1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (C) 2010 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_ialloc.h"
31 #include "xfs_quota.h"
32 #include "xfs_trans.h"
33 #include "xfs_qm.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trace.h"
36
37 /*
38 * A buffer has a format structure overhead in the log in addition
39 * to the data, so we need to take this into account when reserving
40 * space in a transaction for a buffer. Round the space required up
41 * to a multiple of 128 bytes so that we don't change the historical
42 * reservation that has been used for this overhead.
43 */
44 STATIC uint
xfs_buf_log_overhead(void)45 xfs_buf_log_overhead(void)
46 {
47 return round_up(sizeof(struct xlog_op_header) +
48 sizeof(struct xfs_buf_log_format), 128);
49 }
50
51 /*
52 * Calculate out transaction log reservation per item in bytes.
53 *
54 * The nbufs argument is used to indicate the number of items that
55 * will be changed in a transaction. size is used to tell how many
56 * bytes should be reserved per item.
57 */
58 STATIC uint
xfs_calc_buf_res(uint nbufs,uint size)59 xfs_calc_buf_res(
60 uint nbufs,
61 uint size)
62 {
63 return nbufs * (size + xfs_buf_log_overhead());
64 }
65
66 /*
67 * Per-extent log reservation for the btree changes involved in freeing or
68 * allocating an extent. In classic XFS there were two trees that will be
69 * modified (bnobt + cntbt). With rmap enabled, there are three trees
70 * (rmapbt). With reflink, there are four trees (refcountbt). The number of
71 * blocks reserved is based on the formula:
72 *
73 * num trees * ((2 blocks/level * max depth) - 1)
74 *
75 * Keep in mind that max depth is calculated separately for each type of tree.
76 */
77 uint
xfs_allocfree_log_count(struct xfs_mount * mp,uint num_ops)78 xfs_allocfree_log_count(
79 struct xfs_mount *mp,
80 uint num_ops)
81 {
82 uint blocks;
83
84 blocks = num_ops * 2 * (2 * mp->m_ag_maxlevels - 1);
85 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
86 blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1);
87 if (xfs_sb_version_hasreflink(&mp->m_sb))
88 blocks += num_ops * (2 * mp->m_refc_maxlevels - 1);
89
90 return blocks;
91 }
92
93 /*
94 * Logging inodes is really tricksy. They are logged in memory format,
95 * which means that what we write into the log doesn't directly translate into
96 * the amount of space they use on disk.
97 *
98 * Case in point - btree format forks in memory format use more space than the
99 * on-disk format. In memory, the buffer contains a normal btree block header so
100 * the btree code can treat it as though it is just another generic buffer.
101 * However, when we write it to the inode fork, we don't write all of this
102 * header as it isn't needed. e.g. the root is only ever in the inode, so
103 * there's no need for sibling pointers which would waste 16 bytes of space.
104 *
105 * Hence when we have an inode with a maximally sized btree format fork, then
106 * amount of information we actually log is greater than the size of the inode
107 * on disk. Hence we need an inode reservation function that calculates all this
108 * correctly. So, we log:
109 *
110 * - 4 log op headers for object
111 * - for the ilf, the inode core and 2 forks
112 * - inode log format object
113 * - the inode core
114 * - two inode forks containing bmap btree root blocks.
115 * - the btree data contained by both forks will fit into the inode size,
116 * hence when combined with the inode core above, we have a total of the
117 * actual inode size.
118 * - the BMBT headers need to be accounted separately, as they are
119 * additional to the records and pointers that fit inside the inode
120 * forks.
121 */
122 STATIC uint
xfs_calc_inode_res(struct xfs_mount * mp,uint ninodes)123 xfs_calc_inode_res(
124 struct xfs_mount *mp,
125 uint ninodes)
126 {
127 return ninodes *
128 (4 * sizeof(struct xlog_op_header) +
129 sizeof(struct xfs_inode_log_format) +
130 mp->m_sb.sb_inodesize +
131 2 * XFS_BMBT_BLOCK_LEN(mp));
132 }
133
134 /*
135 * The free inode btree is a conditional feature and the log reservation
136 * requirements differ slightly from that of the traditional inode allocation
137 * btree. The finobt tracks records for inode chunks with at least one free
138 * inode. A record can be removed from the tree for an inode allocation
139 * or free and thus the finobt reservation is unconditional across:
140 *
141 * - inode allocation
142 * - inode free
143 * - inode chunk allocation
144 *
145 * The 'modify' param indicates to include the record modification scenario. The
146 * 'alloc' param indicates to include the reservation for free space btree
147 * modifications on behalf of finobt modifications. This is required only for
148 * transactions that do not already account for free space btree modifications.
149 *
150 * the free inode btree: max depth * block size
151 * the allocation btrees: 2 trees * (max depth - 1) * block size
152 * the free inode btree entry: block size
153 */
154 STATIC uint
xfs_calc_finobt_res(struct xfs_mount * mp,int alloc,int modify)155 xfs_calc_finobt_res(
156 struct xfs_mount *mp,
157 int alloc,
158 int modify)
159 {
160 uint res;
161
162 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
163 return 0;
164
165 res = xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1));
166 if (alloc)
167 res += xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
168 XFS_FSB_TO_B(mp, 1));
169 if (modify)
170 res += (uint)XFS_FSB_TO_B(mp, 1);
171
172 return res;
173 }
174
175 /*
176 * Various log reservation values.
177 *
178 * These are based on the size of the file system block because that is what
179 * most transactions manipulate. Each adds in an additional 128 bytes per
180 * item logged to try to account for the overhead of the transaction mechanism.
181 *
182 * Note: Most of the reservations underestimate the number of allocation
183 * groups into which they could free extents in the xfs_defer_finish() call.
184 * This is because the number in the worst case is quite high and quite
185 * unusual. In order to fix this we need to change xfs_defer_finish() to free
186 * extents in only a single AG at a time. This will require changes to the
187 * EFI code as well, however, so that the EFI for the extents not freed is
188 * logged again in each transaction. See SGI PV #261917.
189 *
190 * Reservation functions here avoid a huge stack in xfs_trans_init due to
191 * register overflow from temporaries in the calculations.
192 */
193
194
195 /*
196 * In a write transaction we can allocate a maximum of 2
197 * extents. This gives:
198 * the inode getting the new extents: inode size
199 * the inode's bmap btree: max depth * block size
200 * the agfs of the ags from which the extents are allocated: 2 * sector
201 * the superblock free block counter: sector size
202 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
203 * And the bmap_finish transaction can free bmap blocks in a join:
204 * the agfs of the ags containing the blocks: 2 * sector size
205 * the agfls of the ags containing the blocks: 2 * sector size
206 * the super block free block counter: sector size
207 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
208 */
209 STATIC uint
xfs_calc_write_reservation(struct xfs_mount * mp)210 xfs_calc_write_reservation(
211 struct xfs_mount *mp)
212 {
213 return XFS_DQUOT_LOGRES(mp) +
214 MAX((xfs_calc_inode_res(mp, 1) +
215 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
216 XFS_FSB_TO_B(mp, 1)) +
217 xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
218 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
219 XFS_FSB_TO_B(mp, 1))),
220 (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
221 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
222 XFS_FSB_TO_B(mp, 1))));
223 }
224
225 /*
226 * In truncating a file we free up to two extents at once. We can modify:
227 * the inode being truncated: inode size
228 * the inode's bmap btree: (max depth + 1) * block size
229 * And the bmap_finish transaction can free the blocks and bmap blocks:
230 * the agf for each of the ags: 4 * sector size
231 * the agfl for each of the ags: 4 * sector size
232 * the super block to reflect the freed blocks: sector size
233 * worst case split in allocation btrees per extent assuming 4 extents:
234 * 4 exts * 2 trees * (2 * max depth - 1) * block size
235 */
236 STATIC uint
xfs_calc_itruncate_reservation(struct xfs_mount * mp)237 xfs_calc_itruncate_reservation(
238 struct xfs_mount *mp)
239 {
240 return XFS_DQUOT_LOGRES(mp) +
241 MAX((xfs_calc_inode_res(mp, 1) +
242 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1,
243 XFS_FSB_TO_B(mp, 1))),
244 (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
245 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4),
246 XFS_FSB_TO_B(mp, 1))));
247 }
248
249 /*
250 * In renaming a files we can modify:
251 * the four inodes involved: 4 * inode size
252 * the two directory btrees: 2 * (max depth + v2) * dir block size
253 * the two directory bmap btrees: 2 * max depth * block size
254 * And the bmap_finish transaction can free dir and bmap blocks (two sets
255 * of bmap blocks) giving:
256 * the agf for the ags in which the blocks live: 3 * sector size
257 * the agfl for the ags in which the blocks live: 3 * sector size
258 * the superblock for the free block count: sector size
259 * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
260 */
261 STATIC uint
xfs_calc_rename_reservation(struct xfs_mount * mp)262 xfs_calc_rename_reservation(
263 struct xfs_mount *mp)
264 {
265 return XFS_DQUOT_LOGRES(mp) +
266 MAX((xfs_calc_inode_res(mp, 4) +
267 xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp),
268 XFS_FSB_TO_B(mp, 1))),
269 (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) +
270 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 3),
271 XFS_FSB_TO_B(mp, 1))));
272 }
273
274 /*
275 * For removing an inode from unlinked list at first, we can modify:
276 * the agi hash list and counters: sector size
277 * the on disk inode before ours in the agi hash list: inode cluster size
278 */
279 STATIC uint
xfs_calc_iunlink_remove_reservation(struct xfs_mount * mp)280 xfs_calc_iunlink_remove_reservation(
281 struct xfs_mount *mp)
282 {
283 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
284 max_t(uint, XFS_FSB_TO_B(mp, 1), mp->m_inode_cluster_size);
285 }
286
287 /*
288 * For creating a link to an inode:
289 * the parent directory inode: inode size
290 * the linked inode: inode size
291 * the directory btree could split: (max depth + v2) * dir block size
292 * the directory bmap btree could join or split: (max depth + v2) * blocksize
293 * And the bmap_finish transaction can free some bmap blocks giving:
294 * the agf for the ag in which the blocks live: sector size
295 * the agfl for the ag in which the blocks live: sector size
296 * the superblock for the free block count: sector size
297 * the allocation btrees: 2 trees * (2 * max depth - 1) * block size
298 */
299 STATIC uint
xfs_calc_link_reservation(struct xfs_mount * mp)300 xfs_calc_link_reservation(
301 struct xfs_mount *mp)
302 {
303 return XFS_DQUOT_LOGRES(mp) +
304 xfs_calc_iunlink_remove_reservation(mp) +
305 MAX((xfs_calc_inode_res(mp, 2) +
306 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
307 XFS_FSB_TO_B(mp, 1))),
308 (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
309 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
310 XFS_FSB_TO_B(mp, 1))));
311 }
312
313 /*
314 * For adding an inode to unlinked list we can modify:
315 * the agi hash list: sector size
316 * the unlinked inode: inode size
317 */
318 STATIC uint
xfs_calc_iunlink_add_reservation(xfs_mount_t * mp)319 xfs_calc_iunlink_add_reservation(xfs_mount_t *mp)
320 {
321 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
322 xfs_calc_inode_res(mp, 1);
323 }
324
325 /*
326 * For removing a directory entry we can modify:
327 * the parent directory inode: inode size
328 * the removed inode: inode size
329 * the directory btree could join: (max depth + v2) * dir block size
330 * the directory bmap btree could join or split: (max depth + v2) * blocksize
331 * And the bmap_finish transaction can free the dir and bmap blocks giving:
332 * the agf for the ag in which the blocks live: 2 * sector size
333 * the agfl for the ag in which the blocks live: 2 * sector size
334 * the superblock for the free block count: sector size
335 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
336 */
337 STATIC uint
xfs_calc_remove_reservation(struct xfs_mount * mp)338 xfs_calc_remove_reservation(
339 struct xfs_mount *mp)
340 {
341 return XFS_DQUOT_LOGRES(mp) +
342 xfs_calc_iunlink_add_reservation(mp) +
343 MAX((xfs_calc_inode_res(mp, 1) +
344 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
345 XFS_FSB_TO_B(mp, 1))),
346 (xfs_calc_buf_res(4, mp->m_sb.sb_sectsize) +
347 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
348 XFS_FSB_TO_B(mp, 1))));
349 }
350
351 /*
352 * For create, break it in to the two cases that the transaction
353 * covers. We start with the modify case - allocation done by modification
354 * of the state of existing inodes - and the allocation case.
355 */
356
357 /*
358 * For create we can modify:
359 * the parent directory inode: inode size
360 * the new inode: inode size
361 * the inode btree entry: block size
362 * the superblock for the nlink flag: sector size
363 * the directory btree: (max depth + v2) * dir block size
364 * the directory inode's bmap btree: (max depth + v2) * block size
365 * the finobt (record modification and allocation btrees)
366 */
367 STATIC uint
xfs_calc_create_resv_modify(struct xfs_mount * mp)368 xfs_calc_create_resv_modify(
369 struct xfs_mount *mp)
370 {
371 return xfs_calc_inode_res(mp, 2) +
372 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
373 (uint)XFS_FSB_TO_B(mp, 1) +
374 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), XFS_FSB_TO_B(mp, 1)) +
375 xfs_calc_finobt_res(mp, 1, 1);
376 }
377
378 /*
379 * For create we can allocate some inodes giving:
380 * the agi and agf of the ag getting the new inodes: 2 * sectorsize
381 * the superblock for the nlink flag: sector size
382 * the inode blocks allocated: mp->m_ialloc_blks * blocksize
383 * the inode btree: max depth * blocksize
384 * the allocation btrees: 2 trees * (max depth - 1) * block size
385 */
386 STATIC uint
xfs_calc_create_resv_alloc(struct xfs_mount * mp)387 xfs_calc_create_resv_alloc(
388 struct xfs_mount *mp)
389 {
390 return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
391 mp->m_sb.sb_sectsize +
392 xfs_calc_buf_res(mp->m_ialloc_blks, XFS_FSB_TO_B(mp, 1)) +
393 xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1)) +
394 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
395 XFS_FSB_TO_B(mp, 1));
396 }
397
398 STATIC uint
__xfs_calc_create_reservation(struct xfs_mount * mp)399 __xfs_calc_create_reservation(
400 struct xfs_mount *mp)
401 {
402 return XFS_DQUOT_LOGRES(mp) +
403 MAX(xfs_calc_create_resv_alloc(mp),
404 xfs_calc_create_resv_modify(mp));
405 }
406
407 /*
408 * For icreate we can allocate some inodes giving:
409 * the agi and agf of the ag getting the new inodes: 2 * sectorsize
410 * the superblock for the nlink flag: sector size
411 * the inode btree: max depth * blocksize
412 * the allocation btrees: 2 trees * (max depth - 1) * block size
413 * the finobt (record insertion)
414 */
415 STATIC uint
xfs_calc_icreate_resv_alloc(struct xfs_mount * mp)416 xfs_calc_icreate_resv_alloc(
417 struct xfs_mount *mp)
418 {
419 return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
420 mp->m_sb.sb_sectsize +
421 xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1)) +
422 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
423 XFS_FSB_TO_B(mp, 1)) +
424 xfs_calc_finobt_res(mp, 0, 0);
425 }
426
427 STATIC uint
xfs_calc_icreate_reservation(xfs_mount_t * mp)428 xfs_calc_icreate_reservation(xfs_mount_t *mp)
429 {
430 return XFS_DQUOT_LOGRES(mp) +
431 MAX(xfs_calc_icreate_resv_alloc(mp),
432 xfs_calc_create_resv_modify(mp));
433 }
434
435 STATIC uint
xfs_calc_create_reservation(struct xfs_mount * mp)436 xfs_calc_create_reservation(
437 struct xfs_mount *mp)
438 {
439 if (xfs_sb_version_hascrc(&mp->m_sb))
440 return xfs_calc_icreate_reservation(mp);
441 return __xfs_calc_create_reservation(mp);
442
443 }
444
445 STATIC uint
xfs_calc_create_tmpfile_reservation(struct xfs_mount * mp)446 xfs_calc_create_tmpfile_reservation(
447 struct xfs_mount *mp)
448 {
449 uint res = XFS_DQUOT_LOGRES(mp);
450
451 if (xfs_sb_version_hascrc(&mp->m_sb))
452 res += xfs_calc_icreate_resv_alloc(mp);
453 else
454 res += xfs_calc_create_resv_alloc(mp);
455
456 return res + xfs_calc_iunlink_add_reservation(mp);
457 }
458
459 /*
460 * Making a new directory is the same as creating a new file.
461 */
462 STATIC uint
xfs_calc_mkdir_reservation(struct xfs_mount * mp)463 xfs_calc_mkdir_reservation(
464 struct xfs_mount *mp)
465 {
466 return xfs_calc_create_reservation(mp);
467 }
468
469
470 /*
471 * Making a new symplink is the same as creating a new file, but
472 * with the added blocks for remote symlink data which can be up to 1kB in
473 * length (XFS_SYMLINK_MAXLEN).
474 */
475 STATIC uint
xfs_calc_symlink_reservation(struct xfs_mount * mp)476 xfs_calc_symlink_reservation(
477 struct xfs_mount *mp)
478 {
479 return xfs_calc_create_reservation(mp) +
480 xfs_calc_buf_res(1, XFS_SYMLINK_MAXLEN);
481 }
482
483 /*
484 * In freeing an inode we can modify:
485 * the inode being freed: inode size
486 * the super block free inode counter: sector size
487 * the agi hash list and counters: sector size
488 * the inode btree entry: block size
489 * the on disk inode before ours in the agi hash list: inode cluster size
490 * the inode btree: max depth * blocksize
491 * the allocation btrees: 2 trees * (max depth - 1) * block size
492 * the finobt (record insertion, removal or modification)
493 */
494 STATIC uint
xfs_calc_ifree_reservation(struct xfs_mount * mp)495 xfs_calc_ifree_reservation(
496 struct xfs_mount *mp)
497 {
498 return XFS_DQUOT_LOGRES(mp) +
499 xfs_calc_inode_res(mp, 1) +
500 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
501 xfs_calc_buf_res(1, XFS_FSB_TO_B(mp, 1)) +
502 xfs_calc_iunlink_remove_reservation(mp) +
503 xfs_calc_buf_res(1, 0) +
504 xfs_calc_buf_res(2 + mp->m_ialloc_blks +
505 mp->m_in_maxlevels, 0) +
506 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
507 XFS_FSB_TO_B(mp, 1)) +
508 xfs_calc_finobt_res(mp, 0, 1);
509 }
510
511 /*
512 * When only changing the inode we log the inode and possibly the superblock
513 * We also add a bit of slop for the transaction stuff.
514 */
515 STATIC uint
xfs_calc_ichange_reservation(struct xfs_mount * mp)516 xfs_calc_ichange_reservation(
517 struct xfs_mount *mp)
518 {
519 return XFS_DQUOT_LOGRES(mp) +
520 xfs_calc_inode_res(mp, 1) +
521 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
522
523 }
524
525 /*
526 * Growing the data section of the filesystem.
527 * superblock
528 * agi and agf
529 * allocation btrees
530 */
531 STATIC uint
xfs_calc_growdata_reservation(struct xfs_mount * mp)532 xfs_calc_growdata_reservation(
533 struct xfs_mount *mp)
534 {
535 return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
536 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
537 XFS_FSB_TO_B(mp, 1));
538 }
539
540 /*
541 * Growing the rt section of the filesystem.
542 * In the first set of transactions (ALLOC) we allocate space to the
543 * bitmap or summary files.
544 * superblock: sector size
545 * agf of the ag from which the extent is allocated: sector size
546 * bmap btree for bitmap/summary inode: max depth * blocksize
547 * bitmap/summary inode: inode size
548 * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
549 */
550 STATIC uint
xfs_calc_growrtalloc_reservation(struct xfs_mount * mp)551 xfs_calc_growrtalloc_reservation(
552 struct xfs_mount *mp)
553 {
554 return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
555 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
556 XFS_FSB_TO_B(mp, 1)) +
557 xfs_calc_inode_res(mp, 1) +
558 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
559 XFS_FSB_TO_B(mp, 1));
560 }
561
562 /*
563 * Growing the rt section of the filesystem.
564 * In the second set of transactions (ZERO) we zero the new metadata blocks.
565 * one bitmap/summary block: blocksize
566 */
567 STATIC uint
xfs_calc_growrtzero_reservation(struct xfs_mount * mp)568 xfs_calc_growrtzero_reservation(
569 struct xfs_mount *mp)
570 {
571 return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize);
572 }
573
574 /*
575 * Growing the rt section of the filesystem.
576 * In the third set of transactions (FREE) we update metadata without
577 * allocating any new blocks.
578 * superblock: sector size
579 * bitmap inode: inode size
580 * summary inode: inode size
581 * one bitmap block: blocksize
582 * summary blocks: new summary size
583 */
584 STATIC uint
xfs_calc_growrtfree_reservation(struct xfs_mount * mp)585 xfs_calc_growrtfree_reservation(
586 struct xfs_mount *mp)
587 {
588 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
589 xfs_calc_inode_res(mp, 2) +
590 xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) +
591 xfs_calc_buf_res(1, mp->m_rsumsize);
592 }
593
594 /*
595 * Logging the inode modification timestamp on a synchronous write.
596 * inode
597 */
598 STATIC uint
xfs_calc_swrite_reservation(struct xfs_mount * mp)599 xfs_calc_swrite_reservation(
600 struct xfs_mount *mp)
601 {
602 return xfs_calc_inode_res(mp, 1);
603 }
604
605 /*
606 * Logging the inode mode bits when writing a setuid/setgid file
607 * inode
608 */
609 STATIC uint
xfs_calc_writeid_reservation(struct xfs_mount * mp)610 xfs_calc_writeid_reservation(
611 struct xfs_mount *mp)
612 {
613 return xfs_calc_inode_res(mp, 1);
614 }
615
616 /*
617 * Converting the inode from non-attributed to attributed.
618 * the inode being converted: inode size
619 * agf block and superblock (for block allocation)
620 * the new block (directory sized)
621 * bmap blocks for the new directory block
622 * allocation btrees
623 */
624 STATIC uint
xfs_calc_addafork_reservation(struct xfs_mount * mp)625 xfs_calc_addafork_reservation(
626 struct xfs_mount *mp)
627 {
628 return XFS_DQUOT_LOGRES(mp) +
629 xfs_calc_inode_res(mp, 1) +
630 xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
631 xfs_calc_buf_res(1, mp->m_dir_geo->blksize) +
632 xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1,
633 XFS_FSB_TO_B(mp, 1)) +
634 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1),
635 XFS_FSB_TO_B(mp, 1));
636 }
637
638 /*
639 * Removing the attribute fork of a file
640 * the inode being truncated: inode size
641 * the inode's bmap btree: max depth * block size
642 * And the bmap_finish transaction can free the blocks and bmap blocks:
643 * the agf for each of the ags: 4 * sector size
644 * the agfl for each of the ags: 4 * sector size
645 * the super block to reflect the freed blocks: sector size
646 * worst case split in allocation btrees per extent assuming 4 extents:
647 * 4 exts * 2 trees * (2 * max depth - 1) * block size
648 */
649 STATIC uint
xfs_calc_attrinval_reservation(struct xfs_mount * mp)650 xfs_calc_attrinval_reservation(
651 struct xfs_mount *mp)
652 {
653 return MAX((xfs_calc_inode_res(mp, 1) +
654 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
655 XFS_FSB_TO_B(mp, 1))),
656 (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
657 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4),
658 XFS_FSB_TO_B(mp, 1))));
659 }
660
661 /*
662 * Setting an attribute at mount time.
663 * the inode getting the attribute
664 * the superblock for allocations
665 * the agfs extents are allocated from
666 * the attribute btree * max depth
667 * the inode allocation btree
668 * Since attribute transaction space is dependent on the size of the attribute,
669 * the calculation is done partially at mount time and partially at runtime(see
670 * below).
671 */
672 STATIC uint
xfs_calc_attrsetm_reservation(struct xfs_mount * mp)673 xfs_calc_attrsetm_reservation(
674 struct xfs_mount *mp)
675 {
676 return XFS_DQUOT_LOGRES(mp) +
677 xfs_calc_inode_res(mp, 1) +
678 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
679 xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1));
680 }
681
682 /*
683 * Setting an attribute at runtime, transaction space unit per block.
684 * the superblock for allocations: sector size
685 * the inode bmap btree could join or split: max depth * block size
686 * Since the runtime attribute transaction space is dependent on the total
687 * blocks needed for the 1st bmap, here we calculate out the space unit for
688 * one block so that the caller could figure out the total space according
689 * to the attibute extent length in blocks by:
690 * ext * M_RES(mp)->tr_attrsetrt.tr_logres
691 */
692 STATIC uint
xfs_calc_attrsetrt_reservation(struct xfs_mount * mp)693 xfs_calc_attrsetrt_reservation(
694 struct xfs_mount *mp)
695 {
696 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
697 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
698 XFS_FSB_TO_B(mp, 1));
699 }
700
701 /*
702 * Removing an attribute.
703 * the inode: inode size
704 * the attribute btree could join: max depth * block size
705 * the inode bmap btree could join or split: max depth * block size
706 * And the bmap_finish transaction can free the attr blocks freed giving:
707 * the agf for the ag in which the blocks live: 2 * sector size
708 * the agfl for the ag in which the blocks live: 2 * sector size
709 * the superblock for the free block count: sector size
710 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
711 */
712 STATIC uint
xfs_calc_attrrm_reservation(struct xfs_mount * mp)713 xfs_calc_attrrm_reservation(
714 struct xfs_mount *mp)
715 {
716 return XFS_DQUOT_LOGRES(mp) +
717 MAX((xfs_calc_inode_res(mp, 1) +
718 xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH,
719 XFS_FSB_TO_B(mp, 1)) +
720 (uint)XFS_FSB_TO_B(mp,
721 XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
722 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)),
723 (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
724 xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2),
725 XFS_FSB_TO_B(mp, 1))));
726 }
727
728 /*
729 * Clearing a bad agino number in an agi hash bucket.
730 */
731 STATIC uint
xfs_calc_clear_agi_bucket_reservation(struct xfs_mount * mp)732 xfs_calc_clear_agi_bucket_reservation(
733 struct xfs_mount *mp)
734 {
735 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
736 }
737
738 /*
739 * Adjusting quota limits.
740 * the xfs_disk_dquot_t: sizeof(struct xfs_disk_dquot)
741 */
742 STATIC uint
xfs_calc_qm_setqlim_reservation(struct xfs_mount * mp)743 xfs_calc_qm_setqlim_reservation(
744 struct xfs_mount *mp)
745 {
746 return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot));
747 }
748
749 /*
750 * Allocating quota on disk if needed.
751 * the write transaction log space for quota file extent allocation
752 * the unit of quota allocation: one system block size
753 */
754 STATIC uint
xfs_calc_qm_dqalloc_reservation(struct xfs_mount * mp)755 xfs_calc_qm_dqalloc_reservation(
756 struct xfs_mount *mp)
757 {
758 return xfs_calc_write_reservation(mp) +
759 xfs_calc_buf_res(1,
760 XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1);
761 }
762
763 /*
764 * Turning off quotas.
765 * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2
766 * the superblock for the quota flags: sector size
767 */
768 STATIC uint
xfs_calc_qm_quotaoff_reservation(struct xfs_mount * mp)769 xfs_calc_qm_quotaoff_reservation(
770 struct xfs_mount *mp)
771 {
772 return sizeof(struct xfs_qoff_logitem) * 2 +
773 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
774 }
775
776 /*
777 * End of turning off quotas.
778 * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2
779 */
780 STATIC uint
xfs_calc_qm_quotaoff_end_reservation(struct xfs_mount * mp)781 xfs_calc_qm_quotaoff_end_reservation(
782 struct xfs_mount *mp)
783 {
784 return sizeof(struct xfs_qoff_logitem) * 2;
785 }
786
787 /*
788 * Syncing the incore super block changes to disk.
789 * the super block to reflect the changes: sector size
790 */
791 STATIC uint
xfs_calc_sb_reservation(struct xfs_mount * mp)792 xfs_calc_sb_reservation(
793 struct xfs_mount *mp)
794 {
795 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
796 }
797
798 void
xfs_trans_resv_calc(struct xfs_mount * mp,struct xfs_trans_resv * resp)799 xfs_trans_resv_calc(
800 struct xfs_mount *mp,
801 struct xfs_trans_resv *resp)
802 {
803 /*
804 * The following transactions are logged in physical format and
805 * require a permanent reservation on space.
806 */
807 resp->tr_write.tr_logres = xfs_calc_write_reservation(mp);
808 if (xfs_sb_version_hasreflink(&mp->m_sb))
809 resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
810 else
811 resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
812 resp->tr_write.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
813
814 resp->tr_itruncate.tr_logres = xfs_calc_itruncate_reservation(mp);
815 if (xfs_sb_version_hasreflink(&mp->m_sb))
816 resp->tr_itruncate.tr_logcount =
817 XFS_ITRUNCATE_LOG_COUNT_REFLINK;
818 else
819 resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
820 resp->tr_itruncate.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
821
822 resp->tr_rename.tr_logres = xfs_calc_rename_reservation(mp);
823 resp->tr_rename.tr_logcount = XFS_RENAME_LOG_COUNT;
824 resp->tr_rename.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
825
826 resp->tr_link.tr_logres = xfs_calc_link_reservation(mp);
827 resp->tr_link.tr_logcount = XFS_LINK_LOG_COUNT;
828 resp->tr_link.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
829
830 resp->tr_remove.tr_logres = xfs_calc_remove_reservation(mp);
831 resp->tr_remove.tr_logcount = XFS_REMOVE_LOG_COUNT;
832 resp->tr_remove.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
833
834 resp->tr_symlink.tr_logres = xfs_calc_symlink_reservation(mp);
835 resp->tr_symlink.tr_logcount = XFS_SYMLINK_LOG_COUNT;
836 resp->tr_symlink.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
837
838 resp->tr_create.tr_logres = xfs_calc_create_reservation(mp);
839 resp->tr_create.tr_logcount = XFS_CREATE_LOG_COUNT;
840 resp->tr_create.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
841
842 resp->tr_create_tmpfile.tr_logres =
843 xfs_calc_create_tmpfile_reservation(mp);
844 resp->tr_create_tmpfile.tr_logcount = XFS_CREATE_TMPFILE_LOG_COUNT;
845 resp->tr_create_tmpfile.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
846
847 resp->tr_mkdir.tr_logres = xfs_calc_mkdir_reservation(mp);
848 resp->tr_mkdir.tr_logcount = XFS_MKDIR_LOG_COUNT;
849 resp->tr_mkdir.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
850
851 resp->tr_ifree.tr_logres = xfs_calc_ifree_reservation(mp);
852 resp->tr_ifree.tr_logcount = XFS_INACTIVE_LOG_COUNT;
853 resp->tr_ifree.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
854
855 resp->tr_addafork.tr_logres = xfs_calc_addafork_reservation(mp);
856 resp->tr_addafork.tr_logcount = XFS_ADDAFORK_LOG_COUNT;
857 resp->tr_addafork.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
858
859 resp->tr_attrinval.tr_logres = xfs_calc_attrinval_reservation(mp);
860 resp->tr_attrinval.tr_logcount = XFS_ATTRINVAL_LOG_COUNT;
861 resp->tr_attrinval.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
862
863 resp->tr_attrsetm.tr_logres = xfs_calc_attrsetm_reservation(mp);
864 resp->tr_attrsetm.tr_logcount = XFS_ATTRSET_LOG_COUNT;
865 resp->tr_attrsetm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
866
867 resp->tr_attrrm.tr_logres = xfs_calc_attrrm_reservation(mp);
868 resp->tr_attrrm.tr_logcount = XFS_ATTRRM_LOG_COUNT;
869 resp->tr_attrrm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
870
871 resp->tr_growrtalloc.tr_logres = xfs_calc_growrtalloc_reservation(mp);
872 resp->tr_growrtalloc.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
873 resp->tr_growrtalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
874
875 resp->tr_qm_dqalloc.tr_logres = xfs_calc_qm_dqalloc_reservation(mp);
876 if (xfs_sb_version_hasreflink(&mp->m_sb))
877 resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
878 else
879 resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
880 resp->tr_qm_dqalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
881
882 /*
883 * The following transactions are logged in logical format with
884 * a default log count.
885 */
886 resp->tr_qm_setqlim.tr_logres = xfs_calc_qm_setqlim_reservation(mp);
887 resp->tr_qm_setqlim.tr_logcount = XFS_DEFAULT_LOG_COUNT;
888
889 resp->tr_qm_quotaoff.tr_logres = xfs_calc_qm_quotaoff_reservation(mp);
890 resp->tr_qm_quotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT;
891
892 resp->tr_qm_equotaoff.tr_logres =
893 xfs_calc_qm_quotaoff_end_reservation(mp);
894 resp->tr_qm_equotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT;
895
896 resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp);
897 resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT;
898
899 /* The following transaction are logged in logical format */
900 resp->tr_ichange.tr_logres = xfs_calc_ichange_reservation(mp);
901 resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp);
902 resp->tr_fsyncts.tr_logres = xfs_calc_swrite_reservation(mp);
903 resp->tr_writeid.tr_logres = xfs_calc_writeid_reservation(mp);
904 resp->tr_attrsetrt.tr_logres = xfs_calc_attrsetrt_reservation(mp);
905 resp->tr_clearagi.tr_logres = xfs_calc_clear_agi_bucket_reservation(mp);
906 resp->tr_growrtzero.tr_logres = xfs_calc_growrtzero_reservation(mp);
907 resp->tr_growrtfree.tr_logres = xfs_calc_growrtfree_reservation(mp);
908 }
909