]>
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 the budgeting sub-system which is responsible for UBIFS | |
25 | * space management. | |
26 | * | |
27 | * Factors such as compression, wasted space at the ends of LEBs, space in other | |
28 | * journal heads, the effect of updates on the index, and so on, make it | |
29 | * impossible to accurately predict the amount of space needed. Consequently | |
30 | * approximations are used. | |
31 | */ | |
32 | ||
33 | #include "ubifs.h" | |
34 | #include <linux/writeback.h> | |
4d61db4f | 35 | #include <linux/math64.h> |
1e51764a AB |
36 | |
37 | /* | |
38 | * When pessimistic budget calculations say that there is no enough space, | |
39 | * UBIFS starts writing back dirty inodes and pages, doing garbage collection, | |
2acf8067 | 40 | * or committing. The below constant defines maximum number of times UBIFS |
1e51764a AB |
41 | * repeats the operations. |
42 | */ | |
2acf8067 | 43 | #define MAX_MKSPC_RETRIES 3 |
1e51764a AB |
44 | |
45 | /* | |
46 | * The below constant defines amount of dirty pages which should be written | |
47 | * back at when trying to shrink the liability. | |
48 | */ | |
49 | #define NR_TO_WRITE 16 | |
50 | ||
1e51764a AB |
51 | /** |
52 | * shrink_liability - write-back some dirty pages/inodes. | |
53 | * @c: UBIFS file-system description object | |
54 | * @nr_to_write: how many dirty pages to write-back | |
55 | * | |
56 | * This function shrinks UBIFS liability by means of writing back some amount | |
b6e51316 | 57 | * of dirty inodes and their pages. |
1e51764a AB |
58 | * |
59 | * Note, this function synchronizes even VFS inodes which are locked | |
60 | * (@i_mutex) by the caller of the budgeting function, because write-back does | |
61 | * not touch @i_mutex. | |
62 | */ | |
b6e51316 | 63 | static void shrink_liability(struct ubifs_info *c, int nr_to_write) |
1e51764a | 64 | { |
cf37e972 | 65 | down_read(&c->vfs_sb->s_umount); |
0e175a18 | 66 | writeback_inodes_sb(c->vfs_sb, WB_REASON_FS_FREE_SPACE); |
cf37e972 | 67 | up_read(&c->vfs_sb->s_umount); |
1e51764a AB |
68 | } |
69 | ||
1e51764a AB |
70 | /** |
71 | * run_gc - run garbage collector. | |
72 | * @c: UBIFS file-system description object | |
73 | * | |
74 | * This function runs garbage collector to make some more free space. Returns | |
75 | * zero if a free LEB has been produced, %-EAGAIN if commit is required, and a | |
76 | * negative error code in case of failure. | |
77 | */ | |
78 | static int run_gc(struct ubifs_info *c) | |
79 | { | |
80 | int err, lnum; | |
81 | ||
82 | /* Make some free space by garbage-collecting dirty space */ | |
83 | down_read(&c->commit_sem); | |
84 | lnum = ubifs_garbage_collect(c, 1); | |
85 | up_read(&c->commit_sem); | |
86 | if (lnum < 0) | |
87 | return lnum; | |
88 | ||
89 | /* GC freed one LEB, return it to lprops */ | |
90 | dbg_budg("GC freed LEB %d", lnum); | |
91 | err = ubifs_return_leb(c, lnum); | |
92 | if (err) | |
93 | return err; | |
94 | return 0; | |
95 | } | |
96 | ||
2acf8067 AB |
97 | /** |
98 | * get_liability - calculate current liability. | |
99 | * @c: UBIFS file-system description object | |
100 | * | |
101 | * This function calculates and returns current UBIFS liability, i.e. the | |
102 | * amount of bytes UBIFS has "promised" to write to the media. | |
103 | */ | |
104 | static long long get_liability(struct ubifs_info *c) | |
105 | { | |
106 | long long liab; | |
107 | ||
108 | spin_lock(&c->space_lock); | |
b137545c | 109 | liab = c->bi.idx_growth + c->bi.data_growth + c->bi.dd_growth; |
2acf8067 AB |
110 | spin_unlock(&c->space_lock); |
111 | return liab; | |
112 | } | |
113 | ||
1e51764a AB |
114 | /** |
115 | * make_free_space - make more free space on the file-system. | |
116 | * @c: UBIFS file-system description object | |
1e51764a AB |
117 | * |
118 | * This function is called when an operation cannot be budgeted because there | |
119 | * is supposedly no free space. But in most cases there is some free space: | |
025dfdaf | 120 | * o budgeting is pessimistic, so it always budgets more than it is actually |
1e51764a AB |
121 | * needed, so shrinking the liability is one way to make free space - the |
122 | * cached data will take less space then it was budgeted for; | |
123 | * o GC may turn some dark space into free space (budgeting treats dark space | |
124 | * as not available); | |
125 | * o commit may free some LEB, i.e., turn freeable LEBs into free LEBs. | |
126 | * | |
127 | * So this function tries to do the above. Returns %-EAGAIN if some free space | |
128 | * was presumably made and the caller has to re-try budgeting the operation. | |
129 | * Returns %-ENOSPC if it couldn't do more free space, and other negative error | |
130 | * codes on failures. | |
131 | */ | |
2acf8067 | 132 | static int make_free_space(struct ubifs_info *c) |
1e51764a | 133 | { |
2acf8067 AB |
134 | int err, retries = 0; |
135 | long long liab1, liab2; | |
1e51764a | 136 | |
2acf8067 AB |
137 | do { |
138 | liab1 = get_liability(c); | |
139 | /* | |
140 | * We probably have some dirty pages or inodes (liability), try | |
141 | * to write them back. | |
142 | */ | |
143 | dbg_budg("liability %lld, run write-back", liab1); | |
144 | shrink_liability(c, NR_TO_WRITE); | |
1e51764a | 145 | |
2acf8067 AB |
146 | liab2 = get_liability(c); |
147 | if (liab2 < liab1) | |
148 | return -EAGAIN; | |
1e51764a | 149 | |
25985edc | 150 | dbg_budg("new liability %lld (not shrunk)", liab2); |
1e51764a | 151 | |
2acf8067 AB |
152 | /* Liability did not shrink again, try GC */ |
153 | dbg_budg("Run GC"); | |
1e51764a AB |
154 | err = run_gc(c); |
155 | if (!err) | |
156 | return -EAGAIN; | |
157 | ||
2acf8067 AB |
158 | if (err != -EAGAIN && err != -ENOSPC) |
159 | /* Some real error happened */ | |
1e51764a | 160 | return err; |
1e51764a | 161 | |
2acf8067 | 162 | dbg_budg("Run commit (retries %d)", retries); |
1e51764a AB |
163 | err = ubifs_run_commit(c); |
164 | if (err) | |
165 | return err; | |
2acf8067 AB |
166 | } while (retries++ < MAX_MKSPC_RETRIES); |
167 | ||
1e51764a AB |
168 | return -ENOSPC; |
169 | } | |
170 | ||
171 | /** | |
fb1cd01a | 172 | * ubifs_calc_min_idx_lebs - calculate amount of LEBs for the index. |
1e51764a AB |
173 | * @c: UBIFS file-system description object |
174 | * | |
fb1cd01a AB |
175 | * This function calculates and returns the number of LEBs which should be kept |
176 | * for index usage. | |
1e51764a AB |
177 | */ |
178 | int ubifs_calc_min_idx_lebs(struct ubifs_info *c) | |
179 | { | |
fb1cd01a | 180 | int idx_lebs; |
4d61db4f | 181 | long long idx_size; |
1e51764a | 182 | |
b137545c | 183 | idx_size = c->bi.old_idx_sz + c->bi.idx_growth + c->bi.uncommitted_idx; |
3a13252c | 184 | /* And make sure we have thrice the index size of space reserved */ |
fb1cd01a | 185 | idx_size += idx_size << 1; |
1e51764a AB |
186 | /* |
187 | * We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes' | |
188 | * pair, nor similarly the two variables for the new index size, so we | |
189 | * have to do this costly 64-bit division on fast-path. | |
190 | */ | |
fb1cd01a | 191 | idx_lebs = div_u64(idx_size + c->idx_leb_size - 1, c->idx_leb_size); |
1e51764a AB |
192 | /* |
193 | * The index head is not available for the in-the-gaps method, so add an | |
194 | * extra LEB to compensate. | |
195 | */ | |
4d61db4f AB |
196 | idx_lebs += 1; |
197 | if (idx_lebs < MIN_INDEX_LEBS) | |
198 | idx_lebs = MIN_INDEX_LEBS; | |
199 | return idx_lebs; | |
1e51764a AB |
200 | } |
201 | ||
202 | /** | |
203 | * ubifs_calc_available - calculate available FS space. | |
204 | * @c: UBIFS file-system description object | |
205 | * @min_idx_lebs: minimum number of LEBs reserved for the index | |
206 | * | |
207 | * This function calculates and returns amount of FS space available for use. | |
208 | */ | |
209 | long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs) | |
210 | { | |
211 | int subtract_lebs; | |
212 | long long available; | |
213 | ||
1e51764a AB |
214 | available = c->main_bytes - c->lst.total_used; |
215 | ||
216 | /* | |
217 | * Now 'available' contains theoretically available flash space | |
218 | * assuming there is no index, so we have to subtract the space which | |
219 | * is reserved for the index. | |
220 | */ | |
221 | subtract_lebs = min_idx_lebs; | |
222 | ||
223 | /* Take into account that GC reserves one LEB for its own needs */ | |
224 | subtract_lebs += 1; | |
225 | ||
226 | /* | |
227 | * The GC journal head LEB is not really accessible. And since | |
228 | * different write types go to different heads, we may count only on | |
229 | * one head's space. | |
230 | */ | |
231 | subtract_lebs += c->jhead_cnt - 1; | |
232 | ||
233 | /* We also reserve one LEB for deletions, which bypass budgeting */ | |
234 | subtract_lebs += 1; | |
235 | ||
236 | available -= (long long)subtract_lebs * c->leb_size; | |
237 | ||
238 | /* Subtract the dead space which is not available for use */ | |
239 | available -= c->lst.total_dead; | |
240 | ||
241 | /* | |
242 | * Subtract dark space, which might or might not be usable - it depends | |
243 | * on the data which we have on the media and which will be written. If | |
244 | * this is a lot of uncompressed or not-compressible data, the dark | |
245 | * space cannot be used. | |
246 | */ | |
247 | available -= c->lst.total_dark; | |
248 | ||
249 | /* | |
250 | * However, there is more dark space. The index may be bigger than | |
251 | * @min_idx_lebs. Those extra LEBs are assumed to be available, but | |
252 | * their dark space is not included in total_dark, so it is subtracted | |
253 | * here. | |
254 | */ | |
255 | if (c->lst.idx_lebs > min_idx_lebs) { | |
256 | subtract_lebs = c->lst.idx_lebs - min_idx_lebs; | |
257 | available -= subtract_lebs * c->dark_wm; | |
258 | } | |
259 | ||
260 | /* The calculations are rough and may end up with a negative number */ | |
261 | return available > 0 ? available : 0; | |
262 | } | |
263 | ||
264 | /** | |
265 | * can_use_rp - check whether the user is allowed to use reserved pool. | |
266 | * @c: UBIFS file-system description object | |
267 | * | |
268 | * UBIFS has so-called "reserved pool" which is flash space reserved | |
269 | * for the superuser and for uses whose UID/GID is recorded in UBIFS superblock. | |
270 | * This function checks whether current user is allowed to use reserved pool. | |
271 | * Returns %1 current user is allowed to use reserved pool and %0 otherwise. | |
272 | */ | |
273 | static int can_use_rp(struct ubifs_info *c) | |
274 | { | |
39241beb EB |
275 | if (uid_eq(current_fsuid(), c->rp_uid) || capable(CAP_SYS_RESOURCE) || |
276 | (!gid_eq(c->rp_gid, GLOBAL_ROOT_GID) && in_group_p(c->rp_gid))) | |
1e51764a AB |
277 | return 1; |
278 | return 0; | |
279 | } | |
280 | ||
281 | /** | |
282 | * do_budget_space - reserve flash space for index and data growth. | |
283 | * @c: UBIFS file-system description object | |
284 | * | |
fb1cd01a AB |
285 | * This function makes sure UBIFS has enough free LEBs for index growth and |
286 | * data. | |
1e51764a | 287 | * |
3a13252c | 288 | * When budgeting index space, UBIFS reserves thrice as many LEBs as the index |
1e51764a AB |
289 | * would take if it was consolidated and written to the flash. This guarantees |
290 | * that the "in-the-gaps" commit method always succeeds and UBIFS will always | |
291 | * be able to commit dirty index. So this function basically adds amount of | |
b364b41a | 292 | * budgeted index space to the size of the current index, multiplies this by 3, |
fb1cd01a | 293 | * and makes sure this does not exceed the amount of free LEBs. |
1e51764a | 294 | * |
b137545c | 295 | * Notes about @c->bi.min_idx_lebs and @c->lst.idx_lebs variables: |
1e51764a AB |
296 | * o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might |
297 | * be large, because UBIFS does not do any index consolidation as long as | |
298 | * there is free space. IOW, the index may take a lot of LEBs, but the LEBs | |
299 | * will contain a lot of dirt. | |
b137545c AB |
300 | * o @c->bi.min_idx_lebs is the number of LEBS the index presumably takes. IOW, |
301 | * the index may be consolidated to take up to @c->bi.min_idx_lebs LEBs. | |
1e51764a AB |
302 | * |
303 | * This function returns zero in case of success, and %-ENOSPC in case of | |
304 | * failure. | |
305 | */ | |
306 | static int do_budget_space(struct ubifs_info *c) | |
307 | { | |
308 | long long outstanding, available; | |
309 | int lebs, rsvd_idx_lebs, min_idx_lebs; | |
310 | ||
311 | /* First budget index space */ | |
312 | min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
313 | ||
314 | /* Now 'min_idx_lebs' contains number of LEBs to reserve */ | |
315 | if (min_idx_lebs > c->lst.idx_lebs) | |
316 | rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs; | |
317 | else | |
318 | rsvd_idx_lebs = 0; | |
319 | ||
320 | /* | |
321 | * The number of LEBs that are available to be used by the index is: | |
322 | * | |
323 | * @c->lst.empty_lebs + @c->freeable_cnt + @c->idx_gc_cnt - | |
324 | * @c->lst.taken_empty_lebs | |
325 | * | |
948cfb21 AB |
326 | * @c->lst.empty_lebs are available because they are empty. |
327 | * @c->freeable_cnt are available because they contain only free and | |
328 | * dirty space, @c->idx_gc_cnt are available because they are index | |
329 | * LEBs that have been garbage collected and are awaiting the commit | |
330 | * before they can be used. And the in-the-gaps method will grab these | |
331 | * if it needs them. @c->lst.taken_empty_lebs are empty LEBs that have | |
332 | * already been allocated for some purpose. | |
1e51764a | 333 | * |
948cfb21 AB |
334 | * Note, @c->idx_gc_cnt is included to both @c->lst.empty_lebs (because |
335 | * these LEBs are empty) and to @c->lst.taken_empty_lebs (because they | |
336 | * are taken until after the commit). | |
337 | * | |
338 | * Note, @c->lst.taken_empty_lebs may temporarily be higher by one | |
339 | * because of the way we serialize LEB allocations and budgeting. See a | |
340 | * comment in 'ubifs_find_free_space()'. | |
1e51764a AB |
341 | */ |
342 | lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt - | |
343 | c->lst.taken_empty_lebs; | |
344 | if (unlikely(rsvd_idx_lebs > lebs)) { | |
79fda517 AB |
345 | dbg_budg("out of indexing space: min_idx_lebs %d (old %d), rsvd_idx_lebs %d", |
346 | min_idx_lebs, c->bi.min_idx_lebs, rsvd_idx_lebs); | |
1e51764a AB |
347 | return -ENOSPC; |
348 | } | |
349 | ||
350 | available = ubifs_calc_available(c, min_idx_lebs); | |
b137545c | 351 | outstanding = c->bi.data_growth + c->bi.dd_growth; |
1e51764a AB |
352 | |
353 | if (unlikely(available < outstanding)) { | |
354 | dbg_budg("out of data space: available %lld, outstanding %lld", | |
355 | available, outstanding); | |
356 | return -ENOSPC; | |
357 | } | |
358 | ||
359 | if (available - outstanding <= c->rp_size && !can_use_rp(c)) | |
360 | return -ENOSPC; | |
361 | ||
b137545c | 362 | c->bi.min_idx_lebs = min_idx_lebs; |
1e51764a AB |
363 | return 0; |
364 | } | |
365 | ||
366 | /** | |
367 | * calc_idx_growth - calculate approximate index growth from budgeting request. | |
368 | * @c: UBIFS file-system description object | |
369 | * @req: budgeting request | |
370 | * | |
371 | * For now we assume each new node adds one znode. But this is rather poor | |
372 | * approximation, though. | |
373 | */ | |
374 | static int calc_idx_growth(const struct ubifs_info *c, | |
375 | const struct ubifs_budget_req *req) | |
376 | { | |
377 | int znodes; | |
378 | ||
379 | znodes = req->new_ino + (req->new_page << UBIFS_BLOCKS_PER_PAGE_SHIFT) + | |
380 | req->new_dent; | |
381 | return znodes * c->max_idx_node_sz; | |
382 | } | |
383 | ||
384 | /** | |
385 | * calc_data_growth - calculate approximate amount of new data from budgeting | |
386 | * request. | |
387 | * @c: UBIFS file-system description object | |
388 | * @req: budgeting request | |
389 | */ | |
390 | static int calc_data_growth(const struct ubifs_info *c, | |
391 | const struct ubifs_budget_req *req) | |
392 | { | |
393 | int data_growth; | |
394 | ||
b137545c | 395 | data_growth = req->new_ino ? c->bi.inode_budget : 0; |
1e51764a | 396 | if (req->new_page) |
b137545c | 397 | data_growth += c->bi.page_budget; |
1e51764a | 398 | if (req->new_dent) |
b137545c | 399 | data_growth += c->bi.dent_budget; |
1e51764a AB |
400 | data_growth += req->new_ino_d; |
401 | return data_growth; | |
402 | } | |
403 | ||
404 | /** | |
405 | * calc_dd_growth - calculate approximate amount of data which makes other data | |
406 | * dirty from budgeting request. | |
407 | * @c: UBIFS file-system description object | |
408 | * @req: budgeting request | |
409 | */ | |
410 | static int calc_dd_growth(const struct ubifs_info *c, | |
411 | const struct ubifs_budget_req *req) | |
412 | { | |
413 | int dd_growth; | |
414 | ||
b137545c | 415 | dd_growth = req->dirtied_page ? c->bi.page_budget : 0; |
1e51764a AB |
416 | |
417 | if (req->dirtied_ino) | |
b137545c | 418 | dd_growth += c->bi.inode_budget << (req->dirtied_ino - 1); |
1e51764a | 419 | if (req->mod_dent) |
b137545c | 420 | dd_growth += c->bi.dent_budget; |
1e51764a AB |
421 | dd_growth += req->dirtied_ino_d; |
422 | return dd_growth; | |
423 | } | |
424 | ||
425 | /** | |
426 | * ubifs_budget_space - ensure there is enough space to complete an operation. | |
427 | * @c: UBIFS file-system description object | |
428 | * @req: budget request | |
429 | * | |
430 | * This function allocates budget for an operation. It uses pessimistic | |
431 | * approximation of how much flash space the operation needs. The goal of this | |
432 | * function is to make sure UBIFS always has flash space to flush all dirty | |
433 | * pages, dirty inodes, and dirty znodes (liability). This function may force | |
434 | * commit, garbage-collection or write-back. Returns zero in case of success, | |
435 | * %-ENOSPC if there is no free space and other negative error codes in case of | |
436 | * failures. | |
437 | */ | |
438 | int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req) | |
439 | { | |
2acf8067 | 440 | int err, idx_growth, data_growth, dd_growth, retried = 0; |
1e51764a | 441 | |
547000da AB |
442 | ubifs_assert(req->new_page <= 1); |
443 | ubifs_assert(req->dirtied_page <= 1); | |
444 | ubifs_assert(req->new_dent <= 1); | |
445 | ubifs_assert(req->mod_dent <= 1); | |
446 | ubifs_assert(req->new_ino <= 1); | |
447 | ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA); | |
1e51764a AB |
448 | ubifs_assert(req->dirtied_ino <= 4); |
449 | ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4); | |
dab4b4d2 AB |
450 | ubifs_assert(!(req->new_ino_d & 7)); |
451 | ubifs_assert(!(req->dirtied_ino_d & 7)); | |
1e51764a AB |
452 | |
453 | data_growth = calc_data_growth(c, req); | |
454 | dd_growth = calc_dd_growth(c, req); | |
455 | if (!data_growth && !dd_growth) | |
456 | return 0; | |
457 | idx_growth = calc_idx_growth(c, req); | |
1e51764a AB |
458 | |
459 | again: | |
460 | spin_lock(&c->space_lock); | |
b137545c AB |
461 | ubifs_assert(c->bi.idx_growth >= 0); |
462 | ubifs_assert(c->bi.data_growth >= 0); | |
463 | ubifs_assert(c->bi.dd_growth >= 0); | |
1e51764a | 464 | |
b137545c | 465 | if (unlikely(c->bi.nospace) && (c->bi.nospace_rp || !can_use_rp(c))) { |
1e51764a AB |
466 | dbg_budg("no space"); |
467 | spin_unlock(&c->space_lock); | |
468 | return -ENOSPC; | |
469 | } | |
470 | ||
b137545c AB |
471 | c->bi.idx_growth += idx_growth; |
472 | c->bi.data_growth += data_growth; | |
473 | c->bi.dd_growth += dd_growth; | |
1e51764a AB |
474 | |
475 | err = do_budget_space(c); | |
476 | if (likely(!err)) { | |
477 | req->idx_growth = idx_growth; | |
478 | req->data_growth = data_growth; | |
479 | req->dd_growth = dd_growth; | |
480 | spin_unlock(&c->space_lock); | |
481 | return 0; | |
482 | } | |
483 | ||
484 | /* Restore the old values */ | |
b137545c AB |
485 | c->bi.idx_growth -= idx_growth; |
486 | c->bi.data_growth -= data_growth; | |
487 | c->bi.dd_growth -= dd_growth; | |
1e51764a AB |
488 | spin_unlock(&c->space_lock); |
489 | ||
490 | if (req->fast) { | |
491 | dbg_budg("no space for fast budgeting"); | |
492 | return err; | |
493 | } | |
494 | ||
2acf8067 AB |
495 | err = make_free_space(c); |
496 | cond_resched(); | |
1e51764a AB |
497 | if (err == -EAGAIN) { |
498 | dbg_budg("try again"); | |
1e51764a AB |
499 | goto again; |
500 | } else if (err == -ENOSPC) { | |
2acf8067 AB |
501 | if (!retried) { |
502 | retried = 1; | |
503 | dbg_budg("-ENOSPC, but anyway try once again"); | |
504 | goto again; | |
505 | } | |
1e51764a | 506 | dbg_budg("FS is full, -ENOSPC"); |
b137545c | 507 | c->bi.nospace = 1; |
1e51764a | 508 | if (can_use_rp(c) || c->rp_size == 0) |
b137545c | 509 | c->bi.nospace_rp = 1; |
1e51764a AB |
510 | smp_wmb(); |
511 | } else | |
512 | ubifs_err("cannot budget space, error %d", err); | |
513 | return err; | |
514 | } | |
515 | ||
516 | /** | |
517 | * ubifs_release_budget - release budgeted free space. | |
518 | * @c: UBIFS file-system description object | |
519 | * @req: budget request | |
520 | * | |
521 | * This function releases the space budgeted by 'ubifs_budget_space()'. Note, | |
522 | * since the index changes (which were budgeted for in @req->idx_growth) will | |
523 | * only be written to the media on commit, this function moves the index budget | |
b137545c AB |
524 | * from @c->bi.idx_growth to @c->bi.uncommitted_idx. The latter will be zeroed |
525 | * by the commit operation. | |
1e51764a AB |
526 | */ |
527 | void ubifs_release_budget(struct ubifs_info *c, struct ubifs_budget_req *req) | |
528 | { | |
547000da AB |
529 | ubifs_assert(req->new_page <= 1); |
530 | ubifs_assert(req->dirtied_page <= 1); | |
531 | ubifs_assert(req->new_dent <= 1); | |
532 | ubifs_assert(req->mod_dent <= 1); | |
533 | ubifs_assert(req->new_ino <= 1); | |
534 | ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA); | |
1e51764a AB |
535 | ubifs_assert(req->dirtied_ino <= 4); |
536 | ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4); | |
dab4b4d2 AB |
537 | ubifs_assert(!(req->new_ino_d & 7)); |
538 | ubifs_assert(!(req->dirtied_ino_d & 7)); | |
1e51764a AB |
539 | if (!req->recalculate) { |
540 | ubifs_assert(req->idx_growth >= 0); | |
541 | ubifs_assert(req->data_growth >= 0); | |
542 | ubifs_assert(req->dd_growth >= 0); | |
543 | } | |
544 | ||
545 | if (req->recalculate) { | |
546 | req->data_growth = calc_data_growth(c, req); | |
547 | req->dd_growth = calc_dd_growth(c, req); | |
548 | req->idx_growth = calc_idx_growth(c, req); | |
549 | } | |
550 | ||
551 | if (!req->data_growth && !req->dd_growth) | |
552 | return; | |
553 | ||
b137545c | 554 | c->bi.nospace = c->bi.nospace_rp = 0; |
1e51764a AB |
555 | smp_wmb(); |
556 | ||
557 | spin_lock(&c->space_lock); | |
b137545c AB |
558 | c->bi.idx_growth -= req->idx_growth; |
559 | c->bi.uncommitted_idx += req->idx_growth; | |
560 | c->bi.data_growth -= req->data_growth; | |
561 | c->bi.dd_growth -= req->dd_growth; | |
562 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
563 | ||
564 | ubifs_assert(c->bi.idx_growth >= 0); | |
565 | ubifs_assert(c->bi.data_growth >= 0); | |
566 | ubifs_assert(c->bi.dd_growth >= 0); | |
567 | ubifs_assert(c->bi.min_idx_lebs < c->main_lebs); | |
568 | ubifs_assert(!(c->bi.idx_growth & 7)); | |
569 | ubifs_assert(!(c->bi.data_growth & 7)); | |
570 | ubifs_assert(!(c->bi.dd_growth & 7)); | |
1e51764a AB |
571 | spin_unlock(&c->space_lock); |
572 | } | |
573 | ||
574 | /** | |
575 | * ubifs_convert_page_budget - convert budget of a new page. | |
576 | * @c: UBIFS file-system description object | |
577 | * | |
578 | * This function converts budget which was allocated for a new page of data to | |
025dfdaf | 579 | * the budget of changing an existing page of data. The latter is smaller than |
1e51764a AB |
580 | * the former, so this function only does simple re-calculation and does not |
581 | * involve any write-back. | |
582 | */ | |
583 | void ubifs_convert_page_budget(struct ubifs_info *c) | |
584 | { | |
585 | spin_lock(&c->space_lock); | |
586 | /* Release the index growth reservation */ | |
b137545c | 587 | c->bi.idx_growth -= c->max_idx_node_sz << UBIFS_BLOCKS_PER_PAGE_SHIFT; |
1e51764a | 588 | /* Release the data growth reservation */ |
b137545c | 589 | c->bi.data_growth -= c->bi.page_budget; |
1e51764a | 590 | /* Increase the dirty data growth reservation instead */ |
b137545c | 591 | c->bi.dd_growth += c->bi.page_budget; |
1e51764a | 592 | /* And re-calculate the indexing space reservation */ |
b137545c | 593 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); |
1e51764a AB |
594 | spin_unlock(&c->space_lock); |
595 | } | |
596 | ||
597 | /** | |
598 | * ubifs_release_dirty_inode_budget - release dirty inode budget. | |
599 | * @c: UBIFS file-system description object | |
600 | * @ui: UBIFS inode to release the budget for | |
601 | * | |
602 | * This function releases budget corresponding to a dirty inode. It is usually | |
603 | * called when after the inode has been written to the media and marked as | |
6d6cb0d6 | 604 | * clean. It also causes the "no space" flags to be cleared. |
1e51764a AB |
605 | */ |
606 | void ubifs_release_dirty_inode_budget(struct ubifs_info *c, | |
607 | struct ubifs_inode *ui) | |
608 | { | |
182854b4 | 609 | struct ubifs_budget_req req; |
1e51764a | 610 | |
182854b4 | 611 | memset(&req, 0, sizeof(struct ubifs_budget_req)); |
6d6cb0d6 | 612 | /* The "no space" flags will be cleared because dd_growth is > 0 */ |
b137545c | 613 | req.dd_growth = c->bi.inode_budget + ALIGN(ui->data_len, 8); |
1e51764a AB |
614 | ubifs_release_budget(c, &req); |
615 | } | |
616 | ||
4b5f2762 AB |
617 | /** |
618 | * ubifs_reported_space - calculate reported free space. | |
619 | * @c: the UBIFS file-system description object | |
620 | * @free: amount of free space | |
621 | * | |
622 | * This function calculates amount of free space which will be reported to | |
623 | * user-space. User-space application tend to expect that if the file-system | |
624 | * (e.g., via the 'statfs()' call) reports that it has N bytes available, they | |
625 | * are able to write a file of size N. UBIFS attaches node headers to each data | |
80736d41 AB |
626 | * node and it has to write indexing nodes as well. This introduces additional |
627 | * overhead, and UBIFS has to report slightly less free space to meet the above | |
628 | * expectations. | |
4b5f2762 AB |
629 | * |
630 | * This function assumes free space is made up of uncompressed data nodes and | |
631 | * full index nodes (one per data node, tripled because we always allow enough | |
632 | * space to write the index thrice). | |
633 | * | |
634 | * Note, the calculation is pessimistic, which means that most of the time | |
635 | * UBIFS reports less space than it actually has. | |
636 | */ | |
4d61db4f | 637 | long long ubifs_reported_space(const struct ubifs_info *c, long long free) |
4b5f2762 | 638 | { |
f171d4d7 | 639 | int divisor, factor, f; |
4b5f2762 AB |
640 | |
641 | /* | |
642 | * Reported space size is @free * X, where X is UBIFS block size | |
643 | * divided by UBIFS block size + all overhead one data block | |
644 | * introduces. The overhead is the node header + indexing overhead. | |
645 | * | |
f171d4d7 AB |
646 | * Indexing overhead calculations are based on the following formula: |
647 | * I = N/(f - 1) + 1, where I - number of indexing nodes, N - number | |
648 | * of data nodes, f - fanout. Because effective UBIFS fanout is twice | |
649 | * as less than maximum fanout, we assume that each data node | |
4b5f2762 | 650 | * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes. |
80736d41 | 651 | * Note, the multiplier 3 is because UBIFS reserves thrice as more space |
4b5f2762 AB |
652 | * for the index. |
653 | */ | |
f171d4d7 | 654 | f = c->fanout > 3 ? c->fanout >> 1 : 2; |
4b5f2762 AB |
655 | factor = UBIFS_BLOCK_SIZE; |
656 | divisor = UBIFS_MAX_DATA_NODE_SZ; | |
f171d4d7 | 657 | divisor += (c->max_idx_node_sz * 3) / (f - 1); |
4b5f2762 | 658 | free *= factor; |
4d61db4f | 659 | return div_u64(free, divisor); |
4b5f2762 AB |
660 | } |
661 | ||
1e51764a | 662 | /** |
84abf972 | 663 | * ubifs_get_free_space_nolock - return amount of free space. |
1e51764a AB |
664 | * @c: UBIFS file-system description object |
665 | * | |
7dad181b AB |
666 | * This function calculates amount of free space to report to user-space. |
667 | * | |
668 | * Because UBIFS may introduce substantial overhead (the index, node headers, | |
fb1cd01a AB |
669 | * alignment, wastage at the end of LEBs, etc), it cannot report real amount of |
670 | * free flash space it has (well, because not all dirty space is reclaimable, | |
671 | * UBIFS does not actually know the real amount). If UBIFS did so, it would | |
672 | * bread user expectations about what free space is. Users seem to accustomed | |
673 | * to assume that if the file-system reports N bytes of free space, they would | |
674 | * be able to fit a file of N bytes to the FS. This almost works for | |
7dad181b AB |
675 | * traditional file-systems, because they have way less overhead than UBIFS. |
676 | * So, to keep users happy, UBIFS tries to take the overhead into account. | |
1e51764a | 677 | */ |
84abf972 | 678 | long long ubifs_get_free_space_nolock(struct ubifs_info *c) |
1e51764a | 679 | { |
84abf972 | 680 | int rsvd_idx_lebs, lebs; |
1e51764a AB |
681 | long long available, outstanding, free; |
682 | ||
b137545c AB |
683 | ubifs_assert(c->bi.min_idx_lebs == ubifs_calc_min_idx_lebs(c)); |
684 | outstanding = c->bi.data_growth + c->bi.dd_growth; | |
685 | available = ubifs_calc_available(c, c->bi.min_idx_lebs); | |
7dad181b AB |
686 | |
687 | /* | |
688 | * When reporting free space to user-space, UBIFS guarantees that it is | |
689 | * possible to write a file of free space size. This means that for | |
690 | * empty LEBs we may use more precise calculations than | |
691 | * 'ubifs_calc_available()' is using. Namely, we know that in empty | |
692 | * LEBs we would waste only @c->leb_overhead bytes, not @c->dark_wm. | |
693 | * Thus, amend the available space. | |
694 | * | |
695 | * Note, the calculations below are similar to what we have in | |
696 | * 'do_budget_space()', so refer there for comments. | |
697 | */ | |
b137545c AB |
698 | if (c->bi.min_idx_lebs > c->lst.idx_lebs) |
699 | rsvd_idx_lebs = c->bi.min_idx_lebs - c->lst.idx_lebs; | |
7dad181b AB |
700 | else |
701 | rsvd_idx_lebs = 0; | |
702 | lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt - | |
703 | c->lst.taken_empty_lebs; | |
704 | lebs -= rsvd_idx_lebs; | |
705 | available += lebs * (c->dark_wm - c->leb_overhead); | |
1e51764a AB |
706 | |
707 | if (available > outstanding) | |
708 | free = ubifs_reported_space(c, available - outstanding); | |
709 | else | |
710 | free = 0; | |
711 | return free; | |
712 | } | |
84abf972 AB |
713 | |
714 | /** | |
715 | * ubifs_get_free_space - return amount of free space. | |
716 | * @c: UBIFS file-system description object | |
717 | * | |
055da1b7 | 718 | * This function calculates and returns amount of free space to report to |
84abf972 AB |
719 | * user-space. |
720 | */ | |
721 | long long ubifs_get_free_space(struct ubifs_info *c) | |
722 | { | |
723 | long long free; | |
724 | ||
725 | spin_lock(&c->space_lock); | |
726 | free = ubifs_get_free_space_nolock(c); | |
727 | spin_unlock(&c->space_lock); | |
728 | ||
729 | return free; | |
730 | } |