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>
8 * This is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License version 2.1, as published by the Free Software
11 * Foundation. See file COPYING.
15 #include "include/compat.h"
16 #include "include/types.h"
17 #include "include/buffer.h"
18 #include "osd/osd_types.h"
21 #include "HashIndex.h"
23 #include "common/errno.h"
24 #include "common/debug.h"
25 #define dout_context cct
26 #define dout_subsys ceph_subsys_filestore
28 const string
HashIndex::SUBDIR_ATTR
= "contents";
29 const string
HashIndex::SETTINGS_ATTR
= "settings";
30 const string
HashIndex::IN_PROGRESS_OP_TAG
= "in_progress_op";
32 /// hex digit to integer value
33 int hex_to_int(char c
)
35 if (c
>= '0' && c
<= '9')
37 if (c
>= 'A' && c
<= 'F')
42 /// int value to hex digit
43 char int_to_hex(int v
)
51 /// reverse bits in a nibble (0..15)
52 int reverse_nibble_bits(int in
)
62 /// reverse nibble bits in a hex digit
63 char reverse_hexdigit_bits(char c
)
65 return int_to_hex(reverse_nibble_bits(hex_to_int(c
)));
68 /// reverse nibble bits in a hex string
69 string
reverse_hexdigit_bits_string(string s
)
71 for (unsigned i
=0; i
<s
.size(); ++i
)
72 s
[i
] = reverse_hexdigit_bits(s
[i
]);
76 /// compare hex digit (as length 1 string) bitwise
77 bool cmp_hexdigit_bitwise(const string
& l
, const string
& r
)
79 ceph_assert(l
.length() == 1 && r
.length() == 1);
80 int lv
= hex_to_int(l
[0]);
81 int rv
= hex_to_int(r
[0]);
84 return reverse_nibble_bits(lv
) < reverse_nibble_bits(rv
);
87 /// compare hex digit string bitwise
88 bool cmp_hexdigit_string_bitwise(const string
& l
, const string
& r
)
90 string ll
= reverse_hexdigit_bits_string(l
);
91 string rr
= reverse_hexdigit_bits_string(r
);
95 int HashIndex::cleanup() {
97 int r
= get_attr_path(vector
<string
>(), IN_PROGRESS_OP_TAG
, bl
);
99 // No in progress operations!
102 auto i
= bl
.cbegin();
103 InProgressOp
in_progress(i
);
105 r
= get_info(in_progress
.path
, &info
);
107 return end_split_or_merge(in_progress
.path
);
112 if (in_progress
.is_split())
113 return complete_split(in_progress
.path
, info
);
114 else if (in_progress
.is_merge())
115 return complete_merge(in_progress
.path
, info
);
116 else if (in_progress
.is_col_split()) {
117 for (vector
<string
>::iterator i
= in_progress
.path
.begin();
118 i
!= in_progress
.path
.end();
120 vector
<string
> path(in_progress
.path
.begin(), i
);
121 int r
= reset_attr(path
);
131 int HashIndex::reset_attr(
132 const vector
<string
> &path
)
135 int r
= path_exists(path
, &exists
);
140 map
<string
, ghobject_t
> objects
;
141 vector
<string
> subdirs
;
142 r
= list_objects(path
, 0, 0, &objects
);
145 r
= list_subdirs(path
, &subdirs
);
150 info
.hash_level
= path
.size();
151 info
.objs
= objects
.size();
152 info
.subdirs
= subdirs
.size();
153 return set_info(path
, info
);
156 int HashIndex::col_split_level(
159 const vector
<string
> &path
,
164 /* For each subdir, move, recurse, or ignore based on comparing the low order
165 * bits of the hash represented by the subdir path with inbits, match passed
168 vector
<string
> subdirs
;
169 int r
= from
.list_subdirs(path
, &subdirs
);
172 map
<string
, ghobject_t
> objects
;
173 r
= from
.list_objects(path
, 0, 0, &objects
);
178 for (vector
<string
>::iterator i
= subdirs
.begin();
183 vector
<string
> sub_path(path
.begin(), path
.end());
184 sub_path
.push_back(*i
);
185 path_to_hobject_hash_prefix(sub_path
, &bits
, &hash
);
187 if (hobject_t::match_hash(hash
, bits
, match
)) {
197 if (*mkdirred
> path
.size())
198 *mkdirred
= path
.size();
199 } // else, skip, doesn't need to be moved or recursed into
201 if (hobject_t::match_hash(hash
, inbits
, match
)) {
204 } // else, skip, doesn't need to be moved or recursed into
207 /* Then, do the same for each object */
208 map
<string
, ghobject_t
> objs_to_move
;
209 for (map
<string
, ghobject_t
>::iterator i
= objects
.begin();
212 if (i
->second
.match(inbits
, match
)) {
213 objs_to_move
.insert(*i
);
217 if (objs_to_move
.empty() && to_move
.empty())
220 // Make parent directories as needed
221 while (*mkdirred
< path
.size()) {
224 vector
<string
> creating_path(path
.begin(), path
.begin()+*mkdirred
);
225 r
= to
.path_exists(creating_path
, &exists
);
233 info
.hash_level
= creating_path
.size();
234 if (*mkdirred
< path
.size() - 1)
236 r
= to
.start_col_split(creating_path
);
239 r
= to
.create_path(creating_path
);
242 r
= to
.set_info(creating_path
, info
);
245 r
= to
.end_split_or_merge(creating_path
);
250 subdir_info_s from_info
;
251 subdir_info_s to_info
;
252 r
= from
.get_info(path
, &from_info
);
255 r
= to
.get_info(path
, &to_info
);
259 from
.start_col_split(path
);
260 to
.start_col_split(path
);
263 for (set
<string
>::iterator i
= to_move
.begin();
268 r
= move_subdir(from
, to
, path
, *i
);
273 for (map
<string
, ghobject_t
>::iterator i
= objs_to_move
.begin();
274 i
!= objs_to_move
.end();
278 r
= move_object(from
, to
, path
, *i
);
284 r
= to
.set_info(path
, to_info
);
287 r
= from
.set_info(path
, from_info
);
290 from
.end_split_or_merge(path
);
291 to
.end_split_or_merge(path
);
295 int HashIndex::_merge(
297 CollectionIndex
* dest
) {
298 dout(20) << __func__
<< " bits " << bits
<< dendl
;
299 ceph_assert(collection_version() == dest
->collection_version());
301 vector
<string
> emptypath
;
303 // pre-split to common/target level so that any shared prefix DIR_?
304 // directories already exist at the destination. Since each
305 // directory is a nibble (4 bits),
306 unsigned shared
= bits
/ 4;
307 dout(20) << __func__
<< " pre-splitting to shared level " << shared
<< dendl
;
309 split_dirs(emptypath
, shared
);
310 ((HashIndex
*)dest
)->split_dirs(emptypath
, shared
);
313 // now merge the contents
314 _merge_dirs(*this, *(HashIndex
*)dest
, emptypath
);
319 int HashIndex::_merge_dirs(
322 const vector
<string
>& path
)
324 dout(20) << __func__
<< " path " << path
<< dendl
;
327 vector
<string
> src_subs
, dst_subs
;
328 r
= from
.list_subdirs(path
, &src_subs
);
330 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
331 << " r " << r
<< " from "
332 << "from.list_subdirs"
336 r
= to
.list_subdirs(path
, &dst_subs
);
338 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
339 << " r " << r
<< " from "
345 for (auto& i
: src_subs
) {
346 if (std::find(dst_subs
.begin(), dst_subs
.end(), i
) == dst_subs
.end()) {
348 r
= move_subdir(from
, to
, path
, i
);
350 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
351 << " r " << r
<< " from "
352 << "move_subdir(...,"
353 << path
<< "," << i
<< ")"
359 vector
<string
> nested
= path
;
361 r
= _merge_dirs(from
, to
, nested
);
363 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
364 << " r " << r
<< " from "
371 r
= remove_path(nested
);
373 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
374 << " r " << r
<< " from "
384 map
<string
, ghobject_t
> objects
;
385 r
= from
.list_objects(path
, 0, 0, &objects
);
387 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
388 << " r " << r
<< " from "
389 << "from.list_objects"
394 for (auto& i
: objects
) {
395 r
= move_object(from
, to
, path
, i
);
397 lgeneric_subdout(g_ceph_context
,filestore
,20) << __func__
398 << " r " << r
<< " from "
399 << "move_object(...,"
400 << path
<< "," << i
<< ")"
410 int HashIndex::_split(
413 CollectionIndex
* dest
) {
414 ceph_assert(collection_version() == dest
->collection_version());
415 unsigned mkdirred
= 0;
417 return col_split_level(
419 *static_cast<HashIndex
*>(dest
),
426 int HashIndex::split_dirs(const vector
<string
> &path
, int target_level
) {
427 dout(20) << __func__
<< " " << path
<< " target level: "
428 << target_level
<< dendl
;
430 int r
= get_info(path
, &info
);
432 dout(10) << "error looking up info for " << path
<< ": "
433 << cpp_strerror(r
) << dendl
;
437 if (must_split(info
, target_level
)) {
438 dout(1) << __func__
<< " " << path
<< " has " << info
.objs
439 << " objects, " << info
.hash_level
440 << " level, starting split in pg " << coll() << "." << dendl
;
441 r
= initiate_split(path
, info
);
443 dout(10) << "error initiating split on " << path
<< ": "
444 << cpp_strerror(r
) << dendl
;
448 r
= complete_split(path
, info
);
449 dout(1) << __func__
<< " " << path
<< " split completed in pg " << coll() << "."
452 dout(10) << "error completing split on " << path
<< ": "
453 << cpp_strerror(r
) << dendl
;
458 vector
<string
> subdirs
;
459 r
= list_subdirs(path
, &subdirs
);
461 dout(10) << "error listing subdirs of " << path
<< ": "
462 << cpp_strerror(r
) << dendl
;
465 for (vector
<string
>::const_iterator it
= subdirs
.begin();
466 it
!= subdirs
.end(); ++it
) {
467 vector
<string
> subdir_path(path
);
468 subdir_path
.push_back(*it
);
469 r
= split_dirs(subdir_path
, target_level
);
478 int HashIndex::apply_layout_settings(int target_level
) {
480 dout(10) << __func__
<< " split multiple = " << split_multiplier
481 << " merge threshold = " << merge_threshold
482 << " split rand factor = " << cct
->_conf
->filestore_split_rand_factor
483 << " target level = " << target_level
485 int r
= write_settings();
488 return split_dirs(path
, target_level
);
491 int HashIndex::_init() {
494 int r
= set_info(path
, info
);
497 return write_settings();
500 int HashIndex::write_settings() {
501 if (cct
->_conf
->filestore_split_rand_factor
> 0) {
502 settings
.split_rand_factor
= rand() % cct
->_conf
->filestore_split_rand_factor
;
504 settings
.split_rand_factor
= 0;
509 return add_attr_path(path
, SETTINGS_ATTR
, bl
);
512 int HashIndex::read_settings() {
515 int r
= get_attr_path(path
, SETTINGS_ATTR
, bl
);
519 derr
<< __func__
<< " error reading settings: " << cpp_strerror(r
) << dendl
;
522 auto it
= bl
.cbegin();
524 dout(20) << __func__
<< " split_rand_factor = " << settings
.split_rand_factor
<< dendl
;
528 /* LFNIndex virtual method implementations */
529 int HashIndex::_created(const vector
<string
> &path
,
530 const ghobject_t
&oid
,
531 const string
&mangled_name
) {
534 r
= get_info(path
, &info
);
538 r
= set_info(path
, info
);
542 if (must_split(info
)) {
543 dout(1) << __func__
<< " " << path
<< " has " << info
.objs
544 << " objects, starting split in pg " << coll() << "." << dendl
;
545 int r
= initiate_split(path
, info
);
548 r
= complete_split(path
, info
);
549 dout(1) << __func__
<< " " << path
<< " split completed in pg " << coll() << "."
557 int HashIndex::_remove(const vector
<string
> &path
,
558 const ghobject_t
&oid
,
559 const string
&mangled_name
) {
561 r
= remove_object(path
, oid
);
565 r
= get_info(path
, &info
);
569 r
= set_info(path
, info
);
572 if (must_merge(info
)) {
573 r
= initiate_merge(path
, info
);
576 return complete_merge(path
, info
);
582 int HashIndex::_lookup(const ghobject_t
&oid
,
583 vector
<string
> *path
,
584 string
*mangled_name
,
586 vector
<string
> path_comp
;
587 get_path_components(oid
, &path_comp
);
588 vector
<string
>::iterator next
= path_comp
.begin();
591 int r
= path_exists(*path
, &exists
);
600 if (next
== path_comp
.end())
602 path
->push_back(*(next
++));
604 return get_mangled_name(*path
, oid
, mangled_name
, hardlink
);
607 int HashIndex::_collection_list_partial(const ghobject_t
&start
,
608 const ghobject_t
&end
,
610 vector
<ghobject_t
> *ls
,
617 dout(20) << __func__
<< " start:" << start
<< " end:" << end
<< "-" << max_count
<< " ls.size " << ls
->size() << dendl
;
618 return list_by_hash(path
, end
, max_count
, next
, ls
);
621 int HashIndex::prep_delete() {
622 return recursive_remove(vector
<string
>());
625 int HashIndex::_pre_hash_collection(uint32_t pg_num
, uint64_t expected_num_objs
) {
628 subdir_info_s root_info
;
629 // Make sure there is neither objects nor sub-folders
630 // in this collection
631 ret
= get_info(path
, &root_info
);
635 // Do the folder splitting first
636 ret
= pre_split_folder(pg_num
, expected_num_objs
);
639 // Initialize the folder info starting from root
640 return init_split_folder(path
, 0);
643 int HashIndex::pre_split_folder(uint32_t pg_num
, uint64_t expected_num_objs
)
645 // If folder merging is enabled (by setting the threshold positive),
647 if (merge_threshold
> 0)
649 const coll_t c
= coll();
650 // Do not split if the expected number of objects in this collection is zero (by default)
651 if (expected_num_objs
== 0)
654 // Calculate the number of leaf folders (which actually store files)
655 // need to be created
656 const uint64_t objs_per_folder
= ((uint64_t)(abs(merge_threshold
)) * (uint64_t)split_multiplier
+ settings
.split_rand_factor
) * 16;
657 uint64_t leavies
= expected_num_objs
/ objs_per_folder
;
659 if (leavies
== 0 || expected_num_objs
== objs_per_folder
)
663 if (!c
.is_pg_prefix(&spgid
))
665 const ps_t ps
= spgid
.pgid
.ps();
667 // the most significant bits of pg_num
668 const int pg_num_bits
= calc_num_bits(pg_num
- 1);
670 // calculate the number of levels we only create one sub folder
671 int num
= pg_num_bits
/ 4;
672 // pg num's hex value is like 1xxx,xxxx,xxxx but not 1111,1111,1111,
673 // so that splitting starts at level 3
674 if (pg_num_bits
% 4 == 0 && pg_num
< ((uint32_t)1 << pg_num_bits
)) {
679 // Start with creation that only has one subfolder
680 vector
<string
> paths
;
683 ps_t v
= tmp_id
& 0x0000000f;
684 paths
.push_back(to_hex(v
));
685 ret
= create_path(paths
);
686 if (ret
< 0 && ret
!= -EEXIST
)
688 tmp_id
= tmp_id
>> 4;
691 // Starting from here, we can split by creating multiple subfolders
692 const int left_bits
= pg_num_bits
- dump_num
* 4;
693 // this variable denotes how many bits (for this level) that can be
694 // used for sub folder splitting
695 int split_bits
= 4 - left_bits
;
696 // the below logic is inspired by rados.h#ceph_stable_mod,
697 // it basically determines how many sub-folders should we
698 // create for splitting
699 ceph_assert(pg_num_bits
> 0); // otherwise BAD_SHIFT
700 if (((1 << (pg_num_bits
- 1)) | ps
) >= pg_num
) {
703 const uint32_t subs
= (1 << split_bits
);
704 // Calculate how many levels we create starting from here
706 int level_limit
= MAX_HASH_LEVEL
- dump_num
- 1;
707 uint64_t actual_leaves
= subs
;
708 while (actual_leaves
< leavies
&& level
< level_limit
) {
712 for (uint32_t i
= 0; i
< subs
; ++i
) {
713 ceph_assert(split_bits
<= 4); // otherwise BAD_SHIFT
714 int v
= tmp_id
| (i
<< ((4 - split_bits
) % 4));
715 paths
.push_back(to_hex(v
));
716 ret
= create_path(paths
);
717 if (ret
< 0 && ret
!= -EEXIST
)
719 ret
= recursive_create_path(paths
, level
);
727 int HashIndex::init_split_folder(vector
<string
> &path
, uint32_t hash_level
)
729 // Get the number of sub directories for the current path
730 vector
<string
> subdirs
;
731 int ret
= list_subdirs(path
, &subdirs
);
735 info
.subdirs
= subdirs
.size();
736 info
.hash_level
= hash_level
;
737 ret
= set_info(path
, info
);
740 ret
= fsync_dir(path
);
744 // Do the same for subdirs
745 vector
<string
>::const_iterator iter
;
746 for (iter
= subdirs
.begin(); iter
!= subdirs
.end(); ++iter
) {
747 path
.push_back(*iter
);
748 ret
= init_split_folder(path
, hash_level
+ 1);
756 int HashIndex::recursive_create_path(vector
<string
>& path
, int level
)
760 for (int i
= 0; i
< 16; ++i
) {
761 path
.push_back(to_hex(i
));
762 int ret
= create_path(path
);
763 if (ret
< 0 && ret
!= -EEXIST
)
765 ret
= recursive_create_path(path
, level
- 1);
773 int HashIndex::recursive_remove(const vector
<string
> &path
) {
774 return _recursive_remove(path
, true);
777 int HashIndex::_recursive_remove(const vector
<string
> &path
, bool top
) {
778 vector
<string
> subdirs
;
779 dout(20) << __func__
<< " path=" << path
<< dendl
;
780 int r
= list_subdirs(path
, &subdirs
);
783 map
<string
, ghobject_t
> objects
;
784 r
= list_objects(path
, 0, 0, &objects
);
787 if (!objects
.empty())
789 vector
<string
> subdir(path
);
790 for (vector
<string
>::iterator i
= subdirs
.begin();
793 subdir
.push_back(*i
);
794 r
= _recursive_remove(subdir
, false);
802 return remove_path(path
);
805 int HashIndex::start_col_split(const vector
<string
> &path
) {
807 InProgressOp
op_tag(InProgressOp::COL_SPLIT
, path
);
809 int r
= add_attr_path(vector
<string
>(), IN_PROGRESS_OP_TAG
, bl
);
812 return fsync_dir(vector
<string
>());
815 int HashIndex::start_split(const vector
<string
> &path
) {
817 InProgressOp
op_tag(InProgressOp::SPLIT
, path
);
819 int r
= add_attr_path(vector
<string
>(), IN_PROGRESS_OP_TAG
, bl
);
822 return fsync_dir(vector
<string
>());
825 int HashIndex::start_merge(const vector
<string
> &path
) {
827 InProgressOp
op_tag(InProgressOp::MERGE
, path
);
829 int r
= add_attr_path(vector
<string
>(), IN_PROGRESS_OP_TAG
, bl
);
832 return fsync_dir(vector
<string
>());
835 int HashIndex::end_split_or_merge(const vector
<string
> &path
) {
836 return remove_attr_path(vector
<string
>(), IN_PROGRESS_OP_TAG
);
839 int HashIndex::get_info(const vector
<string
> &path
, subdir_info_s
*info
) {
841 int r
= get_attr_path(path
, SUBDIR_ATTR
, buf
);
844 auto bufiter
= buf
.cbegin();
845 info
->decode(bufiter
);
846 ceph_assert(path
.size() == (unsigned)info
->hash_level
);
850 int HashIndex::set_info(const vector
<string
> &path
, const subdir_info_s
&info
) {
852 ceph_assert(path
.size() == (unsigned)info
.hash_level
);
854 return add_attr_path(path
, SUBDIR_ATTR
, buf
);
857 bool HashIndex::must_merge(const subdir_info_s
&info
) {
858 return (info
.hash_level
> 0 &&
859 merge_threshold
> 0 &&
860 info
.objs
< (unsigned)merge_threshold
&&
864 bool HashIndex::must_split(const subdir_info_s
&info
, int target_level
) {
865 // target_level is used for ceph-objectstore-tool to split dirs offline.
866 // if it is set (defalult is 0) and current hash level < target_level,
867 // this dir would be split no matters how many objects it has.
868 return (info
.hash_level
< (unsigned)MAX_HASH_LEVEL
&&
869 ((target_level
> 0 && info
.hash_level
< (unsigned)target_level
) ||
870 (info
.objs
> ((unsigned)(abs(merge_threshold
) * split_multiplier
+ settings
.split_rand_factor
) * 16))));
873 int HashIndex::initiate_merge(const vector
<string
> &path
, subdir_info_s info
) {
874 return start_merge(path
);
877 int HashIndex::complete_merge(const vector
<string
> &path
, subdir_info_s info
) {
878 vector
<string
> dst
= path
;
880 subdir_info_s dstinfo
;
882 r
= path_exists(path
, &exists
);
885 r
= get_info(dst
, &dstinfo
);
889 r
= move_objects(path
, dst
);
895 r
= remove_path(path
);
899 if (must_merge(dstinfo
)) {
900 r
= initiate_merge(dst
, dstinfo
);
906 return complete_merge(dst
, dstinfo
);
911 return end_split_or_merge(path
);
914 int HashIndex::initiate_split(const vector
<string
> &path
, subdir_info_s info
) {
915 return start_split(path
);
918 int HashIndex::complete_split(const vector
<string
> &path
, subdir_info_s info
) {
919 int level
= info
.hash_level
;
920 map
<string
, ghobject_t
> objects
;
921 vector
<string
> dst
= path
;
924 r
= list_objects(path
, 0, 0, &objects
);
927 vector
<string
> subdirs_vec
;
928 r
= list_subdirs(path
, &subdirs_vec
);
932 subdirs
.insert(subdirs_vec
.begin(), subdirs_vec
.end());
933 map
<string
, map
<string
, ghobject_t
> > mapped
;
934 map
<string
, ghobject_t
> moved
;
936 for (map
<string
, ghobject_t
>::iterator i
= objects
.begin();
939 vector
<string
> new_path
;
940 get_path_components(i
->second
, &new_path
);
941 mapped
[new_path
[level
]][i
->first
] = i
->second
;
943 for (map
<string
, map
<string
, ghobject_t
> >::iterator i
= mapped
.begin();
946 dst
[level
] = i
->first
;
947 /* If the info already exists, it must be correct,
948 * we may be picking up a partially finished split */
950 // subdir has already been fully copied
951 if (subdirs
.count(i
->first
) && !get_info(dst
, &temp
)) {
952 for (map
<string
, ghobject_t
>::iterator j
= i
->second
.begin();
953 j
!= i
->second
.end();
955 moved
[j
->first
] = j
->second
;
957 objects
.erase(j
->first
);
963 subdir_info_s info_new
;
964 info_new
.objs
= i
->second
.size();
965 info_new
.subdirs
= 0;
966 info_new
.hash_level
= level
+ 1;
967 if (must_merge(info_new
) && !subdirs
.count(i
->first
)) {
972 // Subdir doesn't yet exist
973 if (!subdirs
.count(i
->first
)) {
975 r
= create_path(dst
);
978 } // else subdir has been created but only partially copied
980 for (map
<string
, ghobject_t
>::iterator j
= i
->second
.begin();
981 j
!= i
->second
.end();
983 moved
[j
->first
] = j
->second
;
985 objects
.erase(j
->first
);
986 r
= link_object(path
, dst
, j
->second
, j
->first
);
987 // May be a partially finished split
988 if (r
< 0 && r
!= -EEXIST
) {
997 // Presence of info must imply that all objects have been copied
998 r
= set_info(dst
, info_new
);
1008 r
= remove_objects(path
, moved
, &objects
);
1011 info
.objs
= objects
.size();
1012 r
= reset_attr(path
);
1015 r
= fsync_dir(path
);
1018 return end_split_or_merge(path
);
1021 void HashIndex::get_path_components(const ghobject_t
&oid
,
1022 vector
<string
> *path
) {
1023 char buf
[MAX_HASH_LEVEL
+ 1];
1024 snprintf(buf
, sizeof(buf
), "%.*X", MAX_HASH_LEVEL
, (uint32_t)oid
.hobj
.get_nibblewise_key());
1026 // Path components are the hex characters of oid.hobj.hash, least
1027 // significant first
1028 for (int i
= 0; i
< MAX_HASH_LEVEL
; ++i
) {
1029 path
->push_back(string(&buf
[i
], 1));
1033 string
HashIndex::get_hash_str(uint32_t hash
) {
1034 char buf
[MAX_HASH_LEVEL
+ 1];
1035 snprintf(buf
, sizeof(buf
), "%.*X", MAX_HASH_LEVEL
, hash
);
1037 for (int i
= 0; i
< MAX_HASH_LEVEL
; ++i
) {
1038 retval
.push_back(buf
[MAX_HASH_LEVEL
- 1 - i
]);
1043 string
HashIndex::get_path_str(const ghobject_t
&oid
) {
1044 ceph_assert(!oid
.is_max());
1045 return get_hash_str(oid
.hobj
.get_hash());
1048 uint32_t HashIndex::hash_prefix_to_hash(string prefix
) {
1049 while (prefix
.size() < sizeof(uint32_t) * 2) {
1050 prefix
.push_back('0');
1053 sscanf(prefix
.c_str(), "%x", &hash
);
1055 hash
= ((hash
& 0x0f0f0f0f) << 4) | ((hash
& 0xf0f0f0f0) >> 4);
1056 hash
= ((hash
& 0x00ff00ff) << 8) | ((hash
& 0xff00ff00) >> 8);
1057 hash
= ((hash
& 0x0000ffff) << 16) | ((hash
& 0xffff0000) >> 16);
1061 int HashIndex::get_path_contents_by_hash_bitwise(
1062 const vector
<string
> &path
,
1063 const ghobject_t
*next_object
,
1064 set
<string
, CmpHexdigitStringBitwise
> *hash_prefixes
,
1065 set
<pair
<string
, ghobject_t
>, CmpPairBitwise
> *objects
)
1067 map
<string
, ghobject_t
> rev_objects
;
1069 r
= list_objects(path
, 0, 0, &rev_objects
);
1073 for (map
<string
, ghobject_t
>::iterator i
= rev_objects
.begin();
1074 i
!= rev_objects
.end();
1076 if (next_object
&& i
->second
< *next_object
)
1078 string hash_prefix
= get_path_str(i
->second
);
1079 hash_prefixes
->insert(hash_prefix
);
1080 objects
->insert(pair
<string
, ghobject_t
>(hash_prefix
, i
->second
));
1082 vector
<string
> subdirs
;
1083 r
= list_subdirs(path
, &subdirs
);
1087 // sort subdirs bitwise (by reversing hex digit nibbles)
1088 std::sort(subdirs
.begin(), subdirs
.end(), cmp_hexdigit_bitwise
);
1090 // Local to this function, we will convert the prefix strings
1091 // (previously simply the reversed hex digits) to also have each
1092 // digit's nibbles reversed. This will make the strings sort
1095 for (vector
<string
>::const_iterator i
= path
.begin();
1098 cur_prefix
.append(reverse_hexdigit_bits_string(*i
));
1100 string next_object_string
;
1102 next_object_string
= reverse_hexdigit_bits_string(get_path_str(*next_object
));
1103 for (vector
<string
>::iterator i
= subdirs
.begin();
1106 string candidate
= cur_prefix
+ reverse_hexdigit_bits_string(*i
);
1108 if (next_object
->is_max())
1110 if (candidate
< next_object_string
.substr(0, candidate
.size()))
1113 // re-reverse the hex digit nibbles for the caller
1114 hash_prefixes
->insert(reverse_hexdigit_bits_string(candidate
));
1119 int HashIndex::list_by_hash(const vector
<string
> &path
,
1120 const ghobject_t
&end
,
1123 vector
<ghobject_t
> *out
)
1126 return list_by_hash_bitwise(path
, end
, max_count
, next
, out
);
1129 int HashIndex::list_by_hash_bitwise(
1130 const vector
<string
> &path
,
1131 const ghobject_t
& end
,
1134 vector
<ghobject_t
> *out
)
1136 vector
<string
> next_path
= path
;
1137 next_path
.push_back("");
1138 set
<string
, CmpHexdigitStringBitwise
> hash_prefixes
;
1139 set
<pair
<string
, ghobject_t
>, CmpPairBitwise
> objects
;
1140 int r
= get_path_contents_by_hash_bitwise(path
,
1146 for (set
<string
, CmpHexdigitStringBitwise
>::iterator i
= hash_prefixes
.begin();
1147 i
!= hash_prefixes
.end();
1149 dout(20) << __func__
<< " prefix " << *i
<< dendl
;
1150 set
<pair
<string
, ghobject_t
>, CmpPairBitwise
>::iterator j
= objects
.lower_bound(
1151 make_pair(*i
, ghobject_t()));
1152 if (j
== objects
.end() || j
->first
!= *i
) {
1153 *(next_path
.rbegin()) = *(i
->rbegin());
1154 ghobject_t next_recurse
;
1156 next_recurse
= *next
;
1157 r
= list_by_hash_bitwise(next_path
,
1165 if (!next_recurse
.is_max()) {
1167 *next
= next_recurse
;
1171 while (j
!= objects
.end() && j
->first
== *i
) {
1172 if (max_count
> 0 && out
->size() == (unsigned)max_count
) {
1177 if (j
->second
>= end
) {
1182 if (!next
|| j
->second
>= *next
) {
1183 dout(20) << __func__
<< " prefix " << *i
<< " ob " << j
->second
<< dendl
;
1184 out
->push_back(j
->second
);
1191 *next
= ghobject_t::get_max();