Commit | Line | Data |
---|---|---|
2b27bdcc | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1e51764a AB |
2 | /* |
3 | * This file is part of UBIFS. | |
4 | * | |
5 | * Copyright (C) 2006-2008 Nokia Corporation. | |
6 | * | |
1e51764a AB |
7 | * Authors: Adrian Hunter |
8 | * Artem Bityutskiy (Битюцкий Артём) | |
9 | */ | |
10 | ||
11 | /* | |
12 | * This file implements the budgeting sub-system which is responsible for UBIFS | |
13 | * space management. | |
14 | * | |
15 | * Factors such as compression, wasted space at the ends of LEBs, space in other | |
16 | * journal heads, the effect of updates on the index, and so on, make it | |
17 | * impossible to accurately predict the amount of space needed. Consequently | |
18 | * approximations are used. | |
19 | */ | |
20 | ||
21 | #include "ubifs.h" | |
22 | #include <linux/writeback.h> | |
4d61db4f | 23 | #include <linux/math64.h> |
1e51764a AB |
24 | |
25 | /* | |
26 | * When pessimistic budget calculations say that there is no enough space, | |
27 | * UBIFS starts writing back dirty inodes and pages, doing garbage collection, | |
2acf8067 | 28 | * or committing. The below constant defines maximum number of times UBIFS |
1e51764a AB |
29 | * repeats the operations. |
30 | */ | |
2acf8067 | 31 | #define MAX_MKSPC_RETRIES 3 |
1e51764a AB |
32 | |
33 | /* | |
34 | * The below constant defines amount of dirty pages which should be written | |
35 | * back at when trying to shrink the liability. | |
36 | */ | |
37 | #define NR_TO_WRITE 16 | |
38 | ||
1e51764a AB |
39 | /** |
40 | * shrink_liability - write-back some dirty pages/inodes. | |
41 | * @c: UBIFS file-system description object | |
42 | * @nr_to_write: how many dirty pages to write-back | |
43 | * | |
44 | * This function shrinks UBIFS liability by means of writing back some amount | |
b6e51316 | 45 | * of dirty inodes and their pages. |
1e51764a AB |
46 | * |
47 | * Note, this function synchronizes even VFS inodes which are locked | |
48 | * (@i_mutex) by the caller of the budgeting function, because write-back does | |
49 | * not touch @i_mutex. | |
50 | */ | |
b6e51316 | 51 | static void shrink_liability(struct ubifs_info *c, int nr_to_write) |
1e51764a | 52 | { |
cf37e972 | 53 | down_read(&c->vfs_sb->s_umount); |
0af83abb | 54 | writeback_inodes_sb_nr(c->vfs_sb, nr_to_write, WB_REASON_FS_FREE_SPACE); |
cf37e972 | 55 | up_read(&c->vfs_sb->s_umount); |
1e51764a AB |
56 | } |
57 | ||
1e51764a AB |
58 | /** |
59 | * run_gc - run garbage collector. | |
60 | * @c: UBIFS file-system description object | |
61 | * | |
62 | * This function runs garbage collector to make some more free space. Returns | |
63 | * zero if a free LEB has been produced, %-EAGAIN if commit is required, and a | |
64 | * negative error code in case of failure. | |
65 | */ | |
66 | static int run_gc(struct ubifs_info *c) | |
67 | { | |
68 | int err, lnum; | |
69 | ||
70 | /* Make some free space by garbage-collecting dirty space */ | |
71 | down_read(&c->commit_sem); | |
72 | lnum = ubifs_garbage_collect(c, 1); | |
73 | up_read(&c->commit_sem); | |
74 | if (lnum < 0) | |
75 | return lnum; | |
76 | ||
77 | /* GC freed one LEB, return it to lprops */ | |
78 | dbg_budg("GC freed LEB %d", lnum); | |
79 | err = ubifs_return_leb(c, lnum); | |
80 | if (err) | |
81 | return err; | |
82 | return 0; | |
83 | } | |
84 | ||
2acf8067 AB |
85 | /** |
86 | * get_liability - calculate current liability. | |
87 | * @c: UBIFS file-system description object | |
88 | * | |
89 | * This function calculates and returns current UBIFS liability, i.e. the | |
90 | * amount of bytes UBIFS has "promised" to write to the media. | |
91 | */ | |
92 | static long long get_liability(struct ubifs_info *c) | |
93 | { | |
94 | long long liab; | |
95 | ||
96 | spin_lock(&c->space_lock); | |
b137545c | 97 | liab = c->bi.idx_growth + c->bi.data_growth + c->bi.dd_growth; |
2acf8067 AB |
98 | spin_unlock(&c->space_lock); |
99 | return liab; | |
100 | } | |
101 | ||
1e51764a AB |
102 | /** |
103 | * make_free_space - make more free space on the file-system. | |
104 | * @c: UBIFS file-system description object | |
1e51764a AB |
105 | * |
106 | * This function is called when an operation cannot be budgeted because there | |
107 | * is supposedly no free space. But in most cases there is some free space: | |
025dfdaf | 108 | * o budgeting is pessimistic, so it always budgets more than it is actually |
1e51764a AB |
109 | * needed, so shrinking the liability is one way to make free space - the |
110 | * cached data will take less space then it was budgeted for; | |
111 | * o GC may turn some dark space into free space (budgeting treats dark space | |
112 | * as not available); | |
113 | * o commit may free some LEB, i.e., turn freeable LEBs into free LEBs. | |
114 | * | |
115 | * So this function tries to do the above. Returns %-EAGAIN if some free space | |
116 | * was presumably made and the caller has to re-try budgeting the operation. | |
117 | * Returns %-ENOSPC if it couldn't do more free space, and other negative error | |
118 | * codes on failures. | |
119 | */ | |
2acf8067 | 120 | static int make_free_space(struct ubifs_info *c) |
1e51764a | 121 | { |
2acf8067 AB |
122 | int err, retries = 0; |
123 | long long liab1, liab2; | |
1e51764a | 124 | |
2acf8067 AB |
125 | do { |
126 | liab1 = get_liability(c); | |
127 | /* | |
128 | * We probably have some dirty pages or inodes (liability), try | |
129 | * to write them back. | |
130 | */ | |
131 | dbg_budg("liability %lld, run write-back", liab1); | |
132 | shrink_liability(c, NR_TO_WRITE); | |
1e51764a | 133 | |
2acf8067 AB |
134 | liab2 = get_liability(c); |
135 | if (liab2 < liab1) | |
136 | return -EAGAIN; | |
1e51764a | 137 | |
25985edc | 138 | dbg_budg("new liability %lld (not shrunk)", liab2); |
1e51764a | 139 | |
2acf8067 AB |
140 | /* Liability did not shrink again, try GC */ |
141 | dbg_budg("Run GC"); | |
1e51764a AB |
142 | err = run_gc(c); |
143 | if (!err) | |
144 | return -EAGAIN; | |
145 | ||
2acf8067 AB |
146 | if (err != -EAGAIN && err != -ENOSPC) |
147 | /* Some real error happened */ | |
1e51764a | 148 | return err; |
1e51764a | 149 | |
2acf8067 | 150 | dbg_budg("Run commit (retries %d)", retries); |
1e51764a AB |
151 | err = ubifs_run_commit(c); |
152 | if (err) | |
153 | return err; | |
2acf8067 AB |
154 | } while (retries++ < MAX_MKSPC_RETRIES); |
155 | ||
1e51764a AB |
156 | return -ENOSPC; |
157 | } | |
158 | ||
159 | /** | |
fb1cd01a | 160 | * ubifs_calc_min_idx_lebs - calculate amount of LEBs for the index. |
1e51764a AB |
161 | * @c: UBIFS file-system description object |
162 | * | |
fb1cd01a AB |
163 | * This function calculates and returns the number of LEBs which should be kept |
164 | * for index usage. | |
1e51764a AB |
165 | */ |
166 | int ubifs_calc_min_idx_lebs(struct ubifs_info *c) | |
167 | { | |
fb1cd01a | 168 | int idx_lebs; |
4d61db4f | 169 | long long idx_size; |
1e51764a | 170 | |
b137545c | 171 | idx_size = c->bi.old_idx_sz + c->bi.idx_growth + c->bi.uncommitted_idx; |
3a13252c | 172 | /* And make sure we have thrice the index size of space reserved */ |
fb1cd01a | 173 | idx_size += idx_size << 1; |
1e51764a AB |
174 | /* |
175 | * We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes' | |
176 | * pair, nor similarly the two variables for the new index size, so we | |
177 | * have to do this costly 64-bit division on fast-path. | |
178 | */ | |
fb1cd01a | 179 | idx_lebs = div_u64(idx_size + c->idx_leb_size - 1, c->idx_leb_size); |
1e51764a AB |
180 | /* |
181 | * The index head is not available for the in-the-gaps method, so add an | |
182 | * extra LEB to compensate. | |
183 | */ | |
4d61db4f AB |
184 | idx_lebs += 1; |
185 | if (idx_lebs < MIN_INDEX_LEBS) | |
186 | idx_lebs = MIN_INDEX_LEBS; | |
187 | return idx_lebs; | |
1e51764a AB |
188 | } |
189 | ||
190 | /** | |
191 | * ubifs_calc_available - calculate available FS space. | |
192 | * @c: UBIFS file-system description object | |
193 | * @min_idx_lebs: minimum number of LEBs reserved for the index | |
194 | * | |
195 | * This function calculates and returns amount of FS space available for use. | |
196 | */ | |
197 | long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs) | |
198 | { | |
199 | int subtract_lebs; | |
200 | long long available; | |
201 | ||
1e51764a AB |
202 | available = c->main_bytes - c->lst.total_used; |
203 | ||
204 | /* | |
205 | * Now 'available' contains theoretically available flash space | |
206 | * assuming there is no index, so we have to subtract the space which | |
207 | * is reserved for the index. | |
208 | */ | |
209 | subtract_lebs = min_idx_lebs; | |
210 | ||
211 | /* Take into account that GC reserves one LEB for its own needs */ | |
212 | subtract_lebs += 1; | |
213 | ||
214 | /* | |
215 | * The GC journal head LEB is not really accessible. And since | |
216 | * different write types go to different heads, we may count only on | |
217 | * one head's space. | |
218 | */ | |
219 | subtract_lebs += c->jhead_cnt - 1; | |
220 | ||
221 | /* We also reserve one LEB for deletions, which bypass budgeting */ | |
222 | subtract_lebs += 1; | |
223 | ||
224 | available -= (long long)subtract_lebs * c->leb_size; | |
225 | ||
226 | /* Subtract the dead space which is not available for use */ | |
227 | available -= c->lst.total_dead; | |
228 | ||
229 | /* | |
230 | * Subtract dark space, which might or might not be usable - it depends | |
231 | * on the data which we have on the media and which will be written. If | |
232 | * this is a lot of uncompressed or not-compressible data, the dark | |
233 | * space cannot be used. | |
234 | */ | |
235 | available -= c->lst.total_dark; | |
236 | ||
237 | /* | |
238 | * However, there is more dark space. The index may be bigger than | |
239 | * @min_idx_lebs. Those extra LEBs are assumed to be available, but | |
240 | * their dark space is not included in total_dark, so it is subtracted | |
241 | * here. | |
242 | */ | |
243 | if (c->lst.idx_lebs > min_idx_lebs) { | |
244 | subtract_lebs = c->lst.idx_lebs - min_idx_lebs; | |
245 | available -= subtract_lebs * c->dark_wm; | |
246 | } | |
247 | ||
248 | /* The calculations are rough and may end up with a negative number */ | |
249 | return available > 0 ? available : 0; | |
250 | } | |
251 | ||
252 | /** | |
253 | * can_use_rp - check whether the user is allowed to use reserved pool. | |
254 | * @c: UBIFS file-system description object | |
255 | * | |
256 | * UBIFS has so-called "reserved pool" which is flash space reserved | |
257 | * for the superuser and for uses whose UID/GID is recorded in UBIFS superblock. | |
258 | * This function checks whether current user is allowed to use reserved pool. | |
259 | * Returns %1 current user is allowed to use reserved pool and %0 otherwise. | |
260 | */ | |
261 | static int can_use_rp(struct ubifs_info *c) | |
262 | { | |
39241beb EB |
263 | if (uid_eq(current_fsuid(), c->rp_uid) || capable(CAP_SYS_RESOURCE) || |
264 | (!gid_eq(c->rp_gid, GLOBAL_ROOT_GID) && in_group_p(c->rp_gid))) | |
1e51764a AB |
265 | return 1; |
266 | return 0; | |
267 | } | |
268 | ||
269 | /** | |
270 | * do_budget_space - reserve flash space for index and data growth. | |
271 | * @c: UBIFS file-system description object | |
272 | * | |
fb1cd01a AB |
273 | * This function makes sure UBIFS has enough free LEBs for index growth and |
274 | * data. | |
1e51764a | 275 | * |
3a13252c | 276 | * When budgeting index space, UBIFS reserves thrice as many LEBs as the index |
1e51764a AB |
277 | * would take if it was consolidated and written to the flash. This guarantees |
278 | * that the "in-the-gaps" commit method always succeeds and UBIFS will always | |
279 | * be able to commit dirty index. So this function basically adds amount of | |
b364b41a | 280 | * budgeted index space to the size of the current index, multiplies this by 3, |
fb1cd01a | 281 | * and makes sure this does not exceed the amount of free LEBs. |
1e51764a | 282 | * |
b137545c | 283 | * Notes about @c->bi.min_idx_lebs and @c->lst.idx_lebs variables: |
1e51764a AB |
284 | * o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might |
285 | * be large, because UBIFS does not do any index consolidation as long as | |
286 | * there is free space. IOW, the index may take a lot of LEBs, but the LEBs | |
287 | * will contain a lot of dirt. | |
b137545c AB |
288 | * o @c->bi.min_idx_lebs is the number of LEBS the index presumably takes. IOW, |
289 | * the index may be consolidated to take up to @c->bi.min_idx_lebs LEBs. | |
1e51764a AB |
290 | * |
291 | * This function returns zero in case of success, and %-ENOSPC in case of | |
292 | * failure. | |
293 | */ | |
294 | static int do_budget_space(struct ubifs_info *c) | |
295 | { | |
296 | long long outstanding, available; | |
297 | int lebs, rsvd_idx_lebs, min_idx_lebs; | |
298 | ||
299 | /* First budget index space */ | |
300 | min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
301 | ||
302 | /* Now 'min_idx_lebs' contains number of LEBs to reserve */ | |
303 | if (min_idx_lebs > c->lst.idx_lebs) | |
304 | rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs; | |
305 | else | |
306 | rsvd_idx_lebs = 0; | |
307 | ||
308 | /* | |
309 | * The number of LEBs that are available to be used by the index is: | |
310 | * | |
311 | * @c->lst.empty_lebs + @c->freeable_cnt + @c->idx_gc_cnt - | |
312 | * @c->lst.taken_empty_lebs | |
313 | * | |
948cfb21 AB |
314 | * @c->lst.empty_lebs are available because they are empty. |
315 | * @c->freeable_cnt are available because they contain only free and | |
316 | * dirty space, @c->idx_gc_cnt are available because they are index | |
317 | * LEBs that have been garbage collected and are awaiting the commit | |
318 | * before they can be used. And the in-the-gaps method will grab these | |
319 | * if it needs them. @c->lst.taken_empty_lebs are empty LEBs that have | |
320 | * already been allocated for some purpose. | |
1e51764a | 321 | * |
948cfb21 AB |
322 | * Note, @c->idx_gc_cnt is included to both @c->lst.empty_lebs (because |
323 | * these LEBs are empty) and to @c->lst.taken_empty_lebs (because they | |
324 | * are taken until after the commit). | |
325 | * | |
326 | * Note, @c->lst.taken_empty_lebs may temporarily be higher by one | |
327 | * because of the way we serialize LEB allocations and budgeting. See a | |
328 | * comment in 'ubifs_find_free_space()'. | |
1e51764a AB |
329 | */ |
330 | lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt - | |
331 | c->lst.taken_empty_lebs; | |
332 | if (unlikely(rsvd_idx_lebs > lebs)) { | |
79fda517 AB |
333 | dbg_budg("out of indexing space: min_idx_lebs %d (old %d), rsvd_idx_lebs %d", |
334 | min_idx_lebs, c->bi.min_idx_lebs, rsvd_idx_lebs); | |
1e51764a AB |
335 | return -ENOSPC; |
336 | } | |
337 | ||
338 | available = ubifs_calc_available(c, min_idx_lebs); | |
b137545c | 339 | outstanding = c->bi.data_growth + c->bi.dd_growth; |
1e51764a AB |
340 | |
341 | if (unlikely(available < outstanding)) { | |
342 | dbg_budg("out of data space: available %lld, outstanding %lld", | |
343 | available, outstanding); | |
344 | return -ENOSPC; | |
345 | } | |
346 | ||
347 | if (available - outstanding <= c->rp_size && !can_use_rp(c)) | |
348 | return -ENOSPC; | |
349 | ||
b137545c | 350 | c->bi.min_idx_lebs = min_idx_lebs; |
1e51764a AB |
351 | return 0; |
352 | } | |
353 | ||
354 | /** | |
355 | * calc_idx_growth - calculate approximate index growth from budgeting request. | |
356 | * @c: UBIFS file-system description object | |
357 | * @req: budgeting request | |
358 | * | |
359 | * For now we assume each new node adds one znode. But this is rather poor | |
360 | * approximation, though. | |
361 | */ | |
362 | static int calc_idx_growth(const struct ubifs_info *c, | |
363 | const struct ubifs_budget_req *req) | |
364 | { | |
365 | int znodes; | |
366 | ||
367 | znodes = req->new_ino + (req->new_page << UBIFS_BLOCKS_PER_PAGE_SHIFT) + | |
368 | req->new_dent; | |
369 | return znodes * c->max_idx_node_sz; | |
370 | } | |
371 | ||
372 | /** | |
373 | * calc_data_growth - calculate approximate amount of new data from budgeting | |
374 | * request. | |
375 | * @c: UBIFS file-system description object | |
376 | * @req: budgeting request | |
377 | */ | |
378 | static int calc_data_growth(const struct ubifs_info *c, | |
379 | const struct ubifs_budget_req *req) | |
380 | { | |
381 | int data_growth; | |
382 | ||
b137545c | 383 | data_growth = req->new_ino ? c->bi.inode_budget : 0; |
1e51764a | 384 | if (req->new_page) |
b137545c | 385 | data_growth += c->bi.page_budget; |
1e51764a | 386 | if (req->new_dent) |
b137545c | 387 | data_growth += c->bi.dent_budget; |
1e51764a AB |
388 | data_growth += req->new_ino_d; |
389 | return data_growth; | |
390 | } | |
391 | ||
392 | /** | |
393 | * calc_dd_growth - calculate approximate amount of data which makes other data | |
394 | * dirty from budgeting request. | |
395 | * @c: UBIFS file-system description object | |
396 | * @req: budgeting request | |
397 | */ | |
398 | static int calc_dd_growth(const struct ubifs_info *c, | |
399 | const struct ubifs_budget_req *req) | |
400 | { | |
401 | int dd_growth; | |
402 | ||
b137545c | 403 | dd_growth = req->dirtied_page ? c->bi.page_budget : 0; |
1e51764a AB |
404 | |
405 | if (req->dirtied_ino) | |
b137545c | 406 | dd_growth += c->bi.inode_budget << (req->dirtied_ino - 1); |
1e51764a | 407 | if (req->mod_dent) |
b137545c | 408 | dd_growth += c->bi.dent_budget; |
1e51764a AB |
409 | dd_growth += req->dirtied_ino_d; |
410 | return dd_growth; | |
411 | } | |
412 | ||
413 | /** | |
414 | * ubifs_budget_space - ensure there is enough space to complete an operation. | |
415 | * @c: UBIFS file-system description object | |
416 | * @req: budget request | |
417 | * | |
418 | * This function allocates budget for an operation. It uses pessimistic | |
419 | * approximation of how much flash space the operation needs. The goal of this | |
420 | * function is to make sure UBIFS always has flash space to flush all dirty | |
421 | * pages, dirty inodes, and dirty znodes (liability). This function may force | |
422 | * commit, garbage-collection or write-back. Returns zero in case of success, | |
423 | * %-ENOSPC if there is no free space and other negative error codes in case of | |
424 | * failures. | |
425 | */ | |
426 | int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req) | |
427 | { | |
2acf8067 | 428 | int err, idx_growth, data_growth, dd_growth, retried = 0; |
1e51764a | 429 | |
6eb61d58 RW |
430 | ubifs_assert(c, req->new_page <= 1); |
431 | ubifs_assert(c, req->dirtied_page <= 1); | |
432 | ubifs_assert(c, req->new_dent <= 1); | |
433 | ubifs_assert(c, req->mod_dent <= 1); | |
434 | ubifs_assert(c, req->new_ino <= 1); | |
435 | ubifs_assert(c, req->new_ino_d <= UBIFS_MAX_INO_DATA); | |
436 | ubifs_assert(c, req->dirtied_ino <= 4); | |
437 | ubifs_assert(c, req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4); | |
438 | ubifs_assert(c, !(req->new_ino_d & 7)); | |
439 | ubifs_assert(c, !(req->dirtied_ino_d & 7)); | |
1e51764a AB |
440 | |
441 | data_growth = calc_data_growth(c, req); | |
442 | dd_growth = calc_dd_growth(c, req); | |
443 | if (!data_growth && !dd_growth) | |
444 | return 0; | |
445 | idx_growth = calc_idx_growth(c, req); | |
1e51764a AB |
446 | |
447 | again: | |
448 | spin_lock(&c->space_lock); | |
6eb61d58 RW |
449 | ubifs_assert(c, c->bi.idx_growth >= 0); |
450 | ubifs_assert(c, c->bi.data_growth >= 0); | |
451 | ubifs_assert(c, c->bi.dd_growth >= 0); | |
1e51764a | 452 | |
b137545c | 453 | if (unlikely(c->bi.nospace) && (c->bi.nospace_rp || !can_use_rp(c))) { |
1e51764a AB |
454 | dbg_budg("no space"); |
455 | spin_unlock(&c->space_lock); | |
456 | return -ENOSPC; | |
457 | } | |
458 | ||
b137545c AB |
459 | c->bi.idx_growth += idx_growth; |
460 | c->bi.data_growth += data_growth; | |
461 | c->bi.dd_growth += dd_growth; | |
1e51764a AB |
462 | |
463 | err = do_budget_space(c); | |
464 | if (likely(!err)) { | |
465 | req->idx_growth = idx_growth; | |
466 | req->data_growth = data_growth; | |
467 | req->dd_growth = dd_growth; | |
468 | spin_unlock(&c->space_lock); | |
469 | return 0; | |
470 | } | |
471 | ||
472 | /* Restore the old values */ | |
b137545c AB |
473 | c->bi.idx_growth -= idx_growth; |
474 | c->bi.data_growth -= data_growth; | |
475 | c->bi.dd_growth -= dd_growth; | |
1e51764a AB |
476 | spin_unlock(&c->space_lock); |
477 | ||
478 | if (req->fast) { | |
479 | dbg_budg("no space for fast budgeting"); | |
480 | return err; | |
481 | } | |
482 | ||
2acf8067 AB |
483 | err = make_free_space(c); |
484 | cond_resched(); | |
1e51764a AB |
485 | if (err == -EAGAIN) { |
486 | dbg_budg("try again"); | |
1e51764a AB |
487 | goto again; |
488 | } else if (err == -ENOSPC) { | |
2acf8067 AB |
489 | if (!retried) { |
490 | retried = 1; | |
491 | dbg_budg("-ENOSPC, but anyway try once again"); | |
492 | goto again; | |
493 | } | |
1e51764a | 494 | dbg_budg("FS is full, -ENOSPC"); |
b137545c | 495 | c->bi.nospace = 1; |
1e51764a | 496 | if (can_use_rp(c) || c->rp_size == 0) |
b137545c | 497 | c->bi.nospace_rp = 1; |
1e51764a AB |
498 | smp_wmb(); |
499 | } else | |
235c362b | 500 | ubifs_err(c, "cannot budget space, error %d", err); |
1e51764a AB |
501 | return err; |
502 | } | |
503 | ||
504 | /** | |
505 | * ubifs_release_budget - release budgeted free space. | |
506 | * @c: UBIFS file-system description object | |
507 | * @req: budget request | |
508 | * | |
509 | * This function releases the space budgeted by 'ubifs_budget_space()'. Note, | |
510 | * since the index changes (which were budgeted for in @req->idx_growth) will | |
511 | * only be written to the media on commit, this function moves the index budget | |
b137545c AB |
512 | * from @c->bi.idx_growth to @c->bi.uncommitted_idx. The latter will be zeroed |
513 | * by the commit operation. | |
1e51764a AB |
514 | */ |
515 | void ubifs_release_budget(struct ubifs_info *c, struct ubifs_budget_req *req) | |
516 | { | |
6eb61d58 RW |
517 | ubifs_assert(c, req->new_page <= 1); |
518 | ubifs_assert(c, req->dirtied_page <= 1); | |
519 | ubifs_assert(c, req->new_dent <= 1); | |
520 | ubifs_assert(c, req->mod_dent <= 1); | |
521 | ubifs_assert(c, req->new_ino <= 1); | |
522 | ubifs_assert(c, req->new_ino_d <= UBIFS_MAX_INO_DATA); | |
523 | ubifs_assert(c, req->dirtied_ino <= 4); | |
524 | ubifs_assert(c, req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4); | |
525 | ubifs_assert(c, !(req->new_ino_d & 7)); | |
526 | ubifs_assert(c, !(req->dirtied_ino_d & 7)); | |
1e51764a | 527 | if (!req->recalculate) { |
6eb61d58 RW |
528 | ubifs_assert(c, req->idx_growth >= 0); |
529 | ubifs_assert(c, req->data_growth >= 0); | |
530 | ubifs_assert(c, req->dd_growth >= 0); | |
1e51764a AB |
531 | } |
532 | ||
533 | if (req->recalculate) { | |
534 | req->data_growth = calc_data_growth(c, req); | |
535 | req->dd_growth = calc_dd_growth(c, req); | |
536 | req->idx_growth = calc_idx_growth(c, req); | |
537 | } | |
538 | ||
539 | if (!req->data_growth && !req->dd_growth) | |
540 | return; | |
541 | ||
b137545c | 542 | c->bi.nospace = c->bi.nospace_rp = 0; |
1e51764a AB |
543 | smp_wmb(); |
544 | ||
545 | spin_lock(&c->space_lock); | |
b137545c AB |
546 | c->bi.idx_growth -= req->idx_growth; |
547 | c->bi.uncommitted_idx += req->idx_growth; | |
548 | c->bi.data_growth -= req->data_growth; | |
549 | c->bi.dd_growth -= req->dd_growth; | |
550 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
551 | ||
6eb61d58 RW |
552 | ubifs_assert(c, c->bi.idx_growth >= 0); |
553 | ubifs_assert(c, c->bi.data_growth >= 0); | |
554 | ubifs_assert(c, c->bi.dd_growth >= 0); | |
555 | ubifs_assert(c, c->bi.min_idx_lebs < c->main_lebs); | |
556 | ubifs_assert(c, !(c->bi.idx_growth & 7)); | |
557 | ubifs_assert(c, !(c->bi.data_growth & 7)); | |
558 | ubifs_assert(c, !(c->bi.dd_growth & 7)); | |
1e51764a AB |
559 | spin_unlock(&c->space_lock); |
560 | } | |
561 | ||
562 | /** | |
563 | * ubifs_convert_page_budget - convert budget of a new page. | |
564 | * @c: UBIFS file-system description object | |
565 | * | |
566 | * This function converts budget which was allocated for a new page of data to | |
025dfdaf | 567 | * the budget of changing an existing page of data. The latter is smaller than |
1e51764a AB |
568 | * the former, so this function only does simple re-calculation and does not |
569 | * involve any write-back. | |
570 | */ | |
571 | void ubifs_convert_page_budget(struct ubifs_info *c) | |
572 | { | |
573 | spin_lock(&c->space_lock); | |
574 | /* Release the index growth reservation */ | |
b137545c | 575 | c->bi.idx_growth -= c->max_idx_node_sz << UBIFS_BLOCKS_PER_PAGE_SHIFT; |
1e51764a | 576 | /* Release the data growth reservation */ |
b137545c | 577 | c->bi.data_growth -= c->bi.page_budget; |
1e51764a | 578 | /* Increase the dirty data growth reservation instead */ |
b137545c | 579 | c->bi.dd_growth += c->bi.page_budget; |
1e51764a | 580 | /* And re-calculate the indexing space reservation */ |
b137545c | 581 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); |
1e51764a AB |
582 | spin_unlock(&c->space_lock); |
583 | } | |
584 | ||
585 | /** | |
586 | * ubifs_release_dirty_inode_budget - release dirty inode budget. | |
587 | * @c: UBIFS file-system description object | |
588 | * @ui: UBIFS inode to release the budget for | |
589 | * | |
590 | * This function releases budget corresponding to a dirty inode. It is usually | |
591 | * called when after the inode has been written to the media and marked as | |
6d6cb0d6 | 592 | * clean. It also causes the "no space" flags to be cleared. |
1e51764a AB |
593 | */ |
594 | void ubifs_release_dirty_inode_budget(struct ubifs_info *c, | |
595 | struct ubifs_inode *ui) | |
596 | { | |
182854b4 | 597 | struct ubifs_budget_req req; |
1e51764a | 598 | |
182854b4 | 599 | memset(&req, 0, sizeof(struct ubifs_budget_req)); |
6d6cb0d6 | 600 | /* The "no space" flags will be cleared because dd_growth is > 0 */ |
b137545c | 601 | req.dd_growth = c->bi.inode_budget + ALIGN(ui->data_len, 8); |
1e51764a AB |
602 | ubifs_release_budget(c, &req); |
603 | } | |
604 | ||
4b5f2762 AB |
605 | /** |
606 | * ubifs_reported_space - calculate reported free space. | |
607 | * @c: the UBIFS file-system description object | |
608 | * @free: amount of free space | |
609 | * | |
610 | * This function calculates amount of free space which will be reported to | |
611 | * user-space. User-space application tend to expect that if the file-system | |
612 | * (e.g., via the 'statfs()' call) reports that it has N bytes available, they | |
613 | * are able to write a file of size N. UBIFS attaches node headers to each data | |
80736d41 AB |
614 | * node and it has to write indexing nodes as well. This introduces additional |
615 | * overhead, and UBIFS has to report slightly less free space to meet the above | |
616 | * expectations. | |
4b5f2762 AB |
617 | * |
618 | * This function assumes free space is made up of uncompressed data nodes and | |
619 | * full index nodes (one per data node, tripled because we always allow enough | |
620 | * space to write the index thrice). | |
621 | * | |
622 | * Note, the calculation is pessimistic, which means that most of the time | |
623 | * UBIFS reports less space than it actually has. | |
624 | */ | |
4d61db4f | 625 | long long ubifs_reported_space(const struct ubifs_info *c, long long free) |
4b5f2762 | 626 | { |
f171d4d7 | 627 | int divisor, factor, f; |
4b5f2762 AB |
628 | |
629 | /* | |
630 | * Reported space size is @free * X, where X is UBIFS block size | |
631 | * divided by UBIFS block size + all overhead one data block | |
632 | * introduces. The overhead is the node header + indexing overhead. | |
633 | * | |
f171d4d7 AB |
634 | * Indexing overhead calculations are based on the following formula: |
635 | * I = N/(f - 1) + 1, where I - number of indexing nodes, N - number | |
636 | * of data nodes, f - fanout. Because effective UBIFS fanout is twice | |
637 | * as less than maximum fanout, we assume that each data node | |
4b5f2762 | 638 | * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes. |
80736d41 | 639 | * Note, the multiplier 3 is because UBIFS reserves thrice as more space |
4b5f2762 AB |
640 | * for the index. |
641 | */ | |
f171d4d7 | 642 | f = c->fanout > 3 ? c->fanout >> 1 : 2; |
4b5f2762 AB |
643 | factor = UBIFS_BLOCK_SIZE; |
644 | divisor = UBIFS_MAX_DATA_NODE_SZ; | |
f171d4d7 | 645 | divisor += (c->max_idx_node_sz * 3) / (f - 1); |
4b5f2762 | 646 | free *= factor; |
4d61db4f | 647 | return div_u64(free, divisor); |
4b5f2762 AB |
648 | } |
649 | ||
1e51764a | 650 | /** |
84abf972 | 651 | * ubifs_get_free_space_nolock - return amount of free space. |
1e51764a AB |
652 | * @c: UBIFS file-system description object |
653 | * | |
7dad181b AB |
654 | * This function calculates amount of free space to report to user-space. |
655 | * | |
656 | * Because UBIFS may introduce substantial overhead (the index, node headers, | |
fb1cd01a AB |
657 | * alignment, wastage at the end of LEBs, etc), it cannot report real amount of |
658 | * free flash space it has (well, because not all dirty space is reclaimable, | |
659 | * UBIFS does not actually know the real amount). If UBIFS did so, it would | |
660 | * bread user expectations about what free space is. Users seem to accustomed | |
661 | * to assume that if the file-system reports N bytes of free space, they would | |
662 | * be able to fit a file of N bytes to the FS. This almost works for | |
7dad181b AB |
663 | * traditional file-systems, because they have way less overhead than UBIFS. |
664 | * So, to keep users happy, UBIFS tries to take the overhead into account. | |
1e51764a | 665 | */ |
84abf972 | 666 | long long ubifs_get_free_space_nolock(struct ubifs_info *c) |
1e51764a | 667 | { |
84abf972 | 668 | int rsvd_idx_lebs, lebs; |
1e51764a AB |
669 | long long available, outstanding, free; |
670 | ||
6eb61d58 | 671 | ubifs_assert(c, c->bi.min_idx_lebs == ubifs_calc_min_idx_lebs(c)); |
b137545c AB |
672 | outstanding = c->bi.data_growth + c->bi.dd_growth; |
673 | available = ubifs_calc_available(c, c->bi.min_idx_lebs); | |
7dad181b AB |
674 | |
675 | /* | |
676 | * When reporting free space to user-space, UBIFS guarantees that it is | |
677 | * possible to write a file of free space size. This means that for | |
678 | * empty LEBs we may use more precise calculations than | |
679 | * 'ubifs_calc_available()' is using. Namely, we know that in empty | |
680 | * LEBs we would waste only @c->leb_overhead bytes, not @c->dark_wm. | |
681 | * Thus, amend the available space. | |
682 | * | |
683 | * Note, the calculations below are similar to what we have in | |
684 | * 'do_budget_space()', so refer there for comments. | |
685 | */ | |
b137545c AB |
686 | if (c->bi.min_idx_lebs > c->lst.idx_lebs) |
687 | rsvd_idx_lebs = c->bi.min_idx_lebs - c->lst.idx_lebs; | |
7dad181b AB |
688 | else |
689 | rsvd_idx_lebs = 0; | |
690 | lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt - | |
691 | c->lst.taken_empty_lebs; | |
692 | lebs -= rsvd_idx_lebs; | |
693 | available += lebs * (c->dark_wm - c->leb_overhead); | |
1e51764a AB |
694 | |
695 | if (available > outstanding) | |
696 | free = ubifs_reported_space(c, available - outstanding); | |
697 | else | |
698 | free = 0; | |
699 | return free; | |
700 | } | |
84abf972 AB |
701 | |
702 | /** | |
703 | * ubifs_get_free_space - return amount of free space. | |
704 | * @c: UBIFS file-system description object | |
705 | * | |
055da1b7 | 706 | * This function calculates and returns amount of free space to report to |
84abf972 AB |
707 | * user-space. |
708 | */ | |
709 | long long ubifs_get_free_space(struct ubifs_info *c) | |
710 | { | |
711 | long long free; | |
712 | ||
713 | spin_lock(&c->space_lock); | |
714 | free = ubifs_get_free_space_nolock(c); | |
715 | spin_unlock(&c->space_lock); | |
716 | ||
717 | return free; | |
718 | } |