]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/ubifs/gc.c
2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements garbage collection. The procedure for garbage collection
25 * is different depending on whether a LEB as an index LEB (contains index
26 * nodes) or not. For non-index LEBs, garbage collection finds a LEB which
27 * contains a lot of dirty space (obsolete nodes), and copies the non-obsolete
28 * nodes to the journal, at which point the garbage-collected LEB is free to be
29 * reused. For index LEBs, garbage collection marks the non-obsolete index nodes
30 * dirty in the TNC, and after the next commit, the garbage-collected LEB is
31 * to be reused. Garbage collection will cause the number of dirty index nodes
32 * to grow, however sufficient space is reserved for the index to ensure the
33 * commit will never run out of space.
35 * Notes about dead watermark. At current UBIFS implementation we assume that
36 * LEBs which have less than @c->dead_wm bytes of free + dirty space are full
37 * and not worth garbage-collecting. The dead watermark is one min. I/O unit
38 * size, or min. UBIFS node size, depending on what is greater. Indeed, UBIFS
39 * Garbage Collector has to synchronize the GC head's write buffer before
40 * returning, so this is about wasting one min. I/O unit. However, UBIFS GC can
41 * actually reclaim even very small pieces of dirty space by garbage collecting
42 * enough dirty LEBs, but we do not bother doing this at this implementation.
44 * Notes about dark watermark. The results of GC work depends on how big are
45 * the UBIFS nodes GC deals with. Large nodes make GC waste more space. Indeed,
46 * if GC move data from LEB A to LEB B and nodes in LEB A are large, GC would
47 * have to waste large pieces of free space at the end of LEB B, because nodes
48 * from LEB A would not fit. And the worst situation is when all nodes are of
49 * maximum size. So dark watermark is the amount of free + dirty space in LEB
50 * which are guaranteed to be reclaimable. If LEB has less space, the GC might
51 * be unable to reclaim it. So, LEBs with free + dirty greater than dark
52 * watermark are "good" LEBs from GC's point of view. The other LEBs are not so
53 * good, and GC takes extra care when moving them.
56 #include <linux/slab.h>
57 #include <linux/pagemap.h>
58 #include <linux/list_sort.h>
62 * GC may need to move more than one LEB to make progress. The below constants
63 * define "soft" and "hard" limits on the number of LEBs the garbage collector
66 #define SOFT_LEBS_LIMIT 4
67 #define HARD_LEBS_LIMIT 32
70 * switch_gc_head - switch the garbage collection journal head.
71 * @c: UBIFS file-system description object
72 * @buf: buffer to write
73 * @len: length of the buffer to write
74 * @lnum: LEB number written is returned here
75 * @offs: offset written is returned here
77 * This function switch the GC head to the next LEB which is reserved in
78 * @c->gc_lnum. Returns %0 in case of success, %-EAGAIN if commit is required,
79 * and other negative error code in case of failures.
81 static int switch_gc_head(struct ubifs_info
*c
)
83 int err
, gc_lnum
= c
->gc_lnum
;
84 struct ubifs_wbuf
*wbuf
= &c
->jheads
[GCHD
].wbuf
;
86 ubifs_assert(c
, gc_lnum
!= -1);
87 dbg_gc("switch GC head from LEB %d:%d to LEB %d (waste %d bytes)",
88 wbuf
->lnum
, wbuf
->offs
+ wbuf
->used
, gc_lnum
,
89 c
->leb_size
- wbuf
->offs
- wbuf
->used
);
91 err
= ubifs_wbuf_sync_nolock(wbuf
);
96 * The GC write-buffer was synchronized, we may safely unmap
99 err
= ubifs_leb_unmap(c
, gc_lnum
);
103 err
= ubifs_add_bud_to_log(c
, GCHD
, gc_lnum
, 0);
108 err
= ubifs_wbuf_seek_nolock(wbuf
, gc_lnum
, 0);
113 * data_nodes_cmp - compare 2 data nodes.
114 * @priv: UBIFS file-system description object
115 * @a: first data node
116 * @b: second data node
118 * This function compares data nodes @a and @b. Returns %1 if @a has greater
119 * inode or block number, and %-1 otherwise.
121 static int data_nodes_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
124 struct ubifs_info
*c
= priv
;
125 struct ubifs_scan_node
*sa
, *sb
;
131 sa
= list_entry(a
, struct ubifs_scan_node
, list
);
132 sb
= list_entry(b
, struct ubifs_scan_node
, list
);
134 ubifs_assert(c
, key_type(c
, &sa
->key
) == UBIFS_DATA_KEY
);
135 ubifs_assert(c
, key_type(c
, &sb
->key
) == UBIFS_DATA_KEY
);
136 ubifs_assert(c
, sa
->type
== UBIFS_DATA_NODE
);
137 ubifs_assert(c
, sb
->type
== UBIFS_DATA_NODE
);
139 inuma
= key_inum(c
, &sa
->key
);
140 inumb
= key_inum(c
, &sb
->key
);
142 if (inuma
== inumb
) {
143 unsigned int blka
= key_block(c
, &sa
->key
);
144 unsigned int blkb
= key_block(c
, &sb
->key
);
148 } else if (inuma
<= inumb
)
155 * nondata_nodes_cmp - compare 2 non-data nodes.
156 * @priv: UBIFS file-system description object
160 * This function compares nodes @a and @b. It makes sure that inode nodes go
161 * first and sorted by length in descending order. Directory entry nodes go
162 * after inode nodes and are sorted in ascending hash valuer order.
164 static int nondata_nodes_cmp(void *priv
, struct list_head
*a
,
168 struct ubifs_info
*c
= priv
;
169 struct ubifs_scan_node
*sa
, *sb
;
175 sa
= list_entry(a
, struct ubifs_scan_node
, list
);
176 sb
= list_entry(b
, struct ubifs_scan_node
, list
);
178 ubifs_assert(c
, key_type(c
, &sa
->key
) != UBIFS_DATA_KEY
&&
179 key_type(c
, &sb
->key
) != UBIFS_DATA_KEY
);
180 ubifs_assert(c
, sa
->type
!= UBIFS_DATA_NODE
&&
181 sb
->type
!= UBIFS_DATA_NODE
);
183 /* Inodes go before directory entries */
184 if (sa
->type
== UBIFS_INO_NODE
) {
185 if (sb
->type
== UBIFS_INO_NODE
)
186 return sb
->len
- sa
->len
;
189 if (sb
->type
== UBIFS_INO_NODE
)
192 ubifs_assert(c
, key_type(c
, &sa
->key
) == UBIFS_DENT_KEY
||
193 key_type(c
, &sa
->key
) == UBIFS_XENT_KEY
);
194 ubifs_assert(c
, key_type(c
, &sb
->key
) == UBIFS_DENT_KEY
||
195 key_type(c
, &sb
->key
) == UBIFS_XENT_KEY
);
196 ubifs_assert(c
, sa
->type
== UBIFS_DENT_NODE
||
197 sa
->type
== UBIFS_XENT_NODE
);
198 ubifs_assert(c
, sb
->type
== UBIFS_DENT_NODE
||
199 sb
->type
== UBIFS_XENT_NODE
);
201 inuma
= key_inum(c
, &sa
->key
);
202 inumb
= key_inum(c
, &sb
->key
);
204 if (inuma
== inumb
) {
205 uint32_t hasha
= key_hash(c
, &sa
->key
);
206 uint32_t hashb
= key_hash(c
, &sb
->key
);
210 } else if (inuma
<= inumb
)
217 * sort_nodes - sort nodes for GC.
218 * @c: UBIFS file-system description object
219 * @sleb: describes nodes to sort and contains the result on exit
220 * @nondata: contains non-data nodes on exit
221 * @min: minimum node size is returned here
223 * This function sorts the list of inodes to garbage collect. First of all, it
224 * kills obsolete nodes and separates data and non-data nodes to the
225 * @sleb->nodes and @nondata lists correspondingly.
227 * Data nodes are then sorted in block number order - this is important for
228 * bulk-read; data nodes with lower inode number go before data nodes with
229 * higher inode number, and data nodes with lower block number go before data
230 * nodes with higher block number;
232 * Non-data nodes are sorted as follows.
233 * o First go inode nodes - they are sorted in descending length order.
234 * o Then go directory entry nodes - they are sorted in hash order, which
235 * should supposedly optimize 'readdir()'. Direntry nodes with lower parent
236 * inode number go before direntry nodes with higher parent inode number,
237 * and direntry nodes with lower name hash values go before direntry nodes
238 * with higher name hash values.
240 * This function returns zero in case of success and a negative error code in
243 static int sort_nodes(struct ubifs_info
*c
, struct ubifs_scan_leb
*sleb
,
244 struct list_head
*nondata
, int *min
)
247 struct ubifs_scan_node
*snod
, *tmp
;
251 /* Separate data nodes and non-data nodes */
252 list_for_each_entry_safe(snod
, tmp
, &sleb
->nodes
, list
) {
253 ubifs_assert(c
, snod
->type
== UBIFS_INO_NODE
||
254 snod
->type
== UBIFS_DATA_NODE
||
255 snod
->type
== UBIFS_DENT_NODE
||
256 snod
->type
== UBIFS_XENT_NODE
||
257 snod
->type
== UBIFS_TRUN_NODE
||
258 snod
->type
== UBIFS_AUTH_NODE
);
260 if (snod
->type
!= UBIFS_INO_NODE
&&
261 snod
->type
!= UBIFS_DATA_NODE
&&
262 snod
->type
!= UBIFS_DENT_NODE
&&
263 snod
->type
!= UBIFS_XENT_NODE
) {
264 /* Probably truncation node, zap it */
265 list_del(&snod
->list
);
270 ubifs_assert(c
, key_type(c
, &snod
->key
) == UBIFS_DATA_KEY
||
271 key_type(c
, &snod
->key
) == UBIFS_INO_KEY
||
272 key_type(c
, &snod
->key
) == UBIFS_DENT_KEY
||
273 key_type(c
, &snod
->key
) == UBIFS_XENT_KEY
);
275 err
= ubifs_tnc_has_node(c
, &snod
->key
, 0, sleb
->lnum
,
281 /* The node is obsolete, remove it from the list */
282 list_del(&snod
->list
);
287 if (snod
->len
< *min
)
290 if (key_type(c
, &snod
->key
) != UBIFS_DATA_KEY
)
291 list_move_tail(&snod
->list
, nondata
);
294 /* Sort data and non-data nodes */
295 list_sort(c
, &sleb
->nodes
, &data_nodes_cmp
);
296 list_sort(c
, nondata
, &nondata_nodes_cmp
);
298 err
= dbg_check_data_nodes_order(c
, &sleb
->nodes
);
301 err
= dbg_check_nondata_nodes_order(c
, nondata
);
308 * move_node - move a node.
309 * @c: UBIFS file-system description object
310 * @sleb: describes the LEB to move nodes from
311 * @snod: the mode to move
312 * @wbuf: write-buffer to move node to
314 * This function moves node @snod to @wbuf, changes TNC correspondingly, and
315 * destroys @snod. Returns zero in case of success and a negative error code in
318 static int move_node(struct ubifs_info
*c
, struct ubifs_scan_leb
*sleb
,
319 struct ubifs_scan_node
*snod
, struct ubifs_wbuf
*wbuf
)
321 int err
, new_lnum
= wbuf
->lnum
, new_offs
= wbuf
->offs
+ wbuf
->used
;
324 err
= ubifs_wbuf_write_nolock(wbuf
, snod
->node
, snod
->len
);
328 err
= ubifs_tnc_replace(c
, &snod
->key
, sleb
->lnum
,
329 snod
->offs
, new_lnum
, new_offs
,
331 list_del(&snod
->list
);
337 * move_nodes - move nodes.
338 * @c: UBIFS file-system description object
339 * @sleb: describes the LEB to move nodes from
341 * This function moves valid nodes from data LEB described by @sleb to the GC
342 * journal head. This function returns zero in case of success, %-EAGAIN if
343 * commit is required, and other negative error codes in case of other
346 static int move_nodes(struct ubifs_info
*c
, struct ubifs_scan_leb
*sleb
)
350 struct ubifs_wbuf
*wbuf
= &c
->jheads
[GCHD
].wbuf
;
352 if (wbuf
->lnum
== -1) {
354 * The GC journal head is not set, because it is the first GC
355 * invocation since mount.
357 err
= switch_gc_head(c
);
362 err
= sort_nodes(c
, sleb
, &nondata
, &min
);
366 /* Write nodes to their new location. Use the first-fit strategy */
368 int avail
, moved
= 0;
369 struct ubifs_scan_node
*snod
, *tmp
;
371 /* Move data nodes */
372 list_for_each_entry_safe(snod
, tmp
, &sleb
->nodes
, list
) {
373 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
-
374 ubifs_auth_node_sz(c
);
375 if (snod
->len
> avail
)
377 * Do not skip data nodes in order to optimize
382 err
= ubifs_shash_update(c
, c
->jheads
[GCHD
].log_hash
,
383 snod
->node
, snod
->len
);
387 err
= move_node(c
, sleb
, snod
, wbuf
);
393 /* Move non-data nodes */
394 list_for_each_entry_safe(snod
, tmp
, &nondata
, list
) {
395 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
-
396 ubifs_auth_node_sz(c
);
400 if (snod
->len
> avail
) {
402 * Keep going only if this is an inode with
403 * some data. Otherwise stop and switch the GC
404 * head. IOW, we assume that data-less inode
405 * nodes and direntry nodes are roughly of the
408 if (key_type(c
, &snod
->key
) == UBIFS_DENT_KEY
||
409 snod
->len
== UBIFS_INO_NODE_SZ
)
414 err
= ubifs_shash_update(c
, c
->jheads
[GCHD
].log_hash
,
415 snod
->node
, snod
->len
);
419 err
= move_node(c
, sleb
, snod
, wbuf
);
425 if (ubifs_authenticated(c
) && moved
) {
426 struct ubifs_auth_node
*auth
;
428 auth
= kmalloc(ubifs_auth_node_sz(c
), GFP_NOFS
);
434 err
= ubifs_prepare_auth_node(c
, auth
,
435 c
->jheads
[GCHD
].log_hash
);
441 err
= ubifs_wbuf_write_nolock(wbuf
, auth
,
442 ubifs_auth_node_sz(c
));
448 ubifs_add_dirt(c
, wbuf
->lnum
, ubifs_auth_node_sz(c
));
451 if (list_empty(&sleb
->nodes
) && list_empty(&nondata
))
455 * Waste the rest of the space in the LEB and switch to the
458 err
= switch_gc_head(c
);
466 list_splice_tail(&nondata
, &sleb
->nodes
);
471 * gc_sync_wbufs - sync write-buffers for GC.
472 * @c: UBIFS file-system description object
474 * We must guarantee that obsoleting nodes are on flash. Unfortunately they may
475 * be in a write-buffer instead. That is, a node could be written to a
476 * write-buffer, obsoleting another node in a LEB that is GC'd. If that LEB is
477 * erased before the write-buffer is sync'd and then there is an unclean
478 * unmount, then an existing node is lost. To avoid this, we sync all
481 * This function returns %0 on success or a negative error code on failure.
483 static int gc_sync_wbufs(struct ubifs_info
*c
)
487 for (i
= 0; i
< c
->jhead_cnt
; i
++) {
490 err
= ubifs_wbuf_sync(&c
->jheads
[i
].wbuf
);
498 * ubifs_garbage_collect_leb - garbage-collect a logical eraseblock.
499 * @c: UBIFS file-system description object
500 * @lp: describes the LEB to garbage collect
502 * This function garbage-collects an LEB and returns one of the @LEB_FREED,
503 * @LEB_RETAINED, etc positive codes in case of success, %-EAGAIN if commit is
504 * required, and other negative error codes in case of failures.
506 int ubifs_garbage_collect_leb(struct ubifs_info
*c
, struct ubifs_lprops
*lp
)
508 struct ubifs_scan_leb
*sleb
;
509 struct ubifs_scan_node
*snod
;
510 struct ubifs_wbuf
*wbuf
= &c
->jheads
[GCHD
].wbuf
;
511 int err
= 0, lnum
= lp
->lnum
;
513 ubifs_assert(c
, c
->gc_lnum
!= -1 || wbuf
->offs
+ wbuf
->used
== 0 ||
515 ubifs_assert(c
, c
->gc_lnum
!= lnum
);
516 ubifs_assert(c
, wbuf
->lnum
!= lnum
);
518 if (lp
->free
+ lp
->dirty
== c
->leb_size
) {
519 /* Special case - a free LEB */
520 dbg_gc("LEB %d is free, return it", lp
->lnum
);
521 ubifs_assert(c
, !(lp
->flags
& LPROPS_INDEX
));
523 if (lp
->free
!= c
->leb_size
) {
525 * Write buffers must be sync'd before unmapping
526 * freeable LEBs, because one of them may contain data
527 * which obsoletes something in 'lp->lnum'.
529 err
= gc_sync_wbufs(c
);
532 err
= ubifs_change_one_lp(c
, lp
->lnum
, c
->leb_size
,
537 err
= ubifs_leb_unmap(c
, lp
->lnum
);
541 if (c
->gc_lnum
== -1) {
550 * We scan the entire LEB even though we only really need to scan up to
551 * (c->leb_size - lp->free).
553 sleb
= ubifs_scan(c
, lnum
, 0, c
->sbuf
, 0);
555 return PTR_ERR(sleb
);
557 ubifs_assert(c
, !list_empty(&sleb
->nodes
));
558 snod
= list_entry(sleb
->nodes
.next
, struct ubifs_scan_node
, list
);
560 if (snod
->type
== UBIFS_IDX_NODE
) {
561 struct ubifs_gced_idx_leb
*idx_gc
;
563 dbg_gc("indexing LEB %d (free %d, dirty %d)",
564 lnum
, lp
->free
, lp
->dirty
);
565 list_for_each_entry(snod
, &sleb
->nodes
, list
) {
566 struct ubifs_idx_node
*idx
= snod
->node
;
567 int level
= le16_to_cpu(idx
->level
);
569 ubifs_assert(c
, snod
->type
== UBIFS_IDX_NODE
);
570 key_read(c
, ubifs_idx_key(c
, idx
), &snod
->key
);
571 err
= ubifs_dirty_idx_node(c
, &snod
->key
, level
, lnum
,
577 idx_gc
= kmalloc(sizeof(struct ubifs_gced_idx_leb
), GFP_NOFS
);
585 list_add(&idx_gc
->list
, &c
->idx_gc
);
588 * Don't release the LEB until after the next commit, because
589 * it may contain data which is needed for recovery. So
590 * although we freed this LEB, it will become usable only after
593 err
= ubifs_change_one_lp(c
, lnum
, c
->leb_size
, 0, 0,
599 dbg_gc("data LEB %d (free %d, dirty %d)",
600 lnum
, lp
->free
, lp
->dirty
);
602 err
= move_nodes(c
, sleb
);
606 err
= gc_sync_wbufs(c
);
610 err
= ubifs_change_one_lp(c
, lnum
, c
->leb_size
, 0, 0, 0, 0);
614 /* Allow for races with TNC */
620 if (c
->gc_lnum
== -1) {
624 err
= ubifs_wbuf_sync_nolock(wbuf
);
628 err
= ubifs_leb_unmap(c
, lnum
);
637 ubifs_scan_destroy(sleb
);
641 /* We may have moved at least some nodes so allow for races with TNC */
650 * ubifs_garbage_collect - UBIFS garbage collector.
651 * @c: UBIFS file-system description object
652 * @anyway: do GC even if there are free LEBs
654 * This function does out-of-place garbage collection. The return codes are:
655 * o positive LEB number if the LEB has been freed and may be used;
656 * o %-EAGAIN if the caller has to run commit;
657 * o %-ENOSPC if GC failed to make any progress;
658 * o other negative error codes in case of other errors.
660 * Garbage collector writes data to the journal when GC'ing data LEBs, and just
661 * marking indexing nodes dirty when GC'ing indexing LEBs. Thus, at some point
662 * commit may be required. But commit cannot be run from inside GC, because the
663 * caller might be holding the commit lock, so %-EAGAIN is returned instead;
664 * And this error code means that the caller has to run commit, and re-run GC
665 * if there is still no free space.
667 * There are many reasons why this function may return %-EAGAIN:
668 * o the log is full and there is no space to write an LEB reference for
670 * o the journal is too large and exceeds size limitations;
671 * o GC moved indexing LEBs, but they can be used only after the commit;
672 * o the shrinker fails to find clean znodes to free and requests the commit;
675 * Note, if the file-system is close to be full, this function may return
676 * %-EAGAIN infinitely, so the caller has to limit amount of re-invocations of
677 * the function. E.g., this happens if the limits on the journal size are too
678 * tough and GC writes too much to the journal before an LEB is freed. This
679 * might also mean that the journal is too large, and the TNC becomes to big,
680 * so that the shrinker is constantly called, finds not clean znodes to free,
681 * and requests commit. Well, this may also happen if the journal is all right,
682 * but another kernel process consumes too much memory. Anyway, infinite
683 * %-EAGAIN may happen, but in some extreme/misconfiguration cases.
685 int ubifs_garbage_collect(struct ubifs_info
*c
, int anyway
)
687 int i
, err
, ret
, min_space
= c
->dead_wm
;
688 struct ubifs_lprops lp
;
689 struct ubifs_wbuf
*wbuf
= &c
->jheads
[GCHD
].wbuf
;
691 ubifs_assert_cmt_locked(c
);
692 ubifs_assert(c
, !c
->ro_media
&& !c
->ro_mount
);
694 if (ubifs_gc_should_commit(c
))
697 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
704 /* We expect the write-buffer to be empty on entry */
705 ubifs_assert(c
, !wbuf
->used
);
708 int space_before
, space_after
;
712 /* Give the commit an opportunity to run */
713 if (ubifs_gc_should_commit(c
)) {
718 if (i
> SOFT_LEBS_LIMIT
&& !list_empty(&c
->idx_gc
)) {
720 * We've done enough iterations. Indexing LEBs were
721 * moved and will be available after the commit.
723 dbg_gc("soft limit, some index LEBs GC'ed, -EAGAIN");
724 ubifs_commit_required(c
);
729 if (i
> HARD_LEBS_LIMIT
) {
731 * We've moved too many LEBs and have not made
734 dbg_gc("hard limit, -ENOSPC");
740 * Empty and freeable LEBs can turn up while we waited for
741 * the wbuf lock, or while we have been running GC. In that
742 * case, we should just return one of those instead of
743 * continuing to GC dirty LEBs. Hence we request
744 * 'ubifs_find_dirty_leb()' to return an empty LEB if it can.
746 ret
= ubifs_find_dirty_leb(c
, &lp
, min_space
, anyway
? 0 : 1);
749 dbg_gc("no more dirty LEBs");
753 dbg_gc("found LEB %d: free %d, dirty %d, sum %d (min. space %d)",
754 lp
.lnum
, lp
.free
, lp
.dirty
, lp
.free
+ lp
.dirty
,
757 space_before
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
758 if (wbuf
->lnum
== -1)
761 ret
= ubifs_garbage_collect_leb(c
, &lp
);
763 if (ret
== -EAGAIN
) {
765 * This is not error, so we have to return the
766 * LEB to lprops. But if 'ubifs_return_leb()'
767 * fails, its failure code is propagated to the
768 * caller instead of the original '-EAGAIN'.
770 err
= ubifs_return_leb(c
, lp
.lnum
);
778 if (ret
== LEB_FREED
) {
779 /* An LEB has been freed and is ready for use */
780 dbg_gc("LEB %d freed, return", lp
.lnum
);
785 if (ret
== LEB_FREED_IDX
) {
787 * This was an indexing LEB and it cannot be
788 * immediately used. And instead of requesting the
789 * commit straight away, we try to garbage collect some
792 dbg_gc("indexing LEB %d freed, continue", lp
.lnum
);
796 ubifs_assert(c
, ret
== LEB_RETAINED
);
797 space_after
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
798 dbg_gc("LEB %d retained, freed %d bytes", lp
.lnum
,
799 space_after
- space_before
);
801 if (space_after
> space_before
) {
802 /* GC makes progress, keep working */
804 if (min_space
< c
->dead_wm
)
805 min_space
= c
->dead_wm
;
809 dbg_gc("did not make progress");
812 * GC moved an LEB bud have not done any progress. This means
813 * that the previous GC head LEB contained too few free space
814 * and the LEB which was GC'ed contained only large nodes which
815 * did not fit that space.
817 * We can do 2 things:
818 * 1. pick another LEB in a hope it'll contain a small node
819 * which will fit the space we have at the end of current GC
820 * head LEB, but there is no guarantee, so we try this out
821 * unless we have already been working for too long;
822 * 2. request an LEB with more dirty space, which will force
823 * 'ubifs_find_dirty_leb()' to start scanning the lprops
824 * table, instead of just picking one from the heap
825 * (previously it already picked the dirtiest LEB).
827 if (i
< SOFT_LEBS_LIMIT
) {
833 if (min_space
> c
->dark_wm
)
834 min_space
= c
->dark_wm
;
835 dbg_gc("set min. space to %d", min_space
);
838 if (ret
== -ENOSPC
&& !list_empty(&c
->idx_gc
)) {
839 dbg_gc("no space, some index LEBs GC'ed, -EAGAIN");
840 ubifs_commit_required(c
);
844 err
= ubifs_wbuf_sync_nolock(wbuf
);
846 err
= ubifs_leb_unmap(c
, c
->gc_lnum
);
852 mutex_unlock(&wbuf
->io_mutex
);
856 ubifs_assert(c
, ret
< 0);
857 ubifs_assert(c
, ret
!= -ENOSPC
&& ret
!= -EAGAIN
);
858 ubifs_wbuf_sync_nolock(wbuf
);
859 ubifs_ro_mode(c
, ret
);
860 mutex_unlock(&wbuf
->io_mutex
);
861 ubifs_return_leb(c
, lp
.lnum
);
866 * ubifs_gc_start_commit - garbage collection at start of commit.
867 * @c: UBIFS file-system description object
869 * If a LEB has only dirty and free space, then we may safely unmap it and make
870 * it free. Note, we cannot do this with indexing LEBs because dirty space may
871 * correspond index nodes that are required for recovery. In that case, the
872 * LEB cannot be unmapped until after the next commit.
874 * This function returns %0 upon success and a negative error code upon failure.
876 int ubifs_gc_start_commit(struct ubifs_info
*c
)
878 struct ubifs_gced_idx_leb
*idx_gc
;
879 const struct ubifs_lprops
*lp
;
885 * Unmap (non-index) freeable LEBs. Note that recovery requires that all
886 * wbufs are sync'd before this, which is done in 'do_commit()'.
889 lp
= ubifs_fast_find_freeable(c
);
892 ubifs_assert(c
, !(lp
->flags
& LPROPS_TAKEN
));
893 ubifs_assert(c
, !(lp
->flags
& LPROPS_INDEX
));
894 err
= ubifs_leb_unmap(c
, lp
->lnum
);
897 lp
= ubifs_change_lp(c
, lp
, c
->leb_size
, 0, lp
->flags
, 0);
902 ubifs_assert(c
, !(lp
->flags
& LPROPS_TAKEN
));
903 ubifs_assert(c
, !(lp
->flags
& LPROPS_INDEX
));
906 /* Mark GC'd index LEBs OK to unmap after this commit finishes */
907 list_for_each_entry(idx_gc
, &c
->idx_gc
, list
)
910 /* Record index freeable LEBs for unmapping after commit */
912 lp
= ubifs_fast_find_frdi_idx(c
);
919 idx_gc
= kmalloc(sizeof(struct ubifs_gced_idx_leb
), GFP_NOFS
);
924 ubifs_assert(c
, !(lp
->flags
& LPROPS_TAKEN
));
925 ubifs_assert(c
, lp
->flags
& LPROPS_INDEX
);
926 /* Don't release the LEB until after the next commit */
927 flags
= (lp
->flags
| LPROPS_TAKEN
) ^ LPROPS_INDEX
;
928 lp
= ubifs_change_lp(c
, lp
, c
->leb_size
, 0, flags
, 1);
934 ubifs_assert(c
, lp
->flags
& LPROPS_TAKEN
);
935 ubifs_assert(c
, !(lp
->flags
& LPROPS_INDEX
));
936 idx_gc
->lnum
= lp
->lnum
;
938 list_add(&idx_gc
->list
, &c
->idx_gc
);
941 ubifs_release_lprops(c
);
946 * ubifs_gc_end_commit - garbage collection at end of commit.
947 * @c: UBIFS file-system description object
949 * This function completes out-of-place garbage collection of index LEBs.
951 int ubifs_gc_end_commit(struct ubifs_info
*c
)
953 struct ubifs_gced_idx_leb
*idx_gc
, *tmp
;
954 struct ubifs_wbuf
*wbuf
;
957 wbuf
= &c
->jheads
[GCHD
].wbuf
;
958 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
959 list_for_each_entry_safe(idx_gc
, tmp
, &c
->idx_gc
, list
)
961 dbg_gc("LEB %d", idx_gc
->lnum
);
962 err
= ubifs_leb_unmap(c
, idx_gc
->lnum
);
965 err
= ubifs_change_one_lp(c
, idx_gc
->lnum
, LPROPS_NC
,
966 LPROPS_NC
, 0, LPROPS_TAKEN
, -1);
969 list_del(&idx_gc
->list
);
973 mutex_unlock(&wbuf
->io_mutex
);
978 * ubifs_destroy_idx_gc - destroy idx_gc list.
979 * @c: UBIFS file-system description object
981 * This function destroys the @c->idx_gc list. It is called when unmounting
982 * so locks are not needed. Returns zero in case of success and a negative
983 * error code in case of failure.
985 void ubifs_destroy_idx_gc(struct ubifs_info
*c
)
987 while (!list_empty(&c
->idx_gc
)) {
988 struct ubifs_gced_idx_leb
*idx_gc
;
990 idx_gc
= list_entry(c
->idx_gc
.next
, struct ubifs_gced_idx_leb
,
993 list_del(&idx_gc
->list
);
999 * ubifs_get_idx_gc_leb - get a LEB from GC'd index LEB list.
1000 * @c: UBIFS file-system description object
1002 * Called during start commit so locks are not needed.
1004 int ubifs_get_idx_gc_leb(struct ubifs_info
*c
)
1006 struct ubifs_gced_idx_leb
*idx_gc
;
1009 if (list_empty(&c
->idx_gc
))
1011 idx_gc
= list_entry(c
->idx_gc
.next
, struct ubifs_gced_idx_leb
, list
);
1012 lnum
= idx_gc
->lnum
;
1013 /* c->idx_gc_cnt is updated by the caller when lprops are updated */
1014 list_del(&idx_gc
->list
);