace
.init("filestore op", &trace_endpoint
, &osd_op
->pg_trace
);
2177 trace
= osd_op
->store_trace
;
2180 if (journal
&& journal
->is_writeable() && !m_filestore_journal_trailing
) {
2181 Op
*o
= build_op(tls
, onreadable
, onreadable_sync
, osd_op
);
2183 //prepare and encode transactions data out of lock
2185 int orig_len
= journal
->prepare_entry(o
->tls
, &tbl
);
2188 handle
->suspend_tp_timeout();
2190 op_queue_reserve_throttle(o
);
2191 journal
->reserve_throttle_and_backoff(tbl
.length());
2194 handle
->reset_tp_timeout();
2196 uint64_t op_num
= submit_manager
.op_submit_start();
2198 trace
.keyval("opnum", op_num
);
2200 if (m_filestore_do_dump
)
2201 dump_transactions(o
->tls
, o
->op
, osr
);
2203 if (m_filestore_journal_parallel
) {
2204 dout(5) << __FUNC__
<< ": (parallel) " << o
->op
<< " " << o
->tls
<< dendl
;
2206 trace
.keyval("journal mode", "parallel");
2207 trace
.event("journal started");
2208 _op_journal_transactions(tbl
, orig_len
, o
->op
, ondisk
, osd_op
);
2210 // queue inside submit_manager op submission lock
2212 trace
.event("op queued");
2213 } else if (m_filestore_journal_writeahead
) {
2214 dout(5) << __FUNC__
<< ": (writeahead) " << o
->op
<< " " << o
->tls
<< dendl
;
2216 osr
->queue_journal(o
->op
);
2218 trace
.keyval("journal mode", "writeahead");
2219 trace
.event("journal started");
2220 _op_journal_transactions(tbl
, orig_len
, o
->op
,
2221 new C_JournaledAhead(this, osr
, o
, ondisk
),
2226 submit_manager
.op_submit_finish(op_num
);
2227 utime_t end
= ceph_clock_now();
2228 logger
->tinc(l_filestore_queue_transaction_latency_avg
, end
- start
);
2233 Op
*o
= build_op(tls
, onreadable
, onreadable_sync
, osd_op
);
2234 dout(5) << __FUNC__
<< ": (no journal) " << o
<< " " << tls
<< dendl
;
2237 handle
->suspend_tp_timeout();
2239 op_queue_reserve_throttle(o
);
2242 handle
->reset_tp_timeout();
2244 uint64_t op_num
= submit_manager
.op_submit_start();
2247 if (m_filestore_do_dump
)
2248 dump_transactions(o
->tls
, o
->op
, osr
);
2251 trace
.keyval("opnum", op_num
);
2252 trace
.keyval("journal mode", "none");
2253 trace
.event("op queued");
2256 apply_manager
.add_waiter(op_num
, ondisk
);
2257 submit_manager
.op_submit_finish(op_num
);
2258 utime_t end
= ceph_clock_now();
2259 logger
->tinc(l_filestore_queue_transaction_latency_avg
, end
- start
);
2264 //prepare and encode transactions data out of lock
2267 if (journal
->is_writeable()) {
2268 orig_len
= journal
->prepare_entry(tls
, &tbl
);
2270 uint64_t op
= submit_manager
.op_submit_start();
2271 dout(5) << __FUNC__
<< ": (trailing journal) " << op
<< " " << tls
<< dendl
;
2273 if (m_filestore_do_dump
)
2274 dump_transactions(tls
, op
, osr
);
2276 trace
.event("op_apply_start");
2277 trace
.keyval("opnum", op
);
2278 trace
.keyval("journal mode", "trailing");
2279 apply_manager
.op_apply_start(op
);
2280 trace
.event("do_transactions");
2281 int r
= do_transactions(tls
, op
);
2284 trace
.event("journal started");
2285 _op_journal_transactions(tbl
, orig_len
, op
, ondisk
, osd_op
);
2290 // start on_readable finisher after we queue journal item, as on_readable callback
2291 // is allowed to delete the Transaction
2292 if (onreadable_sync
) {
2293 onreadable_sync
->complete(r
);
2295 apply_finishers
[osr
->id
% m_apply_finisher_num
]->queue(onreadable
, r
);
2297 submit_manager
.op_submit_finish(op
);
2298 trace
.event("op_apply_finish");
2299 apply_manager
.op_apply_finish(op
);
2301 utime_t end
= ceph_clock_now();
2302 logger
->tinc(l_filestore_queue_transaction_latency_avg
, end
- start
);
2306 void FileStore::_journaled_ahead(OpSequencer
*osr
, Op
*o
, Context
*ondisk
)
2308 dout(5) << __FUNC__
<< ": " << o
<< " seq " << o
->op
<< " " << *osr
<< " " << o
->tls
<< dendl
;
2310 o
->trace
.event("writeahead journal finished");
2312 // this should queue in order because the journal does it's completions in order.
2315 list
<Context
*> to_queue
;
2316 osr
->dequeue_journal(&to_queue
);
2318 // do ondisk completions async, to prevent any onreadable_sync completions
2319 // getting blocked behind an ondisk completion.
2321 dout(10) << " queueing ondisk " << ondisk
<< dendl
;
2322 ondisk_finishers
[osr
->id
% m_ondisk_finisher_num
]->queue(ondisk
);
2324 if (!to_queue
.empty()) {
2325 ondisk_finishers
[osr
->id
% m_ondisk_finisher_num
]->queue(to_queue
);
2329 int FileStore::_do_transactions(
2330 vector
<Transaction
> &tls
,
2332 ThreadPool::TPHandle
*handle
)
2336 for (vector
<Transaction
>::iterator p
= tls
.begin();
2339 _do_transaction(*p
, op_seq
, trans_num
, handle
);
2341 handle
->reset_tp_timeout();
2347 void FileStore::_set_global_replay_guard(const coll_t
& cid
,
2348 const SequencerPosition
&spos
)
2350 if (backend
->can_checkpoint())
2353 // sync all previous operations on this sequencer
2354 int ret
= object_map
->sync();
2356 derr
<< __FUNC__
<< ": omap sync error " << cpp_strerror(ret
) << dendl
;
2357 assert(0 == "_set_global_replay_guard failed");
2359 ret
= sync_filesystem(basedir_fd
);
2361 derr
<< __FUNC__
<< ": sync_filesystem error " << cpp_strerror(ret
) << dendl
;
2362 assert(0 == "_set_global_replay_guard failed");
2366 get_cdir(cid
, fn
, sizeof(fn
));
2367 int fd
= ::open(fn
, O_RDONLY
);
2370 derr
<< __FUNC__
<< ": " << cid
<< " error " << cpp_strerror(err
) << dendl
;
2371 assert(0 == "_set_global_replay_guard failed");
2376 // then record that we did it
2379 int r
= chain_fsetxattr
<true, true>(
2380 fd
, GLOBAL_REPLAY_GUARD_XATTR
, v
.c_str(), v
.length());
2382 derr
<< __FUNC__
<< ": fsetxattr " << GLOBAL_REPLAY_GUARD_XATTR
2383 << " got " << cpp_strerror(r
) << dendl
;
2384 assert(0 == "fsetxattr failed");
2387 // and make sure our xattr is durable.
2392 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2393 dout(10) << __FUNC__
<< ": " << spos
<< " done" << dendl
;
2396 int FileStore::_check_global_replay_guard(const coll_t
& cid
,
2397 const SequencerPosition
& spos
)
2400 get_cdir(cid
, fn
, sizeof(fn
));
2401 int fd
= ::open(fn
, O_RDONLY
);
2403 dout(10) << __FUNC__
<< ": " << cid
<< " dne" << dendl
;
2404 return 1; // if collection does not exist, there is no guard, and we can replay.
2408 int r
= chain_fgetxattr(fd
, GLOBAL_REPLAY_GUARD_XATTR
, buf
, sizeof(buf
));
2410 dout(20) << __FUNC__
<< ": no xattr" << dendl
;
2411 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
2412 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2413 return 1; // no xattr
2418 SequencerPosition opos
;
2419 bufferlist::iterator p
= bl
.begin();
2422 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2423 return spos
>= opos
? 1 : -1;
2427 void FileStore::_set_replay_guard(const coll_t
& cid
,
2428 const SequencerPosition
&spos
,
2429 bool in_progress
=false)
2432 get_cdir(cid
, fn
, sizeof(fn
));
2433 int fd
= ::open(fn
, O_RDONLY
);
2436 derr
<< __FUNC__
<< ": " << cid
<< " error " << cpp_strerror(err
) << dendl
;
2437 assert(0 == "_set_replay_guard failed");
2439 _set_replay_guard(fd
, spos
, 0, in_progress
);
2440 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2444 void FileStore::_set_replay_guard(int fd
,
2445 const SequencerPosition
& spos
,
2446 const ghobject_t
*hoid
,
2449 if (backend
->can_checkpoint())
2452 dout(10) << __FUNC__
<< ": " << spos
<< (in_progress
? " START" : "") << dendl
;
2456 // first make sure the previous operation commits
2460 // sync object_map too. even if this object has a header or keys,
2461 // it have had them in the past and then removed them, so always
2463 object_map
->sync(hoid
, &spos
);
2468 // then record that we did it
2471 ::encode(in_progress
, v
);
2472 int r
= chain_fsetxattr
<true, true>(
2473 fd
, REPLAY_GUARD_XATTR
, v
.c_str(), v
.length());
2475 derr
<< "fsetxattr " << REPLAY_GUARD_XATTR
<< " got " << cpp_strerror(r
) << dendl
;
2476 assert(0 == "fsetxattr failed");
2479 // and make sure our xattr is durable.
2484 dout(10) << __FUNC__
<< ": " << spos
<< " done" << dendl
;
2487 void FileStore::_close_replay_guard(const coll_t
& cid
,
2488 const SequencerPosition
&spos
)
2491 get_cdir(cid
, fn
, sizeof(fn
));
2492 int fd
= ::open(fn
, O_RDONLY
);
2495 derr
<< __FUNC__
<< ": " << cid
<< " error " << cpp_strerror(err
) << dendl
;
2496 assert(0 == "_close_replay_guard failed");
2498 _close_replay_guard(fd
, spos
);
2499 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2502 void FileStore::_close_replay_guard(int fd
, const SequencerPosition
& spos
,
2503 const ghobject_t
*hoid
)
2505 if (backend
->can_checkpoint())
2508 dout(10) << __FUNC__
<< ": " << spos
<< dendl
;
2512 // sync object_map too. even if this object has a header or keys,
2513 // it have had them in the past and then removed them, so always
2515 object_map
->sync(hoid
, &spos
);
2517 // then record that we are done with this operation
2520 bool in_progress
= false;
2521 ::encode(in_progress
, v
);
2522 int r
= chain_fsetxattr
<true, true>(
2523 fd
, REPLAY_GUARD_XATTR
, v
.c_str(), v
.length());
2525 derr
<< "fsetxattr " << REPLAY_GUARD_XATTR
<< " got " << cpp_strerror(r
) << dendl
;
2526 assert(0 == "fsetxattr failed");
2529 // and make sure our xattr is durable.
2534 dout(10) << __FUNC__
<< ": " << spos
<< " done" << dendl
;
2537 int FileStore::_check_replay_guard(const coll_t
& cid
, const ghobject_t
&oid
,
2538 const SequencerPosition
& spos
)
2540 if (!replaying
|| backend
->can_checkpoint())
2543 int r
= _check_global_replay_guard(cid
, spos
);
2548 r
= lfn_open(cid
, oid
, false, &fd
);
2550 dout(10) << __FUNC__
<< ": " << cid
<< " " << oid
<< " dne" << dendl
;
2551 return 1; // if file does not exist, there is no guard, and we can replay.
2553 int ret
= _check_replay_guard(**fd
, spos
);
2558 int FileStore::_check_replay_guard(const coll_t
& cid
, const SequencerPosition
& spos
)
2560 if (!replaying
|| backend
->can_checkpoint())
2564 get_cdir(cid
, fn
, sizeof(fn
));
2565 int fd
= ::open(fn
, O_RDONLY
);
2567 dout(10) << __FUNC__
<< ": " << cid
<< " dne" << dendl
;
2568 return 1; // if collection does not exist, there is no guard, and we can replay.
2570 int ret
= _check_replay_guard(fd
, spos
);
2571 VOID_TEMP_FAILURE_RETRY(::close(fd
));
2575 int FileStore::_check_replay_guard(int fd
, const SequencerPosition
& spos
)
2577 if (!replaying
|| backend
->can_checkpoint())
2581 int r
= chain_fgetxattr(fd
, REPLAY_GUARD_XATTR
, buf
, sizeof(buf
));
2583 dout(20) << __FUNC__
<< ": no xattr" << dendl
;
2584 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
2585 return 1; // no xattr
2590 SequencerPosition opos
;
2591 bufferlist::iterator p
= bl
.begin();
2593 bool in_progress
= false;
2594 if (!p
.end()) // older journals don't have this
2595 ::decode(in_progress
, p
);
2597 dout(10) << __FUNC__
<< ": object has " << opos
<< " > current pos " << spos
2598 << ", now or in future, SKIPPING REPLAY" << dendl
;
2600 } else if (opos
== spos
) {
2602 dout(10) << __FUNC__
<< ": object has " << opos
<< " == current pos " << spos
2603 << ", in_progress=true, CONDITIONAL REPLAY" << dendl
;
2606 dout(10) << __FUNC__
<< ": object has " << opos
<< " == current pos " << spos
2607 << ", in_progress=false, SKIPPING REPLAY" << dendl
;
2611 dout(10) << __FUNC__
<< ": object has " << opos
<< " < current pos " << spos
2612 << ", in past, will replay" << dendl
;
2617 void FileStore::_do_transaction(
2618 Transaction
& t
, uint64_t op_seq
, int trans_num
,
2619 ThreadPool::TPHandle
*handle
)
2621 dout(10) << __FUNC__
<< ": on " << &t
<< dendl
;
2624 const char *osr_name
= t
.get_osr() ? static_cast<OpSequencer
*>(t
.get_osr())->get_name().c_str() : "<NULL>";
2627 Transaction::iterator i
= t
.begin();
2629 SequencerPosition
spos(op_seq
, trans_num
, 0);
2630 while (i
.have_op()) {
2632 handle
->reset_tp_timeout();
2634 Transaction::Op
*op
= i
.decode_op();
2640 case Transaction::OP_NOP
:
2642 case Transaction::OP_TOUCH
:
2644 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2645 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2646 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2647 _cid
: _cid
.get_temp();
2648 tracepoint(objectstore
, touch_enter
, osr_name
);
2649 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2650 r
= _touch(cid
, oid
);
2651 tracepoint(objectstore
, touch_exit
, r
);
2655 case Transaction::OP_WRITE
:
2657 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2658 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2659 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2660 _cid
: _cid
.get_temp();
2661 uint64_t off
= op
->off
;
2662 uint64_t len
= op
->len
;
2663 uint32_t fadvise_flags
= i
.get_fadvise_flags();
2666 tracepoint(objectstore
, write_enter
, osr_name
, off
, len
);
2667 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2668 r
= _write(cid
, oid
, off
, len
, bl
, fadvise_flags
);
2669 tracepoint(objectstore
, write_exit
, r
);
2673 case Transaction::OP_ZERO
:
2675 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2676 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2677 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2678 _cid
: _cid
.get_temp();
2679 uint64_t off
= op
->off
;
2680 uint64_t len
= op
->len
;
2681 tracepoint(objectstore
, zero_enter
, osr_name
, off
, len
);
2682 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2683 r
= _zero(cid
, oid
, off
, len
);
2684 tracepoint(objectstore
, zero_exit
, r
);
2688 case Transaction::OP_TRIMCACHE
:
2690 // deprecated, no-op
2694 case Transaction::OP_TRUNCATE
:
2696 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2697 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2698 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2699 _cid
: _cid
.get_temp();
2700 uint64_t off
= op
->off
;
2701 tracepoint(objectstore
, truncate_enter
, osr_name
, off
);
2702 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2703 r
= _truncate(cid
, oid
, off
);
2704 tracepoint(objectstore
, truncate_exit
, r
);
2708 case Transaction::OP_REMOVE
:
2710 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2711 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2712 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2713 _cid
: _cid
.get_temp();
2714 tracepoint(objectstore
, remove_enter
, osr_name
);
2715 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2716 r
= _remove(cid
, oid
, spos
);
2717 tracepoint(objectstore
, remove_exit
, r
);
2721 case Transaction::OP_SETATTR
:
2723 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2724 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2725 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2726 _cid
: _cid
.get_temp();
2727 string name
= i
.decode_string();
2730 tracepoint(objectstore
, setattr_enter
, osr_name
);
2731 if (_check_replay_guard(cid
, oid
, spos
) > 0) {
2732 map
<string
, bufferptr
> to_set
;
2733 to_set
[name
] = bufferptr(bl
.c_str(), bl
.length());
2734 r
= _setattrs(cid
, oid
, to_set
, spos
);
2736 dout(0) << " ENOSPC on setxattr on " << cid
<< "/" << oid
2737 << " name " << name
<< " size " << bl
.length() << dendl
;
2739 tracepoint(objectstore
, setattr_exit
, r
);
2743 case Transaction::OP_SETATTRS
:
2745 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2746 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2747 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2748 _cid
: _cid
.get_temp();
2749 map
<string
, bufferptr
> aset
;
2750 i
.decode_attrset(aset
);
2751 tracepoint(objectstore
, setattrs_enter
, osr_name
);
2752 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2753 r
= _setattrs(cid
, oid
, aset
, spos
);
2754 tracepoint(objectstore
, setattrs_exit
, r
);
2756 dout(0) << " ENOSPC on setxattrs on " << cid
<< "/" << oid
<< dendl
;
2760 case Transaction::OP_RMATTR
:
2762 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2763 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2764 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2765 _cid
: _cid
.get_temp();
2766 string name
= i
.decode_string();
2767 tracepoint(objectstore
, rmattr_enter
, osr_name
);
2768 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2769 r
= _rmattr(cid
, oid
, name
.c_str(), spos
);
2770 tracepoint(objectstore
, rmattr_exit
, r
);
2774 case Transaction::OP_RMATTRS
:
2776 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2777 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2778 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2779 _cid
: _cid
.get_temp();
2780 tracepoint(objectstore
, rmattrs_enter
, osr_name
);
2781 if (_check_replay_guard(cid
, oid
, spos
) > 0)
2782 r
= _rmattrs(cid
, oid
, spos
);
2783 tracepoint(objectstore
, rmattrs_exit
, r
);
2787 case Transaction::OP_CLONE
:
2789 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2790 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2791 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2792 _cid
: _cid
.get_temp();
2793 const ghobject_t
&noid
= i
.get_oid(op
->dest_oid
);
2794 tracepoint(objectstore
, clone_enter
, osr_name
);
2795 r
= _clone(cid
, oid
, noid
, spos
);
2796 tracepoint(objectstore
, clone_exit
, r
);
2800 case Transaction::OP_CLONERANGE
:
2802 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2803 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2804 const ghobject_t
&noid
= i
.get_oid(op
->dest_oid
);
2805 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2806 _cid
: _cid
.get_temp();
2807 const coll_t
&ncid
= !_need_temp_object_collection(_cid
, noid
) ?
2808 _cid
: _cid
.get_temp();
2809 uint64_t off
= op
->off
;
2810 uint64_t len
= op
->len
;
2811 tracepoint(objectstore
, clone_range_enter
, osr_name
, len
);
2812 r
= _clone_range(cid
, oid
, ncid
, noid
, off
, len
, off
, spos
);
2813 tracepoint(objectstore
, clone_range_exit
, r
);
2817 case Transaction::OP_CLONERANGE2
:
2819 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2820 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2821 const ghobject_t
&noid
= i
.get_oid(op
->dest_oid
);
2822 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2823 _cid
: _cid
.get_temp();
2824 const coll_t
&ncid
= !_need_temp_object_collection(_cid
, noid
) ?
2825 _cid
: _cid
.get_temp();
2826 uint64_t srcoff
= op
->off
;
2827 uint64_t len
= op
->len
;
2828 uint64_t dstoff
= op
->dest_off
;
2829 tracepoint(objectstore
, clone_range2_enter
, osr_name
, len
);
2830 r
= _clone_range(cid
, oid
, ncid
, noid
, srcoff
, len
, dstoff
, spos
);
2831 tracepoint(objectstore
, clone_range2_exit
, r
);
2835 case Transaction::OP_MKCOLL
:
2837 const coll_t
&cid
= i
.get_cid(op
->cid
);
2838 tracepoint(objectstore
, mkcoll_enter
, osr_name
);
2839 if (_check_replay_guard(cid
, spos
) > 0)
2840 r
= _create_collection(cid
, op
->split_bits
, spos
);
2841 tracepoint(objectstore
, mkcoll_exit
, r
);
2845 case Transaction::OP_COLL_SET_BITS
:
2847 const coll_t
&cid
= i
.get_cid(op
->cid
);
2848 int bits
= op
->split_bits
;
2849 r
= _collection_set_bits(cid
, bits
);
2853 case Transaction::OP_COLL_HINT
:
2855 const coll_t
&cid
= i
.get_cid(op
->cid
);
2856 uint32_t type
= op
->hint_type
;
2859 bufferlist::iterator hiter
= hint
.begin();
2860 if (type
== Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS
) {
2863 ::decode(pg_num
, hiter
);
2864 ::decode(num_objs
, hiter
);
2865 if (_check_replay_guard(cid
, spos
) > 0) {
2866 r
= _collection_hint_expected_num_objs(cid
, pg_num
, num_objs
, spos
);
2870 dout(10) << "Unrecognized collection hint type: " << type
<< dendl
;
2875 case Transaction::OP_RMCOLL
:
2877 const coll_t
&cid
= i
.get_cid(op
->cid
);
2878 tracepoint(objectstore
, rmcoll_enter
, osr_name
);
2879 if (_check_replay_guard(cid
, spos
) > 0)
2880 r
= _destroy_collection(cid
);
2881 tracepoint(objectstore
, rmcoll_exit
, r
);
2885 case Transaction::OP_COLL_ADD
:
2887 const coll_t
&ocid
= i
.get_cid(op
->cid
);
2888 const coll_t
&ncid
= i
.get_cid(op
->dest_cid
);
2889 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2891 assert(oid
.hobj
.pool
>= -1);
2893 // always followed by OP_COLL_REMOVE
2894 Transaction::Op
*op2
= i
.decode_op();
2895 const coll_t
&ocid2
= i
.get_cid(op2
->cid
);
2896 const ghobject_t
&oid2
= i
.get_oid(op2
->oid
);
2897 assert(op2
->op
== Transaction::OP_COLL_REMOVE
);
2898 assert(ocid2
== ocid
);
2899 assert(oid2
== oid
);
2901 tracepoint(objectstore
, coll_add_enter
);
2902 r
= _collection_add(ncid
, ocid
, oid
, spos
);
2903 tracepoint(objectstore
, coll_add_exit
, r
);
2907 tracepoint(objectstore
, coll_remove_enter
, osr_name
);
2908 if (_check_replay_guard(ocid
, oid
, spos
) > 0)
2909 r
= _remove(ocid
, oid
, spos
);
2910 tracepoint(objectstore
, coll_remove_exit
, r
);
2914 case Transaction::OP_COLL_MOVE
:
2916 // WARNING: this is deprecated and buggy; only here to replay old journals.
2917 const coll_t
&ocid
= i
.get_cid(op
->cid
);
2918 const coll_t
&ncid
= i
.get_cid(op
->dest_cid
);
2919 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2920 tracepoint(objectstore
, coll_move_enter
);
2921 r
= _collection_add(ocid
, ncid
, oid
, spos
);
2923 (_check_replay_guard(ocid
, oid
, spos
) > 0))
2924 r
= _remove(ocid
, oid
, spos
);
2925 tracepoint(objectstore
, coll_move_exit
, r
);
2929 case Transaction::OP_COLL_MOVE_RENAME
:
2931 const coll_t
&_oldcid
= i
.get_cid(op
->cid
);
2932 const ghobject_t
&oldoid
= i
.get_oid(op
->oid
);
2933 const coll_t
&_newcid
= i
.get_cid(op
->dest_cid
);
2934 const ghobject_t
&newoid
= i
.get_oid(op
->dest_oid
);
2935 const coll_t
&oldcid
= !_need_temp_object_collection(_oldcid
, oldoid
) ?
2936 _oldcid
: _oldcid
.get_temp();
2937 const coll_t
&newcid
= !_need_temp_object_collection(_newcid
, newoid
) ?
2938 _oldcid
: _newcid
.get_temp();
2939 tracepoint(objectstore
, coll_move_rename_enter
);
2940 r
= _collection_move_rename(oldcid
, oldoid
, newcid
, newoid
, spos
);
2941 tracepoint(objectstore
, coll_move_rename_exit
, r
);
2945 case Transaction::OP_TRY_RENAME
:
2947 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2948 const ghobject_t
&oldoid
= i
.get_oid(op
->oid
);
2949 const ghobject_t
&newoid
= i
.get_oid(op
->dest_oid
);
2950 const coll_t
&oldcid
= !_need_temp_object_collection(_cid
, oldoid
) ?
2951 _cid
: _cid
.get_temp();
2952 const coll_t
&newcid
= !_need_temp_object_collection(_cid
, newoid
) ?
2953 _cid
: _cid
.get_temp();
2954 tracepoint(objectstore
, coll_try_rename_enter
);
2955 r
= _collection_move_rename(oldcid
, oldoid
, newcid
, newoid
, spos
, true);
2956 tracepoint(objectstore
, coll_try_rename_exit
, r
);
2960 case Transaction::OP_COLL_SETATTR
:
2961 case Transaction::OP_COLL_RMATTR
:
2962 assert(0 == "collection attr methods no longer implemented");
2965 case Transaction::OP_STARTSYNC
:
2966 tracepoint(objectstore
, startsync_enter
, osr_name
);
2968 tracepoint(objectstore
, startsync_exit
);
2971 case Transaction::OP_COLL_RENAME
:
2977 case Transaction::OP_OMAP_CLEAR
:
2979 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2980 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2981 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2982 _cid
: _cid
.get_temp();
2983 tracepoint(objectstore
, omap_clear_enter
, osr_name
);
2984 r
= _omap_clear(cid
, oid
, spos
);
2985 tracepoint(objectstore
, omap_clear_exit
, r
);
2988 case Transaction::OP_OMAP_SETKEYS
:
2990 const coll_t
&_cid
= i
.get_cid(op
->cid
);
2991 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
2992 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
2993 _cid
: _cid
.get_temp();
2994 map
<string
, bufferlist
> aset
;
2995 i
.decode_attrset(aset
);
2996 tracepoint(objectstore
, omap_setkeys_enter
, osr_name
);
2997 r
= _omap_setkeys(cid
, oid
, aset
, spos
);
2998 tracepoint(objectstore
, omap_setkeys_exit
, r
);
3001 case Transaction::OP_OMAP_RMKEYS
:
3003 const coll_t
&_cid
= i
.get_cid(op
->cid
);
3004 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
3005 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
3006 _cid
: _cid
.get_temp();
3008 i
.decode_keyset(keys
);
3009 tracepoint(objectstore
, omap_rmkeys_enter
, osr_name
);
3010 r
= _omap_rmkeys(cid
, oid
, keys
, spos
);
3011 tracepoint(objectstore
, omap_rmkeys_exit
, r
);
3014 case Transaction::OP_OMAP_RMKEYRANGE
:
3016 const coll_t
&_cid
= i
.get_cid(op
->cid
);
3017 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
3018 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
3019 _cid
: _cid
.get_temp();
3021 first
= i
.decode_string();
3022 last
= i
.decode_string();
3023 tracepoint(objectstore
, omap_rmkeyrange_enter
, osr_name
);
3024 r
= _omap_rmkeyrange(cid
, oid
, first
, last
, spos
);
3025 tracepoint(objectstore
, omap_rmkeyrange_exit
, r
);
3028 case Transaction::OP_OMAP_SETHEADER
:
3030 const coll_t
&_cid
= i
.get_cid(op
->cid
);
3031 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
3032 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
3033 _cid
: _cid
.get_temp();
3036 tracepoint(objectstore
, omap_setheader_enter
, osr_name
);
3037 r
= _omap_setheader(cid
, oid
, bl
, spos
);
3038 tracepoint(objectstore
, omap_setheader_exit
, r
);
3041 case Transaction::OP_SPLIT_COLLECTION
:
3043 assert(0 == "not legacy journal; upgrade to firefly first");
3046 case Transaction::OP_SPLIT_COLLECTION2
:
3048 coll_t cid
= i
.get_cid(op
->cid
);
3049 uint32_t bits
= op
->split_bits
;
3050 uint32_t rem
= op
->split_rem
;
3051 coll_t dest
= i
.get_cid(op
->dest_cid
);
3052 tracepoint(objectstore
, split_coll2_enter
, osr_name
);
3053 r
= _split_collection(cid
, bits
, rem
, dest
, spos
);
3054 tracepoint(objectstore
, split_coll2_exit
, r
);
3058 case Transaction::OP_SETALLOCHINT
:
3060 const coll_t
&_cid
= i
.get_cid(op
->cid
);
3061 const ghobject_t
&oid
= i
.get_oid(op
->oid
);
3062 const coll_t
&cid
= !_need_temp_object_collection(_cid
, oid
) ?
3063 _cid
: _cid
.get_temp();
3064 uint64_t expected_object_size
= op
->expected_object_size
;
3065 uint64_t expected_write_size
= op
->expected_write_size
;
3066 tracepoint(objectstore
, setallochint_enter
, osr_name
);
3067 if (_check_replay_guard(cid
, oid
, spos
) > 0)
3068 r
= _set_alloc_hint(cid
, oid
, expected_object_size
,
3069 expected_write_size
);
3070 tracepoint(objectstore
, setallochint_exit
, r
);
3075 derr
<< "bad op " << op
->op
<< dendl
;
3082 if (r
== -ENOENT
&& !(op
->op
== Transaction::OP_CLONERANGE
||
3083 op
->op
== Transaction::OP_CLONE
||
3084 op
->op
== Transaction::OP_CLONERANGE2
||
3085 op
->op
== Transaction::OP_COLL_ADD
||
3086 op
->op
== Transaction::OP_SETATTR
||
3087 op
->op
== Transaction::OP_SETATTRS
||
3088 op
->op
== Transaction::OP_RMATTR
||
3089 op
->op
== Transaction::OP_OMAP_SETKEYS
||
3090 op
->op
== Transaction::OP_OMAP_RMKEYS
||
3091 op
->op
== Transaction::OP_OMAP_RMKEYRANGE
||
3092 op
->op
== Transaction::OP_OMAP_SETHEADER
))
3093 // -ENOENT is normally okay
3094 // ...including on a replayed OP_RMCOLL with checkpoint mode
3099 if (op
->op
== Transaction::OP_SETALLOCHINT
)
3100 // Either EOPNOTSUPP or EINVAL most probably. EINVAL in most
3101 // cases means invalid hint size (e.g. too big, not a multiple
3102 // of block size, etc) or, at least on xfs, an attempt to set
3103 // or change it when the file is not empty. However,
3104 // OP_SETALLOCHINT is advisory, so ignore all errors.
3107 if (replaying
&& !backend
->can_checkpoint()) {
3108 if (r
== -EEXIST
&& op
->op
== Transaction::OP_MKCOLL
) {
3109 dout(10) << "tolerating EEXIST during journal replay since checkpoint is not enabled" << dendl
;
3112 if (r
== -EEXIST
&& op
->op
== Transaction::OP_COLL_ADD
) {
3113 dout(10) << "tolerating EEXIST during journal replay since checkpoint is not enabled" << dendl
;
3116 if (r
== -EEXIST
&& op
->op
== Transaction::OP_COLL_MOVE
) {
3117 dout(10) << "tolerating EEXIST during journal replay since checkpoint is not enabled" << dendl
;
3121 dout(10) << "tolerating ERANGE on replay" << dendl
;
3125 dout(10) << "tolerating ENOENT on replay" << dendl
;
3131 const char *msg
= "unexpected error code";
3133 if (r
== -ENOENT
&& (op
->op
== Transaction::OP_CLONERANGE
||
3134 op
->op
== Transaction::OP_CLONE
||
3135 op
->op
== Transaction::OP_CLONERANGE2
)) {
3136 msg
= "ENOENT on clone suggests osd bug";
3137 } else if (r
== -ENOSPC
) {
3138 // For now, if we hit _any_ ENOSPC, crash, before we do any damage
3139 // by partially applying transactions.
3140 msg
= "ENOSPC from disk filesystem, misconfigured cluster";
3141 } else if (r
== -ENOTEMPTY
) {
3142 msg
= "ENOTEMPTY suggests garbage data in osd data dir";
3143 } else if (r
== -EPERM
) {
3144 msg
= "EPERM suggests file(s) in osd data dir not owned by ceph user, or leveldb corruption";
3147 derr
<< " error " << cpp_strerror(r
) << " not handled on operation " << op
3148 << " (" << spos
<< ", or op " << spos
.op
<< ", counting from 0)" << dendl
;
3149 dout(0) << msg
<< dendl
;
3150 dout(0) << " transaction dump:\n";
3151 JSONFormatter
f(true);
3152 f
.open_object_section("transaction");
3162 assert(0 == "unexpected error");
3172 /*********************************************/
3176 // --------------------
3179 bool FileStore::exists(const coll_t
& _cid
, const ghobject_t
& oid
)
3181 tracepoint(objectstore
, exists_enter
, _cid
.c_str());
3182 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
3184 bool retval
= stat(cid
, oid
, &st
) == 0;
3185 tracepoint(objectstore
, exists_exit
, retval
);
3189 int FileStore::stat(
3190 const coll_t
& _cid
, const ghobject_t
& oid
, struct stat
*st
, bool allow_eio
)
3192 tracepoint(objectstore
, stat_enter
, _cid
.c_str());
3193 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
3194 int r
= lfn_stat(cid
, oid
, st
);
3195 assert(allow_eio
|| !m_filestore_fail_eio
|| r
!= -EIO
);
3197 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
3198 << " = " << r
<< dendl
;
3200 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
3202 << " (size " << st
->st_size
<< ")" << dendl
;
3204 if (cct
->_conf
->filestore_debug_inject_read_err
&&
3205 debug_mdata_eio(oid
)) {
3208 tracepoint(objectstore
, stat_exit
, r
);
3213 int FileStore::set_collection_opts(
3215 const pool_opts_t
& opts
)
3220 int FileStore::read(
3222 const ghobject_t
& oid
,
3229 tracepoint(objectstore
, read_enter
, _cid
.c_str(), offset
, len
);
3230 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
3232 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3235 int r
= lfn_open(cid
, oid
, false, &fd
);
3237 dout(10) << __FUNC__
<< ": (" << cid
<< "/" << oid
<< ") open error: "
3238 << cpp_strerror(r
) << dendl
;
3242 if (offset
== 0 && len
== 0) {
3244 memset(&st
, 0, sizeof(struct stat
));
3245 int r
= ::fstat(**fd
, &st
);
3250 #ifdef HAVE_POSIX_FADVISE
3251 if (op_flags
& CEPH_OSD_OP_FLAG_FADVISE_RANDOM
)
3252 posix_fadvise(**fd
, offset
, len
, POSIX_FADV_RANDOM
);
3253 if (op_flags
& CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL
)
3254 posix_fadvise(**fd
, offset
, len
, POSIX_FADV_SEQUENTIAL
);
3257 bufferptr
bptr(len
); // prealloc space for entire read
3258 got
= safe_pread(**fd
, bptr
.c_str(), len
, offset
);
3260 dout(10) << __FUNC__
<< ": (" << cid
<< "/" << oid
<< ") pread error: " << cpp_strerror(got
) << dendl
;
3264 bptr
.set_length(got
); // properly size the buffer
3266 bl
.push_back(std::move(bptr
)); // put it in the target bufferlist
3268 #ifdef HAVE_POSIX_FADVISE
3269 if (op_flags
& CEPH_OSD_OP_FLAG_FADVISE_DONTNEED
)
3270 posix_fadvise(**fd
, offset
, len
, POSIX_FADV_DONTNEED
);
3271 if (op_flags
& (CEPH_OSD_OP_FLAG_FADVISE_RANDOM
| CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL
))
3272 posix_fadvise(**fd
, offset
, len
, POSIX_FADV_NORMAL
);
3275 if (m_filestore_sloppy_crc
&& (!replaying
|| backend
->can_checkpoint())) {
3277 int errors
= backend
->_crc_verify_read(**fd
, offset
, got
, bl
, &ss
);
3279 dout(0) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~"
3280 << got
<< " ... BAD CRC:\n" << ss
.str() << dendl
;
3281 assert(0 == "bad crc on read");
3287 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~"
3288 << got
<< "/" << len
<< dendl
;
3289 if (cct
->_conf
->filestore_debug_inject_read_err
&&
3290 debug_data_eio(oid
)) {
3292 } else if (cct
->_conf
->filestore_debug_random_read_err
&&
3293 (rand() % (int)(cct
->_conf
->filestore_debug_random_read_err
* 100.0)) == 0) {
3294 dout(0) << __func__
<< ": inject random EIO" << dendl
;
3297 tracepoint(objectstore
, read_exit
, got
);
3302 int FileStore::_do_fiemap(int fd
, uint64_t offset
, size_t len
,
3303 map
<uint64_t, uint64_t> *m
)
3306 struct fiemap_extent
*extent
= NULL
;
3307 struct fiemap
*fiemap
= NULL
;
3311 r
= backend
->do_fiemap(fd
, offset
, len
, &fiemap
);
3315 if (fiemap
->fm_mapped_extents
== 0) {
3320 extent
= &fiemap
->fm_extents
[0];
3322 /* start where we were asked to start */
3323 if (extent
->fe_logical
< offset
) {
3324 extent
->fe_length
-= offset
- extent
->fe_logical
;
3325 extent
->fe_logical
= offset
;
3330 struct fiemap_extent
*last
= nullptr;
3331 while (i
< fiemap
->fm_mapped_extents
) {
3332 struct fiemap_extent
*next
= extent
+ 1;
3334 dout(10) << __FUNC__
<< ": fm_mapped_extents=" << fiemap
->fm_mapped_extents
3335 << " fe_logical=" << extent
->fe_logical
<< " fe_length=" << extent
->fe_length
<< dendl
;
3337 /* try to merge extents */
3338 while ((i
< fiemap
->fm_mapped_extents
- 1) &&
3339 (extent
->fe_logical
+ extent
->fe_length
== next
->fe_logical
)) {
3340 next
->fe_length
+= extent
->fe_length
;
3341 next
->fe_logical
= extent
->fe_logical
;
3347 if (extent
->fe_logical
+ extent
->fe_length
> offset
+ len
)
3348 extent
->fe_length
= offset
+ len
- extent
->fe_logical
;
3349 (*m
)[extent
->fe_logical
] = extent
->fe_length
;
3353 uint64_t xoffset
= last
->fe_logical
+ last
->fe_length
- offset
;
3354 offset
= last
->fe_logical
+ last
->fe_length
;
3356 const bool is_last
= (last
->fe_flags
& FIEMAP_EXTENT_LAST
) || (len
== 0);
3365 int FileStore::_do_seek_hole_data(int fd
, uint64_t offset
, size_t len
,
3366 map
<uint64_t, uint64_t> *m
)
3368 #if defined(__linux__) && defined(SEEK_HOLE) && defined(SEEK_DATA)
3369 off_t hole_pos
, data_pos
;
3372 // If lseek fails with errno setting to be ENXIO, this means the current
3373 // file offset is beyond the end of the file.
3374 off_t start
= offset
;
3375 while(start
< (off_t
)(offset
+ len
)) {
3376 data_pos
= lseek(fd
, start
, SEEK_DATA
);
3382 dout(10) << "failed to lseek: " << cpp_strerror(r
) << dendl
;
3385 } else if (data_pos
> (off_t
)(offset
+ len
)) {
3389 hole_pos
= lseek(fd
, data_pos
, SEEK_HOLE
);
3391 if (errno
== ENXIO
) {
3395 dout(10) << "failed to lseek: " << cpp_strerror(r
) << dendl
;
3400 if (hole_pos
>= (off_t
)(offset
+ len
)) {
3401 (*m
)[data_pos
] = offset
+ len
- data_pos
;
3404 (*m
)[data_pos
] = hole_pos
- data_pos
;
3415 int FileStore::fiemap(const coll_t
& _cid
, const ghobject_t
& oid
,
3416 uint64_t offset
, size_t len
,
3419 map
<uint64_t, uint64_t> exomap
;
3420 int r
= fiemap(_cid
, oid
, offset
, len
, exomap
);
3422 ::encode(exomap
, bl
);
3427 int FileStore::fiemap(const coll_t
& _cid
, const ghobject_t
& oid
,
3428 uint64_t offset
, size_t len
,
3429 map
<uint64_t, uint64_t>& destmap
)
3431 tracepoint(objectstore
, fiemap_enter
, _cid
.c_str(), offset
, len
);
3432 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
3435 if ((!backend
->has_seek_data_hole() && !backend
->has_fiemap()) ||
3436 len
<= (size_t)m_filestore_fiemap_threshold
) {
3437 destmap
[offset
] = len
;
3441 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3445 int r
= lfn_open(cid
, oid
, false, &fd
);
3447 dout(10) << "read couldn't open " << cid
<< "/" << oid
<< ": " << cpp_strerror(r
) << dendl
;
3451 if (backend
->has_seek_data_hole()) {
3452 dout(15) << "seek_data/seek_hole " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3453 r
= _do_seek_hole_data(**fd
, offset
, len
, &destmap
);
3454 } else if (backend
->has_fiemap()) {
3455 dout(15) << "fiemap ioctl" << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3456 r
= _do_fiemap(**fd
, offset
, len
, &destmap
);
3463 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< " = " << r
<< " num_extents=" << destmap
.size() << " " << destmap
<< dendl
;
3464 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
3465 tracepoint(objectstore
, fiemap_exit
, r
);
3469 int FileStore::_remove(const coll_t
& cid
, const ghobject_t
& oid
,
3470 const SequencerPosition
&spos
)
3472 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< dendl
;
3473 int r
= lfn_unlink(cid
, oid
, spos
);
3474 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " = " << r
<< dendl
;
3478 int FileStore::_truncate(const coll_t
& cid
, const ghobject_t
& oid
, uint64_t size
)
3480 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " size " << size
<< dendl
;
3481 int r
= lfn_truncate(cid
, oid
, size
);
3482 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " size " << size
<< " = " << r
<< dendl
;
3487 int FileStore::_touch(const coll_t
& cid
, const ghobject_t
& oid
)
3489 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< dendl
;
3492 int r
= lfn_open(cid
, oid
, true, &fd
);
3498 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " = " << r
<< dendl
;
3502 int FileStore::_write(const coll_t
& cid
, const ghobject_t
& oid
,
3503 uint64_t offset
, size_t len
,
3504 const bufferlist
& bl
, uint32_t fadvise_flags
)
3506 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3510 r
= lfn_open(cid
, oid
, true, &fd
);
3512 dout(0) << __FUNC__
<< ": couldn't open " << cid
<< "/"
3514 << cpp_strerror(r
) << dendl
;
3519 r
= bl
.write_fd(**fd
, offset
);
3521 derr
<< __FUNC__
<< ": write_fd on " << cid
<< "/" << oid
3522 << " error: " << cpp_strerror(r
) << dendl
;
3528 if (r
>= 0 && m_filestore_sloppy_crc
) {
3529 int rc
= backend
->_crc_update_write(**fd
, offset
, len
, bl
);
3533 if (replaying
|| m_disable_wbthrottle
) {
3534 if (fadvise_flags
& CEPH_OSD_OP_FLAG_FADVISE_DONTNEED
) {
3535 #ifdef HAVE_POSIX_FADVISE
3536 posix_fadvise(**fd
, 0, 0, POSIX_FADV_DONTNEED
);
3540 wbthrottle
.queue_wb(fd
, oid
, offset
, len
,
3541 fadvise_flags
& CEPH_OSD_OP_FLAG_FADVISE_DONTNEED
);
3547 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< " = " << r
<< dendl
;
3551 int FileStore::_zero(const coll_t
& cid
, const ghobject_t
& oid
, uint64_t offset
, size_t len
)
3553 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< dendl
;
3556 if (cct
->_conf
->filestore_punch_hole
) {
3557 #ifdef CEPH_HAVE_FALLOCATE
3558 # if !defined(DARWIN) && !defined(__FreeBSD__)
3559 # ifdef FALLOC_FL_KEEP_SIZE
3560 // first try to punch a hole.
3562 ret
= lfn_open(cid
, oid
, false, &fd
);
3568 ret
= ::fstat(**fd
, &st
);
3575 // first try fallocate
3576 ret
= fallocate(**fd
, FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
,
3581 // ensure we extend file size, if needed
3582 if (len
> 0 && offset
+ len
> (uint64_t)st
.st_size
) {
3583 ret
= ::ftruncate(**fd
, offset
+ len
);
3593 if (ret
>= 0 && m_filestore_sloppy_crc
) {
3594 int rc
= backend
->_crc_update_zero(**fd
, offset
, len
);
3600 if (ret
!= -EOPNOTSUPP
)
3601 goto out
; // some other error
3607 // lame, kernel is old and doesn't support it.
3608 // write zeros.. yuck!
3609 dout(20) << __FUNC__
<< ": falling back to writing zeros" << dendl
;
3612 bl
.append_zero(len
);
3613 ret
= _write(cid
, oid
, offset
, len
, bl
);
3616 #ifdef CEPH_HAVE_FALLOCATE
3617 # if !defined(DARWIN) && !defined(__FreeBSD__)
3618 # ifdef FALLOC_FL_KEEP_SIZE
3623 dout(20) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " " << offset
<< "~" << len
<< " = " << ret
<< dendl
;
3627 int FileStore::_clone(const coll_t
& cid
, const ghobject_t
& oldoid
, const ghobject_t
& newoid
,
3628 const SequencerPosition
& spos
)
3630 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oldoid
<< " -> " << cid
<< "/" << newoid
<< dendl
;
3632 if (_check_replay_guard(cid
, newoid
, spos
) < 0)
3639 r
= lfn_open(cid
, oldoid
, false, &o
, &index
);
3643 assert(NULL
!= (index
.index
));
3644 RWLock::WLocker
l((index
.index
)->access_lock
);
3646 r
= lfn_open(cid
, newoid
, true, &n
, &index
);
3650 r
= ::ftruncate(**n
, 0);
3656 r
= ::fstat(**o
, &st
);
3662 r
= _do_clone_range(**o
, **n
, 0, st
.st_size
, 0);
3667 dout(20) << "objectmap clone" << dendl
;
3668 r
= object_map
->clone(oldoid
, newoid
, &spos
);
3669 if (r
< 0 && r
!= -ENOENT
)
3675 map
<string
, bufferptr
> aset
;
3676 r
= _fgetattrs(**o
, aset
);
3680 r
= chain_fgetxattr(**o
, XATTR_SPILL_OUT_NAME
, buf
, sizeof(buf
));
3681 if (r
>= 0 && !strncmp(buf
, XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
))) {
3682 r
= chain_fsetxattr
<true, true>(**n
, XATTR_SPILL_OUT_NAME
, XATTR_NO_SPILL_OUT
,
3683 sizeof(XATTR_NO_SPILL_OUT
));
3685 r
= chain_fsetxattr
<true, true>(**n
, XATTR_SPILL_OUT_NAME
, XATTR_SPILL_OUT
,
3686 sizeof(XATTR_SPILL_OUT
));
3691 r
= _fsetattrs(**n
, aset
);
3696 // clone is non-idempotent; record our work.
3697 _set_replay_guard(**n
, spos
, &newoid
);
3704 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oldoid
<< " -> " << cid
<< "/" << newoid
<< " = " << r
<< dendl
;
3705 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
3709 int FileStore::_do_clone_range(int from
, int to
, uint64_t srcoff
, uint64_t len
, uint64_t dstoff
)
3711 dout(20) << __FUNC__
<< ": copy " << srcoff
<< "~" << len
<< " to " << dstoff
<< dendl
;
3712 return backend
->clone_range(from
, to
, srcoff
, len
, dstoff
);
3715 int FileStore::_do_sparse_copy_range(int from
, int to
, uint64_t srcoff
, uint64_t len
, uint64_t dstoff
)
3717 dout(20) << __FUNC__
<< ": " << srcoff
<< "~" << len
<< " to " << dstoff
<< dendl
;
3719 map
<uint64_t, uint64_t> exomap
;
3720 // fiemap doesn't allow zero length
3724 if (backend
->has_seek_data_hole()) {
3725 dout(15) << "seek_data/seek_hole " << from
<< " " << srcoff
<< "~" << len
<< dendl
;
3726 r
= _do_seek_hole_data(from
, srcoff
, len
, &exomap
);
3727 } else if (backend
->has_fiemap()) {
3728 dout(15) << "fiemap ioctl" << from
<< " " << srcoff
<< "~" << len
<< dendl
;
3729 r
= _do_fiemap(from
, srcoff
, len
, &exomap
);
3733 int64_t written
= 0;
3737 for (map
<uint64_t, uint64_t>::iterator miter
= exomap
.begin(); miter
!= exomap
.end(); ++miter
) {
3738 uint64_t it_off
= miter
->first
- srcoff
+ dstoff
;
3739 r
= _do_copy_range(from
, to
, miter
->first
, miter
->second
, it_off
, true);
3741 derr
<< __FUNC__
<< ": copy error at " << miter
->first
<< "~" << miter
->second
3742 << " to " << it_off
<< ", " << cpp_strerror(r
) << dendl
;
3745 written
+= miter
->second
;
3749 if (m_filestore_sloppy_crc
) {
3750 int rc
= backend
->_crc_update_clone_range(from
, to
, srcoff
, len
, dstoff
);
3754 r
= ::fstat(to
, &st
);
3757 derr
<< __FUNC__
<< ": fstat error at " << to
<< " " << cpp_strerror(r
) << dendl
;
3760 if (st
.st_size
< (int)(dstoff
+ len
)) {
3761 r
= ::ftruncate(to
, dstoff
+ len
);
3764 derr
<< __FUNC__
<< ": ftruncate error at " << dstoff
+len
<< " " << cpp_strerror(r
) << dendl
;
3772 dout(20) << __FUNC__
<< ": " << srcoff
<< "~" << len
<< " to " << dstoff
<< " = " << r
<< dendl
;
3776 int FileStore::_do_copy_range(int from
, int to
, uint64_t srcoff
, uint64_t len
, uint64_t dstoff
, bool skip_sloppycrc
)
3778 dout(20) << __FUNC__
<< ": " << srcoff
<< "~" << len
<< " to " << dstoff
<< dendl
;
3780 loff_t pos
= srcoff
;
3781 loff_t end
= srcoff
+ len
;
3782 int buflen
= 4096 * 16; //limit by pipe max size.see fcntl
3784 #ifdef CEPH_HAVE_SPLICE
3785 if (backend
->has_splice()) {
3787 if (pipe(pipefd
) < 0) {
3789 derr
<< " pipe " << " got " << cpp_strerror(r
) << dendl
;
3793 loff_t dstpos
= dstoff
;
3795 int l
= MIN(end
-pos
, buflen
);
3796 r
= safe_splice(from
, &pos
, pipefd
[1], NULL
, l
, SPLICE_F_NONBLOCK
);
3797 dout(10) << " safe_splice read from " << pos
<< "~" << l
<< " got " << r
<< dendl
;
3799 derr
<< __FUNC__
<< ": safe_splice read error at " << pos
<< "~" << len
3800 << ", " << cpp_strerror(r
) << dendl
;
3804 // hrm, bad source range, wtf.
3806 derr
<< __FUNC__
<< ": got short read result at " << pos
3807 << " of fd " << from
<< " len " << len
<< dendl
;
3811 r
= safe_splice(pipefd
[0], NULL
, to
, &dstpos
, r
, 0);
3812 dout(10) << " safe_splice write to " << to
<< " len " << r
3813 << " got " << r
<< dendl
;
3815 derr
<< __FUNC__
<< ": write error at " << pos
<< "~"
3816 << r
<< ", " << cpp_strerror(r
) << dendl
;
3827 actual
= ::lseek64(from
, srcoff
, SEEK_SET
);
3828 if (actual
!= (int64_t)srcoff
) {
3833 derr
<< "lseek64 to " << srcoff
<< " got " << cpp_strerror(r
) << dendl
;
3836 actual
= ::lseek64(to
, dstoff
, SEEK_SET
);
3837 if (actual
!= (int64_t)dstoff
) {
3842 derr
<< "lseek64 to " << dstoff
<< " got " << cpp_strerror(r
) << dendl
;
3848 int l
= MIN(end
-pos
, buflen
);
3849 r
= ::read(from
, buf
, l
);
3850 dout(25) << " read from " << pos
<< "~" << l
<< " got " << r
<< dendl
;
3852 if (errno
== EINTR
) {
3856 derr
<< __FUNC__
<< ": read error at " << pos
<< "~" << len
3857 << ", " << cpp_strerror(r
) << dendl
;
3862 // hrm, bad source range, wtf.
3864 derr
<< __FUNC__
<< ": got short read result at " << pos
3865 << " of fd " << from
<< " len " << len
<< dendl
;
3870 int r2
= safe_write(to
, buf
+op
, r
-op
);
3871 dout(25) << " write to " << to
<< " len " << (r
-op
)
3872 << " got " << r2
<< dendl
;
3875 derr
<< __FUNC__
<< ": write error at " << pos
<< "~"
3876 << r
-op
<< ", " << cpp_strerror(r
) << dendl
;
3888 if (r
< 0 && replaying
) {
3889 assert(r
== -ERANGE
);
3890 derr
<< __FUNC__
<< ": short source tolerated because we are replaying" << dendl
;
3893 assert(replaying
|| pos
== end
);
3894 if (r
>= 0 && !skip_sloppycrc
&& m_filestore_sloppy_crc
) {
3895 int rc
= backend
->_crc_update_clone_range(from
, to
, srcoff
, len
, dstoff
);
3898 dout(20) << __FUNC__
<< ": " << srcoff
<< "~" << len
<< " to " << dstoff
<< " = " << r
<< dendl
;
3902 int FileStore::_clone_range(const coll_t
& oldcid
, const ghobject_t
& oldoid
, const coll_t
& newcid
, const ghobject_t
& newoid
,
3903 uint64_t srcoff
, uint64_t len
, uint64_t dstoff
,
3904 const SequencerPosition
& spos
)
3906 dout(15) << __FUNC__
<< ": " << oldcid
<< "/" << oldoid
<< " -> " << newcid
<< "/" << newoid
<< " " << srcoff
<< "~" << len
<< " to " << dstoff
<< dendl
;
3908 if (_check_replay_guard(newcid
, newoid
, spos
) < 0)
3913 r
= lfn_open(oldcid
, oldoid
, false, &o
);
3917 r
= lfn_open(newcid
, newoid
, true, &n
);
3921 r
= _do_clone_range(**o
, **n
, srcoff
, len
, dstoff
);
3926 // clone is non-idempotent; record our work.
3927 _set_replay_guard(**n
, spos
, &newoid
);
3934 dout(10) << __FUNC__
<< ": " << oldcid
<< "/" << oldoid
<< " -> " << newcid
<< "/" << newoid
<< " "
3935 << srcoff
<< "~" << len
<< " to " << dstoff
<< " = " << r
<< dendl
;
3939 class SyncEntryTimeout
: public Context
{
3942 explicit SyncEntryTimeout(CephContext
* cct
, int commit_timeo
)
3943 : cct(cct
), m_commit_timeo(commit_timeo
)
3947 void finish(int r
) override
{
3948 BackTrace
*bt
= new BackTrace(1);
3949 generic_dout(-1) << "FileStore: sync_entry timed out after "
3950 << m_commit_timeo
<< " seconds.\n";
3960 void FileStore::sync_entry()
3964 utime_t max_interval
;
3965 max_interval
.set_from_double(m_filestore_max_sync_interval
);
3966 utime_t min_interval
;
3967 min_interval
.set_from_double(m_filestore_min_sync_interval
);
3969 utime_t startwait
= ceph_clock_now();
3971 dout(20) << __FUNC__
<< ": waiting for max_interval " << max_interval
<< dendl
;
3972 sync_cond
.WaitInterval(lock
, max_interval
);
3974 dout(20) << __FUNC__
<< ": not waiting, force_sync set" << dendl
;
3978 dout(20) << __FUNC__
<< ": force_sync set" << dendl
;
3981 dout(20) << __FUNC__
<< ": stop set" << dendl
;
3984 // wait for at least the min interval
3985 utime_t woke
= ceph_clock_now();
3987 dout(20) << __FUNC__
<< ": woke after " << woke
<< dendl
;
3988 if (woke
< min_interval
) {
3989 utime_t t
= min_interval
;
3991 dout(20) << __FUNC__
<< ": waiting for another " << t
3992 << " to reach min interval " << min_interval
<< dendl
;
3993 sync_cond
.WaitInterval(lock
, t
);
3999 fin
.swap(sync_waiters
);
4003 if (apply_manager
.commit_start()) {
4004 utime_t start
= ceph_clock_now();
4005 uint64_t cp
= apply_manager
.get_committing_seq();
4007 sync_entry_timeo_lock
.Lock();
4008 SyncEntryTimeout
*sync_entry_timeo
=
4009 new SyncEntryTimeout(cct
, m_filestore_commit_timeout
);
4010 if (!timer
.add_event_after(m_filestore_commit_timeout
,
4011 sync_entry_timeo
)) {
4012 sync_entry_timeo
= nullptr;
4014 sync_entry_timeo_lock
.Unlock();
4016 logger
->set(l_filestore_committing
, 1);
4018 dout(15) << __FUNC__
<< ": committing " << cp
<< dendl
;
4019 stringstream errstream
;
4020 if (cct
->_conf
->filestore_debug_omap_check
&& !object_map
->check(errstream
)) {
4021 derr
<< errstream
.str() << dendl
;
4025 if (backend
->can_checkpoint()) {
4026 int err
= write_op_seq(op_fd
, cp
);
4028 derr
<< "Error during write_op_seq: " << cpp_strerror(err
) << dendl
;
4029 assert(0 == "error during write_op_seq");
4033 snprintf(s
, sizeof(s
), COMMIT_SNAP_ITEM
, (long long unsigned)cp
);
4035 err
= backend
->create_checkpoint(s
, &cid
);
4038 derr
<< "snap create '" << s
<< "' got error " << err
<< dendl
;
4042 snaps
.push_back(cp
);
4043 apply_manager
.commit_started();
4047 dout(20) << " waiting for checkpoint " << cid
<< " to complete" << dendl
;
4048 err
= backend
->sync_checkpoint(cid
);
4050 derr
<< "ioctl WAIT_SYNC got " << cpp_strerror(err
) << dendl
;
4051 assert(0 == "wait_sync got error");
4053 dout(20) << " done waiting for checkpoint " << cid
<< " to complete" << dendl
;
4056 apply_manager
.commit_started();
4059 int err
= object_map
->sync();
4061 derr
<< "object_map sync got " << cpp_strerror(err
) << dendl
;
4062 assert(0 == "object_map sync returned error");
4065 err
= backend
->syncfs();
4067 derr
<< "syncfs got " << cpp_strerror(err
) << dendl
;
4068 assert(0 == "syncfs returned error");
4071 err
= write_op_seq(op_fd
, cp
);
4073 derr
<< "Error during write_op_seq: " << cpp_strerror(err
) << dendl
;
4074 assert(0 == "error during write_op_seq");
4076 err
= ::fsync(op_fd
);
4078 derr
<< "Error during fsync of op_seq: " << cpp_strerror(err
) << dendl
;
4079 assert(0 == "error during fsync of op_seq");
4083 utime_t done
= ceph_clock_now();
4084 utime_t lat
= done
- start
;
4085 utime_t dur
= done
- startwait
;
4086 dout(10) << __FUNC__
<< ": commit took " << lat
<< ", interval was " << dur
<< dendl
;
4087 utime_t max_pause_lat
= logger
->tget(l_filestore_sync_pause_max_lat
);
4088 if (max_pause_lat
< dur
- lat
) {
4089 logger
->tinc(l_filestore_sync_pause_max_lat
, dur
- lat
);
4092 logger
->inc(l_filestore_commitcycle
);
4093 logger
->tinc(l_filestore_commitcycle_latency
, lat
);
4094 logger
->tinc(l_filestore_commitcycle_interval
, dur
);
4096 apply_manager
.commit_finish();
4097 if (!m_disable_wbthrottle
) {
4101 logger
->set(l_filestore_committing
, 0);
4103 // remove old snaps?
4104 if (backend
->can_checkpoint()) {
4106 while (snaps
.size() > 2) {
4107 snprintf(s
, sizeof(s
), COMMIT_SNAP_ITEM
, (long long unsigned)snaps
.front());
4109 dout(10) << "removing snap '" << s
<< "'" << dendl
;
4110 int r
= backend
->destroy_checkpoint(s
);
4113 derr
<< "unable to destroy snap '" << s
<< "' got " << cpp_strerror(err
) << dendl
;
4118 dout(15) << __FUNC__
<< ": committed to op_seq " << cp
<< dendl
;
4120 if (sync_entry_timeo
) {
4121 Mutex::Locker
lock(sync_entry_timeo_lock
);
4122 timer
.cancel_event(sync_entry_timeo
);
4129 finish_contexts(cct
, fin
, 0);
4131 if (!sync_waiters
.empty()) {
4132 dout(10) << __FUNC__
<< ": more waiters, committing again" << dendl
;
4135 if (!stop
&& journal
&& journal
->should_commit_now()) {
4136 dout(10) << __FUNC__
<< ": journal says we should commit again (probably is/was full)" << dendl
;
4144 void FileStore::_start_sync()
4146 if (!journal
) { // don't do a big sync if the journal is on
4147 dout(10) << __FUNC__
<< dendl
;
4150 dout(10) << __FUNC__
<< ": - NOOP (journal is on)" << dendl
;
4154 void FileStore::do_force_sync()
4156 dout(10) << __FUNC__
<< dendl
;
4157 Mutex::Locker
l(lock
);
4162 void FileStore::start_sync(Context
*onsafe
)
4164 Mutex::Locker
l(lock
);
4165 sync_waiters
.push_back(onsafe
);
4168 dout(10) << __FUNC__
<< dendl
;
4171 void FileStore::sync()
4173 Mutex
l("FileStore::sync");
4176 C_SafeCond
*fin
= new C_SafeCond(&l
, &c
, &done
);
4182 dout(10) << "sync waiting" << dendl
;
4186 dout(10) << "sync done" << dendl
;
4189 void FileStore::_flush_op_queue()
4191 dout(10) << __FUNC__
<< ": draining op tp" << dendl
;
4193 dout(10) << __FUNC__
<< ": waiting for apply finisher" << dendl
;
4194 for (vector
<Finisher
*>::iterator it
= apply_finishers
.begin(); it
!= apply_finishers
.end(); ++it
) {
4195 (*it
)->wait_for_empty();
4200 * flush - make every queued write readable
4202 void FileStore::flush()
4204 dout(10) << __FUNC__
<< dendl
;
4206 if (cct
->_conf
->filestore_blackhole
) {
4208 Mutex
lock("FileStore::flush::lock");
4216 if (m_filestore_journal_writeahead
) {
4219 dout(10) << __FUNC__
<< ": draining ondisk finisher" << dendl
;
4220 for (vector
<Finisher
*>::iterator it
= ondisk_finishers
.begin(); it
!= ondisk_finishers
.end(); ++it
) {
4221 (*it
)->wait_for_empty();
4226 dout(10) << __FUNC__
<< ": complete" << dendl
;
4230 * sync_and_flush - make every queued write readable AND committed to disk
4232 void FileStore::sync_and_flush()
4234 dout(10) << __FUNC__
<< dendl
;
4236 if (m_filestore_journal_writeahead
) {
4241 // includes m_filestore_journal_parallel
4245 dout(10) << __FUNC__
<< ": done" << dendl
;
4248 int FileStore::flush_journal()
4250 dout(10) << __FUNC__
<< dendl
;
4256 int FileStore::snapshot(const string
& name
)
4258 dout(10) << __FUNC__
<< ": " << name
<< dendl
;
4261 if (!backend
->can_checkpoint()) {
4262 dout(0) << __FUNC__
<< ": " << name
<< " failed, not supported" << dendl
;
4267 snprintf(s
, sizeof(s
), CLUSTER_SNAP_ITEM
, name
.c_str());
4269 int r
= backend
->create_checkpoint(s
, NULL
);
4271 derr
<< __FUNC__
<< ": " << name
<< " failed: " << cpp_strerror(r
) << dendl
;
4277 // -------------------------------
4280 int FileStore::_fgetattr(int fd
, const char *name
, bufferptr
& bp
)
4282 char val
[CHAIN_XATTR_MAX_BLOCK_LEN
];
4283 int l
= chain_fgetxattr(fd
, name
, val
, sizeof(val
));
4285 bp
= buffer::create(l
);
4286 memcpy(bp
.c_str(), val
, l
);
4287 } else if (l
== -ERANGE
) {
4288 l
= chain_fgetxattr(fd
, name
, 0, 0);
4290 bp
= buffer::create(l
);
4291 l
= chain_fgetxattr(fd
, name
, bp
.c_str(), l
);
4294 assert(!m_filestore_fail_eio
|| l
!= -EIO
);
4298 int FileStore::_fgetattrs(int fd
, map
<string
,bufferptr
>& aset
)
4302 int len
= chain_flistxattr(fd
, names1
, sizeof(names1
)-1);
4305 if (len
== -ERANGE
) {
4306 len
= chain_flistxattr(fd
, 0, 0);
4308 assert(!m_filestore_fail_eio
|| len
!= -EIO
);
4311 dout(10) << " -ERANGE, len is " << len
<< dendl
;
4312 names2
= new char[len
+1];
4313 len
= chain_flistxattr(fd
, names2
, len
);
4314 dout(10) << " -ERANGE, got " << len
<< dendl
;
4316 assert(!m_filestore_fail_eio
|| len
!= -EIO
);
4321 } else if (len
< 0) {
4322 assert(!m_filestore_fail_eio
|| len
!= -EIO
);
4329 char *end
= name
+ len
;
4330 while (name
< end
) {
4331 char *attrname
= name
;
4332 if (parse_attrname(&name
)) {
4334 dout(20) << __FUNC__
<< ": " << fd
<< " getting '" << name
<< "'" << dendl
;
4335 int r
= _fgetattr(fd
, attrname
, aset
[name
]);
4342 name
+= strlen(name
) + 1;
4349 int FileStore::_fsetattrs(int fd
, map
<string
, bufferptr
> &aset
)
4351 for (map
<string
, bufferptr
>::iterator p
= aset
.begin();
4354 char n
[CHAIN_XATTR_MAX_NAME_LEN
];
4355 get_attrname(p
->first
.c_str(), n
, CHAIN_XATTR_MAX_NAME_LEN
);
4357 if (p
->second
.length())
4358 val
= p
->second
.c_str();
4361 // ??? Why do we skip setting all the other attrs if one fails?
4362 int r
= chain_fsetxattr(fd
, n
, val
, p
->second
.length());
4364 derr
<< __FUNC__
<< ": chain_setxattr returned " << r
<< dendl
;
4371 // debug EIO injection
4372 void FileStore::inject_data_error(const ghobject_t
&oid
) {
4373 Mutex::Locker
l(read_error_lock
);
4374 dout(10) << __FUNC__
<< ": init error on " << oid
<< dendl
;
4375 data_error_set
.insert(oid
);
4377 void FileStore::inject_mdata_error(const ghobject_t
&oid
) {
4378 Mutex::Locker
l(read_error_lock
);
4379 dout(10) << __FUNC__
<< ": init error on " << oid
<< dendl
;
4380 mdata_error_set
.insert(oid
);
4383 void FileStore::debug_obj_on_delete(const ghobject_t
&oid
) {
4384 Mutex::Locker
l(read_error_lock
);
4385 dout(10) << __FUNC__
<< ": clear error on " << oid
<< dendl
;
4386 data_error_set
.erase(oid
);
4387 mdata_error_set
.erase(oid
);
4389 bool FileStore::debug_data_eio(const ghobject_t
&oid
) {
4390 Mutex::Locker
l(read_error_lock
);
4391 if (data_error_set
.count(oid
)) {
4392 dout(10) << __FUNC__
<< ": inject error on " << oid
<< dendl
;
4398 bool FileStore::debug_mdata_eio(const ghobject_t
&oid
) {
4399 Mutex::Locker
l(read_error_lock
);
4400 if (mdata_error_set
.count(oid
)) {
4401 dout(10) << __FUNC__
<< ": inject error on " << oid
<< dendl
;
4411 int FileStore::getattr(const coll_t
& _cid
, const ghobject_t
& oid
, const char *name
, bufferptr
&bp
)
4413 tracepoint(objectstore
, getattr_enter
, _cid
.c_str());
4414 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
4415 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " '" << name
<< "'" << dendl
;
4417 int r
= lfn_open(cid
, oid
, false, &fd
);
4421 char n
[CHAIN_XATTR_MAX_NAME_LEN
];
4422 get_attrname(name
, n
, CHAIN_XATTR_MAX_NAME_LEN
);
4423 r
= _fgetattr(**fd
, n
, bp
);
4425 if (r
== -ENODATA
) {
4426 map
<string
, bufferlist
> got
;
4428 to_get
.insert(string(name
));
4430 r
= get_index(cid
, &index
);
4432 dout(10) << __FUNC__
<< ": could not get index r = " << r
<< dendl
;
4435 r
= object_map
->get_xattrs(oid
, to_get
, &got
);
4436 if (r
< 0 && r
!= -ENOENT
) {
4437 dout(10) << __FUNC__
<< ": get_xattrs err r =" << r
<< dendl
;
4441 dout(10) << __FUNC__
<< ": got.size() is 0" << dendl
;
4444 bp
= bufferptr(got
.begin()->second
.c_str(),
4445 got
.begin()->second
.length());
4449 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " '" << name
<< "' = " << r
<< dendl
;
4450 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4451 if (cct
->_conf
->filestore_debug_inject_read_err
&&
4452 debug_mdata_eio(oid
)) {
4455 tracepoint(objectstore
, getattr_exit
, r
);
4456 return r
< 0 ? r
: 0;
4460 int FileStore::getattrs(const coll_t
& _cid
, const ghobject_t
& oid
, map
<string
,bufferptr
>& aset
)
4462 tracepoint(objectstore
, getattrs_enter
, _cid
.c_str());
4463 const coll_t
& cid
= !_need_temp_object_collection(_cid
, oid
) ? _cid
: _cid
.get_temp();
4464 set
<string
> omap_attrs
;
4465 map
<string
, bufferlist
> omap_aset
;
4467 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< dendl
;
4469 bool spill_out
= true;
4472 int r
= lfn_open(cid
, oid
, false, &fd
);
4477 r
= chain_fgetxattr(**fd
, XATTR_SPILL_OUT_NAME
, buf
, sizeof(buf
));
4478 if (r
>= 0 && !strncmp(buf
, XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
)))
4481 r
= _fgetattrs(**fd
, aset
);
4483 fd
= FDRef(); // defensive
4489 dout(10) << __FUNC__
<< ": no xattr exists in object_map r = " << r
<< dendl
;
4493 r
= get_index(cid
, &index
);
4495 dout(10) << __FUNC__
<< ": could not get index r = " << r
<< dendl
;
4499 r
= object_map
->get_all_xattrs(oid
, &omap_attrs
);
4500 if (r
< 0 && r
!= -ENOENT
) {
4501 dout(10) << __FUNC__
<< ": could not get omap_attrs r = " << r
<< dendl
;
4505 r
= object_map
->get_xattrs(oid
, omap_attrs
, &omap_aset
);
4506 if (r
< 0 && r
!= -ENOENT
) {
4507 dout(10) << __FUNC__
<< ": could not get omap_attrs r = " << r
<< dendl
;
4513 assert(omap_attrs
.size() == omap_aset
.size());
4514 for (map
<string
, bufferlist
>::iterator i
= omap_aset
.begin();
4515 i
!= omap_aset
.end();
4517 string
key(i
->first
);
4518 aset
.insert(make_pair(key
,
4519 bufferptr(i
->second
.c_str(), i
->second
.length())));
4522 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " = " << r
<< dendl
;
4523 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4525 if (cct
->_conf
->filestore_debug_inject_read_err
&&
4526 debug_mdata_eio(oid
)) {
4529 tracepoint(objectstore
, getattrs_exit
, r
);
4534 int FileStore::_setattrs(const coll_t
& cid
, const ghobject_t
& oid
, map
<string
,bufferptr
>& aset
,
4535 const SequencerPosition
&spos
)
4537 map
<string
, bufferlist
> omap_set
;
4538 set
<string
> omap_remove
;
4539 map
<string
, bufferptr
> inline_set
;
4540 map
<string
, bufferptr
> inline_to_set
;
4543 bool incomplete_inline
= false;
4545 int r
= lfn_open(cid
, oid
, false, &fd
);
4551 r
= chain_fgetxattr(**fd
, XATTR_SPILL_OUT_NAME
, buf
, sizeof(buf
));
4552 if (r
>= 0 && !strncmp(buf
, XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
)))
4557 r
= _fgetattrs(**fd
, inline_set
);
4558 incomplete_inline
= (r
== -E2BIG
);
4559 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4560 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
4561 << (incomplete_inline
? " (incomplete_inline, forcing omap)" : "")
4564 for (map
<string
,bufferptr
>::iterator p
= aset
.begin();
4567 char n
[CHAIN_XATTR_MAX_NAME_LEN
];
4568 get_attrname(p
->first
.c_str(), n
, CHAIN_XATTR_MAX_NAME_LEN
);
4570 if (incomplete_inline
) {
4571 chain_fremovexattr(**fd
, n
); // ignore any error
4572 omap_set
[p
->first
].push_back(p
->second
);
4576 if (p
->second
.length() > m_filestore_max_inline_xattr_size
) {
4577 if (inline_set
.count(p
->first
)) {
4578 inline_set
.erase(p
->first
);
4579 r
= chain_fremovexattr(**fd
, n
);
4583 omap_set
[p
->first
].push_back(p
->second
);
4587 if (!inline_set
.count(p
->first
) &&
4588 inline_set
.size() >= m_filestore_max_inline_xattrs
) {
4589 omap_set
[p
->first
].push_back(p
->second
);
4592 omap_remove
.insert(p
->first
);
4593 inline_set
.insert(*p
);
4595 inline_to_set
.insert(*p
);
4598 if (spill_out
!= 1 && !omap_set
.empty()) {
4599 chain_fsetxattr(**fd
, XATTR_SPILL_OUT_NAME
, XATTR_SPILL_OUT
,
4600 sizeof(XATTR_SPILL_OUT
));
4603 r
= _fsetattrs(**fd
, inline_to_set
);
4607 if (spill_out
&& !omap_remove
.empty()) {
4608 r
= object_map
->remove_xattrs(oid
, omap_remove
, &spos
);
4609 if (r
< 0 && r
!= -ENOENT
) {
4610 dout(10) << __FUNC__
<< ": could not remove_xattrs r = " << r
<< dendl
;
4611 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4614 r
= 0; // don't confuse the debug output
4618 if (!omap_set
.empty()) {
4619 r
= object_map
->set_xattrs(oid
, omap_set
, &spos
);
4621 dout(10) << __FUNC__
<< ": could not set_xattrs r = " << r
<< dendl
;
4622 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4629 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " = " << r
<< dendl
;
4634 int FileStore::_rmattr(const coll_t
& cid
, const ghobject_t
& oid
, const char *name
,
4635 const SequencerPosition
&spos
)
4637 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " '" << name
<< "'" << dendl
;
4639 bool spill_out
= true;
4641 int r
= lfn_open(cid
, oid
, false, &fd
);
4647 r
= chain_fgetxattr(**fd
, XATTR_SPILL_OUT_NAME
, buf
, sizeof(buf
));
4648 if (r
>= 0 && !strncmp(buf
, XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
))) {
4652 char n
[CHAIN_XATTR_MAX_NAME_LEN
];
4653 get_attrname(name
, n
, CHAIN_XATTR_MAX_NAME_LEN
);
4654 r
= chain_fremovexattr(**fd
, n
);
4655 if (r
== -ENODATA
&& spill_out
) {
4657 r
= get_index(cid
, &index
);
4659 dout(10) << __FUNC__
<< ": could not get index r = " << r
<< dendl
;
4662 set
<string
> to_remove
;
4663 to_remove
.insert(string(name
));
4664 r
= object_map
->remove_xattrs(oid
, to_remove
, &spos
);
4665 if (r
< 0 && r
!= -ENOENT
) {
4666 dout(10) << __FUNC__
<< ": could not remove_xattrs index r = " << r
<< dendl
;
4667 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4674 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " '" << name
<< "' = " << r
<< dendl
;
4678 int FileStore::_rmattrs(const coll_t
& cid
, const ghobject_t
& oid
,
4679 const SequencerPosition
&spos
)
4681 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< dendl
;
4683 map
<string
,bufferptr
> aset
;
4685 set
<string
> omap_attrs
;
4687 bool spill_out
= true;
4689 int r
= lfn_open(cid
, oid
, false, &fd
);
4695 r
= chain_fgetxattr(**fd
, XATTR_SPILL_OUT_NAME
, buf
, sizeof(buf
));
4696 if (r
>= 0 && !strncmp(buf
, XATTR_NO_SPILL_OUT
, sizeof(XATTR_NO_SPILL_OUT
))) {
4700 r
= _fgetattrs(**fd
, aset
);
4702 for (map
<string
,bufferptr
>::iterator p
= aset
.begin(); p
!= aset
.end(); ++p
) {
4703 char n
[CHAIN_XATTR_MAX_NAME_LEN
];
4704 get_attrname(p
->first
.c_str(), n
, CHAIN_XATTR_MAX_NAME_LEN
);
4705 r
= chain_fremovexattr(**fd
, n
);
4707 dout(10) << __FUNC__
<< ": could not remove xattr r = " << r
<< dendl
;
4714 dout(10) << __FUNC__
<< ": no xattr exists in object_map r = " << r
<< dendl
;
4718 r
= get_index(cid
, &index
);
4720 dout(10) << __FUNC__
<< ": could not get index r = " << r
<< dendl
;
4724 r
= object_map
->get_all_xattrs(oid
, &omap_attrs
);
4725 if (r
< 0 && r
!= -ENOENT
) {
4726 dout(10) << __FUNC__
<< ": could not get omap_attrs r = " << r
<< dendl
;
4727 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4730 r
= object_map
->remove_xattrs(oid
, omap_attrs
, &spos
);
4731 if (r
< 0 && r
!= -ENOENT
) {
4732 dout(10) << __FUNC__
<< ": could not remove omap_attrs r = " << r
<< dendl
;
4737 chain_fsetxattr(**fd
, XATTR_SPILL_OUT_NAME
, XATTR_NO_SPILL_OUT
,
4738 sizeof(XATTR_NO_SPILL_OUT
));
4744 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " = " << r
<< dendl
;
4751 int FileStore::_collection_remove_recursive(const coll_t
&cid
,
4752 const SequencerPosition
&spos
)
4755 int r
= collection_stat(cid
, &st
);
4762 vector
<ghobject_t
> objects
;
4764 while (!max
.is_max()) {
4765 r
= collection_list(cid
, max
, ghobject_t::get_max(),
4766 300, &objects
, &max
);
4769 for (vector
<ghobject_t
>::iterator i
= objects
.begin();
4772 assert(_check_replay_guard(cid
, *i
, spos
));
4773 r
= _remove(cid
, *i
, spos
);
4779 return _destroy_collection(cid
);
4782 // --------------------------
4785 int FileStore::list_collections(vector
<coll_t
>& ls
)
4787 return list_collections(ls
, false);
4790 int FileStore::list_collections(vector
<coll_t
>& ls
, bool include_temp
)
4792 tracepoint(objectstore
, list_collections_enter
);
4793 dout(10) << __FUNC__
<< dendl
;
4796 snprintf(fn
, sizeof(fn
), "%s/current", basedir
.c_str());
4799 DIR *dir
= ::opendir(fn
);
4802 derr
<< "tried opening directory " << fn
<< ": " << cpp_strerror(-r
) << dendl
;
4803 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4807 struct dirent
*de
= nullptr;
4808 while ((de
= ::readdir(dir
))) {
4809 if (de
->d_type
== DT_UNKNOWN
) {
4810 // d_type not supported (non-ext[234], btrfs), must stat
4812 char filename
[PATH_MAX
];
4813 snprintf(filename
, sizeof(filename
), "%s/%s", fn
, de
->d_name
);
4815 r
= ::stat(filename
, &sb
);
4818 derr
<< "stat on " << filename
<< ": " << cpp_strerror(-r
) << dendl
;
4819 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4822 if (!S_ISDIR(sb
.st_mode
)) {
4825 } else if (de
->d_type
!= DT_DIR
) {
4828 if (strcmp(de
->d_name
, "omap") == 0) {
4831 if (de
->d_name
[0] == '.' &&
4832 (de
->d_name
[1] == '\0' ||
4833 (de
->d_name
[1] == '.' &&
4834 de
->d_name
[2] == '\0')))
4837 if (!cid
.parse(de
->d_name
)) {
4838 derr
<< "ignoring invalid collection '" << de
->d_name
<< "'" << dendl
;
4841 if (!cid
.is_temp() || include_temp
)
4846 derr
<< "trying readdir " << fn
<< ": " << cpp_strerror(r
) << dendl
;
4851 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4852 tracepoint(objectstore
, list_collections_exit
, r
);
4856 int FileStore::collection_stat(const coll_t
& c
, struct stat
*st
)
4858 tracepoint(objectstore
, collection_stat_enter
, c
.c_str());
4860 get_cdir(c
, fn
, sizeof(fn
));
4861 dout(15) << __FUNC__
<< ": " << fn
<< dendl
;
4862 int r
= ::stat(fn
, st
);
4865 dout(10) << __FUNC__
<< ": " << fn
<< " = " << r
<< dendl
;
4866 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4867 tracepoint(objectstore
, collection_stat_exit
, r
);
4871 bool FileStore::collection_exists(const coll_t
& c
)
4873 tracepoint(objectstore
, collection_exists_enter
, c
.c_str());
4875 bool ret
= collection_stat(c
, &st
) == 0;
4876 tracepoint(objectstore
, collection_exists_exit
, ret
);
4880 int FileStore::collection_empty(const coll_t
& c
, bool *empty
)
4882 tracepoint(objectstore
, collection_empty_enter
, c
.c_str());
4883 dout(15) << __FUNC__
<< ": " << c
<< dendl
;
4885 int r
= get_index(c
, &index
);
4887 derr
<< __FUNC__
<< ": get_index returned: " << cpp_strerror(r
)
4892 assert(NULL
!= index
.index
);
4893 RWLock::RLocker
l((index
.index
)->access_lock
);
4895 vector
<ghobject_t
> ls
;
4896 r
= index
->collection_list_partial(ghobject_t(), ghobject_t::get_max(),
4899 derr
<< __FUNC__
<< ": collection_list_partial returned: "
4900 << cpp_strerror(r
) << dendl
;
4901 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
4904 *empty
= ls
.empty();
4905 tracepoint(objectstore
, collection_empty_exit
, *empty
);
4909 int FileStore::_collection_set_bits(const coll_t
& c
, int bits
)
4912 get_cdir(c
, fn
, sizeof(fn
));
4913 dout(10) << __FUNC__
<< ": " << fn
<< " " << bits
<< dendl
;
4917 int fd
= ::open(fn
, O_RDONLY
);
4922 get_attrname("bits", n
, PATH_MAX
);
4923 r
= chain_fsetxattr(fd
, n
, (char*)&v
, sizeof(v
));
4924 VOID_TEMP_FAILURE_RETRY(::close(fd
));
4926 dout(10) << __FUNC__
<< ": " << fn
<< " " << bits
<< " = " << r
<< dendl
;
4930 int FileStore::collection_bits(const coll_t
& c
)
4933 get_cdir(c
, fn
, sizeof(fn
));
4934 dout(15) << __FUNC__
<< ": " << fn
<< dendl
;
4938 int fd
= ::open(fn
, O_RDONLY
);
4943 get_attrname("bits", n
, PATH_MAX
);
4944 r
= chain_fgetxattr(fd
, n
, (char*)&bits
, sizeof(bits
));
4945 VOID_TEMP_FAILURE_RETRY(::close(fd
));
4951 dout(10) << __FUNC__
<< ": " << fn
<< " = " << bits
<< dendl
;
4955 int FileStore::collection_list(const coll_t
& c
,
4956 const ghobject_t
& orig_start
,
4957 const ghobject_t
& end
,
4959 vector
<ghobject_t
> *ls
, ghobject_t
*next
)
4961 ghobject_t start
= orig_start
;
4965 ghobject_t temp_next
;
4968 // figure out the pool id. we need this in order to generate a
4969 // meaningful 'next' value.
4974 if (c
.is_temp(&pgid
)) {
4975 pool
= -2 - pgid
.pool();
4977 } else if (c
.is_pg(&pgid
)) {
4980 } else if (c
.is_meta()) {
4982 shard
= shard_id_t::NO_SHARD
;
4984 // hrm, the caller is test code! we should get kill it off. for now,
4987 shard
= shard_id_t::NO_SHARD
;
4989 dout(20) << __FUNC__
<< ": pool is " << pool
<< " shard is " << shard
4990 << " pgid " << pgid
<< dendl
;
4994 sep
.set_shard(shard
);
4995 if (!c
.is_temp() && !c
.is_meta()) {
4997 dout(10) << __FUNC__
<< ": first checking temp pool" << dendl
;
4998 coll_t temp
= c
.get_temp();
4999 int r
= collection_list(temp
, start
, end
, max
, ls
, next
);
5002 if (*next
!= ghobject_t::get_max())
5005 dout(10) << __FUNC__
<< ": fall through to non-temp collection, start "
5008 dout(10) << __FUNC__
<< ": start " << start
<< " >= sep " << sep
<< dendl
;
5013 int r
= get_index(c
, &index
);
5017 assert(NULL
!= index
.index
);
5018 RWLock::RLocker
l((index
.index
)->access_lock
);
5020 r
= index
->collection_list_partial(start
, end
, max
, ls
, next
);
5023 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
5026 dout(20) << "objects: " << *ls
<< dendl
;
5028 // HashIndex doesn't know the pool when constructing a 'next' value
5029 if (next
&& !next
->is_max()) {
5030 next
->hobj
.pool
= pool
;
5031 next
->set_shard(shard
);
5032 dout(20) << " next " << *next
<< dendl
;
5038 int FileStore::omap_get(const coll_t
& _c
, const ghobject_t
&hoid
,
5040 map
<string
, bufferlist
> *out
)
5042 tracepoint(objectstore
, omap_get_enter
, _c
.c_str());
5043 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5044 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5046 int r
= get_index(c
, &index
);
5050 assert(NULL
!= index
.index
);
5051 RWLock::RLocker
l((index
.index
)->access_lock
);
5052 r
= lfn_find(hoid
, index
);
5056 r
= object_map
->get(hoid
, header
, out
);
5057 if (r
< 0 && r
!= -ENOENT
) {
5058 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
5061 tracepoint(objectstore
, omap_get_exit
, 0);
5065 int FileStore::omap_get_header(
5067 const ghobject_t
&hoid
,
5071 tracepoint(objectstore
, omap_get_header_enter
, _c
.c_str());
5072 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5073 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5075 int r
= get_index(c
, &index
);
5079 assert(NULL
!= index
.index
);
5080 RWLock::RLocker
l((index
.index
)->access_lock
);
5081 r
= lfn_find(hoid
, index
);
5085 r
= object_map
->get_header(hoid
, bl
);
5086 if (r
< 0 && r
!= -ENOENT
) {
5087 assert(allow_eio
|| !m_filestore_fail_eio
|| r
!= -EIO
);
5090 tracepoint(objectstore
, omap_get_header_exit
, 0);
5094 int FileStore::omap_get_keys(const coll_t
& _c
, const ghobject_t
&hoid
, set
<string
> *keys
)
5096 tracepoint(objectstore
, omap_get_keys_enter
, _c
.c_str());
5097 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5098 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5100 int r
= get_index(c
, &index
);
5104 assert(NULL
!= index
.index
);
5105 RWLock::RLocker
l((index
.index
)->access_lock
);
5106 r
= lfn_find(hoid
, index
);
5110 r
= object_map
->get_keys(hoid
, keys
);
5111 if (r
< 0 && r
!= -ENOENT
) {
5112 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
5115 tracepoint(objectstore
, omap_get_keys_exit
, 0);
5119 int FileStore::omap_get_values(const coll_t
& _c
, const ghobject_t
&hoid
,
5120 const set
<string
> &keys
,
5121 map
<string
, bufferlist
> *out
)
5123 tracepoint(objectstore
, omap_get_values_enter
, _c
.c_str());
5124 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5125 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5127 const char *where
= "()";
5128 int r
= get_index(c
, &index
);
5130 where
= " (get_index)";
5134 assert(NULL
!= index
.index
);
5135 RWLock::RLocker
l((index
.index
)->access_lock
);
5136 r
= lfn_find(hoid
, index
);
5138 where
= " (lfn_find)";
5142 r
= object_map
->get_values(hoid
, keys
, out
);
5143 if (r
< 0 && r
!= -ENOENT
) {
5144 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
5145 where
= " (get_values)";
5150 tracepoint(objectstore
, omap_get_values_exit
, r
);
5151 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< " = " << r
5156 int FileStore::omap_check_keys(const coll_t
& _c
, const ghobject_t
&hoid
,
5157 const set
<string
> &keys
,
5160 tracepoint(objectstore
, omap_check_keys_enter
, _c
.c_str());
5161 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5162 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5165 int r
= get_index(c
, &index
);
5169 assert(NULL
!= index
.index
);
5170 RWLock::RLocker
l((index
.index
)->access_lock
);
5171 r
= lfn_find(hoid
, index
);
5175 r
= object_map
->check_keys(hoid
, keys
, out
);
5176 if (r
< 0 && r
!= -ENOENT
) {
5177 assert(!m_filestore_fail_eio
|| r
!= -EIO
);
5180 tracepoint(objectstore
, omap_check_keys_exit
, 0);
5184 ObjectMap::ObjectMapIterator
FileStore::get_omap_iterator(const coll_t
& _c
,
5185 const ghobject_t
&hoid
)
5187 tracepoint(objectstore
, get_omap_iterator
, _c
.c_str());
5188 const coll_t
& c
= !_need_temp_object_collection(_c
, hoid
) ? _c
: _c
.get_temp();
5189 dout(15) << __FUNC__
<< ": " << c
<< "/" << hoid
<< dendl
;
5191 int r
= get_index(c
, &index
);
5193 dout(10) << __FUNC__
<< ": " << c
<< "/" << hoid
<< " = 0 "
5194 << "(get_index failed with " << cpp_strerror(r
) << ")" << dendl
;
5195 return ObjectMap::ObjectMapIterator();
5198 assert(NULL
!= index
.index
);
5199 RWLock::RLocker
l((index
.index
)->access_lock
);
5200 r
= lfn_find(hoid
, index
);
5202 dout(10) << __FUNC__
<< ": " << c
<< "/" << hoid
<< " = 0 "
5203 << "(lfn_find failed with " << cpp_strerror(r
) << ")" << dendl
;
5204 return ObjectMap::ObjectMapIterator();
5207 return object_map
->get_iterator(hoid
);
5210 int FileStore::_collection_hint_expected_num_objs(const coll_t
& c
, uint32_t pg_num
,
5211 uint64_t expected_num_objs
,
5212 const SequencerPosition
&spos
)
5214 dout(15) << __FUNC__
<< ": collection: " << c
<< " pg number: "
5215 << pg_num
<< " expected number of objects: " << expected_num_objs
<< dendl
;
5218 int ret
= collection_empty(c
, &empty
);
5221 if (!empty
&& !replaying
) {
5222 dout(0) << "Failed to give an expected number of objects hint to collection : "
5223 << c
<< ", only empty collection can take such type of hint. " << dendl
;
5228 ret
= get_index(c
, &index
);
5231 // Pre-hash the collection
5232 ret
= index
->pre_hash_collection(pg_num
, expected_num_objs
);
5233 dout(10) << "pre_hash_collection " << c
<< " = " << ret
<< dendl
;
5236 _set_replay_guard(c
, spos
);
5241 int FileStore::_create_collection(
5244 const SequencerPosition
&spos
)
5247 get_cdir(c
, fn
, sizeof(fn
));
5248 dout(15) << __FUNC__
<< ": " << fn
<< dendl
;
5249 int r
= ::mkdir(fn
, 0755);
5252 if (r
== -EEXIST
&& replaying
)
5254 dout(10) << __FUNC__
<< ": " << fn
<< " = " << r
<< dendl
;
5261 r
= _collection_set_bits(c
, bits
);
5264 // create parallel temp collection, too
5265 if (!c
.is_meta() && !c
.is_temp()) {
5266 coll_t temp
= c
.get_temp();
5267 r
= _create_collection(temp
, 0, spos
);
5272 _set_replay_guard(c
, spos
);
5276 int FileStore::_destroy_collection(const coll_t
& c
)
5280 get_cdir(c
, fn
, sizeof(fn
));
5281 dout(15) << __FUNC__
<< ": " << fn
<< dendl
;
5284 r
= get_index(c
, &from
);
5287 assert(NULL
!= from
.index
);
5288 RWLock::WLocker
l((from
.index
)->access_lock
);
5290 r
= from
->prep_delete();
5301 // destroy parallel temp collection, too
5302 if (!c
.is_meta() && !c
.is_temp()) {
5303 coll_t temp
= c
.get_temp();
5304 int r2
= _destroy_collection(temp
);
5312 dout(10) << __FUNC__
<< ": " << fn
<< " = " << r
<< dendl
;
5317 int FileStore::_collection_add(const coll_t
& c
, const coll_t
& oldcid
, const ghobject_t
& o
,
5318 const SequencerPosition
& spos
)
5320 dout(15) << __FUNC__
<< ": " << c
<< "/" << o
<< " from " << oldcid
<< "/" << o
<< dendl
;
5322 int dstcmp
= _check_replay_guard(c
, o
, spos
);
5326 // check the src name too; it might have a newer guard, and we don't
5327 // want to clobber it
5328 int srccmp
= _check_replay_guard(oldcid
, o
, spos
);
5332 // open guard on object so we don't any previous operations on the
5333 // new name that will modify the source inode.
5335 int r
= lfn_open(oldcid
, o
, 0, &fd
);
5337 // the source collection/object does not exist. If we are replaying, we
5338 // should be safe, so just return 0 and move on.
5340 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from "
5341 << oldcid
<< "/" << o
<< " (dne, continue replay) " << dendl
;
5344 if (dstcmp
> 0) { // if dstcmp == 0 the guard already says "in-progress"
5345 _set_replay_guard(**fd
, spos
, &o
, true);
5348 r
= lfn_link(oldcid
, c
, o
, o
);
5349 if (replaying
&& !backend
->can_checkpoint() &&
5350 r
== -EEXIST
) // crashed between link() and set_replay_guard()
5355 // close guard on object so we don't do this again
5357 _close_replay_guard(**fd
, spos
);
5361 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from " << oldcid
<< "/" << o
<< " = " << r
<< dendl
;
5365 int FileStore::_collection_move_rename(const coll_t
& oldcid
, const ghobject_t
& oldoid
,
5366 coll_t c
, const ghobject_t
& o
,
5367 const SequencerPosition
& spos
,
5370 dout(15) << __FUNC__
<< ": " << c
<< "/" << o
<< " from " << oldcid
<< "/" << oldoid
<< dendl
;
5375 /* If the destination collection doesn't exist during replay,
5376 * we need to delete the src object and continue on
5378 if (!collection_exists(c
))
5382 dstcmp
= _check_replay_guard(c
, o
, spos
);
5386 // check the src name too; it might have a newer guard, and we don't
5387 // want to clobber it
5388 srccmp
= _check_replay_guard(oldcid
, oldoid
, spos
);
5393 // open guard on object so we don't any previous operations on the
5394 // new name that will modify the source inode.
5396 r
= lfn_open(oldcid
, oldoid
, 0, &fd
);
5398 // the source collection/object does not exist. If we are replaying, we
5399 // should be safe, so just return 0 and move on.
5401 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from "
5402 << oldcid
<< "/" << oldoid
<< " (dne, continue replay) " << dendl
;
5403 } else if (allow_enoent
) {
5404 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from "
5405 << oldcid
<< "/" << oldoid
<< " (dne, ignoring enoent)"
5408 assert(0 == "ERROR: source must exist");
5414 if (allow_enoent
&& dstcmp
> 0) { // if dstcmp == 0, try_rename was started.
5418 r
= 0; // don't know if object_map was cloned
5420 if (dstcmp
> 0) { // if dstcmp == 0 the guard already says "in-progress"
5421 _set_replay_guard(**fd
, spos
, &o
, true);
5424 r
= lfn_link(oldcid
, c
, oldoid
, o
);
5425 if (replaying
&& !backend
->can_checkpoint() &&
5426 r
== -EEXIST
) // crashed between link() and set_replay_guard()
5436 // the name changed; link the omap content
5437 r
= object_map
->rename(oldoid
, o
, &spos
);
5445 r
= lfn_unlink(oldcid
, oldoid
, spos
, true);
5448 r
= lfn_open(c
, o
, 0, &fd
);
5450 // close guard on object so we don't do this again
5452 _close_replay_guard(**fd
, spos
, &o
);
5457 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from " << oldcid
<< "/" << oldoid
5458 << " = " << r
<< dendl
;
5463 if (_check_replay_guard(oldcid
, oldoid
, spos
) > 0) {
5464 r
= lfn_unlink(oldcid
, oldoid
, spos
, true);
5467 dout(10) << __FUNC__
<< ": " << c
<< "/" << o
<< " from " << oldcid
<< "/" << oldoid
5468 << " = " << r
<< dendl
;
5472 void FileStore::_inject_failure()
5474 if (m_filestore_kill_at
) {
5475 int final
= --m_filestore_kill_at
;
5476 dout(5) << __FUNC__
<< ": " << (final
+1) << " -> " << final
<< dendl
;
5478 derr
<< __FUNC__
<< ": KILLING" << dendl
;
5485 int FileStore::_omap_clear(const coll_t
& cid
, const ghobject_t
&hoid
,
5486 const SequencerPosition
&spos
) {
5487 dout(15) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< dendl
;
5489 int r
= get_index(cid
, &index
);
5493 assert(NULL
!= index
.index
);
5494 RWLock::RLocker
l((index
.index
)->access_lock
);
5495 r
= lfn_find(hoid
, index
);
5499 r
= object_map
->clear_keys_header(hoid
, &spos
);
5500 if (r
< 0 && r
!= -ENOENT
)
5505 int FileStore::_omap_setkeys(const coll_t
& cid
, const ghobject_t
&hoid
,
5506 const map
<string
, bufferlist
> &aset
,
5507 const SequencerPosition
&spos
) {
5508 dout(15) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< dendl
;
5511 //treat pgmeta as a logical object, skip to check exist
5512 if (hoid
.is_pgmeta())
5515 r
= get_index(cid
, &index
);
5517 dout(20) << __FUNC__
<< ": get_index got " << cpp_strerror(r
) << dendl
;
5521 assert(NULL
!= index
.index
);
5522 RWLock::RLocker
l((index
.index
)->access_lock
);
5523 r
= lfn_find(hoid
, index
);
5525 dout(20) << __FUNC__
<< ": lfn_find got " << cpp_strerror(r
) << dendl
;
5530 if (g_conf
->subsys
.should_gather(ceph_subsys_filestore
, 20)) {
5531 for (auto& p
: aset
) {
5532 dout(20) << __FUNC__
<< ": set " << p
.first
<< dendl
;
5535 r
= object_map
->set_keys(hoid
, aset
, &spos
);
5536 dout(20) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< " = " << r
<< dendl
;
5540 int FileStore::_omap_rmkeys(const coll_t
& cid
, const ghobject_t
&hoid
,
5541 const set
<string
> &keys
,
5542 const SequencerPosition
&spos
) {
5543 dout(15) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< dendl
;
5546 //treat pgmeta as a logical object, skip to check exist
5547 if (hoid
.is_pgmeta())
5550 r
= get_index(cid
, &index
);
5554 assert(NULL
!= index
.index
);
5555 RWLock::RLocker
l((index
.index
)->access_lock
);
5556 r
= lfn_find(hoid
, index
);
5561 r
= object_map
->rm_keys(hoid
, keys
, &spos
);
5562 if (r
< 0 && r
!= -ENOENT
)
5567 int FileStore::_omap_rmkeyrange(const coll_t
& cid
, const ghobject_t
&hoid
,
5568 const string
& first
, const string
& last
,
5569 const SequencerPosition
&spos
) {
5570 dout(15) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< " [" << first
<< "," << last
<< "]" << dendl
;
5573 ObjectMap::ObjectMapIterator iter
= get_omap_iterator(cid
, hoid
);
5576 for (iter
->lower_bound(first
); iter
->valid() && iter
->key() < last
;
5578 keys
.insert(iter
->key());
5581 return _omap_rmkeys(cid
, hoid
, keys
, spos
);
5584 int FileStore::_omap_setheader(const coll_t
& cid
, const ghobject_t
&hoid
,
5585 const bufferlist
&bl
,
5586 const SequencerPosition
&spos
)
5588 dout(15) << __FUNC__
<< ": " << cid
<< "/" << hoid
<< dendl
;
5590 int r
= get_index(cid
, &index
);
5594 assert(NULL
!= index
.index
);
5595 RWLock::RLocker
l((index
.index
)->access_lock
);
5596 r
= lfn_find(hoid
, index
);
5600 return object_map
->set_header(hoid
, bl
, &spos
);
5603 int FileStore::_split_collection(const coll_t
& cid
,
5607 const SequencerPosition
&spos
)
5611 dout(15) << __FUNC__
<< ": " << cid
<< " bits: " << bits
<< dendl
;
5612 if (!collection_exists(cid
)) {
5613 dout(2) << __FUNC__
<< ": " << cid
<< " DNE" << dendl
;
5617 if (!collection_exists(dest
)) {
5618 dout(2) << __FUNC__
<< ": " << dest
<< " DNE" << dendl
;
5623 int dstcmp
= _check_replay_guard(dest
, spos
);
5627 int srccmp
= _check_replay_guard(cid
, spos
);
5631 _set_global_replay_guard(cid
, spos
);
5632 _set_replay_guard(cid
, spos
, true);
5633 _set_replay_guard(dest
, spos
, true);
5636 r
= get_index(cid
, &from
);
5640 r
= get_index(dest
, &to
);
5643 assert(NULL
!= from
.index
);
5644 RWLock::WLocker
l1((from
.index
)->access_lock
);
5646 assert(NULL
!= to
.index
);
5647 RWLock::WLocker
l2((to
.index
)->access_lock
);
5649 r
= from
->split(rem
, bits
, to
.index
);
5652 _close_replay_guard(cid
, spos
);
5653 _close_replay_guard(dest
, spos
);
5655 _collection_set_bits(cid
, bits
);
5656 if (!r
&& cct
->_conf
->filestore_debug_verify_split
) {
5657 vector
<ghobject_t
> objects
;
5662 next
, ghobject_t::get_max(),
5663 get_ideal_list_max(),
5666 if (objects
.empty())
5668 for (vector
<ghobject_t
>::iterator i
= objects
.begin();
5671 dout(20) << __FUNC__
<< ": " << *i
<< " still in source "
5673 assert(!i
->match(bits
, rem
));
5677 next
= ghobject_t();
5681 next
, ghobject_t::get_max(),
5682 get_ideal_list_max(),
5685 if (objects
.empty())
5687 for (vector
<ghobject_t
>::iterator i
= objects
.begin();
5690 dout(20) << __FUNC__
<< ": " << *i
<< " now in dest "
5692 assert(i
->match(bits
, rem
));
5700 int FileStore::_set_alloc_hint(const coll_t
& cid
, const ghobject_t
& oid
,
5701 uint64_t expected_object_size
,
5702 uint64_t expected_write_size
)
5704 dout(15) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " object_size " << expected_object_size
<< " write_size " << expected_write_size
<< dendl
;
5709 if (expected_object_size
== 0 || expected_write_size
== 0)
5712 ret
= lfn_open(cid
, oid
, false, &fd
);
5717 // TODO: a more elaborate hint calculation
5718 uint64_t hint
= MIN(expected_write_size
, m_filestore_max_alloc_hint_size
);
5720 ret
= backend
->set_alloc_hint(**fd
, hint
);
5721 dout(20) << __FUNC__
<< ": hint " << hint
<< " ret " << ret
<< dendl
;
5726 dout(10) << __FUNC__
<< ": " << cid
<< "/" << oid
<< " object_size " << expected_object_size
<< " write_size " << expected_write_size
<< " = " << ret
<< dendl
;
5727 assert(!m_filestore_fail_eio
|| ret
!= -EIO
);
5731 const char** FileStore::get_tracked_conf_keys() const
5733 static const char* KEYS
[] = {
5734 "filestore_max_inline_xattr_size",
5735 "filestore_max_inline_xattr_size_xfs",
5736 "filestore_max_inline_xattr_size_btrfs",
5737 "filestore_max_inline_xattr_size_other",
5738 "filestore_max_inline_xattrs",
5739 "filestore_max_inline_xattrs_xfs",
5740 "filestore_max_inline_xattrs_btrfs",
5741 "filestore_max_inline_xattrs_other",
5742 "filestore_max_xattr_value_size",
5743 "filestore_max_xattr_value_size_xfs",
5744 "filestore_max_xattr_value_size_btrfs",
5745 "filestore_max_xattr_value_size_other",
5746 "filestore_min_sync_interval",
5747 "filestore_max_sync_interval",
5748 "filestore_queue_max_ops",
5749 "filestore_queue_max_bytes",
5750 "filestore_expected_throughput_bytes",
5751 "filestore_expected_throughput_ops",
5752 "filestore_queue_low_threshhold",
5753 "filestore_queue_high_threshhold",
5754 "filestore_queue_high_delay_multiple",
5755 "filestore_queue_max_delay_multiple",
5756 "filestore_commit_timeout",
5757 "filestore_dump_file",
5758 "filestore_kill_at",
5759 "filestore_fail_eio",
5760 "filestore_fadvise",
5761 "filestore_sloppy_crc",
5762 "filestore_sloppy_crc_block_size",
5763 "filestore_max_alloc_hint_size",
5769 void FileStore::handle_conf_change(const struct md_config_t
*conf
,
5770 const std::set
<std::string
> &changed
)
5772 if (changed
.count("filestore_max_inline_xattr_size") ||
5773 changed
.count("filestore_max_inline_xattr_size_xfs") ||
5774 changed
.count("filestore_max_inline_xattr_size_btrfs") ||
5775 changed
.count("filestore_max_inline_xattr_size_other") ||
5776 changed
.count("filestore_max_inline_xattrs") ||
5777 changed
.count("filestore_max_inline_xattrs_xfs") ||
5778 changed
.count("filestore_max_inline_xattrs_btrfs") ||
5779 changed
.count("filestore_max_inline_xattrs_other") ||
5780 changed
.count("filestore_max_xattr_value_size") ||
5781 changed
.count("filestore_max_xattr_value_size_xfs") ||
5782 changed
.count("filestore_max_xattr_value_size_btrfs") ||
5783 changed
.count("filestore_max_xattr_value_size_other")) {
5785 Mutex::Locker
l(lock
);
5786 set_xattr_limits_via_conf();
5790 if (changed
.count("filestore_queue_max_bytes") ||
5791 changed
.count("filestore_queue_max_ops") ||
5792 changed
.count("filestore_expected_throughput_bytes") ||
5793 changed
.count("filestore_expected_throughput_ops") ||
5794 changed
.count("filestore_queue_low_threshhold") ||
5795 changed
.count("filestore_queue_high_threshhold") ||
5796 changed
.count("filestore_queue_high_delay_multiple") ||
5797 changed
.count("filestore_queue_max_delay_multiple")) {
5798 Mutex::Locker
l(lock
);
5799 set_throttle_params();
5802 if (changed
.count("filestore_min_sync_interval") ||
5803 changed
.count("filestore_max_sync_interval") ||
5804 changed
.count("filestore_kill_at") ||
5805 changed
.count("filestore_fail_eio") ||
5806 changed
.count("filestore_sloppy_crc") ||
5807 changed
.count("filestore_sloppy_crc_block_size") ||
5808 changed
.count("filestore_max_alloc_hint_size") ||
5809 changed
.count("filestore_fadvise")) {
5810 Mutex::Locker
l(lock
);
5811 m_filestore_min_sync_interval
= conf
->filestore_min_sync_interval
;
5812 m_filestore_max_sync_interval
= conf
->filestore_max_sync_interval
;
5813 m_filestore_kill_at
= conf
->filestore_kill_at
;
5814 m_filestore_fail_eio
= conf
->filestore_fail_eio
;
5815 m_filestore_fadvise
= conf
->filestore_fadvise
;
5816 m_filestore_sloppy_crc
= conf
->filestore_sloppy_crc
;
5817 m_filestore_sloppy_crc_block_size
= conf
->filestore_sloppy_crc_block_size
;
5818 m_filestore_max_alloc_hint_size
= conf
->filestore_max_alloc_hint_size
;
5820 if (changed
.count("filestore_commit_timeout")) {
5821 Mutex::Locker
l(sync_entry_timeo_lock
);
5822 m_filestore_commit_timeout
= conf
->filestore_commit_timeout
;
5824 if (changed
.count("filestore_dump_file")) {
5825 if (conf
->filestore_dump_file
.length() &&
5826 conf
->filestore_dump_file
!= "-") {
5827 dump_start(conf
->filestore_dump_file
);
5834 int FileStore::set_throttle_params()
5837 bool valid
= throttle_bytes
.set_params(
5838 cct
->_conf
->filestore_queue_low_threshhold
,
5839 cct
->_conf
->filestore_queue_high_threshhold
,
5840 cct
->_conf
->filestore_expected_throughput_bytes
,
5841 cct
->_conf
->filestore_queue_high_delay_multiple
,
5842 cct
->_conf
->filestore_queue_max_delay_multiple
,
5843 cct
->_conf
->filestore_queue_max_bytes
,
5846 valid
&= throttle_ops
.set_params(
5847 cct
->_conf
->filestore_queue_low_threshhold
,
5848 cct
->_conf
->filestore_queue_high_threshhold
,
5849 cct
->_conf
->filestore_expected_throughput_ops
,
5850 cct
->_conf
->filestore_queue_high_delay_multiple
,
5851 cct
->_conf
->filestore_queue_max_delay_multiple
,
5852 cct
->_conf
->filestore_queue_max_ops
,
5855 logger
->set(l_filestore_op_queue_max_ops
, throttle_ops
.get_max());
5856 logger
->set(l_filestore_op_queue_max_bytes
, throttle_bytes
.get_max());
5859 derr
<< "tried to set invalid params: "
5863 return valid
? 0 : -EINVAL
;
5866 void FileStore::dump_start(const std::string
& file
)
5868 dout(10) << __FUNC__
<< ": " << file
<< dendl
;
5869 if (m_filestore_do_dump
) {
5872 m_filestore_dump_fmt
.reset();
5873 m_filestore_dump_fmt
.open_array_section("dump");
5874 m_filestore_dump
.open(file
.c_str());
5875 m_filestore_do_dump
= true;
5878 void FileStore::dump_stop()
5880 dout(10) << __FUNC__
<< dendl
;
5881 m_filestore_do_dump
= false;
5882 if (m_filestore_dump
.is_open()) {
5883 m_filestore_dump_fmt
.close_section();
5884 m_filestore_dump_fmt
.flush(m_filestore_dump
);
5885 m_filestore_dump
.flush();
5886 m_filestore_dump
.close();
5890 void FileStore::dump_transactions(vector
<ObjectStore::Transaction
>& ls
, uint64_t seq
, OpSequencer
*osr
)
5892 m_filestore_dump_fmt
.open_array_section("transactions");
5893 unsigned trans_num
= 0;
5894 for (vector
<ObjectStore::Transaction
>::iterator i
= ls
.begin(); i
!= ls
.end(); ++i
, ++trans_num
) {
5895 m_filestore_dump_fmt
.open_object_section("transaction");
5896 m_filestore_dump_fmt
.dump_string("osr", osr
->get_name());
5897 m_filestore_dump_fmt
.dump_unsigned("seq", seq
);
5898 m_filestore_dump_fmt
.dump_unsigned("trans_num", trans_num
);
5899 (*i
).dump(&m_filestore_dump_fmt
);
5900 m_filestore_dump_fmt
.close_section();
5902 m_filestore_dump_fmt
.close_section();
5903 m_filestore_dump_fmt
.flush(m_filestore_dump
);
5904 m_filestore_dump
.flush();
5907 void FileStore::set_xattr_limits_via_conf()
5909 uint32_t fs_xattr_size
;
5911 uint32_t fs_xattr_max_value_size
;
5913 switch (m_fs_type
) {
5914 #if defined(__linux__)
5915 case XFS_SUPER_MAGIC
:
5916 fs_xattr_size
= cct
->_conf
->filestore_max_inline_xattr_size_xfs
;
5917 fs_xattrs
= cct
->_conf
->filestore_max_inline_xattrs_xfs
;
5918 fs_xattr_max_value_size
= cct
->_conf
->filestore_max_xattr_value_size_xfs
;
5920 case BTRFS_SUPER_MAGIC
:
5921 fs_xattr_size
= cct
->_conf
->filestore_max_inline_xattr_size_btrfs
;
5922 fs_xattrs
= cct
->_conf
->filestore_max_inline_xattrs_btrfs
;
5923 fs_xattr_max_value_size
= cct
->_conf
->filestore_max_xattr_value_size_btrfs
;
5927 fs_xattr_size
= cct
->_conf
->filestore_max_inline_xattr_size_other
;
5928 fs_xattrs
= cct
->_conf
->filestore_max_inline_xattrs_other
;
5929 fs_xattr_max_value_size
= cct
->_conf
->filestore_max_xattr_value_size_other
;
5933 // Use override value if set
5934 if (cct
->_conf
->filestore_max_inline_xattr_size
)
5935 m_filestore_max_inline_xattr_size
= cct
->_conf
->filestore_max_inline_xattr_size
;
5937 m_filestore_max_inline_xattr_size
= fs_xattr_size
;
5939 // Use override value if set
5940 if (cct
->_conf
->filestore_max_inline_xattrs
)
5941 m_filestore_max_inline_xattrs
= cct
->_conf
->filestore_max_inline_xattrs
;
5943 m_filestore_max_inline_xattrs
= fs_xattrs
;
5945 // Use override value if set
5946 if (cct
->_conf
->filestore_max_xattr_value_size
)
5947 m_filestore_max_xattr_value_size
= cct
->_conf
->filestore_max_xattr_value_size
;
5949 m_filestore_max_xattr_value_size
= fs_xattr_max_value_size
;
5951 if (m_filestore_max_xattr_value_size
< cct
->_conf
->osd_max_object_name_len
) {
5952 derr
<< "WARNING: max attr value size ("
5953 << m_filestore_max_xattr_value_size
5954 << ") is smaller than osd_max_object_name_len ("
5955 << cct
->_conf
->osd_max_object_name_len
5956 << "). Your backend filesystem appears to not support attrs large "
5957 << "enough to handle the configured max rados name size. You may get "
5958 << "unexpected ENAMETOOLONG errors on rados operations or buggy "
5964 uint64_t FileStore::estimate_objects_overhead(uint64_t num_objects
)
5966 uint64_t res
= num_objects
* blk_size
/ 2; //assumes that each object uses ( in average ) additional 1/2 block due to FS allocation granularity.
5970 int FileStore::apply_layout_settings(const coll_t
&cid
)
5972 dout(20) << __FUNC__
<< ": " << cid
<< dendl
;
5974 int r
= get_index(cid
, &index
);
5976 dout(10) << "Error getting index for " << cid
<< ": " << cpp_strerror(r
)
5981 return index
->apply_layout_settings();
5985 // -- FSSuperblock --
5987 void FSSuperblock::encode(bufferlist
&bl
) const
5989 ENCODE_START(2, 1, bl
);
5990 compat_features
.encode(bl
);
5991 ::encode(omap_backend
, bl
);
5995 void FSSuperblock::decode(bufferlist::iterator
&bl
)
5997 DECODE_START(2, bl
);
5998 compat_features
.decode(bl
);
6000 ::decode(omap_backend
, bl
);
6002 omap_backend
= "leveldb";
6006 void FSSuperblock::dump(Formatter
*f
) const
6008 f
->open_object_section("compat");
6009 compat_features
.dump(f
);
6010 f
->dump_string("omap_backend", omap_backend
);
6014 void FSSuperblock::generate_test_instances(list
<FSSuperblock
*>& o
)
6017 o
.push_back(new FSSuperblock(z
));
6018 CompatSet::FeatureSet feature_compat
;
6019 CompatSet::FeatureSet feature_ro_compat
;
6020 CompatSet::FeatureSet feature_incompat
;
6021 feature_incompat
.insert(CEPH_FS_FEATURE_INCOMPAT_SHARDS
);
6022 z
.compat_features
= CompatSet(feature_compat
, feature_ro_compat
,
6024 o
.push_back(new FSSuperblock(z
));
6025 z
.omap_backend
= "rocksdb";
6026 o
.push_back(new FSSuperblock(z
));