]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_mount.h" |
28 | #include "xfs_error.h" | |
a844f451 | 29 | #include "xfs_da_btree.h" |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_btree.h" |
36 | #include "xfs_ialloc.h" | |
37 | #include "xfs_alloc.h" | |
1da177e4 | 38 | #include "xfs_bmap.h" |
1da177e4 | 39 | #include "xfs_quota.h" |
a844f451 | 40 | #include "xfs_trans_priv.h" |
1da177e4 | 41 | #include "xfs_trans_space.h" |
322ff6b8 | 42 | #include "xfs_inode_item.h" |
ed3b4d6c | 43 | #include "xfs_trace.h" |
1da177e4 | 44 | |
8f794055 | 45 | kmem_zone_t *xfs_trans_zone; |
1da177e4 | 46 | |
025101dc | 47 | |
8f794055 | 48 | /* |
025101dc CH |
49 | * Various log reservation values. |
50 | * | |
51 | * These are based on the size of the file system block because that is what | |
52 | * most transactions manipulate. Each adds in an additional 128 bytes per | |
53 | * item logged to try to account for the overhead of the transaction mechanism. | |
54 | * | |
55 | * Note: Most of the reservations underestimate the number of allocation | |
56 | * groups into which they could free extents in the xfs_bmap_finish() call. | |
57 | * This is because the number in the worst case is quite high and quite | |
58 | * unusual. In order to fix this we need to change xfs_bmap_finish() to free | |
59 | * extents in only a single AG at a time. This will require changes to the | |
60 | * EFI code as well, however, so that the EFI for the extents not freed is | |
61 | * logged again in each transaction. See SGI PV #261917. | |
62 | * | |
63 | * Reservation functions here avoid a huge stack in xfs_trans_init due to | |
64 | * register overflow from temporaries in the calculations. | |
65 | */ | |
66 | ||
67 | ||
68 | /* | |
69 | * In a write transaction we can allocate a maximum of 2 | |
70 | * extents. This gives: | |
71 | * the inode getting the new extents: inode size | |
72 | * the inode's bmap btree: max depth * block size | |
73 | * the agfs of the ags from which the extents are allocated: 2 * sector | |
74 | * the superblock free block counter: sector size | |
75 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
76 | * And the bmap_finish transaction can free bmap blocks in a join: | |
77 | * the agfs of the ags containing the blocks: 2 * sector size | |
78 | * the agfls of the ags containing the blocks: 2 * sector size | |
79 | * the super block free block counter: sector size | |
80 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
8f794055 | 81 | */ |
8f794055 | 82 | STATIC uint |
025101dc CH |
83 | xfs_calc_write_reservation( |
84 | struct xfs_mount *mp) | |
8f794055 | 85 | { |
025101dc CH |
86 | return XFS_DQUOT_LOGRES(mp) + |
87 | MAX((mp->m_sb.sb_inodesize + | |
88 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) + | |
89 | 2 * mp->m_sb.sb_sectsize + | |
90 | mp->m_sb.sb_sectsize + | |
91 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
92 | 128 * (4 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + | |
93 | XFS_ALLOCFREE_LOG_COUNT(mp, 2))), | |
94 | (2 * mp->m_sb.sb_sectsize + | |
95 | 2 * mp->m_sb.sb_sectsize + | |
96 | mp->m_sb.sb_sectsize + | |
97 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
98 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
99 | } |
100 | ||
025101dc CH |
101 | /* |
102 | * In truncating a file we free up to two extents at once. We can modify: | |
103 | * the inode being truncated: inode size | |
104 | * the inode's bmap btree: (max depth + 1) * block size | |
105 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
106 | * the agf for each of the ags: 4 * sector size | |
107 | * the agfl for each of the ags: 4 * sector size | |
108 | * the super block to reflect the freed blocks: sector size | |
109 | * worst case split in allocation btrees per extent assuming 4 extents: | |
110 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
111 | * the inode btree: max depth * blocksize | |
112 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
113 | */ | |
8f794055 | 114 | STATIC uint |
025101dc CH |
115 | xfs_calc_itruncate_reservation( |
116 | struct xfs_mount *mp) | |
8f794055 | 117 | { |
025101dc CH |
118 | return XFS_DQUOT_LOGRES(mp) + |
119 | MAX((mp->m_sb.sb_inodesize + | |
120 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1) + | |
121 | 128 * (2 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))), | |
122 | (4 * mp->m_sb.sb_sectsize + | |
123 | 4 * mp->m_sb.sb_sectsize + | |
124 | mp->m_sb.sb_sectsize + | |
125 | XFS_ALLOCFREE_LOG_RES(mp, 4) + | |
126 | 128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)) + | |
127 | 128 * 5 + | |
128 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
129 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
130 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
131 | } |
132 | ||
025101dc CH |
133 | /* |
134 | * In renaming a files we can modify: | |
135 | * the four inodes involved: 4 * inode size | |
136 | * the two directory btrees: 2 * (max depth + v2) * dir block size | |
137 | * the two directory bmap btrees: 2 * max depth * block size | |
138 | * And the bmap_finish transaction can free dir and bmap blocks (two sets | |
139 | * of bmap blocks) giving: | |
140 | * the agf for the ags in which the blocks live: 3 * sector size | |
141 | * the agfl for the ags in which the blocks live: 3 * sector size | |
142 | * the superblock for the free block count: sector size | |
143 | * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size | |
144 | */ | |
8f794055 | 145 | STATIC uint |
025101dc CH |
146 | xfs_calc_rename_reservation( |
147 | struct xfs_mount *mp) | |
8f794055 | 148 | { |
025101dc CH |
149 | return XFS_DQUOT_LOGRES(mp) + |
150 | MAX((4 * mp->m_sb.sb_inodesize + | |
151 | 2 * XFS_DIROP_LOG_RES(mp) + | |
152 | 128 * (4 + 2 * XFS_DIROP_LOG_COUNT(mp))), | |
153 | (3 * mp->m_sb.sb_sectsize + | |
154 | 3 * mp->m_sb.sb_sectsize + | |
155 | mp->m_sb.sb_sectsize + | |
156 | XFS_ALLOCFREE_LOG_RES(mp, 3) + | |
157 | 128 * (7 + XFS_ALLOCFREE_LOG_COUNT(mp, 3)))); | |
8f794055 NS |
158 | } |
159 | ||
025101dc CH |
160 | /* |
161 | * For creating a link to an inode: | |
162 | * the parent directory inode: inode size | |
163 | * the linked inode: inode size | |
164 | * the directory btree could split: (max depth + v2) * dir block size | |
165 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
166 | * And the bmap_finish transaction can free some bmap blocks giving: | |
167 | * the agf for the ag in which the blocks live: sector size | |
168 | * the agfl for the ag in which the blocks live: sector size | |
169 | * the superblock for the free block count: sector size | |
170 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
171 | */ | |
8f794055 | 172 | STATIC uint |
025101dc CH |
173 | xfs_calc_link_reservation( |
174 | struct xfs_mount *mp) | |
8f794055 | 175 | { |
025101dc CH |
176 | return XFS_DQUOT_LOGRES(mp) + |
177 | MAX((mp->m_sb.sb_inodesize + | |
178 | mp->m_sb.sb_inodesize + | |
179 | XFS_DIROP_LOG_RES(mp) + | |
180 | 128 * (2 + XFS_DIROP_LOG_COUNT(mp))), | |
181 | (mp->m_sb.sb_sectsize + | |
182 | mp->m_sb.sb_sectsize + | |
183 | mp->m_sb.sb_sectsize + | |
184 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
185 | 128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
186 | } |
187 | ||
025101dc CH |
188 | /* |
189 | * For removing a directory entry we can modify: | |
190 | * the parent directory inode: inode size | |
191 | * the removed inode: inode size | |
192 | * the directory btree could join: (max depth + v2) * dir block size | |
193 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
194 | * And the bmap_finish transaction can free the dir and bmap blocks giving: | |
195 | * the agf for the ag in which the blocks live: 2 * sector size | |
196 | * the agfl for the ag in which the blocks live: 2 * sector size | |
197 | * the superblock for the free block count: sector size | |
198 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
199 | */ | |
8f794055 | 200 | STATIC uint |
025101dc CH |
201 | xfs_calc_remove_reservation( |
202 | struct xfs_mount *mp) | |
8f794055 | 203 | { |
025101dc CH |
204 | return XFS_DQUOT_LOGRES(mp) + |
205 | MAX((mp->m_sb.sb_inodesize + | |
206 | mp->m_sb.sb_inodesize + | |
207 | XFS_DIROP_LOG_RES(mp) + | |
208 | 128 * (2 + XFS_DIROP_LOG_COUNT(mp))), | |
209 | (2 * mp->m_sb.sb_sectsize + | |
210 | 2 * mp->m_sb.sb_sectsize + | |
211 | mp->m_sb.sb_sectsize + | |
212 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
213 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
214 | } |
215 | ||
025101dc CH |
216 | /* |
217 | * For symlink we can modify: | |
218 | * the parent directory inode: inode size | |
219 | * the new inode: inode size | |
220 | * the inode btree entry: 1 block | |
221 | * the directory btree: (max depth + v2) * dir block size | |
222 | * the directory inode's bmap btree: (max depth + v2) * block size | |
223 | * the blocks for the symlink: 1 kB | |
224 | * Or in the first xact we allocate some inodes giving: | |
225 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
226 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
227 | * the inode btree: max depth * blocksize | |
228 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
229 | */ | |
8f794055 | 230 | STATIC uint |
025101dc CH |
231 | xfs_calc_symlink_reservation( |
232 | struct xfs_mount *mp) | |
8f794055 | 233 | { |
025101dc CH |
234 | return XFS_DQUOT_LOGRES(mp) + |
235 | MAX((mp->m_sb.sb_inodesize + | |
236 | mp->m_sb.sb_inodesize + | |
237 | XFS_FSB_TO_B(mp, 1) + | |
238 | XFS_DIROP_LOG_RES(mp) + | |
239 | 1024 + | |
240 | 128 * (4 + XFS_DIROP_LOG_COUNT(mp))), | |
241 | (2 * mp->m_sb.sb_sectsize + | |
242 | XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) + | |
243 | XFS_FSB_TO_B(mp, mp->m_in_maxlevels) + | |
244 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
245 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
246 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
247 | } |
248 | ||
025101dc CH |
249 | /* |
250 | * For create we can modify: | |
251 | * the parent directory inode: inode size | |
252 | * the new inode: inode size | |
253 | * the inode btree entry: block size | |
254 | * the superblock for the nlink flag: sector size | |
255 | * the directory btree: (max depth + v2) * dir block size | |
256 | * the directory inode's bmap btree: (max depth + v2) * block size | |
257 | * Or in the first xact we allocate some inodes giving: | |
258 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
259 | * the superblock for the nlink flag: sector size | |
260 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
261 | * the inode btree: max depth * blocksize | |
262 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
263 | */ | |
8f794055 | 264 | STATIC uint |
025101dc CH |
265 | xfs_calc_create_reservation( |
266 | struct xfs_mount *mp) | |
8f794055 | 267 | { |
025101dc CH |
268 | return XFS_DQUOT_LOGRES(mp) + |
269 | MAX((mp->m_sb.sb_inodesize + | |
270 | mp->m_sb.sb_inodesize + | |
271 | mp->m_sb.sb_sectsize + | |
272 | XFS_FSB_TO_B(mp, 1) + | |
273 | XFS_DIROP_LOG_RES(mp) + | |
274 | 128 * (3 + XFS_DIROP_LOG_COUNT(mp))), | |
275 | (3 * mp->m_sb.sb_sectsize + | |
276 | XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) + | |
277 | XFS_FSB_TO_B(mp, mp->m_in_maxlevels) + | |
278 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
279 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
280 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
281 | } |
282 | ||
025101dc CH |
283 | /* |
284 | * Making a new directory is the same as creating a new file. | |
285 | */ | |
8f794055 | 286 | STATIC uint |
025101dc CH |
287 | xfs_calc_mkdir_reservation( |
288 | struct xfs_mount *mp) | |
8f794055 | 289 | { |
025101dc | 290 | return xfs_calc_create_reservation(mp); |
8f794055 NS |
291 | } |
292 | ||
025101dc CH |
293 | /* |
294 | * In freeing an inode we can modify: | |
295 | * the inode being freed: inode size | |
296 | * the super block free inode counter: sector size | |
297 | * the agi hash list and counters: sector size | |
298 | * the inode btree entry: block size | |
299 | * the on disk inode before ours in the agi hash list: inode cluster size | |
300 | * the inode btree: max depth * blocksize | |
301 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
302 | */ | |
8f794055 | 303 | STATIC uint |
025101dc CH |
304 | xfs_calc_ifree_reservation( |
305 | struct xfs_mount *mp) | |
8f794055 | 306 | { |
025101dc CH |
307 | return XFS_DQUOT_LOGRES(mp) + |
308 | mp->m_sb.sb_inodesize + | |
309 | mp->m_sb.sb_sectsize + | |
310 | mp->m_sb.sb_sectsize + | |
311 | XFS_FSB_TO_B(mp, 1) + | |
312 | MAX((__uint16_t)XFS_FSB_TO_B(mp, 1), | |
313 | XFS_INODE_CLUSTER_SIZE(mp)) + | |
314 | 128 * 5 + | |
315 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
316 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
317 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
318 | } |
319 | ||
025101dc CH |
320 | /* |
321 | * When only changing the inode we log the inode and possibly the superblock | |
322 | * We also add a bit of slop for the transaction stuff. | |
323 | */ | |
8f794055 | 324 | STATIC uint |
025101dc CH |
325 | xfs_calc_ichange_reservation( |
326 | struct xfs_mount *mp) | |
8f794055 | 327 | { |
025101dc CH |
328 | return XFS_DQUOT_LOGRES(mp) + |
329 | mp->m_sb.sb_inodesize + | |
330 | mp->m_sb.sb_sectsize + | |
331 | 512; | |
332 | ||
8f794055 NS |
333 | } |
334 | ||
025101dc CH |
335 | /* |
336 | * Growing the data section of the filesystem. | |
337 | * superblock | |
338 | * agi and agf | |
339 | * allocation btrees | |
340 | */ | |
8f794055 | 341 | STATIC uint |
025101dc CH |
342 | xfs_calc_growdata_reservation( |
343 | struct xfs_mount *mp) | |
8f794055 | 344 | { |
025101dc CH |
345 | return mp->m_sb.sb_sectsize * 3 + |
346 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
347 | 128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
348 | } |
349 | ||
025101dc CH |
350 | /* |
351 | * Growing the rt section of the filesystem. | |
352 | * In the first set of transactions (ALLOC) we allocate space to the | |
353 | * bitmap or summary files. | |
354 | * superblock: sector size | |
355 | * agf of the ag from which the extent is allocated: sector size | |
356 | * bmap btree for bitmap/summary inode: max depth * blocksize | |
357 | * bitmap/summary inode: inode size | |
358 | * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize | |
359 | */ | |
8f794055 | 360 | STATIC uint |
025101dc CH |
361 | xfs_calc_growrtalloc_reservation( |
362 | struct xfs_mount *mp) | |
8f794055 | 363 | { |
025101dc CH |
364 | return 2 * mp->m_sb.sb_sectsize + |
365 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) + | |
366 | mp->m_sb.sb_inodesize + | |
367 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
368 | 128 * (3 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + | |
369 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
370 | } |
371 | ||
025101dc CH |
372 | /* |
373 | * Growing the rt section of the filesystem. | |
374 | * In the second set of transactions (ZERO) we zero the new metadata blocks. | |
375 | * one bitmap/summary block: blocksize | |
376 | */ | |
8f794055 | 377 | STATIC uint |
025101dc CH |
378 | xfs_calc_growrtzero_reservation( |
379 | struct xfs_mount *mp) | |
8f794055 | 380 | { |
025101dc | 381 | return mp->m_sb.sb_blocksize + 128; |
8f794055 NS |
382 | } |
383 | ||
025101dc CH |
384 | /* |
385 | * Growing the rt section of the filesystem. | |
386 | * In the third set of transactions (FREE) we update metadata without | |
387 | * allocating any new blocks. | |
388 | * superblock: sector size | |
389 | * bitmap inode: inode size | |
390 | * summary inode: inode size | |
391 | * one bitmap block: blocksize | |
392 | * summary blocks: new summary size | |
393 | */ | |
8f794055 | 394 | STATIC uint |
025101dc CH |
395 | xfs_calc_growrtfree_reservation( |
396 | struct xfs_mount *mp) | |
8f794055 | 397 | { |
025101dc CH |
398 | return mp->m_sb.sb_sectsize + |
399 | 2 * mp->m_sb.sb_inodesize + | |
400 | mp->m_sb.sb_blocksize + | |
401 | mp->m_rsumsize + | |
402 | 128 * 5; | |
8f794055 NS |
403 | } |
404 | ||
025101dc CH |
405 | /* |
406 | * Logging the inode modification timestamp on a synchronous write. | |
407 | * inode | |
408 | */ | |
8f794055 | 409 | STATIC uint |
025101dc CH |
410 | xfs_calc_swrite_reservation( |
411 | struct xfs_mount *mp) | |
8f794055 | 412 | { |
025101dc | 413 | return mp->m_sb.sb_inodesize + 128; |
8f794055 NS |
414 | } |
415 | ||
025101dc CH |
416 | /* |
417 | * Logging the inode mode bits when writing a setuid/setgid file | |
418 | * inode | |
419 | */ | |
8f794055 NS |
420 | STATIC uint |
421 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
422 | { | |
025101dc | 423 | return mp->m_sb.sb_inodesize + 128; |
8f794055 NS |
424 | } |
425 | ||
025101dc CH |
426 | /* |
427 | * Converting the inode from non-attributed to attributed. | |
428 | * the inode being converted: inode size | |
429 | * agf block and superblock (for block allocation) | |
430 | * the new block (directory sized) | |
431 | * bmap blocks for the new directory block | |
432 | * allocation btrees | |
433 | */ | |
8f794055 | 434 | STATIC uint |
025101dc CH |
435 | xfs_calc_addafork_reservation( |
436 | struct xfs_mount *mp) | |
8f794055 | 437 | { |
025101dc CH |
438 | return XFS_DQUOT_LOGRES(mp) + |
439 | mp->m_sb.sb_inodesize + | |
440 | mp->m_sb.sb_sectsize * 2 + | |
441 | mp->m_dirblksize + | |
442 | XFS_FSB_TO_B(mp, XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1) + | |
443 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
444 | 128 * (4 + XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1 + | |
445 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
446 | } |
447 | ||
025101dc CH |
448 | /* |
449 | * Removing the attribute fork of a file | |
450 | * the inode being truncated: inode size | |
451 | * the inode's bmap btree: max depth * block size | |
452 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
453 | * the agf for each of the ags: 4 * sector size | |
454 | * the agfl for each of the ags: 4 * sector size | |
455 | * the super block to reflect the freed blocks: sector size | |
456 | * worst case split in allocation btrees per extent assuming 4 extents: | |
457 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
458 | */ | |
8f794055 | 459 | STATIC uint |
025101dc CH |
460 | xfs_calc_attrinval_reservation( |
461 | struct xfs_mount *mp) | |
8f794055 | 462 | { |
025101dc CH |
463 | return MAX((mp->m_sb.sb_inodesize + |
464 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
465 | 128 * (1 + XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))), | |
466 | (4 * mp->m_sb.sb_sectsize + | |
467 | 4 * mp->m_sb.sb_sectsize + | |
468 | mp->m_sb.sb_sectsize + | |
469 | XFS_ALLOCFREE_LOG_RES(mp, 4) + | |
470 | 128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)))); | |
8f794055 NS |
471 | } |
472 | ||
025101dc CH |
473 | /* |
474 | * Setting an attribute. | |
475 | * the inode getting the attribute | |
476 | * the superblock for allocations | |
477 | * the agfs extents are allocated from | |
478 | * the attribute btree * max depth | |
479 | * the inode allocation btree | |
480 | * Since attribute transaction space is dependent on the size of the attribute, | |
481 | * the calculation is done partially at mount time and partially at runtime. | |
482 | */ | |
8f794055 | 483 | STATIC uint |
025101dc CH |
484 | xfs_calc_attrset_reservation( |
485 | struct xfs_mount *mp) | |
8f794055 | 486 | { |
025101dc CH |
487 | return XFS_DQUOT_LOGRES(mp) + |
488 | mp->m_sb.sb_inodesize + | |
489 | mp->m_sb.sb_sectsize + | |
490 | XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) + | |
491 | 128 * (2 + XFS_DA_NODE_MAXDEPTH); | |
8f794055 NS |
492 | } |
493 | ||
025101dc CH |
494 | /* |
495 | * Removing an attribute. | |
496 | * the inode: inode size | |
497 | * the attribute btree could join: max depth * block size | |
498 | * the inode bmap btree could join or split: max depth * block size | |
499 | * And the bmap_finish transaction can free the attr blocks freed giving: | |
500 | * the agf for the ag in which the blocks live: 2 * sector size | |
501 | * the agfl for the ag in which the blocks live: 2 * sector size | |
502 | * the superblock for the free block count: sector size | |
503 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
504 | */ | |
8f794055 | 505 | STATIC uint |
025101dc CH |
506 | xfs_calc_attrrm_reservation( |
507 | struct xfs_mount *mp) | |
8f794055 | 508 | { |
025101dc CH |
509 | return XFS_DQUOT_LOGRES(mp) + |
510 | MAX((mp->m_sb.sb_inodesize + | |
511 | XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) + | |
512 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
513 | 128 * (1 + XFS_DA_NODE_MAXDEPTH + | |
514 | XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))), | |
515 | (2 * mp->m_sb.sb_sectsize + | |
516 | 2 * mp->m_sb.sb_sectsize + | |
517 | mp->m_sb.sb_sectsize + | |
518 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
519 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
520 | } |
521 | ||
025101dc CH |
522 | /* |
523 | * Clearing a bad agino number in an agi hash bucket. | |
524 | */ | |
8f794055 | 525 | STATIC uint |
025101dc CH |
526 | xfs_calc_clear_agi_bucket_reservation( |
527 | struct xfs_mount *mp) | |
8f794055 | 528 | { |
025101dc | 529 | return mp->m_sb.sb_sectsize + 128; |
8f794055 NS |
530 | } |
531 | ||
1da177e4 LT |
532 | /* |
533 | * Initialize the precomputed transaction reservation values | |
534 | * in the mount structure. | |
535 | */ | |
536 | void | |
537 | xfs_trans_init( | |
025101dc | 538 | struct xfs_mount *mp) |
1da177e4 | 539 | { |
025101dc | 540 | struct xfs_trans_reservations *resp = &mp->m_reservations; |
1da177e4 | 541 | |
8f794055 NS |
542 | resp->tr_write = xfs_calc_write_reservation(mp); |
543 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
544 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
545 | resp->tr_link = xfs_calc_link_reservation(mp); | |
546 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
547 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
548 | resp->tr_create = xfs_calc_create_reservation(mp); | |
549 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
550 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
551 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
552 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
553 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
554 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
555 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
556 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
557 | resp->tr_attrset = xfs_calc_attrset_reservation(mp); | |
558 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); | |
559 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
560 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
561 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
562 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
1da177e4 LT |
563 | } |
564 | ||
565 | /* | |
566 | * This routine is called to allocate a transaction structure. | |
567 | * The type parameter indicates the type of the transaction. These | |
568 | * are enumerated in xfs_trans.h. | |
569 | * | |
570 | * Dynamically allocate the transaction structure from the transaction | |
571 | * zone, initialize it, and return it to the caller. | |
572 | */ | |
573 | xfs_trans_t * | |
574 | xfs_trans_alloc( | |
575 | xfs_mount_t *mp, | |
576 | uint type) | |
577 | { | |
b267ce99 | 578 | xfs_wait_for_freeze(mp, SB_FREEZE_TRANS); |
80641dc6 | 579 | return _xfs_trans_alloc(mp, type, KM_SLEEP); |
1da177e4 LT |
580 | } |
581 | ||
582 | xfs_trans_t * | |
583 | _xfs_trans_alloc( | |
584 | xfs_mount_t *mp, | |
80641dc6 CH |
585 | uint type, |
586 | uint memflags) | |
1da177e4 LT |
587 | { |
588 | xfs_trans_t *tp; | |
589 | ||
34327e13 | 590 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 591 | |
80641dc6 | 592 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
593 | tp->t_magic = XFS_TRANS_MAGIC; |
594 | tp->t_type = type; | |
595 | tp->t_mountp = mp; | |
596 | tp->t_items_free = XFS_LIC_NUM_SLOTS; | |
39dab9d7 | 597 | xfs_lic_init(&(tp->t_items)); |
ed3b4d6c | 598 | INIT_LIST_HEAD(&tp->t_busy); |
34327e13 | 599 | return tp; |
1da177e4 LT |
600 | } |
601 | ||
b1c1b5b6 DC |
602 | /* |
603 | * Free the transaction structure. If there is more clean up | |
604 | * to do when the structure is freed, add it here. | |
605 | */ | |
606 | STATIC void | |
607 | xfs_trans_free( | |
ed3b4d6c | 608 | struct xfs_trans *tp) |
b1c1b5b6 | 609 | { |
ed3b4d6c DC |
610 | struct xfs_busy_extent *busyp, *n; |
611 | ||
612 | list_for_each_entry_safe(busyp, n, &tp->t_busy, list) | |
613 | xfs_alloc_busy_clear(tp->t_mountp, busyp); | |
614 | ||
b1c1b5b6 DC |
615 | atomic_dec(&tp->t_mountp->m_active_trans); |
616 | xfs_trans_free_dqinfo(tp); | |
617 | kmem_zone_free(xfs_trans_zone, tp); | |
618 | } | |
619 | ||
1da177e4 LT |
620 | /* |
621 | * This is called to create a new transaction which will share the | |
622 | * permanent log reservation of the given transaction. The remaining | |
623 | * unused block and rt extent reservations are also inherited. This | |
624 | * implies that the original transaction is no longer allowed to allocate | |
625 | * blocks. Locks and log items, however, are no inherited. They must | |
626 | * be added to the new transaction explicitly. | |
627 | */ | |
628 | xfs_trans_t * | |
629 | xfs_trans_dup( | |
630 | xfs_trans_t *tp) | |
631 | { | |
632 | xfs_trans_t *ntp; | |
633 | ||
634 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
635 | ||
636 | /* | |
637 | * Initialize the new transaction structure. | |
638 | */ | |
639 | ntp->t_magic = XFS_TRANS_MAGIC; | |
640 | ntp->t_type = tp->t_type; | |
641 | ntp->t_mountp = tp->t_mountp; | |
642 | ntp->t_items_free = XFS_LIC_NUM_SLOTS; | |
39dab9d7 | 643 | xfs_lic_init(&(ntp->t_items)); |
ed3b4d6c | 644 | INIT_LIST_HEAD(&ntp->t_busy); |
1da177e4 LT |
645 | |
646 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 647 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 648 | |
1da177e4 | 649 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE); |
cc09c0dc | 650 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
651 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
652 | tp->t_blk_res = tp->t_blk_res_used; | |
653 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
654 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 655 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 656 | |
7d095257 | 657 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
658 | |
659 | atomic_inc(&tp->t_mountp->m_active_trans); | |
660 | return ntp; | |
661 | } | |
662 | ||
663 | /* | |
664 | * This is called to reserve free disk blocks and log space for the | |
665 | * given transaction. This must be done before allocating any resources | |
666 | * within the transaction. | |
667 | * | |
668 | * This will return ENOSPC if there are not enough blocks available. | |
669 | * It will sleep waiting for available log space. | |
670 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
671 | * is used by long running transactions. If any one of the reservations | |
672 | * fails then they will all be backed out. | |
673 | * | |
674 | * This does not do quota reservations. That typically is done by the | |
675 | * caller afterwards. | |
676 | */ | |
677 | int | |
678 | xfs_trans_reserve( | |
679 | xfs_trans_t *tp, | |
680 | uint blocks, | |
681 | uint logspace, | |
682 | uint rtextents, | |
683 | uint flags, | |
684 | uint logcount) | |
685 | { | |
686 | int log_flags; | |
59c1b082 NS |
687 | int error = 0; |
688 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
689 | |
690 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 691 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
692 | |
693 | /* | |
694 | * Attempt to reserve the needed disk blocks by decrementing | |
695 | * the number needed from the number available. This will | |
696 | * fail if the count would go below zero. | |
697 | */ | |
698 | if (blocks > 0) { | |
699 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 700 | -((int64_t)blocks), rsvd); |
1da177e4 | 701 | if (error != 0) { |
59c1b082 | 702 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
703 | return (XFS_ERROR(ENOSPC)); |
704 | } | |
705 | tp->t_blk_res += blocks; | |
706 | } | |
707 | ||
708 | /* | |
709 | * Reserve the log space needed for this transaction. | |
710 | */ | |
711 | if (logspace > 0) { | |
712 | ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace)); | |
713 | ASSERT((tp->t_log_count == 0) || | |
714 | (tp->t_log_count == logcount)); | |
715 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
716 | log_flags = XFS_LOG_PERM_RESERV; | |
717 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; | |
718 | } else { | |
719 | ASSERT(tp->t_ticket == NULL); | |
720 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
721 | log_flags = 0; | |
722 | } | |
723 | ||
724 | error = xfs_log_reserve(tp->t_mountp, logspace, logcount, | |
725 | &tp->t_ticket, | |
7e9c6396 | 726 | XFS_TRANSACTION, log_flags, tp->t_type); |
1da177e4 LT |
727 | if (error) { |
728 | goto undo_blocks; | |
729 | } | |
730 | tp->t_log_res = logspace; | |
731 | tp->t_log_count = logcount; | |
732 | } | |
733 | ||
734 | /* | |
735 | * Attempt to reserve the needed realtime extents by decrementing | |
736 | * the number needed from the number available. This will | |
737 | * fail if the count would go below zero. | |
738 | */ | |
739 | if (rtextents > 0) { | |
740 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 741 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
742 | if (error) { |
743 | error = XFS_ERROR(ENOSPC); | |
744 | goto undo_log; | |
745 | } | |
746 | tp->t_rtx_res += rtextents; | |
747 | } | |
748 | ||
749 | return 0; | |
750 | ||
751 | /* | |
752 | * Error cases jump to one of these labels to undo any | |
753 | * reservations which have already been performed. | |
754 | */ | |
755 | undo_log: | |
756 | if (logspace > 0) { | |
757 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
758 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
759 | } else { | |
760 | log_flags = 0; | |
761 | } | |
762 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
763 | tp->t_ticket = NULL; | |
764 | tp->t_log_res = 0; | |
765 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
766 | } | |
767 | ||
768 | undo_blocks: | |
769 | if (blocks > 0) { | |
770 | (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 771 | (int64_t)blocks, rsvd); |
1da177e4 LT |
772 | tp->t_blk_res = 0; |
773 | } | |
774 | ||
59c1b082 | 775 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 776 | |
59c1b082 | 777 | return error; |
1da177e4 LT |
778 | } |
779 | ||
1da177e4 LT |
780 | /* |
781 | * Record the indicated change to the given field for application | |
782 | * to the file system's superblock when the transaction commits. | |
783 | * For now, just store the change in the transaction structure. | |
784 | * | |
785 | * Mark the transaction structure to indicate that the superblock | |
786 | * needs to be updated before committing. | |
92821e2b DC |
787 | * |
788 | * Because we may not be keeping track of allocated/free inodes and | |
789 | * used filesystem blocks in the superblock, we do not mark the | |
790 | * superblock dirty in this transaction if we modify these fields. | |
791 | * We still need to update the transaction deltas so that they get | |
792 | * applied to the incore superblock, but we don't want them to | |
793 | * cause the superblock to get locked and logged if these are the | |
794 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
795 | */ |
796 | void | |
797 | xfs_trans_mod_sb( | |
798 | xfs_trans_t *tp, | |
799 | uint field, | |
20f4ebf2 | 800 | int64_t delta) |
1da177e4 | 801 | { |
92821e2b DC |
802 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
803 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
804 | |
805 | switch (field) { | |
806 | case XFS_TRANS_SB_ICOUNT: | |
807 | tp->t_icount_delta += delta; | |
92821e2b DC |
808 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
809 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
810 | break; |
811 | case XFS_TRANS_SB_IFREE: | |
812 | tp->t_ifree_delta += delta; | |
92821e2b DC |
813 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
814 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
815 | break; |
816 | case XFS_TRANS_SB_FDBLOCKS: | |
817 | /* | |
818 | * Track the number of blocks allocated in the | |
819 | * transaction. Make sure it does not exceed the | |
820 | * number reserved. | |
821 | */ | |
822 | if (delta < 0) { | |
823 | tp->t_blk_res_used += (uint)-delta; | |
824 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
825 | } | |
826 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
827 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
828 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
829 | break; |
830 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
831 | /* | |
832 | * The allocation has already been applied to the | |
833 | * in-core superblock's counter. This should only | |
834 | * be applied to the on-disk superblock. | |
835 | */ | |
836 | ASSERT(delta < 0); | |
837 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
838 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
839 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
840 | break; |
841 | case XFS_TRANS_SB_FREXTENTS: | |
842 | /* | |
843 | * Track the number of blocks allocated in the | |
844 | * transaction. Make sure it does not exceed the | |
845 | * number reserved. | |
846 | */ | |
847 | if (delta < 0) { | |
848 | tp->t_rtx_res_used += (uint)-delta; | |
849 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
850 | } | |
851 | tp->t_frextents_delta += delta; | |
852 | break; | |
853 | case XFS_TRANS_SB_RES_FREXTENTS: | |
854 | /* | |
855 | * The allocation has already been applied to the | |
c41564b5 | 856 | * in-core superblock's counter. This should only |
1da177e4 LT |
857 | * be applied to the on-disk superblock. |
858 | */ | |
859 | ASSERT(delta < 0); | |
860 | tp->t_res_frextents_delta += delta; | |
861 | break; | |
862 | case XFS_TRANS_SB_DBLOCKS: | |
863 | ASSERT(delta > 0); | |
864 | tp->t_dblocks_delta += delta; | |
865 | break; | |
866 | case XFS_TRANS_SB_AGCOUNT: | |
867 | ASSERT(delta > 0); | |
868 | tp->t_agcount_delta += delta; | |
869 | break; | |
870 | case XFS_TRANS_SB_IMAXPCT: | |
871 | tp->t_imaxpct_delta += delta; | |
872 | break; | |
873 | case XFS_TRANS_SB_REXTSIZE: | |
874 | tp->t_rextsize_delta += delta; | |
875 | break; | |
876 | case XFS_TRANS_SB_RBMBLOCKS: | |
877 | tp->t_rbmblocks_delta += delta; | |
878 | break; | |
879 | case XFS_TRANS_SB_RBLOCKS: | |
880 | tp->t_rblocks_delta += delta; | |
881 | break; | |
882 | case XFS_TRANS_SB_REXTENTS: | |
883 | tp->t_rextents_delta += delta; | |
884 | break; | |
885 | case XFS_TRANS_SB_REXTSLOG: | |
886 | tp->t_rextslog_delta += delta; | |
887 | break; | |
888 | default: | |
889 | ASSERT(0); | |
890 | return; | |
891 | } | |
892 | ||
210c6f1c | 893 | tp->t_flags |= flags; |
1da177e4 LT |
894 | } |
895 | ||
896 | /* | |
897 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
898 | * to bring the superblock buffer into the current transaction | |
899 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
900 | * | |
901 | * For now we just look at each field allowed to change and change | |
902 | * it if necessary. | |
903 | */ | |
904 | STATIC void | |
905 | xfs_trans_apply_sb_deltas( | |
906 | xfs_trans_t *tp) | |
907 | { | |
2bdf7cd0 | 908 | xfs_dsb_t *sbp; |
1da177e4 LT |
909 | xfs_buf_t *bp; |
910 | int whole = 0; | |
911 | ||
912 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
913 | sbp = XFS_BUF_TO_SBP(bp); | |
914 | ||
915 | /* | |
916 | * Check that superblock mods match the mods made to AGF counters. | |
917 | */ | |
918 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
919 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
920 | tp->t_ag_btree_delta)); | |
921 | ||
92821e2b DC |
922 | /* |
923 | * Only update the superblock counters if we are logging them | |
924 | */ | |
925 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 926 | if (tp->t_icount_delta) |
413d57c9 | 927 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 928 | if (tp->t_ifree_delta) |
413d57c9 | 929 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 930 | if (tp->t_fdblocks_delta) |
413d57c9 | 931 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 932 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 933 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
934 | } |
935 | ||
2bdf7cd0 | 936 | if (tp->t_frextents_delta) |
413d57c9 | 937 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 938 | if (tp->t_res_frextents_delta) |
413d57c9 | 939 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
940 | |
941 | if (tp->t_dblocks_delta) { | |
413d57c9 | 942 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
943 | whole = 1; |
944 | } | |
2bdf7cd0 | 945 | if (tp->t_agcount_delta) { |
413d57c9 | 946 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
947 | whole = 1; |
948 | } | |
2bdf7cd0 CH |
949 | if (tp->t_imaxpct_delta) { |
950 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
951 | whole = 1; |
952 | } | |
2bdf7cd0 | 953 | if (tp->t_rextsize_delta) { |
413d57c9 | 954 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
955 | whole = 1; |
956 | } | |
2bdf7cd0 | 957 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 958 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
959 | whole = 1; |
960 | } | |
2bdf7cd0 | 961 | if (tp->t_rblocks_delta) { |
413d57c9 | 962 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
963 | whole = 1; |
964 | } | |
2bdf7cd0 | 965 | if (tp->t_rextents_delta) { |
413d57c9 | 966 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
967 | whole = 1; |
968 | } | |
2bdf7cd0 CH |
969 | if (tp->t_rextslog_delta) { |
970 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
971 | whole = 1; |
972 | } | |
973 | ||
974 | if (whole) | |
975 | /* | |
c41564b5 | 976 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 977 | */ |
2bdf7cd0 | 978 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
979 | else |
980 | /* | |
981 | * Since all the modifiable fields are contiguous, we | |
982 | * can get away with this. | |
983 | */ | |
2bdf7cd0 CH |
984 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
985 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 986 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
987 | } |
988 | ||
989 | /* | |
45c34141 DC |
990 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
991 | * and apply superblock counter changes to the in-core superblock. The | |
992 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
993 | * applied to the in-core superblock. The idea is that that has already been | |
994 | * done. | |
1da177e4 LT |
995 | * |
996 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
997 | * However, we have to ensure that we only modify each superblock field only |
998 | * once because the application of the delta values may not be atomic. That can | |
999 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
1000 | * free space counter to put back the entire reservation and then take away | |
1001 | * what we used. | |
1002 | * | |
1003 | * If we are not logging superblock counters, then the inode allocated/free and | |
1004 | * used block counts are not updated in the on disk superblock. In this case, | |
1005 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
1006 | * still need to update the incore superblock with the changes. | |
1da177e4 | 1007 | */ |
71e330b5 | 1008 | void |
1da177e4 LT |
1009 | xfs_trans_unreserve_and_mod_sb( |
1010 | xfs_trans_t *tp) | |
1011 | { | |
1012 | xfs_mod_sb_t msb[14]; /* If you add cases, add entries */ | |
1013 | xfs_mod_sb_t *msbp; | |
92821e2b | 1014 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1015 | /* REFERENCED */ |
1016 | int error; | |
1017 | int rsvd; | |
45c34141 DC |
1018 | int64_t blkdelta = 0; |
1019 | int64_t rtxdelta = 0; | |
1da177e4 LT |
1020 | |
1021 | msbp = msb; | |
1022 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1023 | ||
45c34141 DC |
1024 | /* calculate free blocks delta */ |
1025 | if (tp->t_blk_res > 0) | |
1026 | blkdelta = tp->t_blk_res; | |
1027 | ||
1028 | if ((tp->t_fdblocks_delta != 0) && | |
1029 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
1030 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
1031 | blkdelta += tp->t_fdblocks_delta; | |
1032 | ||
1033 | if (blkdelta != 0) { | |
1da177e4 | 1034 | msbp->msb_field = XFS_SBS_FDBLOCKS; |
45c34141 | 1035 | msbp->msb_delta = blkdelta; |
1da177e4 LT |
1036 | msbp++; |
1037 | } | |
1038 | ||
45c34141 DC |
1039 | /* calculate free realtime extents delta */ |
1040 | if (tp->t_rtx_res > 0) | |
1041 | rtxdelta = tp->t_rtx_res; | |
1042 | ||
1043 | if ((tp->t_frextents_delta != 0) && | |
1044 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
1045 | rtxdelta += tp->t_frextents_delta; | |
1046 | ||
1047 | if (rtxdelta != 0) { | |
1da177e4 | 1048 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 1049 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
1050 | msbp++; |
1051 | } | |
1052 | ||
45c34141 DC |
1053 | /* apply remaining deltas */ |
1054 | ||
92821e2b DC |
1055 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
1056 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1da177e4 LT |
1057 | if (tp->t_icount_delta != 0) { |
1058 | msbp->msb_field = XFS_SBS_ICOUNT; | |
20f4ebf2 | 1059 | msbp->msb_delta = tp->t_icount_delta; |
1da177e4 LT |
1060 | msbp++; |
1061 | } | |
1062 | if (tp->t_ifree_delta != 0) { | |
1063 | msbp->msb_field = XFS_SBS_IFREE; | |
20f4ebf2 | 1064 | msbp->msb_delta = tp->t_ifree_delta; |
1da177e4 LT |
1065 | msbp++; |
1066 | } | |
92821e2b DC |
1067 | } |
1068 | ||
1069 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { | |
1da177e4 LT |
1070 | if (tp->t_dblocks_delta != 0) { |
1071 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 1072 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
1073 | msbp++; |
1074 | } | |
1075 | if (tp->t_agcount_delta != 0) { | |
1076 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 1077 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
1078 | msbp++; |
1079 | } | |
1080 | if (tp->t_imaxpct_delta != 0) { | |
1081 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 1082 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
1083 | msbp++; |
1084 | } | |
1085 | if (tp->t_rextsize_delta != 0) { | |
1086 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 1087 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
1088 | msbp++; |
1089 | } | |
1090 | if (tp->t_rbmblocks_delta != 0) { | |
1091 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 1092 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
1093 | msbp++; |
1094 | } | |
1095 | if (tp->t_rblocks_delta != 0) { | |
1096 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 1097 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
1098 | msbp++; |
1099 | } | |
1100 | if (tp->t_rextents_delta != 0) { | |
1101 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 1102 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
1103 | msbp++; |
1104 | } | |
1105 | if (tp->t_rextslog_delta != 0) { | |
1106 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 1107 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
1108 | msbp++; |
1109 | } | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * If we need to change anything, do it. | |
1114 | */ | |
1115 | if (msbp > msb) { | |
1116 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
1117 | (uint)(msbp - msb), rsvd); | |
1118 | ASSERT(error == 0); | |
1119 | } | |
1120 | } | |
1121 | ||
1da177e4 | 1122 | /* |
0924378a DC |
1123 | * Total up the number of log iovecs needed to commit this |
1124 | * transaction. The transaction itself needs one for the | |
1125 | * transaction header. Ask each dirty item in turn how many | |
1126 | * it needs to get the total. | |
1da177e4 | 1127 | */ |
0924378a DC |
1128 | static uint |
1129 | xfs_trans_count_vecs( | |
b1c1b5b6 | 1130 | struct xfs_trans *tp) |
1da177e4 | 1131 | { |
0924378a DC |
1132 | int nvecs; |
1133 | xfs_log_item_desc_t *lidp; | |
1da177e4 | 1134 | |
0924378a DC |
1135 | nvecs = 1; |
1136 | lidp = xfs_trans_first_item(tp); | |
1137 | ASSERT(lidp != NULL); | |
1da177e4 | 1138 | |
0924378a DC |
1139 | /* In the non-debug case we need to start bailing out if we |
1140 | * didn't find a log_item here, return zero and let trans_commit | |
1141 | * deal with it. | |
1da177e4 | 1142 | */ |
0924378a DC |
1143 | if (lidp == NULL) |
1144 | return 0; | |
1145 | ||
1146 | while (lidp != NULL) { | |
1147 | /* | |
1148 | * Skip items which aren't dirty in this transaction. | |
1149 | */ | |
1150 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
1151 | lidp = xfs_trans_next_item(tp, lidp); | |
1152 | continue; | |
1153 | } | |
1154 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
1155 | nvecs += lidp->lid_size; | |
1156 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 1157 | } |
0924378a DC |
1158 | |
1159 | return nvecs; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Fill in the vector with pointers to data to be logged | |
1164 | * by this transaction. The transaction header takes | |
1165 | * the first vector, and then each dirty item takes the | |
1166 | * number of vectors it indicated it needed in xfs_trans_count_vecs(). | |
1167 | * | |
1168 | * As each item fills in the entries it needs, also pin the item | |
1169 | * so that it cannot be flushed out until the log write completes. | |
1170 | */ | |
1171 | static void | |
1172 | xfs_trans_fill_vecs( | |
1173 | struct xfs_trans *tp, | |
1174 | struct xfs_log_iovec *log_vector) | |
1175 | { | |
1176 | xfs_log_item_desc_t *lidp; | |
1177 | struct xfs_log_iovec *vecp; | |
1178 | uint nitems; | |
1da177e4 LT |
1179 | |
1180 | /* | |
0924378a DC |
1181 | * Skip over the entry for the transaction header, we'll |
1182 | * fill that in at the end. | |
1da177e4 | 1183 | */ |
0924378a DC |
1184 | vecp = log_vector + 1; |
1185 | ||
1186 | nitems = 0; | |
1187 | lidp = xfs_trans_first_item(tp); | |
1188 | ASSERT(lidp); | |
1189 | while (lidp) { | |
1190 | /* Skip items which aren't dirty in this transaction. */ | |
1191 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
1192 | lidp = xfs_trans_next_item(tp, lidp); | |
1193 | continue; | |
1194 | } | |
1195 | ||
1da177e4 | 1196 | /* |
0924378a DC |
1197 | * The item may be marked dirty but not log anything. This can |
1198 | * be used to get called when a transaction is committed. | |
1da177e4 | 1199 | */ |
0924378a DC |
1200 | if (lidp->lid_size) |
1201 | nitems++; | |
1202 | IOP_FORMAT(lidp->lid_item, vecp); | |
1203 | vecp += lidp->lid_size; | |
1204 | IOP_PIN(lidp->lid_item); | |
1205 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 1206 | } |
1da177e4 LT |
1207 | |
1208 | /* | |
0924378a DC |
1209 | * Now that we've counted the number of items in this transaction, fill |
1210 | * in the transaction header. Note that the transaction header does not | |
1211 | * have a log item. | |
1da177e4 | 1212 | */ |
0924378a DC |
1213 | tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC; |
1214 | tp->t_header.th_type = tp->t_type; | |
1215 | tp->t_header.th_num_items = nitems; | |
1216 | log_vector->i_addr = (xfs_caddr_t)&tp->t_header; | |
1217 | log_vector->i_len = sizeof(xfs_trans_header_t); | |
1218 | log_vector->i_type = XLOG_REG_TYPE_TRANSHDR; | |
1219 | } | |
1220 | ||
b1c1b5b6 DC |
1221 | /* |
1222 | * The committed item processing consists of calling the committed routine of | |
1223 | * each logged item, updating the item's position in the AIL if necessary, and | |
1224 | * unpinning each item. If the committed routine returns -1, then do nothing | |
1225 | * further with the item because it may have been freed. | |
1226 | * | |
1227 | * Since items are unlocked when they are copied to the incore log, it is | |
1228 | * possible for two transactions to be completing and manipulating the same | |
1229 | * item simultaneously. The AIL lock will protect the lsn field of each item. | |
1230 | * The value of this field can never go backwards. | |
1231 | * | |
1232 | * We unpin the items after repositioning them in the AIL, because otherwise | |
1233 | * they could be immediately flushed and we'd have to race with the flusher | |
1234 | * trying to pull the item from the AIL as we add it. | |
1235 | */ | |
71e330b5 | 1236 | void |
b1c1b5b6 DC |
1237 | xfs_trans_item_committed( |
1238 | struct xfs_log_item *lip, | |
1239 | xfs_lsn_t commit_lsn, | |
1240 | int aborted) | |
1241 | { | |
1242 | xfs_lsn_t item_lsn; | |
1243 | struct xfs_ail *ailp; | |
1244 | ||
1245 | if (aborted) | |
1246 | lip->li_flags |= XFS_LI_ABORTED; | |
1247 | item_lsn = IOP_COMMITTED(lip, commit_lsn); | |
1248 | ||
1249 | /* If the committed routine returns -1, item has been freed. */ | |
1250 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) | |
1251 | return; | |
1252 | ||
1253 | /* | |
1254 | * If the returned lsn is greater than what it contained before, update | |
1255 | * the location of the item in the AIL. If it is not, then do nothing. | |
1256 | * Items can never move backwards in the AIL. | |
1257 | * | |
1258 | * While the new lsn should usually be greater, it is possible that a | |
1259 | * later transaction completing simultaneously with an earlier one | |
1260 | * using the same item could complete first with a higher lsn. This | |
1261 | * would cause the earlier transaction to fail the test below. | |
1262 | */ | |
1263 | ailp = lip->li_ailp; | |
1264 | spin_lock(&ailp->xa_lock); | |
1265 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) { | |
1266 | /* | |
1267 | * This will set the item's lsn to item_lsn and update the | |
1268 | * position of the item in the AIL. | |
1269 | * | |
1270 | * xfs_trans_ail_update() drops the AIL lock. | |
1271 | */ | |
1272 | xfs_trans_ail_update(ailp, lip, item_lsn); | |
1273 | } else { | |
1274 | spin_unlock(&ailp->xa_lock); | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Now that we've repositioned the item in the AIL, unpin it so it can | |
1279 | * be flushed. Pass information about buffer stale state down from the | |
1280 | * log item flags, if anyone else stales the buffer we do not want to | |
1281 | * pay any attention to it. | |
1282 | */ | |
1283 | IOP_UNPIN(lip); | |
1284 | } | |
1285 | ||
b1c1b5b6 DC |
1286 | /* |
1287 | * This is typically called by the LM when a transaction has been fully | |
1288 | * committed to disk. It needs to unpin the items which have | |
1289 | * been logged by the transaction and update their positions | |
1290 | * in the AIL if necessary. | |
1291 | * | |
1292 | * This also gets called when the transactions didn't get written out | |
1293 | * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then. | |
1294 | */ | |
1295 | STATIC void | |
1296 | xfs_trans_committed( | |
1297 | struct xfs_trans *tp, | |
1298 | int abortflag) | |
1299 | { | |
1300 | xfs_log_item_desc_t *lidp; | |
1301 | xfs_log_item_chunk_t *licp; | |
1302 | xfs_log_item_chunk_t *next_licp; | |
1303 | ||
1304 | /* Call the transaction's completion callback if there is one. */ | |
1305 | if (tp->t_callback != NULL) | |
1306 | tp->t_callback(tp, tp->t_callarg); | |
1307 | ||
1308 | for (lidp = xfs_trans_first_item(tp); | |
1309 | lidp != NULL; | |
1310 | lidp = xfs_trans_next_item(tp, lidp)) { | |
1311 | xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag); | |
1312 | } | |
1313 | ||
1314 | /* free the item chunks, ignoring the embedded chunk */ | |
1315 | for (licp = tp->t_items.lic_next; licp != NULL; licp = next_licp) { | |
1316 | next_licp = licp->lic_next; | |
1317 | kmem_free(licp); | |
1318 | } | |
1319 | ||
b1c1b5b6 DC |
1320 | xfs_trans_free(tp); |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * Called from the trans_commit code when we notice that | |
1325 | * the filesystem is in the middle of a forced shutdown. | |
1326 | */ | |
1327 | STATIC void | |
1328 | xfs_trans_uncommit( | |
1329 | struct xfs_trans *tp, | |
1330 | uint flags) | |
1331 | { | |
1332 | xfs_log_item_desc_t *lidp; | |
1333 | ||
1334 | for (lidp = xfs_trans_first_item(tp); | |
1335 | lidp != NULL; | |
1336 | lidp = xfs_trans_next_item(tp, lidp)) { | |
1337 | /* | |
1338 | * Unpin all but those that aren't dirty. | |
1339 | */ | |
1340 | if (lidp->lid_flags & XFS_LID_DIRTY) | |
1341 | IOP_UNPIN_REMOVE(lidp->lid_item, tp); | |
1342 | } | |
1343 | ||
1344 | xfs_trans_unreserve_and_mod_sb(tp); | |
1345 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1346 | ||
71e330b5 | 1347 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
b1c1b5b6 DC |
1348 | xfs_trans_free(tp); |
1349 | } | |
1350 | ||
0924378a DC |
1351 | /* |
1352 | * Format the transaction direct to the iclog. This isolates the physical | |
1353 | * transaction commit operation from the logical operation and hence allows | |
1354 | * other methods to be introduced without affecting the existing commit path. | |
1355 | */ | |
1356 | static int | |
1357 | xfs_trans_commit_iclog( | |
1358 | struct xfs_mount *mp, | |
1359 | struct xfs_trans *tp, | |
1360 | xfs_lsn_t *commit_lsn, | |
1361 | int flags) | |
1362 | { | |
1363 | int shutdown; | |
1364 | int error; | |
1365 | int log_flags = 0; | |
1366 | struct xlog_in_core *commit_iclog; | |
1367 | #define XFS_TRANS_LOGVEC_COUNT 16 | |
1368 | struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT]; | |
1369 | struct xfs_log_iovec *log_vector; | |
1370 | uint nvec; | |
1371 | ||
1da177e4 LT |
1372 | |
1373 | /* | |
1374 | * Ask each log item how many log_vector entries it will | |
1375 | * need so we can figure out how many to allocate. | |
1376 | * Try to avoid the kmem_alloc() call in the common case | |
1377 | * by using a vector from the stack when it fits. | |
1378 | */ | |
1379 | nvec = xfs_trans_count_vecs(tp); | |
1da177e4 | 1380 | if (nvec == 0) { |
0924378a | 1381 | return ENOMEM; /* triggers a shutdown! */ |
cfcbbbd0 | 1382 | } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) { |
1da177e4 LT |
1383 | log_vector = log_vector_fast; |
1384 | } else { | |
1385 | log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec * | |
1386 | sizeof(xfs_log_iovec_t), | |
1387 | KM_SLEEP); | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * Fill in the log_vector and pin the logged items, and | |
1392 | * then write the transaction to the log. | |
1393 | */ | |
1394 | xfs_trans_fill_vecs(tp, log_vector); | |
1395 | ||
0924378a DC |
1396 | if (flags & XFS_TRANS_RELEASE_LOG_RES) |
1397 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1398 | ||
cfcbbbd0 | 1399 | error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn)); |
1da177e4 | 1400 | |
1da177e4 | 1401 | /* |
cfcbbbd0 NS |
1402 | * The transaction is committed incore here, and can go out to disk |
1403 | * at any time after this call. However, all the items associated | |
1404 | * with the transaction are still locked and pinned in memory. | |
1da177e4 | 1405 | */ |
0924378a | 1406 | *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); |
1da177e4 | 1407 | |
0924378a | 1408 | tp->t_commit_lsn = *commit_lsn; |
ed3b4d6c DC |
1409 | trace_xfs_trans_commit_lsn(tp); |
1410 | ||
0924378a | 1411 | if (nvec > XFS_TRANS_LOGVEC_COUNT) |
f0e2d93c | 1412 | kmem_free(log_vector); |
1da177e4 | 1413 | |
1da177e4 LT |
1414 | /* |
1415 | * If we got a log write error. Unpin the logitems that we | |
1416 | * had pinned, clean up, free trans structure, and return error. | |
1417 | */ | |
0924378a | 1418 | if (error || *commit_lsn == -1) { |
59c1b082 | 1419 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1420 | xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT); |
1421 | return XFS_ERROR(EIO); | |
1422 | } | |
1423 | ||
1424 | /* | |
1425 | * Once the transaction has committed, unused | |
1426 | * reservations need to be released and changes to | |
1427 | * the superblock need to be reflected in the in-core | |
1428 | * version. Do that now. | |
1429 | */ | |
1430 | xfs_trans_unreserve_and_mod_sb(tp); | |
1431 | ||
1da177e4 LT |
1432 | /* |
1433 | * Tell the LM to call the transaction completion routine | |
1434 | * when the log write with LSN commit_lsn completes (e.g. | |
1435 | * when the transaction commit really hits the on-disk log). | |
1436 | * After this call we cannot reference tp, because the call | |
1437 | * can happen at any time and the call will free the transaction | |
1438 | * structure pointed to by tp. The only case where we call | |
1439 | * the completion routine (xfs_trans_committed) directly is | |
1440 | * if the log is turned off on a debug kernel or we're | |
1441 | * running in simulation mode (the log is explicitly turned | |
1442 | * off). | |
1443 | */ | |
1444 | tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed; | |
1445 | tp->t_logcb.cb_arg = tp; | |
1446 | ||
1447 | /* | |
1448 | * We need to pass the iclog buffer which was used for the | |
1449 | * transaction commit record into this function, and attach | |
1450 | * the callback to it. The callback must be attached before | |
1451 | * the items are unlocked to avoid racing with other threads | |
1452 | * waiting for an item to unlock. | |
1453 | */ | |
1454 | shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb)); | |
1455 | ||
1456 | /* | |
1457 | * Mark this thread as no longer being in a transaction | |
1458 | */ | |
59c1b082 | 1459 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1460 | |
1461 | /* | |
1462 | * Once all the items of the transaction have been copied | |
1463 | * to the in core log and the callback is attached, the | |
1464 | * items can be unlocked. | |
1465 | * | |
1466 | * This will free descriptors pointing to items which were | |
1467 | * not logged since there is nothing more to do with them. | |
1468 | * For items which were logged, we will keep pointers to them | |
1469 | * so they can be unpinned after the transaction commits to disk. | |
1470 | * This will also stamp each modified meta-data item with | |
1471 | * the commit lsn of this transaction for dependency tracking | |
1472 | * purposes. | |
1473 | */ | |
0924378a | 1474 | xfs_trans_unlock_items(tp, *commit_lsn); |
1da177e4 LT |
1475 | |
1476 | /* | |
1477 | * If we detected a log error earlier, finish committing | |
1478 | * the transaction now (unpin log items, etc). | |
1479 | * | |
1480 | * Order is critical here, to avoid using the transaction | |
1481 | * pointer after its been freed (by xfs_trans_committed | |
1482 | * either here now, or as a callback). We cannot do this | |
1483 | * step inside xfs_log_notify as was done earlier because | |
1484 | * of this issue. | |
1485 | */ | |
1486 | if (shutdown) | |
1487 | xfs_trans_committed(tp, XFS_LI_ABORTED); | |
1488 | ||
1489 | /* | |
1490 | * Now that the xfs_trans_committed callback has been attached, | |
1491 | * and the items are released we can finally allow the iclog to | |
1492 | * go to disk. | |
1493 | */ | |
0924378a DC |
1494 | return xfs_log_release_iclog(mp, commit_iclog); |
1495 | } | |
1496 | ||
71e330b5 DC |
1497 | /* |
1498 | * Walk the log items and allocate log vector structures for | |
1499 | * each item large enough to fit all the vectors they require. | |
1500 | * Note that this format differs from the old log vector format in | |
1501 | * that there is no transaction header in these log vectors. | |
1502 | */ | |
1503 | STATIC struct xfs_log_vec * | |
1504 | xfs_trans_alloc_log_vecs( | |
1505 | xfs_trans_t *tp) | |
1506 | { | |
1507 | xfs_log_item_desc_t *lidp; | |
1508 | struct xfs_log_vec *lv = NULL; | |
1509 | struct xfs_log_vec *ret_lv = NULL; | |
1510 | ||
1511 | lidp = xfs_trans_first_item(tp); | |
1512 | ||
1513 | /* Bail out if we didn't find a log item. */ | |
1514 | if (!lidp) { | |
1515 | ASSERT(0); | |
1516 | return NULL; | |
1517 | } | |
1518 | ||
1519 | while (lidp != NULL) { | |
1520 | struct xfs_log_vec *new_lv; | |
1521 | ||
1522 | /* Skip items which aren't dirty in this transaction. */ | |
1523 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
1524 | lidp = xfs_trans_next_item(tp, lidp); | |
1525 | continue; | |
1526 | } | |
1527 | ||
1528 | /* Skip items that do not have any vectors for writing */ | |
1529 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
1530 | if (!lidp->lid_size) { | |
1531 | lidp = xfs_trans_next_item(tp, lidp); | |
1532 | continue; | |
1533 | } | |
1534 | ||
1535 | new_lv = kmem_zalloc(sizeof(*new_lv) + | |
1536 | lidp->lid_size * sizeof(struct xfs_log_iovec), | |
1537 | KM_SLEEP); | |
1538 | ||
1539 | /* The allocated iovec region lies beyond the log vector. */ | |
1540 | new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1]; | |
1541 | new_lv->lv_niovecs = lidp->lid_size; | |
1542 | new_lv->lv_item = lidp->lid_item; | |
1543 | if (!ret_lv) | |
1544 | ret_lv = new_lv; | |
1545 | else | |
1546 | lv->lv_next = new_lv; | |
1547 | lv = new_lv; | |
1548 | lidp = xfs_trans_next_item(tp, lidp); | |
1549 | } | |
1550 | ||
1551 | return ret_lv; | |
1552 | } | |
1553 | ||
1554 | static int | |
1555 | xfs_trans_commit_cil( | |
1556 | struct xfs_mount *mp, | |
1557 | struct xfs_trans *tp, | |
1558 | xfs_lsn_t *commit_lsn, | |
1559 | int flags) | |
1560 | { | |
1561 | struct xfs_log_vec *log_vector; | |
1562 | int error; | |
1563 | ||
1564 | /* | |
1565 | * Get each log item to allocate a vector structure for | |
1566 | * the log item to to pass to the log write code. The | |
1567 | * CIL commit code will format the vector and save it away. | |
1568 | */ | |
1569 | log_vector = xfs_trans_alloc_log_vecs(tp); | |
1570 | if (!log_vector) | |
1571 | return ENOMEM; | |
1572 | ||
1573 | error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags); | |
1574 | if (error) | |
1575 | return error; | |
1576 | ||
1577 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
1578 | ||
1579 | /* xfs_trans_free_items() unlocks them first */ | |
1580 | xfs_trans_free_items(tp, *commit_lsn, 0); | |
1581 | xfs_trans_free(tp); | |
1582 | return 0; | |
1583 | } | |
0924378a DC |
1584 | |
1585 | /* | |
1586 | * xfs_trans_commit | |
1587 | * | |
1588 | * Commit the given transaction to the log a/synchronously. | |
1589 | * | |
1590 | * XFS disk error handling mechanism is not based on a typical | |
1591 | * transaction abort mechanism. Logically after the filesystem | |
1592 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1593 | * be durable - ie. committed to disk - because some metadata might | |
1594 | * be inconsistent. In such cases, this returns an error, and the | |
1595 | * caller may assume that all locked objects joined to the transaction | |
1596 | * have already been unlocked as if the commit had succeeded. | |
1597 | * Do not reference the transaction structure after this call. | |
1598 | */ | |
0924378a DC |
1599 | int |
1600 | _xfs_trans_commit( | |
a3ccd2ca CH |
1601 | struct xfs_trans *tp, |
1602 | uint flags, | |
1603 | int *log_flushed) | |
0924378a | 1604 | { |
a3ccd2ca | 1605 | struct xfs_mount *mp = tp->t_mountp; |
0924378a | 1606 | xfs_lsn_t commit_lsn = -1; |
a3ccd2ca | 1607 | int error = 0; |
0924378a DC |
1608 | int log_flags = 0; |
1609 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
0924378a DC |
1610 | |
1611 | /* | |
1612 | * Determine whether this commit is releasing a permanent | |
1613 | * log reservation or not. | |
1614 | */ | |
1615 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1616 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1617 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1618 | } | |
1619 | ||
1620 | /* | |
1621 | * If there is nothing to be logged by the transaction, | |
1622 | * then unlock all of the items associated with the | |
1623 | * transaction and free the transaction structure. | |
1624 | * Also make sure to return any reserved blocks to | |
1625 | * the free pool. | |
1626 | */ | |
a3ccd2ca CH |
1627 | if (!(tp->t_flags & XFS_TRANS_DIRTY)) |
1628 | goto out_unreserve; | |
1629 | ||
1630 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1631 | error = XFS_ERROR(EIO); | |
1632 | goto out_unreserve; | |
0924378a | 1633 | } |
a3ccd2ca | 1634 | |
0924378a DC |
1635 | ASSERT(tp->t_ticket != NULL); |
1636 | ||
1637 | /* | |
1638 | * If we need to update the superblock, then do it now. | |
1639 | */ | |
1640 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1641 | xfs_trans_apply_sb_deltas(tp); | |
1642 | xfs_trans_apply_dquot_deltas(tp); | |
1643 | ||
71e330b5 DC |
1644 | if (mp->m_flags & XFS_MOUNT_DELAYLOG) |
1645 | error = xfs_trans_commit_cil(mp, tp, &commit_lsn, flags); | |
1646 | else | |
1647 | error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags); | |
1648 | ||
0924378a DC |
1649 | if (error == ENOMEM) { |
1650 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
a3ccd2ca CH |
1651 | error = XFS_ERROR(EIO); |
1652 | goto out_unreserve; | |
0924378a | 1653 | } |
1da177e4 LT |
1654 | |
1655 | /* | |
1656 | * If the transaction needs to be synchronous, then force the | |
1657 | * log out now and wait for it. | |
1658 | */ | |
1659 | if (sync) { | |
f538d4da | 1660 | if (!error) { |
a14a348b CH |
1661 | error = _xfs_log_force_lsn(mp, commit_lsn, |
1662 | XFS_LOG_SYNC, log_flushed); | |
f538d4da | 1663 | } |
1da177e4 LT |
1664 | XFS_STATS_INC(xs_trans_sync); |
1665 | } else { | |
1666 | XFS_STATS_INC(xs_trans_async); | |
1667 | } | |
1668 | ||
a3ccd2ca CH |
1669 | return error; |
1670 | ||
1671 | out_unreserve: | |
1672 | xfs_trans_unreserve_and_mod_sb(tp); | |
1673 | ||
1674 | /* | |
1675 | * It is indeed possible for the transaction to be not dirty but | |
1676 | * the dqinfo portion to be. All that means is that we have some | |
1677 | * (non-persistent) quota reservations that need to be unreserved. | |
1678 | */ | |
1679 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1680 | if (tp->t_ticket) { | |
1681 | commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
1682 | if (commit_lsn == -1 && !error) | |
1683 | error = XFS_ERROR(EIO); | |
1684 | } | |
1685 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
71e330b5 | 1686 | xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0); |
a3ccd2ca CH |
1687 | xfs_trans_free(tp); |
1688 | ||
1689 | XFS_STATS_INC(xs_trans_empty); | |
1690 | return error; | |
1da177e4 LT |
1691 | } |
1692 | ||
1da177e4 LT |
1693 | /* |
1694 | * Unlock all of the transaction's items and free the transaction. | |
1695 | * The transaction must not have modified any of its items, because | |
1696 | * there is no way to restore them to their previous state. | |
1697 | * | |
1698 | * If the transaction has made a log reservation, make sure to release | |
1699 | * it as well. | |
1700 | */ | |
1701 | void | |
1702 | xfs_trans_cancel( | |
1703 | xfs_trans_t *tp, | |
1704 | int flags) | |
1705 | { | |
1706 | int log_flags; | |
1707 | #ifdef DEBUG | |
1708 | xfs_log_item_chunk_t *licp; | |
1709 | xfs_log_item_desc_t *lidp; | |
1710 | xfs_log_item_t *lip; | |
1711 | int i; | |
1712 | #endif | |
0733af21 | 1713 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1714 | |
1715 | /* | |
1716 | * See if the caller is being too lazy to figure out if | |
1717 | * the transaction really needs an abort. | |
1718 | */ | |
1719 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1720 | flags &= ~XFS_TRANS_ABORT; | |
1721 | /* | |
1722 | * See if the caller is relying on us to shut down the | |
1723 | * filesystem. This happens in paths where we detect | |
1724 | * corruption and decide to give up. | |
1725 | */ | |
60a204f0 | 1726 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1727 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1728 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1729 | } |
1da177e4 LT |
1730 | #ifdef DEBUG |
1731 | if (!(flags & XFS_TRANS_ABORT)) { | |
1732 | licp = &(tp->t_items); | |
1733 | while (licp != NULL) { | |
1734 | lidp = licp->lic_descs; | |
1735 | for (i = 0; i < licp->lic_unused; i++, lidp++) { | |
39dab9d7 | 1736 | if (xfs_lic_isfree(licp, i)) { |
1da177e4 LT |
1737 | continue; |
1738 | } | |
1739 | ||
1740 | lip = lidp->lid_item; | |
0733af21 | 1741 | if (!XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 LT |
1742 | ASSERT(!(lip->li_type == XFS_LI_EFD)); |
1743 | } | |
1744 | licp = licp->lic_next; | |
1745 | } | |
1746 | } | |
1747 | #endif | |
1748 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1749 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1750 | |
1751 | if (tp->t_ticket) { | |
1752 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1753 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1754 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1755 | } else { | |
1756 | log_flags = 0; | |
1757 | } | |
0733af21 | 1758 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1759 | } |
1760 | ||
1761 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1762 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 1763 | |
71e330b5 | 1764 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
1da177e4 LT |
1765 | xfs_trans_free(tp); |
1766 | } | |
1767 | ||
322ff6b8 NS |
1768 | /* |
1769 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1770 | * the next: permanent transactions are only flushed out when | |
1771 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1772 | * as possible to let chunks of it go to the log. So we commit the | |
1773 | * chunk we've been working on and get a new transaction to continue. | |
1774 | */ | |
1775 | int | |
1776 | xfs_trans_roll( | |
1777 | struct xfs_trans **tpp, | |
1778 | struct xfs_inode *dp) | |
1779 | { | |
1780 | struct xfs_trans *trans; | |
1781 | unsigned int logres, count; | |
1782 | int error; | |
1783 | ||
1784 | /* | |
1785 | * Ensure that the inode is always logged. | |
1786 | */ | |
1787 | trans = *tpp; | |
1788 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1789 | ||
1790 | /* | |
1791 | * Copy the critical parameters from one trans to the next. | |
1792 | */ | |
1793 | logres = trans->t_log_res; | |
1794 | count = trans->t_log_count; | |
1795 | *tpp = xfs_trans_dup(trans); | |
1796 | ||
1797 | /* | |
1798 | * Commit the current transaction. | |
1799 | * If this commit failed, then it'd just unlock those items that | |
1800 | * are not marked ihold. That also means that a filesystem shutdown | |
1801 | * is in progress. The caller takes the responsibility to cancel | |
1802 | * the duplicate transaction that gets returned. | |
1803 | */ | |
1804 | error = xfs_trans_commit(trans, 0); | |
1805 | if (error) | |
1806 | return (error); | |
1807 | ||
1808 | trans = *tpp; | |
1809 | ||
cc09c0dc DC |
1810 | /* |
1811 | * transaction commit worked ok so we can drop the extra ticket | |
1812 | * reference that we gained in xfs_trans_dup() | |
1813 | */ | |
1814 | xfs_log_ticket_put(trans->t_ticket); | |
1815 | ||
1816 | ||
322ff6b8 NS |
1817 | /* |
1818 | * Reserve space in the log for th next transaction. | |
1819 | * This also pushes items in the "AIL", the list of logged items, | |
1820 | * out to disk if they are taking up space at the tail of the log | |
1821 | * that we want to use. This requires that either nothing be locked | |
1822 | * across this call, or that anything that is locked be logged in | |
1823 | * the prior and the next transactions. | |
1824 | */ | |
1825 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1826 | XFS_TRANS_PERM_LOG_RES, count); | |
1827 | /* | |
1828 | * Ensure that the inode is in the new transaction and locked. | |
1829 | */ | |
1830 | if (error) | |
1831 | return error; | |
1832 | ||
1833 | xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL); | |
1834 | xfs_trans_ihold(trans, dp); | |
1835 | return 0; | |
1836 | } |