1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Node local data allocation
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/bitops.h>
31 #include <linux/debugfs.h>
33 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
34 #include <cluster/masklog.h>
42 #include "localalloc.h"
47 #include "buffer_head_io.h"
49 #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
51 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
53 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
54 struct ocfs2_dinode *alloc,
57 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
59 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
61 struct ocfs2_dinode *alloc,
62 struct inode *main_bm_inode,
63 struct buffer_head *main_bm_bh);
65 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
66 struct ocfs2_alloc_context **ac,
67 struct inode **bitmap_inode,
68 struct buffer_head **bitmap_bh);
70 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
72 struct ocfs2_alloc_context *ac);
74 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
75 struct inode *local_alloc_inode);
77 #ifdef CONFIG_OCFS2_FS_STATS
79 DEFINE_MUTEX(la_debug_mutex);
81 static int ocfs2_la_debug_open(struct inode *inode, struct file *file)
83 file->private_data = inode->i_private;
87 #define LA_DEBUG_BUF_SZ PAGE_CACHE_SIZE
88 #define LA_DEBUG_VER 1
89 static ssize_t ocfs2_la_debug_read(struct file *file, char __user *userbuf,
90 size_t count, loff_t *ppos)
92 struct ocfs2_super *osb = file->private_data;
94 char *buf = osb->local_alloc_debug_buf;
96 mutex_lock(&la_debug_mutex);
97 memset(buf, 0, LA_DEBUG_BUF_SZ);
99 written = snprintf(buf, LA_DEBUG_BUF_SZ,
100 "0x%x\t0x%llx\t%u\t%u\t0x%x\n",
102 (unsigned long long)osb->la_last_gd,
103 osb->local_alloc_default_bits,
104 osb->local_alloc_bits, osb->local_alloc_state);
106 ret = simple_read_from_buffer(userbuf, count, ppos, buf, written);
108 mutex_unlock(&la_debug_mutex);
112 static const struct file_operations ocfs2_la_debug_fops = {
113 .open = ocfs2_la_debug_open,
114 .read = ocfs2_la_debug_read,
117 static void ocfs2_init_la_debug(struct ocfs2_super *osb)
119 osb->local_alloc_debug_buf = kmalloc(LA_DEBUG_BUF_SZ, GFP_NOFS);
120 if (!osb->local_alloc_debug_buf)
123 osb->local_alloc_debug = debugfs_create_file("local_alloc_stats",
127 &ocfs2_la_debug_fops);
128 if (!osb->local_alloc_debug) {
129 kfree(osb->local_alloc_debug_buf);
130 osb->local_alloc_debug_buf = NULL;
134 static void ocfs2_shutdown_la_debug(struct ocfs2_super *osb)
136 if (osb->local_alloc_debug)
137 debugfs_remove(osb->local_alloc_debug);
139 if (osb->local_alloc_debug_buf)
140 kfree(osb->local_alloc_debug_buf);
142 osb->local_alloc_debug_buf = NULL;
143 osb->local_alloc_debug = NULL;
145 #else /* CONFIG_OCFS2_FS_STATS */
146 static void ocfs2_init_la_debug(struct ocfs2_super *osb)
150 static void ocfs2_shutdown_la_debug(struct ocfs2_super *osb)
156 static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
158 return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
159 osb->local_alloc_state == OCFS2_LA_ENABLED);
162 void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
163 unsigned int num_clusters)
165 spin_lock(&osb->osb_lock);
166 if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
167 osb->local_alloc_state == OCFS2_LA_THROTTLED)
168 if (num_clusters >= osb->local_alloc_default_bits) {
169 cancel_delayed_work(&osb->la_enable_wq);
170 osb->local_alloc_state = OCFS2_LA_ENABLED;
172 spin_unlock(&osb->osb_lock);
175 void ocfs2_la_enable_worker(struct work_struct *work)
177 struct ocfs2_super *osb =
178 container_of(work, struct ocfs2_super,
180 spin_lock(&osb->osb_lock);
181 osb->local_alloc_state = OCFS2_LA_ENABLED;
182 spin_unlock(&osb->osb_lock);
186 * Tell us whether a given allocation should use the local alloc
187 * file. Otherwise, it has to go to the main bitmap.
189 * This function does semi-dirty reads of local alloc size and state!
190 * This is ok however, as the values are re-checked once under mutex.
192 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
197 spin_lock(&osb->osb_lock);
198 la_bits = osb->local_alloc_bits;
200 if (!ocfs2_la_state_enabled(osb))
203 /* la_bits should be at least twice the size (in clusters) of
204 * a new block group. We want to be sure block group
205 * allocations go through the local alloc, so allow an
206 * allocation to take up to half the bitmap. */
207 if (bits > (la_bits / 2))
212 mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
213 osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
214 spin_unlock(&osb->osb_lock);
218 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
221 struct ocfs2_dinode *alloc = NULL;
222 struct buffer_head *alloc_bh = NULL;
224 struct inode *inode = NULL;
225 struct ocfs2_local_alloc *la;
229 ocfs2_init_la_debug(osb);
231 if (osb->local_alloc_bits == 0)
234 if (osb->local_alloc_bits >= osb->bitmap_cpg) {
235 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
236 "than max possible %u. Using defaults.\n",
237 osb->local_alloc_bits, (osb->bitmap_cpg - 1));
238 osb->local_alloc_bits =
239 ocfs2_megabytes_to_clusters(osb->sb,
240 OCFS2_DEFAULT_LOCAL_ALLOC_SIZE);
243 /* read the alloc off disk */
244 inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
252 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
253 &alloc_bh, 0, inode);
259 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
260 la = OCFS2_LOCAL_ALLOC(alloc);
262 if (!(le32_to_cpu(alloc->i_flags) &
263 (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
264 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
265 (unsigned long long)OCFS2_I(inode)->ip_blkno);
270 if ((la->la_size == 0) ||
271 (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
272 mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
273 le16_to_cpu(la->la_size));
278 /* do a little verification. */
279 num_used = ocfs2_local_alloc_count_bits(alloc);
281 /* hopefully the local alloc has always been recovered before
284 || alloc->id1.bitmap1.i_used
285 || alloc->id1.bitmap1.i_total
287 mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
288 "found = %u, set = %u, taken = %u, off = %u\n",
289 num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
290 le32_to_cpu(alloc->id1.bitmap1.i_total),
291 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
293 osb->local_alloc_bh = alloc_bh;
294 osb->local_alloc_state = OCFS2_LA_ENABLED;
304 ocfs2_shutdown_la_debug(osb);
306 mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits);
313 * return any unused bits to the bitmap and write out a clean
316 * local_alloc_bh is optional. If not passed, we will simply use the
317 * one off osb. If you do pass it however, be warned that it *will* be
318 * returned brelse'd and NULL'd out.*/
319 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
323 struct inode *local_alloc_inode = NULL;
324 struct buffer_head *bh = NULL;
325 struct buffer_head *main_bm_bh = NULL;
326 struct inode *main_bm_inode = NULL;
327 struct ocfs2_dinode *alloc_copy = NULL;
328 struct ocfs2_dinode *alloc = NULL;
332 cancel_delayed_work(&osb->la_enable_wq);
333 flush_workqueue(ocfs2_wq);
335 ocfs2_shutdown_la_debug(osb);
337 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
341 ocfs2_get_system_file_inode(osb,
342 LOCAL_ALLOC_SYSTEM_INODE,
344 if (!local_alloc_inode) {
350 osb->local_alloc_state = OCFS2_LA_DISABLED;
352 main_bm_inode = ocfs2_get_system_file_inode(osb,
353 GLOBAL_BITMAP_SYSTEM_INODE,
355 if (!main_bm_inode) {
361 mutex_lock(&main_bm_inode->i_mutex);
363 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
369 /* WINDOW_MOVE_CREDITS is a bit heavy... */
370 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
371 if (IS_ERR(handle)) {
372 mlog_errno(PTR_ERR(handle));
377 bh = osb->local_alloc_bh;
378 alloc = (struct ocfs2_dinode *) bh->b_data;
380 alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
385 memcpy(alloc_copy, alloc, bh->b_size);
387 status = ocfs2_journal_access(handle, local_alloc_inode, bh,
388 OCFS2_JOURNAL_ACCESS_WRITE);
394 ocfs2_clear_local_alloc(alloc);
396 status = ocfs2_journal_dirty(handle, bh);
403 osb->local_alloc_bh = NULL;
404 osb->local_alloc_state = OCFS2_LA_UNUSED;
406 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
407 main_bm_inode, main_bm_bh);
412 ocfs2_commit_trans(osb, handle);
418 ocfs2_inode_unlock(main_bm_inode, 1);
421 mutex_unlock(&main_bm_inode->i_mutex);
425 if (local_alloc_inode)
426 iput(local_alloc_inode);
435 * We want to free the bitmap bits outside of any recovery context as
436 * we'll need a cluster lock to do so, but we must clear the local
437 * alloc before giving up the recovered nodes journal. To solve this,
438 * we kmalloc a copy of the local alloc before it's change for the
439 * caller to process with ocfs2_complete_local_alloc_recovery
441 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
443 struct ocfs2_dinode **alloc_copy)
446 struct buffer_head *alloc_bh = NULL;
447 struct inode *inode = NULL;
448 struct ocfs2_dinode *alloc;
450 mlog_entry("(slot_num = %d)\n", slot_num);
454 inode = ocfs2_get_system_file_inode(osb,
455 LOCAL_ALLOC_SYSTEM_INODE,
463 mutex_lock(&inode->i_mutex);
465 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
466 &alloc_bh, 0, inode);
472 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
473 if (!(*alloc_copy)) {
477 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
479 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
480 ocfs2_clear_local_alloc(alloc);
482 status = ocfs2_write_block(osb, alloc_bh, inode);
487 if ((status < 0) && (*alloc_copy)) {
496 mutex_unlock(&inode->i_mutex);
505 * Step 2: By now, we've completed the journal recovery, we've stamped
506 * a clean local alloc on disk and dropped the node out of the
507 * recovery map. Dlm locks will no longer stall, so lets clear out the
510 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
511 struct ocfs2_dinode *alloc)
515 struct buffer_head *main_bm_bh = NULL;
516 struct inode *main_bm_inode;
520 main_bm_inode = ocfs2_get_system_file_inode(osb,
521 GLOBAL_BITMAP_SYSTEM_INODE,
523 if (!main_bm_inode) {
529 mutex_lock(&main_bm_inode->i_mutex);
531 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
537 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
538 if (IS_ERR(handle)) {
539 status = PTR_ERR(handle);
545 /* we want the bitmap change to be recorded on disk asap */
548 status = ocfs2_sync_local_to_main(osb, handle, alloc,
549 main_bm_inode, main_bm_bh);
553 ocfs2_commit_trans(osb, handle);
556 ocfs2_inode_unlock(main_bm_inode, 1);
559 mutex_unlock(&main_bm_inode->i_mutex);
568 ocfs2_init_inode_steal_slot(osb);
573 /* Check to see if the local alloc window is within ac->ac_max_block */
574 static int ocfs2_local_alloc_in_range(struct inode *inode,
575 struct ocfs2_alloc_context *ac,
578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
579 struct ocfs2_dinode *alloc;
580 struct ocfs2_local_alloc *la;
584 if (!ac->ac_max_block)
587 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
588 la = OCFS2_LOCAL_ALLOC(alloc);
590 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
597 * Converting (bm_off + start + bits_wanted) to blocks gives us
598 * the blkno just past our actual allocation. This is perfect
599 * to compare with ac_max_block.
601 block_off = ocfs2_clusters_to_blocks(inode->i_sb,
602 le32_to_cpu(la->la_bm_off) +
603 start + bits_wanted);
604 mlog(0, "Checking %llu against %llu\n",
605 (unsigned long long)block_off,
606 (unsigned long long)ac->ac_max_block);
607 if (block_off > ac->ac_max_block)
614 * make sure we've got at least bits_wanted contiguous bits in the
615 * local alloc. You lose them when you drop i_mutex.
617 * We will add ourselves to the transaction passed in, but may start
618 * our own in order to shift windows.
620 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
622 struct ocfs2_alloc_context *ac)
625 struct ocfs2_dinode *alloc;
626 struct inode *local_alloc_inode;
627 unsigned int free_bits;
634 ocfs2_get_system_file_inode(osb,
635 LOCAL_ALLOC_SYSTEM_INODE,
637 if (!local_alloc_inode) {
643 mutex_lock(&local_alloc_inode->i_mutex);
646 * We must double check state and allocator bits because
647 * another process may have changed them while holding i_mutex.
649 spin_lock(&osb->osb_lock);
650 if (!ocfs2_la_state_enabled(osb) ||
651 (bits_wanted > osb->local_alloc_bits)) {
652 spin_unlock(&osb->osb_lock);
656 spin_unlock(&osb->osb_lock);
658 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
660 #ifdef CONFIG_OCFS2_DEBUG_FS
661 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
662 ocfs2_local_alloc_count_bits(alloc)) {
663 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
664 "%u free bits, but a count shows %u",
665 (unsigned long long)le64_to_cpu(alloc->i_blkno),
666 le32_to_cpu(alloc->id1.bitmap1.i_used),
667 ocfs2_local_alloc_count_bits(alloc));
673 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
674 le32_to_cpu(alloc->id1.bitmap1.i_used);
675 if (bits_wanted > free_bits) {
676 /* uhoh, window change time. */
678 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
680 if (status != -ENOSPC)
686 * Under certain conditions, the window slide code
687 * might have reduced the number of bits available or
688 * disabled the the local alloc entirely. Re-check
689 * here and return -ENOSPC if necessary.
692 if (!ocfs2_la_state_enabled(osb))
695 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
696 le32_to_cpu(alloc->id1.bitmap1.i_used);
697 if (bits_wanted > free_bits)
701 if (ac->ac_max_block)
702 mlog(0, "Calling in_range for max block %llu\n",
703 (unsigned long long)ac->ac_max_block);
705 if (!ocfs2_local_alloc_in_range(local_alloc_inode, ac,
708 * The window is outside ac->ac_max_block.
709 * This errno tells the caller to keep localalloc enabled
710 * but to get the allocation from the main bitmap.
716 ac->ac_inode = local_alloc_inode;
717 /* We should never use localalloc from another slot */
718 ac->ac_alloc_slot = osb->slot_num;
719 ac->ac_which = OCFS2_AC_USE_LOCAL;
720 get_bh(osb->local_alloc_bh);
721 ac->ac_bh = osb->local_alloc_bh;
724 if (status < 0 && local_alloc_inode) {
725 mutex_unlock(&local_alloc_inode->i_mutex);
726 iput(local_alloc_inode);
729 mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num,
736 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
738 struct ocfs2_alloc_context *ac,
744 struct inode *local_alloc_inode;
746 struct ocfs2_dinode *alloc;
747 struct ocfs2_local_alloc *la;
750 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
752 local_alloc_inode = ac->ac_inode;
753 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
754 la = OCFS2_LOCAL_ALLOC(alloc);
756 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
758 /* TODO: Shouldn't we just BUG here? */
764 bitmap = la->la_bitmap;
765 *bit_off = le32_to_cpu(la->la_bm_off) + start;
766 /* local alloc is always contiguous by nature -- we never
767 * delete bits from it! */
768 *num_bits = bits_wanted;
770 status = ocfs2_journal_access(handle, local_alloc_inode,
772 OCFS2_JOURNAL_ACCESS_WRITE);
779 ocfs2_set_bit(start++, bitmap);
781 le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
783 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
795 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
800 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
804 buffer = la->la_bitmap;
805 for (i = 0; i < le16_to_cpu(la->la_size); i++)
806 count += hweight8(buffer[i]);
812 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
813 struct ocfs2_dinode *alloc,
816 int numfound, bitoff, left, startoff, lastzero;
819 mlog_entry("(numbits wanted = %u)\n", numbits);
821 if (!alloc->id1.bitmap1.i_total) {
822 mlog(0, "No bits in my window!\n");
827 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
829 numfound = bitoff = startoff = 0;
831 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
832 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
833 if (bitoff == left) {
834 /* mlog(0, "bitoff (%d) == left", bitoff); */
837 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
838 "numfound = %d\n", bitoff, startoff, numfound);*/
840 /* Ok, we found a zero bit... is it contig. or do we
842 if (bitoff == startoff) {
843 /* we found a zero */
847 /* got a zero after some ones */
851 /* we got everything we needed */
852 if (numfound == numbits) {
853 /* mlog(0, "Found it all!\n"); */
858 mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
861 if (numfound == numbits)
862 bitoff = startoff - numfound;
871 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
873 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
877 alloc->id1.bitmap1.i_total = 0;
878 alloc->id1.bitmap1.i_used = 0;
880 for(i = 0; i < le16_to_cpu(la->la_size); i++)
881 la->la_bitmap[i] = 0;
887 /* turn this on and uncomment below to aid debugging window shifts. */
888 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
892 unsigned int tmp = count;
894 if (ocfs2_test_bit(start + tmp, bitmap)) {
895 printk("ocfs2_verify_zero_bits: start = %u, count = "
896 "%u\n", start, count);
897 printk("ocfs2_verify_zero_bits: bit %u is set!",
906 * sync the local alloc to main bitmap.
908 * assumes you've already locked the main bitmap -- the bitmap inode
909 * passed is used for caching.
911 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
913 struct ocfs2_dinode *alloc,
914 struct inode *main_bm_inode,
915 struct buffer_head *main_bm_bh)
918 int bit_off, left, count, start;
922 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
924 mlog_entry("total = %u, used = %u\n",
925 le32_to_cpu(alloc->id1.bitmap1.i_total),
926 le32_to_cpu(alloc->id1.bitmap1.i_used));
928 if (!alloc->id1.bitmap1.i_total) {
929 mlog(0, "nothing to sync!\n");
933 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
934 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
935 mlog(0, "all bits were taken!\n");
939 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
940 le32_to_cpu(la->la_bm_off));
941 bitmap = la->la_bitmap;
942 start = count = bit_off = 0;
943 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
945 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
947 if ((bit_off < left) && (bit_off == start)) {
953 blkno = la_start_blk +
954 ocfs2_clusters_to_blocks(osb->sb,
957 mlog(0, "freeing %u bits starting at local alloc bit "
958 "%u (la_start_blk = %llu, blkno = %llu)\n",
959 count, start - count,
960 (unsigned long long)la_start_blk,
961 (unsigned long long)blkno);
963 status = ocfs2_free_clusters(handle, main_bm_inode,
964 main_bm_bh, blkno, count);
981 enum ocfs2_la_event {
982 OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */
983 OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has
984 * enough bits theoretically
985 * free, but a contiguous
986 * allocation could not be
988 OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have
989 * enough bits free to satisfy
992 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
994 * Given an event, calculate the size of our next local alloc window.
996 * This should always be called under i_mutex of the local alloc inode
997 * so that local alloc disabling doesn't race with processes trying to
1000 * Returns the state which the local alloc was left in. This value can
1001 * be ignored by some paths.
1003 static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
1004 enum ocfs2_la_event event)
1009 spin_lock(&osb->osb_lock);
1010 if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
1011 WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
1016 * ENOSPC and fragmentation are treated similarly for now.
1018 if (event == OCFS2_LA_EVENT_ENOSPC ||
1019 event == OCFS2_LA_EVENT_FRAGMENTED) {
1021 * We ran out of contiguous space in the primary
1022 * bitmap. Drastically reduce the number of bits used
1023 * by local alloc until we have to disable it.
1025 bits = osb->local_alloc_bits >> 1;
1026 if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
1028 * By setting state to THROTTLED, we'll keep
1029 * the number of local alloc bits used down
1030 * until an event occurs which would give us
1031 * reason to assume the bitmap situation might
1034 osb->local_alloc_state = OCFS2_LA_THROTTLED;
1035 osb->local_alloc_bits = bits;
1037 osb->local_alloc_state = OCFS2_LA_DISABLED;
1039 queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
1040 OCFS2_LA_ENABLE_INTERVAL);
1045 * Don't increase the size of the local alloc window until we
1046 * know we might be able to fulfill the request. Otherwise, we
1047 * risk bouncing around the global bitmap during periods of
1050 if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
1051 osb->local_alloc_bits = osb->local_alloc_default_bits;
1054 state = osb->local_alloc_state;
1055 spin_unlock(&osb->osb_lock);
1060 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
1061 struct ocfs2_alloc_context **ac,
1062 struct inode **bitmap_inode,
1063 struct buffer_head **bitmap_bh)
1067 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1075 (*ac)->ac_bits_wanted = osb->local_alloc_bits;
1077 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1078 if (status == -ENOSPC) {
1079 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
1083 ocfs2_free_ac_resource(*ac);
1084 memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
1092 *bitmap_inode = (*ac)->ac_inode;
1093 igrab(*bitmap_inode);
1094 *bitmap_bh = (*ac)->ac_bh;
1098 if ((status < 0) && *ac) {
1099 ocfs2_free_alloc_context(*ac);
1108 * pass it the bitmap lock in lock_bh if you have it.
1110 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
1112 struct ocfs2_alloc_context *ac)
1115 u32 cluster_off, cluster_count;
1116 struct ocfs2_dinode *alloc = NULL;
1117 struct ocfs2_local_alloc *la;
1121 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1122 la = OCFS2_LOCAL_ALLOC(alloc);
1124 if (alloc->id1.bitmap1.i_total)
1125 mlog(0, "asking me to alloc a new window over a non-empty "
1128 mlog(0, "Allocating %u clusters for a new window.\n",
1129 osb->local_alloc_bits);
1131 /* Instruct the allocation code to try the most recently used
1132 * cluster group. We'll re-record the group used this pass
1134 ac->ac_last_group = osb->la_last_gd;
1136 /* we used the generic suballoc reserve function, but we set
1137 * everything up nicely, so there's no reason why we can't use
1138 * the more specific cluster api to claim bits. */
1139 status = ocfs2_claim_clusters(osb, handle, ac, osb->local_alloc_bits,
1140 &cluster_off, &cluster_count);
1141 if (status == -ENOSPC) {
1144 * Note: We could also try syncing the journal here to
1145 * allow use of any free bits which the current
1146 * transaction can't give us access to. --Mark
1148 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1152 status = ocfs2_claim_clusters(osb, handle, ac,
1153 osb->local_alloc_bits,
1156 if (status == -ENOSPC)
1159 * We only shrunk the *minimum* number of in our
1160 * request - it's entirely possible that the allocator
1161 * might give us more than we asked for.
1164 spin_lock(&osb->osb_lock);
1165 osb->local_alloc_bits = cluster_count;
1166 spin_unlock(&osb->osb_lock);
1170 if (status != -ENOSPC)
1175 osb->la_last_gd = ac->ac_last_group;
1177 la->la_bm_off = cpu_to_le32(cluster_off);
1178 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1179 /* just in case... In the future when we find space ourselves,
1180 * we don't have to get all contiguous -- but we'll have to
1181 * set all previously used bits in bitmap and update
1182 * la_bits_set before setting the bits in the main bitmap. */
1183 alloc->id1.bitmap1.i_used = 0;
1184 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1185 le16_to_cpu(la->la_size));
1187 mlog(0, "New window allocated:\n");
1188 mlog(0, "window la_bm_off = %u\n",
1189 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
1190 mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
1197 /* Note that we do *NOT* lock the local alloc inode here as
1198 * it's been locked already for us. */
1199 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1200 struct inode *local_alloc_inode)
1203 struct buffer_head *main_bm_bh = NULL;
1204 struct inode *main_bm_inode = NULL;
1205 handle_t *handle = NULL;
1206 struct ocfs2_dinode *alloc;
1207 struct ocfs2_dinode *alloc_copy = NULL;
1208 struct ocfs2_alloc_context *ac = NULL;
1212 ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1214 /* This will lock the main bitmap for us. */
1215 status = ocfs2_local_alloc_reserve_for_window(osb,
1220 if (status != -ENOSPC)
1225 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
1226 if (IS_ERR(handle)) {
1227 status = PTR_ERR(handle);
1233 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1235 /* We want to clear the local alloc before doing anything
1236 * else, so that if we error later during this operation,
1237 * local alloc shutdown won't try to double free main bitmap
1238 * bits. Make a copy so the sync function knows which bits to
1240 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
1246 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1248 status = ocfs2_journal_access(handle, local_alloc_inode,
1249 osb->local_alloc_bh,
1250 OCFS2_JOURNAL_ACCESS_WRITE);
1256 ocfs2_clear_local_alloc(alloc);
1258 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
1264 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1265 main_bm_inode, main_bm_bh);
1271 status = ocfs2_local_alloc_new_window(osb, handle, ac);
1273 if (status != -ENOSPC)
1278 atomic_inc(&osb->alloc_stats.moves);
1283 ocfs2_commit_trans(osb, handle);
1289 iput(main_bm_inode);
1295 ocfs2_free_alloc_context(ac);