blob: d0231ba783dfaaa4a881e7bdfd9eb91abaf7b3c2 [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
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>
Artem Bityutskiy4d61db42008-12-18 14:06:51 +020035#include <linux/math64.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030036
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,
Artem Bityutskiy2acf8062008-12-09 11:04:40 -050040 * or committing. The below constant defines maximum number of times UBIFS
Artem Bityutskiy1e517642008-07-14 19:08:37 +030041 * repeats the operations.
42 */
Artem Bityutskiy2acf8062008-12-09 11:04:40 -050043#define MAX_MKSPC_RETRIES 3
Artem Bityutskiy1e517642008-07-14 19:08:37 +030044
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
51/**
Artem Bityutskiy1e517642008-07-14 19:08:37 +030052 * 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
57 * of dirty inodes and their pages. Returns the amount of pages which were
58 * written back. The returned value does not include dirty inodes which were
59 * synchronized.
60 *
61 * Note, this function synchronizes even VFS inodes which are locked
62 * (@i_mutex) by the caller of the budgeting function, because write-back does
63 * not touch @i_mutex.
64 */
65static int shrink_liability(struct ubifs_info *c, int nr_to_write)
66{
67 int nr_written;
68 struct writeback_control wbc = {
69 .sync_mode = WB_SYNC_NONE,
70 .range_end = LLONG_MAX,
71 .nr_to_write = nr_to_write,
72 };
73
74 generic_sync_sb_inodes(c->vfs_sb, &wbc);
75 nr_written = nr_to_write - wbc.nr_to_write;
76
77 if (!nr_written) {
78 /*
79 * Re-try again but wait on pages/inodes which are being
80 * written-back concurrently (e.g., by pdflush).
81 */
82 memset(&wbc, 0, sizeof(struct writeback_control));
83 wbc.sync_mode = WB_SYNC_ALL;
84 wbc.range_end = LLONG_MAX;
85 wbc.nr_to_write = nr_to_write;
86 generic_sync_sb_inodes(c->vfs_sb, &wbc);
87 nr_written = nr_to_write - wbc.nr_to_write;
88 }
89
90 dbg_budg("%d pages were written back", nr_written);
91 return nr_written;
92}
93
94
95/**
96 * run_gc - run garbage collector.
97 * @c: UBIFS file-system description object
98 *
99 * This function runs garbage collector to make some more free space. Returns
100 * zero if a free LEB has been produced, %-EAGAIN if commit is required, and a
101 * negative error code in case of failure.
102 */
103static int run_gc(struct ubifs_info *c)
104{
105 int err, lnum;
106
107 /* Make some free space by garbage-collecting dirty space */
108 down_read(&c->commit_sem);
109 lnum = ubifs_garbage_collect(c, 1);
110 up_read(&c->commit_sem);
111 if (lnum < 0)
112 return lnum;
113
114 /* GC freed one LEB, return it to lprops */
115 dbg_budg("GC freed LEB %d", lnum);
116 err = ubifs_return_leb(c, lnum);
117 if (err)
118 return err;
119 return 0;
120}
121
122/**
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500123 * get_liability - calculate current liability.
124 * @c: UBIFS file-system description object
125 *
126 * This function calculates and returns current UBIFS liability, i.e. the
127 * amount of bytes UBIFS has "promised" to write to the media.
128 */
129static long long get_liability(struct ubifs_info *c)
130{
131 long long liab;
132
133 spin_lock(&c->space_lock);
134 liab = c->budg_idx_growth + c->budg_data_growth + c->budg_dd_growth;
135 spin_unlock(&c->space_lock);
136 return liab;
137}
138
139/**
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300140 * make_free_space - make more free space on the file-system.
141 * @c: UBIFS file-system description object
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300142 *
143 * This function is called when an operation cannot be budgeted because there
144 * is supposedly no free space. But in most cases there is some free space:
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200145 * o budgeting is pessimistic, so it always budgets more than it is actually
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300146 * needed, so shrinking the liability is one way to make free space - the
147 * cached data will take less space then it was budgeted for;
148 * o GC may turn some dark space into free space (budgeting treats dark space
149 * as not available);
150 * o commit may free some LEB, i.e., turn freeable LEBs into free LEBs.
151 *
152 * So this function tries to do the above. Returns %-EAGAIN if some free space
153 * was presumably made and the caller has to re-try budgeting the operation.
154 * Returns %-ENOSPC if it couldn't do more free space, and other negative error
155 * codes on failures.
156 */
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500157static int make_free_space(struct ubifs_info *c)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300158{
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500159 int err, retries = 0;
160 long long liab1, liab2;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300161
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500162 do {
163 liab1 = get_liability(c);
164 /*
165 * We probably have some dirty pages or inodes (liability), try
166 * to write them back.
167 */
168 dbg_budg("liability %lld, run write-back", liab1);
169 shrink_liability(c, NR_TO_WRITE);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300170
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500171 liab2 = get_liability(c);
172 if (liab2 < liab1)
173 return -EAGAIN;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300174
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500175 dbg_budg("new liability %lld (not shrinked)", liab2);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300176
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500177 /* Liability did not shrink again, try GC */
178 dbg_budg("Run GC");
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300179 err = run_gc(c);
180 if (!err)
181 return -EAGAIN;
182
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500183 if (err != -EAGAIN && err != -ENOSPC)
184 /* Some real error happened */
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300185 return err;
186
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500187 dbg_budg("Run commit (retries %d)", retries);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300188 err = ubifs_run_commit(c);
189 if (err)
190 return err;
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500191 } while (retries++ < MAX_MKSPC_RETRIES);
192
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300193 return -ENOSPC;
194}
195
196/**
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200197 * ubifs_calc_min_idx_lebs - calculate amount of LEBs for the index.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300198 * @c: UBIFS file-system description object
199 *
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200200 * This function calculates and returns the number of LEBs which should be kept
201 * for index usage.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300202 */
203int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
204{
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200205 int idx_lebs;
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200206 long long idx_size;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300207
208 idx_size = c->old_idx_sz + c->budg_idx_growth + c->budg_uncommitted_idx;
Adrian Hunter3a132522008-07-30 12:18:02 +0300209 /* And make sure we have thrice the index size of space reserved */
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200210 idx_size += idx_size << 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300211 /*
212 * We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes'
213 * pair, nor similarly the two variables for the new index size, so we
214 * have to do this costly 64-bit division on fast-path.
215 */
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200216 idx_lebs = div_u64(idx_size + c->idx_leb_size - 1, c->idx_leb_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300217 /*
218 * The index head is not available for the in-the-gaps method, so add an
219 * extra LEB to compensate.
220 */
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200221 idx_lebs += 1;
222 if (idx_lebs < MIN_INDEX_LEBS)
223 idx_lebs = MIN_INDEX_LEBS;
224 return idx_lebs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300225}
226
227/**
228 * ubifs_calc_available - calculate available FS space.
229 * @c: UBIFS file-system description object
230 * @min_idx_lebs: minimum number of LEBs reserved for the index
231 *
232 * This function calculates and returns amount of FS space available for use.
233 */
234long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs)
235{
236 int subtract_lebs;
237 long long available;
238
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300239 available = c->main_bytes - c->lst.total_used;
240
241 /*
242 * Now 'available' contains theoretically available flash space
243 * assuming there is no index, so we have to subtract the space which
244 * is reserved for the index.
245 */
246 subtract_lebs = min_idx_lebs;
247
248 /* Take into account that GC reserves one LEB for its own needs */
249 subtract_lebs += 1;
250
251 /*
252 * The GC journal head LEB is not really accessible. And since
253 * different write types go to different heads, we may count only on
254 * one head's space.
255 */
256 subtract_lebs += c->jhead_cnt - 1;
257
258 /* We also reserve one LEB for deletions, which bypass budgeting */
259 subtract_lebs += 1;
260
261 available -= (long long)subtract_lebs * c->leb_size;
262
263 /* Subtract the dead space which is not available for use */
264 available -= c->lst.total_dead;
265
266 /*
267 * Subtract dark space, which might or might not be usable - it depends
268 * on the data which we have on the media and which will be written. If
269 * this is a lot of uncompressed or not-compressible data, the dark
270 * space cannot be used.
271 */
272 available -= c->lst.total_dark;
273
274 /*
275 * However, there is more dark space. The index may be bigger than
276 * @min_idx_lebs. Those extra LEBs are assumed to be available, but
277 * their dark space is not included in total_dark, so it is subtracted
278 * here.
279 */
280 if (c->lst.idx_lebs > min_idx_lebs) {
281 subtract_lebs = c->lst.idx_lebs - min_idx_lebs;
282 available -= subtract_lebs * c->dark_wm;
283 }
284
285 /* The calculations are rough and may end up with a negative number */
286 return available > 0 ? available : 0;
287}
288
289/**
290 * can_use_rp - check whether the user is allowed to use reserved pool.
291 * @c: UBIFS file-system description object
292 *
293 * UBIFS has so-called "reserved pool" which is flash space reserved
294 * for the superuser and for uses whose UID/GID is recorded in UBIFS superblock.
295 * This function checks whether current user is allowed to use reserved pool.
296 * Returns %1 current user is allowed to use reserved pool and %0 otherwise.
297 */
298static int can_use_rp(struct ubifs_info *c)
299{
David Howells26bf1942008-11-14 10:39:03 +1100300 if (current_fsuid() == c->rp_uid || capable(CAP_SYS_RESOURCE) ||
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300301 (c->rp_gid != 0 && in_group_p(c->rp_gid)))
302 return 1;
303 return 0;
304}
305
306/**
307 * do_budget_space - reserve flash space for index and data growth.
308 * @c: UBIFS file-system description object
309 *
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200310 * This function makes sure UBIFS has enough free LEBs for index growth and
311 * data.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300312 *
Adrian Hunter3a132522008-07-30 12:18:02 +0300313 * When budgeting index space, UBIFS reserves thrice as many LEBs as the index
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300314 * would take if it was consolidated and written to the flash. This guarantees
315 * that the "in-the-gaps" commit method always succeeds and UBIFS will always
316 * be able to commit dirty index. So this function basically adds amount of
Artem Bityutskiyb364b412008-07-25 14:38:51 +0300317 * budgeted index space to the size of the current index, multiplies this by 3,
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200318 * and makes sure this does not exceed the amount of free LEBs.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300319 *
320 * Notes about @c->min_idx_lebs and @c->lst.idx_lebs variables:
321 * o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might
322 * be large, because UBIFS does not do any index consolidation as long as
323 * there is free space. IOW, the index may take a lot of LEBs, but the LEBs
324 * will contain a lot of dirt.
Artem Bityutskiy7d4e9cc2009-03-20 19:11:12 +0200325 * o @c->min_idx_lebs is the number of LEBS the index presumably takes. IOW,
326 * the index may be consolidated to take up to @c->min_idx_lebs LEBs.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300327 *
328 * This function returns zero in case of success, and %-ENOSPC in case of
329 * failure.
330 */
331static int do_budget_space(struct ubifs_info *c)
332{
333 long long outstanding, available;
334 int lebs, rsvd_idx_lebs, min_idx_lebs;
335
336 /* First budget index space */
337 min_idx_lebs = ubifs_calc_min_idx_lebs(c);
338
339 /* Now 'min_idx_lebs' contains number of LEBs to reserve */
340 if (min_idx_lebs > c->lst.idx_lebs)
341 rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs;
342 else
343 rsvd_idx_lebs = 0;
344
345 /*
346 * The number of LEBs that are available to be used by the index is:
347 *
348 * @c->lst.empty_lebs + @c->freeable_cnt + @c->idx_gc_cnt -
349 * @c->lst.taken_empty_lebs
350 *
Artem Bityutskiy948cfb22008-08-20 11:56:33 +0300351 * @c->lst.empty_lebs are available because they are empty.
352 * @c->freeable_cnt are available because they contain only free and
353 * dirty space, @c->idx_gc_cnt are available because they are index
354 * LEBs that have been garbage collected and are awaiting the commit
355 * before they can be used. And the in-the-gaps method will grab these
356 * if it needs them. @c->lst.taken_empty_lebs are empty LEBs that have
357 * already been allocated for some purpose.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300358 *
Artem Bityutskiy948cfb22008-08-20 11:56:33 +0300359 * Note, @c->idx_gc_cnt is included to both @c->lst.empty_lebs (because
360 * these LEBs are empty) and to @c->lst.taken_empty_lebs (because they
361 * are taken until after the commit).
362 *
363 * Note, @c->lst.taken_empty_lebs may temporarily be higher by one
364 * because of the way we serialize LEB allocations and budgeting. See a
365 * comment in 'ubifs_find_free_space()'.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300366 */
367 lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt -
368 c->lst.taken_empty_lebs;
369 if (unlikely(rsvd_idx_lebs > lebs)) {
370 dbg_budg("out of indexing space: min_idx_lebs %d (old %d), "
371 "rsvd_idx_lebs %d", min_idx_lebs, c->min_idx_lebs,
372 rsvd_idx_lebs);
373 return -ENOSPC;
374 }
375
376 available = ubifs_calc_available(c, min_idx_lebs);
377 outstanding = c->budg_data_growth + c->budg_dd_growth;
378
379 if (unlikely(available < outstanding)) {
380 dbg_budg("out of data space: available %lld, outstanding %lld",
381 available, outstanding);
382 return -ENOSPC;
383 }
384
385 if (available - outstanding <= c->rp_size && !can_use_rp(c))
386 return -ENOSPC;
387
388 c->min_idx_lebs = min_idx_lebs;
389 return 0;
390}
391
392/**
393 * calc_idx_growth - calculate approximate index growth from budgeting request.
394 * @c: UBIFS file-system description object
395 * @req: budgeting request
396 *
397 * For now we assume each new node adds one znode. But this is rather poor
398 * approximation, though.
399 */
400static int calc_idx_growth(const struct ubifs_info *c,
401 const struct ubifs_budget_req *req)
402{
403 int znodes;
404
405 znodes = req->new_ino + (req->new_page << UBIFS_BLOCKS_PER_PAGE_SHIFT) +
406 req->new_dent;
407 return znodes * c->max_idx_node_sz;
408}
409
410/**
411 * calc_data_growth - calculate approximate amount of new data from budgeting
412 * request.
413 * @c: UBIFS file-system description object
414 * @req: budgeting request
415 */
416static int calc_data_growth(const struct ubifs_info *c,
417 const struct ubifs_budget_req *req)
418{
419 int data_growth;
420
421 data_growth = req->new_ino ? c->inode_budget : 0;
422 if (req->new_page)
423 data_growth += c->page_budget;
424 if (req->new_dent)
425 data_growth += c->dent_budget;
426 data_growth += req->new_ino_d;
427 return data_growth;
428}
429
430/**
431 * calc_dd_growth - calculate approximate amount of data which makes other data
432 * dirty from budgeting request.
433 * @c: UBIFS file-system description object
434 * @req: budgeting request
435 */
436static int calc_dd_growth(const struct ubifs_info *c,
437 const struct ubifs_budget_req *req)
438{
439 int dd_growth;
440
441 dd_growth = req->dirtied_page ? c->page_budget : 0;
442
443 if (req->dirtied_ino)
444 dd_growth += c->inode_budget << (req->dirtied_ino - 1);
445 if (req->mod_dent)
446 dd_growth += c->dent_budget;
447 dd_growth += req->dirtied_ino_d;
448 return dd_growth;
449}
450
451/**
452 * ubifs_budget_space - ensure there is enough space to complete an operation.
453 * @c: UBIFS file-system description object
454 * @req: budget request
455 *
456 * This function allocates budget for an operation. It uses pessimistic
457 * approximation of how much flash space the operation needs. The goal of this
458 * function is to make sure UBIFS always has flash space to flush all dirty
459 * pages, dirty inodes, and dirty znodes (liability). This function may force
460 * commit, garbage-collection or write-back. Returns zero in case of success,
461 * %-ENOSPC if there is no free space and other negative error codes in case of
462 * failures.
463 */
464int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req)
465{
466 int uninitialized_var(cmt_retries), uninitialized_var(wb_retries);
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500467 int err, idx_growth, data_growth, dd_growth, retried = 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300468
Artem Bityutskiy547000d2008-07-24 14:42:05 +0300469 ubifs_assert(req->new_page <= 1);
470 ubifs_assert(req->dirtied_page <= 1);
471 ubifs_assert(req->new_dent <= 1);
472 ubifs_assert(req->mod_dent <= 1);
473 ubifs_assert(req->new_ino <= 1);
474 ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300475 ubifs_assert(req->dirtied_ino <= 4);
476 ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300477 ubifs_assert(!(req->new_ino_d & 7));
478 ubifs_assert(!(req->dirtied_ino_d & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300479
480 data_growth = calc_data_growth(c, req);
481 dd_growth = calc_dd_growth(c, req);
482 if (!data_growth && !dd_growth)
483 return 0;
484 idx_growth = calc_idx_growth(c, req);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300485
486again:
487 spin_lock(&c->space_lock);
488 ubifs_assert(c->budg_idx_growth >= 0);
489 ubifs_assert(c->budg_data_growth >= 0);
490 ubifs_assert(c->budg_dd_growth >= 0);
491
492 if (unlikely(c->nospace) && (c->nospace_rp || !can_use_rp(c))) {
493 dbg_budg("no space");
494 spin_unlock(&c->space_lock);
495 return -ENOSPC;
496 }
497
498 c->budg_idx_growth += idx_growth;
499 c->budg_data_growth += data_growth;
500 c->budg_dd_growth += dd_growth;
501
502 err = do_budget_space(c);
503 if (likely(!err)) {
504 req->idx_growth = idx_growth;
505 req->data_growth = data_growth;
506 req->dd_growth = dd_growth;
507 spin_unlock(&c->space_lock);
508 return 0;
509 }
510
511 /* Restore the old values */
512 c->budg_idx_growth -= idx_growth;
513 c->budg_data_growth -= data_growth;
514 c->budg_dd_growth -= dd_growth;
515 spin_unlock(&c->space_lock);
516
517 if (req->fast) {
518 dbg_budg("no space for fast budgeting");
519 return err;
520 }
521
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500522 err = make_free_space(c);
523 cond_resched();
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300524 if (err == -EAGAIN) {
525 dbg_budg("try again");
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300526 goto again;
527 } else if (err == -ENOSPC) {
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500528 if (!retried) {
529 retried = 1;
530 dbg_budg("-ENOSPC, but anyway try once again");
531 goto again;
532 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300533 dbg_budg("FS is full, -ENOSPC");
534 c->nospace = 1;
535 if (can_use_rp(c) || c->rp_size == 0)
536 c->nospace_rp = 1;
537 smp_wmb();
538 } else
539 ubifs_err("cannot budget space, error %d", err);
540 return err;
541}
542
543/**
544 * ubifs_release_budget - release budgeted free space.
545 * @c: UBIFS file-system description object
546 * @req: budget request
547 *
548 * This function releases the space budgeted by 'ubifs_budget_space()'. Note,
549 * since the index changes (which were budgeted for in @req->idx_growth) will
550 * only be written to the media on commit, this function moves the index budget
551 * from @c->budg_idx_growth to @c->budg_uncommitted_idx. The latter will be
552 * zeroed by the commit operation.
553 */
554void ubifs_release_budget(struct ubifs_info *c, struct ubifs_budget_req *req)
555{
Artem Bityutskiy547000d2008-07-24 14:42:05 +0300556 ubifs_assert(req->new_page <= 1);
557 ubifs_assert(req->dirtied_page <= 1);
558 ubifs_assert(req->new_dent <= 1);
559 ubifs_assert(req->mod_dent <= 1);
560 ubifs_assert(req->new_ino <= 1);
561 ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300562 ubifs_assert(req->dirtied_ino <= 4);
563 ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300564 ubifs_assert(!(req->new_ino_d & 7));
565 ubifs_assert(!(req->dirtied_ino_d & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300566 if (!req->recalculate) {
567 ubifs_assert(req->idx_growth >= 0);
568 ubifs_assert(req->data_growth >= 0);
569 ubifs_assert(req->dd_growth >= 0);
570 }
571
572 if (req->recalculate) {
573 req->data_growth = calc_data_growth(c, req);
574 req->dd_growth = calc_dd_growth(c, req);
575 req->idx_growth = calc_idx_growth(c, req);
576 }
577
578 if (!req->data_growth && !req->dd_growth)
579 return;
580
581 c->nospace = c->nospace_rp = 0;
582 smp_wmb();
583
584 spin_lock(&c->space_lock);
585 c->budg_idx_growth -= req->idx_growth;
586 c->budg_uncommitted_idx += req->idx_growth;
587 c->budg_data_growth -= req->data_growth;
588 c->budg_dd_growth -= req->dd_growth;
589 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
590
591 ubifs_assert(c->budg_idx_growth >= 0);
592 ubifs_assert(c->budg_data_growth >= 0);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300593 ubifs_assert(c->budg_dd_growth >= 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300594 ubifs_assert(c->min_idx_lebs < c->main_lebs);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300595 ubifs_assert(!(c->budg_idx_growth & 7));
596 ubifs_assert(!(c->budg_data_growth & 7));
597 ubifs_assert(!(c->budg_dd_growth & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300598 spin_unlock(&c->space_lock);
599}
600
601/**
602 * ubifs_convert_page_budget - convert budget of a new page.
603 * @c: UBIFS file-system description object
604 *
605 * This function converts budget which was allocated for a new page of data to
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200606 * the budget of changing an existing page of data. The latter is smaller than
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300607 * the former, so this function only does simple re-calculation and does not
608 * involve any write-back.
609 */
610void ubifs_convert_page_budget(struct ubifs_info *c)
611{
612 spin_lock(&c->space_lock);
613 /* Release the index growth reservation */
614 c->budg_idx_growth -= c->max_idx_node_sz << UBIFS_BLOCKS_PER_PAGE_SHIFT;
615 /* Release the data growth reservation */
616 c->budg_data_growth -= c->page_budget;
617 /* Increase the dirty data growth reservation instead */
618 c->budg_dd_growth += c->page_budget;
619 /* And re-calculate the indexing space reservation */
620 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
621 spin_unlock(&c->space_lock);
622}
623
624/**
625 * ubifs_release_dirty_inode_budget - release dirty inode budget.
626 * @c: UBIFS file-system description object
627 * @ui: UBIFS inode to release the budget for
628 *
629 * This function releases budget corresponding to a dirty inode. It is usually
630 * called when after the inode has been written to the media and marked as
Adrian Hunter6d6cb0d2009-04-08 14:07:57 +0200631 * clean. It also causes the "no space" flags to be cleared.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300632 */
633void ubifs_release_dirty_inode_budget(struct ubifs_info *c,
634 struct ubifs_inode *ui)
635{
Artem Bityutskiy182854b2008-07-18 18:54:29 +0300636 struct ubifs_budget_req req;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300637
Artem Bityutskiy182854b2008-07-18 18:54:29 +0300638 memset(&req, 0, sizeof(struct ubifs_budget_req));
Adrian Hunter6d6cb0d2009-04-08 14:07:57 +0200639 /* The "no space" flags will be cleared because dd_growth is > 0 */
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300640 req.dd_growth = c->inode_budget + ALIGN(ui->data_len, 8);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300641 ubifs_release_budget(c, &req);
642}
643
644/**
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300645 * ubifs_reported_space - calculate reported free space.
646 * @c: the UBIFS file-system description object
647 * @free: amount of free space
648 *
649 * This function calculates amount of free space which will be reported to
650 * user-space. User-space application tend to expect that if the file-system
651 * (e.g., via the 'statfs()' call) reports that it has N bytes available, they
652 * are able to write a file of size N. UBIFS attaches node headers to each data
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200653 * node and it has to write indexing nodes as well. This introduces additional
654 * overhead, and UBIFS has to report slightly less free space to meet the above
655 * expectations.
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300656 *
657 * This function assumes free space is made up of uncompressed data nodes and
658 * full index nodes (one per data node, tripled because we always allow enough
659 * space to write the index thrice).
660 *
661 * Note, the calculation is pessimistic, which means that most of the time
662 * UBIFS reports less space than it actually has.
663 */
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200664long long ubifs_reported_space(const struct ubifs_info *c, long long free)
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300665{
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300666 int divisor, factor, f;
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300667
668 /*
669 * Reported space size is @free * X, where X is UBIFS block size
670 * divided by UBIFS block size + all overhead one data block
671 * introduces. The overhead is the node header + indexing overhead.
672 *
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300673 * Indexing overhead calculations are based on the following formula:
674 * I = N/(f - 1) + 1, where I - number of indexing nodes, N - number
675 * of data nodes, f - fanout. Because effective UBIFS fanout is twice
676 * as less than maximum fanout, we assume that each data node
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300677 * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes.
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200678 * Note, the multiplier 3 is because UBIFS reserves thrice as more space
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300679 * for the index.
680 */
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300681 f = c->fanout > 3 ? c->fanout >> 1 : 2;
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300682 factor = UBIFS_BLOCK_SIZE;
683 divisor = UBIFS_MAX_DATA_NODE_SZ;
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300684 divisor += (c->max_idx_node_sz * 3) / (f - 1);
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300685 free *= factor;
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200686 return div_u64(free, divisor);
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300687}
688
689/**
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200690 * ubifs_get_free_space_nolock - return amount of free space.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300691 * @c: UBIFS file-system description object
692 *
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300693 * This function calculates amount of free space to report to user-space.
694 *
695 * Because UBIFS may introduce substantial overhead (the index, node headers,
Artem Bityutskiyfb1cd012009-03-16 09:56:57 +0200696 * alignment, wastage at the end of LEBs, etc), it cannot report real amount of
697 * free flash space it has (well, because not all dirty space is reclaimable,
698 * UBIFS does not actually know the real amount). If UBIFS did so, it would
699 * bread user expectations about what free space is. Users seem to accustomed
700 * to assume that if the file-system reports N bytes of free space, they would
701 * be able to fit a file of N bytes to the FS. This almost works for
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300702 * traditional file-systems, because they have way less overhead than UBIFS.
703 * So, to keep users happy, UBIFS tries to take the overhead into account.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300704 */
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200705long long ubifs_get_free_space_nolock(struct ubifs_info *c)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300706{
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200707 int rsvd_idx_lebs, lebs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300708 long long available, outstanding, free;
709
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200710 ubifs_assert(c->min_idx_lebs == ubifs_calc_min_idx_lebs(c));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300711 outstanding = c->budg_data_growth + c->budg_dd_growth;
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200712 available = ubifs_calc_available(c, c->min_idx_lebs);
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300713
714 /*
715 * When reporting free space to user-space, UBIFS guarantees that it is
716 * possible to write a file of free space size. This means that for
717 * empty LEBs we may use more precise calculations than
718 * 'ubifs_calc_available()' is using. Namely, we know that in empty
719 * LEBs we would waste only @c->leb_overhead bytes, not @c->dark_wm.
720 * Thus, amend the available space.
721 *
722 * Note, the calculations below are similar to what we have in
723 * 'do_budget_space()', so refer there for comments.
724 */
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200725 if (c->min_idx_lebs > c->lst.idx_lebs)
726 rsvd_idx_lebs = c->min_idx_lebs - c->lst.idx_lebs;
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300727 else
728 rsvd_idx_lebs = 0;
729 lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt -
730 c->lst.taken_empty_lebs;
731 lebs -= rsvd_idx_lebs;
732 available += lebs * (c->dark_wm - c->leb_overhead);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300733
734 if (available > outstanding)
735 free = ubifs_reported_space(c, available - outstanding);
736 else
737 free = 0;
738 return free;
739}
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200740
741/**
742 * ubifs_get_free_space - return amount of free space.
743 * @c: UBIFS file-system description object
744 *
745 * This function calculates and retuns amount of free space to report to
746 * user-space.
747 */
748long long ubifs_get_free_space(struct ubifs_info *c)
749{
750 long long free;
751
752 spin_lock(&c->space_lock);
753 free = ubifs_get_free_space_nolock(c);
754 spin_unlock(&c->space_lock);
755
756 return free;
757}