1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * metadata alloc and free
7 * Inspired by ext3 block groups.
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
38 #include "blockcheck.h"
42 #include "localalloc.h"
48 #include "buffer_head_io.h"
50 #define NOT_ALLOC_NEW_GROUP 0
51 #define ALLOC_NEW_GROUP 0x1
52 #define ALLOC_GROUPS_FROM_GLOBAL 0x2
54 #define OCFS2_MAX_TO_STEAL 1024
56 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
57 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
58 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
59 static int ocfs2_block_group_fill(handle_t *handle,
60 struct inode *alloc_inode,
61 struct buffer_head *bg_bh,
64 struct ocfs2_chain_list *cl);
65 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
66 struct inode *alloc_inode,
67 struct buffer_head *bh,
69 u64 *last_alloc_group,
72 static int ocfs2_cluster_group_search(struct inode *inode,
73 struct buffer_head *group_bh,
74 u32 bits_wanted, u32 min_bits,
76 u16 *bit_off, u16 *bits_found);
77 static int ocfs2_block_group_search(struct inode *inode,
78 struct buffer_head *group_bh,
79 u32 bits_wanted, u32 min_bits,
81 u16 *bit_off, u16 *bits_found);
82 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
83 struct ocfs2_alloc_context *ac,
88 unsigned int *num_bits,
90 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
92 static inline int ocfs2_block_group_set_bits(handle_t *handle,
93 struct inode *alloc_inode,
94 struct ocfs2_group_desc *bg,
95 struct buffer_head *group_bh,
97 unsigned int num_bits);
98 static int ocfs2_relink_block_group(handle_t *handle,
99 struct inode *alloc_inode,
100 struct buffer_head *fe_bh,
101 struct buffer_head *bg_bh,
102 struct buffer_head *prev_bg_bh,
104 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
106 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
109 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
113 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
114 u32 bits_wanted, u64 max_block,
116 struct ocfs2_alloc_context **ac);
118 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
120 struct inode *inode = ac->ac_inode;
123 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
124 ocfs2_inode_unlock(inode, 1);
126 mutex_unlock(&inode->i_mutex);
136 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
138 ocfs2_free_ac_resource(ac);
142 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
144 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
147 #define do_error(fmt, ...) \
150 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
152 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
155 static int ocfs2_validate_gd_self(struct super_block *sb,
156 struct buffer_head *bh,
159 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
161 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
162 do_error("Group descriptor #%llu has bad signature %.*s",
163 (unsigned long long)bh->b_blocknr, 7,
168 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
169 do_error("Group descriptor #%llu has an invalid bg_blkno "
171 (unsigned long long)bh->b_blocknr,
172 (unsigned long long)le64_to_cpu(gd->bg_blkno));
176 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
177 do_error("Group descriptor #%llu has an invalid "
178 "fs_generation of #%u",
179 (unsigned long long)bh->b_blocknr,
180 le32_to_cpu(gd->bg_generation));
184 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
185 do_error("Group descriptor #%llu has bit count %u but "
186 "claims that %u are free",
187 (unsigned long long)bh->b_blocknr,
188 le16_to_cpu(gd->bg_bits),
189 le16_to_cpu(gd->bg_free_bits_count));
193 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
194 do_error("Group descriptor #%llu has bit count %u but "
195 "max bitmap bits of %u",
196 (unsigned long long)bh->b_blocknr,
197 le16_to_cpu(gd->bg_bits),
198 8 * le16_to_cpu(gd->bg_size));
205 static int ocfs2_validate_gd_parent(struct super_block *sb,
206 struct ocfs2_dinode *di,
207 struct buffer_head *bh,
210 unsigned int max_bits;
211 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
213 if (di->i_blkno != gd->bg_parent_dinode) {
214 do_error("Group descriptor #%llu has bad parent "
215 "pointer (%llu, expected %llu)",
216 (unsigned long long)bh->b_blocknr,
217 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
218 (unsigned long long)le64_to_cpu(di->i_blkno));
222 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
223 if (le16_to_cpu(gd->bg_bits) > max_bits) {
224 do_error("Group descriptor #%llu has bit count of %u",
225 (unsigned long long)bh->b_blocknr,
226 le16_to_cpu(gd->bg_bits));
230 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
231 if ((le16_to_cpu(gd->bg_chain) >
232 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
233 ((le16_to_cpu(gd->bg_chain) ==
234 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
235 do_error("Group descriptor #%llu has bad chain %u",
236 (unsigned long long)bh->b_blocknr,
237 le16_to_cpu(gd->bg_chain));
247 * This version only prints errors. It does not fail the filesystem, and
248 * exists only for resize.
250 int ocfs2_check_group_descriptor(struct super_block *sb,
251 struct ocfs2_dinode *di,
252 struct buffer_head *bh)
255 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
257 BUG_ON(!buffer_uptodate(bh));
260 * If the ecc fails, we return the error but otherwise
261 * leave the filesystem running. We know any error is
262 * local to this block.
264 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
267 "Checksum failed for group descriptor %llu\n",
268 (unsigned long long)bh->b_blocknr);
270 rc = ocfs2_validate_gd_self(sb, bh, 1);
272 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
277 static int ocfs2_validate_group_descriptor(struct super_block *sb,
278 struct buffer_head *bh)
281 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
283 mlog(0, "Validating group descriptor %llu\n",
284 (unsigned long long)bh->b_blocknr);
286 BUG_ON(!buffer_uptodate(bh));
289 * If the ecc fails, we return the error but otherwise
290 * leave the filesystem running. We know any error is
291 * local to this block.
293 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
298 * Errors after here are fatal.
301 return ocfs2_validate_gd_self(sb, bh, 0);
304 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
305 u64 gd_blkno, struct buffer_head **bh)
308 struct buffer_head *tmp = *bh;
310 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
311 ocfs2_validate_group_descriptor);
315 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
321 /* If ocfs2_read_block() got us a new bh, pass it up. */
329 static int ocfs2_block_group_fill(handle_t *handle,
330 struct inode *alloc_inode,
331 struct buffer_head *bg_bh,
334 struct ocfs2_chain_list *cl)
337 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
338 struct super_block * sb = alloc_inode->i_sb;
342 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
343 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
345 (unsigned long long)group_blkno,
346 (unsigned long long) bg_bh->b_blocknr);
351 status = ocfs2_journal_access_gd(handle,
352 INODE_CACHE(alloc_inode),
354 OCFS2_JOURNAL_ACCESS_CREATE);
360 memset(bg, 0, sb->s_blocksize);
361 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
362 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
363 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
364 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
365 bg->bg_chain = cpu_to_le16(my_chain);
366 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
367 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
368 bg->bg_blkno = cpu_to_le64(group_blkno);
369 /* set the 1st bit in the bitmap to account for the descriptor block */
370 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
371 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
373 ocfs2_journal_dirty(handle, bg_bh);
375 /* There is no need to zero out or otherwise initialize the
376 * other blocks in a group - All valid FS metadata in a block
377 * group stores the superblock fs_generation value at
378 * allocation time. */
385 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
390 while (curr < le16_to_cpu(cl->cl_count)) {
391 if (le32_to_cpu(cl->cl_recs[best].c_total) >
392 le32_to_cpu(cl->cl_recs[curr].c_total))
400 * We expect the block group allocator to already be locked.
402 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
403 struct inode *alloc_inode,
404 struct buffer_head *bh,
406 u64 *last_alloc_group,
410 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
411 struct ocfs2_chain_list *cl;
412 struct ocfs2_alloc_context *ac = NULL;
413 handle_t *handle = NULL;
414 u32 bit_off, num_bits;
417 struct buffer_head *bg_bh = NULL;
418 struct ocfs2_group_desc *bg;
420 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
424 cl = &fe->id2.i_chain;
425 status = ocfs2_reserve_clusters_with_limit(osb,
426 le16_to_cpu(cl->cl_cpg),
427 max_block, flags, &ac);
429 if (status != -ENOSPC)
434 credits = ocfs2_calc_group_alloc_credits(osb->sb,
435 le16_to_cpu(cl->cl_cpg));
436 handle = ocfs2_start_trans(osb, credits);
437 if (IS_ERR(handle)) {
438 status = PTR_ERR(handle);
444 if (last_alloc_group && *last_alloc_group != 0) {
445 mlog(0, "use old allocation group %llu for block group alloc\n",
446 (unsigned long long)*last_alloc_group);
447 ac->ac_last_group = *last_alloc_group;
449 status = ocfs2_claim_clusters(osb,
452 le16_to_cpu(cl->cl_cpg),
456 if (status != -ENOSPC)
461 alloc_rec = ocfs2_find_smallest_chain(cl);
463 /* setup the group */
464 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
465 mlog(0, "new descriptor, record %u, at block %llu\n",
466 alloc_rec, (unsigned long long)bg_blkno);
468 bg_bh = sb_getblk(osb->sb, bg_blkno);
474 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
476 status = ocfs2_block_group_fill(handle,
487 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
489 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
490 bh, OCFS2_JOURNAL_ACCESS_WRITE);
496 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
497 le16_to_cpu(bg->bg_free_bits_count));
498 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
499 cl->cl_recs[alloc_rec].c_blkno = cpu_to_le64(bg_blkno);
500 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
501 le16_add_cpu(&cl->cl_next_free_rec, 1);
503 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
504 le16_to_cpu(bg->bg_free_bits_count));
505 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
506 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
508 ocfs2_journal_dirty(handle, bh);
510 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
511 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
512 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
513 le32_to_cpu(fe->i_clusters)));
514 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
515 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
516 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
520 /* save the new last alloc group so that the caller can cache it. */
521 if (last_alloc_group)
522 *last_alloc_group = ac->ac_last_group;
526 ocfs2_commit_trans(osb, handle);
529 ocfs2_free_alloc_context(ac);
537 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
538 struct ocfs2_alloc_context *ac,
541 u64 *last_alloc_group,
545 u32 bits_wanted = ac->ac_bits_wanted;
546 struct inode *alloc_inode;
547 struct buffer_head *bh = NULL;
548 struct ocfs2_dinode *fe;
553 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
559 mutex_lock(&alloc_inode->i_mutex);
561 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
563 mutex_unlock(&alloc_inode->i_mutex);
570 ac->ac_inode = alloc_inode;
571 ac->ac_alloc_slot = slot;
573 fe = (struct ocfs2_dinode *) bh->b_data;
575 /* The bh was validated by the inode read inside
576 * ocfs2_inode_lock(). Any corruption is a code bug. */
577 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
579 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
580 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
581 (unsigned long long)le64_to_cpu(fe->i_blkno));
586 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
587 le32_to_cpu(fe->id1.bitmap1.i_used);
589 if (bits_wanted > free_bits) {
590 /* cluster bitmap never grows */
591 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
592 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
593 bits_wanted, free_bits);
598 if (!(flags & ALLOC_NEW_GROUP)) {
599 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
600 "and we don't alloc a new group for it.\n",
601 slot, bits_wanted, free_bits);
606 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
608 last_alloc_group, flags);
610 if (status != -ENOSPC)
614 atomic_inc(&osb->alloc_stats.bg_extends);
616 /* You should never ask for this much metadata */
618 (le32_to_cpu(fe->id1.bitmap1.i_total)
619 - le32_to_cpu(fe->id1.bitmap1.i_used)));
631 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
633 spin_lock(&osb->osb_lock);
634 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
635 spin_unlock(&osb->osb_lock);
636 atomic_set(&osb->s_num_inodes_stolen, 0);
639 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
641 spin_lock(&osb->osb_lock);
642 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
643 spin_unlock(&osb->osb_lock);
644 atomic_set(&osb->s_num_meta_stolen, 0);
647 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
649 ocfs2_init_inode_steal_slot(osb);
650 ocfs2_init_meta_steal_slot(osb);
653 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
655 spin_lock(&osb->osb_lock);
656 if (type == INODE_ALLOC_SYSTEM_INODE)
657 osb->s_inode_steal_slot = slot;
658 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
659 osb->s_meta_steal_slot = slot;
660 spin_unlock(&osb->osb_lock);
663 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
665 int slot = OCFS2_INVALID_SLOT;
667 spin_lock(&osb->osb_lock);
668 if (type == INODE_ALLOC_SYSTEM_INODE)
669 slot = osb->s_inode_steal_slot;
670 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
671 slot = osb->s_meta_steal_slot;
672 spin_unlock(&osb->osb_lock);
677 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
679 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
682 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
684 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
687 static int ocfs2_steal_resource(struct ocfs2_super *osb,
688 struct ocfs2_alloc_context *ac,
691 int i, status = -ENOSPC;
692 int slot = __ocfs2_get_steal_slot(osb, type);
694 /* Start to steal resource from the first slot after ours. */
695 if (slot == OCFS2_INVALID_SLOT)
696 slot = osb->slot_num + 1;
698 for (i = 0; i < osb->max_slots; i++, slot++) {
699 if (slot == osb->max_slots)
702 if (slot == osb->slot_num)
705 status = ocfs2_reserve_suballoc_bits(osb, ac,
708 NOT_ALLOC_NEW_GROUP);
710 __ocfs2_set_steal_slot(osb, slot, type);
714 ocfs2_free_ac_resource(ac);
720 static int ocfs2_steal_inode(struct ocfs2_super *osb,
721 struct ocfs2_alloc_context *ac)
723 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
726 static int ocfs2_steal_meta(struct ocfs2_super *osb,
727 struct ocfs2_alloc_context *ac)
729 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
732 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
734 struct ocfs2_alloc_context **ac)
737 int slot = ocfs2_get_meta_steal_slot(osb);
739 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
746 (*ac)->ac_bits_wanted = blocks;
747 (*ac)->ac_which = OCFS2_AC_USE_META;
748 (*ac)->ac_group_search = ocfs2_block_group_search;
750 if (slot != OCFS2_INVALID_SLOT &&
751 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
754 atomic_set(&osb->s_num_meta_stolen, 0);
755 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
756 EXTENT_ALLOC_SYSTEM_INODE,
757 (u32)osb->slot_num, NULL,
758 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
763 if (slot != OCFS2_INVALID_SLOT)
764 ocfs2_init_meta_steal_slot(osb);
766 } else if (status < 0 && status != -ENOSPC) {
771 ocfs2_free_ac_resource(*ac);
774 status = ocfs2_steal_meta(osb, *ac);
775 atomic_inc(&osb->s_num_meta_stolen);
777 if (status != -ENOSPC)
784 if ((status < 0) && *ac) {
785 ocfs2_free_alloc_context(*ac);
793 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
794 struct ocfs2_extent_list *root_el,
795 struct ocfs2_alloc_context **ac)
797 return ocfs2_reserve_new_metadata_blocks(osb,
798 ocfs2_extend_meta_needed(root_el),
802 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
803 struct ocfs2_alloc_context **ac)
806 int slot = ocfs2_get_inode_steal_slot(osb);
809 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
816 (*ac)->ac_bits_wanted = 1;
817 (*ac)->ac_which = OCFS2_AC_USE_INODE;
819 (*ac)->ac_group_search = ocfs2_block_group_search;
822 * stat(2) can't handle i_ino > 32bits, so we tell the
823 * lower levels not to allocate us a block group past that
824 * limit. The 'inode64' mount option avoids this behavior.
826 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
827 (*ac)->ac_max_block = (u32)~0U;
830 * slot is set when we successfully steal inode from other nodes.
831 * It is reset in 3 places:
832 * 1. when we flush the truncate log
833 * 2. when we complete local alloc recovery.
834 * 3. when we successfully allocate from our own slot.
835 * After it is set, we will go on stealing inodes until we find the
836 * need to check our slots to see whether there is some space for us.
838 if (slot != OCFS2_INVALID_SLOT &&
839 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
842 atomic_set(&osb->s_num_inodes_stolen, 0);
843 alloc_group = osb->osb_inode_alloc_group;
844 status = ocfs2_reserve_suballoc_bits(osb, *ac,
845 INODE_ALLOC_SYSTEM_INODE,
849 ALLOC_GROUPS_FROM_GLOBAL);
853 spin_lock(&osb->osb_lock);
854 osb->osb_inode_alloc_group = alloc_group;
855 spin_unlock(&osb->osb_lock);
856 mlog(0, "after reservation, new allocation group is "
857 "%llu\n", (unsigned long long)alloc_group);
860 * Some inodes must be freed by us, so try to allocate
861 * from our own next time.
863 if (slot != OCFS2_INVALID_SLOT)
864 ocfs2_init_inode_steal_slot(osb);
866 } else if (status < 0 && status != -ENOSPC) {
871 ocfs2_free_ac_resource(*ac);
874 status = ocfs2_steal_inode(osb, *ac);
875 atomic_inc(&osb->s_num_inodes_stolen);
877 if (status != -ENOSPC)
884 if ((status < 0) && *ac) {
885 ocfs2_free_alloc_context(*ac);
893 /* local alloc code has to do the same thing, so rather than do this
895 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
896 struct ocfs2_alloc_context *ac)
900 ac->ac_which = OCFS2_AC_USE_MAIN;
901 ac->ac_group_search = ocfs2_cluster_group_search;
903 status = ocfs2_reserve_suballoc_bits(osb, ac,
904 GLOBAL_BITMAP_SYSTEM_INODE,
905 OCFS2_INVALID_SLOT, NULL,
907 if (status < 0 && status != -ENOSPC) {
916 /* Callers don't need to care which bitmap (local alloc or main) to
917 * use so we figure it out for them, but unfortunately this clutters
919 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
920 u32 bits_wanted, u64 max_block,
922 struct ocfs2_alloc_context **ac)
928 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
935 (*ac)->ac_bits_wanted = bits_wanted;
936 (*ac)->ac_max_block = max_block;
939 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
940 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
941 status = ocfs2_reserve_local_alloc_bits(osb,
944 if ((status < 0) && (status != -ENOSPC)) {
950 if (status == -ENOSPC) {
951 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
953 if (status != -ENOSPC)
961 if ((status < 0) && *ac) {
962 ocfs2_free_alloc_context(*ac);
970 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
972 struct ocfs2_alloc_context **ac)
974 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
975 ALLOC_NEW_GROUP, ac);
979 * More or less lifted from ext3. I'll leave their description below:
981 * "For ext3 allocations, we must not reuse any blocks which are
982 * allocated in the bitmap buffer's "last committed data" copy. This
983 * prevents deletes from freeing up the page for reuse until we have
984 * committed the delete transaction.
986 * If we didn't do this, then deleting something and reallocating it as
987 * data would allow the old block to be overwritten before the
988 * transaction committed (because we force data to disk before commit).
989 * This would lead to corruption if we crashed between overwriting the
990 * data and committing the delete.
992 * @@@ We may want to make this allocation behaviour conditional on
993 * data-writes at some point, and disable it for metadata allocations or
996 * Note: OCFS2 already does this differently for metadata vs data
997 * allocations, as those bitmaps are separate and undo access is never
998 * called on a metadata group descriptor.
1000 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1003 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1006 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1009 if (!buffer_jbd(bg_bh))
1012 jbd_lock_bh_state(bg_bh);
1013 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1015 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1018 jbd_unlock_bh_state(bg_bh);
1023 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1024 struct buffer_head *bg_bh,
1025 unsigned int bits_wanted,
1026 unsigned int total_bits,
1031 u16 best_offset, best_size;
1032 int offset, start, found, status = 0;
1033 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1035 /* Callers got this descriptor from
1036 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1037 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1039 found = start = best_offset = best_size = 0;
1040 bitmap = bg->bg_bitmap;
1042 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1043 if (offset == total_bits)
1046 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1047 /* We found a zero, but we can't use it as it
1048 * hasn't been put to disk yet! */
1051 } else if (offset == start) {
1052 /* we found a zero */
1054 /* move start to the next bit to test */
1057 /* got a zero after some ones */
1061 if (found > best_size) {
1063 best_offset = start - found;
1065 /* we got everything we needed */
1066 if (found == bits_wanted) {
1067 /* mlog(0, "Found it all!\n"); */
1072 /* XXX: I think the first clause is equivalent to the second
1074 if (found == bits_wanted) {
1075 *bit_off = start - found;
1076 *bits_found = found;
1077 } else if (best_size) {
1078 *bit_off = best_offset;
1079 *bits_found = best_size;
1082 /* No error log here -- see the comment above
1083 * ocfs2_test_bg_bit_allocatable */
1089 static inline int ocfs2_block_group_set_bits(handle_t *handle,
1090 struct inode *alloc_inode,
1091 struct ocfs2_group_desc *bg,
1092 struct buffer_head *group_bh,
1093 unsigned int bit_off,
1094 unsigned int num_bits)
1097 void *bitmap = bg->bg_bitmap;
1098 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1102 /* All callers get the descriptor via
1103 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1104 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1105 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1107 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1110 if (ocfs2_is_cluster_bitmap(alloc_inode))
1111 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1113 status = ocfs2_journal_access_gd(handle,
1114 INODE_CACHE(alloc_inode),
1122 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1124 ocfs2_set_bit(bit_off++, bitmap);
1126 ocfs2_journal_dirty(handle, group_bh);
1133 /* find the one with the most empty bits */
1134 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1138 BUG_ON(!cl->cl_next_free_rec);
1141 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1142 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1143 le32_to_cpu(cl->cl_recs[best].c_free))
1148 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1152 static int ocfs2_relink_block_group(handle_t *handle,
1153 struct inode *alloc_inode,
1154 struct buffer_head *fe_bh,
1155 struct buffer_head *bg_bh,
1156 struct buffer_head *prev_bg_bh,
1160 /* there is a really tiny chance the journal calls could fail,
1161 * but we wouldn't want inconsistent blocks in *any* case. */
1162 u64 fe_ptr, bg_ptr, prev_bg_ptr;
1163 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1164 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1165 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1167 /* The caller got these descriptors from
1168 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1169 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1170 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1172 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1173 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1174 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1175 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1177 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1178 bg_ptr = le64_to_cpu(bg->bg_next_group);
1179 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1181 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1183 OCFS2_JOURNAL_ACCESS_WRITE);
1189 prev_bg->bg_next_group = bg->bg_next_group;
1190 ocfs2_journal_dirty(handle, prev_bg_bh);
1192 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1193 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1199 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1200 ocfs2_journal_dirty(handle, bg_bh);
1202 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1203 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1209 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1210 ocfs2_journal_dirty(handle, fe_bh);
1214 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1215 bg->bg_next_group = cpu_to_le64(bg_ptr);
1216 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1223 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1226 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1229 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1230 * value on error. */
1231 static int ocfs2_cluster_group_search(struct inode *inode,
1232 struct buffer_head *group_bh,
1233 u32 bits_wanted, u32 min_bits,
1235 u16 *bit_off, u16 *bits_found)
1237 int search = -ENOSPC;
1240 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1241 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1242 u16 tmp_off, tmp_found;
1243 unsigned int max_bits, gd_cluster_off;
1245 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1247 if (gd->bg_free_bits_count) {
1248 max_bits = le16_to_cpu(gd->bg_bits);
1250 /* Tail groups in cluster bitmaps which aren't cpg
1251 * aligned are prone to partial extention by a failed
1252 * fs resize. If the file system resize never got to
1253 * update the dinode cluster count, then we don't want
1254 * to trust any clusters past it, regardless of what
1255 * the group descriptor says. */
1256 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1257 le64_to_cpu(gd->bg_blkno));
1258 if ((gd_cluster_off + max_bits) >
1259 OCFS2_I(inode)->ip_clusters) {
1260 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1261 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1262 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1263 le16_to_cpu(gd->bg_bits),
1264 OCFS2_I(inode)->ip_clusters, max_bits);
1267 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1268 group_bh, bits_wanted,
1270 &tmp_off, &tmp_found);
1275 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1277 tmp_off + tmp_found);
1278 mlog(0, "Checking %llu against %llu\n",
1279 (unsigned long long)blkoff,
1280 (unsigned long long)max_block);
1281 if (blkoff > max_block)
1285 /* ocfs2_block_group_find_clear_bits() might
1286 * return success, but we still want to return
1287 * -ENOSPC unless it found the minimum number
1289 if (min_bits <= tmp_found) {
1291 *bits_found = tmp_found;
1292 search = 0; /* success */
1293 } else if (tmp_found) {
1295 * Don't show bits which we'll be returning
1296 * for allocation to the local alloc bitmap.
1298 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1305 static int ocfs2_block_group_search(struct inode *inode,
1306 struct buffer_head *group_bh,
1307 u32 bits_wanted, u32 min_bits,
1309 u16 *bit_off, u16 *bits_found)
1313 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1315 BUG_ON(min_bits != 1);
1316 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1318 if (bg->bg_free_bits_count) {
1319 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1320 group_bh, bits_wanted,
1321 le16_to_cpu(bg->bg_bits),
1322 bit_off, bits_found);
1323 if (!ret && max_block) {
1324 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1326 mlog(0, "Checking %llu against %llu\n",
1327 (unsigned long long)blkoff,
1328 (unsigned long long)max_block);
1329 if (blkoff > max_block)
1337 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1339 struct buffer_head *di_bh,
1345 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1346 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1348 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1349 OCFS2_JOURNAL_ACCESS_WRITE);
1355 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1356 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1357 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1358 ocfs2_journal_dirty(handle, di_bh);
1364 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1369 unsigned int *num_bits,
1375 struct buffer_head *group_bh = NULL;
1376 struct ocfs2_group_desc *gd;
1377 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1378 struct inode *alloc_inode = ac->ac_inode;
1380 ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1387 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1388 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1389 ac->ac_max_block, bit_off, &found);
1398 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1400 le16_to_cpu(gd->bg_chain));
1406 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1407 *bit_off, *num_bits);
1411 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1419 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1424 unsigned int *num_bits,
1429 u16 chain, tmp_bits;
1432 struct inode *alloc_inode = ac->ac_inode;
1433 struct buffer_head *group_bh = NULL;
1434 struct buffer_head *prev_group_bh = NULL;
1435 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1436 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1437 struct ocfs2_group_desc *bg;
1439 chain = ac->ac_chain;
1440 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1442 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1444 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1445 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1451 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1454 /* for now, the chain search is a bit simplistic. We just use
1455 * the 1st group with any empty bits. */
1456 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1457 bits_wanted, min_bits,
1458 ac->ac_max_block, bit_off,
1459 &tmp_bits)) == -ENOSPC) {
1460 if (!bg->bg_next_group)
1463 brelse(prev_group_bh);
1464 prev_group_bh = NULL;
1466 next_group = le64_to_cpu(bg->bg_next_group);
1467 prev_group_bh = group_bh;
1469 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1470 next_group, &group_bh);
1475 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1478 if (status != -ENOSPC)
1483 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1484 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1486 *num_bits = tmp_bits;
1488 BUG_ON(*num_bits == 0);
1491 * Keep track of previous block descriptor read. When
1492 * we find a target, if we have read more than X
1493 * number of descriptors, and the target is reasonably
1494 * empty, relink him to top of his chain.
1496 * We've read 0 extra blocks and only send one more to
1497 * the transaction, yet the next guy to search has a
1500 * Do this *after* figuring out how many bits we're taking out
1501 * of our target group.
1503 if (ac->ac_allow_chain_relink &&
1505 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1506 status = ocfs2_relink_block_group(handle, alloc_inode,
1507 ac->ac_bh, group_bh,
1508 prev_group_bh, chain);
1515 /* Ok, claim our bits now: set the info on dinode, chainlist
1516 * and then the group */
1517 status = ocfs2_journal_access_di(handle,
1518 INODE_CACHE(alloc_inode),
1520 OCFS2_JOURNAL_ACCESS_WRITE);
1526 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1527 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1528 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1529 ocfs2_journal_dirty(handle, ac->ac_bh);
1531 status = ocfs2_block_group_set_bits(handle,
1542 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1543 (unsigned long long)le64_to_cpu(fe->i_blkno));
1545 *bg_blkno = le64_to_cpu(bg->bg_blkno);
1546 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1549 brelse(prev_group_bh);
1555 /* will give out up to bits_wanted contiguous bits. */
1556 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1557 struct ocfs2_alloc_context *ac,
1562 unsigned int *num_bits,
1568 u64 hint_blkno = ac->ac_last_group;
1569 struct ocfs2_chain_list *cl;
1570 struct ocfs2_dinode *fe;
1574 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1575 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1578 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1580 /* The bh was validated by the inode read during
1581 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1582 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1584 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1585 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1586 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1587 "bits but only %u total.",
1588 (unsigned long long)le64_to_cpu(fe->i_blkno),
1589 le32_to_cpu(fe->id1.bitmap1.i_used),
1590 le32_to_cpu(fe->id1.bitmap1.i_total));
1596 /* Attempt to short-circuit the usual search mechanism
1597 * by jumping straight to the most recently used
1598 * allocation group. This helps us mantain some
1599 * contiguousness across allocations. */
1600 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1601 min_bits, bit_off, num_bits,
1602 hint_blkno, &bits_left);
1604 /* Be careful to update *bg_blkno here as the
1605 * caller is expecting it to be filled in, and
1606 * ocfs2_search_one_group() won't do that for
1608 *bg_blkno = hint_blkno;
1611 if (status < 0 && status != -ENOSPC) {
1617 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1619 victim = ocfs2_find_victim_chain(cl);
1620 ac->ac_chain = victim;
1621 ac->ac_allow_chain_relink = 1;
1623 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1624 num_bits, bg_blkno, &bits_left);
1627 if (status < 0 && status != -ENOSPC) {
1632 mlog(0, "Search of victim chain %u came up with nothing, "
1633 "trying all chains now.\n", victim);
1635 /* If we didn't pick a good victim, then just default to
1636 * searching each chain in order. Don't allow chain relinking
1637 * because we only calculate enough journal credits for one
1638 * relink per alloc. */
1639 ac->ac_allow_chain_relink = 0;
1640 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1643 if (!cl->cl_recs[i].c_free)
1647 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1648 bit_off, num_bits, bg_blkno,
1652 if (status < 0 && status != -ENOSPC) {
1659 if (status != -ENOSPC) {
1660 /* If the next search of this group is not likely to
1661 * yield a suitable extent, then we reset the last
1662 * group hint so as to not waste a disk read */
1663 if (bits_left < min_bits)
1664 ac->ac_last_group = 0;
1666 ac->ac_last_group = *bg_blkno;
1674 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1676 struct ocfs2_alloc_context *ac,
1678 u16 *suballoc_bit_start,
1679 unsigned int *num_bits,
1686 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1687 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1689 status = ocfs2_claim_suballoc_bits(osb,
1701 atomic_inc(&osb->alloc_stats.bg_allocs);
1703 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1704 ac->ac_bits_given += (*num_bits);
1711 static void ocfs2_init_inode_ac_group(struct inode *dir,
1712 struct buffer_head *parent_fe_bh,
1713 struct ocfs2_alloc_context *ac)
1715 struct ocfs2_dinode *fe = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1717 * Try to allocate inodes from some specific group.
1719 * If the parent dir has recorded the last group used in allocation,
1720 * cool, use it. Otherwise if we try to allocate new inode from the
1721 * same slot the parent dir belongs to, use the same chunk.
1723 * We are very careful here to avoid the mistake of setting
1724 * ac_last_group to a group descriptor from a different (unlocked) slot.
1726 if (OCFS2_I(dir)->ip_last_used_group &&
1727 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
1728 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
1729 else if (le16_to_cpu(fe->i_suballoc_slot) == ac->ac_alloc_slot)
1730 ac->ac_last_group = ocfs2_which_suballoc_group(
1731 le64_to_cpu(fe->i_blkno),
1732 le16_to_cpu(fe->i_suballoc_bit));
1735 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
1736 struct ocfs2_alloc_context *ac)
1738 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
1739 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
1742 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1745 struct buffer_head *parent_fe_bh,
1746 struct ocfs2_alloc_context *ac,
1751 unsigned int num_bits;
1757 BUG_ON(ac->ac_bits_given != 0);
1758 BUG_ON(ac->ac_bits_wanted != 1);
1759 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1761 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
1763 status = ocfs2_claim_suballoc_bits(osb,
1775 atomic_inc(&osb->alloc_stats.bg_allocs);
1777 BUG_ON(num_bits != 1);
1779 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1780 ac->ac_bits_given++;
1781 ocfs2_save_inode_ac_group(dir, ac);
1788 /* translate a group desc. blkno and it's bitmap offset into
1789 * disk cluster offset. */
1790 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1794 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1797 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1799 if (bg_blkno != osb->first_cluster_group_blkno)
1800 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1801 cluster += (u32) bg_bit_off;
1805 /* given a cluster offset, calculate which block group it belongs to
1806 * and return that block offset. */
1807 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1809 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1812 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1814 group_no = cluster / osb->bitmap_cpg;
1816 return osb->first_cluster_group_blkno;
1817 return ocfs2_clusters_to_blocks(inode->i_sb,
1818 group_no * osb->bitmap_cpg);
1821 /* given the block number of a cluster start, calculate which cluster
1822 * group and descriptor bitmap offset that corresponds to. */
1823 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1828 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1829 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1831 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1833 *bg_blkno = ocfs2_which_cluster_group(inode,
1836 if (*bg_blkno == osb->first_cluster_group_blkno)
1837 *bg_bit_off = (u16) data_cluster;
1839 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1840 data_blkno - *bg_blkno);
1844 * min_bits - minimum contiguous chunk from this total allocation we
1845 * can handle. set to what we asked for originally for a full
1846 * contig. allocation, set to '1' to indicate we can deal with extents
1849 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1851 struct ocfs2_alloc_context *ac,
1858 unsigned int bits_wanted = max_clusters;
1864 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1866 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1867 && ac->ac_which != OCFS2_AC_USE_MAIN);
1869 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1870 WARN_ON(min_clusters > 1);
1872 status = ocfs2_claim_local_alloc_bits(osb,
1879 atomic_inc(&osb->alloc_stats.local_data);
1881 if (min_clusters > (osb->bitmap_cpg - 1)) {
1882 /* The only paths asking for contiguousness
1883 * should know about this already. */
1884 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1885 "group bitmap size %u!\n", min_clusters,
1890 /* clamp the current request down to a realistic size. */
1891 if (bits_wanted > (osb->bitmap_cpg - 1))
1892 bits_wanted = osb->bitmap_cpg - 1;
1894 status = ocfs2_claim_suballoc_bits(osb,
1904 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1907 atomic_inc(&osb->alloc_stats.bitmap_data);
1911 if (status != -ENOSPC)
1916 ac->ac_bits_given += *num_clusters;
1923 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1925 struct ocfs2_alloc_context *ac,
1930 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1932 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1933 bits_wanted, cluster_start, num_clusters);
1936 static int ocfs2_block_group_clear_bits(handle_t *handle,
1937 struct inode *alloc_inode,
1938 struct ocfs2_group_desc *bg,
1939 struct buffer_head *group_bh,
1940 unsigned int bit_off,
1941 unsigned int num_bits,
1942 void (*undo_fn)(unsigned int bit,
1943 unsigned long *bmap))
1947 struct ocfs2_group_desc *undo_bg = NULL;
1951 /* The caller got this descriptor from
1952 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1953 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1955 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1957 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
1958 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1961 OCFS2_JOURNAL_ACCESS_UNDO :
1962 OCFS2_JOURNAL_ACCESS_WRITE);
1969 jbd_lock_bh_state(group_bh);
1970 undo_bg = (struct ocfs2_group_desc *)
1971 bh2jh(group_bh)->b_committed_data;
1977 ocfs2_clear_bit((bit_off + tmp),
1978 (unsigned long *) bg->bg_bitmap);
1980 undo_fn(bit_off + tmp,
1981 (unsigned long *) undo_bg->bg_bitmap);
1983 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1986 jbd_unlock_bh_state(group_bh);
1988 ocfs2_journal_dirty(handle, group_bh);
1994 * expects the suballoc inode to already be locked.
1996 static int _ocfs2_free_suballoc_bits(handle_t *handle,
1997 struct inode *alloc_inode,
1998 struct buffer_head *alloc_bh,
1999 unsigned int start_bit,
2002 void (*undo_fn)(unsigned int bit,
2003 unsigned long *bitmap))
2007 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2008 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2009 struct buffer_head *group_bh = NULL;
2010 struct ocfs2_group_desc *group;
2014 /* The alloc_bh comes from ocfs2_free_dinode() or
2015 * ocfs2_free_clusters(). The callers have all locked the
2016 * allocator and gotten alloc_bh from the lock call. This
2017 * validates the dinode buffer. Any corruption that has happended
2019 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2020 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2022 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
2023 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
2024 (unsigned long long)bg_blkno, start_bit);
2026 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2032 group = (struct ocfs2_group_desc *) group_bh->b_data;
2034 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2036 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2038 start_bit, count, undo_fn);
2044 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2045 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2051 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2053 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2054 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2055 ocfs2_journal_dirty(handle, alloc_bh);
2064 int ocfs2_free_suballoc_bits(handle_t *handle,
2065 struct inode *alloc_inode,
2066 struct buffer_head *alloc_bh,
2067 unsigned int start_bit,
2071 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2072 start_bit, bg_blkno, count, NULL);
2075 int ocfs2_free_dinode(handle_t *handle,
2076 struct inode *inode_alloc_inode,
2077 struct buffer_head *inode_alloc_bh,
2078 struct ocfs2_dinode *di)
2080 u64 blk = le64_to_cpu(di->i_blkno);
2081 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2082 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2084 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2085 inode_alloc_bh, bit, bg_blkno, 1);
2088 static int _ocfs2_free_clusters(handle_t *handle,
2089 struct inode *bitmap_inode,
2090 struct buffer_head *bitmap_bh,
2092 unsigned int num_clusters,
2093 void (*undo_fn)(unsigned int bit,
2094 unsigned long *bitmap))
2099 struct ocfs2_dinode *fe;
2101 /* You can't ever have a contiguous set of clusters
2102 * bigger than a block group bitmap so we never have to worry
2103 * about looping on them. */
2107 /* This is expensive. We can safely remove once this stuff has
2108 * gotten tested really well. */
2109 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2111 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2113 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2116 mlog(0, "want to free %u clusters starting at block %llu\n",
2117 num_clusters, (unsigned long long)start_blk);
2118 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2119 (unsigned long long)bg_blkno, bg_start_bit);
2121 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2122 bg_start_bit, bg_blkno,
2123 num_clusters, undo_fn);
2129 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2137 int ocfs2_free_clusters(handle_t *handle,
2138 struct inode *bitmap_inode,
2139 struct buffer_head *bitmap_bh,
2141 unsigned int num_clusters)
2143 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2144 start_blk, num_clusters,
2149 * Give never-used clusters back to the global bitmap. We don't need
2150 * to protect these bits in the undo buffer.
2152 int ocfs2_release_clusters(handle_t *handle,
2153 struct inode *bitmap_inode,
2154 struct buffer_head *bitmap_bh,
2156 unsigned int num_clusters)
2158 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2159 start_blk, num_clusters,
2163 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2165 printk("Block Group:\n");
2166 printk("bg_signature: %s\n", bg->bg_signature);
2167 printk("bg_size: %u\n", bg->bg_size);
2168 printk("bg_bits: %u\n", bg->bg_bits);
2169 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2170 printk("bg_chain: %u\n", bg->bg_chain);
2171 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
2172 printk("bg_next_group: %llu\n",
2173 (unsigned long long)bg->bg_next_group);
2174 printk("bg_parent_dinode: %llu\n",
2175 (unsigned long long)bg->bg_parent_dinode);
2176 printk("bg_blkno: %llu\n",
2177 (unsigned long long)bg->bg_blkno);
2180 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2184 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2185 printk("i_signature: %s\n", fe->i_signature);
2186 printk("i_size: %llu\n",
2187 (unsigned long long)fe->i_size);
2188 printk("i_clusters: %u\n", fe->i_clusters);
2189 printk("i_generation: %u\n",
2190 le32_to_cpu(fe->i_generation));
2191 printk("id1.bitmap1.i_used: %u\n",
2192 le32_to_cpu(fe->id1.bitmap1.i_used));
2193 printk("id1.bitmap1.i_total: %u\n",
2194 le32_to_cpu(fe->id1.bitmap1.i_total));
2195 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2196 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2197 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2198 printk("id2.i_chain.cl_next_free_rec: %u\n",
2199 fe->id2.i_chain.cl_next_free_rec);
2200 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2201 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2202 fe->id2.i_chain.cl_recs[i].c_free);
2203 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2204 fe->id2.i_chain.cl_recs[i].c_total);
2205 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2206 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2211 * For a given allocation, determine which allocators will need to be
2212 * accessed, and lock them, reserving the appropriate number of bits.
2214 * Sparse file systems call this from ocfs2_write_begin_nolock()
2215 * and ocfs2_allocate_unwritten_extents().
2217 * File systems which don't support holes call this from
2218 * ocfs2_extend_allocation().
2220 int ocfs2_lock_allocators(struct inode *inode,
2221 struct ocfs2_extent_tree *et,
2222 u32 clusters_to_add, u32 extents_to_split,
2223 struct ocfs2_alloc_context **data_ac,
2224 struct ocfs2_alloc_context **meta_ac)
2226 int ret = 0, num_free_extents;
2227 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2228 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2234 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2236 num_free_extents = ocfs2_num_free_extents(osb, et);
2237 if (num_free_extents < 0) {
2238 ret = num_free_extents;
2244 * Sparse allocation file systems need to be more conservative
2245 * with reserving room for expansion - the actual allocation
2246 * happens while we've got a journal handle open so re-taking
2247 * a cluster lock (because we ran out of room for another
2248 * extent) will violate ordering rules.
2250 * Most of the time we'll only be seeing this 1 cluster at a time
2253 * Always lock for any unwritten extents - we might want to
2254 * add blocks during a split.
2256 if (!num_free_extents ||
2257 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2258 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2266 if (clusters_to_add == 0)
2269 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2279 ocfs2_free_alloc_context(*meta_ac);
2284 * We cannot have an error and a non null *data_ac.
2292 * Read the inode specified by blkno to get suballoc_slot and
2295 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2296 u16 *suballoc_slot, u16 *suballoc_bit)
2299 struct buffer_head *inode_bh = NULL;
2300 struct ocfs2_dinode *inode_fe;
2302 mlog_entry("blkno: %llu\n", (unsigned long long)blkno);
2304 /* dirty read disk */
2305 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2307 mlog(ML_ERROR, "read block %llu failed %d\n",
2308 (unsigned long long)blkno, status);
2312 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2313 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2314 mlog(ML_ERROR, "invalid inode %llu requested\n",
2315 (unsigned long long)blkno);
2320 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2321 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2322 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2323 (unsigned long long)blkno,
2324 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2330 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2332 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2342 * test whether bit is SET in allocator bitmap or not. on success, 0
2343 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2344 * is returned and *res is meaningless. Call this after you have
2345 * cluster locked against suballoc, or you may get a result based on
2346 * non-up2date contents
2348 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2349 struct inode *suballoc,
2350 struct buffer_head *alloc_bh, u64 blkno,
2353 struct ocfs2_dinode *alloc_fe;
2354 struct ocfs2_group_desc *group;
2355 struct buffer_head *group_bh = NULL;
2359 mlog_entry("blkno: %llu bit: %u\n", (unsigned long long)blkno,
2362 alloc_fe = (struct ocfs2_dinode *)alloc_bh->b_data;
2363 if ((bit + 1) > ocfs2_bits_per_group(&alloc_fe->id2.i_chain)) {
2364 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2366 ocfs2_bits_per_group(&alloc_fe->id2.i_chain));
2371 bg_blkno = ocfs2_which_suballoc_group(blkno, bit);
2372 status = ocfs2_read_group_descriptor(suballoc, alloc_fe, bg_blkno,
2375 mlog(ML_ERROR, "read group %llu failed %d\n",
2376 (unsigned long long)bg_blkno, status);
2380 group = (struct ocfs2_group_desc *) group_bh->b_data;
2381 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2391 * Test if the bit representing this inode (blkno) is set in the
2394 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2396 * In the event of failure, a negative value is returned and *res is
2399 * Callers must make sure to hold nfs_sync_lock to prevent
2400 * ocfs2_delete_inode() on another node from accessing the same
2401 * suballocator concurrently.
2403 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2406 u16 suballoc_bit = 0, suballoc_slot = 0;
2407 struct inode *inode_alloc_inode;
2408 struct buffer_head *alloc_bh = NULL;
2410 mlog_entry("blkno: %llu", (unsigned long long)blkno);
2412 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2415 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2420 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2422 if (!inode_alloc_inode) {
2423 /* the error code could be inaccurate, but we are not able to
2424 * get the correct one. */
2426 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2427 (u32)suballoc_slot);
2431 mutex_lock(&inode_alloc_inode->i_mutex);
2432 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2434 mutex_unlock(&inode_alloc_inode->i_mutex);
2435 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2436 (u32)suballoc_slot, status);
2440 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2441 blkno, suballoc_bit, res);
2443 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2445 ocfs2_inode_unlock(inode_alloc_inode, 0);
2446 mutex_unlock(&inode_alloc_inode->i_mutex);
2448 iput(inode_alloc_inode);