blob: f393620890ee4dfd8cb4c2bde0e32dce25390635 [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/**
197 * ubifs_calc_min_idx_lebs - calculate amount of eraseblocks for the index.
198 * @c: UBIFS file-system description object
199 *
200 * This function calculates and returns the number of eraseblocks which should
201 * be kept for index usage.
202 */
203int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
204{
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200205 int idx_lebs, eff_leb_size = c->leb_size - c->max_idx_node_sz;
206 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;
209
Adrian Hunter3a132522008-07-30 12:18:02 +0300210 /* And make sure we have thrice the index size of space reserved */
Artem Bityutskiyb364b412008-07-25 14:38:51 +0300211 idx_size = idx_size + (idx_size << 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300212
213 /*
214 * We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes'
215 * pair, nor similarly the two variables for the new index size, so we
216 * have to do this costly 64-bit division on fast-path.
217 */
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200218 idx_size += eff_leb_size - 1;
219 idx_lebs = div_u64(idx_size, eff_leb_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300220 /*
221 * The index head is not available for the in-the-gaps method, so add an
222 * extra LEB to compensate.
223 */
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200224 idx_lebs += 1;
225 if (idx_lebs < MIN_INDEX_LEBS)
226 idx_lebs = MIN_INDEX_LEBS;
227 return idx_lebs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300228}
229
230/**
231 * ubifs_calc_available - calculate available FS space.
232 * @c: UBIFS file-system description object
233 * @min_idx_lebs: minimum number of LEBs reserved for the index
234 *
235 * This function calculates and returns amount of FS space available for use.
236 */
237long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs)
238{
239 int subtract_lebs;
240 long long available;
241
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300242 available = c->main_bytes - c->lst.total_used;
243
244 /*
245 * Now 'available' contains theoretically available flash space
246 * assuming there is no index, so we have to subtract the space which
247 * is reserved for the index.
248 */
249 subtract_lebs = min_idx_lebs;
250
251 /* Take into account that GC reserves one LEB for its own needs */
252 subtract_lebs += 1;
253
254 /*
255 * The GC journal head LEB is not really accessible. And since
256 * different write types go to different heads, we may count only on
257 * one head's space.
258 */
259 subtract_lebs += c->jhead_cnt - 1;
260
261 /* We also reserve one LEB for deletions, which bypass budgeting */
262 subtract_lebs += 1;
263
264 available -= (long long)subtract_lebs * c->leb_size;
265
266 /* Subtract the dead space which is not available for use */
267 available -= c->lst.total_dead;
268
269 /*
270 * Subtract dark space, which might or might not be usable - it depends
271 * on the data which we have on the media and which will be written. If
272 * this is a lot of uncompressed or not-compressible data, the dark
273 * space cannot be used.
274 */
275 available -= c->lst.total_dark;
276
277 /*
278 * However, there is more dark space. The index may be bigger than
279 * @min_idx_lebs. Those extra LEBs are assumed to be available, but
280 * their dark space is not included in total_dark, so it is subtracted
281 * here.
282 */
283 if (c->lst.idx_lebs > min_idx_lebs) {
284 subtract_lebs = c->lst.idx_lebs - min_idx_lebs;
285 available -= subtract_lebs * c->dark_wm;
286 }
287
288 /* The calculations are rough and may end up with a negative number */
289 return available > 0 ? available : 0;
290}
291
292/**
293 * can_use_rp - check whether the user is allowed to use reserved pool.
294 * @c: UBIFS file-system description object
295 *
296 * UBIFS has so-called "reserved pool" which is flash space reserved
297 * for the superuser and for uses whose UID/GID is recorded in UBIFS superblock.
298 * This function checks whether current user is allowed to use reserved pool.
299 * Returns %1 current user is allowed to use reserved pool and %0 otherwise.
300 */
301static int can_use_rp(struct ubifs_info *c)
302{
David Howells26bf1942008-11-14 10:39:03 +1100303 if (current_fsuid() == c->rp_uid || capable(CAP_SYS_RESOURCE) ||
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300304 (c->rp_gid != 0 && in_group_p(c->rp_gid)))
305 return 1;
306 return 0;
307}
308
309/**
310 * do_budget_space - reserve flash space for index and data growth.
311 * @c: UBIFS file-system description object
312 *
313 * This function makes sure UBIFS has enough free eraseblocks for index growth
314 * and data.
315 *
Adrian Hunter3a132522008-07-30 12:18:02 +0300316 * When budgeting index space, UBIFS reserves thrice as many LEBs as the index
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300317 * would take if it was consolidated and written to the flash. This guarantees
318 * that the "in-the-gaps" commit method always succeeds and UBIFS will always
319 * be able to commit dirty index. So this function basically adds amount of
Artem Bityutskiyb364b412008-07-25 14:38:51 +0300320 * budgeted index space to the size of the current index, multiplies this by 3,
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300321 * and makes sure this does not exceed the amount of free eraseblocks.
322 *
323 * Notes about @c->min_idx_lebs and @c->lst.idx_lebs variables:
324 * o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might
325 * be large, because UBIFS does not do any index consolidation as long as
326 * there is free space. IOW, the index may take a lot of LEBs, but the LEBs
327 * will contain a lot of dirt.
328 * o @c->min_idx_lebs is the the index presumably takes. IOW, the index may be
329 * consolidated to take up to @c->min_idx_lebs LEBs.
330 *
331 * This function returns zero in case of success, and %-ENOSPC in case of
332 * failure.
333 */
334static int do_budget_space(struct ubifs_info *c)
335{
336 long long outstanding, available;
337 int lebs, rsvd_idx_lebs, min_idx_lebs;
338
339 /* First budget index space */
340 min_idx_lebs = ubifs_calc_min_idx_lebs(c);
341
342 /* Now 'min_idx_lebs' contains number of LEBs to reserve */
343 if (min_idx_lebs > c->lst.idx_lebs)
344 rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs;
345 else
346 rsvd_idx_lebs = 0;
347
348 /*
349 * The number of LEBs that are available to be used by the index is:
350 *
351 * @c->lst.empty_lebs + @c->freeable_cnt + @c->idx_gc_cnt -
352 * @c->lst.taken_empty_lebs
353 *
Artem Bityutskiy948cfb22008-08-20 11:56:33 +0300354 * @c->lst.empty_lebs are available because they are empty.
355 * @c->freeable_cnt are available because they contain only free and
356 * dirty space, @c->idx_gc_cnt are available because they are index
357 * LEBs that have been garbage collected and are awaiting the commit
358 * before they can be used. And the in-the-gaps method will grab these
359 * if it needs them. @c->lst.taken_empty_lebs are empty LEBs that have
360 * already been allocated for some purpose.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300361 *
Artem Bityutskiy948cfb22008-08-20 11:56:33 +0300362 * Note, @c->idx_gc_cnt is included to both @c->lst.empty_lebs (because
363 * these LEBs are empty) and to @c->lst.taken_empty_lebs (because they
364 * are taken until after the commit).
365 *
366 * Note, @c->lst.taken_empty_lebs may temporarily be higher by one
367 * because of the way we serialize LEB allocations and budgeting. See a
368 * comment in 'ubifs_find_free_space()'.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300369 */
370 lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt -
371 c->lst.taken_empty_lebs;
372 if (unlikely(rsvd_idx_lebs > lebs)) {
373 dbg_budg("out of indexing space: min_idx_lebs %d (old %d), "
374 "rsvd_idx_lebs %d", min_idx_lebs, c->min_idx_lebs,
375 rsvd_idx_lebs);
376 return -ENOSPC;
377 }
378
379 available = ubifs_calc_available(c, min_idx_lebs);
380 outstanding = c->budg_data_growth + c->budg_dd_growth;
381
382 if (unlikely(available < outstanding)) {
383 dbg_budg("out of data space: available %lld, outstanding %lld",
384 available, outstanding);
385 return -ENOSPC;
386 }
387
388 if (available - outstanding <= c->rp_size && !can_use_rp(c))
389 return -ENOSPC;
390
391 c->min_idx_lebs = min_idx_lebs;
392 return 0;
393}
394
395/**
396 * calc_idx_growth - calculate approximate index growth from budgeting request.
397 * @c: UBIFS file-system description object
398 * @req: budgeting request
399 *
400 * For now we assume each new node adds one znode. But this is rather poor
401 * approximation, though.
402 */
403static int calc_idx_growth(const struct ubifs_info *c,
404 const struct ubifs_budget_req *req)
405{
406 int znodes;
407
408 znodes = req->new_ino + (req->new_page << UBIFS_BLOCKS_PER_PAGE_SHIFT) +
409 req->new_dent;
410 return znodes * c->max_idx_node_sz;
411}
412
413/**
414 * calc_data_growth - calculate approximate amount of new data from budgeting
415 * request.
416 * @c: UBIFS file-system description object
417 * @req: budgeting request
418 */
419static int calc_data_growth(const struct ubifs_info *c,
420 const struct ubifs_budget_req *req)
421{
422 int data_growth;
423
424 data_growth = req->new_ino ? c->inode_budget : 0;
425 if (req->new_page)
426 data_growth += c->page_budget;
427 if (req->new_dent)
428 data_growth += c->dent_budget;
429 data_growth += req->new_ino_d;
430 return data_growth;
431}
432
433/**
434 * calc_dd_growth - calculate approximate amount of data which makes other data
435 * dirty from budgeting request.
436 * @c: UBIFS file-system description object
437 * @req: budgeting request
438 */
439static int calc_dd_growth(const struct ubifs_info *c,
440 const struct ubifs_budget_req *req)
441{
442 int dd_growth;
443
444 dd_growth = req->dirtied_page ? c->page_budget : 0;
445
446 if (req->dirtied_ino)
447 dd_growth += c->inode_budget << (req->dirtied_ino - 1);
448 if (req->mod_dent)
449 dd_growth += c->dent_budget;
450 dd_growth += req->dirtied_ino_d;
451 return dd_growth;
452}
453
454/**
455 * ubifs_budget_space - ensure there is enough space to complete an operation.
456 * @c: UBIFS file-system description object
457 * @req: budget request
458 *
459 * This function allocates budget for an operation. It uses pessimistic
460 * approximation of how much flash space the operation needs. The goal of this
461 * function is to make sure UBIFS always has flash space to flush all dirty
462 * pages, dirty inodes, and dirty znodes (liability). This function may force
463 * commit, garbage-collection or write-back. Returns zero in case of success,
464 * %-ENOSPC if there is no free space and other negative error codes in case of
465 * failures.
466 */
467int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req)
468{
469 int uninitialized_var(cmt_retries), uninitialized_var(wb_retries);
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500470 int err, idx_growth, data_growth, dd_growth, retried = 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300471
Artem Bityutskiy547000d2008-07-24 14:42:05 +0300472 ubifs_assert(req->new_page <= 1);
473 ubifs_assert(req->dirtied_page <= 1);
474 ubifs_assert(req->new_dent <= 1);
475 ubifs_assert(req->mod_dent <= 1);
476 ubifs_assert(req->new_ino <= 1);
477 ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300478 ubifs_assert(req->dirtied_ino <= 4);
479 ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300480 ubifs_assert(!(req->new_ino_d & 7));
481 ubifs_assert(!(req->dirtied_ino_d & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300482
483 data_growth = calc_data_growth(c, req);
484 dd_growth = calc_dd_growth(c, req);
485 if (!data_growth && !dd_growth)
486 return 0;
487 idx_growth = calc_idx_growth(c, req);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300488
489again:
490 spin_lock(&c->space_lock);
491 ubifs_assert(c->budg_idx_growth >= 0);
492 ubifs_assert(c->budg_data_growth >= 0);
493 ubifs_assert(c->budg_dd_growth >= 0);
494
495 if (unlikely(c->nospace) && (c->nospace_rp || !can_use_rp(c))) {
496 dbg_budg("no space");
497 spin_unlock(&c->space_lock);
498 return -ENOSPC;
499 }
500
501 c->budg_idx_growth += idx_growth;
502 c->budg_data_growth += data_growth;
503 c->budg_dd_growth += dd_growth;
504
505 err = do_budget_space(c);
506 if (likely(!err)) {
507 req->idx_growth = idx_growth;
508 req->data_growth = data_growth;
509 req->dd_growth = dd_growth;
510 spin_unlock(&c->space_lock);
511 return 0;
512 }
513
514 /* Restore the old values */
515 c->budg_idx_growth -= idx_growth;
516 c->budg_data_growth -= data_growth;
517 c->budg_dd_growth -= dd_growth;
518 spin_unlock(&c->space_lock);
519
520 if (req->fast) {
521 dbg_budg("no space for fast budgeting");
522 return err;
523 }
524
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500525 err = make_free_space(c);
526 cond_resched();
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300527 if (err == -EAGAIN) {
528 dbg_budg("try again");
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300529 goto again;
530 } else if (err == -ENOSPC) {
Artem Bityutskiy2acf8062008-12-09 11:04:40 -0500531 if (!retried) {
532 retried = 1;
533 dbg_budg("-ENOSPC, but anyway try once again");
534 goto again;
535 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300536 dbg_budg("FS is full, -ENOSPC");
537 c->nospace = 1;
538 if (can_use_rp(c) || c->rp_size == 0)
539 c->nospace_rp = 1;
540 smp_wmb();
541 } else
542 ubifs_err("cannot budget space, error %d", err);
543 return err;
544}
545
546/**
547 * ubifs_release_budget - release budgeted free space.
548 * @c: UBIFS file-system description object
549 * @req: budget request
550 *
551 * This function releases the space budgeted by 'ubifs_budget_space()'. Note,
552 * since the index changes (which were budgeted for in @req->idx_growth) will
553 * only be written to the media on commit, this function moves the index budget
554 * from @c->budg_idx_growth to @c->budg_uncommitted_idx. The latter will be
555 * zeroed by the commit operation.
556 */
557void ubifs_release_budget(struct ubifs_info *c, struct ubifs_budget_req *req)
558{
Artem Bityutskiy547000d2008-07-24 14:42:05 +0300559 ubifs_assert(req->new_page <= 1);
560 ubifs_assert(req->dirtied_page <= 1);
561 ubifs_assert(req->new_dent <= 1);
562 ubifs_assert(req->mod_dent <= 1);
563 ubifs_assert(req->new_ino <= 1);
564 ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300565 ubifs_assert(req->dirtied_ino <= 4);
566 ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300567 ubifs_assert(!(req->new_ino_d & 7));
568 ubifs_assert(!(req->dirtied_ino_d & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300569 if (!req->recalculate) {
570 ubifs_assert(req->idx_growth >= 0);
571 ubifs_assert(req->data_growth >= 0);
572 ubifs_assert(req->dd_growth >= 0);
573 }
574
575 if (req->recalculate) {
576 req->data_growth = calc_data_growth(c, req);
577 req->dd_growth = calc_dd_growth(c, req);
578 req->idx_growth = calc_idx_growth(c, req);
579 }
580
581 if (!req->data_growth && !req->dd_growth)
582 return;
583
584 c->nospace = c->nospace_rp = 0;
585 smp_wmb();
586
587 spin_lock(&c->space_lock);
588 c->budg_idx_growth -= req->idx_growth;
589 c->budg_uncommitted_idx += req->idx_growth;
590 c->budg_data_growth -= req->data_growth;
591 c->budg_dd_growth -= req->dd_growth;
592 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
593
594 ubifs_assert(c->budg_idx_growth >= 0);
595 ubifs_assert(c->budg_data_growth >= 0);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300596 ubifs_assert(c->budg_dd_growth >= 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300597 ubifs_assert(c->min_idx_lebs < c->main_lebs);
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300598 ubifs_assert(!(c->budg_idx_growth & 7));
599 ubifs_assert(!(c->budg_data_growth & 7));
600 ubifs_assert(!(c->budg_dd_growth & 7));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300601 spin_unlock(&c->space_lock);
602}
603
604/**
605 * ubifs_convert_page_budget - convert budget of a new page.
606 * @c: UBIFS file-system description object
607 *
608 * This function converts budget which was allocated for a new page of data to
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200609 * the budget of changing an existing page of data. The latter is smaller than
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300610 * the former, so this function only does simple re-calculation and does not
611 * involve any write-back.
612 */
613void ubifs_convert_page_budget(struct ubifs_info *c)
614{
615 spin_lock(&c->space_lock);
616 /* Release the index growth reservation */
617 c->budg_idx_growth -= c->max_idx_node_sz << UBIFS_BLOCKS_PER_PAGE_SHIFT;
618 /* Release the data growth reservation */
619 c->budg_data_growth -= c->page_budget;
620 /* Increase the dirty data growth reservation instead */
621 c->budg_dd_growth += c->page_budget;
622 /* And re-calculate the indexing space reservation */
623 c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
624 spin_unlock(&c->space_lock);
625}
626
627/**
628 * ubifs_release_dirty_inode_budget - release dirty inode budget.
629 * @c: UBIFS file-system description object
630 * @ui: UBIFS inode to release the budget for
631 *
632 * This function releases budget corresponding to a dirty inode. It is usually
633 * called when after the inode has been written to the media and marked as
634 * clean.
635 */
636void ubifs_release_dirty_inode_budget(struct ubifs_info *c,
637 struct ubifs_inode *ui)
638{
Artem Bityutskiy182854b2008-07-18 18:54:29 +0300639 struct ubifs_budget_req req;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300640
Artem Bityutskiy182854b2008-07-18 18:54:29 +0300641 memset(&req, 0, sizeof(struct ubifs_budget_req));
Artem Bityutskiydab4b4d2008-07-24 14:52:45 +0300642 req.dd_growth = c->inode_budget + ALIGN(ui->data_len, 8);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300643 ubifs_release_budget(c, &req);
644}
645
646/**
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300647 * ubifs_reported_space - calculate reported free space.
648 * @c: the UBIFS file-system description object
649 * @free: amount of free space
650 *
651 * This function calculates amount of free space which will be reported to
652 * user-space. User-space application tend to expect that if the file-system
653 * (e.g., via the 'statfs()' call) reports that it has N bytes available, they
654 * are able to write a file of size N. UBIFS attaches node headers to each data
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200655 * node and it has to write indexing nodes as well. This introduces additional
656 * overhead, and UBIFS has to report slightly less free space to meet the above
657 * expectations.
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300658 *
659 * This function assumes free space is made up of uncompressed data nodes and
660 * full index nodes (one per data node, tripled because we always allow enough
661 * space to write the index thrice).
662 *
663 * Note, the calculation is pessimistic, which means that most of the time
664 * UBIFS reports less space than it actually has.
665 */
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200666long long ubifs_reported_space(const struct ubifs_info *c, long long free)
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300667{
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300668 int divisor, factor, f;
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300669
670 /*
671 * Reported space size is @free * X, where X is UBIFS block size
672 * divided by UBIFS block size + all overhead one data block
673 * introduces. The overhead is the node header + indexing overhead.
674 *
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300675 * Indexing overhead calculations are based on the following formula:
676 * I = N/(f - 1) + 1, where I - number of indexing nodes, N - number
677 * of data nodes, f - fanout. Because effective UBIFS fanout is twice
678 * as less than maximum fanout, we assume that each data node
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300679 * introduces 3 * @c->max_idx_node_sz / (@c->fanout/2 - 1) bytes.
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200680 * Note, the multiplier 3 is because UBIFS reserves thrice as more space
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300681 * for the index.
682 */
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300683 f = c->fanout > 3 ? c->fanout >> 1 : 2;
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300684 factor = UBIFS_BLOCK_SIZE;
685 divisor = UBIFS_MAX_DATA_NODE_SZ;
Artem Bityutskiyf171d4d2008-09-03 16:17:14 +0300686 divisor += (c->max_idx_node_sz * 3) / (f - 1);
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300687 free *= factor;
Artem Bityutskiy4d61db42008-12-18 14:06:51 +0200688 return div_u64(free, divisor);
Artem Bityutskiy4b5f2762008-08-25 16:15:56 +0300689}
690
691/**
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200692 * ubifs_get_free_space_nolock - return amount of free space.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300693 * @c: UBIFS file-system description object
694 *
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300695 * This function calculates amount of free space to report to user-space.
696 *
697 * Because UBIFS may introduce substantial overhead (the index, node headers,
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200698 * alignment, wastage at the end of eraseblocks, etc), it cannot report real
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300699 * amount of free flash space it has (well, because not all dirty space is
Artem Bityutskiy80736d42008-12-30 17:44:02 +0200700 * reclaimable, UBIFS does not actually know the real amount). If UBIFS did so,
701 * it would bread user expectations about what free space is. Users seem to
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300702 * accustomed to assume that if the file-system reports N bytes of free space,
703 * they would be able to fit a file of N bytes to the FS. This almost works for
704 * traditional file-systems, because they have way less overhead than UBIFS.
705 * So, to keep users happy, UBIFS tries to take the overhead into account.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300706 */
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200707long long ubifs_get_free_space_nolock(struct ubifs_info *c)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300708{
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200709 int rsvd_idx_lebs, lebs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300710 long long available, outstanding, free;
711
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200712 ubifs_assert(c->min_idx_lebs == ubifs_calc_min_idx_lebs(c));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300713 outstanding = c->budg_data_growth + c->budg_dd_growth;
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200714 available = ubifs_calc_available(c, c->min_idx_lebs);
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300715
716 /*
717 * When reporting free space to user-space, UBIFS guarantees that it is
718 * possible to write a file of free space size. This means that for
719 * empty LEBs we may use more precise calculations than
720 * 'ubifs_calc_available()' is using. Namely, we know that in empty
721 * LEBs we would waste only @c->leb_overhead bytes, not @c->dark_wm.
722 * Thus, amend the available space.
723 *
724 * Note, the calculations below are similar to what we have in
725 * 'do_budget_space()', so refer there for comments.
726 */
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200727 if (c->min_idx_lebs > c->lst.idx_lebs)
728 rsvd_idx_lebs = c->min_idx_lebs - c->lst.idx_lebs;
Artem Bityutskiy7dad1812008-08-25 18:58:19 +0300729 else
730 rsvd_idx_lebs = 0;
731 lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt -
732 c->lst.taken_empty_lebs;
733 lebs -= rsvd_idx_lebs;
734 available += lebs * (c->dark_wm - c->leb_overhead);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300735
736 if (available > outstanding)
737 free = ubifs_reported_space(c, available - outstanding);
738 else
739 free = 0;
740 return free;
741}
Artem Bityutskiy84abf972009-01-23 14:54:59 +0200742
743/**
744 * ubifs_get_free_space - return amount of free space.
745 * @c: UBIFS file-system description object
746 *
747 * This function calculates and retuns amount of free space to report to
748 * user-space.
749 */
750long long ubifs_get_free_space(struct ubifs_info *c)
751{
752 long long free;
753
754 spin_lock(&c->space_lock);
755 free = ubifs_get_free_space_nolock(c);
756 spin_unlock(&c->space_lock);
757
758 return free;
759}