Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 | 2 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. |
e98c414f | 3 | * Copyright (C) 2010 Red Hat, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | * |
7b718769 NS |
6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
8 | * published by the Free Software Foundation. |
9 | * | |
7b718769 NS |
10 | * This program is distributed in the hope that it would be useful, |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
1da177e4 | 14 | * |
7b718769 NS |
15 | * You should have received a copy of the GNU General Public License |
16 | * along with this program; if not, write the Free Software Foundation, | |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 18 | */ |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 LT |
26 | #include "xfs_mount.h" |
27 | #include "xfs_error.h" | |
a844f451 | 28 | #include "xfs_da_btree.h" |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
a844f451 | 30 | #include "xfs_alloc_btree.h" |
1da177e4 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
32 | #include "xfs_dinode.h" |
33 | #include "xfs_inode.h" | |
a844f451 NS |
34 | #include "xfs_btree.h" |
35 | #include "xfs_ialloc.h" | |
36 | #include "xfs_alloc.h" | |
efc27b52 | 37 | #include "xfs_extent_busy.h" |
1da177e4 | 38 | #include "xfs_bmap.h" |
1da177e4 | 39 | #include "xfs_quota.h" |
4f3b5783 | 40 | #include "xfs_qm.h" |
a844f451 | 41 | #include "xfs_trans_priv.h" |
1da177e4 | 42 | #include "xfs_trans_space.h" |
322ff6b8 | 43 | #include "xfs_inode_item.h" |
4f3b5783 JL |
44 | #include "xfs_log_priv.h" |
45 | #include "xfs_buf_item.h" | |
ed3b4d6c | 46 | #include "xfs_trace.h" |
1da177e4 | 47 | |
8f794055 | 48 | kmem_zone_t *xfs_trans_zone; |
e98c414f | 49 | kmem_zone_t *xfs_log_item_desc_zone; |
1da177e4 | 50 | |
4f3b5783 JL |
51 | /* |
52 | * A buffer has a format structure overhead in the log in addition | |
53 | * to the data, so we need to take this into account when reserving | |
54 | * space in a transaction for a buffer. Round the space required up | |
55 | * to a multiple of 128 bytes so that we don't change the historical | |
56 | * reservation that has been used for this overhead. | |
57 | */ | |
58 | STATIC uint | |
59 | xfs_buf_log_overhead(void) | |
60 | { | |
61 | return round_up(sizeof(struct xlog_op_header) + | |
62 | sizeof(struct xfs_buf_log_format), 128); | |
63 | } | |
64 | ||
65 | /* | |
66 | * Calculate out transaction log reservation per item in bytes. | |
67 | * | |
68 | * The nbufs argument is used to indicate the number of items that | |
69 | * will be changed in a transaction. size is used to tell how many | |
70 | * bytes should be reserved per item. | |
71 | */ | |
72 | STATIC uint | |
73 | xfs_calc_buf_res( | |
74 | uint nbufs, | |
75 | uint size) | |
76 | { | |
77 | return nbufs * (size + xfs_buf_log_overhead()); | |
78 | } | |
025101dc | 79 | |
8f794055 | 80 | /* |
025101dc CH |
81 | * Various log reservation values. |
82 | * | |
83 | * These are based on the size of the file system block because that is what | |
84 | * most transactions manipulate. Each adds in an additional 128 bytes per | |
85 | * item logged to try to account for the overhead of the transaction mechanism. | |
86 | * | |
87 | * Note: Most of the reservations underestimate the number of allocation | |
88 | * groups into which they could free extents in the xfs_bmap_finish() call. | |
89 | * This is because the number in the worst case is quite high and quite | |
90 | * unusual. In order to fix this we need to change xfs_bmap_finish() to free | |
91 | * extents in only a single AG at a time. This will require changes to the | |
92 | * EFI code as well, however, so that the EFI for the extents not freed is | |
93 | * logged again in each transaction. See SGI PV #261917. | |
94 | * | |
95 | * Reservation functions here avoid a huge stack in xfs_trans_init due to | |
96 | * register overflow from temporaries in the calculations. | |
97 | */ | |
98 | ||
99 | ||
100 | /* | |
101 | * In a write transaction we can allocate a maximum of 2 | |
102 | * extents. This gives: | |
103 | * the inode getting the new extents: inode size | |
104 | * the inode's bmap btree: max depth * block size | |
105 | * the agfs of the ags from which the extents are allocated: 2 * sector | |
106 | * the superblock free block counter: sector size | |
107 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
108 | * And the bmap_finish transaction can free bmap blocks in a join: | |
109 | * the agfs of the ags containing the blocks: 2 * sector size | |
110 | * the agfls of the ags containing the blocks: 2 * sector size | |
111 | * the super block free block counter: sector size | |
112 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
8f794055 | 113 | */ |
8f794055 | 114 | STATIC uint |
025101dc CH |
115 | xfs_calc_write_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)) + | |
121 | 2 * mp->m_sb.sb_sectsize + | |
122 | mp->m_sb.sb_sectsize + | |
123 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
124 | 128 * (4 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + | |
125 | XFS_ALLOCFREE_LOG_COUNT(mp, 2))), | |
126 | (2 * mp->m_sb.sb_sectsize + | |
127 | 2 * mp->m_sb.sb_sectsize + | |
128 | mp->m_sb.sb_sectsize + | |
129 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
130 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
131 | } |
132 | ||
025101dc CH |
133 | /* |
134 | * In truncating a file we free up to two extents at once. We can modify: | |
135 | * the inode being truncated: inode size | |
136 | * the inode's bmap btree: (max depth + 1) * block size | |
137 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
138 | * the agf for each of the ags: 4 * sector size | |
139 | * the agfl for each of the ags: 4 * sector size | |
140 | * the super block to reflect the freed blocks: sector size | |
141 | * worst case split in allocation btrees per extent assuming 4 extents: | |
142 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
143 | * the inode btree: max depth * blocksize | |
144 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
145 | */ | |
8f794055 | 146 | STATIC uint |
025101dc CH |
147 | xfs_calc_itruncate_reservation( |
148 | struct xfs_mount *mp) | |
8f794055 | 149 | { |
025101dc CH |
150 | return XFS_DQUOT_LOGRES(mp) + |
151 | MAX((mp->m_sb.sb_inodesize + | |
152 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1) + | |
153 | 128 * (2 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))), | |
154 | (4 * mp->m_sb.sb_sectsize + | |
155 | 4 * mp->m_sb.sb_sectsize + | |
156 | mp->m_sb.sb_sectsize + | |
157 | XFS_ALLOCFREE_LOG_RES(mp, 4) + | |
158 | 128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)) + | |
159 | 128 * 5 + | |
160 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
161 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
162 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
163 | } |
164 | ||
025101dc CH |
165 | /* |
166 | * In renaming a files we can modify: | |
167 | * the four inodes involved: 4 * inode size | |
168 | * the two directory btrees: 2 * (max depth + v2) * dir block size | |
169 | * the two directory bmap btrees: 2 * max depth * block size | |
170 | * And the bmap_finish transaction can free dir and bmap blocks (two sets | |
171 | * of bmap blocks) giving: | |
172 | * the agf for the ags in which the blocks live: 3 * sector size | |
173 | * the agfl for the ags in which the blocks live: 3 * sector size | |
174 | * the superblock for the free block count: sector size | |
175 | * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size | |
176 | */ | |
8f794055 | 177 | STATIC uint |
025101dc CH |
178 | xfs_calc_rename_reservation( |
179 | struct xfs_mount *mp) | |
8f794055 | 180 | { |
025101dc CH |
181 | return XFS_DQUOT_LOGRES(mp) + |
182 | MAX((4 * mp->m_sb.sb_inodesize + | |
183 | 2 * XFS_DIROP_LOG_RES(mp) + | |
184 | 128 * (4 + 2 * XFS_DIROP_LOG_COUNT(mp))), | |
185 | (3 * mp->m_sb.sb_sectsize + | |
186 | 3 * mp->m_sb.sb_sectsize + | |
187 | mp->m_sb.sb_sectsize + | |
188 | XFS_ALLOCFREE_LOG_RES(mp, 3) + | |
189 | 128 * (7 + XFS_ALLOCFREE_LOG_COUNT(mp, 3)))); | |
8f794055 NS |
190 | } |
191 | ||
025101dc CH |
192 | /* |
193 | * For creating a link to an inode: | |
194 | * the parent directory inode: inode size | |
195 | * the linked inode: inode size | |
196 | * the directory btree could split: (max depth + v2) * dir block size | |
197 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
198 | * And the bmap_finish transaction can free some bmap blocks giving: | |
199 | * the agf for the ag in which the blocks live: sector size | |
200 | * the agfl for the ag in which the blocks live: sector size | |
201 | * the superblock for the free block count: sector size | |
202 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
203 | */ | |
8f794055 | 204 | STATIC uint |
025101dc CH |
205 | xfs_calc_link_reservation( |
206 | struct xfs_mount *mp) | |
8f794055 | 207 | { |
025101dc CH |
208 | return XFS_DQUOT_LOGRES(mp) + |
209 | MAX((mp->m_sb.sb_inodesize + | |
210 | mp->m_sb.sb_inodesize + | |
211 | XFS_DIROP_LOG_RES(mp) + | |
212 | 128 * (2 + XFS_DIROP_LOG_COUNT(mp))), | |
213 | (mp->m_sb.sb_sectsize + | |
214 | mp->m_sb.sb_sectsize + | |
215 | mp->m_sb.sb_sectsize + | |
216 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
217 | 128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
218 | } |
219 | ||
025101dc CH |
220 | /* |
221 | * For removing a directory entry we can modify: | |
222 | * the parent directory inode: inode size | |
223 | * the removed inode: inode size | |
224 | * the directory btree could join: (max depth + v2) * dir block size | |
225 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
226 | * And the bmap_finish transaction can free the dir and bmap blocks giving: | |
227 | * the agf for the ag in which the blocks live: 2 * sector size | |
228 | * the agfl for the ag in which the blocks live: 2 * sector size | |
229 | * the superblock for the free block count: sector size | |
230 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
231 | */ | |
8f794055 | 232 | STATIC uint |
025101dc CH |
233 | xfs_calc_remove_reservation( |
234 | struct xfs_mount *mp) | |
8f794055 | 235 | { |
025101dc CH |
236 | return XFS_DQUOT_LOGRES(mp) + |
237 | MAX((mp->m_sb.sb_inodesize + | |
238 | mp->m_sb.sb_inodesize + | |
239 | XFS_DIROP_LOG_RES(mp) + | |
240 | 128 * (2 + XFS_DIROP_LOG_COUNT(mp))), | |
241 | (2 * mp->m_sb.sb_sectsize + | |
242 | 2 * mp->m_sb.sb_sectsize + | |
243 | mp->m_sb.sb_sectsize + | |
244 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
245 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
246 | } |
247 | ||
025101dc CH |
248 | /* |
249 | * For symlink we can modify: | |
250 | * the parent directory inode: inode size | |
251 | * the new inode: inode size | |
252 | * the inode btree entry: 1 block | |
253 | * the directory btree: (max depth + v2) * dir block size | |
254 | * the directory inode's bmap btree: (max depth + v2) * block size | |
255 | * the blocks for the symlink: 1 kB | |
256 | * Or in the first xact we allocate some inodes giving: | |
257 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
258 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
259 | * the inode btree: max depth * blocksize | |
260 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
261 | */ | |
8f794055 | 262 | STATIC uint |
025101dc CH |
263 | xfs_calc_symlink_reservation( |
264 | struct xfs_mount *mp) | |
8f794055 | 265 | { |
025101dc CH |
266 | return XFS_DQUOT_LOGRES(mp) + |
267 | MAX((mp->m_sb.sb_inodesize + | |
268 | mp->m_sb.sb_inodesize + | |
269 | XFS_FSB_TO_B(mp, 1) + | |
270 | XFS_DIROP_LOG_RES(mp) + | |
271 | 1024 + | |
272 | 128 * (4 + XFS_DIROP_LOG_COUNT(mp))), | |
273 | (2 * mp->m_sb.sb_sectsize + | |
274 | XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) + | |
275 | XFS_FSB_TO_B(mp, mp->m_in_maxlevels) + | |
276 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
277 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
278 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
279 | } |
280 | ||
025101dc CH |
281 | /* |
282 | * For create we can modify: | |
283 | * the parent directory inode: inode size | |
284 | * the new inode: inode size | |
285 | * the inode btree entry: block size | |
286 | * the superblock for the nlink flag: sector size | |
287 | * the directory btree: (max depth + v2) * dir block size | |
288 | * the directory inode's bmap btree: (max depth + v2) * block size | |
289 | * Or in the first xact we allocate some inodes giving: | |
290 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
291 | * the superblock for the nlink flag: sector size | |
292 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
293 | * the inode btree: max depth * blocksize | |
294 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
295 | */ | |
8f794055 | 296 | STATIC uint |
025101dc CH |
297 | xfs_calc_create_reservation( |
298 | struct xfs_mount *mp) | |
8f794055 | 299 | { |
025101dc CH |
300 | return XFS_DQUOT_LOGRES(mp) + |
301 | MAX((mp->m_sb.sb_inodesize + | |
302 | mp->m_sb.sb_inodesize + | |
303 | mp->m_sb.sb_sectsize + | |
304 | XFS_FSB_TO_B(mp, 1) + | |
305 | XFS_DIROP_LOG_RES(mp) + | |
306 | 128 * (3 + XFS_DIROP_LOG_COUNT(mp))), | |
307 | (3 * mp->m_sb.sb_sectsize + | |
308 | XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) + | |
309 | XFS_FSB_TO_B(mp, mp->m_in_maxlevels) + | |
310 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
311 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
312 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)))); | |
8f794055 NS |
313 | } |
314 | ||
025101dc CH |
315 | /* |
316 | * Making a new directory is the same as creating a new file. | |
317 | */ | |
8f794055 | 318 | STATIC uint |
025101dc CH |
319 | xfs_calc_mkdir_reservation( |
320 | struct xfs_mount *mp) | |
8f794055 | 321 | { |
025101dc | 322 | return xfs_calc_create_reservation(mp); |
8f794055 NS |
323 | } |
324 | ||
025101dc CH |
325 | /* |
326 | * In freeing an inode we can modify: | |
327 | * the inode being freed: inode size | |
328 | * the super block free inode counter: sector size | |
329 | * the agi hash list and counters: sector size | |
330 | * the inode btree entry: block size | |
331 | * the on disk inode before ours in the agi hash list: inode cluster size | |
332 | * the inode btree: max depth * blocksize | |
333 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
334 | */ | |
8f794055 | 335 | STATIC uint |
025101dc CH |
336 | xfs_calc_ifree_reservation( |
337 | struct xfs_mount *mp) | |
8f794055 | 338 | { |
025101dc CH |
339 | return XFS_DQUOT_LOGRES(mp) + |
340 | mp->m_sb.sb_inodesize + | |
341 | mp->m_sb.sb_sectsize + | |
342 | mp->m_sb.sb_sectsize + | |
343 | XFS_FSB_TO_B(mp, 1) + | |
344 | MAX((__uint16_t)XFS_FSB_TO_B(mp, 1), | |
345 | XFS_INODE_CLUSTER_SIZE(mp)) + | |
346 | 128 * 5 + | |
347 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
348 | 128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels + | |
349 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
350 | } |
351 | ||
025101dc CH |
352 | /* |
353 | * When only changing the inode we log the inode and possibly the superblock | |
354 | * We also add a bit of slop for the transaction stuff. | |
355 | */ | |
8f794055 | 356 | STATIC uint |
025101dc CH |
357 | xfs_calc_ichange_reservation( |
358 | struct xfs_mount *mp) | |
8f794055 | 359 | { |
025101dc CH |
360 | return XFS_DQUOT_LOGRES(mp) + |
361 | mp->m_sb.sb_inodesize + | |
362 | mp->m_sb.sb_sectsize + | |
363 | 512; | |
364 | ||
8f794055 NS |
365 | } |
366 | ||
025101dc CH |
367 | /* |
368 | * Growing the data section of the filesystem. | |
369 | * superblock | |
370 | * agi and agf | |
371 | * allocation btrees | |
372 | */ | |
8f794055 | 373 | STATIC uint |
025101dc CH |
374 | xfs_calc_growdata_reservation( |
375 | struct xfs_mount *mp) | |
8f794055 | 376 | { |
025101dc CH |
377 | return mp->m_sb.sb_sectsize * 3 + |
378 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
379 | 128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
380 | } |
381 | ||
025101dc CH |
382 | /* |
383 | * Growing the rt section of the filesystem. | |
384 | * In the first set of transactions (ALLOC) we allocate space to the | |
385 | * bitmap or summary files. | |
386 | * superblock: sector size | |
387 | * agf of the ag from which the extent is allocated: sector size | |
388 | * bmap btree for bitmap/summary inode: max depth * blocksize | |
389 | * bitmap/summary inode: inode size | |
390 | * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize | |
391 | */ | |
8f794055 | 392 | STATIC uint |
025101dc CH |
393 | xfs_calc_growrtalloc_reservation( |
394 | struct xfs_mount *mp) | |
8f794055 | 395 | { |
025101dc CH |
396 | return 2 * mp->m_sb.sb_sectsize + |
397 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) + | |
398 | mp->m_sb.sb_inodesize + | |
399 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
400 | 128 * (3 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + | |
401 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
402 | } |
403 | ||
025101dc CH |
404 | /* |
405 | * Growing the rt section of the filesystem. | |
406 | * In the second set of transactions (ZERO) we zero the new metadata blocks. | |
407 | * one bitmap/summary block: blocksize | |
408 | */ | |
8f794055 | 409 | STATIC uint |
025101dc CH |
410 | xfs_calc_growrtzero_reservation( |
411 | struct xfs_mount *mp) | |
8f794055 | 412 | { |
025101dc | 413 | return mp->m_sb.sb_blocksize + 128; |
8f794055 NS |
414 | } |
415 | ||
025101dc CH |
416 | /* |
417 | * Growing the rt section of the filesystem. | |
418 | * In the third set of transactions (FREE) we update metadata without | |
419 | * allocating any new blocks. | |
420 | * superblock: sector size | |
421 | * bitmap inode: inode size | |
422 | * summary inode: inode size | |
423 | * one bitmap block: blocksize | |
424 | * summary blocks: new summary size | |
425 | */ | |
8f794055 | 426 | STATIC uint |
025101dc CH |
427 | xfs_calc_growrtfree_reservation( |
428 | struct xfs_mount *mp) | |
8f794055 | 429 | { |
025101dc CH |
430 | return mp->m_sb.sb_sectsize + |
431 | 2 * mp->m_sb.sb_inodesize + | |
432 | mp->m_sb.sb_blocksize + | |
433 | mp->m_rsumsize + | |
434 | 128 * 5; | |
8f794055 NS |
435 | } |
436 | ||
025101dc CH |
437 | /* |
438 | * Logging the inode modification timestamp on a synchronous write. | |
439 | * inode | |
440 | */ | |
8f794055 | 441 | STATIC uint |
025101dc CH |
442 | xfs_calc_swrite_reservation( |
443 | struct xfs_mount *mp) | |
8f794055 | 444 | { |
025101dc | 445 | return mp->m_sb.sb_inodesize + 128; |
8f794055 NS |
446 | } |
447 | ||
025101dc CH |
448 | /* |
449 | * Logging the inode mode bits when writing a setuid/setgid file | |
450 | * inode | |
451 | */ | |
8f794055 NS |
452 | STATIC uint |
453 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
454 | { | |
025101dc | 455 | return mp->m_sb.sb_inodesize + 128; |
8f794055 NS |
456 | } |
457 | ||
025101dc CH |
458 | /* |
459 | * Converting the inode from non-attributed to attributed. | |
460 | * the inode being converted: inode size | |
461 | * agf block and superblock (for block allocation) | |
462 | * the new block (directory sized) | |
463 | * bmap blocks for the new directory block | |
464 | * allocation btrees | |
465 | */ | |
8f794055 | 466 | STATIC uint |
025101dc CH |
467 | xfs_calc_addafork_reservation( |
468 | struct xfs_mount *mp) | |
8f794055 | 469 | { |
025101dc CH |
470 | return XFS_DQUOT_LOGRES(mp) + |
471 | mp->m_sb.sb_inodesize + | |
472 | mp->m_sb.sb_sectsize * 2 + | |
473 | mp->m_dirblksize + | |
474 | XFS_FSB_TO_B(mp, XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1) + | |
475 | XFS_ALLOCFREE_LOG_RES(mp, 1) + | |
476 | 128 * (4 + XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1 + | |
477 | XFS_ALLOCFREE_LOG_COUNT(mp, 1)); | |
8f794055 NS |
478 | } |
479 | ||
025101dc CH |
480 | /* |
481 | * Removing the attribute fork of a file | |
482 | * the inode being truncated: inode size | |
483 | * the inode's bmap btree: max depth * block size | |
484 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
485 | * the agf for each of the ags: 4 * sector size | |
486 | * the agfl for each of the ags: 4 * sector size | |
487 | * the super block to reflect the freed blocks: sector size | |
488 | * worst case split in allocation btrees per extent assuming 4 extents: | |
489 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
490 | */ | |
8f794055 | 491 | STATIC uint |
025101dc CH |
492 | xfs_calc_attrinval_reservation( |
493 | struct xfs_mount *mp) | |
8f794055 | 494 | { |
025101dc CH |
495 | return MAX((mp->m_sb.sb_inodesize + |
496 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
497 | 128 * (1 + XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))), | |
498 | (4 * mp->m_sb.sb_sectsize + | |
499 | 4 * mp->m_sb.sb_sectsize + | |
500 | mp->m_sb.sb_sectsize + | |
501 | XFS_ALLOCFREE_LOG_RES(mp, 4) + | |
502 | 128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)))); | |
8f794055 NS |
503 | } |
504 | ||
025101dc CH |
505 | /* |
506 | * Setting an attribute. | |
507 | * the inode getting the attribute | |
508 | * the superblock for allocations | |
509 | * the agfs extents are allocated from | |
510 | * the attribute btree * max depth | |
511 | * the inode allocation btree | |
512 | * Since attribute transaction space is dependent on the size of the attribute, | |
513 | * the calculation is done partially at mount time and partially at runtime. | |
514 | */ | |
8f794055 | 515 | STATIC uint |
025101dc CH |
516 | xfs_calc_attrset_reservation( |
517 | struct xfs_mount *mp) | |
8f794055 | 518 | { |
025101dc CH |
519 | return XFS_DQUOT_LOGRES(mp) + |
520 | mp->m_sb.sb_inodesize + | |
521 | mp->m_sb.sb_sectsize + | |
522 | XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) + | |
523 | 128 * (2 + XFS_DA_NODE_MAXDEPTH); | |
8f794055 NS |
524 | } |
525 | ||
025101dc CH |
526 | /* |
527 | * Removing an attribute. | |
528 | * the inode: inode size | |
529 | * the attribute btree could join: max depth * block size | |
530 | * the inode bmap btree could join or split: max depth * block size | |
531 | * And the bmap_finish transaction can free the attr blocks freed giving: | |
532 | * the agf for the ag in which the blocks live: 2 * sector size | |
533 | * the agfl for the ag in which the blocks live: 2 * sector size | |
534 | * the superblock for the free block count: sector size | |
535 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
536 | */ | |
8f794055 | 537 | STATIC uint |
025101dc CH |
538 | xfs_calc_attrrm_reservation( |
539 | struct xfs_mount *mp) | |
8f794055 | 540 | { |
025101dc CH |
541 | return XFS_DQUOT_LOGRES(mp) + |
542 | MAX((mp->m_sb.sb_inodesize + | |
543 | XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) + | |
544 | XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
545 | 128 * (1 + XFS_DA_NODE_MAXDEPTH + | |
546 | XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))), | |
547 | (2 * mp->m_sb.sb_sectsize + | |
548 | 2 * mp->m_sb.sb_sectsize + | |
549 | mp->m_sb.sb_sectsize + | |
550 | XFS_ALLOCFREE_LOG_RES(mp, 2) + | |
551 | 128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2)))); | |
8f794055 NS |
552 | } |
553 | ||
025101dc CH |
554 | /* |
555 | * Clearing a bad agino number in an agi hash bucket. | |
556 | */ | |
8f794055 | 557 | STATIC uint |
025101dc CH |
558 | xfs_calc_clear_agi_bucket_reservation( |
559 | struct xfs_mount *mp) | |
8f794055 | 560 | { |
025101dc | 561 | return mp->m_sb.sb_sectsize + 128; |
8f794055 NS |
562 | } |
563 | ||
1da177e4 LT |
564 | /* |
565 | * Initialize the precomputed transaction reservation values | |
566 | * in the mount structure. | |
567 | */ | |
568 | void | |
569 | xfs_trans_init( | |
025101dc | 570 | struct xfs_mount *mp) |
1da177e4 | 571 | { |
025101dc | 572 | struct xfs_trans_reservations *resp = &mp->m_reservations; |
1da177e4 | 573 | |
8f794055 NS |
574 | resp->tr_write = xfs_calc_write_reservation(mp); |
575 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
576 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
577 | resp->tr_link = xfs_calc_link_reservation(mp); | |
578 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
579 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
580 | resp->tr_create = xfs_calc_create_reservation(mp); | |
581 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
582 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
583 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
584 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
585 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
586 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
587 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
588 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
589 | resp->tr_attrset = xfs_calc_attrset_reservation(mp); | |
590 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); | |
591 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
592 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
593 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
594 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
1da177e4 LT |
595 | } |
596 | ||
597 | /* | |
598 | * This routine is called to allocate a transaction structure. | |
599 | * The type parameter indicates the type of the transaction. These | |
600 | * are enumerated in xfs_trans.h. | |
b2ce3974 AE |
601 | * |
602 | * Dynamically allocate the transaction structure from the transaction | |
603 | * zone, initialize it, and return it to the caller. | |
1da177e4 | 604 | */ |
b2ce3974 AE |
605 | xfs_trans_t * |
606 | xfs_trans_alloc( | |
607 | xfs_mount_t *mp, | |
608 | uint type) | |
609 | { | |
d9457dc0 JK |
610 | xfs_trans_t *tp; |
611 | ||
612 | sb_start_intwrite(mp->m_super); | |
613 | tp = _xfs_trans_alloc(mp, type, KM_SLEEP); | |
614 | tp->t_flags |= XFS_TRANS_FREEZE_PROT; | |
615 | return tp; | |
b2ce3974 AE |
616 | } |
617 | ||
618 | xfs_trans_t * | |
1da177e4 | 619 | _xfs_trans_alloc( |
b2ce3974 AE |
620 | xfs_mount_t *mp, |
621 | uint type, | |
77ba7877 | 622 | xfs_km_flags_t memflags) |
1da177e4 | 623 | { |
b2ce3974 | 624 | xfs_trans_t *tp; |
1da177e4 | 625 | |
d9457dc0 | 626 | WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE); |
34327e13 | 627 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 628 | |
80641dc6 | 629 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
630 | tp->t_magic = XFS_TRANS_MAGIC; |
631 | tp->t_type = type; | |
632 | tp->t_mountp = mp; | |
e98c414f | 633 | INIT_LIST_HEAD(&tp->t_items); |
ed3b4d6c | 634 | INIT_LIST_HEAD(&tp->t_busy); |
34327e13 | 635 | return tp; |
1da177e4 LT |
636 | } |
637 | ||
b1c1b5b6 DC |
638 | /* |
639 | * Free the transaction structure. If there is more clean up | |
640 | * to do when the structure is freed, add it here. | |
641 | */ | |
642 | STATIC void | |
643 | xfs_trans_free( | |
ed3b4d6c | 644 | struct xfs_trans *tp) |
b1c1b5b6 | 645 | { |
4ecbfe63 DC |
646 | xfs_extent_busy_sort(&tp->t_busy); |
647 | xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false); | |
ed3b4d6c | 648 | |
b1c1b5b6 | 649 | atomic_dec(&tp->t_mountp->m_active_trans); |
d9457dc0 JK |
650 | if (tp->t_flags & XFS_TRANS_FREEZE_PROT) |
651 | sb_end_intwrite(tp->t_mountp->m_super); | |
b1c1b5b6 DC |
652 | xfs_trans_free_dqinfo(tp); |
653 | kmem_zone_free(xfs_trans_zone, tp); | |
654 | } | |
655 | ||
1da177e4 LT |
656 | /* |
657 | * This is called to create a new transaction which will share the | |
658 | * permanent log reservation of the given transaction. The remaining | |
659 | * unused block and rt extent reservations are also inherited. This | |
660 | * implies that the original transaction is no longer allowed to allocate | |
661 | * blocks. Locks and log items, however, are no inherited. They must | |
662 | * be added to the new transaction explicitly. | |
663 | */ | |
664 | xfs_trans_t * | |
665 | xfs_trans_dup( | |
666 | xfs_trans_t *tp) | |
667 | { | |
668 | xfs_trans_t *ntp; | |
669 | ||
670 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
671 | ||
672 | /* | |
673 | * Initialize the new transaction structure. | |
674 | */ | |
675 | ntp->t_magic = XFS_TRANS_MAGIC; | |
676 | ntp->t_type = tp->t_type; | |
677 | ntp->t_mountp = tp->t_mountp; | |
e98c414f | 678 | INIT_LIST_HEAD(&ntp->t_items); |
ed3b4d6c | 679 | INIT_LIST_HEAD(&ntp->t_busy); |
1da177e4 LT |
680 | |
681 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 682 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 683 | |
d9457dc0 JK |
684 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | |
685 | (tp->t_flags & XFS_TRANS_RESERVE) | | |
686 | (tp->t_flags & XFS_TRANS_FREEZE_PROT); | |
687 | /* We gave our writer reference to the new transaction */ | |
688 | tp->t_flags &= ~XFS_TRANS_FREEZE_PROT; | |
cc09c0dc | 689 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
690 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
691 | tp->t_blk_res = tp->t_blk_res_used; | |
692 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
693 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 694 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 695 | |
7d095257 | 696 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
697 | |
698 | atomic_inc(&tp->t_mountp->m_active_trans); | |
699 | return ntp; | |
700 | } | |
701 | ||
702 | /* | |
703 | * This is called to reserve free disk blocks and log space for the | |
704 | * given transaction. This must be done before allocating any resources | |
705 | * within the transaction. | |
706 | * | |
707 | * This will return ENOSPC if there are not enough blocks available. | |
708 | * It will sleep waiting for available log space. | |
709 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
710 | * is used by long running transactions. If any one of the reservations | |
711 | * fails then they will all be backed out. | |
712 | * | |
713 | * This does not do quota reservations. That typically is done by the | |
714 | * caller afterwards. | |
715 | */ | |
716 | int | |
717 | xfs_trans_reserve( | |
718 | xfs_trans_t *tp, | |
719 | uint blocks, | |
720 | uint logspace, | |
721 | uint rtextents, | |
722 | uint flags, | |
723 | uint logcount) | |
724 | { | |
59c1b082 NS |
725 | int error = 0; |
726 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
727 | |
728 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 729 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
730 | |
731 | /* | |
732 | * Attempt to reserve the needed disk blocks by decrementing | |
733 | * the number needed from the number available. This will | |
734 | * fail if the count would go below zero. | |
735 | */ | |
736 | if (blocks > 0) { | |
96540c78 | 737 | error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, |
20f4ebf2 | 738 | -((int64_t)blocks), rsvd); |
1da177e4 | 739 | if (error != 0) { |
59c1b082 | 740 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
741 | return (XFS_ERROR(ENOSPC)); |
742 | } | |
743 | tp->t_blk_res += blocks; | |
744 | } | |
745 | ||
746 | /* | |
747 | * Reserve the log space needed for this transaction. | |
748 | */ | |
749 | if (logspace > 0) { | |
9006fb91 CH |
750 | bool permanent = false; |
751 | ||
752 | ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace); | |
753 | ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount); | |
754 | ||
1da177e4 | 755 | if (flags & XFS_TRANS_PERM_LOG_RES) { |
1da177e4 | 756 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; |
9006fb91 | 757 | permanent = true; |
1da177e4 LT |
758 | } else { |
759 | ASSERT(tp->t_ticket == NULL); | |
760 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
1da177e4 LT |
761 | } |
762 | ||
9006fb91 CH |
763 | if (tp->t_ticket != NULL) { |
764 | ASSERT(flags & XFS_TRANS_PERM_LOG_RES); | |
765 | error = xfs_log_regrant(tp->t_mountp, tp->t_ticket); | |
766 | } else { | |
767 | error = xfs_log_reserve(tp->t_mountp, logspace, | |
768 | logcount, &tp->t_ticket, | |
769 | XFS_TRANSACTION, permanent, | |
770 | tp->t_type); | |
1da177e4 | 771 | } |
9006fb91 CH |
772 | |
773 | if (error) | |
774 | goto undo_blocks; | |
775 | ||
1da177e4 LT |
776 | tp->t_log_res = logspace; |
777 | tp->t_log_count = logcount; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Attempt to reserve the needed realtime extents by decrementing | |
782 | * the number needed from the number available. This will | |
783 | * fail if the count would go below zero. | |
784 | */ | |
785 | if (rtextents > 0) { | |
786 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 787 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
788 | if (error) { |
789 | error = XFS_ERROR(ENOSPC); | |
790 | goto undo_log; | |
791 | } | |
792 | tp->t_rtx_res += rtextents; | |
793 | } | |
794 | ||
795 | return 0; | |
796 | ||
797 | /* | |
798 | * Error cases jump to one of these labels to undo any | |
799 | * reservations which have already been performed. | |
800 | */ | |
801 | undo_log: | |
802 | if (logspace > 0) { | |
9006fb91 CH |
803 | int log_flags; |
804 | ||
1da177e4 LT |
805 | if (flags & XFS_TRANS_PERM_LOG_RES) { |
806 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
807 | } else { | |
808 | log_flags = 0; | |
809 | } | |
810 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
811 | tp->t_ticket = NULL; | |
812 | tp->t_log_res = 0; | |
813 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
814 | } | |
815 | ||
816 | undo_blocks: | |
817 | if (blocks > 0) { | |
96540c78 | 818 | xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, |
20f4ebf2 | 819 | (int64_t)blocks, rsvd); |
1da177e4 LT |
820 | tp->t_blk_res = 0; |
821 | } | |
822 | ||
59c1b082 | 823 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 824 | |
59c1b082 | 825 | return error; |
1da177e4 LT |
826 | } |
827 | ||
1da177e4 LT |
828 | /* |
829 | * Record the indicated change to the given field for application | |
830 | * to the file system's superblock when the transaction commits. | |
831 | * For now, just store the change in the transaction structure. | |
832 | * | |
833 | * Mark the transaction structure to indicate that the superblock | |
834 | * needs to be updated before committing. | |
92821e2b DC |
835 | * |
836 | * Because we may not be keeping track of allocated/free inodes and | |
837 | * used filesystem blocks in the superblock, we do not mark the | |
838 | * superblock dirty in this transaction if we modify these fields. | |
839 | * We still need to update the transaction deltas so that they get | |
840 | * applied to the incore superblock, but we don't want them to | |
841 | * cause the superblock to get locked and logged if these are the | |
842 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
843 | */ |
844 | void | |
845 | xfs_trans_mod_sb( | |
846 | xfs_trans_t *tp, | |
847 | uint field, | |
20f4ebf2 | 848 | int64_t delta) |
1da177e4 | 849 | { |
92821e2b DC |
850 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
851 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
852 | |
853 | switch (field) { | |
854 | case XFS_TRANS_SB_ICOUNT: | |
855 | tp->t_icount_delta += delta; | |
92821e2b DC |
856 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
857 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
858 | break; |
859 | case XFS_TRANS_SB_IFREE: | |
860 | tp->t_ifree_delta += delta; | |
92821e2b DC |
861 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
862 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
863 | break; |
864 | case XFS_TRANS_SB_FDBLOCKS: | |
865 | /* | |
866 | * Track the number of blocks allocated in the | |
867 | * transaction. Make sure it does not exceed the | |
868 | * number reserved. | |
869 | */ | |
870 | if (delta < 0) { | |
871 | tp->t_blk_res_used += (uint)-delta; | |
872 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
873 | } | |
874 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
875 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
876 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
877 | break; |
878 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
879 | /* | |
880 | * The allocation has already been applied to the | |
881 | * in-core superblock's counter. This should only | |
882 | * be applied to the on-disk superblock. | |
883 | */ | |
884 | ASSERT(delta < 0); | |
885 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
886 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
887 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
888 | break; |
889 | case XFS_TRANS_SB_FREXTENTS: | |
890 | /* | |
891 | * Track the number of blocks allocated in the | |
892 | * transaction. Make sure it does not exceed the | |
893 | * number reserved. | |
894 | */ | |
895 | if (delta < 0) { | |
896 | tp->t_rtx_res_used += (uint)-delta; | |
897 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
898 | } | |
899 | tp->t_frextents_delta += delta; | |
900 | break; | |
901 | case XFS_TRANS_SB_RES_FREXTENTS: | |
902 | /* | |
903 | * The allocation has already been applied to the | |
c41564b5 | 904 | * in-core superblock's counter. This should only |
1da177e4 LT |
905 | * be applied to the on-disk superblock. |
906 | */ | |
907 | ASSERT(delta < 0); | |
908 | tp->t_res_frextents_delta += delta; | |
909 | break; | |
910 | case XFS_TRANS_SB_DBLOCKS: | |
911 | ASSERT(delta > 0); | |
912 | tp->t_dblocks_delta += delta; | |
913 | break; | |
914 | case XFS_TRANS_SB_AGCOUNT: | |
915 | ASSERT(delta > 0); | |
916 | tp->t_agcount_delta += delta; | |
917 | break; | |
918 | case XFS_TRANS_SB_IMAXPCT: | |
919 | tp->t_imaxpct_delta += delta; | |
920 | break; | |
921 | case XFS_TRANS_SB_REXTSIZE: | |
922 | tp->t_rextsize_delta += delta; | |
923 | break; | |
924 | case XFS_TRANS_SB_RBMBLOCKS: | |
925 | tp->t_rbmblocks_delta += delta; | |
926 | break; | |
927 | case XFS_TRANS_SB_RBLOCKS: | |
928 | tp->t_rblocks_delta += delta; | |
929 | break; | |
930 | case XFS_TRANS_SB_REXTENTS: | |
931 | tp->t_rextents_delta += delta; | |
932 | break; | |
933 | case XFS_TRANS_SB_REXTSLOG: | |
934 | tp->t_rextslog_delta += delta; | |
935 | break; | |
936 | default: | |
937 | ASSERT(0); | |
938 | return; | |
939 | } | |
940 | ||
210c6f1c | 941 | tp->t_flags |= flags; |
1da177e4 LT |
942 | } |
943 | ||
944 | /* | |
945 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
946 | * to bring the superblock buffer into the current transaction | |
947 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
948 | * | |
949 | * For now we just look at each field allowed to change and change | |
950 | * it if necessary. | |
951 | */ | |
952 | STATIC void | |
953 | xfs_trans_apply_sb_deltas( | |
954 | xfs_trans_t *tp) | |
955 | { | |
2bdf7cd0 | 956 | xfs_dsb_t *sbp; |
1da177e4 LT |
957 | xfs_buf_t *bp; |
958 | int whole = 0; | |
959 | ||
960 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
961 | sbp = XFS_BUF_TO_SBP(bp); | |
962 | ||
963 | /* | |
964 | * Check that superblock mods match the mods made to AGF counters. | |
965 | */ | |
966 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
967 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
968 | tp->t_ag_btree_delta)); | |
969 | ||
92821e2b DC |
970 | /* |
971 | * Only update the superblock counters if we are logging them | |
972 | */ | |
973 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 974 | if (tp->t_icount_delta) |
413d57c9 | 975 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 976 | if (tp->t_ifree_delta) |
413d57c9 | 977 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 978 | if (tp->t_fdblocks_delta) |
413d57c9 | 979 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 980 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 981 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
982 | } |
983 | ||
2bdf7cd0 | 984 | if (tp->t_frextents_delta) |
413d57c9 | 985 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 986 | if (tp->t_res_frextents_delta) |
413d57c9 | 987 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
988 | |
989 | if (tp->t_dblocks_delta) { | |
413d57c9 | 990 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
991 | whole = 1; |
992 | } | |
2bdf7cd0 | 993 | if (tp->t_agcount_delta) { |
413d57c9 | 994 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
995 | whole = 1; |
996 | } | |
2bdf7cd0 CH |
997 | if (tp->t_imaxpct_delta) { |
998 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
999 | whole = 1; |
1000 | } | |
2bdf7cd0 | 1001 | if (tp->t_rextsize_delta) { |
413d57c9 | 1002 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
1003 | whole = 1; |
1004 | } | |
2bdf7cd0 | 1005 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 1006 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
1007 | whole = 1; |
1008 | } | |
2bdf7cd0 | 1009 | if (tp->t_rblocks_delta) { |
413d57c9 | 1010 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
1011 | whole = 1; |
1012 | } | |
2bdf7cd0 | 1013 | if (tp->t_rextents_delta) { |
413d57c9 | 1014 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
1015 | whole = 1; |
1016 | } | |
2bdf7cd0 CH |
1017 | if (tp->t_rextslog_delta) { |
1018 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
1019 | whole = 1; |
1020 | } | |
1021 | ||
1022 | if (whole) | |
1023 | /* | |
c41564b5 | 1024 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 1025 | */ |
2bdf7cd0 | 1026 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
1027 | else |
1028 | /* | |
1029 | * Since all the modifiable fields are contiguous, we | |
1030 | * can get away with this. | |
1031 | */ | |
2bdf7cd0 CH |
1032 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
1033 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 1034 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | /* | |
45c34141 DC |
1038 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
1039 | * and apply superblock counter changes to the in-core superblock. The | |
1040 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
1041 | * applied to the in-core superblock. The idea is that that has already been | |
1042 | * done. | |
1da177e4 LT |
1043 | * |
1044 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
1045 | * However, we have to ensure that we only modify each superblock field only |
1046 | * once because the application of the delta values may not be atomic. That can | |
1047 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
1048 | * free space counter to put back the entire reservation and then take away | |
1049 | * what we used. | |
1050 | * | |
1051 | * If we are not logging superblock counters, then the inode allocated/free and | |
1052 | * used block counts are not updated in the on disk superblock. In this case, | |
1053 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
1054 | * still need to update the incore superblock with the changes. | |
1da177e4 | 1055 | */ |
71e330b5 | 1056 | void |
1da177e4 LT |
1057 | xfs_trans_unreserve_and_mod_sb( |
1058 | xfs_trans_t *tp) | |
1059 | { | |
1b040712 | 1060 | xfs_mod_sb_t msb[9]; /* If you add cases, add entries */ |
1da177e4 | 1061 | xfs_mod_sb_t *msbp; |
92821e2b | 1062 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1063 | /* REFERENCED */ |
1064 | int error; | |
1065 | int rsvd; | |
45c34141 DC |
1066 | int64_t blkdelta = 0; |
1067 | int64_t rtxdelta = 0; | |
1b040712 CH |
1068 | int64_t idelta = 0; |
1069 | int64_t ifreedelta = 0; | |
1da177e4 LT |
1070 | |
1071 | msbp = msb; | |
1072 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1073 | ||
1b040712 | 1074 | /* calculate deltas */ |
45c34141 DC |
1075 | if (tp->t_blk_res > 0) |
1076 | blkdelta = tp->t_blk_res; | |
45c34141 DC |
1077 | if ((tp->t_fdblocks_delta != 0) && |
1078 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
1079 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
1080 | blkdelta += tp->t_fdblocks_delta; | |
1081 | ||
45c34141 DC |
1082 | if (tp->t_rtx_res > 0) |
1083 | rtxdelta = tp->t_rtx_res; | |
45c34141 DC |
1084 | if ((tp->t_frextents_delta != 0) && |
1085 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
1086 | rtxdelta += tp->t_frextents_delta; | |
1087 | ||
1b040712 CH |
1088 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
1089 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1090 | idelta = tp->t_icount_delta; | |
1091 | ifreedelta = tp->t_ifree_delta; | |
1092 | } | |
1093 | ||
1094 | /* apply the per-cpu counters */ | |
1095 | if (blkdelta) { | |
1096 | error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, | |
1097 | blkdelta, rsvd); | |
1098 | if (error) | |
1099 | goto out; | |
1100 | } | |
1101 | ||
1102 | if (idelta) { | |
1103 | error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, | |
1104 | idelta, rsvd); | |
1105 | if (error) | |
1106 | goto out_undo_fdblocks; | |
1107 | } | |
1108 | ||
1109 | if (ifreedelta) { | |
1110 | error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, | |
1111 | ifreedelta, rsvd); | |
1112 | if (error) | |
1113 | goto out_undo_icount; | |
1114 | } | |
1115 | ||
1116 | /* apply remaining deltas */ | |
45c34141 | 1117 | if (rtxdelta != 0) { |
1da177e4 | 1118 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 1119 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
1120 | msbp++; |
1121 | } | |
1122 | ||
92821e2b | 1123 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { |
1da177e4 LT |
1124 | if (tp->t_dblocks_delta != 0) { |
1125 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 1126 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
1127 | msbp++; |
1128 | } | |
1129 | if (tp->t_agcount_delta != 0) { | |
1130 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 1131 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
1132 | msbp++; |
1133 | } | |
1134 | if (tp->t_imaxpct_delta != 0) { | |
1135 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 1136 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
1137 | msbp++; |
1138 | } | |
1139 | if (tp->t_rextsize_delta != 0) { | |
1140 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 1141 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
1142 | msbp++; |
1143 | } | |
1144 | if (tp->t_rbmblocks_delta != 0) { | |
1145 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 1146 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
1147 | msbp++; |
1148 | } | |
1149 | if (tp->t_rblocks_delta != 0) { | |
1150 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 1151 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
1152 | msbp++; |
1153 | } | |
1154 | if (tp->t_rextents_delta != 0) { | |
1155 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 1156 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
1157 | msbp++; |
1158 | } | |
1159 | if (tp->t_rextslog_delta != 0) { | |
1160 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 1161 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
1162 | msbp++; |
1163 | } | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * If we need to change anything, do it. | |
1168 | */ | |
1169 | if (msbp > msb) { | |
1170 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
1171 | (uint)(msbp - msb), rsvd); | |
1b040712 CH |
1172 | if (error) |
1173 | goto out_undo_ifreecount; | |
1da177e4 | 1174 | } |
1b040712 CH |
1175 | |
1176 | return; | |
1177 | ||
1178 | out_undo_ifreecount: | |
1179 | if (ifreedelta) | |
1180 | xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd); | |
1181 | out_undo_icount: | |
1182 | if (idelta) | |
1183 | xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd); | |
1184 | out_undo_fdblocks: | |
1185 | if (blkdelta) | |
1186 | xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd); | |
1187 | out: | |
1884bd83 | 1188 | ASSERT(error == 0); |
1b040712 | 1189 | return; |
1da177e4 LT |
1190 | } |
1191 | ||
e98c414f CH |
1192 | /* |
1193 | * Add the given log item to the transaction's list of log items. | |
1194 | * | |
1195 | * The log item will now point to its new descriptor with its li_desc field. | |
1196 | */ | |
1197 | void | |
1198 | xfs_trans_add_item( | |
1199 | struct xfs_trans *tp, | |
1200 | struct xfs_log_item *lip) | |
1201 | { | |
1202 | struct xfs_log_item_desc *lidp; | |
1203 | ||
f65020a8 JJ |
1204 | ASSERT(lip->li_mountp == tp->t_mountp); |
1205 | ASSERT(lip->li_ailp == tp->t_mountp->m_ail); | |
e98c414f | 1206 | |
43869706 | 1207 | lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS); |
e98c414f CH |
1208 | |
1209 | lidp->lid_item = lip; | |
1210 | lidp->lid_flags = 0; | |
e98c414f CH |
1211 | list_add_tail(&lidp->lid_trans, &tp->t_items); |
1212 | ||
1213 | lip->li_desc = lidp; | |
1214 | } | |
1215 | ||
1216 | STATIC void | |
1217 | xfs_trans_free_item_desc( | |
1218 | struct xfs_log_item_desc *lidp) | |
1219 | { | |
1220 | list_del_init(&lidp->lid_trans); | |
1221 | kmem_zone_free(xfs_log_item_desc_zone, lidp); | |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * Unlink and free the given descriptor. | |
1226 | */ | |
1227 | void | |
1228 | xfs_trans_del_item( | |
1229 | struct xfs_log_item *lip) | |
1230 | { | |
1231 | xfs_trans_free_item_desc(lip->li_desc); | |
1232 | lip->li_desc = NULL; | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * Unlock all of the items of a transaction and free all the descriptors | |
1237 | * of that transaction. | |
1238 | */ | |
d17c701c | 1239 | void |
e98c414f CH |
1240 | xfs_trans_free_items( |
1241 | struct xfs_trans *tp, | |
1242 | xfs_lsn_t commit_lsn, | |
1243 | int flags) | |
1244 | { | |
1245 | struct xfs_log_item_desc *lidp, *next; | |
1246 | ||
1247 | list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) { | |
1248 | struct xfs_log_item *lip = lidp->lid_item; | |
1249 | ||
1250 | lip->li_desc = NULL; | |
1251 | ||
1252 | if (commit_lsn != NULLCOMMITLSN) | |
1253 | IOP_COMMITTING(lip, commit_lsn); | |
1254 | if (flags & XFS_TRANS_ABORT) | |
1255 | lip->li_flags |= XFS_LI_ABORTED; | |
1256 | IOP_UNLOCK(lip); | |
1257 | ||
1258 | xfs_trans_free_item_desc(lidp); | |
1259 | } | |
1260 | } | |
1261 | ||
0e57f6a3 DC |
1262 | static inline void |
1263 | xfs_log_item_batch_insert( | |
1264 | struct xfs_ail *ailp, | |
1d8c95a3 | 1265 | struct xfs_ail_cursor *cur, |
0e57f6a3 DC |
1266 | struct xfs_log_item **log_items, |
1267 | int nr_items, | |
1268 | xfs_lsn_t commit_lsn) | |
1269 | { | |
1270 | int i; | |
1271 | ||
1272 | spin_lock(&ailp->xa_lock); | |
1273 | /* xfs_trans_ail_update_bulk drops ailp->xa_lock */ | |
1d8c95a3 | 1274 | xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn); |
0e57f6a3 DC |
1275 | |
1276 | for (i = 0; i < nr_items; i++) | |
1277 | IOP_UNPIN(log_items[i], 0); | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Bulk operation version of xfs_trans_committed that takes a log vector of | |
1282 | * items to insert into the AIL. This uses bulk AIL insertion techniques to | |
1283 | * minimise lock traffic. | |
e34a314c DC |
1284 | * |
1285 | * If we are called with the aborted flag set, it is because a log write during | |
1286 | * a CIL checkpoint commit has failed. In this case, all the items in the | |
1287 | * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which | |
1288 | * means that checkpoint commit abort handling is treated exactly the same | |
1289 | * as an iclog write error even though we haven't started any IO yet. Hence in | |
1290 | * this case all we need to do is IOP_COMMITTED processing, followed by an | |
1291 | * IOP_UNPIN(aborted) call. | |
1d8c95a3 DC |
1292 | * |
1293 | * The AIL cursor is used to optimise the insert process. If commit_lsn is not | |
1294 | * at the end of the AIL, the insert cursor avoids the need to walk | |
1295 | * the AIL to find the insertion point on every xfs_log_item_batch_insert() | |
1296 | * call. This saves a lot of needless list walking and is a net win, even | |
1297 | * though it slightly increases that amount of AIL lock traffic to set it up | |
1298 | * and tear it down. | |
0e57f6a3 DC |
1299 | */ |
1300 | void | |
1301 | xfs_trans_committed_bulk( | |
1302 | struct xfs_ail *ailp, | |
1303 | struct xfs_log_vec *log_vector, | |
1304 | xfs_lsn_t commit_lsn, | |
1305 | int aborted) | |
1306 | { | |
1307 | #define LOG_ITEM_BATCH_SIZE 32 | |
1308 | struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE]; | |
1309 | struct xfs_log_vec *lv; | |
1d8c95a3 | 1310 | struct xfs_ail_cursor cur; |
0e57f6a3 DC |
1311 | int i = 0; |
1312 | ||
1d8c95a3 DC |
1313 | spin_lock(&ailp->xa_lock); |
1314 | xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn); | |
1315 | spin_unlock(&ailp->xa_lock); | |
1316 | ||
0e57f6a3 DC |
1317 | /* unpin all the log items */ |
1318 | for (lv = log_vector; lv; lv = lv->lv_next ) { | |
1319 | struct xfs_log_item *lip = lv->lv_item; | |
1320 | xfs_lsn_t item_lsn; | |
1321 | ||
1322 | if (aborted) | |
1323 | lip->li_flags |= XFS_LI_ABORTED; | |
1324 | item_lsn = IOP_COMMITTED(lip, commit_lsn); | |
1325 | ||
1316d4da | 1326 | /* item_lsn of -1 means the item needs no further processing */ |
0e57f6a3 DC |
1327 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) |
1328 | continue; | |
1329 | ||
e34a314c DC |
1330 | /* |
1331 | * if we are aborting the operation, no point in inserting the | |
1332 | * object into the AIL as we are in a shutdown situation. | |
1333 | */ | |
1334 | if (aborted) { | |
1335 | ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount)); | |
1336 | IOP_UNPIN(lip, 1); | |
1337 | continue; | |
1338 | } | |
1339 | ||
0e57f6a3 DC |
1340 | if (item_lsn != commit_lsn) { |
1341 | ||
1342 | /* | |
1343 | * Not a bulk update option due to unusual item_lsn. | |
1344 | * Push into AIL immediately, rechecking the lsn once | |
1d8c95a3 DC |
1345 | * we have the ail lock. Then unpin the item. This does |
1346 | * not affect the AIL cursor the bulk insert path is | |
1347 | * using. | |
0e57f6a3 DC |
1348 | */ |
1349 | spin_lock(&ailp->xa_lock); | |
1350 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) | |
1351 | xfs_trans_ail_update(ailp, lip, item_lsn); | |
1352 | else | |
1353 | spin_unlock(&ailp->xa_lock); | |
1354 | IOP_UNPIN(lip, 0); | |
1355 | continue; | |
1356 | } | |
1357 | ||
1358 | /* Item is a candidate for bulk AIL insert. */ | |
1359 | log_items[i++] = lv->lv_item; | |
1360 | if (i >= LOG_ITEM_BATCH_SIZE) { | |
1d8c95a3 | 1361 | xfs_log_item_batch_insert(ailp, &cur, log_items, |
0e57f6a3 DC |
1362 | LOG_ITEM_BATCH_SIZE, commit_lsn); |
1363 | i = 0; | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | /* make sure we insert the remainder! */ | |
1368 | if (i) | |
1d8c95a3 DC |
1369 | xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn); |
1370 | ||
1371 | spin_lock(&ailp->xa_lock); | |
1372 | xfs_trans_ail_cursor_done(ailp, &cur); | |
1373 | spin_unlock(&ailp->xa_lock); | |
0e57f6a3 DC |
1374 | } |
1375 | ||
0924378a | 1376 | /* |
b1037058 | 1377 | * Commit the given transaction to the log. |
0924378a DC |
1378 | * |
1379 | * XFS disk error handling mechanism is not based on a typical | |
1380 | * transaction abort mechanism. Logically after the filesystem | |
1381 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1382 | * be durable - ie. committed to disk - because some metadata might | |
1383 | * be inconsistent. In such cases, this returns an error, and the | |
1384 | * caller may assume that all locked objects joined to the transaction | |
1385 | * have already been unlocked as if the commit had succeeded. | |
1386 | * Do not reference the transaction structure after this call. | |
1387 | */ | |
0924378a | 1388 | int |
b1037058 | 1389 | xfs_trans_commit( |
a3ccd2ca | 1390 | struct xfs_trans *tp, |
b1037058 | 1391 | uint flags) |
0924378a | 1392 | { |
a3ccd2ca | 1393 | struct xfs_mount *mp = tp->t_mountp; |
0924378a | 1394 | xfs_lsn_t commit_lsn = -1; |
a3ccd2ca | 1395 | int error = 0; |
0924378a DC |
1396 | int log_flags = 0; |
1397 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
0924378a DC |
1398 | |
1399 | /* | |
1400 | * Determine whether this commit is releasing a permanent | |
1401 | * log reservation or not. | |
1402 | */ | |
1403 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1404 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1405 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * If there is nothing to be logged by the transaction, | |
1410 | * then unlock all of the items associated with the | |
1411 | * transaction and free the transaction structure. | |
1412 | * Also make sure to return any reserved blocks to | |
1413 | * the free pool. | |
1414 | */ | |
a3ccd2ca CH |
1415 | if (!(tp->t_flags & XFS_TRANS_DIRTY)) |
1416 | goto out_unreserve; | |
1417 | ||
1418 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1419 | error = XFS_ERROR(EIO); | |
1420 | goto out_unreserve; | |
0924378a | 1421 | } |
a3ccd2ca | 1422 | |
0924378a DC |
1423 | ASSERT(tp->t_ticket != NULL); |
1424 | ||
1425 | /* | |
1426 | * If we need to update the superblock, then do it now. | |
1427 | */ | |
1428 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1429 | xfs_trans_apply_sb_deltas(tp); | |
1430 | xfs_trans_apply_dquot_deltas(tp); | |
1431 | ||
0244b960 | 1432 | error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags); |
0924378a DC |
1433 | if (error == ENOMEM) { |
1434 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
a3ccd2ca CH |
1435 | error = XFS_ERROR(EIO); |
1436 | goto out_unreserve; | |
0924378a | 1437 | } |
1da177e4 | 1438 | |
0244b960 CH |
1439 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1440 | xfs_trans_free(tp); | |
1441 | ||
1da177e4 LT |
1442 | /* |
1443 | * If the transaction needs to be synchronous, then force the | |
1444 | * log out now and wait for it. | |
1445 | */ | |
1446 | if (sync) { | |
f538d4da | 1447 | if (!error) { |
a14a348b | 1448 | error = _xfs_log_force_lsn(mp, commit_lsn, |
b1037058 | 1449 | XFS_LOG_SYNC, NULL); |
f538d4da | 1450 | } |
1da177e4 LT |
1451 | XFS_STATS_INC(xs_trans_sync); |
1452 | } else { | |
1453 | XFS_STATS_INC(xs_trans_async); | |
1454 | } | |
1455 | ||
a3ccd2ca CH |
1456 | return error; |
1457 | ||
1458 | out_unreserve: | |
1459 | xfs_trans_unreserve_and_mod_sb(tp); | |
1460 | ||
1461 | /* | |
1462 | * It is indeed possible for the transaction to be not dirty but | |
1463 | * the dqinfo portion to be. All that means is that we have some | |
1464 | * (non-persistent) quota reservations that need to be unreserved. | |
1465 | */ | |
1466 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1467 | if (tp->t_ticket) { | |
1468 | commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
1469 | if (commit_lsn == -1 && !error) | |
1470 | error = XFS_ERROR(EIO); | |
1471 | } | |
1472 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
71e330b5 | 1473 | xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0); |
a3ccd2ca CH |
1474 | xfs_trans_free(tp); |
1475 | ||
1476 | XFS_STATS_INC(xs_trans_empty); | |
1477 | return error; | |
1da177e4 LT |
1478 | } |
1479 | ||
1da177e4 LT |
1480 | /* |
1481 | * Unlock all of the transaction's items and free the transaction. | |
1482 | * The transaction must not have modified any of its items, because | |
1483 | * there is no way to restore them to their previous state. | |
1484 | * | |
1485 | * If the transaction has made a log reservation, make sure to release | |
1486 | * it as well. | |
1487 | */ | |
1488 | void | |
1489 | xfs_trans_cancel( | |
1490 | xfs_trans_t *tp, | |
1491 | int flags) | |
1492 | { | |
1493 | int log_flags; | |
0733af21 | 1494 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1495 | |
1496 | /* | |
1497 | * See if the caller is being too lazy to figure out if | |
1498 | * the transaction really needs an abort. | |
1499 | */ | |
1500 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1501 | flags &= ~XFS_TRANS_ABORT; | |
1502 | /* | |
1503 | * See if the caller is relying on us to shut down the | |
1504 | * filesystem. This happens in paths where we detect | |
1505 | * corruption and decide to give up. | |
1506 | */ | |
60a204f0 | 1507 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1508 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1509 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1510 | } |
1da177e4 | 1511 | #ifdef DEBUG |
e98c414f CH |
1512 | if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) { |
1513 | struct xfs_log_item_desc *lidp; | |
1514 | ||
1515 | list_for_each_entry(lidp, &tp->t_items, lid_trans) | |
1516 | ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD)); | |
1da177e4 LT |
1517 | } |
1518 | #endif | |
1519 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1520 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1521 | |
1522 | if (tp->t_ticket) { | |
1523 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1524 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1525 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1526 | } else { | |
1527 | log_flags = 0; | |
1528 | } | |
0733af21 | 1529 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1533 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 1534 | |
71e330b5 | 1535 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
1da177e4 LT |
1536 | xfs_trans_free(tp); |
1537 | } | |
1538 | ||
322ff6b8 NS |
1539 | /* |
1540 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1541 | * the next: permanent transactions are only flushed out when | |
1542 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1543 | * as possible to let chunks of it go to the log. So we commit the | |
1544 | * chunk we've been working on and get a new transaction to continue. | |
1545 | */ | |
1546 | int | |
1547 | xfs_trans_roll( | |
1548 | struct xfs_trans **tpp, | |
1549 | struct xfs_inode *dp) | |
1550 | { | |
1551 | struct xfs_trans *trans; | |
1552 | unsigned int logres, count; | |
1553 | int error; | |
1554 | ||
1555 | /* | |
1556 | * Ensure that the inode is always logged. | |
1557 | */ | |
1558 | trans = *tpp; | |
1559 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1560 | ||
1561 | /* | |
1562 | * Copy the critical parameters from one trans to the next. | |
1563 | */ | |
1564 | logres = trans->t_log_res; | |
1565 | count = trans->t_log_count; | |
1566 | *tpp = xfs_trans_dup(trans); | |
1567 | ||
1568 | /* | |
1569 | * Commit the current transaction. | |
1570 | * If this commit failed, then it'd just unlock those items that | |
1571 | * are not marked ihold. That also means that a filesystem shutdown | |
1572 | * is in progress. The caller takes the responsibility to cancel | |
1573 | * the duplicate transaction that gets returned. | |
1574 | */ | |
1575 | error = xfs_trans_commit(trans, 0); | |
1576 | if (error) | |
1577 | return (error); | |
1578 | ||
1579 | trans = *tpp; | |
1580 | ||
cc09c0dc DC |
1581 | /* |
1582 | * transaction commit worked ok so we can drop the extra ticket | |
1583 | * reference that we gained in xfs_trans_dup() | |
1584 | */ | |
1585 | xfs_log_ticket_put(trans->t_ticket); | |
1586 | ||
1587 | ||
322ff6b8 NS |
1588 | /* |
1589 | * Reserve space in the log for th next transaction. | |
1590 | * This also pushes items in the "AIL", the list of logged items, | |
1591 | * out to disk if they are taking up space at the tail of the log | |
1592 | * that we want to use. This requires that either nothing be locked | |
1593 | * across this call, or that anything that is locked be logged in | |
1594 | * the prior and the next transactions. | |
1595 | */ | |
1596 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1597 | XFS_TRANS_PERM_LOG_RES, count); | |
1598 | /* | |
1599 | * Ensure that the inode is in the new transaction and locked. | |
1600 | */ | |
1601 | if (error) | |
1602 | return error; | |
1603 | ||
ddc3415a | 1604 | xfs_trans_ijoin(trans, dp, 0); |
322ff6b8 NS |
1605 | return 0; |
1606 | } |