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 inline int ocfs2_block_group_clear_bits(handle_t *handle,
99 struct inode *alloc_inode,
100 struct ocfs2_group_desc *bg,
101 struct buffer_head *group_bh,
102 unsigned int bit_off,
103 unsigned int num_bits);
105 static int ocfs2_relink_block_group(handle_t *handle,
106 struct inode *alloc_inode,
107 struct buffer_head *fe_bh,
108 struct buffer_head *bg_bh,
109 struct buffer_head *prev_bg_bh,
111 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
113 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
116 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
120 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
121 u32 bits_wanted, u64 max_block,
123 struct ocfs2_alloc_context **ac);
125 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
127 struct inode *inode = ac->ac_inode;
130 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
131 ocfs2_inode_unlock(inode, 1);
133 mutex_unlock(&inode->i_mutex);
142 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
144 ocfs2_free_ac_resource(ac);
148 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
150 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
153 #define do_error(fmt, ...) \
156 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
158 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
161 static int ocfs2_validate_gd_self(struct super_block *sb,
162 struct buffer_head *bh,
165 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
167 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
168 do_error("Group descriptor #%llu has bad signature %.*s",
169 (unsigned long long)bh->b_blocknr, 7,
174 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
175 do_error("Group descriptor #%llu has an invalid bg_blkno "
177 (unsigned long long)bh->b_blocknr,
178 (unsigned long long)le64_to_cpu(gd->bg_blkno));
182 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
183 do_error("Group descriptor #%llu has an invalid "
184 "fs_generation of #%u",
185 (unsigned long long)bh->b_blocknr,
186 le32_to_cpu(gd->bg_generation));
190 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
191 do_error("Group descriptor #%llu has bit count %u but "
192 "claims that %u are free",
193 (unsigned long long)bh->b_blocknr,
194 le16_to_cpu(gd->bg_bits),
195 le16_to_cpu(gd->bg_free_bits_count));
199 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
200 do_error("Group descriptor #%llu has bit count %u but "
201 "max bitmap bits of %u",
202 (unsigned long long)bh->b_blocknr,
203 le16_to_cpu(gd->bg_bits),
204 8 * le16_to_cpu(gd->bg_size));
211 static int ocfs2_validate_gd_parent(struct super_block *sb,
212 struct ocfs2_dinode *di,
213 struct buffer_head *bh,
216 unsigned int max_bits;
217 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
219 if (di->i_blkno != gd->bg_parent_dinode) {
220 do_error("Group descriptor #%llu has bad parent "
221 "pointer (%llu, expected %llu)",
222 (unsigned long long)bh->b_blocknr,
223 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
224 (unsigned long long)le64_to_cpu(di->i_blkno));
228 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
229 if (le16_to_cpu(gd->bg_bits) > max_bits) {
230 do_error("Group descriptor #%llu has bit count of %u",
231 (unsigned long long)bh->b_blocknr,
232 le16_to_cpu(gd->bg_bits));
236 if (le16_to_cpu(gd->bg_chain) >=
237 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
238 do_error("Group descriptor #%llu has bad chain %u",
239 (unsigned long long)bh->b_blocknr,
240 le16_to_cpu(gd->bg_chain));
250 * This version only prints errors. It does not fail the filesystem, and
251 * exists only for resize.
253 int ocfs2_check_group_descriptor(struct super_block *sb,
254 struct ocfs2_dinode *di,
255 struct buffer_head *bh)
258 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
260 BUG_ON(!buffer_uptodate(bh));
263 * If the ecc fails, we return the error but otherwise
264 * leave the filesystem running. We know any error is
265 * local to this block.
267 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
270 "Checksum failed for group descriptor %llu\n",
271 (unsigned long long)bh->b_blocknr);
273 rc = ocfs2_validate_gd_self(sb, bh, 1);
275 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
280 static int ocfs2_validate_group_descriptor(struct super_block *sb,
281 struct buffer_head *bh)
284 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
286 mlog(0, "Validating group descriptor %llu\n",
287 (unsigned long long)bh->b_blocknr);
289 BUG_ON(!buffer_uptodate(bh));
292 * If the ecc fails, we return the error but otherwise
293 * leave the filesystem running. We know any error is
294 * local to this block.
296 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
301 * Errors after here are fatal.
304 return ocfs2_validate_gd_self(sb, bh, 0);
307 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
308 u64 gd_blkno, struct buffer_head **bh)
311 struct buffer_head *tmp = *bh;
313 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
314 ocfs2_validate_group_descriptor);
318 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
324 /* If ocfs2_read_block() got us a new bh, pass it up. */
332 static int ocfs2_block_group_fill(handle_t *handle,
333 struct inode *alloc_inode,
334 struct buffer_head *bg_bh,
337 struct ocfs2_chain_list *cl)
340 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
341 struct super_block * sb = alloc_inode->i_sb;
345 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
346 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
348 (unsigned long long)group_blkno,
349 (unsigned long long) bg_bh->b_blocknr);
354 status = ocfs2_journal_access_gd(handle,
355 INODE_CACHE(alloc_inode),
357 OCFS2_JOURNAL_ACCESS_CREATE);
363 memset(bg, 0, sb->s_blocksize);
364 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
365 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
366 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
367 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
368 bg->bg_chain = cpu_to_le16(my_chain);
369 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
370 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
371 bg->bg_blkno = cpu_to_le64(group_blkno);
372 /* set the 1st bit in the bitmap to account for the descriptor block */
373 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
374 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
376 status = ocfs2_journal_dirty(handle, bg_bh);
380 /* There is no need to zero out or otherwise initialize the
381 * other blocks in a group - All valid FS metadata in a block
382 * group stores the superblock fs_generation value at
383 * allocation time. */
390 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
395 while (curr < le16_to_cpu(cl->cl_count)) {
396 if (le32_to_cpu(cl->cl_recs[best].c_total) >
397 le32_to_cpu(cl->cl_recs[curr].c_total))
405 * We expect the block group allocator to already be locked.
407 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
408 struct inode *alloc_inode,
409 struct buffer_head *bh,
411 u64 *last_alloc_group,
415 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
416 struct ocfs2_chain_list *cl;
417 struct ocfs2_alloc_context *ac = NULL;
418 handle_t *handle = NULL;
419 u32 bit_off, num_bits;
422 struct buffer_head *bg_bh = NULL;
423 struct ocfs2_group_desc *bg;
425 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
429 cl = &fe->id2.i_chain;
430 status = ocfs2_reserve_clusters_with_limit(osb,
431 le16_to_cpu(cl->cl_cpg),
432 max_block, flags, &ac);
434 if (status != -ENOSPC)
439 credits = ocfs2_calc_group_alloc_credits(osb->sb,
440 le16_to_cpu(cl->cl_cpg));
441 handle = ocfs2_start_trans(osb, credits);
442 if (IS_ERR(handle)) {
443 status = PTR_ERR(handle);
449 if (last_alloc_group && *last_alloc_group != 0) {
450 mlog(0, "use old allocation group %llu for block group alloc\n",
451 (unsigned long long)*last_alloc_group);
452 ac->ac_last_group = *last_alloc_group;
454 status = ocfs2_claim_clusters(osb,
457 le16_to_cpu(cl->cl_cpg),
461 if (status != -ENOSPC)
466 alloc_rec = ocfs2_find_smallest_chain(cl);
468 /* setup the group */
469 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
470 mlog(0, "new descriptor, record %u, at block %llu\n",
471 alloc_rec, (unsigned long long)bg_blkno);
473 bg_bh = sb_getblk(osb->sb, bg_blkno);
479 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
481 status = ocfs2_block_group_fill(handle,
492 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
494 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
495 bh, OCFS2_JOURNAL_ACCESS_WRITE);
501 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
502 le16_to_cpu(bg->bg_free_bits_count));
503 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
504 cl->cl_recs[alloc_rec].c_blkno = cpu_to_le64(bg_blkno);
505 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
506 le16_add_cpu(&cl->cl_next_free_rec, 1);
508 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
509 le16_to_cpu(bg->bg_free_bits_count));
510 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
511 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
513 status = ocfs2_journal_dirty(handle, bh);
519 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
520 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
521 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
522 le32_to_cpu(fe->i_clusters)));
523 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
524 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
525 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
529 /* save the new last alloc group so that the caller can cache it. */
530 if (last_alloc_group)
531 *last_alloc_group = ac->ac_last_group;
535 ocfs2_commit_trans(osb, handle);
538 ocfs2_free_alloc_context(ac);
546 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
547 struct ocfs2_alloc_context *ac,
550 u64 *last_alloc_group,
554 u32 bits_wanted = ac->ac_bits_wanted;
555 struct inode *alloc_inode;
556 struct buffer_head *bh = NULL;
557 struct ocfs2_dinode *fe;
562 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
568 mutex_lock(&alloc_inode->i_mutex);
570 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
572 mutex_unlock(&alloc_inode->i_mutex);
579 ac->ac_inode = alloc_inode;
580 ac->ac_alloc_slot = slot;
582 fe = (struct ocfs2_dinode *) bh->b_data;
584 /* The bh was validated by the inode read inside
585 * ocfs2_inode_lock(). Any corruption is a code bug. */
586 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
588 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
589 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
590 (unsigned long long)le64_to_cpu(fe->i_blkno));
595 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
596 le32_to_cpu(fe->id1.bitmap1.i_used);
598 if (bits_wanted > free_bits) {
599 /* cluster bitmap never grows */
600 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
601 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
602 bits_wanted, free_bits);
607 if (!(flags & ALLOC_NEW_GROUP)) {
608 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
609 "and we don't alloc a new group for it.\n",
610 slot, bits_wanted, free_bits);
615 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
617 last_alloc_group, flags);
619 if (status != -ENOSPC)
623 atomic_inc(&osb->alloc_stats.bg_extends);
625 /* You should never ask for this much metadata */
627 (le32_to_cpu(fe->id1.bitmap1.i_total)
628 - le32_to_cpu(fe->id1.bitmap1.i_used)));
640 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
642 spin_lock(&osb->osb_lock);
643 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
644 spin_unlock(&osb->osb_lock);
645 atomic_set(&osb->s_num_inodes_stolen, 0);
648 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
650 spin_lock(&osb->osb_lock);
651 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
652 spin_unlock(&osb->osb_lock);
653 atomic_set(&osb->s_num_meta_stolen, 0);
656 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
658 ocfs2_init_inode_steal_slot(osb);
659 ocfs2_init_meta_steal_slot(osb);
662 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
664 spin_lock(&osb->osb_lock);
665 if (type == INODE_ALLOC_SYSTEM_INODE)
666 osb->s_inode_steal_slot = slot;
667 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
668 osb->s_meta_steal_slot = slot;
669 spin_unlock(&osb->osb_lock);
672 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
674 int slot = OCFS2_INVALID_SLOT;
676 spin_lock(&osb->osb_lock);
677 if (type == INODE_ALLOC_SYSTEM_INODE)
678 slot = osb->s_inode_steal_slot;
679 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
680 slot = osb->s_meta_steal_slot;
681 spin_unlock(&osb->osb_lock);
686 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
688 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
691 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
693 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
696 static int ocfs2_steal_resource(struct ocfs2_super *osb,
697 struct ocfs2_alloc_context *ac,
700 int i, status = -ENOSPC;
701 int slot = __ocfs2_get_steal_slot(osb, type);
703 /* Start to steal resource from the first slot after ours. */
704 if (slot == OCFS2_INVALID_SLOT)
705 slot = osb->slot_num + 1;
707 for (i = 0; i < osb->max_slots; i++, slot++) {
708 if (slot == osb->max_slots)
711 if (slot == osb->slot_num)
714 status = ocfs2_reserve_suballoc_bits(osb, ac,
717 NOT_ALLOC_NEW_GROUP);
719 __ocfs2_set_steal_slot(osb, slot, type);
723 ocfs2_free_ac_resource(ac);
729 static int ocfs2_steal_inode(struct ocfs2_super *osb,
730 struct ocfs2_alloc_context *ac)
732 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
735 static int ocfs2_steal_meta(struct ocfs2_super *osb,
736 struct ocfs2_alloc_context *ac)
738 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
741 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
743 struct ocfs2_alloc_context **ac)
746 int slot = ocfs2_get_meta_steal_slot(osb);
748 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
755 (*ac)->ac_bits_wanted = blocks;
756 (*ac)->ac_which = OCFS2_AC_USE_META;
757 (*ac)->ac_group_search = ocfs2_block_group_search;
759 if (slot != OCFS2_INVALID_SLOT &&
760 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
763 atomic_set(&osb->s_num_meta_stolen, 0);
764 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
765 EXTENT_ALLOC_SYSTEM_INODE,
766 (u32)osb->slot_num, NULL,
772 if (slot != OCFS2_INVALID_SLOT)
773 ocfs2_init_meta_steal_slot(osb);
775 } else if (status < 0 && status != -ENOSPC) {
780 ocfs2_free_ac_resource(*ac);
783 status = ocfs2_steal_meta(osb, *ac);
784 atomic_inc(&osb->s_num_meta_stolen);
786 if (status != -ENOSPC)
793 if ((status < 0) && *ac) {
794 ocfs2_free_alloc_context(*ac);
802 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
803 struct ocfs2_extent_list *root_el,
804 struct ocfs2_alloc_context **ac)
806 return ocfs2_reserve_new_metadata_blocks(osb,
807 ocfs2_extend_meta_needed(root_el),
811 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
812 struct ocfs2_alloc_context **ac)
815 int slot = ocfs2_get_inode_steal_slot(osb);
818 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
825 (*ac)->ac_bits_wanted = 1;
826 (*ac)->ac_which = OCFS2_AC_USE_INODE;
828 (*ac)->ac_group_search = ocfs2_block_group_search;
831 * stat(2) can't handle i_ino > 32bits, so we tell the
832 * lower levels not to allocate us a block group past that
833 * limit. The 'inode64' mount option avoids this behavior.
835 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
836 (*ac)->ac_max_block = (u32)~0U;
839 * slot is set when we successfully steal inode from other nodes.
840 * It is reset in 3 places:
841 * 1. when we flush the truncate log
842 * 2. when we complete local alloc recovery.
843 * 3. when we successfully allocate from our own slot.
844 * After it is set, we will go on stealing inodes until we find the
845 * need to check our slots to see whether there is some space for us.
847 if (slot != OCFS2_INVALID_SLOT &&
848 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
851 atomic_set(&osb->s_num_inodes_stolen, 0);
852 alloc_group = osb->osb_inode_alloc_group;
853 status = ocfs2_reserve_suballoc_bits(osb, *ac,
854 INODE_ALLOC_SYSTEM_INODE,
858 ALLOC_GROUPS_FROM_GLOBAL);
862 spin_lock(&osb->osb_lock);
863 osb->osb_inode_alloc_group = alloc_group;
864 spin_unlock(&osb->osb_lock);
865 mlog(0, "after reservation, new allocation group is "
866 "%llu\n", (unsigned long long)alloc_group);
869 * Some inodes must be freed by us, so try to allocate
870 * from our own next time.
872 if (slot != OCFS2_INVALID_SLOT)
873 ocfs2_init_inode_steal_slot(osb);
875 } else if (status < 0 && status != -ENOSPC) {
880 ocfs2_free_ac_resource(*ac);
883 status = ocfs2_steal_inode(osb, *ac);
884 atomic_inc(&osb->s_num_inodes_stolen);
886 if (status != -ENOSPC)
893 if ((status < 0) && *ac) {
894 ocfs2_free_alloc_context(*ac);
902 /* local alloc code has to do the same thing, so rather than do this
904 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
905 struct ocfs2_alloc_context *ac)
909 ac->ac_which = OCFS2_AC_USE_MAIN;
910 ac->ac_group_search = ocfs2_cluster_group_search;
912 status = ocfs2_reserve_suballoc_bits(osb, ac,
913 GLOBAL_BITMAP_SYSTEM_INODE,
914 OCFS2_INVALID_SLOT, NULL,
916 if (status < 0 && status != -ENOSPC) {
925 /* Callers don't need to care which bitmap (local alloc or main) to
926 * use so we figure it out for them, but unfortunately this clutters
928 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
929 u32 bits_wanted, u64 max_block,
931 struct ocfs2_alloc_context **ac)
937 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
944 (*ac)->ac_bits_wanted = bits_wanted;
945 (*ac)->ac_max_block = max_block;
948 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
949 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
950 status = ocfs2_reserve_local_alloc_bits(osb,
953 if (status == -EFBIG) {
954 /* The local alloc window is outside ac_max_block.
955 * use the main bitmap. */
957 } else if ((status < 0) && (status != -ENOSPC)) {
963 if (status == -ENOSPC) {
964 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
966 if (status != -ENOSPC)
974 if ((status < 0) && *ac) {
975 ocfs2_free_alloc_context(*ac);
983 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
985 struct ocfs2_alloc_context **ac)
987 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
988 ALLOC_NEW_GROUP, ac);
992 * More or less lifted from ext3. I'll leave their description below:
994 * "For ext3 allocations, we must not reuse any blocks which are
995 * allocated in the bitmap buffer's "last committed data" copy. This
996 * prevents deletes from freeing up the page for reuse until we have
997 * committed the delete transaction.
999 * If we didn't do this, then deleting something and reallocating it as
1000 * data would allow the old block to be overwritten before the
1001 * transaction committed (because we force data to disk before commit).
1002 * This would lead to corruption if we crashed between overwriting the
1003 * data and committing the delete.
1005 * @@@ We may want to make this allocation behaviour conditional on
1006 * data-writes at some point, and disable it for metadata allocations or
1007 * sync-data inodes."
1009 * Note: OCFS2 already does this differently for metadata vs data
1010 * allocations, as those bitmaps are separate and undo access is never
1011 * called on a metadata group descriptor.
1013 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1016 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1019 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1022 if (!buffer_jbd(bg_bh))
1025 jbd_lock_bh_state(bg_bh);
1026 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1028 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1031 jbd_unlock_bh_state(bg_bh);
1036 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1037 struct buffer_head *bg_bh,
1038 unsigned int bits_wanted,
1039 unsigned int total_bits,
1044 u16 best_offset, best_size;
1045 int offset, start, found, status = 0;
1046 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1048 /* Callers got this descriptor from
1049 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1050 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1052 found = start = best_offset = best_size = 0;
1053 bitmap = bg->bg_bitmap;
1055 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1056 if (offset == total_bits)
1059 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1060 /* We found a zero, but we can't use it as it
1061 * hasn't been put to disk yet! */
1064 } else if (offset == start) {
1065 /* we found a zero */
1067 /* move start to the next bit to test */
1070 /* got a zero after some ones */
1074 if (found > best_size) {
1076 best_offset = start - found;
1078 /* we got everything we needed */
1079 if (found == bits_wanted) {
1080 /* mlog(0, "Found it all!\n"); */
1085 /* XXX: I think the first clause is equivalent to the second
1087 if (found == bits_wanted) {
1088 *bit_off = start - found;
1089 *bits_found = found;
1090 } else if (best_size) {
1091 *bit_off = best_offset;
1092 *bits_found = best_size;
1095 /* No error log here -- see the comment above
1096 * ocfs2_test_bg_bit_allocatable */
1102 static inline int ocfs2_block_group_set_bits(handle_t *handle,
1103 struct inode *alloc_inode,
1104 struct ocfs2_group_desc *bg,
1105 struct buffer_head *group_bh,
1106 unsigned int bit_off,
1107 unsigned int num_bits)
1110 void *bitmap = bg->bg_bitmap;
1111 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1115 /* All callers get the descriptor via
1116 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1117 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1118 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1120 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1123 if (ocfs2_is_cluster_bitmap(alloc_inode))
1124 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1126 status = ocfs2_journal_access_gd(handle,
1127 INODE_CACHE(alloc_inode),
1135 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1138 ocfs2_set_bit(bit_off++, bitmap);
1140 status = ocfs2_journal_dirty(handle,
1152 /* find the one with the most empty bits */
1153 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1157 BUG_ON(!cl->cl_next_free_rec);
1160 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1161 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1162 le32_to_cpu(cl->cl_recs[best].c_free))
1167 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1171 static int ocfs2_relink_block_group(handle_t *handle,
1172 struct inode *alloc_inode,
1173 struct buffer_head *fe_bh,
1174 struct buffer_head *bg_bh,
1175 struct buffer_head *prev_bg_bh,
1179 /* there is a really tiny chance the journal calls could fail,
1180 * but we wouldn't want inconsistent blocks in *any* case. */
1181 u64 fe_ptr, bg_ptr, prev_bg_ptr;
1182 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1183 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1184 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1186 /* The caller got these descriptors from
1187 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1188 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1189 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1191 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1192 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1193 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1194 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1196 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1197 bg_ptr = le64_to_cpu(bg->bg_next_group);
1198 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1200 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1202 OCFS2_JOURNAL_ACCESS_WRITE);
1208 prev_bg->bg_next_group = bg->bg_next_group;
1210 status = ocfs2_journal_dirty(handle, prev_bg_bh);
1216 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1217 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1223 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1225 status = ocfs2_journal_dirty(handle, bg_bh);
1231 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1232 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1238 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1240 status = ocfs2_journal_dirty(handle, fe_bh);
1249 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1250 bg->bg_next_group = cpu_to_le64(bg_ptr);
1251 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1258 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1261 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1264 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1265 * value on error. */
1266 static int ocfs2_cluster_group_search(struct inode *inode,
1267 struct buffer_head *group_bh,
1268 u32 bits_wanted, u32 min_bits,
1270 u16 *bit_off, u16 *bits_found)
1272 int search = -ENOSPC;
1275 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1276 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1277 u16 tmp_off, tmp_found;
1278 unsigned int max_bits, gd_cluster_off;
1280 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1282 if (gd->bg_free_bits_count) {
1283 max_bits = le16_to_cpu(gd->bg_bits);
1285 /* Tail groups in cluster bitmaps which aren't cpg
1286 * aligned are prone to partial extention by a failed
1287 * fs resize. If the file system resize never got to
1288 * update the dinode cluster count, then we don't want
1289 * to trust any clusters past it, regardless of what
1290 * the group descriptor says. */
1291 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1292 le64_to_cpu(gd->bg_blkno));
1293 if ((gd_cluster_off + max_bits) >
1294 OCFS2_I(inode)->ip_clusters) {
1295 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1296 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1297 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1298 le16_to_cpu(gd->bg_bits),
1299 OCFS2_I(inode)->ip_clusters, max_bits);
1302 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1303 group_bh, bits_wanted,
1305 &tmp_off, &tmp_found);
1310 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1312 tmp_off + tmp_found);
1313 mlog(0, "Checking %llu against %llu\n",
1314 (unsigned long long)blkoff,
1315 (unsigned long long)max_block);
1316 if (blkoff > max_block)
1320 /* ocfs2_block_group_find_clear_bits() might
1321 * return success, but we still want to return
1322 * -ENOSPC unless it found the minimum number
1324 if (min_bits <= tmp_found) {
1326 *bits_found = tmp_found;
1327 search = 0; /* success */
1328 } else if (tmp_found) {
1330 * Don't show bits which we'll be returning
1331 * for allocation to the local alloc bitmap.
1333 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1340 static int ocfs2_block_group_search(struct inode *inode,
1341 struct buffer_head *group_bh,
1342 u32 bits_wanted, u32 min_bits,
1344 u16 *bit_off, u16 *bits_found)
1348 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1350 BUG_ON(min_bits != 1);
1351 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1353 if (bg->bg_free_bits_count) {
1354 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1355 group_bh, bits_wanted,
1356 le16_to_cpu(bg->bg_bits),
1357 bit_off, bits_found);
1358 if (!ret && max_block) {
1359 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1361 mlog(0, "Checking %llu against %llu\n",
1362 (unsigned long long)blkoff,
1363 (unsigned long long)max_block);
1364 if (blkoff > max_block)
1372 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1374 struct buffer_head *di_bh,
1380 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1381 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1383 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1384 OCFS2_JOURNAL_ACCESS_WRITE);
1390 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1391 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1392 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1394 ret = ocfs2_journal_dirty(handle, di_bh);
1402 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1407 unsigned int *num_bits,
1413 struct buffer_head *group_bh = NULL;
1414 struct ocfs2_group_desc *gd;
1415 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1416 struct inode *alloc_inode = ac->ac_inode;
1418 ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1425 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1426 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1427 ac->ac_max_block, bit_off, &found);
1436 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1438 le16_to_cpu(gd->bg_chain));
1444 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1445 *bit_off, *num_bits);
1449 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1457 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1462 unsigned int *num_bits,
1467 u16 chain, tmp_bits;
1470 struct inode *alloc_inode = ac->ac_inode;
1471 struct buffer_head *group_bh = NULL;
1472 struct buffer_head *prev_group_bh = NULL;
1473 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1474 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1475 struct ocfs2_group_desc *bg;
1477 chain = ac->ac_chain;
1478 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1480 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1482 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1483 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1489 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1492 /* for now, the chain search is a bit simplistic. We just use
1493 * the 1st group with any empty bits. */
1494 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1495 bits_wanted, min_bits,
1496 ac->ac_max_block, bit_off,
1497 &tmp_bits)) == -ENOSPC) {
1498 if (!bg->bg_next_group)
1501 brelse(prev_group_bh);
1502 prev_group_bh = NULL;
1504 next_group = le64_to_cpu(bg->bg_next_group);
1505 prev_group_bh = group_bh;
1507 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1508 next_group, &group_bh);
1513 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1516 if (status != -ENOSPC)
1521 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1522 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1524 *num_bits = tmp_bits;
1526 BUG_ON(*num_bits == 0);
1529 * Keep track of previous block descriptor read. When
1530 * we find a target, if we have read more than X
1531 * number of descriptors, and the target is reasonably
1532 * empty, relink him to top of his chain.
1534 * We've read 0 extra blocks and only send one more to
1535 * the transaction, yet the next guy to search has a
1538 * Do this *after* figuring out how many bits we're taking out
1539 * of our target group.
1541 if (ac->ac_allow_chain_relink &&
1543 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1544 status = ocfs2_relink_block_group(handle, alloc_inode,
1545 ac->ac_bh, group_bh,
1546 prev_group_bh, chain);
1553 /* Ok, claim our bits now: set the info on dinode, chainlist
1554 * and then the group */
1555 status = ocfs2_journal_access_di(handle,
1556 INODE_CACHE(alloc_inode),
1558 OCFS2_JOURNAL_ACCESS_WRITE);
1564 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1565 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1566 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1568 status = ocfs2_journal_dirty(handle,
1575 status = ocfs2_block_group_set_bits(handle,
1586 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1587 (unsigned long long)le64_to_cpu(fe->i_blkno));
1589 *bg_blkno = le64_to_cpu(bg->bg_blkno);
1590 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1593 brelse(prev_group_bh);
1599 /* will give out up to bits_wanted contiguous bits. */
1600 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1601 struct ocfs2_alloc_context *ac,
1606 unsigned int *num_bits,
1612 u64 hint_blkno = ac->ac_last_group;
1613 struct ocfs2_chain_list *cl;
1614 struct ocfs2_dinode *fe;
1618 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1619 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1622 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1624 /* The bh was validated by the inode read during
1625 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1626 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1628 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1629 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1630 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1631 "bits but only %u total.",
1632 (unsigned long long)le64_to_cpu(fe->i_blkno),
1633 le32_to_cpu(fe->id1.bitmap1.i_used),
1634 le32_to_cpu(fe->id1.bitmap1.i_total));
1640 /* Attempt to short-circuit the usual search mechanism
1641 * by jumping straight to the most recently used
1642 * allocation group. This helps us mantain some
1643 * contiguousness across allocations. */
1644 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1645 min_bits, bit_off, num_bits,
1646 hint_blkno, &bits_left);
1648 /* Be careful to update *bg_blkno here as the
1649 * caller is expecting it to be filled in, and
1650 * ocfs2_search_one_group() won't do that for
1652 *bg_blkno = hint_blkno;
1655 if (status < 0 && status != -ENOSPC) {
1661 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1663 victim = ocfs2_find_victim_chain(cl);
1664 ac->ac_chain = victim;
1665 ac->ac_allow_chain_relink = 1;
1667 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1668 num_bits, bg_blkno, &bits_left);
1671 if (status < 0 && status != -ENOSPC) {
1676 mlog(0, "Search of victim chain %u came up with nothing, "
1677 "trying all chains now.\n", victim);
1679 /* If we didn't pick a good victim, then just default to
1680 * searching each chain in order. Don't allow chain relinking
1681 * because we only calculate enough journal credits for one
1682 * relink per alloc. */
1683 ac->ac_allow_chain_relink = 0;
1684 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1687 if (!cl->cl_recs[i].c_free)
1691 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1692 bit_off, num_bits, bg_blkno,
1696 if (status < 0 && status != -ENOSPC) {
1703 if (status != -ENOSPC) {
1704 /* If the next search of this group is not likely to
1705 * yield a suitable extent, then we reset the last
1706 * group hint so as to not waste a disk read */
1707 if (bits_left < min_bits)
1708 ac->ac_last_group = 0;
1710 ac->ac_last_group = *bg_blkno;
1718 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1720 struct ocfs2_alloc_context *ac,
1722 u16 *suballoc_bit_start,
1723 unsigned int *num_bits,
1730 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1731 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1733 status = ocfs2_claim_suballoc_bits(osb,
1745 atomic_inc(&osb->alloc_stats.bg_allocs);
1747 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1748 ac->ac_bits_given += (*num_bits);
1755 static void ocfs2_init_inode_ac_group(struct inode *dir,
1756 struct buffer_head *parent_fe_bh,
1757 struct ocfs2_alloc_context *ac)
1759 struct ocfs2_dinode *fe = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1761 * Try to allocate inodes from some specific group.
1763 * If the parent dir has recorded the last group used in allocation,
1764 * cool, use it. Otherwise if we try to allocate new inode from the
1765 * same slot the parent dir belongs to, use the same chunk.
1767 * We are very careful here to avoid the mistake of setting
1768 * ac_last_group to a group descriptor from a different (unlocked) slot.
1770 if (OCFS2_I(dir)->ip_last_used_group &&
1771 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
1772 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
1773 else if (le16_to_cpu(fe->i_suballoc_slot) == ac->ac_alloc_slot)
1774 ac->ac_last_group = ocfs2_which_suballoc_group(
1775 le64_to_cpu(fe->i_blkno),
1776 le16_to_cpu(fe->i_suballoc_bit));
1779 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
1780 struct ocfs2_alloc_context *ac)
1782 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
1783 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
1786 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1789 struct buffer_head *parent_fe_bh,
1790 struct ocfs2_alloc_context *ac,
1795 unsigned int num_bits;
1801 BUG_ON(ac->ac_bits_given != 0);
1802 BUG_ON(ac->ac_bits_wanted != 1);
1803 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1805 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
1807 status = ocfs2_claim_suballoc_bits(osb,
1819 atomic_inc(&osb->alloc_stats.bg_allocs);
1821 BUG_ON(num_bits != 1);
1823 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1824 ac->ac_bits_given++;
1825 ocfs2_save_inode_ac_group(dir, ac);
1832 /* translate a group desc. blkno and it's bitmap offset into
1833 * disk cluster offset. */
1834 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1838 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1841 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1843 if (bg_blkno != osb->first_cluster_group_blkno)
1844 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1845 cluster += (u32) bg_bit_off;
1849 /* given a cluster offset, calculate which block group it belongs to
1850 * and return that block offset. */
1851 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1853 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1856 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1858 group_no = cluster / osb->bitmap_cpg;
1860 return osb->first_cluster_group_blkno;
1861 return ocfs2_clusters_to_blocks(inode->i_sb,
1862 group_no * osb->bitmap_cpg);
1865 /* given the block number of a cluster start, calculate which cluster
1866 * group and descriptor bitmap offset that corresponds to. */
1867 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1872 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1873 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1875 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1877 *bg_blkno = ocfs2_which_cluster_group(inode,
1880 if (*bg_blkno == osb->first_cluster_group_blkno)
1881 *bg_bit_off = (u16) data_cluster;
1883 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1884 data_blkno - *bg_blkno);
1888 * min_bits - minimum contiguous chunk from this total allocation we
1889 * can handle. set to what we asked for originally for a full
1890 * contig. allocation, set to '1' to indicate we can deal with extents
1893 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1895 struct ocfs2_alloc_context *ac,
1902 unsigned int bits_wanted = max_clusters;
1908 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1910 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1911 && ac->ac_which != OCFS2_AC_USE_MAIN);
1913 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1914 status = ocfs2_claim_local_alloc_bits(osb,
1921 atomic_inc(&osb->alloc_stats.local_data);
1923 if (min_clusters > (osb->bitmap_cpg - 1)) {
1924 /* The only paths asking for contiguousness
1925 * should know about this already. */
1926 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1927 "group bitmap size %u!\n", min_clusters,
1932 /* clamp the current request down to a realistic size. */
1933 if (bits_wanted > (osb->bitmap_cpg - 1))
1934 bits_wanted = osb->bitmap_cpg - 1;
1936 status = ocfs2_claim_suballoc_bits(osb,
1946 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1949 atomic_inc(&osb->alloc_stats.bitmap_data);
1953 if (status != -ENOSPC)
1958 ac->ac_bits_given += *num_clusters;
1965 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1967 struct ocfs2_alloc_context *ac,
1972 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1974 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1975 bits_wanted, cluster_start, num_clusters);
1978 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1979 struct inode *alloc_inode,
1980 struct ocfs2_group_desc *bg,
1981 struct buffer_head *group_bh,
1982 unsigned int bit_off,
1983 unsigned int num_bits)
1987 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1988 struct ocfs2_group_desc *undo_bg = NULL;
1989 int cluster_bitmap = 0;
1993 /* The caller got this descriptor from
1994 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1995 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1997 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1999 if (ocfs2_is_cluster_bitmap(alloc_inode))
2000 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
2002 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2003 group_bh, journal_type);
2009 if (ocfs2_is_cluster_bitmap(alloc_inode))
2012 if (cluster_bitmap) {
2013 jbd_lock_bh_state(group_bh);
2014 undo_bg = (struct ocfs2_group_desc *)
2015 bh2jh(group_bh)->b_committed_data;
2021 ocfs2_clear_bit((bit_off + tmp),
2022 (unsigned long *) bg->bg_bitmap);
2024 ocfs2_set_bit(bit_off + tmp,
2025 (unsigned long *) undo_bg->bg_bitmap);
2027 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2030 jbd_unlock_bh_state(group_bh);
2032 status = ocfs2_journal_dirty(handle, group_bh);
2040 * expects the suballoc inode to already be locked.
2042 int ocfs2_free_suballoc_bits(handle_t *handle,
2043 struct inode *alloc_inode,
2044 struct buffer_head *alloc_bh,
2045 unsigned int start_bit,
2051 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2052 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2053 struct buffer_head *group_bh = NULL;
2054 struct ocfs2_group_desc *group;
2058 /* The alloc_bh comes from ocfs2_free_dinode() or
2059 * ocfs2_free_clusters(). The callers have all locked the
2060 * allocator and gotten alloc_bh from the lock call. This
2061 * validates the dinode buffer. Any corruption that has happended
2063 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2064 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2066 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
2067 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
2068 (unsigned long long)bg_blkno, start_bit);
2070 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2076 group = (struct ocfs2_group_desc *) group_bh->b_data;
2078 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2080 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2088 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2089 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2095 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2097 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2098 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2100 status = ocfs2_journal_dirty(handle, alloc_bh);
2113 int ocfs2_free_dinode(handle_t *handle,
2114 struct inode *inode_alloc_inode,
2115 struct buffer_head *inode_alloc_bh,
2116 struct ocfs2_dinode *di)
2118 u64 blk = le64_to_cpu(di->i_blkno);
2119 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2120 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2122 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2123 inode_alloc_bh, bit, bg_blkno, 1);
2126 int ocfs2_free_clusters(handle_t *handle,
2127 struct inode *bitmap_inode,
2128 struct buffer_head *bitmap_bh,
2130 unsigned int num_clusters)
2135 struct ocfs2_dinode *fe;
2137 /* You can't ever have a contiguous set of clusters
2138 * bigger than a block group bitmap so we never have to worry
2139 * about looping on them. */
2143 /* This is expensive. We can safely remove once this stuff has
2144 * gotten tested really well. */
2145 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2147 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2149 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2152 mlog(0, "want to free %u clusters starting at block %llu\n",
2153 num_clusters, (unsigned long long)start_blk);
2154 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2155 (unsigned long long)bg_blkno, bg_start_bit);
2157 status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2158 bg_start_bit, bg_blkno,
2165 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2173 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2175 printk("Block Group:\n");
2176 printk("bg_signature: %s\n", bg->bg_signature);
2177 printk("bg_size: %u\n", bg->bg_size);
2178 printk("bg_bits: %u\n", bg->bg_bits);
2179 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2180 printk("bg_chain: %u\n", bg->bg_chain);
2181 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
2182 printk("bg_next_group: %llu\n",
2183 (unsigned long long)bg->bg_next_group);
2184 printk("bg_parent_dinode: %llu\n",
2185 (unsigned long long)bg->bg_parent_dinode);
2186 printk("bg_blkno: %llu\n",
2187 (unsigned long long)bg->bg_blkno);
2190 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2194 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2195 printk("i_signature: %s\n", fe->i_signature);
2196 printk("i_size: %llu\n",
2197 (unsigned long long)fe->i_size);
2198 printk("i_clusters: %u\n", fe->i_clusters);
2199 printk("i_generation: %u\n",
2200 le32_to_cpu(fe->i_generation));
2201 printk("id1.bitmap1.i_used: %u\n",
2202 le32_to_cpu(fe->id1.bitmap1.i_used));
2203 printk("id1.bitmap1.i_total: %u\n",
2204 le32_to_cpu(fe->id1.bitmap1.i_total));
2205 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2206 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2207 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2208 printk("id2.i_chain.cl_next_free_rec: %u\n",
2209 fe->id2.i_chain.cl_next_free_rec);
2210 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2211 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2212 fe->id2.i_chain.cl_recs[i].c_free);
2213 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2214 fe->id2.i_chain.cl_recs[i].c_total);
2215 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2216 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2221 * For a given allocation, determine which allocators will need to be
2222 * accessed, and lock them, reserving the appropriate number of bits.
2224 * Sparse file systems call this from ocfs2_write_begin_nolock()
2225 * and ocfs2_allocate_unwritten_extents().
2227 * File systems which don't support holes call this from
2228 * ocfs2_extend_allocation().
2230 int ocfs2_lock_allocators(struct inode *inode,
2231 struct ocfs2_extent_tree *et,
2232 u32 clusters_to_add, u32 extents_to_split,
2233 struct ocfs2_alloc_context **data_ac,
2234 struct ocfs2_alloc_context **meta_ac)
2236 int ret = 0, num_free_extents;
2237 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2238 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2244 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2246 num_free_extents = ocfs2_num_free_extents(osb, et);
2247 if (num_free_extents < 0) {
2248 ret = num_free_extents;
2254 * Sparse allocation file systems need to be more conservative
2255 * with reserving room for expansion - the actual allocation
2256 * happens while we've got a journal handle open so re-taking
2257 * a cluster lock (because we ran out of room for another
2258 * extent) will violate ordering rules.
2260 * Most of the time we'll only be seeing this 1 cluster at a time
2263 * Always lock for any unwritten extents - we might want to
2264 * add blocks during a split.
2266 if (!num_free_extents ||
2267 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2268 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2276 if (clusters_to_add == 0)
2279 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2289 ocfs2_free_alloc_context(*meta_ac);
2294 * We cannot have an error and a non null *data_ac.
2302 * Read the inode specified by blkno to get suballoc_slot and
2305 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2306 u16 *suballoc_slot, u16 *suballoc_bit)
2309 struct buffer_head *inode_bh = NULL;
2310 struct ocfs2_dinode *inode_fe;
2312 mlog_entry("blkno: %llu\n", (unsigned long long)blkno);
2314 /* dirty read disk */
2315 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2317 mlog(ML_ERROR, "read block %llu failed %d\n",
2318 (unsigned long long)blkno, status);
2322 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2323 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2324 mlog(ML_ERROR, "invalid inode %llu requested\n",
2325 (unsigned long long)blkno);
2330 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2331 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2332 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2333 (unsigned long long)blkno,
2334 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2340 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2342 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2352 * test whether bit is SET in allocator bitmap or not. on success, 0
2353 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2354 * is returned and *res is meaningless. Call this after you have
2355 * cluster locked against suballoc, or you may get a result based on
2356 * non-up2date contents
2358 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2359 struct inode *suballoc,
2360 struct buffer_head *alloc_bh, u64 blkno,
2363 struct ocfs2_dinode *alloc_fe;
2364 struct ocfs2_group_desc *group;
2365 struct buffer_head *group_bh = NULL;
2369 mlog_entry("blkno: %llu bit: %u\n", (unsigned long long)blkno,
2372 alloc_fe = (struct ocfs2_dinode *)alloc_bh->b_data;
2373 if ((bit + 1) > ocfs2_bits_per_group(&alloc_fe->id2.i_chain)) {
2374 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2376 ocfs2_bits_per_group(&alloc_fe->id2.i_chain));
2381 bg_blkno = ocfs2_which_suballoc_group(blkno, bit);
2382 status = ocfs2_read_group_descriptor(suballoc, alloc_fe, bg_blkno,
2385 mlog(ML_ERROR, "read group %llu failed %d\n",
2386 (unsigned long long)bg_blkno, status);
2390 group = (struct ocfs2_group_desc *) group_bh->b_data;
2391 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2401 * Test if the bit representing this inode (blkno) is set in the
2404 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2406 * In the event of failure, a negative value is returned and *res is
2409 * Callers must make sure to hold nfs_sync_lock to prevent
2410 * ocfs2_delete_inode() on another node from accessing the same
2411 * suballocator concurrently.
2413 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2416 u16 suballoc_bit = 0, suballoc_slot = 0;
2417 struct inode *inode_alloc_inode;
2418 struct buffer_head *alloc_bh = NULL;
2420 mlog_entry("blkno: %llu", (unsigned long long)blkno);
2422 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2425 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2430 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2432 if (!inode_alloc_inode) {
2433 /* the error code could be inaccurate, but we are not able to
2434 * get the correct one. */
2436 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2437 (u32)suballoc_slot);
2441 mutex_lock(&inode_alloc_inode->i_mutex);
2442 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2444 mutex_unlock(&inode_alloc_inode->i_mutex);
2445 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2446 (u32)suballoc_slot, status);
2450 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2451 blkno, suballoc_bit, res);
2453 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2455 ocfs2_inode_unlock(inode_alloc_inode, 0);
2456 mutex_unlock(&inode_alloc_inode->i_mutex);
2458 iput(inode_alloc_inode);