Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation. | |
5 | * | |
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. | |
9 | * | |
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 | |
13 | * more details. | |
14 | * | |
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 | |
18 | * | |
19 | * Authors: Adrian Hunter | |
20 | * Artem Bityutskiy (Битюцкий Артём) | |
21 | */ | |
22 | ||
23 | /* | |
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. | |
7078202e AB |
34 | * |
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. | |
43 | * | |
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 migh | |
51 | * be unable to reclaim it. So, LEBs with free + dirty greater than dark | |
52 | * watermark are "good" LEBs from GC's point of few. The other LEBs are not so | |
53 | * good, and GC takes extra care when moving them. | |
1e51764a AB |
54 | */ |
55 | ||
56 | #include <linux/pagemap.h> | |
57 | #include "ubifs.h" | |
58 | ||
59 | /* | |
60 | * GC tries to optimize the way it fit nodes to available space, and it sorts | |
61 | * nodes a little. The below constants are watermarks which define "large", | |
62 | * "medium", and "small" nodes. | |
63 | */ | |
64 | #define MEDIUM_NODE_WM (UBIFS_BLOCK_SIZE / 4) | |
65 | #define SMALL_NODE_WM UBIFS_MAX_DENT_NODE_SZ | |
66 | ||
67 | /* | |
025dfdaf | 68 | * GC may need to move more than one LEB to make progress. The below constants |
1e51764a AB |
69 | * define "soft" and "hard" limits on the number of LEBs the garbage collector |
70 | * may move. | |
71 | */ | |
72 | #define SOFT_LEBS_LIMIT 4 | |
73 | #define HARD_LEBS_LIMIT 32 | |
74 | ||
75 | /** | |
76 | * switch_gc_head - switch the garbage collection journal head. | |
77 | * @c: UBIFS file-system description object | |
78 | * @buf: buffer to write | |
79 | * @len: length of the buffer to write | |
80 | * @lnum: LEB number written is returned here | |
81 | * @offs: offset written is returned here | |
82 | * | |
83 | * This function switch the GC head to the next LEB which is reserved in | |
84 | * @c->gc_lnum. Returns %0 in case of success, %-EAGAIN if commit is required, | |
85 | * and other negative error code in case of failures. | |
86 | */ | |
87 | static int switch_gc_head(struct ubifs_info *c) | |
88 | { | |
89 | int err, gc_lnum = c->gc_lnum; | |
90 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
91 | ||
92 | ubifs_assert(gc_lnum != -1); | |
93 | dbg_gc("switch GC head from LEB %d:%d to LEB %d (waste %d bytes)", | |
94 | wbuf->lnum, wbuf->offs + wbuf->used, gc_lnum, | |
95 | c->leb_size - wbuf->offs - wbuf->used); | |
96 | ||
97 | err = ubifs_wbuf_sync_nolock(wbuf); | |
98 | if (err) | |
99 | return err; | |
100 | ||
101 | /* | |
102 | * The GC write-buffer was synchronized, we may safely unmap | |
103 | * 'c->gc_lnum'. | |
104 | */ | |
105 | err = ubifs_leb_unmap(c, gc_lnum); | |
106 | if (err) | |
107 | return err; | |
108 | ||
109 | err = ubifs_add_bud_to_log(c, GCHD, gc_lnum, 0); | |
110 | if (err) | |
111 | return err; | |
112 | ||
113 | c->gc_lnum = -1; | |
114 | err = ubifs_wbuf_seek_nolock(wbuf, gc_lnum, 0, UBI_LONGTERM); | |
115 | return err; | |
116 | } | |
117 | ||
46773be4 AH |
118 | /** |
119 | * joinup - bring data nodes for an inode together. | |
120 | * @c: UBIFS file-system description object | |
121 | * @sleb: describes scanned LEB | |
122 | * @inum: inode number | |
123 | * @blk: block number | |
124 | * @data: list to which to add data nodes | |
125 | * | |
126 | * This function looks at the first few nodes in the scanned LEB @sleb and adds | |
127 | * them to @data if they are data nodes from @inum and have a larger block | |
128 | * number than @blk. This function returns %0 on success and a negative error | |
129 | * code on failure. | |
130 | */ | |
131 | static int joinup(struct ubifs_info *c, struct ubifs_scan_leb *sleb, ino_t inum, | |
132 | unsigned int blk, struct list_head *data) | |
133 | { | |
134 | int err, cnt = 6, lnum = sleb->lnum, offs; | |
135 | struct ubifs_scan_node *snod, *tmp; | |
136 | union ubifs_key *key; | |
137 | ||
138 | list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) { | |
139 | key = &snod->key; | |
140 | if (key_inum(c, key) == inum && | |
141 | key_type(c, key) == UBIFS_DATA_KEY && | |
142 | key_block(c, key) > blk) { | |
143 | offs = snod->offs; | |
144 | err = ubifs_tnc_has_node(c, key, 0, lnum, offs, 0); | |
145 | if (err < 0) | |
146 | return err; | |
147 | list_del(&snod->list); | |
148 | if (err) { | |
149 | list_add_tail(&snod->list, data); | |
150 | blk = key_block(c, key); | |
151 | } else | |
152 | kfree(snod); | |
153 | cnt = 6; | |
154 | } else if (--cnt == 0) | |
155 | break; | |
156 | } | |
157 | return 0; | |
158 | } | |
159 | ||
1e51764a AB |
160 | /** |
161 | * move_nodes - move nodes. | |
162 | * @c: UBIFS file-system description object | |
163 | * @sleb: describes nodes to move | |
164 | * | |
165 | * This function moves valid nodes from data LEB described by @sleb to the GC | |
166 | * journal head. The obsolete nodes are dropped. | |
167 | * | |
168 | * When moving nodes we have to deal with classical bin-packing problem: the | |
169 | * space in the current GC journal head LEB and in @c->gc_lnum are the "bins", | |
170 | * where the nodes in the @sleb->nodes list are the elements which should be | |
171 | * fit optimally to the bins. This function uses the "first fit decreasing" | |
172 | * strategy, although it does not really sort the nodes but just split them on | |
173 | * 3 classes - large, medium, and small, so they are roughly sorted. | |
174 | * | |
175 | * This function returns zero in case of success, %-EAGAIN if commit is | |
176 | * required, and other negative error codes in case of other failures. | |
177 | */ | |
178 | static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb) | |
179 | { | |
180 | struct ubifs_scan_node *snod, *tmp; | |
46773be4 | 181 | struct list_head data, large, medium, small; |
1e51764a AB |
182 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; |
183 | int avail, err, min = INT_MAX; | |
46773be4 AH |
184 | unsigned int blk = 0; |
185 | ino_t inum = 0; | |
1e51764a | 186 | |
46773be4 | 187 | INIT_LIST_HEAD(&data); |
1e51764a AB |
188 | INIT_LIST_HEAD(&large); |
189 | INIT_LIST_HEAD(&medium); | |
190 | INIT_LIST_HEAD(&small); | |
191 | ||
46773be4 AH |
192 | while (!list_empty(&sleb->nodes)) { |
193 | struct list_head *lst = sleb->nodes.next; | |
194 | ||
195 | snod = list_entry(lst, struct ubifs_scan_node, list); | |
1e51764a AB |
196 | |
197 | ubifs_assert(snod->type != UBIFS_IDX_NODE); | |
198 | ubifs_assert(snod->type != UBIFS_REF_NODE); | |
199 | ubifs_assert(snod->type != UBIFS_CS_NODE); | |
200 | ||
201 | err = ubifs_tnc_has_node(c, &snod->key, 0, sleb->lnum, | |
202 | snod->offs, 0); | |
203 | if (err < 0) | |
204 | goto out; | |
205 | ||
1e51764a AB |
206 | list_del(lst); |
207 | if (!err) { | |
208 | /* The node is obsolete, remove it from the list */ | |
209 | kfree(snod); | |
210 | continue; | |
211 | } | |
212 | ||
213 | /* | |
46773be4 AH |
214 | * Sort the list of nodes so that data nodes go first, large |
215 | * nodes go second, and small nodes go last. | |
1e51764a | 216 | */ |
46773be4 AH |
217 | if (key_type(c, &snod->key) == UBIFS_DATA_KEY) { |
218 | if (inum != key_inum(c, &snod->key)) { | |
219 | if (inum) { | |
220 | /* | |
221 | * Try to move data nodes from the same | |
222 | * inode together. | |
223 | */ | |
224 | err = joinup(c, sleb, inum, blk, &data); | |
225 | if (err) | |
226 | goto out; | |
227 | } | |
228 | inum = key_inum(c, &snod->key); | |
229 | blk = key_block(c, &snod->key); | |
230 | } | |
231 | list_add_tail(lst, &data); | |
232 | } else if (snod->len > MEDIUM_NODE_WM) | |
233 | list_add_tail(lst, &large); | |
1e51764a | 234 | else if (snod->len > SMALL_NODE_WM) |
46773be4 | 235 | list_add_tail(lst, &medium); |
1e51764a | 236 | else |
46773be4 | 237 | list_add_tail(lst, &small); |
1e51764a AB |
238 | |
239 | /* And find the smallest node */ | |
240 | if (snod->len < min) | |
241 | min = snod->len; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Join the tree lists so that we'd have one roughly sorted list | |
246 | * ('large' will be the head of the joined list). | |
247 | */ | |
46773be4 | 248 | list_splice(&data, &large); |
1e51764a AB |
249 | list_splice(&medium, large.prev); |
250 | list_splice(&small, large.prev); | |
251 | ||
252 | if (wbuf->lnum == -1) { | |
253 | /* | |
254 | * The GC journal head is not set, because it is the first GC | |
255 | * invocation since mount. | |
256 | */ | |
257 | err = switch_gc_head(c); | |
258 | if (err) | |
259 | goto out; | |
260 | } | |
261 | ||
262 | /* Write nodes to their new location. Use the first-fit strategy */ | |
263 | while (1) { | |
264 | avail = c->leb_size - wbuf->offs - wbuf->used; | |
265 | list_for_each_entry_safe(snod, tmp, &large, list) { | |
266 | int new_lnum, new_offs; | |
267 | ||
268 | if (avail < min) | |
269 | break; | |
270 | ||
271 | if (snod->len > avail) | |
272 | /* This node does not fit */ | |
273 | continue; | |
274 | ||
275 | cond_resched(); | |
276 | ||
277 | new_lnum = wbuf->lnum; | |
278 | new_offs = wbuf->offs + wbuf->used; | |
279 | err = ubifs_wbuf_write_nolock(wbuf, snod->node, | |
280 | snod->len); | |
281 | if (err) | |
282 | goto out; | |
283 | err = ubifs_tnc_replace(c, &snod->key, sleb->lnum, | |
284 | snod->offs, new_lnum, new_offs, | |
285 | snod->len); | |
286 | if (err) | |
287 | goto out; | |
288 | ||
289 | avail = c->leb_size - wbuf->offs - wbuf->used; | |
290 | list_del(&snod->list); | |
291 | kfree(snod); | |
292 | } | |
293 | ||
294 | if (list_empty(&large)) | |
295 | break; | |
296 | ||
297 | /* | |
298 | * Waste the rest of the space in the LEB and switch to the | |
299 | * next LEB. | |
300 | */ | |
301 | err = switch_gc_head(c); | |
302 | if (err) | |
303 | goto out; | |
304 | } | |
305 | ||
306 | return 0; | |
307 | ||
308 | out: | |
309 | list_for_each_entry_safe(snod, tmp, &large, list) { | |
310 | list_del(&snod->list); | |
311 | kfree(snod); | |
312 | } | |
313 | return err; | |
314 | } | |
315 | ||
316 | /** | |
317 | * gc_sync_wbufs - sync write-buffers for GC. | |
318 | * @c: UBIFS file-system description object | |
319 | * | |
320 | * We must guarantee that obsoleting nodes are on flash. Unfortunately they may | |
321 | * be in a write-buffer instead. That is, a node could be written to a | |
322 | * write-buffer, obsoleting another node in a LEB that is GC'd. If that LEB is | |
323 | * erased before the write-buffer is sync'd and then there is an unclean | |
324 | * unmount, then an existing node is lost. To avoid this, we sync all | |
325 | * write-buffers. | |
326 | * | |
327 | * This function returns %0 on success or a negative error code on failure. | |
328 | */ | |
329 | static int gc_sync_wbufs(struct ubifs_info *c) | |
330 | { | |
331 | int err, i; | |
332 | ||
333 | for (i = 0; i < c->jhead_cnt; i++) { | |
334 | if (i == GCHD) | |
335 | continue; | |
336 | err = ubifs_wbuf_sync(&c->jheads[i].wbuf); | |
337 | if (err) | |
338 | return err; | |
339 | } | |
340 | return 0; | |
341 | } | |
342 | ||
343 | /** | |
344 | * ubifs_garbage_collect_leb - garbage-collect a logical eraseblock. | |
345 | * @c: UBIFS file-system description object | |
346 | * @lp: describes the LEB to garbage collect | |
347 | * | |
348 | * This function garbage-collects an LEB and returns one of the @LEB_FREED, | |
349 | * @LEB_RETAINED, etc positive codes in case of success, %-EAGAIN if commit is | |
350 | * required, and other negative error codes in case of failures. | |
351 | */ | |
352 | int ubifs_garbage_collect_leb(struct ubifs_info *c, struct ubifs_lprops *lp) | |
353 | { | |
354 | struct ubifs_scan_leb *sleb; | |
355 | struct ubifs_scan_node *snod; | |
356 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
357 | int err = 0, lnum = lp->lnum; | |
358 | ||
359 | ubifs_assert(c->gc_lnum != -1 || wbuf->offs + wbuf->used == 0 || | |
360 | c->need_recovery); | |
361 | ubifs_assert(c->gc_lnum != lnum); | |
362 | ubifs_assert(wbuf->lnum != lnum); | |
363 | ||
364 | /* | |
365 | * We scan the entire LEB even though we only really need to scan up to | |
366 | * (c->leb_size - lp->free). | |
367 | */ | |
368 | sleb = ubifs_scan(c, lnum, 0, c->sbuf); | |
369 | if (IS_ERR(sleb)) | |
370 | return PTR_ERR(sleb); | |
371 | ||
372 | ubifs_assert(!list_empty(&sleb->nodes)); | |
373 | snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list); | |
374 | ||
375 | if (snod->type == UBIFS_IDX_NODE) { | |
376 | struct ubifs_gced_idx_leb *idx_gc; | |
377 | ||
378 | dbg_gc("indexing LEB %d (free %d, dirty %d)", | |
379 | lnum, lp->free, lp->dirty); | |
380 | list_for_each_entry(snod, &sleb->nodes, list) { | |
381 | struct ubifs_idx_node *idx = snod->node; | |
382 | int level = le16_to_cpu(idx->level); | |
383 | ||
384 | ubifs_assert(snod->type == UBIFS_IDX_NODE); | |
385 | key_read(c, ubifs_idx_key(c, idx), &snod->key); | |
386 | err = ubifs_dirty_idx_node(c, &snod->key, level, lnum, | |
387 | snod->offs); | |
388 | if (err) | |
389 | goto out; | |
390 | } | |
391 | ||
392 | idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS); | |
393 | if (!idx_gc) { | |
394 | err = -ENOMEM; | |
395 | goto out; | |
396 | } | |
397 | ||
398 | idx_gc->lnum = lnum; | |
399 | idx_gc->unmap = 0; | |
400 | list_add(&idx_gc->list, &c->idx_gc); | |
401 | ||
402 | /* | |
403 | * Don't release the LEB until after the next commit, because | |
227c75c9 | 404 | * it may contain data which is needed for recovery. So |
1e51764a AB |
405 | * although we freed this LEB, it will become usable only after |
406 | * the commit. | |
407 | */ | |
408 | err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0, | |
409 | LPROPS_INDEX, 1); | |
410 | if (err) | |
411 | goto out; | |
412 | err = LEB_FREED_IDX; | |
413 | } else { | |
414 | dbg_gc("data LEB %d (free %d, dirty %d)", | |
415 | lnum, lp->free, lp->dirty); | |
416 | ||
417 | err = move_nodes(c, sleb); | |
418 | if (err) | |
6dcfac4f | 419 | goto out_inc_seq; |
1e51764a AB |
420 | |
421 | err = gc_sync_wbufs(c); | |
422 | if (err) | |
6dcfac4f | 423 | goto out_inc_seq; |
1e51764a AB |
424 | |
425 | err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0, 0, 0); | |
426 | if (err) | |
6dcfac4f | 427 | goto out_inc_seq; |
1e51764a | 428 | |
601c0bc4 AH |
429 | /* Allow for races with TNC */ |
430 | c->gced_lnum = lnum; | |
431 | smp_wmb(); | |
432 | c->gc_seq += 1; | |
433 | smp_wmb(); | |
434 | ||
1e51764a AB |
435 | if (c->gc_lnum == -1) { |
436 | c->gc_lnum = lnum; | |
437 | err = LEB_RETAINED; | |
438 | } else { | |
439 | err = ubifs_wbuf_sync_nolock(wbuf); | |
440 | if (err) | |
441 | goto out; | |
442 | ||
443 | err = ubifs_leb_unmap(c, lnum); | |
444 | if (err) | |
445 | goto out; | |
446 | ||
447 | err = LEB_FREED; | |
448 | } | |
449 | } | |
450 | ||
451 | out: | |
452 | ubifs_scan_destroy(sleb); | |
453 | return err; | |
6dcfac4f AH |
454 | |
455 | out_inc_seq: | |
456 | /* We may have moved at least some nodes so allow for races with TNC */ | |
457 | c->gced_lnum = lnum; | |
458 | smp_wmb(); | |
459 | c->gc_seq += 1; | |
460 | smp_wmb(); | |
461 | goto out; | |
1e51764a AB |
462 | } |
463 | ||
464 | /** | |
465 | * ubifs_garbage_collect - UBIFS garbage collector. | |
466 | * @c: UBIFS file-system description object | |
467 | * @anyway: do GC even if there are free LEBs | |
468 | * | |
469 | * This function does out-of-place garbage collection. The return codes are: | |
470 | * o positive LEB number if the LEB has been freed and may be used; | |
471 | * o %-EAGAIN if the caller has to run commit; | |
472 | * o %-ENOSPC if GC failed to make any progress; | |
473 | * o other negative error codes in case of other errors. | |
474 | * | |
475 | * Garbage collector writes data to the journal when GC'ing data LEBs, and just | |
476 | * marking indexing nodes dirty when GC'ing indexing LEBs. Thus, at some point | |
477 | * commit may be required. But commit cannot be run from inside GC, because the | |
478 | * caller might be holding the commit lock, so %-EAGAIN is returned instead; | |
479 | * And this error code means that the caller has to run commit, and re-run GC | |
480 | * if there is still no free space. | |
481 | * | |
482 | * There are many reasons why this function may return %-EAGAIN: | |
483 | * o the log is full and there is no space to write an LEB reference for | |
484 | * @c->gc_lnum; | |
485 | * o the journal is too large and exceeds size limitations; | |
486 | * o GC moved indexing LEBs, but they can be used only after the commit; | |
487 | * o the shrinker fails to find clean znodes to free and requests the commit; | |
488 | * o etc. | |
489 | * | |
490 | * Note, if the file-system is close to be full, this function may return | |
491 | * %-EAGAIN infinitely, so the caller has to limit amount of re-invocations of | |
492 | * the function. E.g., this happens if the limits on the journal size are too | |
493 | * tough and GC writes too much to the journal before an LEB is freed. This | |
494 | * might also mean that the journal is too large, and the TNC becomes to big, | |
495 | * so that the shrinker is constantly called, finds not clean znodes to free, | |
496 | * and requests commit. Well, this may also happen if the journal is all right, | |
497 | * but another kernel process consumes too much memory. Anyway, infinite | |
498 | * %-EAGAIN may happen, but in some extreme/misconfiguration cases. | |
499 | */ | |
500 | int ubifs_garbage_collect(struct ubifs_info *c, int anyway) | |
501 | { | |
502 | int i, err, ret, min_space = c->dead_wm; | |
503 | struct ubifs_lprops lp; | |
504 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
505 | ||
506 | ubifs_assert_cmt_locked(c); | |
507 | ||
508 | if (ubifs_gc_should_commit(c)) | |
509 | return -EAGAIN; | |
510 | ||
511 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
512 | ||
513 | if (c->ro_media) { | |
514 | ret = -EROFS; | |
515 | goto out_unlock; | |
516 | } | |
517 | ||
518 | /* We expect the write-buffer to be empty on entry */ | |
519 | ubifs_assert(!wbuf->used); | |
520 | ||
521 | for (i = 0; ; i++) { | |
522 | int space_before = c->leb_size - wbuf->offs - wbuf->used; | |
523 | int space_after; | |
524 | ||
525 | cond_resched(); | |
526 | ||
527 | /* Give the commit an opportunity to run */ | |
528 | if (ubifs_gc_should_commit(c)) { | |
529 | ret = -EAGAIN; | |
530 | break; | |
531 | } | |
532 | ||
533 | if (i > SOFT_LEBS_LIMIT && !list_empty(&c->idx_gc)) { | |
534 | /* | |
535 | * We've done enough iterations. Indexing LEBs were | |
536 | * moved and will be available after the commit. | |
537 | */ | |
538 | dbg_gc("soft limit, some index LEBs GC'ed, -EAGAIN"); | |
539 | ubifs_commit_required(c); | |
540 | ret = -EAGAIN; | |
541 | break; | |
542 | } | |
543 | ||
544 | if (i > HARD_LEBS_LIMIT) { | |
545 | /* | |
546 | * We've moved too many LEBs and have not made | |
547 | * progress, give up. | |
548 | */ | |
549 | dbg_gc("hard limit, -ENOSPC"); | |
550 | ret = -ENOSPC; | |
551 | break; | |
552 | } | |
553 | ||
554 | /* | |
555 | * Empty and freeable LEBs can turn up while we waited for | |
556 | * the wbuf lock, or while we have been running GC. In that | |
557 | * case, we should just return one of those instead of | |
558 | * continuing to GC dirty LEBs. Hence we request | |
559 | * 'ubifs_find_dirty_leb()' to return an empty LEB if it can. | |
560 | */ | |
561 | ret = ubifs_find_dirty_leb(c, &lp, min_space, anyway ? 0 : 1); | |
562 | if (ret) { | |
563 | if (ret == -ENOSPC) | |
564 | dbg_gc("no more dirty LEBs"); | |
565 | break; | |
566 | } | |
567 | ||
568 | dbg_gc("found LEB %d: free %d, dirty %d, sum %d " | |
569 | "(min. space %d)", lp.lnum, lp.free, lp.dirty, | |
570 | lp.free + lp.dirty, min_space); | |
571 | ||
572 | if (lp.free + lp.dirty == c->leb_size) { | |
573 | /* An empty LEB was returned */ | |
574 | dbg_gc("LEB %d is free, return it", lp.lnum); | |
575 | /* | |
576 | * ubifs_find_dirty_leb() doesn't return freeable index | |
577 | * LEBs. | |
578 | */ | |
579 | ubifs_assert(!(lp.flags & LPROPS_INDEX)); | |
580 | if (lp.free != c->leb_size) { | |
581 | /* | |
582 | * Write buffers must be sync'd before | |
583 | * unmapping freeable LEBs, because one of them | |
584 | * may contain data which obsoletes something | |
585 | * in 'lp.pnum'. | |
586 | */ | |
587 | ret = gc_sync_wbufs(c); | |
588 | if (ret) | |
589 | goto out; | |
590 | ret = ubifs_change_one_lp(c, lp.lnum, | |
591 | c->leb_size, 0, 0, 0, | |
592 | 0); | |
593 | if (ret) | |
594 | goto out; | |
595 | } | |
596 | ret = ubifs_leb_unmap(c, lp.lnum); | |
597 | if (ret) | |
598 | goto out; | |
599 | ret = lp.lnum; | |
600 | break; | |
601 | } | |
602 | ||
603 | space_before = c->leb_size - wbuf->offs - wbuf->used; | |
604 | if (wbuf->lnum == -1) | |
605 | space_before = 0; | |
606 | ||
607 | ret = ubifs_garbage_collect_leb(c, &lp); | |
608 | if (ret < 0) { | |
609 | if (ret == -EAGAIN || ret == -ENOSPC) { | |
610 | /* | |
611 | * These codes are not errors, so we have to | |
612 | * return the LEB to lprops. But if the | |
613 | * 'ubifs_return_leb()' function fails, its | |
614 | * failure code is propagated to the caller | |
615 | * instead of the original '-EAGAIN' or | |
616 | * '-ENOSPC'. | |
617 | */ | |
618 | err = ubifs_return_leb(c, lp.lnum); | |
619 | if (err) | |
620 | ret = err; | |
621 | break; | |
622 | } | |
623 | goto out; | |
624 | } | |
625 | ||
626 | if (ret == LEB_FREED) { | |
627 | /* An LEB has been freed and is ready for use */ | |
628 | dbg_gc("LEB %d freed, return", lp.lnum); | |
629 | ret = lp.lnum; | |
630 | break; | |
631 | } | |
632 | ||
633 | if (ret == LEB_FREED_IDX) { | |
634 | /* | |
635 | * This was an indexing LEB and it cannot be | |
636 | * immediately used. And instead of requesting the | |
637 | * commit straight away, we try to garbage collect some | |
638 | * more. | |
639 | */ | |
640 | dbg_gc("indexing LEB %d freed, continue", lp.lnum); | |
641 | continue; | |
642 | } | |
643 | ||
644 | ubifs_assert(ret == LEB_RETAINED); | |
645 | space_after = c->leb_size - wbuf->offs - wbuf->used; | |
646 | dbg_gc("LEB %d retained, freed %d bytes", lp.lnum, | |
647 | space_after - space_before); | |
648 | ||
649 | if (space_after > space_before) { | |
650 | /* GC makes progress, keep working */ | |
651 | min_space >>= 1; | |
652 | if (min_space < c->dead_wm) | |
653 | min_space = c->dead_wm; | |
654 | continue; | |
655 | } | |
656 | ||
657 | dbg_gc("did not make progress"); | |
658 | ||
659 | /* | |
660 | * GC moved an LEB bud have not done any progress. This means | |
661 | * that the previous GC head LEB contained too few free space | |
662 | * and the LEB which was GC'ed contained only large nodes which | |
663 | * did not fit that space. | |
664 | * | |
665 | * We can do 2 things: | |
666 | * 1. pick another LEB in a hope it'll contain a small node | |
667 | * which will fit the space we have at the end of current GC | |
668 | * head LEB, but there is no guarantee, so we try this out | |
669 | * unless we have already been working for too long; | |
670 | * 2. request an LEB with more dirty space, which will force | |
671 | * 'ubifs_find_dirty_leb()' to start scanning the lprops | |
672 | * table, instead of just picking one from the heap | |
673 | * (previously it already picked the dirtiest LEB). | |
674 | */ | |
675 | if (i < SOFT_LEBS_LIMIT) { | |
676 | dbg_gc("try again"); | |
677 | continue; | |
678 | } | |
679 | ||
680 | min_space <<= 1; | |
681 | if (min_space > c->dark_wm) | |
682 | min_space = c->dark_wm; | |
683 | dbg_gc("set min. space to %d", min_space); | |
684 | } | |
685 | ||
686 | if (ret == -ENOSPC && !list_empty(&c->idx_gc)) { | |
687 | dbg_gc("no space, some index LEBs GC'ed, -EAGAIN"); | |
688 | ubifs_commit_required(c); | |
689 | ret = -EAGAIN; | |
690 | } | |
691 | ||
692 | err = ubifs_wbuf_sync_nolock(wbuf); | |
693 | if (!err) | |
694 | err = ubifs_leb_unmap(c, c->gc_lnum); | |
695 | if (err) { | |
696 | ret = err; | |
697 | goto out; | |
698 | } | |
699 | out_unlock: | |
700 | mutex_unlock(&wbuf->io_mutex); | |
701 | return ret; | |
702 | ||
703 | out: | |
704 | ubifs_assert(ret < 0); | |
705 | ubifs_assert(ret != -ENOSPC && ret != -EAGAIN); | |
706 | ubifs_ro_mode(c, ret); | |
707 | ubifs_wbuf_sync_nolock(wbuf); | |
708 | mutex_unlock(&wbuf->io_mutex); | |
709 | ubifs_return_leb(c, lp.lnum); | |
710 | return ret; | |
711 | } | |
712 | ||
713 | /** | |
714 | * ubifs_gc_start_commit - garbage collection at start of commit. | |
715 | * @c: UBIFS file-system description object | |
716 | * | |
717 | * If a LEB has only dirty and free space, then we may safely unmap it and make | |
718 | * it free. Note, we cannot do this with indexing LEBs because dirty space may | |
719 | * correspond index nodes that are required for recovery. In that case, the | |
720 | * LEB cannot be unmapped until after the next commit. | |
721 | * | |
722 | * This function returns %0 upon success and a negative error code upon failure. | |
723 | */ | |
724 | int ubifs_gc_start_commit(struct ubifs_info *c) | |
725 | { | |
726 | struct ubifs_gced_idx_leb *idx_gc; | |
727 | const struct ubifs_lprops *lp; | |
728 | int err = 0, flags; | |
729 | ||
730 | ubifs_get_lprops(c); | |
731 | ||
732 | /* | |
733 | * Unmap (non-index) freeable LEBs. Note that recovery requires that all | |
734 | * wbufs are sync'd before this, which is done in 'do_commit()'. | |
735 | */ | |
736 | while (1) { | |
737 | lp = ubifs_fast_find_freeable(c); | |
8d47aef4 | 738 | if (IS_ERR(lp)) { |
1e51764a AB |
739 | err = PTR_ERR(lp); |
740 | goto out; | |
741 | } | |
742 | if (!lp) | |
743 | break; | |
744 | ubifs_assert(!(lp->flags & LPROPS_TAKEN)); | |
745 | ubifs_assert(!(lp->flags & LPROPS_INDEX)); | |
746 | err = ubifs_leb_unmap(c, lp->lnum); | |
747 | if (err) | |
748 | goto out; | |
749 | lp = ubifs_change_lp(c, lp, c->leb_size, 0, lp->flags, 0); | |
8d47aef4 | 750 | if (IS_ERR(lp)) { |
1e51764a AB |
751 | err = PTR_ERR(lp); |
752 | goto out; | |
753 | } | |
754 | ubifs_assert(!(lp->flags & LPROPS_TAKEN)); | |
755 | ubifs_assert(!(lp->flags & LPROPS_INDEX)); | |
756 | } | |
757 | ||
758 | /* Mark GC'd index LEBs OK to unmap after this commit finishes */ | |
759 | list_for_each_entry(idx_gc, &c->idx_gc, list) | |
760 | idx_gc->unmap = 1; | |
761 | ||
762 | /* Record index freeable LEBs for unmapping after commit */ | |
763 | while (1) { | |
764 | lp = ubifs_fast_find_frdi_idx(c); | |
8d47aef4 | 765 | if (IS_ERR(lp)) { |
1e51764a AB |
766 | err = PTR_ERR(lp); |
767 | goto out; | |
768 | } | |
769 | if (!lp) | |
770 | break; | |
771 | idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS); | |
772 | if (!idx_gc) { | |
773 | err = -ENOMEM; | |
774 | goto out; | |
775 | } | |
776 | ubifs_assert(!(lp->flags & LPROPS_TAKEN)); | |
777 | ubifs_assert(lp->flags & LPROPS_INDEX); | |
778 | /* Don't release the LEB until after the next commit */ | |
779 | flags = (lp->flags | LPROPS_TAKEN) ^ LPROPS_INDEX; | |
780 | lp = ubifs_change_lp(c, lp, c->leb_size, 0, flags, 1); | |
8d47aef4 | 781 | if (IS_ERR(lp)) { |
1e51764a AB |
782 | err = PTR_ERR(lp); |
783 | kfree(idx_gc); | |
784 | goto out; | |
785 | } | |
786 | ubifs_assert(lp->flags & LPROPS_TAKEN); | |
787 | ubifs_assert(!(lp->flags & LPROPS_INDEX)); | |
788 | idx_gc->lnum = lp->lnum; | |
789 | idx_gc->unmap = 1; | |
790 | list_add(&idx_gc->list, &c->idx_gc); | |
791 | } | |
792 | out: | |
793 | ubifs_release_lprops(c); | |
794 | return err; | |
795 | } | |
796 | ||
797 | /** | |
798 | * ubifs_gc_end_commit - garbage collection at end of commit. | |
799 | * @c: UBIFS file-system description object | |
800 | * | |
801 | * This function completes out-of-place garbage collection of index LEBs. | |
802 | */ | |
803 | int ubifs_gc_end_commit(struct ubifs_info *c) | |
804 | { | |
805 | struct ubifs_gced_idx_leb *idx_gc, *tmp; | |
806 | struct ubifs_wbuf *wbuf; | |
807 | int err = 0; | |
808 | ||
809 | wbuf = &c->jheads[GCHD].wbuf; | |
810 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
811 | list_for_each_entry_safe(idx_gc, tmp, &c->idx_gc, list) | |
812 | if (idx_gc->unmap) { | |
813 | dbg_gc("LEB %d", idx_gc->lnum); | |
814 | err = ubifs_leb_unmap(c, idx_gc->lnum); | |
815 | if (err) | |
816 | goto out; | |
817 | err = ubifs_change_one_lp(c, idx_gc->lnum, LPROPS_NC, | |
818 | LPROPS_NC, 0, LPROPS_TAKEN, -1); | |
819 | if (err) | |
820 | goto out; | |
821 | list_del(&idx_gc->list); | |
822 | kfree(idx_gc); | |
823 | } | |
824 | out: | |
825 | mutex_unlock(&wbuf->io_mutex); | |
826 | return err; | |
827 | } | |
828 | ||
829 | /** | |
830 | * ubifs_destroy_idx_gc - destroy idx_gc list. | |
831 | * @c: UBIFS file-system description object | |
832 | * | |
e4d9b6cb AB |
833 | * This function destroys the @c->idx_gc list. It is called when unmounting or |
834 | * remounting read-only so locks are not needed. Returns zero in case of | |
835 | * success and a negative error code in case of failure. | |
1e51764a | 836 | */ |
e4d9b6cb | 837 | int ubifs_destroy_idx_gc(struct ubifs_info *c) |
1e51764a | 838 | { |
e4d9b6cb AB |
839 | int ret = 0; |
840 | ||
1e51764a | 841 | while (!list_empty(&c->idx_gc)) { |
e4d9b6cb | 842 | int err; |
1e51764a AB |
843 | struct ubifs_gced_idx_leb *idx_gc; |
844 | ||
845 | idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb, | |
846 | list); | |
e4d9b6cb AB |
847 | err = ubifs_change_one_lp(c, idx_gc->lnum, LPROPS_NC, |
848 | LPROPS_NC, 0, LPROPS_TAKEN, -1); | |
849 | if (err && !ret) | |
850 | ret = err; | |
1e51764a AB |
851 | list_del(&idx_gc->list); |
852 | kfree(idx_gc); | |
853 | } | |
854 | ||
e4d9b6cb | 855 | return ret; |
1e51764a AB |
856 | } |
857 | ||
858 | /** | |
859 | * ubifs_get_idx_gc_leb - get a LEB from GC'd index LEB list. | |
860 | * @c: UBIFS file-system description object | |
861 | * | |
862 | * Called during start commit so locks are not needed. | |
863 | */ | |
864 | int ubifs_get_idx_gc_leb(struct ubifs_info *c) | |
865 | { | |
866 | struct ubifs_gced_idx_leb *idx_gc; | |
867 | int lnum; | |
868 | ||
869 | if (list_empty(&c->idx_gc)) | |
870 | return -ENOSPC; | |
871 | idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb, list); | |
872 | lnum = idx_gc->lnum; | |
873 | /* c->idx_gc_cnt is updated by the caller when lprops are updated */ | |
874 | list_del(&idx_gc->list); | |
875 | kfree(idx_gc); | |
876 | return lnum; | |
877 | } |