1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (C) 2010 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_inode.h"
17 #include "xfs_bmap_btree.h"
18 #include "xfs_quota.h"
19 #include "xfs_trans.h"
20 #include "xfs_qm.h"
21 #include "xfs_trans_space.h"
22
23 #define _ALLOC true
24 #define _FREE false
25
26 /*
27 * A buffer has a format structure overhead in the log in addition
28 * to the data, so we need to take this into account when reserving
29 * space in a transaction for a buffer. Round the space required up
30 * to a multiple of 128 bytes so that we don't change the historical
31 * reservation that has been used for this overhead.
32 */
33 STATIC uint
xfs_buf_log_overhead(void)34 xfs_buf_log_overhead(void)
35 {
36 return round_up(sizeof(struct xlog_op_header) +
37 sizeof(struct xfs_buf_log_format), 128);
38 }
39
40 /*
41 * Calculate out transaction log reservation per item in bytes.
42 *
43 * The nbufs argument is used to indicate the number of items that
44 * will be changed in a transaction. size is used to tell how many
45 * bytes should be reserved per item.
46 */
47 STATIC uint
xfs_calc_buf_res(uint nbufs,uint size)48 xfs_calc_buf_res(
49 uint nbufs,
50 uint size)
51 {
52 return nbufs * (size + xfs_buf_log_overhead());
53 }
54
55 /*
56 * Per-extent log reservation for the btree changes involved in freeing or
57 * allocating an extent. In classic XFS there were two trees that will be
58 * modified (bnobt + cntbt). With rmap enabled, there are three trees
59 * (rmapbt). The number of blocks reserved is based on the formula:
60 *
61 * num trees * ((2 blocks/level * max depth) - 1)
62 *
63 * Keep in mind that max depth is calculated separately for each type of tree.
64 */
65 uint
xfs_allocfree_block_count(struct xfs_mount * mp,uint num_ops)66 xfs_allocfree_block_count(
67 struct xfs_mount *mp,
68 uint num_ops)
69 {
70 uint blocks;
71
72 blocks = num_ops * 2 * (2 * mp->m_alloc_maxlevels - 1);
73 if (xfs_has_rmapbt(mp))
74 blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1);
75
76 return blocks;
77 }
78
79 /*
80 * Per-extent log reservation for refcount btree changes. These are never done
81 * in the same transaction as an allocation or a free, so we compute them
82 * separately.
83 */
84 static unsigned int
xfs_refcountbt_block_count(struct xfs_mount * mp,unsigned int num_ops)85 xfs_refcountbt_block_count(
86 struct xfs_mount *mp,
87 unsigned int num_ops)
88 {
89 return num_ops * (2 * mp->m_refc_maxlevels - 1);
90 }
91
92 /*
93 * Logging inodes is really tricksy. They are logged in memory format,
94 * which means that what we write into the log doesn't directly translate into
95 * the amount of space they use on disk.
96 *
97 * Case in point - btree format forks in memory format use more space than the
98 * on-disk format. In memory, the buffer contains a normal btree block header so
99 * the btree code can treat it as though it is just another generic buffer.
100 * However, when we write it to the inode fork, we don't write all of this
101 * header as it isn't needed. e.g. the root is only ever in the inode, so
102 * there's no need for sibling pointers which would waste 16 bytes of space.
103 *
104 * Hence when we have an inode with a maximally sized btree format fork, then
105 * amount of information we actually log is greater than the size of the inode
106 * on disk. Hence we need an inode reservation function that calculates all this
107 * correctly. So, we log:
108 *
109 * - 4 log op headers for object
110 * - for the ilf, the inode core and 2 forks
111 * - inode log format object
112 * - the inode core
113 * - two inode forks containing bmap btree root blocks.
114 * - the btree data contained by both forks will fit into the inode size,
115 * hence when combined with the inode core above, we have a total of the
116 * actual inode size.
117 * - the BMBT headers need to be accounted separately, as they are
118 * additional to the records and pointers that fit inside the inode
119 * forks.
120 */
121 STATIC uint
xfs_calc_inode_res(struct xfs_mount * mp,uint ninodes)122 xfs_calc_inode_res(
123 struct xfs_mount *mp,
124 uint ninodes)
125 {
126 return ninodes *
127 (4 * sizeof(struct xlog_op_header) +
128 sizeof(struct xfs_inode_log_format) +
129 mp->m_sb.sb_inodesize +
130 2 * XFS_BMBT_BLOCK_LEN(mp));
131 }
132
133 /*
134 * Inode btree record insertion/removal modifies the inode btree and free space
135 * btrees (since the inobt does not use the agfl). This requires the following
136 * reservation:
137 *
138 * the inode btree: max depth * blocksize
139 * the allocation btrees: 2 trees * (max depth - 1) * block size
140 *
141 * The caller must account for SB and AG header modifications, etc.
142 */
143 STATIC uint
xfs_calc_inobt_res(struct xfs_mount * mp)144 xfs_calc_inobt_res(
145 struct xfs_mount *mp)
146 {
147 return xfs_calc_buf_res(M_IGEO(mp)->inobt_maxlevels,
148 XFS_FSB_TO_B(mp, 1)) +
149 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
150 XFS_FSB_TO_B(mp, 1));
151 }
152
153 /*
154 * The free inode btree is a conditional feature. The behavior differs slightly
155 * from that of the traditional inode btree in that the finobt tracks records
156 * for inode chunks with at least one free inode. A record can be removed from
157 * the tree during individual inode allocation. Therefore the finobt
158 * reservation is unconditional for both the inode chunk allocation and
159 * individual inode allocation (modify) cases.
160 *
161 * Behavior aside, the reservation for finobt modification is equivalent to the
162 * traditional inobt: cover a full finobt shape change plus block allocation.
163 */
164 STATIC uint
xfs_calc_finobt_res(struct xfs_mount * mp)165 xfs_calc_finobt_res(
166 struct xfs_mount *mp)
167 {
168 if (!xfs_has_finobt(mp))
169 return 0;
170
171 return xfs_calc_inobt_res(mp);
172 }
173
174 /*
175 * Calculate the reservation required to allocate or free an inode chunk. This
176 * includes:
177 *
178 * the allocation btrees: 2 trees * (max depth - 1) * block size
179 * the inode chunk: m_ino_geo.ialloc_blks * N
180 *
181 * The size N of the inode chunk reservation depends on whether it is for
182 * allocation or free and which type of create transaction is in use. An inode
183 * chunk free always invalidates the buffers and only requires reservation for
184 * headers (N == 0). An inode chunk allocation requires a chunk sized
185 * reservation on v4 and older superblocks to initialize the chunk. No chunk
186 * reservation is required for allocation on v5 supers, which use ordered
187 * buffers to initialize.
188 */
189 STATIC uint
xfs_calc_inode_chunk_res(struct xfs_mount * mp,bool alloc)190 xfs_calc_inode_chunk_res(
191 struct xfs_mount *mp,
192 bool alloc)
193 {
194 uint res, size = 0;
195
196 res = xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
197 XFS_FSB_TO_B(mp, 1));
198 if (alloc) {
199 /* icreate tx uses ordered buffers */
200 if (xfs_has_v3inodes(mp))
201 return res;
202 size = XFS_FSB_TO_B(mp, 1);
203 }
204
205 res += xfs_calc_buf_res(M_IGEO(mp)->ialloc_blks, size);
206 return res;
207 }
208
209 /*
210 * Per-extent log reservation for the btree changes involved in freeing or
211 * allocating a realtime extent. We have to be able to log as many rtbitmap
212 * blocks as needed to mark inuse XFS_BMBT_MAX_EXTLEN blocks' worth of realtime
213 * extents, as well as the realtime summary block.
214 */
215 static unsigned int
xfs_rtalloc_block_count(struct xfs_mount * mp,unsigned int num_ops)216 xfs_rtalloc_block_count(
217 struct xfs_mount *mp,
218 unsigned int num_ops)
219 {
220 unsigned int blksz = XFS_FSB_TO_B(mp, 1);
221 unsigned int rtbmp_bytes;
222
223 rtbmp_bytes = (XFS_MAX_BMBT_EXTLEN / mp->m_sb.sb_rextsize) / NBBY;
224 return (howmany(rtbmp_bytes, blksz) + 1) * num_ops;
225 }
226
227 /*
228 * Various log reservation values.
229 *
230 * These are based on the size of the file system block because that is what
231 * most transactions manipulate. Each adds in an additional 128 bytes per
232 * item logged to try to account for the overhead of the transaction mechanism.
233 *
234 * Note: Most of the reservations underestimate the number of allocation
235 * groups into which they could free extents in the xfs_defer_finish() call.
236 * This is because the number in the worst case is quite high and quite
237 * unusual. In order to fix this we need to change xfs_defer_finish() to free
238 * extents in only a single AG at a time. This will require changes to the
239 * EFI code as well, however, so that the EFI for the extents not freed is
240 * logged again in each transaction. See SGI PV #261917.
241 *
242 * Reservation functions here avoid a huge stack in xfs_trans_init due to
243 * register overflow from temporaries in the calculations.
244 */
245
246 /*
247 * Compute the log reservation required to handle the refcount update
248 * transaction. Refcount updates are always done via deferred log items.
249 *
250 * This is calculated as:
251 * Data device refcount updates (t1):
252 * the agfs of the ags containing the blocks: nr_ops * sector size
253 * the refcount btrees: nr_ops * 1 trees * (2 * max depth - 1) * block size
254 */
255 static unsigned int
xfs_calc_refcountbt_reservation(struct xfs_mount * mp,unsigned int nr_ops)256 xfs_calc_refcountbt_reservation(
257 struct xfs_mount *mp,
258 unsigned int nr_ops)
259 {
260 unsigned int blksz = XFS_FSB_TO_B(mp, 1);
261
262 if (!xfs_has_reflink(mp))
263 return 0;
264
265 return xfs_calc_buf_res(nr_ops, mp->m_sb.sb_sectsize) +
266 xfs_calc_buf_res(xfs_refcountbt_block_count(mp, nr_ops), blksz);
267 }
268
269 /*
270 * In a write transaction we can allocate a maximum of 2
271 * extents. This gives (t1):
272 * the inode getting the new extents: inode size
273 * the inode's bmap btree: max depth * block size
274 * the agfs of the ags from which the extents are allocated: 2 * sector
275 * the superblock free block counter: sector size
276 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
277 * Or, if we're writing to a realtime file (t2):
278 * the inode getting the new extents: inode size
279 * the inode's bmap btree: max depth * block size
280 * the agfs of the ags from which the extents are allocated: 2 * sector
281 * the superblock free block counter: sector size
282 * the realtime bitmap: ((XFS_BMBT_MAX_EXTLEN / rtextsize) / NBBY) bytes
283 * the realtime summary: 1 block
284 * the allocation btrees: 2 trees * (2 * max depth - 1) * block size
285 * And the bmap_finish transaction can free bmap blocks in a join (t3):
286 * the agfs of the ags containing the blocks: 2 * sector size
287 * the agfls of the ags containing the blocks: 2 * sector size
288 * the super block free block counter: sector size
289 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
290 * And any refcount updates that happen in a separate transaction (t4).
291 */
292 STATIC uint
xfs_calc_write_reservation(struct xfs_mount * mp,bool for_minlogsize)293 xfs_calc_write_reservation(
294 struct xfs_mount *mp,
295 bool for_minlogsize)
296 {
297 unsigned int t1, t2, t3, t4;
298 unsigned int blksz = XFS_FSB_TO_B(mp, 1);
299
300 t1 = xfs_calc_inode_res(mp, 1) +
301 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), blksz) +
302 xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
303 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 2), blksz);
304
305 if (xfs_has_realtime(mp)) {
306 t2 = xfs_calc_inode_res(mp, 1) +
307 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
308 blksz) +
309 xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
310 xfs_calc_buf_res(xfs_rtalloc_block_count(mp, 1), blksz) +
311 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1), blksz);
312 } else {
313 t2 = 0;
314 }
315
316 t3 = xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
317 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 2), blksz);
318
319 /*
320 * In the early days of reflink, we included enough reservation to log
321 * two refcountbt splits for each transaction. The codebase runs
322 * refcountbt updates in separate transactions now, so to compute the
323 * minimum log size, add the refcountbtree splits back to t1 and t3 and
324 * do not account them separately as t4. Reflink did not support
325 * realtime when the reservations were established, so no adjustment to
326 * t2 is needed.
327 */
328 if (for_minlogsize) {
329 unsigned int adj = 0;
330
331 if (xfs_has_reflink(mp))
332 adj = xfs_calc_buf_res(
333 xfs_refcountbt_block_count(mp, 2),
334 blksz);
335 t1 += adj;
336 t3 += adj;
337 return XFS_DQUOT_LOGRES(mp) + max3(t1, t2, t3);
338 }
339
340 t4 = xfs_calc_refcountbt_reservation(mp, 1);
341 return XFS_DQUOT_LOGRES(mp) + max(t4, max3(t1, t2, t3));
342 }
343
344 unsigned int
xfs_calc_write_reservation_minlogsize(struct xfs_mount * mp)345 xfs_calc_write_reservation_minlogsize(
346 struct xfs_mount *mp)
347 {
348 return xfs_calc_write_reservation(mp, true);
349 }
350
351 /*
352 * In truncating a file we free up to two extents at once. We can modify (t1):
353 * the inode being truncated: inode size
354 * the inode's bmap btree: (max depth + 1) * block size
355 * And the bmap_finish transaction can free the blocks and bmap blocks (t2):
356 * the agf for each of the ags: 4 * sector size
357 * the agfl for each of the ags: 4 * sector size
358 * the super block to reflect the freed blocks: sector size
359 * worst case split in allocation btrees per extent assuming 4 extents:
360 * 4 exts * 2 trees * (2 * max depth - 1) * block size
361 * Or, if it's a realtime file (t3):
362 * the agf for each of the ags: 2 * sector size
363 * the agfl for each of the ags: 2 * sector size
364 * the super block to reflect the freed blocks: sector size
365 * the realtime bitmap:
366 * 2 exts * ((XFS_BMBT_MAX_EXTLEN / rtextsize) / NBBY) bytes
367 * the realtime summary: 2 exts * 1 block
368 * worst case split in allocation btrees per extent assuming 2 extents:
369 * 2 exts * 2 trees * (2 * max depth - 1) * block size
370 * And any refcount updates that happen in a separate transaction (t4).
371 */
372 STATIC uint
xfs_calc_itruncate_reservation(struct xfs_mount * mp,bool for_minlogsize)373 xfs_calc_itruncate_reservation(
374 struct xfs_mount *mp,
375 bool for_minlogsize)
376 {
377 unsigned int t1, t2, t3, t4;
378 unsigned int blksz = XFS_FSB_TO_B(mp, 1);
379
380 t1 = xfs_calc_inode_res(mp, 1) +
381 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1, blksz);
382
383 t2 = xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
384 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 4), blksz);
385
386 if (xfs_has_realtime(mp)) {
387 t3 = xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
388 xfs_calc_buf_res(xfs_rtalloc_block_count(mp, 2), blksz) +
389 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 2), blksz);
390 } else {
391 t3 = 0;
392 }
393
394 /*
395 * In the early days of reflink, we included enough reservation to log
396 * four refcountbt splits in the same transaction as bnobt/cntbt
397 * updates. The codebase runs refcountbt updates in separate
398 * transactions now, so to compute the minimum log size, add the
399 * refcount btree splits back here and do not compute them separately
400 * as t4. Reflink did not support realtime when the reservations were
401 * established, so do not adjust t3.
402 */
403 if (for_minlogsize) {
404 if (xfs_has_reflink(mp))
405 t2 += xfs_calc_buf_res(
406 xfs_refcountbt_block_count(mp, 4),
407 blksz);
408
409 return XFS_DQUOT_LOGRES(mp) + max3(t1, t2, t3);
410 }
411
412 t4 = xfs_calc_refcountbt_reservation(mp, 2);
413 return XFS_DQUOT_LOGRES(mp) + max(t4, max3(t1, t2, t3));
414 }
415
416 unsigned int
xfs_calc_itruncate_reservation_minlogsize(struct xfs_mount * mp)417 xfs_calc_itruncate_reservation_minlogsize(
418 struct xfs_mount *mp)
419 {
420 return xfs_calc_itruncate_reservation(mp, true);
421 }
422
423 /*
424 * In renaming a files we can modify:
425 * the five inodes involved: 5 * inode size
426 * the two directory btrees: 2 * (max depth + v2) * dir block size
427 * the two directory bmap btrees: 2 * max depth * block size
428 * And the bmap_finish transaction can free dir and bmap blocks (two sets
429 * of bmap blocks) giving:
430 * the agf for the ags in which the blocks live: 3 * sector size
431 * the agfl for the ags in which the blocks live: 3 * sector size
432 * the superblock for the free block count: sector size
433 * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
434 */
435 STATIC uint
xfs_calc_rename_reservation(struct xfs_mount * mp)436 xfs_calc_rename_reservation(
437 struct xfs_mount *mp)
438 {
439 return XFS_DQUOT_LOGRES(mp) +
440 max((xfs_calc_inode_res(mp, 5) +
441 xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp),
442 XFS_FSB_TO_B(mp, 1))),
443 (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) +
444 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 3),
445 XFS_FSB_TO_B(mp, 1))));
446 }
447
448 /*
449 * For removing an inode from unlinked list at first, we can modify:
450 * the agi hash list and counters: sector size
451 * the on disk inode before ours in the agi hash list: inode cluster size
452 * the on disk inode in the agi hash list: inode cluster size
453 */
454 STATIC uint
xfs_calc_iunlink_remove_reservation(struct xfs_mount * mp)455 xfs_calc_iunlink_remove_reservation(
456 struct xfs_mount *mp)
457 {
458 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
459 2 * M_IGEO(mp)->inode_cluster_size;
460 }
461
462 /*
463 * For creating a link to an inode:
464 * the parent directory inode: inode size
465 * the linked inode: inode size
466 * the directory btree could split: (max depth + v2) * dir block size
467 * the directory bmap btree could join or split: (max depth + v2) * blocksize
468 * And the bmap_finish transaction can free some bmap blocks giving:
469 * the agf for the ag in which the blocks live: sector size
470 * the agfl for the ag in which the blocks live: sector size
471 * the superblock for the free block count: sector size
472 * the allocation btrees: 2 trees * (2 * max depth - 1) * block size
473 */
474 STATIC uint
xfs_calc_link_reservation(struct xfs_mount * mp)475 xfs_calc_link_reservation(
476 struct xfs_mount *mp)
477 {
478 return XFS_DQUOT_LOGRES(mp) +
479 xfs_calc_iunlink_remove_reservation(mp) +
480 max((xfs_calc_inode_res(mp, 2) +
481 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
482 XFS_FSB_TO_B(mp, 1))),
483 (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
484 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
485 XFS_FSB_TO_B(mp, 1))));
486 }
487
488 /*
489 * For adding an inode to unlinked list we can modify:
490 * the agi hash list: sector size
491 * the on disk inode: inode cluster size
492 */
493 STATIC uint
xfs_calc_iunlink_add_reservation(xfs_mount_t * mp)494 xfs_calc_iunlink_add_reservation(xfs_mount_t *mp)
495 {
496 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
497 M_IGEO(mp)->inode_cluster_size;
498 }
499
500 /*
501 * For removing a directory entry we can modify:
502 * the parent directory inode: inode size
503 * the removed inode: inode size
504 * the directory btree could join: (max depth + v2) * dir block size
505 * the directory bmap btree could join or split: (max depth + v2) * blocksize
506 * And the bmap_finish transaction can free the dir and bmap blocks giving:
507 * the agf for the ag in which the blocks live: 2 * sector size
508 * the agfl for the ag in which the blocks live: 2 * sector size
509 * the superblock for the free block count: sector size
510 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
511 */
512 STATIC uint
xfs_calc_remove_reservation(struct xfs_mount * mp)513 xfs_calc_remove_reservation(
514 struct xfs_mount *mp)
515 {
516 return XFS_DQUOT_LOGRES(mp) +
517 xfs_calc_iunlink_add_reservation(mp) +
518 max((xfs_calc_inode_res(mp, 2) +
519 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
520 XFS_FSB_TO_B(mp, 1))),
521 (xfs_calc_buf_res(4, mp->m_sb.sb_sectsize) +
522 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 2),
523 XFS_FSB_TO_B(mp, 1))));
524 }
525
526 /*
527 * For create, break it in to the two cases that the transaction
528 * covers. We start with the modify case - allocation done by modification
529 * of the state of existing inodes - and the allocation case.
530 */
531
532 /*
533 * For create we can modify:
534 * the parent directory inode: inode size
535 * the new inode: inode size
536 * the inode btree entry: block size
537 * the superblock for the nlink flag: sector size
538 * the directory btree: (max depth + v2) * dir block size
539 * the directory inode's bmap btree: (max depth + v2) * block size
540 * the finobt (record modification and allocation btrees)
541 */
542 STATIC uint
xfs_calc_create_resv_modify(struct xfs_mount * mp)543 xfs_calc_create_resv_modify(
544 struct xfs_mount *mp)
545 {
546 return xfs_calc_inode_res(mp, 2) +
547 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
548 (uint)XFS_FSB_TO_B(mp, 1) +
549 xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), XFS_FSB_TO_B(mp, 1)) +
550 xfs_calc_finobt_res(mp);
551 }
552
553 /*
554 * For icreate we can allocate some inodes giving:
555 * the agi and agf of the ag getting the new inodes: 2 * sectorsize
556 * the superblock for the nlink flag: sector size
557 * the inode chunk (allocation, optional init)
558 * the inobt (record insertion)
559 * the finobt (optional, record insertion)
560 */
561 STATIC uint
xfs_calc_icreate_resv_alloc(struct xfs_mount * mp)562 xfs_calc_icreate_resv_alloc(
563 struct xfs_mount *mp)
564 {
565 return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
566 mp->m_sb.sb_sectsize +
567 xfs_calc_inode_chunk_res(mp, _ALLOC) +
568 xfs_calc_inobt_res(mp) +
569 xfs_calc_finobt_res(mp);
570 }
571
572 STATIC uint
xfs_calc_icreate_reservation(xfs_mount_t * mp)573 xfs_calc_icreate_reservation(xfs_mount_t *mp)
574 {
575 return XFS_DQUOT_LOGRES(mp) +
576 max(xfs_calc_icreate_resv_alloc(mp),
577 xfs_calc_create_resv_modify(mp));
578 }
579
580 STATIC uint
xfs_calc_create_tmpfile_reservation(struct xfs_mount * mp)581 xfs_calc_create_tmpfile_reservation(
582 struct xfs_mount *mp)
583 {
584 uint res = XFS_DQUOT_LOGRES(mp);
585
586 res += xfs_calc_icreate_resv_alloc(mp);
587 return res + xfs_calc_iunlink_add_reservation(mp);
588 }
589
590 /*
591 * Making a new directory is the same as creating a new file.
592 */
593 STATIC uint
xfs_calc_mkdir_reservation(struct xfs_mount * mp)594 xfs_calc_mkdir_reservation(
595 struct xfs_mount *mp)
596 {
597 return xfs_calc_icreate_reservation(mp);
598 }
599
600
601 /*
602 * Making a new symplink is the same as creating a new file, but
603 * with the added blocks for remote symlink data which can be up to 1kB in
604 * length (XFS_SYMLINK_MAXLEN).
605 */
606 STATIC uint
xfs_calc_symlink_reservation(struct xfs_mount * mp)607 xfs_calc_symlink_reservation(
608 struct xfs_mount *mp)
609 {
610 return xfs_calc_icreate_reservation(mp) +
611 xfs_calc_buf_res(1, XFS_SYMLINK_MAXLEN);
612 }
613
614 /*
615 * In freeing an inode we can modify:
616 * the inode being freed: inode size
617 * the super block free inode counter, AGF and AGFL: sector size
618 * the on disk inode (agi unlinked list removal)
619 * the inode chunk (invalidated, headers only)
620 * the inode btree
621 * the finobt (record insertion, removal or modification)
622 *
623 * Note that the inode chunk res. includes an allocfree res. for freeing of the
624 * inode chunk. This is technically extraneous because the inode chunk free is
625 * deferred (it occurs after a transaction roll). Include the extra reservation
626 * anyways since we've had reports of ifree transaction overruns due to too many
627 * agfl fixups during inode chunk frees.
628 */
629 STATIC uint
xfs_calc_ifree_reservation(struct xfs_mount * mp)630 xfs_calc_ifree_reservation(
631 struct xfs_mount *mp)
632 {
633 return XFS_DQUOT_LOGRES(mp) +
634 xfs_calc_inode_res(mp, 1) +
635 xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
636 xfs_calc_iunlink_remove_reservation(mp) +
637 xfs_calc_inode_chunk_res(mp, _FREE) +
638 xfs_calc_inobt_res(mp) +
639 xfs_calc_finobt_res(mp);
640 }
641
642 /*
643 * When only changing the inode we log the inode and possibly the superblock
644 * We also add a bit of slop for the transaction stuff.
645 */
646 STATIC uint
xfs_calc_ichange_reservation(struct xfs_mount * mp)647 xfs_calc_ichange_reservation(
648 struct xfs_mount *mp)
649 {
650 return XFS_DQUOT_LOGRES(mp) +
651 xfs_calc_inode_res(mp, 1) +
652 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
653
654 }
655
656 /*
657 * Growing the data section of the filesystem.
658 * superblock
659 * agi and agf
660 * allocation btrees
661 */
662 STATIC uint
xfs_calc_growdata_reservation(struct xfs_mount * mp)663 xfs_calc_growdata_reservation(
664 struct xfs_mount *mp)
665 {
666 return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
667 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
668 XFS_FSB_TO_B(mp, 1));
669 }
670
671 /*
672 * Growing the rt section of the filesystem.
673 * In the first set of transactions (ALLOC) we allocate space to the
674 * bitmap or summary files.
675 * superblock: sector size
676 * agf of the ag from which the extent is allocated: sector size
677 * bmap btree for bitmap/summary inode: max depth * blocksize
678 * bitmap/summary inode: inode size
679 * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
680 */
681 STATIC uint
xfs_calc_growrtalloc_reservation(struct xfs_mount * mp)682 xfs_calc_growrtalloc_reservation(
683 struct xfs_mount *mp)
684 {
685 return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
686 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
687 XFS_FSB_TO_B(mp, 1)) +
688 xfs_calc_inode_res(mp, 1) +
689 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
690 XFS_FSB_TO_B(mp, 1));
691 }
692
693 /*
694 * Growing the rt section of the filesystem.
695 * In the second set of transactions (ZERO) we zero the new metadata blocks.
696 * one bitmap/summary block: blocksize
697 */
698 STATIC uint
xfs_calc_growrtzero_reservation(struct xfs_mount * mp)699 xfs_calc_growrtzero_reservation(
700 struct xfs_mount *mp)
701 {
702 return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize);
703 }
704
705 /*
706 * Growing the rt section of the filesystem.
707 * In the third set of transactions (FREE) we update metadata without
708 * allocating any new blocks.
709 * superblock: sector size
710 * bitmap inode: inode size
711 * summary inode: inode size
712 * one bitmap block: blocksize
713 * summary blocks: new summary size
714 */
715 STATIC uint
xfs_calc_growrtfree_reservation(struct xfs_mount * mp)716 xfs_calc_growrtfree_reservation(
717 struct xfs_mount *mp)
718 {
719 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
720 xfs_calc_inode_res(mp, 2) +
721 xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) +
722 xfs_calc_buf_res(1, mp->m_rsumsize);
723 }
724
725 /*
726 * Logging the inode modification timestamp on a synchronous write.
727 * inode
728 */
729 STATIC uint
xfs_calc_swrite_reservation(struct xfs_mount * mp)730 xfs_calc_swrite_reservation(
731 struct xfs_mount *mp)
732 {
733 return xfs_calc_inode_res(mp, 1);
734 }
735
736 /*
737 * Logging the inode mode bits when writing a setuid/setgid file
738 * inode
739 */
740 STATIC uint
xfs_calc_writeid_reservation(struct xfs_mount * mp)741 xfs_calc_writeid_reservation(
742 struct xfs_mount *mp)
743 {
744 return xfs_calc_inode_res(mp, 1);
745 }
746
747 /*
748 * Converting the inode from non-attributed to attributed.
749 * the inode being converted: inode size
750 * agf block and superblock (for block allocation)
751 * the new block (directory sized)
752 * bmap blocks for the new directory block
753 * allocation btrees
754 */
755 STATIC uint
xfs_calc_addafork_reservation(struct xfs_mount * mp)756 xfs_calc_addafork_reservation(
757 struct xfs_mount *mp)
758 {
759 return XFS_DQUOT_LOGRES(mp) +
760 xfs_calc_inode_res(mp, 1) +
761 xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
762 xfs_calc_buf_res(1, mp->m_dir_geo->blksize) +
763 xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1,
764 XFS_FSB_TO_B(mp, 1)) +
765 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 1),
766 XFS_FSB_TO_B(mp, 1));
767 }
768
769 /*
770 * Removing the attribute fork of a file
771 * the inode being truncated: inode size
772 * the inode's bmap btree: max depth * block size
773 * And the bmap_finish transaction can free the blocks and bmap blocks:
774 * the agf for each of the ags: 4 * sector size
775 * the agfl for each of the ags: 4 * sector size
776 * the super block to reflect the freed blocks: sector size
777 * worst case split in allocation btrees per extent assuming 4 extents:
778 * 4 exts * 2 trees * (2 * max depth - 1) * block size
779 */
780 STATIC uint
xfs_calc_attrinval_reservation(struct xfs_mount * mp)781 xfs_calc_attrinval_reservation(
782 struct xfs_mount *mp)
783 {
784 return max((xfs_calc_inode_res(mp, 1) +
785 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
786 XFS_FSB_TO_B(mp, 1))),
787 (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
788 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 4),
789 XFS_FSB_TO_B(mp, 1))));
790 }
791
792 /*
793 * Setting an attribute at mount time.
794 * the inode getting the attribute
795 * the superblock for allocations
796 * the agfs extents are allocated from
797 * the attribute btree * max depth
798 * the inode allocation btree
799 * Since attribute transaction space is dependent on the size of the attribute,
800 * the calculation is done partially at mount time and partially at runtime(see
801 * below).
802 */
803 STATIC uint
xfs_calc_attrsetm_reservation(struct xfs_mount * mp)804 xfs_calc_attrsetm_reservation(
805 struct xfs_mount *mp)
806 {
807 return XFS_DQUOT_LOGRES(mp) +
808 xfs_calc_inode_res(mp, 1) +
809 xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
810 xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1));
811 }
812
813 /*
814 * Setting an attribute at runtime, transaction space unit per block.
815 * the superblock for allocations: sector size
816 * the inode bmap btree could join or split: max depth * block size
817 * Since the runtime attribute transaction space is dependent on the total
818 * blocks needed for the 1st bmap, here we calculate out the space unit for
819 * one block so that the caller could figure out the total space according
820 * to the attibute extent length in blocks by:
821 * ext * M_RES(mp)->tr_attrsetrt.tr_logres
822 */
823 STATIC uint
xfs_calc_attrsetrt_reservation(struct xfs_mount * mp)824 xfs_calc_attrsetrt_reservation(
825 struct xfs_mount *mp)
826 {
827 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
828 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
829 XFS_FSB_TO_B(mp, 1));
830 }
831
832 /*
833 * Removing an attribute.
834 * the inode: inode size
835 * the attribute btree could join: max depth * block size
836 * the inode bmap btree could join or split: max depth * block size
837 * And the bmap_finish transaction can free the attr blocks freed giving:
838 * the agf for the ag in which the blocks live: 2 * sector size
839 * the agfl for the ag in which the blocks live: 2 * sector size
840 * the superblock for the free block count: sector size
841 * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
842 */
843 STATIC uint
xfs_calc_attrrm_reservation(struct xfs_mount * mp)844 xfs_calc_attrrm_reservation(
845 struct xfs_mount *mp)
846 {
847 return XFS_DQUOT_LOGRES(mp) +
848 max((xfs_calc_inode_res(mp, 1) +
849 xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH,
850 XFS_FSB_TO_B(mp, 1)) +
851 (uint)XFS_FSB_TO_B(mp,
852 XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
853 xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)),
854 (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
855 xfs_calc_buf_res(xfs_allocfree_block_count(mp, 2),
856 XFS_FSB_TO_B(mp, 1))));
857 }
858
859 /*
860 * Clearing a bad agino number in an agi hash bucket.
861 */
862 STATIC uint
xfs_calc_clear_agi_bucket_reservation(struct xfs_mount * mp)863 xfs_calc_clear_agi_bucket_reservation(
864 struct xfs_mount *mp)
865 {
866 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
867 }
868
869 /*
870 * Adjusting quota limits.
871 * the disk quota buffer: sizeof(struct xfs_disk_dquot)
872 */
873 STATIC uint
xfs_calc_qm_setqlim_reservation(void)874 xfs_calc_qm_setqlim_reservation(void)
875 {
876 return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot));
877 }
878
879 /*
880 * Allocating quota on disk if needed.
881 * the write transaction log space for quota file extent allocation
882 * the unit of quota allocation: one system block size
883 */
884 STATIC uint
xfs_calc_qm_dqalloc_reservation(struct xfs_mount * mp,bool for_minlogsize)885 xfs_calc_qm_dqalloc_reservation(
886 struct xfs_mount *mp,
887 bool for_minlogsize)
888 {
889 return xfs_calc_write_reservation(mp, for_minlogsize) +
890 xfs_calc_buf_res(1,
891 XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1);
892 }
893
894 unsigned int
xfs_calc_qm_dqalloc_reservation_minlogsize(struct xfs_mount * mp)895 xfs_calc_qm_dqalloc_reservation_minlogsize(
896 struct xfs_mount *mp)
897 {
898 return xfs_calc_qm_dqalloc_reservation(mp, true);
899 }
900
901 /*
902 * Syncing the incore super block changes to disk.
903 * the super block to reflect the changes: sector size
904 */
905 STATIC uint
xfs_calc_sb_reservation(struct xfs_mount * mp)906 xfs_calc_sb_reservation(
907 struct xfs_mount *mp)
908 {
909 return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
910 }
911
912 void
xfs_trans_resv_calc(struct xfs_mount * mp,struct xfs_trans_resv * resp)913 xfs_trans_resv_calc(
914 struct xfs_mount *mp,
915 struct xfs_trans_resv *resp)
916 {
917 int logcount_adj = 0;
918
919 /*
920 * The following transactions are logged in physical format and
921 * require a permanent reservation on space.
922 */
923 resp->tr_write.tr_logres = xfs_calc_write_reservation(mp, false);
924 resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
925 resp->tr_write.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
926
927 resp->tr_itruncate.tr_logres = xfs_calc_itruncate_reservation(mp, false);
928 resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
929 resp->tr_itruncate.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
930
931 resp->tr_rename.tr_logres = xfs_calc_rename_reservation(mp);
932 resp->tr_rename.tr_logcount = XFS_RENAME_LOG_COUNT;
933 resp->tr_rename.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
934
935 resp->tr_link.tr_logres = xfs_calc_link_reservation(mp);
936 resp->tr_link.tr_logcount = XFS_LINK_LOG_COUNT;
937 resp->tr_link.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
938
939 resp->tr_remove.tr_logres = xfs_calc_remove_reservation(mp);
940 resp->tr_remove.tr_logcount = XFS_REMOVE_LOG_COUNT;
941 resp->tr_remove.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
942
943 resp->tr_symlink.tr_logres = xfs_calc_symlink_reservation(mp);
944 resp->tr_symlink.tr_logcount = XFS_SYMLINK_LOG_COUNT;
945 resp->tr_symlink.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
946
947 resp->tr_create.tr_logres = xfs_calc_icreate_reservation(mp);
948 resp->tr_create.tr_logcount = XFS_CREATE_LOG_COUNT;
949 resp->tr_create.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
950
951 resp->tr_create_tmpfile.tr_logres =
952 xfs_calc_create_tmpfile_reservation(mp);
953 resp->tr_create_tmpfile.tr_logcount = XFS_CREATE_TMPFILE_LOG_COUNT;
954 resp->tr_create_tmpfile.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
955
956 resp->tr_mkdir.tr_logres = xfs_calc_mkdir_reservation(mp);
957 resp->tr_mkdir.tr_logcount = XFS_MKDIR_LOG_COUNT;
958 resp->tr_mkdir.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
959
960 resp->tr_ifree.tr_logres = xfs_calc_ifree_reservation(mp);
961 resp->tr_ifree.tr_logcount = XFS_INACTIVE_LOG_COUNT;
962 resp->tr_ifree.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
963
964 resp->tr_addafork.tr_logres = xfs_calc_addafork_reservation(mp);
965 resp->tr_addafork.tr_logcount = XFS_ADDAFORK_LOG_COUNT;
966 resp->tr_addafork.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
967
968 resp->tr_attrinval.tr_logres = xfs_calc_attrinval_reservation(mp);
969 resp->tr_attrinval.tr_logcount = XFS_ATTRINVAL_LOG_COUNT;
970 resp->tr_attrinval.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
971
972 resp->tr_attrsetm.tr_logres = xfs_calc_attrsetm_reservation(mp);
973 resp->tr_attrsetm.tr_logcount = XFS_ATTRSET_LOG_COUNT;
974 resp->tr_attrsetm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
975
976 resp->tr_attrrm.tr_logres = xfs_calc_attrrm_reservation(mp);
977 resp->tr_attrrm.tr_logcount = XFS_ATTRRM_LOG_COUNT;
978 resp->tr_attrrm.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
979
980 resp->tr_growrtalloc.tr_logres = xfs_calc_growrtalloc_reservation(mp);
981 resp->tr_growrtalloc.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
982 resp->tr_growrtalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
983
984 resp->tr_qm_dqalloc.tr_logres = xfs_calc_qm_dqalloc_reservation(mp,
985 false);
986 resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
987 resp->tr_qm_dqalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
988
989 /*
990 * The following transactions are logged in logical format with
991 * a default log count.
992 */
993 resp->tr_qm_setqlim.tr_logres = xfs_calc_qm_setqlim_reservation();
994 resp->tr_qm_setqlim.tr_logcount = XFS_DEFAULT_LOG_COUNT;
995
996 resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp);
997 resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT;
998
999 /* growdata requires permanent res; it can free space to the last AG */
1000 resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp);
1001 resp->tr_growdata.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
1002 resp->tr_growdata.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
1003
1004 /* The following transaction are logged in logical format */
1005 resp->tr_ichange.tr_logres = xfs_calc_ichange_reservation(mp);
1006 resp->tr_fsyncts.tr_logres = xfs_calc_swrite_reservation(mp);
1007 resp->tr_writeid.tr_logres = xfs_calc_writeid_reservation(mp);
1008 resp->tr_attrsetrt.tr_logres = xfs_calc_attrsetrt_reservation(mp);
1009 resp->tr_clearagi.tr_logres = xfs_calc_clear_agi_bucket_reservation(mp);
1010 resp->tr_growrtzero.tr_logres = xfs_calc_growrtzero_reservation(mp);
1011 resp->tr_growrtfree.tr_logres = xfs_calc_growrtfree_reservation(mp);
1012
1013 /*
1014 * Add one logcount for BUI items that appear with rmap or reflink,
1015 * one logcount for refcount intent items, and one logcount for rmap
1016 * intent items.
1017 */
1018 if (xfs_has_reflink(mp) || xfs_has_rmapbt(mp))
1019 logcount_adj++;
1020 if (xfs_has_reflink(mp))
1021 logcount_adj++;
1022 if (xfs_has_rmapbt(mp))
1023 logcount_adj++;
1024
1025 resp->tr_itruncate.tr_logcount += logcount_adj;
1026 resp->tr_write.tr_logcount += logcount_adj;
1027 resp->tr_qm_dqalloc.tr_logcount += logcount_adj;
1028 }
1029