]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
7fd36c44 DC |
2 | /* |
3 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. | |
4 | * Copyright (C) 2010 Red Hat, Inc. | |
5 | * All Rights Reserved. | |
7fd36c44 DC |
6 | */ |
7 | #include "xfs.h" | |
8 | #include "xfs_fs.h" | |
70a9883c | 9 | #include "xfs_shared.h" |
239880ef DC |
10 | #include "xfs_format.h" |
11 | #include "xfs_log_format.h" | |
7fd36c44 | 12 | #include "xfs_trans_resv.h" |
7fd36c44 | 13 | #include "xfs_mount.h" |
57062787 | 14 | #include "xfs_da_format.h" |
d6cf1305 | 15 | #include "xfs_da_btree.h" |
7fd36c44 | 16 | #include "xfs_inode.h" |
a4fbe6ab | 17 | #include "xfs_bmap_btree.h" |
7fd36c44 | 18 | #include "xfs_ialloc.h" |
7fd36c44 | 19 | #include "xfs_quota.h" |
239880ef | 20 | #include "xfs_trans.h" |
7fd36c44 DC |
21 | #include "xfs_qm.h" |
22 | #include "xfs_trans_space.h" | |
23 | #include "xfs_trace.h" | |
24 | ||
57af33e4 BF |
25 | #define _ALLOC true |
26 | #define _FREE false | |
27 | ||
7fd36c44 DC |
28 | /* |
29 | * A buffer has a format structure overhead in the log in addition | |
30 | * to the data, so we need to take this into account when reserving | |
31 | * space in a transaction for a buffer. Round the space required up | |
32 | * to a multiple of 128 bytes so that we don't change the historical | |
33 | * reservation that has been used for this overhead. | |
34 | */ | |
35 | STATIC uint | |
36 | xfs_buf_log_overhead(void) | |
37 | { | |
38 | return round_up(sizeof(struct xlog_op_header) + | |
39 | sizeof(struct xfs_buf_log_format), 128); | |
40 | } | |
41 | ||
42 | /* | |
43 | * Calculate out transaction log reservation per item in bytes. | |
44 | * | |
45 | * The nbufs argument is used to indicate the number of items that | |
46 | * will be changed in a transaction. size is used to tell how many | |
47 | * bytes should be reserved per item. | |
48 | */ | |
49 | STATIC uint | |
50 | xfs_calc_buf_res( | |
51 | uint nbufs, | |
52 | uint size) | |
53 | { | |
54 | return nbufs * (size + xfs_buf_log_overhead()); | |
55 | } | |
56 | ||
fa30f03c DW |
57 | /* |
58 | * Per-extent log reservation for the btree changes involved in freeing or | |
59 | * allocating an extent. In classic XFS there were two trees that will be | |
60 | * modified (bnobt + cntbt). With rmap enabled, there are three trees | |
f310bd2e DW |
61 | * (rmapbt). With reflink, there are four trees (refcountbt). The number of |
62 | * blocks reserved is based on the formula: | |
fa30f03c DW |
63 | * |
64 | * num trees * ((2 blocks/level * max depth) - 1) | |
65 | * | |
66 | * Keep in mind that max depth is calculated separately for each type of tree. | |
67 | */ | |
1946b91c | 68 | uint |
fa30f03c DW |
69 | xfs_allocfree_log_count( |
70 | struct xfs_mount *mp, | |
71 | uint num_ops) | |
72 | { | |
73 | uint blocks; | |
74 | ||
75 | blocks = num_ops * 2 * (2 * mp->m_ag_maxlevels - 1); | |
76 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
77 | blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1); | |
f310bd2e DW |
78 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
79 | blocks += num_ops * (2 * mp->m_refc_maxlevels - 1); | |
fa30f03c DW |
80 | |
81 | return blocks; | |
82 | } | |
83 | ||
23956703 DC |
84 | /* |
85 | * Logging inodes is really tricksy. They are logged in memory format, | |
86 | * which means that what we write into the log doesn't directly translate into | |
87 | * the amount of space they use on disk. | |
88 | * | |
89 | * Case in point - btree format forks in memory format use more space than the | |
90 | * on-disk format. In memory, the buffer contains a normal btree block header so | |
91 | * the btree code can treat it as though it is just another generic buffer. | |
92 | * However, when we write it to the inode fork, we don't write all of this | |
93 | * header as it isn't needed. e.g. the root is only ever in the inode, so | |
94 | * there's no need for sibling pointers which would waste 16 bytes of space. | |
95 | * | |
96 | * Hence when we have an inode with a maximally sized btree format fork, then | |
97 | * amount of information we actually log is greater than the size of the inode | |
98 | * on disk. Hence we need an inode reservation function that calculates all this | |
99 | * correctly. So, we log: | |
100 | * | |
fe4c224a DC |
101 | * - 4 log op headers for object |
102 | * - for the ilf, the inode core and 2 forks | |
23956703 | 103 | * - inode log format object |
fe4c224a DC |
104 | * - the inode core |
105 | * - two inode forks containing bmap btree root blocks. | |
106 | * - the btree data contained by both forks will fit into the inode size, | |
107 | * hence when combined with the inode core above, we have a total of the | |
108 | * actual inode size. | |
109 | * - the BMBT headers need to be accounted separately, as they are | |
110 | * additional to the records and pointers that fit inside the inode | |
111 | * forks. | |
23956703 DC |
112 | */ |
113 | STATIC uint | |
114 | xfs_calc_inode_res( | |
115 | struct xfs_mount *mp, | |
116 | uint ninodes) | |
117 | { | |
fe4c224a DC |
118 | return ninodes * |
119 | (4 * sizeof(struct xlog_op_header) + | |
120 | sizeof(struct xfs_inode_log_format) + | |
121 | mp->m_sb.sb_inodesize + | |
122 | 2 * XFS_BMBT_BLOCK_LEN(mp)); | |
23956703 DC |
123 | } |
124 | ||
9d43b180 | 125 | /* |
f03c78f3 BF |
126 | * Inode btree record insertion/removal modifies the inode btree and free space |
127 | * btrees (since the inobt does not use the agfl). This requires the following | |
128 | * reservation: | |
9d43b180 | 129 | * |
f03c78f3 BF |
130 | * the inode btree: max depth * blocksize |
131 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
9d43b180 | 132 | * |
f03c78f3 BF |
133 | * The caller must account for SB and AG header modifications, etc. |
134 | */ | |
135 | STATIC uint | |
136 | xfs_calc_inobt_res( | |
137 | struct xfs_mount *mp) | |
138 | { | |
139 | return xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1)) + | |
140 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), | |
141 | XFS_FSB_TO_B(mp, 1)); | |
142 | } | |
143 | ||
144 | /* | |
145 | * The free inode btree is a conditional feature. The behavior differs slightly | |
146 | * from that of the traditional inode btree in that the finobt tracks records | |
147 | * for inode chunks with at least one free inode. A record can be removed from | |
148 | * the tree during individual inode allocation. Therefore the finobt | |
149 | * reservation is unconditional for both the inode chunk allocation and | |
150 | * individual inode allocation (modify) cases. | |
9d43b180 | 151 | * |
f03c78f3 BF |
152 | * Behavior aside, the reservation for finobt modification is equivalent to the |
153 | * traditional inobt: cover a full finobt shape change plus block allocation. | |
9d43b180 BF |
154 | */ |
155 | STATIC uint | |
156 | xfs_calc_finobt_res( | |
f03c78f3 | 157 | struct xfs_mount *mp) |
9d43b180 | 158 | { |
9d43b180 BF |
159 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) |
160 | return 0; | |
161 | ||
f03c78f3 | 162 | return xfs_calc_inobt_res(mp); |
9d43b180 BF |
163 | } |
164 | ||
57af33e4 BF |
165 | /* |
166 | * Calculate the reservation required to allocate or free an inode chunk. This | |
167 | * includes: | |
168 | * | |
169 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
170 | * the inode chunk: m_ialloc_blks * N | |
171 | * | |
172 | * The size N of the inode chunk reservation depends on whether it is for | |
173 | * allocation or free and which type of create transaction is in use. An inode | |
174 | * chunk free always invalidates the buffers and only requires reservation for | |
175 | * headers (N == 0). An inode chunk allocation requires a chunk sized | |
176 | * reservation on v4 and older superblocks to initialize the chunk. No chunk | |
177 | * reservation is required for allocation on v5 supers, which use ordered | |
178 | * buffers to initialize. | |
179 | */ | |
180 | STATIC uint | |
181 | xfs_calc_inode_chunk_res( | |
182 | struct xfs_mount *mp, | |
183 | bool alloc) | |
184 | { | |
185 | uint res, size = 0; | |
186 | ||
187 | res = xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), | |
188 | XFS_FSB_TO_B(mp, 1)); | |
189 | if (alloc) { | |
190 | /* icreate tx uses ordered buffers */ | |
191 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
192 | return res; | |
193 | size = XFS_FSB_TO_B(mp, 1); | |
194 | } | |
195 | ||
196 | res += xfs_calc_buf_res(mp->m_ialloc_blks, size); | |
197 | return res; | |
198 | } | |
199 | ||
7fd36c44 DC |
200 | /* |
201 | * Various log reservation values. | |
202 | * | |
203 | * These are based on the size of the file system block because that is what | |
204 | * most transactions manipulate. Each adds in an additional 128 bytes per | |
205 | * item logged to try to account for the overhead of the transaction mechanism. | |
206 | * | |
207 | * Note: Most of the reservations underestimate the number of allocation | |
310a75a3 | 208 | * groups into which they could free extents in the xfs_defer_finish() call. |
7fd36c44 | 209 | * This is because the number in the worst case is quite high and quite |
310a75a3 | 210 | * unusual. In order to fix this we need to change xfs_defer_finish() to free |
7fd36c44 DC |
211 | * extents in only a single AG at a time. This will require changes to the |
212 | * EFI code as well, however, so that the EFI for the extents not freed is | |
213 | * logged again in each transaction. See SGI PV #261917. | |
214 | * | |
215 | * Reservation functions here avoid a huge stack in xfs_trans_init due to | |
216 | * register overflow from temporaries in the calculations. | |
217 | */ | |
218 | ||
219 | ||
220 | /* | |
221 | * In a write transaction we can allocate a maximum of 2 | |
222 | * extents. This gives: | |
223 | * the inode getting the new extents: inode size | |
224 | * the inode's bmap btree: max depth * block size | |
225 | * the agfs of the ags from which the extents are allocated: 2 * sector | |
226 | * the superblock free block counter: sector size | |
227 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
228 | * And the bmap_finish transaction can free bmap blocks in a join: | |
229 | * the agfs of the ags containing the blocks: 2 * sector size | |
230 | * the agfls of the ags containing the blocks: 2 * sector size | |
231 | * the super block free block counter: sector size | |
232 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
233 | */ | |
234 | STATIC uint | |
235 | xfs_calc_write_reservation( | |
236 | struct xfs_mount *mp) | |
237 | { | |
238 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 239 | MAX((xfs_calc_inode_res(mp, 1) + |
7fd36c44 DC |
240 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), |
241 | XFS_FSB_TO_B(mp, 1)) + | |
242 | xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + | |
fa30f03c | 243 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), |
7fd36c44 DC |
244 | XFS_FSB_TO_B(mp, 1))), |
245 | (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) + | |
fa30f03c | 246 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), |
7fd36c44 DC |
247 | XFS_FSB_TO_B(mp, 1)))); |
248 | } | |
249 | ||
250 | /* | |
251 | * In truncating a file we free up to two extents at once. We can modify: | |
252 | * the inode being truncated: inode size | |
253 | * the inode's bmap btree: (max depth + 1) * block size | |
254 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
255 | * the agf for each of the ags: 4 * sector size | |
256 | * the agfl for each of the ags: 4 * sector size | |
257 | * the super block to reflect the freed blocks: sector size | |
258 | * worst case split in allocation btrees per extent assuming 4 extents: | |
259 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
7fd36c44 DC |
260 | */ |
261 | STATIC uint | |
262 | xfs_calc_itruncate_reservation( | |
263 | struct xfs_mount *mp) | |
264 | { | |
265 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 266 | MAX((xfs_calc_inode_res(mp, 1) + |
7fd36c44 DC |
267 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1, |
268 | XFS_FSB_TO_B(mp, 1))), | |
269 | (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) + | |
fa30f03c | 270 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4), |
a606ebdb | 271 | XFS_FSB_TO_B(mp, 1)))); |
7fd36c44 DC |
272 | } |
273 | ||
274 | /* | |
275 | * In renaming a files we can modify: | |
276 | * the four inodes involved: 4 * inode size | |
277 | * the two directory btrees: 2 * (max depth + v2) * dir block size | |
278 | * the two directory bmap btrees: 2 * max depth * block size | |
279 | * And the bmap_finish transaction can free dir and bmap blocks (two sets | |
280 | * of bmap blocks) giving: | |
281 | * the agf for the ags in which the blocks live: 3 * sector size | |
282 | * the agfl for the ags in which the blocks live: 3 * sector size | |
283 | * the superblock for the free block count: sector size | |
284 | * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size | |
285 | */ | |
286 | STATIC uint | |
287 | xfs_calc_rename_reservation( | |
288 | struct xfs_mount *mp) | |
289 | { | |
290 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 291 | MAX((xfs_calc_inode_res(mp, 4) + |
7fd36c44 DC |
292 | xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp), |
293 | XFS_FSB_TO_B(mp, 1))), | |
294 | (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) + | |
fa30f03c | 295 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 3), |
7fd36c44 DC |
296 | XFS_FSB_TO_B(mp, 1)))); |
297 | } | |
298 | ||
ab297431 ZYW |
299 | /* |
300 | * For removing an inode from unlinked list at first, we can modify: | |
301 | * the agi hash list and counters: sector size | |
302 | * the on disk inode before ours in the agi hash list: inode cluster size | |
e8341d9f | 303 | * the on disk inode in the agi hash list: inode cluster size |
ab297431 ZYW |
304 | */ |
305 | STATIC uint | |
306 | xfs_calc_iunlink_remove_reservation( | |
307 | struct xfs_mount *mp) | |
308 | { | |
309 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
e8341d9f | 310 | 2 * max_t(uint, XFS_FSB_TO_B(mp, 1), mp->m_inode_cluster_size); |
ab297431 ZYW |
311 | } |
312 | ||
7fd36c44 DC |
313 | /* |
314 | * For creating a link to an inode: | |
315 | * the parent directory inode: inode size | |
316 | * the linked inode: inode size | |
317 | * the directory btree could split: (max depth + v2) * dir block size | |
318 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
319 | * And the bmap_finish transaction can free some bmap blocks giving: | |
320 | * the agf for the ag in which the blocks live: sector size | |
321 | * the agfl for the ag in which the blocks live: sector size | |
322 | * the superblock for the free block count: sector size | |
323 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
324 | */ | |
325 | STATIC uint | |
326 | xfs_calc_link_reservation( | |
327 | struct xfs_mount *mp) | |
328 | { | |
329 | return XFS_DQUOT_LOGRES(mp) + | |
ab297431 | 330 | xfs_calc_iunlink_remove_reservation(mp) + |
23956703 | 331 | MAX((xfs_calc_inode_res(mp, 2) + |
7fd36c44 DC |
332 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), |
333 | XFS_FSB_TO_B(mp, 1))), | |
334 | (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + | |
fa30f03c | 335 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), |
7fd36c44 DC |
336 | XFS_FSB_TO_B(mp, 1)))); |
337 | } | |
338 | ||
99b6436b ZYW |
339 | /* |
340 | * For adding an inode to unlinked list we can modify: | |
341 | * the agi hash list: sector size | |
e8341d9f | 342 | * the on disk inode: inode cluster size |
99b6436b ZYW |
343 | */ |
344 | STATIC uint | |
345 | xfs_calc_iunlink_add_reservation(xfs_mount_t *mp) | |
346 | { | |
347 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
e8341d9f | 348 | max_t(uint, XFS_FSB_TO_B(mp, 1), mp->m_inode_cluster_size); |
99b6436b ZYW |
349 | } |
350 | ||
7fd36c44 DC |
351 | /* |
352 | * For removing a directory entry we can modify: | |
353 | * the parent directory inode: inode size | |
354 | * the removed inode: inode size | |
355 | * the directory btree could join: (max depth + v2) * dir block size | |
356 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
357 | * And the bmap_finish transaction can free the dir and bmap blocks giving: | |
358 | * the agf for the ag in which the blocks live: 2 * sector size | |
359 | * the agfl for the ag in which the blocks live: 2 * sector size | |
360 | * the superblock for the free block count: sector size | |
361 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
362 | */ | |
363 | STATIC uint | |
364 | xfs_calc_remove_reservation( | |
365 | struct xfs_mount *mp) | |
366 | { | |
367 | return XFS_DQUOT_LOGRES(mp) + | |
99b6436b ZYW |
368 | xfs_calc_iunlink_add_reservation(mp) + |
369 | MAX((xfs_calc_inode_res(mp, 1) + | |
7fd36c44 DC |
370 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), |
371 | XFS_FSB_TO_B(mp, 1))), | |
99b6436b | 372 | (xfs_calc_buf_res(4, mp->m_sb.sb_sectsize) + |
fa30f03c | 373 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), |
7fd36c44 DC |
374 | XFS_FSB_TO_B(mp, 1)))); |
375 | } | |
376 | ||
377 | /* | |
378 | * For create, break it in to the two cases that the transaction | |
379 | * covers. We start with the modify case - allocation done by modification | |
380 | * of the state of existing inodes - and the allocation case. | |
381 | */ | |
382 | ||
383 | /* | |
384 | * For create we can modify: | |
385 | * the parent directory inode: inode size | |
386 | * the new inode: inode size | |
387 | * the inode btree entry: block size | |
388 | * the superblock for the nlink flag: sector size | |
389 | * the directory btree: (max depth + v2) * dir block size | |
390 | * the directory inode's bmap btree: (max depth + v2) * block size | |
9d43b180 | 391 | * the finobt (record modification and allocation btrees) |
7fd36c44 DC |
392 | */ |
393 | STATIC uint | |
394 | xfs_calc_create_resv_modify( | |
395 | struct xfs_mount *mp) | |
396 | { | |
23956703 | 397 | return xfs_calc_inode_res(mp, 2) + |
7fd36c44 DC |
398 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + |
399 | (uint)XFS_FSB_TO_B(mp, 1) + | |
9d43b180 | 400 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), XFS_FSB_TO_B(mp, 1)) + |
f03c78f3 | 401 | xfs_calc_finobt_res(mp); |
7fd36c44 DC |
402 | } |
403 | ||
7fd36c44 DC |
404 | /* |
405 | * For icreate we can allocate some inodes giving: | |
406 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
407 | * the superblock for the nlink flag: sector size | |
c017cb5d | 408 | * the inode chunk (allocation, optional init) |
f03c78f3 | 409 | * the inobt (record insertion) |
c017cb5d | 410 | * the finobt (optional, record insertion) |
7fd36c44 DC |
411 | */ |
412 | STATIC uint | |
413 | xfs_calc_icreate_resv_alloc( | |
414 | struct xfs_mount *mp) | |
415 | { | |
416 | return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
417 | mp->m_sb.sb_sectsize + | |
57af33e4 | 418 | xfs_calc_inode_chunk_res(mp, _ALLOC) + |
f03c78f3 BF |
419 | xfs_calc_inobt_res(mp) + |
420 | xfs_calc_finobt_res(mp); | |
7fd36c44 DC |
421 | } |
422 | ||
423 | STATIC uint | |
424 | xfs_calc_icreate_reservation(xfs_mount_t *mp) | |
425 | { | |
426 | return XFS_DQUOT_LOGRES(mp) + | |
427 | MAX(xfs_calc_icreate_resv_alloc(mp), | |
428 | xfs_calc_create_resv_modify(mp)); | |
429 | } | |
430 | ||
99b6436b ZYW |
431 | STATIC uint |
432 | xfs_calc_create_tmpfile_reservation( | |
433 | struct xfs_mount *mp) | |
434 | { | |
435 | uint res = XFS_DQUOT_LOGRES(mp); | |
436 | ||
c017cb5d | 437 | res += xfs_calc_icreate_resv_alloc(mp); |
99b6436b ZYW |
438 | return res + xfs_calc_iunlink_add_reservation(mp); |
439 | } | |
440 | ||
7fd36c44 DC |
441 | /* |
442 | * Making a new directory is the same as creating a new file. | |
443 | */ | |
444 | STATIC uint | |
445 | xfs_calc_mkdir_reservation( | |
446 | struct xfs_mount *mp) | |
447 | { | |
c017cb5d | 448 | return xfs_calc_icreate_reservation(mp); |
7fd36c44 DC |
449 | } |
450 | ||
451 | ||
452 | /* | |
453 | * Making a new symplink is the same as creating a new file, but | |
454 | * with the added blocks for remote symlink data which can be up to 1kB in | |
6eb0b8df | 455 | * length (XFS_SYMLINK_MAXLEN). |
7fd36c44 DC |
456 | */ |
457 | STATIC uint | |
458 | xfs_calc_symlink_reservation( | |
459 | struct xfs_mount *mp) | |
460 | { | |
c017cb5d | 461 | return xfs_calc_icreate_reservation(mp) + |
6eb0b8df | 462 | xfs_calc_buf_res(1, XFS_SYMLINK_MAXLEN); |
7fd36c44 DC |
463 | } |
464 | ||
465 | /* | |
466 | * In freeing an inode we can modify: | |
467 | * the inode being freed: inode size | |
a6f48590 BF |
468 | * the super block free inode counter, AGF and AGFL: sector size |
469 | * the on disk inode (agi unlinked list removal) | |
57af33e4 | 470 | * the inode chunk (invalidated, headers only) |
f03c78f3 | 471 | * the inode btree |
9d43b180 | 472 | * the finobt (record insertion, removal or modification) |
f03c78f3 | 473 | * |
57af33e4 BF |
474 | * Note that the inode chunk res. includes an allocfree res. for freeing of the |
475 | * inode chunk. This is technically extraneous because the inode chunk free is | |
476 | * deferred (it occurs after a transaction roll). Include the extra reservation | |
477 | * anyways since we've had reports of ifree transaction overruns due to too many | |
478 | * agfl fixups during inode chunk frees. | |
7fd36c44 DC |
479 | */ |
480 | STATIC uint | |
481 | xfs_calc_ifree_reservation( | |
482 | struct xfs_mount *mp) | |
483 | { | |
484 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 485 | xfs_calc_inode_res(mp, 1) + |
a6f48590 | 486 | xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + |
ab297431 | 487 | xfs_calc_iunlink_remove_reservation(mp) + |
57af33e4 | 488 | xfs_calc_inode_chunk_res(mp, _FREE) + |
f03c78f3 BF |
489 | xfs_calc_inobt_res(mp) + |
490 | xfs_calc_finobt_res(mp); | |
7fd36c44 DC |
491 | } |
492 | ||
493 | /* | |
494 | * When only changing the inode we log the inode and possibly the superblock | |
495 | * We also add a bit of slop for the transaction stuff. | |
496 | */ | |
497 | STATIC uint | |
498 | xfs_calc_ichange_reservation( | |
499 | struct xfs_mount *mp) | |
500 | { | |
501 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 DC |
502 | xfs_calc_inode_res(mp, 1) + |
503 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
7fd36c44 DC |
504 | |
505 | } | |
506 | ||
507 | /* | |
508 | * Growing the data section of the filesystem. | |
509 | * superblock | |
510 | * agi and agf | |
511 | * allocation btrees | |
512 | */ | |
513 | STATIC uint | |
514 | xfs_calc_growdata_reservation( | |
515 | struct xfs_mount *mp) | |
516 | { | |
517 | return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + | |
fa30f03c | 518 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), |
7fd36c44 DC |
519 | XFS_FSB_TO_B(mp, 1)); |
520 | } | |
521 | ||
522 | /* | |
523 | * Growing the rt section of the filesystem. | |
524 | * In the first set of transactions (ALLOC) we allocate space to the | |
525 | * bitmap or summary files. | |
526 | * superblock: sector size | |
527 | * agf of the ag from which the extent is allocated: sector size | |
528 | * bmap btree for bitmap/summary inode: max depth * blocksize | |
529 | * bitmap/summary inode: inode size | |
530 | * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize | |
531 | */ | |
532 | STATIC uint | |
533 | xfs_calc_growrtalloc_reservation( | |
534 | struct xfs_mount *mp) | |
535 | { | |
536 | return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
537 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), | |
538 | XFS_FSB_TO_B(mp, 1)) + | |
23956703 | 539 | xfs_calc_inode_res(mp, 1) + |
fa30f03c | 540 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), |
7fd36c44 DC |
541 | XFS_FSB_TO_B(mp, 1)); |
542 | } | |
543 | ||
544 | /* | |
545 | * Growing the rt section of the filesystem. | |
546 | * In the second set of transactions (ZERO) we zero the new metadata blocks. | |
547 | * one bitmap/summary block: blocksize | |
548 | */ | |
549 | STATIC uint | |
550 | xfs_calc_growrtzero_reservation( | |
551 | struct xfs_mount *mp) | |
552 | { | |
553 | return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize); | |
554 | } | |
555 | ||
556 | /* | |
557 | * Growing the rt section of the filesystem. | |
558 | * In the third set of transactions (FREE) we update metadata without | |
559 | * allocating any new blocks. | |
560 | * superblock: sector size | |
561 | * bitmap inode: inode size | |
562 | * summary inode: inode size | |
563 | * one bitmap block: blocksize | |
564 | * summary blocks: new summary size | |
565 | */ | |
566 | STATIC uint | |
567 | xfs_calc_growrtfree_reservation( | |
568 | struct xfs_mount *mp) | |
569 | { | |
570 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
23956703 | 571 | xfs_calc_inode_res(mp, 2) + |
7fd36c44 DC |
572 | xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) + |
573 | xfs_calc_buf_res(1, mp->m_rsumsize); | |
574 | } | |
575 | ||
576 | /* | |
577 | * Logging the inode modification timestamp on a synchronous write. | |
578 | * inode | |
579 | */ | |
580 | STATIC uint | |
581 | xfs_calc_swrite_reservation( | |
582 | struct xfs_mount *mp) | |
583 | { | |
23956703 | 584 | return xfs_calc_inode_res(mp, 1); |
7fd36c44 DC |
585 | } |
586 | ||
587 | /* | |
588 | * Logging the inode mode bits when writing a setuid/setgid file | |
589 | * inode | |
590 | */ | |
591 | STATIC uint | |
23956703 DC |
592 | xfs_calc_writeid_reservation( |
593 | struct xfs_mount *mp) | |
7fd36c44 | 594 | { |
23956703 | 595 | return xfs_calc_inode_res(mp, 1); |
7fd36c44 DC |
596 | } |
597 | ||
598 | /* | |
599 | * Converting the inode from non-attributed to attributed. | |
600 | * the inode being converted: inode size | |
601 | * agf block and superblock (for block allocation) | |
602 | * the new block (directory sized) | |
603 | * bmap blocks for the new directory block | |
604 | * allocation btrees | |
605 | */ | |
606 | STATIC uint | |
607 | xfs_calc_addafork_reservation( | |
608 | struct xfs_mount *mp) | |
609 | { | |
610 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 611 | xfs_calc_inode_res(mp, 1) + |
7fd36c44 | 612 | xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + |
8f66193c | 613 | xfs_calc_buf_res(1, mp->m_dir_geo->blksize) + |
7fd36c44 DC |
614 | xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1, |
615 | XFS_FSB_TO_B(mp, 1)) + | |
fa30f03c | 616 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 1), |
7fd36c44 DC |
617 | XFS_FSB_TO_B(mp, 1)); |
618 | } | |
619 | ||
620 | /* | |
621 | * Removing the attribute fork of a file | |
622 | * the inode being truncated: inode size | |
623 | * the inode's bmap btree: max depth * block size | |
624 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
625 | * the agf for each of the ags: 4 * sector size | |
626 | * the agfl for each of the ags: 4 * sector size | |
627 | * the super block to reflect the freed blocks: sector size | |
628 | * worst case split in allocation btrees per extent assuming 4 extents: | |
629 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
630 | */ | |
631 | STATIC uint | |
632 | xfs_calc_attrinval_reservation( | |
633 | struct xfs_mount *mp) | |
634 | { | |
23956703 | 635 | return MAX((xfs_calc_inode_res(mp, 1) + |
7fd36c44 DC |
636 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK), |
637 | XFS_FSB_TO_B(mp, 1))), | |
638 | (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) + | |
fa30f03c | 639 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 4), |
7fd36c44 DC |
640 | XFS_FSB_TO_B(mp, 1)))); |
641 | } | |
642 | ||
643 | /* | |
644 | * Setting an attribute at mount time. | |
645 | * the inode getting the attribute | |
646 | * the superblock for allocations | |
647 | * the agfs extents are allocated from | |
648 | * the attribute btree * max depth | |
649 | * the inode allocation btree | |
650 | * Since attribute transaction space is dependent on the size of the attribute, | |
651 | * the calculation is done partially at mount time and partially at runtime(see | |
652 | * below). | |
653 | */ | |
654 | STATIC uint | |
655 | xfs_calc_attrsetm_reservation( | |
656 | struct xfs_mount *mp) | |
657 | { | |
658 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 659 | xfs_calc_inode_res(mp, 1) + |
7fd36c44 DC |
660 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + |
661 | xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1)); | |
662 | } | |
663 | ||
664 | /* | |
665 | * Setting an attribute at runtime, transaction space unit per block. | |
666 | * the superblock for allocations: sector size | |
667 | * the inode bmap btree could join or split: max depth * block size | |
668 | * Since the runtime attribute transaction space is dependent on the total | |
669 | * blocks needed for the 1st bmap, here we calculate out the space unit for | |
670 | * one block so that the caller could figure out the total space according | |
3d3c8b52 JL |
671 | * to the attibute extent length in blocks by: |
672 | * ext * M_RES(mp)->tr_attrsetrt.tr_logres | |
7fd36c44 DC |
673 | */ |
674 | STATIC uint | |
675 | xfs_calc_attrsetrt_reservation( | |
676 | struct xfs_mount *mp) | |
677 | { | |
678 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
679 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK), | |
680 | XFS_FSB_TO_B(mp, 1)); | |
681 | } | |
682 | ||
683 | /* | |
684 | * Removing an attribute. | |
685 | * the inode: inode size | |
686 | * the attribute btree could join: max depth * block size | |
687 | * the inode bmap btree could join or split: max depth * block size | |
688 | * And the bmap_finish transaction can free the attr blocks freed giving: | |
689 | * the agf for the ag in which the blocks live: 2 * sector size | |
690 | * the agfl for the ag in which the blocks live: 2 * sector size | |
691 | * the superblock for the free block count: sector size | |
692 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
693 | */ | |
694 | STATIC uint | |
695 | xfs_calc_attrrm_reservation( | |
696 | struct xfs_mount *mp) | |
697 | { | |
698 | return XFS_DQUOT_LOGRES(mp) + | |
23956703 | 699 | MAX((xfs_calc_inode_res(mp, 1) + |
7fd36c44 DC |
700 | xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, |
701 | XFS_FSB_TO_B(mp, 1)) + | |
702 | (uint)XFS_FSB_TO_B(mp, | |
703 | XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
704 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)), | |
705 | (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) + | |
fa30f03c | 706 | xfs_calc_buf_res(xfs_allocfree_log_count(mp, 2), |
7fd36c44 DC |
707 | XFS_FSB_TO_B(mp, 1)))); |
708 | } | |
709 | ||
710 | /* | |
711 | * Clearing a bad agino number in an agi hash bucket. | |
712 | */ | |
713 | STATIC uint | |
714 | xfs_calc_clear_agi_bucket_reservation( | |
715 | struct xfs_mount *mp) | |
716 | { | |
717 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
718 | } | |
719 | ||
7fd36c44 DC |
720 | /* |
721 | * Adjusting quota limits. | |
722 | * the xfs_disk_dquot_t: sizeof(struct xfs_disk_dquot) | |
723 | */ | |
724 | STATIC uint | |
a1f69417 | 725 | xfs_calc_qm_setqlim_reservation(void) |
7fd36c44 DC |
726 | { |
727 | return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot)); | |
728 | } | |
729 | ||
730 | /* | |
731 | * Allocating quota on disk if needed. | |
410b11a6 | 732 | * the write transaction log space for quota file extent allocation |
7fd36c44 DC |
733 | * the unit of quota allocation: one system block size |
734 | */ | |
735 | STATIC uint | |
736 | xfs_calc_qm_dqalloc_reservation( | |
737 | struct xfs_mount *mp) | |
738 | { | |
410b11a6 | 739 | return xfs_calc_write_reservation(mp) + |
7fd36c44 DC |
740 | xfs_calc_buf_res(1, |
741 | XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1); | |
742 | } | |
743 | ||
744 | /* | |
745 | * Turning off quotas. | |
746 | * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2 | |
747 | * the superblock for the quota flags: sector size | |
748 | */ | |
749 | STATIC uint | |
750 | xfs_calc_qm_quotaoff_reservation( | |
751 | struct xfs_mount *mp) | |
752 | { | |
753 | return sizeof(struct xfs_qoff_logitem) * 2 + | |
754 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
755 | } | |
756 | ||
757 | /* | |
758 | * End of turning off quotas. | |
759 | * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2 | |
760 | */ | |
761 | STATIC uint | |
a1f69417 | 762 | xfs_calc_qm_quotaoff_end_reservation(void) |
7fd36c44 DC |
763 | { |
764 | return sizeof(struct xfs_qoff_logitem) * 2; | |
765 | } | |
766 | ||
767 | /* | |
768 | * Syncing the incore super block changes to disk. | |
769 | * the super block to reflect the changes: sector size | |
770 | */ | |
771 | STATIC uint | |
772 | xfs_calc_sb_reservation( | |
773 | struct xfs_mount *mp) | |
774 | { | |
775 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
776 | } | |
777 | ||
778 | void | |
779 | xfs_trans_resv_calc( | |
780 | struct xfs_mount *mp, | |
781 | struct xfs_trans_resv *resp) | |
782 | { | |
0eadd102 JL |
783 | /* |
784 | * The following transactions are logged in physical format and | |
785 | * require a permanent reservation on space. | |
786 | */ | |
787 | resp->tr_write.tr_logres = xfs_calc_write_reservation(mp); | |
80de462e DW |
788 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
789 | resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK; | |
790 | else | |
791 | resp->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT; | |
0eadd102 JL |
792 | resp->tr_write.tr_logflags |= XFS_TRANS_PERM_LOG_RES; |
793 | ||
794 | resp->tr_itruncate.tr_logres = xfs_calc_itruncate_reservation(mp); | |
80de462e DW |
795 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
796 | resp->tr_itruncate.tr_logcount = | |
797 | XFS_ITRUNCATE_LOG_COUNT_REFLINK; | |
798 | else | |
799 | resp->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT; | |
0eadd102 JL |
800 | resp->tr_itruncate.tr_logflags |= XFS_TRANS_PERM_LOG_RES; |
801 | ||
802 | resp->tr_rename.tr_logres = xfs_calc_rename_reservation(mp); | |
803 | resp->tr_rename.tr_logcount = XFS_RENAME_LOG_COUNT; | |
804 | resp->tr_rename.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
805 | ||
806 | resp->tr_link.tr_logres = xfs_calc_link_reservation(mp); | |
807 | resp->tr_link.tr_logcount = XFS_LINK_LOG_COUNT; | |
808 | resp->tr_link.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
809 | ||
810 | resp->tr_remove.tr_logres = xfs_calc_remove_reservation(mp); | |
811 | resp->tr_remove.tr_logcount = XFS_REMOVE_LOG_COUNT; | |
812 | resp->tr_remove.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
813 | ||
814 | resp->tr_symlink.tr_logres = xfs_calc_symlink_reservation(mp); | |
815 | resp->tr_symlink.tr_logcount = XFS_SYMLINK_LOG_COUNT; | |
816 | resp->tr_symlink.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
817 | ||
c017cb5d | 818 | resp->tr_create.tr_logres = xfs_calc_icreate_reservation(mp); |
0eadd102 JL |
819 | resp->tr_create.tr_logcount = XFS_CREATE_LOG_COUNT; |
820 | resp->tr_create.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
821 | ||
99b6436b ZYW |
822 | resp->tr_create_tmpfile.tr_logres = |
823 | xfs_calc_create_tmpfile_reservation(mp); | |
824 | resp->tr_create_tmpfile.tr_logcount = XFS_CREATE_TMPFILE_LOG_COUNT; | |
825 | resp->tr_create_tmpfile.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
826 | ||
0eadd102 JL |
827 | resp->tr_mkdir.tr_logres = xfs_calc_mkdir_reservation(mp); |
828 | resp->tr_mkdir.tr_logcount = XFS_MKDIR_LOG_COUNT; | |
829 | resp->tr_mkdir.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
830 | ||
831 | resp->tr_ifree.tr_logres = xfs_calc_ifree_reservation(mp); | |
832 | resp->tr_ifree.tr_logcount = XFS_INACTIVE_LOG_COUNT; | |
833 | resp->tr_ifree.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
834 | ||
835 | resp->tr_addafork.tr_logres = xfs_calc_addafork_reservation(mp); | |
836 | resp->tr_addafork.tr_logcount = XFS_ADDAFORK_LOG_COUNT; | |
837 | resp->tr_addafork.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
838 | ||
839 | resp->tr_attrinval.tr_logres = xfs_calc_attrinval_reservation(mp); | |
840 | resp->tr_attrinval.tr_logcount = XFS_ATTRINVAL_LOG_COUNT; | |
841 | resp->tr_attrinval.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
842 | ||
843 | resp->tr_attrsetm.tr_logres = xfs_calc_attrsetm_reservation(mp); | |
844 | resp->tr_attrsetm.tr_logcount = XFS_ATTRSET_LOG_COUNT; | |
845 | resp->tr_attrsetm.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
846 | ||
847 | resp->tr_attrrm.tr_logres = xfs_calc_attrrm_reservation(mp); | |
848 | resp->tr_attrrm.tr_logcount = XFS_ATTRRM_LOG_COUNT; | |
849 | resp->tr_attrrm.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
850 | ||
851 | resp->tr_growrtalloc.tr_logres = xfs_calc_growrtalloc_reservation(mp); | |
852 | resp->tr_growrtalloc.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT; | |
853 | resp->tr_growrtalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES; | |
854 | ||
855 | resp->tr_qm_dqalloc.tr_logres = xfs_calc_qm_dqalloc_reservation(mp); | |
80de462e DW |
856 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
857 | resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK; | |
858 | else | |
859 | resp->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT; | |
0eadd102 JL |
860 | resp->tr_qm_dqalloc.tr_logflags |= XFS_TRANS_PERM_LOG_RES; |
861 | ||
862 | /* | |
863 | * The following transactions are logged in logical format with | |
864 | * a default log count. | |
865 | */ | |
a1f69417 | 866 | resp->tr_qm_setqlim.tr_logres = xfs_calc_qm_setqlim_reservation(); |
0eadd102 JL |
867 | resp->tr_qm_setqlim.tr_logcount = XFS_DEFAULT_LOG_COUNT; |
868 | ||
869 | resp->tr_qm_quotaoff.tr_logres = xfs_calc_qm_quotaoff_reservation(mp); | |
870 | resp->tr_qm_quotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT; | |
871 | ||
872 | resp->tr_qm_equotaoff.tr_logres = | |
a1f69417 | 873 | xfs_calc_qm_quotaoff_end_reservation(); |
0eadd102 JL |
874 | resp->tr_qm_equotaoff.tr_logcount = XFS_DEFAULT_LOG_COUNT; |
875 | ||
876 | resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp); | |
877 | resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT; | |
878 | ||
879 | /* The following transaction are logged in logical format */ | |
880 | resp->tr_ichange.tr_logres = xfs_calc_ichange_reservation(mp); | |
881 | resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp); | |
20996c93 | 882 | resp->tr_fsyncts.tr_logres = xfs_calc_swrite_reservation(mp); |
0eadd102 JL |
883 | resp->tr_writeid.tr_logres = xfs_calc_writeid_reservation(mp); |
884 | resp->tr_attrsetrt.tr_logres = xfs_calc_attrsetrt_reservation(mp); | |
885 | resp->tr_clearagi.tr_logres = xfs_calc_clear_agi_bucket_reservation(mp); | |
886 | resp->tr_growrtzero.tr_logres = xfs_calc_growrtzero_reservation(mp); | |
887 | resp->tr_growrtfree.tr_logres = xfs_calc_growrtfree_reservation(mp); | |
7fd36c44 | 888 | } |