1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Node local data allocation
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/bitops.h>
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
38 #include "blockcheck.h"
42 #include "localalloc.h"
47 #include "buffer_head_io.h"
49 #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
51 static u32
ocfs2_local_alloc_count_bits(struct ocfs2_dinode
*alloc
);
53 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super
*osb
,
54 struct ocfs2_dinode
*alloc
,
56 struct ocfs2_alloc_reservation
*resv
);
58 static void ocfs2_clear_local_alloc(struct ocfs2_dinode
*alloc
);
60 static int ocfs2_sync_local_to_main(struct ocfs2_super
*osb
,
62 struct ocfs2_dinode
*alloc
,
63 struct inode
*main_bm_inode
,
64 struct buffer_head
*main_bm_bh
);
66 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super
*osb
,
67 struct ocfs2_alloc_context
**ac
,
68 struct inode
**bitmap_inode
,
69 struct buffer_head
**bitmap_bh
);
71 static int ocfs2_local_alloc_new_window(struct ocfs2_super
*osb
,
73 struct ocfs2_alloc_context
*ac
);
75 static int ocfs2_local_alloc_slide_window(struct ocfs2_super
*osb
,
76 struct inode
*local_alloc_inode
);
79 * ocfs2_la_default_mb() - determine a default size, in megabytes of
82 * Generally, we'd like to pick as large a local alloc as
83 * possible. Performance on large workloads tends to scale
84 * proportionally to la size. In addition to that, the reservations
85 * code functions more efficiently as it can reserve more windows for
88 * Some things work against us when trying to choose a large local alloc:
90 * - We need to ensure our sizing is picked to leave enough space in
91 * group descriptors for other allocations (such as block groups,
92 * etc). Picking default sizes which are a multiple of 4 could help
93 * - block groups are allocated in 2mb and 4mb chunks.
95 * - Likewise, we don't want to starve other nodes of bits on small
96 * file systems. This can easily be taken care of by limiting our
97 * default to a reasonable size (256M) on larger cluster sizes.
99 * - Some file systems can't support very large sizes - 4k and 8k in
100 * particular are limited to less than 128 and 256 megabytes respectively.
102 * The following reference table shows group descriptor and local
103 * alloc maximums at various cluster sizes (4k blocksize)
105 * csize: 4K group: 126M la: 121M
106 * csize: 8K group: 252M la: 243M
107 * csize: 16K group: 504M la: 486M
108 * csize: 32K group: 1008M la: 972M
109 * csize: 64K group: 2016M la: 1944M
110 * csize: 128K group: 4032M la: 3888M
111 * csize: 256K group: 8064M la: 7776M
112 * csize: 512K group: 16128M la: 15552M
113 * csize: 1024K group: 32256M la: 31104M
115 #define OCFS2_LA_MAX_DEFAULT_MB 256
116 #define OCFS2_LA_OLD_DEFAULT 8
117 unsigned int ocfs2_la_default_mb(struct ocfs2_super
*osb
)
121 unsigned int megs_per_slot
;
122 struct super_block
*sb
= osb
->sb
;
124 gd_mb
= ocfs2_clusters_to_megabytes(osb
->sb
,
125 8 * ocfs2_group_bitmap_size(sb
, 0, osb
->s_feature_incompat
));
128 * This takes care of files systems with very small group
129 * descriptors - 512 byte blocksize at cluster sizes lower
130 * than 16K and also 1k blocksize with 4k cluster size.
132 if ((sb
->s_blocksize
== 512 && osb
->s_clustersize
<= 8192)
133 || (sb
->s_blocksize
== 1024 && osb
->s_clustersize
== 4096))
134 return OCFS2_LA_OLD_DEFAULT
;
137 * Leave enough room for some block groups and make the final
138 * value we work from a multiple of 4.
146 * Keep window sizes down to a reasonable default
148 if (la_mb
> OCFS2_LA_MAX_DEFAULT_MB
) {
150 * Some clustersize / blocksize combinations will have
151 * given us a larger than OCFS2_LA_MAX_DEFAULT_MB
152 * default size, but get poor distribution when
153 * limited to exactly 256 megabytes.
155 * As an example, 16K clustersize at 4K blocksize
156 * gives us a cluster group size of 504M. Paring the
157 * local alloc size down to 256 however, would give us
158 * only one window and around 200MB left in the
159 * cluster group. Instead, find the first size below
160 * 256 which would give us an even distribution.
162 * Larger cluster group sizes actually work out pretty
163 * well when pared to 256, so we don't have to do this
164 * for any group that fits more than two
165 * OCFS2_LA_MAX_DEFAULT_MB windows.
167 if (gd_mb
> (2 * OCFS2_LA_MAX_DEFAULT_MB
))
170 unsigned int gd_mult
= gd_mb
;
172 while (gd_mult
> 256)
173 gd_mult
= gd_mult
>> 1;
179 megs_per_slot
= osb
->osb_clusters_at_boot
/ osb
->max_slots
;
180 megs_per_slot
= ocfs2_clusters_to_megabytes(osb
->sb
, megs_per_slot
);
181 /* Too many nodes, too few disk clusters. */
182 if (megs_per_slot
< la_mb
)
183 la_mb
= megs_per_slot
;
188 void ocfs2_la_set_sizes(struct ocfs2_super
*osb
, int requested_mb
)
190 struct super_block
*sb
= osb
->sb
;
191 unsigned int la_default_mb
= ocfs2_la_default_mb(osb
);
192 unsigned int la_max_mb
;
194 la_max_mb
= ocfs2_clusters_to_megabytes(sb
,
195 ocfs2_local_alloc_size(sb
) * 8);
197 mlog(0, "requested: %dM, max: %uM, default: %uM\n",
198 requested_mb
, la_max_mb
, la_default_mb
);
200 if (requested_mb
== -1) {
201 /* No user request - use defaults */
202 osb
->local_alloc_default_bits
=
203 ocfs2_megabytes_to_clusters(sb
, la_default_mb
);
204 } else if (requested_mb
> la_max_mb
) {
205 /* Request is too big, we give the maximum available */
206 osb
->local_alloc_default_bits
=
207 ocfs2_megabytes_to_clusters(sb
, la_max_mb
);
209 osb
->local_alloc_default_bits
=
210 ocfs2_megabytes_to_clusters(sb
, requested_mb
);
213 osb
->local_alloc_bits
= osb
->local_alloc_default_bits
;
216 static inline int ocfs2_la_state_enabled(struct ocfs2_super
*osb
)
218 return (osb
->local_alloc_state
== OCFS2_LA_THROTTLED
||
219 osb
->local_alloc_state
== OCFS2_LA_ENABLED
);
222 void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super
*osb
,
223 unsigned int num_clusters
)
225 spin_lock(&osb
->osb_lock
);
226 if (osb
->local_alloc_state
== OCFS2_LA_DISABLED
||
227 osb
->local_alloc_state
== OCFS2_LA_THROTTLED
)
228 if (num_clusters
>= osb
->local_alloc_default_bits
) {
229 cancel_delayed_work(&osb
->la_enable_wq
);
230 osb
->local_alloc_state
= OCFS2_LA_ENABLED
;
232 spin_unlock(&osb
->osb_lock
);
235 void ocfs2_la_enable_worker(struct work_struct
*work
)
237 struct ocfs2_super
*osb
=
238 container_of(work
, struct ocfs2_super
,
240 spin_lock(&osb
->osb_lock
);
241 osb
->local_alloc_state
= OCFS2_LA_ENABLED
;
242 spin_unlock(&osb
->osb_lock
);
246 * Tell us whether a given allocation should use the local alloc
247 * file. Otherwise, it has to go to the main bitmap.
249 * This function does semi-dirty reads of local alloc size and state!
250 * This is ok however, as the values are re-checked once under mutex.
252 int ocfs2_alloc_should_use_local(struct ocfs2_super
*osb
, u64 bits
)
257 spin_lock(&osb
->osb_lock
);
258 la_bits
= osb
->local_alloc_bits
;
260 if (!ocfs2_la_state_enabled(osb
))
263 /* la_bits should be at least twice the size (in clusters) of
264 * a new block group. We want to be sure block group
265 * allocations go through the local alloc, so allow an
266 * allocation to take up to half the bitmap. */
267 if (bits
> (la_bits
/ 2))
272 mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
273 osb
->local_alloc_state
, (unsigned long long)bits
, la_bits
, ret
);
274 spin_unlock(&osb
->osb_lock
);
278 int ocfs2_load_local_alloc(struct ocfs2_super
*osb
)
281 struct ocfs2_dinode
*alloc
= NULL
;
282 struct buffer_head
*alloc_bh
= NULL
;
284 struct inode
*inode
= NULL
;
285 struct ocfs2_local_alloc
*la
;
289 if (osb
->local_alloc_bits
== 0)
292 if (osb
->local_alloc_bits
>= osb
->bitmap_cpg
) {
293 mlog(ML_NOTICE
, "Requested local alloc window %d is larger "
294 "than max possible %u. Using defaults.\n",
295 osb
->local_alloc_bits
, (osb
->bitmap_cpg
- 1));
296 osb
->local_alloc_bits
=
297 ocfs2_megabytes_to_clusters(osb
->sb
,
298 ocfs2_la_default_mb(osb
));
301 /* read the alloc off disk */
302 inode
= ocfs2_get_system_file_inode(osb
, LOCAL_ALLOC_SYSTEM_INODE
,
310 status
= ocfs2_read_inode_block_full(inode
, &alloc_bh
,
311 OCFS2_BH_IGNORE_CACHE
);
317 alloc
= (struct ocfs2_dinode
*) alloc_bh
->b_data
;
318 la
= OCFS2_LOCAL_ALLOC(alloc
);
320 if (!(le32_to_cpu(alloc
->i_flags
) &
321 (OCFS2_LOCAL_ALLOC_FL
|OCFS2_BITMAP_FL
))) {
322 mlog(ML_ERROR
, "Invalid local alloc inode, %llu\n",
323 (unsigned long long)OCFS2_I(inode
)->ip_blkno
);
328 if ((la
->la_size
== 0) ||
329 (le16_to_cpu(la
->la_size
) > ocfs2_local_alloc_size(inode
->i_sb
))) {
330 mlog(ML_ERROR
, "Local alloc size is invalid (la_size = %u)\n",
331 le16_to_cpu(la
->la_size
));
336 /* do a little verification. */
337 num_used
= ocfs2_local_alloc_count_bits(alloc
);
339 /* hopefully the local alloc has always been recovered before
342 || alloc
->id1
.bitmap1
.i_used
343 || alloc
->id1
.bitmap1
.i_total
345 mlog(ML_ERROR
, "Local alloc hasn't been recovered!\n"
346 "found = %u, set = %u, taken = %u, off = %u\n",
347 num_used
, le32_to_cpu(alloc
->id1
.bitmap1
.i_used
),
348 le32_to_cpu(alloc
->id1
.bitmap1
.i_total
),
349 OCFS2_LOCAL_ALLOC(alloc
)->la_bm_off
);
351 osb
->local_alloc_bh
= alloc_bh
;
352 osb
->local_alloc_state
= OCFS2_LA_ENABLED
;
360 mlog(0, "Local alloc window bits = %d\n", osb
->local_alloc_bits
);
367 * return any unused bits to the bitmap and write out a clean
370 * local_alloc_bh is optional. If not passed, we will simply use the
371 * one off osb. If you do pass it however, be warned that it *will* be
372 * returned brelse'd and NULL'd out.*/
373 void ocfs2_shutdown_local_alloc(struct ocfs2_super
*osb
)
377 struct inode
*local_alloc_inode
= NULL
;
378 struct buffer_head
*bh
= NULL
;
379 struct buffer_head
*main_bm_bh
= NULL
;
380 struct inode
*main_bm_inode
= NULL
;
381 struct ocfs2_dinode
*alloc_copy
= NULL
;
382 struct ocfs2_dinode
*alloc
= NULL
;
386 cancel_delayed_work(&osb
->la_enable_wq
);
387 flush_workqueue(ocfs2_wq
);
389 if (osb
->local_alloc_state
== OCFS2_LA_UNUSED
)
393 ocfs2_get_system_file_inode(osb
,
394 LOCAL_ALLOC_SYSTEM_INODE
,
396 if (!local_alloc_inode
) {
402 osb
->local_alloc_state
= OCFS2_LA_DISABLED
;
404 ocfs2_resmap_uninit(&osb
->osb_la_resmap
);
406 main_bm_inode
= ocfs2_get_system_file_inode(osb
,
407 GLOBAL_BITMAP_SYSTEM_INODE
,
409 if (!main_bm_inode
) {
415 mutex_lock(&main_bm_inode
->i_mutex
);
417 status
= ocfs2_inode_lock(main_bm_inode
, &main_bm_bh
, 1);
423 /* WINDOW_MOVE_CREDITS is a bit heavy... */
424 handle
= ocfs2_start_trans(osb
, OCFS2_WINDOW_MOVE_CREDITS
);
425 if (IS_ERR(handle
)) {
426 mlog_errno(PTR_ERR(handle
));
431 bh
= osb
->local_alloc_bh
;
432 alloc
= (struct ocfs2_dinode
*) bh
->b_data
;
434 alloc_copy
= kmalloc(bh
->b_size
, GFP_NOFS
);
439 memcpy(alloc_copy
, alloc
, bh
->b_size
);
441 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(local_alloc_inode
),
442 bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
448 ocfs2_clear_local_alloc(alloc
);
449 ocfs2_journal_dirty(handle
, bh
);
452 osb
->local_alloc_bh
= NULL
;
453 osb
->local_alloc_state
= OCFS2_LA_UNUSED
;
455 status
= ocfs2_sync_local_to_main(osb
, handle
, alloc_copy
,
456 main_bm_inode
, main_bm_bh
);
461 ocfs2_commit_trans(osb
, handle
);
466 ocfs2_inode_unlock(main_bm_inode
, 1);
469 mutex_unlock(&main_bm_inode
->i_mutex
);
473 if (local_alloc_inode
)
474 iput(local_alloc_inode
);
483 * We want to free the bitmap bits outside of any recovery context as
484 * we'll need a cluster lock to do so, but we must clear the local
485 * alloc before giving up the recovered nodes journal. To solve this,
486 * we kmalloc a copy of the local alloc before it's change for the
487 * caller to process with ocfs2_complete_local_alloc_recovery
489 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super
*osb
,
491 struct ocfs2_dinode
**alloc_copy
)
494 struct buffer_head
*alloc_bh
= NULL
;
495 struct inode
*inode
= NULL
;
496 struct ocfs2_dinode
*alloc
;
498 mlog_entry("(slot_num = %d)\n", slot_num
);
502 inode
= ocfs2_get_system_file_inode(osb
,
503 LOCAL_ALLOC_SYSTEM_INODE
,
511 mutex_lock(&inode
->i_mutex
);
513 status
= ocfs2_read_inode_block_full(inode
, &alloc_bh
,
514 OCFS2_BH_IGNORE_CACHE
);
520 *alloc_copy
= kmalloc(alloc_bh
->b_size
, GFP_KERNEL
);
521 if (!(*alloc_copy
)) {
525 memcpy((*alloc_copy
), alloc_bh
->b_data
, alloc_bh
->b_size
);
527 alloc
= (struct ocfs2_dinode
*) alloc_bh
->b_data
;
528 ocfs2_clear_local_alloc(alloc
);
530 ocfs2_compute_meta_ecc(osb
->sb
, alloc_bh
->b_data
, &alloc
->i_check
);
531 status
= ocfs2_write_block(osb
, alloc_bh
, INODE_CACHE(inode
));
536 if ((status
< 0) && (*alloc_copy
)) {
544 mutex_unlock(&inode
->i_mutex
);
553 * Step 2: By now, we've completed the journal recovery, we've stamped
554 * a clean local alloc on disk and dropped the node out of the
555 * recovery map. Dlm locks will no longer stall, so lets clear out the
558 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super
*osb
,
559 struct ocfs2_dinode
*alloc
)
563 struct buffer_head
*main_bm_bh
= NULL
;
564 struct inode
*main_bm_inode
;
568 main_bm_inode
= ocfs2_get_system_file_inode(osb
,
569 GLOBAL_BITMAP_SYSTEM_INODE
,
571 if (!main_bm_inode
) {
577 mutex_lock(&main_bm_inode
->i_mutex
);
579 status
= ocfs2_inode_lock(main_bm_inode
, &main_bm_bh
, 1);
585 handle
= ocfs2_start_trans(osb
, OCFS2_WINDOW_MOVE_CREDITS
);
586 if (IS_ERR(handle
)) {
587 status
= PTR_ERR(handle
);
593 /* we want the bitmap change to be recorded on disk asap */
596 status
= ocfs2_sync_local_to_main(osb
, handle
, alloc
,
597 main_bm_inode
, main_bm_bh
);
601 ocfs2_commit_trans(osb
, handle
);
604 ocfs2_inode_unlock(main_bm_inode
, 1);
607 mutex_unlock(&main_bm_inode
->i_mutex
);
615 ocfs2_init_steal_slots(osb
);
621 * make sure we've got at least bits_wanted contiguous bits in the
622 * local alloc. You lose them when you drop i_mutex.
624 * We will add ourselves to the transaction passed in, but may start
625 * our own in order to shift windows.
627 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super
*osb
,
629 struct ocfs2_alloc_context
*ac
)
632 struct ocfs2_dinode
*alloc
;
633 struct inode
*local_alloc_inode
;
634 unsigned int free_bits
;
641 ocfs2_get_system_file_inode(osb
,
642 LOCAL_ALLOC_SYSTEM_INODE
,
644 if (!local_alloc_inode
) {
650 mutex_lock(&local_alloc_inode
->i_mutex
);
653 * We must double check state and allocator bits because
654 * another process may have changed them while holding i_mutex.
656 spin_lock(&osb
->osb_lock
);
657 if (!ocfs2_la_state_enabled(osb
) ||
658 (bits_wanted
> osb
->local_alloc_bits
)) {
659 spin_unlock(&osb
->osb_lock
);
663 spin_unlock(&osb
->osb_lock
);
665 alloc
= (struct ocfs2_dinode
*) osb
->local_alloc_bh
->b_data
;
667 #ifdef CONFIG_OCFS2_DEBUG_FS
668 if (le32_to_cpu(alloc
->id1
.bitmap1
.i_used
) !=
669 ocfs2_local_alloc_count_bits(alloc
)) {
670 ocfs2_error(osb
->sb
, "local alloc inode %llu says it has "
671 "%u free bits, but a count shows %u",
672 (unsigned long long)le64_to_cpu(alloc
->i_blkno
),
673 le32_to_cpu(alloc
->id1
.bitmap1
.i_used
),
674 ocfs2_local_alloc_count_bits(alloc
));
680 free_bits
= le32_to_cpu(alloc
->id1
.bitmap1
.i_total
) -
681 le32_to_cpu(alloc
->id1
.bitmap1
.i_used
);
682 if (bits_wanted
> free_bits
) {
683 /* uhoh, window change time. */
685 ocfs2_local_alloc_slide_window(osb
, local_alloc_inode
);
687 if (status
!= -ENOSPC
)
693 * Under certain conditions, the window slide code
694 * might have reduced the number of bits available or
695 * disabled the the local alloc entirely. Re-check
696 * here and return -ENOSPC if necessary.
699 if (!ocfs2_la_state_enabled(osb
))
702 free_bits
= le32_to_cpu(alloc
->id1
.bitmap1
.i_total
) -
703 le32_to_cpu(alloc
->id1
.bitmap1
.i_used
);
704 if (bits_wanted
> free_bits
)
708 if (ac
->ac_max_block
)
709 mlog(0, "Calling in_range for max block %llu\n",
710 (unsigned long long)ac
->ac_max_block
);
712 ac
->ac_inode
= local_alloc_inode
;
713 /* We should never use localalloc from another slot */
714 ac
->ac_alloc_slot
= osb
->slot_num
;
715 ac
->ac_which
= OCFS2_AC_USE_LOCAL
;
716 get_bh(osb
->local_alloc_bh
);
717 ac
->ac_bh
= osb
->local_alloc_bh
;
720 if (status
< 0 && local_alloc_inode
) {
721 mutex_unlock(&local_alloc_inode
->i_mutex
);
722 iput(local_alloc_inode
);
725 mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted
, osb
->slot_num
,
732 int ocfs2_claim_local_alloc_bits(struct ocfs2_super
*osb
,
734 struct ocfs2_alloc_context
*ac
,
740 struct inode
*local_alloc_inode
;
742 struct ocfs2_dinode
*alloc
;
743 struct ocfs2_local_alloc
*la
;
746 BUG_ON(ac
->ac_which
!= OCFS2_AC_USE_LOCAL
);
748 local_alloc_inode
= ac
->ac_inode
;
749 alloc
= (struct ocfs2_dinode
*) osb
->local_alloc_bh
->b_data
;
750 la
= OCFS2_LOCAL_ALLOC(alloc
);
752 start
= ocfs2_local_alloc_find_clear_bits(osb
, alloc
, &bits_wanted
,
755 /* TODO: Shouldn't we just BUG here? */
761 bitmap
= la
->la_bitmap
;
762 *bit_off
= le32_to_cpu(la
->la_bm_off
) + start
;
763 *num_bits
= bits_wanted
;
765 status
= ocfs2_journal_access_di(handle
,
766 INODE_CACHE(local_alloc_inode
),
768 OCFS2_JOURNAL_ACCESS_WRITE
);
774 ocfs2_resmap_claimed_bits(&osb
->osb_la_resmap
, ac
->ac_resv
, start
,
778 ocfs2_set_bit(start
++, bitmap
);
780 le32_add_cpu(&alloc
->id1
.bitmap1
.i_used
, *num_bits
);
781 ocfs2_journal_dirty(handle
, osb
->local_alloc_bh
);
788 static u32
ocfs2_local_alloc_count_bits(struct ocfs2_dinode
*alloc
)
793 struct ocfs2_local_alloc
*la
= OCFS2_LOCAL_ALLOC(alloc
);
797 buffer
= la
->la_bitmap
;
798 for (i
= 0; i
< le16_to_cpu(la
->la_size
); i
++)
799 count
+= hweight8(buffer
[i
]);
805 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super
*osb
,
806 struct ocfs2_dinode
*alloc
,
808 struct ocfs2_alloc_reservation
*resv
)
810 int numfound
, bitoff
, left
, startoff
, lastzero
;
812 struct ocfs2_alloc_reservation r
;
814 struct ocfs2_reservation_map
*resmap
= &osb
->osb_la_resmap
;
816 mlog_entry("(numbits wanted = %u)\n", *numbits
);
818 if (!alloc
->id1
.bitmap1
.i_total
) {
819 mlog(0, "No bits in my window!\n");
826 ocfs2_resv_init_once(&r
);
827 ocfs2_resv_set_type(&r
, OCFS2_RESV_FLAG_TMP
);
832 if (ocfs2_resmap_resv_bits(resmap
, resv
, &bitoff
, &numfound
) == 0) {
833 if (numfound
< *numbits
)
839 * Code error. While reservations are enabled, local
840 * allocation should _always_ go through them.
842 BUG_ON(osb
->osb_resv_level
!= 0);
845 * Reservations are disabled. Handle this the old way.
848 bitmap
= OCFS2_LOCAL_ALLOC(alloc
)->la_bitmap
;
850 numfound
= bitoff
= startoff
= 0;
852 left
= le32_to_cpu(alloc
->id1
.bitmap1
.i_total
);
853 while ((bitoff
= ocfs2_find_next_zero_bit(bitmap
, left
, startoff
)) != -1) {
854 if (bitoff
== left
) {
855 /* mlog(0, "bitoff (%d) == left", bitoff); */
858 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
859 "numfound = %d\n", bitoff, startoff, numfound);*/
861 /* Ok, we found a zero bit... is it contig. or do we
863 if (bitoff
== startoff
) {
864 /* we found a zero */
868 /* got a zero after some ones */
872 /* we got everything we needed */
873 if (numfound
== *numbits
) {
874 /* mlog(0, "Found it all!\n"); */
879 mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff
,
882 if (numfound
== *numbits
)
883 bitoff
= startoff
- numfound
;
889 ocfs2_resv_discard(resmap
, resv
);
895 static void ocfs2_clear_local_alloc(struct ocfs2_dinode
*alloc
)
897 struct ocfs2_local_alloc
*la
= OCFS2_LOCAL_ALLOC(alloc
);
901 alloc
->id1
.bitmap1
.i_total
= 0;
902 alloc
->id1
.bitmap1
.i_used
= 0;
904 for(i
= 0; i
< le16_to_cpu(la
->la_size
); i
++)
905 la
->la_bitmap
[i
] = 0;
911 /* turn this on and uncomment below to aid debugging window shifts. */
912 static void ocfs2_verify_zero_bits(unsigned long *bitmap
,
916 unsigned int tmp
= count
;
918 if (ocfs2_test_bit(start
+ tmp
, bitmap
)) {
919 printk("ocfs2_verify_zero_bits: start = %u, count = "
920 "%u\n", start
, count
);
921 printk("ocfs2_verify_zero_bits: bit %u is set!",
930 * sync the local alloc to main bitmap.
932 * assumes you've already locked the main bitmap -- the bitmap inode
933 * passed is used for caching.
935 static int ocfs2_sync_local_to_main(struct ocfs2_super
*osb
,
937 struct ocfs2_dinode
*alloc
,
938 struct inode
*main_bm_inode
,
939 struct buffer_head
*main_bm_bh
)
942 int bit_off
, left
, count
, start
;
946 struct ocfs2_local_alloc
*la
= OCFS2_LOCAL_ALLOC(alloc
);
948 mlog_entry("total = %u, used = %u\n",
949 le32_to_cpu(alloc
->id1
.bitmap1
.i_total
),
950 le32_to_cpu(alloc
->id1
.bitmap1
.i_used
));
952 if (!alloc
->id1
.bitmap1
.i_total
) {
953 mlog(0, "nothing to sync!\n");
957 if (le32_to_cpu(alloc
->id1
.bitmap1
.i_used
) ==
958 le32_to_cpu(alloc
->id1
.bitmap1
.i_total
)) {
959 mlog(0, "all bits were taken!\n");
963 la_start_blk
= ocfs2_clusters_to_blocks(osb
->sb
,
964 le32_to_cpu(la
->la_bm_off
));
965 bitmap
= la
->la_bitmap
;
966 start
= count
= bit_off
= 0;
967 left
= le32_to_cpu(alloc
->id1
.bitmap1
.i_total
);
969 while ((bit_off
= ocfs2_find_next_zero_bit(bitmap
, left
, start
))
971 if ((bit_off
< left
) && (bit_off
== start
)) {
977 blkno
= la_start_blk
+
978 ocfs2_clusters_to_blocks(osb
->sb
,
981 mlog(0, "freeing %u bits starting at local alloc bit "
982 "%u (la_start_blk = %llu, blkno = %llu)\n",
983 count
, start
- count
,
984 (unsigned long long)la_start_blk
,
985 (unsigned long long)blkno
);
987 status
= ocfs2_release_clusters(handle
,
1007 enum ocfs2_la_event
{
1008 OCFS2_LA_EVENT_SLIDE
, /* Normal window slide. */
1009 OCFS2_LA_EVENT_FRAGMENTED
, /* The global bitmap has
1010 * enough bits theoretically
1011 * free, but a contiguous
1012 * allocation could not be
1014 OCFS2_LA_EVENT_ENOSPC
, /* Global bitmap doesn't have
1015 * enough bits free to satisfy
1018 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
1020 * Given an event, calculate the size of our next local alloc window.
1022 * This should always be called under i_mutex of the local alloc inode
1023 * so that local alloc disabling doesn't race with processes trying to
1024 * use the allocator.
1026 * Returns the state which the local alloc was left in. This value can
1027 * be ignored by some paths.
1029 static int ocfs2_recalc_la_window(struct ocfs2_super
*osb
,
1030 enum ocfs2_la_event event
)
1035 spin_lock(&osb
->osb_lock
);
1036 if (osb
->local_alloc_state
== OCFS2_LA_DISABLED
) {
1037 WARN_ON_ONCE(osb
->local_alloc_state
== OCFS2_LA_DISABLED
);
1042 * ENOSPC and fragmentation are treated similarly for now.
1044 if (event
== OCFS2_LA_EVENT_ENOSPC
||
1045 event
== OCFS2_LA_EVENT_FRAGMENTED
) {
1047 * We ran out of contiguous space in the primary
1048 * bitmap. Drastically reduce the number of bits used
1049 * by local alloc until we have to disable it.
1051 bits
= osb
->local_alloc_bits
>> 1;
1052 if (bits
> ocfs2_megabytes_to_clusters(osb
->sb
, 1)) {
1054 * By setting state to THROTTLED, we'll keep
1055 * the number of local alloc bits used down
1056 * until an event occurs which would give us
1057 * reason to assume the bitmap situation might
1060 osb
->local_alloc_state
= OCFS2_LA_THROTTLED
;
1061 osb
->local_alloc_bits
= bits
;
1063 osb
->local_alloc_state
= OCFS2_LA_DISABLED
;
1065 queue_delayed_work(ocfs2_wq
, &osb
->la_enable_wq
,
1066 OCFS2_LA_ENABLE_INTERVAL
);
1071 * Don't increase the size of the local alloc window until we
1072 * know we might be able to fulfill the request. Otherwise, we
1073 * risk bouncing around the global bitmap during periods of
1076 if (osb
->local_alloc_state
!= OCFS2_LA_THROTTLED
)
1077 osb
->local_alloc_bits
= osb
->local_alloc_default_bits
;
1080 state
= osb
->local_alloc_state
;
1081 spin_unlock(&osb
->osb_lock
);
1086 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super
*osb
,
1087 struct ocfs2_alloc_context
**ac
,
1088 struct inode
**bitmap_inode
,
1089 struct buffer_head
**bitmap_bh
)
1093 *ac
= kzalloc(sizeof(struct ocfs2_alloc_context
), GFP_KERNEL
);
1101 (*ac
)->ac_bits_wanted
= osb
->local_alloc_default_bits
;
1102 status
= ocfs2_reserve_cluster_bitmap_bits(osb
, *ac
);
1103 if (status
== -ENOSPC
) {
1104 if (ocfs2_recalc_la_window(osb
, OCFS2_LA_EVENT_ENOSPC
) ==
1108 ocfs2_free_ac_resource(*ac
);
1109 memset(*ac
, 0, sizeof(struct ocfs2_alloc_context
));
1117 *bitmap_inode
= (*ac
)->ac_inode
;
1118 igrab(*bitmap_inode
);
1119 *bitmap_bh
= (*ac
)->ac_bh
;
1123 if ((status
< 0) && *ac
) {
1124 ocfs2_free_alloc_context(*ac
);
1133 * pass it the bitmap lock in lock_bh if you have it.
1135 static int ocfs2_local_alloc_new_window(struct ocfs2_super
*osb
,
1137 struct ocfs2_alloc_context
*ac
)
1140 u32 cluster_off
, cluster_count
;
1141 struct ocfs2_dinode
*alloc
= NULL
;
1142 struct ocfs2_local_alloc
*la
;
1146 alloc
= (struct ocfs2_dinode
*) osb
->local_alloc_bh
->b_data
;
1147 la
= OCFS2_LOCAL_ALLOC(alloc
);
1149 if (alloc
->id1
.bitmap1
.i_total
)
1150 mlog(0, "asking me to alloc a new window over a non-empty "
1153 mlog(0, "Allocating %u clusters for a new window.\n",
1154 osb
->local_alloc_bits
);
1156 /* Instruct the allocation code to try the most recently used
1157 * cluster group. We'll re-record the group used this pass
1159 ac
->ac_last_group
= osb
->la_last_gd
;
1161 /* we used the generic suballoc reserve function, but we set
1162 * everything up nicely, so there's no reason why we can't use
1163 * the more specific cluster api to claim bits. */
1164 status
= ocfs2_claim_clusters(handle
, ac
, osb
->local_alloc_bits
,
1165 &cluster_off
, &cluster_count
);
1166 if (status
== -ENOSPC
) {
1169 * Note: We could also try syncing the journal here to
1170 * allow use of any free bits which the current
1171 * transaction can't give us access to. --Mark
1173 if (ocfs2_recalc_la_window(osb
, OCFS2_LA_EVENT_FRAGMENTED
) ==
1177 ac
->ac_bits_wanted
= osb
->local_alloc_default_bits
;
1178 status
= ocfs2_claim_clusters(handle
, ac
,
1179 osb
->local_alloc_bits
,
1182 if (status
== -ENOSPC
)
1185 * We only shrunk the *minimum* number of in our
1186 * request - it's entirely possible that the allocator
1187 * might give us more than we asked for.
1190 spin_lock(&osb
->osb_lock
);
1191 osb
->local_alloc_bits
= cluster_count
;
1192 spin_unlock(&osb
->osb_lock
);
1196 if (status
!= -ENOSPC
)
1201 osb
->la_last_gd
= ac
->ac_last_group
;
1203 la
->la_bm_off
= cpu_to_le32(cluster_off
);
1204 alloc
->id1
.bitmap1
.i_total
= cpu_to_le32(cluster_count
);
1205 /* just in case... In the future when we find space ourselves,
1206 * we don't have to get all contiguous -- but we'll have to
1207 * set all previously used bits in bitmap and update
1208 * la_bits_set before setting the bits in the main bitmap. */
1209 alloc
->id1
.bitmap1
.i_used
= 0;
1210 memset(OCFS2_LOCAL_ALLOC(alloc
)->la_bitmap
, 0,
1211 le16_to_cpu(la
->la_size
));
1213 ocfs2_resmap_restart(&osb
->osb_la_resmap
, cluster_count
,
1214 OCFS2_LOCAL_ALLOC(alloc
)->la_bitmap
);
1216 mlog(0, "New window allocated:\n");
1217 mlog(0, "window la_bm_off = %u\n",
1218 OCFS2_LOCAL_ALLOC(alloc
)->la_bm_off
);
1219 mlog(0, "window bits = %u\n", le32_to_cpu(alloc
->id1
.bitmap1
.i_total
));
1226 /* Note that we do *NOT* lock the local alloc inode here as
1227 * it's been locked already for us. */
1228 static int ocfs2_local_alloc_slide_window(struct ocfs2_super
*osb
,
1229 struct inode
*local_alloc_inode
)
1232 struct buffer_head
*main_bm_bh
= NULL
;
1233 struct inode
*main_bm_inode
= NULL
;
1234 handle_t
*handle
= NULL
;
1235 struct ocfs2_dinode
*alloc
;
1236 struct ocfs2_dinode
*alloc_copy
= NULL
;
1237 struct ocfs2_alloc_context
*ac
= NULL
;
1241 ocfs2_recalc_la_window(osb
, OCFS2_LA_EVENT_SLIDE
);
1243 /* This will lock the main bitmap for us. */
1244 status
= ocfs2_local_alloc_reserve_for_window(osb
,
1249 if (status
!= -ENOSPC
)
1254 handle
= ocfs2_start_trans(osb
, OCFS2_WINDOW_MOVE_CREDITS
);
1255 if (IS_ERR(handle
)) {
1256 status
= PTR_ERR(handle
);
1262 alloc
= (struct ocfs2_dinode
*) osb
->local_alloc_bh
->b_data
;
1264 /* We want to clear the local alloc before doing anything
1265 * else, so that if we error later during this operation,
1266 * local alloc shutdown won't try to double free main bitmap
1267 * bits. Make a copy so the sync function knows which bits to
1269 alloc_copy
= kmalloc(osb
->local_alloc_bh
->b_size
, GFP_NOFS
);
1275 memcpy(alloc_copy
, alloc
, osb
->local_alloc_bh
->b_size
);
1277 status
= ocfs2_journal_access_di(handle
,
1278 INODE_CACHE(local_alloc_inode
),
1279 osb
->local_alloc_bh
,
1280 OCFS2_JOURNAL_ACCESS_WRITE
);
1286 ocfs2_clear_local_alloc(alloc
);
1287 ocfs2_journal_dirty(handle
, osb
->local_alloc_bh
);
1289 status
= ocfs2_sync_local_to_main(osb
, handle
, alloc_copy
,
1290 main_bm_inode
, main_bm_bh
);
1296 status
= ocfs2_local_alloc_new_window(osb
, handle
, ac
);
1298 if (status
!= -ENOSPC
)
1303 atomic_inc(&osb
->alloc_stats
.moves
);
1307 ocfs2_commit_trans(osb
, handle
);
1312 iput(main_bm_inode
);
1318 ocfs2_free_alloc_context(ac
);