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 | 118 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
119 | MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
120 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), | |
121 | XFS_FSB_TO_B(mp, 1)) + | |
122 | xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + | |
123 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2), | |
124 | XFS_FSB_TO_B(mp, 1))), | |
125 | (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) + | |
126 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2), | |
127 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
128 | } |
129 | ||
025101dc CH |
130 | /* |
131 | * In truncating a file we free up to two extents at once. We can modify: | |
132 | * the inode being truncated: inode size | |
133 | * the inode's bmap btree: (max depth + 1) * block size | |
134 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
135 | * the agf for each of the ags: 4 * sector size | |
136 | * the agfl for each of the ags: 4 * sector size | |
137 | * the super block to reflect the freed blocks: sector size | |
138 | * worst case split in allocation btrees per extent assuming 4 extents: | |
139 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
140 | * the inode btree: max depth * blocksize | |
141 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
142 | */ | |
8f794055 | 143 | STATIC uint |
025101dc CH |
144 | xfs_calc_itruncate_reservation( |
145 | struct xfs_mount *mp) | |
8f794055 | 146 | { |
025101dc | 147 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
148 | MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
149 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1, | |
150 | XFS_FSB_TO_B(mp, 1))), | |
151 | (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) + | |
152 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 4), | |
153 | XFS_FSB_TO_B(mp, 1)) + | |
154 | xfs_calc_buf_res(5, 0) + | |
155 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
156 | XFS_FSB_TO_B(mp, 1)) + | |
157 | xfs_calc_buf_res(2 + XFS_IALLOC_BLOCKS(mp) + | |
158 | mp->m_in_maxlevels, 0))); | |
8f794055 NS |
159 | } |
160 | ||
025101dc CH |
161 | /* |
162 | * In renaming a files we can modify: | |
163 | * the four inodes involved: 4 * inode size | |
164 | * the two directory btrees: 2 * (max depth + v2) * dir block size | |
165 | * the two directory bmap btrees: 2 * max depth * block size | |
166 | * And the bmap_finish transaction can free dir and bmap blocks (two sets | |
167 | * of bmap blocks) giving: | |
168 | * the agf for the ags in which the blocks live: 3 * sector size | |
169 | * the agfl for the ags in which the blocks live: 3 * sector size | |
170 | * the superblock for the free block count: sector size | |
171 | * the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size | |
172 | */ | |
8f794055 | 173 | STATIC uint |
025101dc CH |
174 | xfs_calc_rename_reservation( |
175 | struct xfs_mount *mp) | |
8f794055 | 176 | { |
025101dc | 177 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
178 | MAX((xfs_calc_buf_res(4, mp->m_sb.sb_inodesize) + |
179 | xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp), | |
180 | XFS_FSB_TO_B(mp, 1))), | |
181 | (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) + | |
182 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 3), | |
183 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
184 | } |
185 | ||
025101dc CH |
186 | /* |
187 | * For creating a link to an inode: | |
188 | * the parent directory inode: inode size | |
189 | * the linked inode: inode size | |
190 | * the directory btree could split: (max depth + v2) * dir block size | |
191 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
192 | * And the bmap_finish transaction can free some bmap blocks giving: | |
193 | * the agf for the ag in which the blocks live: sector size | |
194 | * the agfl for the ag in which the blocks live: sector size | |
195 | * the superblock for the free block count: sector size | |
196 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
197 | */ | |
8f794055 | 198 | STATIC uint |
025101dc CH |
199 | xfs_calc_link_reservation( |
200 | struct xfs_mount *mp) | |
8f794055 | 201 | { |
025101dc | 202 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
203 | MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) + |
204 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), | |
205 | XFS_FSB_TO_B(mp, 1))), | |
206 | (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + | |
207 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
208 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
209 | } |
210 | ||
025101dc CH |
211 | /* |
212 | * For removing a directory entry we can modify: | |
213 | * the parent directory inode: inode size | |
214 | * the removed inode: inode size | |
215 | * the directory btree could join: (max depth + v2) * dir block size | |
216 | * the directory bmap btree could join or split: (max depth + v2) * blocksize | |
217 | * And the bmap_finish transaction can free the dir and bmap blocks giving: | |
218 | * the agf for the ag in which the blocks live: 2 * sector size | |
219 | * the agfl for the ag in which the blocks live: 2 * sector size | |
220 | * the superblock for the free block count: sector size | |
221 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
222 | */ | |
8f794055 | 223 | STATIC uint |
025101dc CH |
224 | xfs_calc_remove_reservation( |
225 | struct xfs_mount *mp) | |
8f794055 | 226 | { |
025101dc | 227 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
228 | MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) + |
229 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), | |
230 | XFS_FSB_TO_B(mp, 1))), | |
231 | (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) + | |
232 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2), | |
233 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
234 | } |
235 | ||
025101dc CH |
236 | /* |
237 | * For symlink we can modify: | |
238 | * the parent directory inode: inode size | |
239 | * the new inode: inode size | |
240 | * the inode btree entry: 1 block | |
241 | * the directory btree: (max depth + v2) * dir block size | |
242 | * the directory inode's bmap btree: (max depth + v2) * block size | |
243 | * the blocks for the symlink: 1 kB | |
244 | * Or in the first xact we allocate some inodes giving: | |
245 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
246 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
247 | * the inode btree: max depth * blocksize | |
248 | * the allocation btrees: 2 trees * (2 * max depth - 1) * block size | |
249 | */ | |
8f794055 | 250 | STATIC uint |
025101dc CH |
251 | xfs_calc_symlink_reservation( |
252 | struct xfs_mount *mp) | |
8f794055 | 253 | { |
025101dc | 254 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
255 | MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) + |
256 | xfs_calc_buf_res(1, XFS_FSB_TO_B(mp, 1)) + | |
257 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), | |
258 | XFS_FSB_TO_B(mp, 1)) + | |
259 | xfs_calc_buf_res(1, 1024)), | |
260 | (xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
261 | xfs_calc_buf_res(XFS_IALLOC_BLOCKS(mp), | |
262 | XFS_FSB_TO_B(mp, 1)) + | |
263 | xfs_calc_buf_res(mp->m_in_maxlevels, | |
264 | XFS_FSB_TO_B(mp, 1)) + | |
265 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
266 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
267 | } |
268 | ||
025101dc CH |
269 | /* |
270 | * For create we can modify: | |
271 | * the parent directory inode: inode size | |
272 | * the new inode: inode size | |
273 | * the inode btree entry: block size | |
274 | * the superblock for the nlink flag: sector size | |
275 | * the directory btree: (max depth + v2) * dir block size | |
276 | * the directory inode's bmap btree: (max depth + v2) * block size | |
277 | * Or in the first xact we allocate some inodes giving: | |
278 | * the agi and agf of the ag getting the new inodes: 2 * sectorsize | |
279 | * the superblock for the nlink flag: sector size | |
280 | * the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize | |
281 | * the inode btree: max depth * blocksize | |
282 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
283 | */ | |
8f794055 | 284 | STATIC uint |
025101dc CH |
285 | xfs_calc_create_reservation( |
286 | struct xfs_mount *mp) | |
8f794055 | 287 | { |
025101dc | 288 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
289 | MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) + |
290 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
291 | (uint)XFS_FSB_TO_B(mp, 1) + | |
292 | xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), | |
293 | XFS_FSB_TO_B(mp, 1))), | |
294 | (xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
025101dc | 295 | mp->m_sb.sb_sectsize + |
5b292ae3 JL |
296 | xfs_calc_buf_res(XFS_IALLOC_BLOCKS(mp), |
297 | XFS_FSB_TO_B(mp, 1)) + | |
298 | xfs_calc_buf_res(mp->m_in_maxlevels, | |
299 | XFS_FSB_TO_B(mp, 1)) + | |
300 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
301 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
302 | } |
303 | ||
025101dc CH |
304 | /* |
305 | * Making a new directory is the same as creating a new file. | |
306 | */ | |
8f794055 | 307 | STATIC uint |
025101dc CH |
308 | xfs_calc_mkdir_reservation( |
309 | struct xfs_mount *mp) | |
8f794055 | 310 | { |
025101dc | 311 | return xfs_calc_create_reservation(mp); |
8f794055 NS |
312 | } |
313 | ||
025101dc CH |
314 | /* |
315 | * In freeing an inode we can modify: | |
316 | * the inode being freed: inode size | |
317 | * the super block free inode counter: sector size | |
318 | * the agi hash list and counters: sector size | |
319 | * the inode btree entry: block size | |
320 | * the on disk inode before ours in the agi hash list: inode cluster size | |
321 | * the inode btree: max depth * blocksize | |
322 | * the allocation btrees: 2 trees * (max depth - 1) * block size | |
323 | */ | |
8f794055 | 324 | STATIC uint |
025101dc CH |
325 | xfs_calc_ifree_reservation( |
326 | struct xfs_mount *mp) | |
8f794055 | 327 | { |
025101dc | 328 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
329 | xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
330 | xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
331 | xfs_calc_buf_res(1, XFS_FSB_TO_B(mp, 1)) + | |
025101dc CH |
332 | MAX((__uint16_t)XFS_FSB_TO_B(mp, 1), |
333 | XFS_INODE_CLUSTER_SIZE(mp)) + | |
5b292ae3 JL |
334 | xfs_calc_buf_res(1, 0) + |
335 | xfs_calc_buf_res(2 + XFS_IALLOC_BLOCKS(mp) + | |
336 | mp->m_in_maxlevels, 0) + | |
337 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
338 | XFS_FSB_TO_B(mp, 1)); | |
8f794055 NS |
339 | } |
340 | ||
025101dc CH |
341 | /* |
342 | * When only changing the inode we log the inode and possibly the superblock | |
343 | * We also add a bit of slop for the transaction stuff. | |
344 | */ | |
8f794055 | 345 | STATIC uint |
025101dc CH |
346 | xfs_calc_ichange_reservation( |
347 | struct xfs_mount *mp) | |
8f794055 | 348 | { |
025101dc CH |
349 | return XFS_DQUOT_LOGRES(mp) + |
350 | mp->m_sb.sb_inodesize + | |
351 | mp->m_sb.sb_sectsize + | |
352 | 512; | |
353 | ||
8f794055 NS |
354 | } |
355 | ||
025101dc CH |
356 | /* |
357 | * Growing the data section of the filesystem. | |
358 | * superblock | |
359 | * agi and agf | |
360 | * allocation btrees | |
361 | */ | |
8f794055 | 362 | STATIC uint |
025101dc CH |
363 | xfs_calc_growdata_reservation( |
364 | struct xfs_mount *mp) | |
8f794055 | 365 | { |
5b292ae3 JL |
366 | return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) + |
367 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
368 | XFS_FSB_TO_B(mp, 1)); | |
8f794055 NS |
369 | } |
370 | ||
025101dc CH |
371 | /* |
372 | * Growing the rt section of the filesystem. | |
373 | * In the first set of transactions (ALLOC) we allocate space to the | |
374 | * bitmap or summary files. | |
375 | * superblock: sector size | |
376 | * agf of the ag from which the extent is allocated: sector size | |
377 | * bmap btree for bitmap/summary inode: max depth * blocksize | |
378 | * bitmap/summary inode: inode size | |
379 | * allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize | |
380 | */ | |
8f794055 | 381 | STATIC uint |
025101dc CH |
382 | xfs_calc_growrtalloc_reservation( |
383 | struct xfs_mount *mp) | |
8f794055 | 384 | { |
5b292ae3 JL |
385 | return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + |
386 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), | |
387 | XFS_FSB_TO_B(mp, 1)) + | |
388 | xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + | |
389 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
390 | XFS_FSB_TO_B(mp, 1)); | |
8f794055 NS |
391 | } |
392 | ||
025101dc CH |
393 | /* |
394 | * Growing the rt section of the filesystem. | |
395 | * In the second set of transactions (ZERO) we zero the new metadata blocks. | |
396 | * one bitmap/summary block: blocksize | |
397 | */ | |
8f794055 | 398 | STATIC uint |
025101dc CH |
399 | xfs_calc_growrtzero_reservation( |
400 | struct xfs_mount *mp) | |
8f794055 | 401 | { |
5b292ae3 | 402 | return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize); |
8f794055 NS |
403 | } |
404 | ||
025101dc CH |
405 | /* |
406 | * Growing the rt section of the filesystem. | |
407 | * In the third set of transactions (FREE) we update metadata without | |
408 | * allocating any new blocks. | |
409 | * superblock: sector size | |
410 | * bitmap inode: inode size | |
411 | * summary inode: inode size | |
412 | * one bitmap block: blocksize | |
413 | * summary blocks: new summary size | |
414 | */ | |
8f794055 | 415 | STATIC uint |
025101dc CH |
416 | xfs_calc_growrtfree_reservation( |
417 | struct xfs_mount *mp) | |
8f794055 | 418 | { |
5b292ae3 JL |
419 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + |
420 | xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) + | |
421 | xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) + | |
422 | xfs_calc_buf_res(1, mp->m_rsumsize); | |
8f794055 NS |
423 | } |
424 | ||
025101dc CH |
425 | /* |
426 | * Logging the inode modification timestamp on a synchronous write. | |
427 | * inode | |
428 | */ | |
8f794055 | 429 | STATIC uint |
025101dc CH |
430 | xfs_calc_swrite_reservation( |
431 | struct xfs_mount *mp) | |
8f794055 | 432 | { |
5b292ae3 | 433 | return xfs_calc_buf_res(1, mp->m_sb.sb_inodesize); |
8f794055 NS |
434 | } |
435 | ||
025101dc CH |
436 | /* |
437 | * Logging the inode mode bits when writing a setuid/setgid file | |
438 | * inode | |
439 | */ | |
8f794055 NS |
440 | STATIC uint |
441 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
442 | { | |
5b292ae3 | 443 | return xfs_calc_buf_res(1, mp->m_sb.sb_inodesize); |
8f794055 NS |
444 | } |
445 | ||
025101dc CH |
446 | /* |
447 | * Converting the inode from non-attributed to attributed. | |
448 | * the inode being converted: inode size | |
449 | * agf block and superblock (for block allocation) | |
450 | * the new block (directory sized) | |
451 | * bmap blocks for the new directory block | |
452 | * allocation btrees | |
453 | */ | |
8f794055 | 454 | STATIC uint |
025101dc CH |
455 | xfs_calc_addafork_reservation( |
456 | struct xfs_mount *mp) | |
8f794055 | 457 | { |
025101dc | 458 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
459 | xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
460 | xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) + | |
461 | xfs_calc_buf_res(1, mp->m_dirblksize) + | |
462 | xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1, | |
463 | XFS_FSB_TO_B(mp, 1)) + | |
464 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1), | |
465 | XFS_FSB_TO_B(mp, 1)); | |
8f794055 NS |
466 | } |
467 | ||
025101dc CH |
468 | /* |
469 | * Removing the attribute fork of a file | |
470 | * the inode being truncated: inode size | |
471 | * the inode's bmap btree: max depth * block size | |
472 | * And the bmap_finish transaction can free the blocks and bmap blocks: | |
473 | * the agf for each of the ags: 4 * sector size | |
474 | * the agfl for each of the ags: 4 * sector size | |
475 | * the super block to reflect the freed blocks: sector size | |
476 | * worst case split in allocation btrees per extent assuming 4 extents: | |
477 | * 4 exts * 2 trees * (2 * max depth - 1) * block size | |
478 | */ | |
8f794055 | 479 | STATIC uint |
025101dc CH |
480 | xfs_calc_attrinval_reservation( |
481 | struct xfs_mount *mp) | |
8f794055 | 482 | { |
5b292ae3 JL |
483 | return MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
484 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK), | |
485 | XFS_FSB_TO_B(mp, 1))), | |
486 | (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) + | |
487 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 4), | |
488 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
489 | } |
490 | ||
025101dc | 491 | /* |
a21cd503 | 492 | * Setting an attribute at mount time. |
025101dc CH |
493 | * the inode getting the attribute |
494 | * the superblock for allocations | |
495 | * the agfs extents are allocated from | |
496 | * the attribute btree * max depth | |
497 | * the inode allocation btree | |
498 | * Since attribute transaction space is dependent on the size of the attribute, | |
a21cd503 JL |
499 | * the calculation is done partially at mount time and partially at runtime(see |
500 | * below). | |
025101dc | 501 | */ |
8f794055 | 502 | STATIC uint |
a21cd503 | 503 | xfs_calc_attrsetm_reservation( |
025101dc | 504 | struct xfs_mount *mp) |
8f794055 | 505 | { |
025101dc | 506 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
507 | xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
508 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
509 | xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1)); | |
8f794055 NS |
510 | } |
511 | ||
a21cd503 JL |
512 | /* |
513 | * Setting an attribute at runtime, transaction space unit per block. | |
514 | * the superblock for allocations: sector size | |
515 | * the inode bmap btree could join or split: max depth * block size | |
516 | * Since the runtime attribute transaction space is dependent on the total | |
517 | * blocks needed for the 1st bmap, here we calculate out the space unit for | |
518 | * one block so that the caller could figure out the total space according | |
519 | * to the attibute extent length in blocks by: ext * XFS_ATTRSETRT_LOG_RES(mp). | |
520 | */ | |
521 | STATIC uint | |
522 | xfs_calc_attrsetrt_reservation( | |
523 | struct xfs_mount *mp) | |
524 | { | |
525 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) + | |
526 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK), | |
527 | XFS_FSB_TO_B(mp, 1)); | |
528 | } | |
529 | ||
025101dc CH |
530 | /* |
531 | * Removing an attribute. | |
532 | * the inode: inode size | |
533 | * the attribute btree could join: max depth * block size | |
534 | * the inode bmap btree could join or split: max depth * block size | |
535 | * And the bmap_finish transaction can free the attr blocks freed giving: | |
536 | * the agf for the ag in which the blocks live: 2 * sector size | |
537 | * the agfl for the ag in which the blocks live: 2 * sector size | |
538 | * the superblock for the free block count: sector size | |
539 | * the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size | |
540 | */ | |
8f794055 | 541 | STATIC uint |
025101dc CH |
542 | xfs_calc_attrrm_reservation( |
543 | struct xfs_mount *mp) | |
8f794055 | 544 | { |
025101dc | 545 | return XFS_DQUOT_LOGRES(mp) + |
5b292ae3 JL |
546 | MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) + |
547 | xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, | |
548 | XFS_FSB_TO_B(mp, 1)) + | |
549 | (uint)XFS_FSB_TO_B(mp, | |
550 | XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) + | |
551 | xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)), | |
552 | (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) + | |
553 | xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2), | |
554 | XFS_FSB_TO_B(mp, 1)))); | |
8f794055 NS |
555 | } |
556 | ||
025101dc CH |
557 | /* |
558 | * Clearing a bad agino number in an agi hash bucket. | |
559 | */ | |
8f794055 | 560 | STATIC uint |
025101dc CH |
561 | xfs_calc_clear_agi_bucket_reservation( |
562 | struct xfs_mount *mp) | |
8f794055 | 563 | { |
5b292ae3 | 564 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); |
8f794055 NS |
565 | } |
566 | ||
b0c10b98 JL |
567 | /* |
568 | * Clearing the quotaflags in the superblock. | |
569 | * the super block for changing quota flags: sector size | |
570 | */ | |
571 | STATIC uint | |
572 | xfs_calc_qm_sbchange_reservation( | |
573 | struct xfs_mount *mp) | |
574 | { | |
575 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
576 | } | |
577 | ||
f0f2df94 JL |
578 | /* |
579 | * Adjusting quota limits. | |
580 | * the xfs_disk_dquot_t: sizeof(struct xfs_disk_dquot) | |
581 | */ | |
582 | STATIC uint | |
583 | xfs_calc_qm_setqlim_reservation( | |
584 | struct xfs_mount *mp) | |
585 | { | |
586 | return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot)); | |
587 | } | |
588 | ||
48001044 JL |
589 | /* |
590 | * Allocating quota on disk if needed. | |
591 | * the write transaction log space: XFS_WRITE_LOG_RES(mp) | |
592 | * the unit of quota allocation: one system block size | |
593 | */ | |
594 | STATIC uint | |
595 | xfs_calc_qm_dqalloc_reservation( | |
596 | struct xfs_mount *mp) | |
597 | { | |
598 | return XFS_WRITE_LOG_RES(mp) + | |
599 | xfs_calc_buf_res(1, | |
600 | XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1); | |
601 | } | |
602 | ||
a1bd9557 JL |
603 | /* |
604 | * Turning off quotas. | |
605 | * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2 | |
606 | * the superblock for the quota flags: sector size | |
607 | */ | |
608 | STATIC uint | |
609 | xfs_calc_qm_quotaoff_reservation( | |
610 | struct xfs_mount *mp) | |
611 | { | |
612 | return sizeof(struct xfs_qoff_logitem) * 2 + | |
613 | xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
614 | } | |
615 | ||
762d7ba6 JL |
616 | /* |
617 | * End of turning off quotas. | |
618 | * the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2 | |
619 | */ | |
620 | STATIC uint | |
621 | xfs_calc_qm_quotaoff_end_reservation( | |
622 | struct xfs_mount *mp) | |
623 | { | |
624 | return sizeof(struct xfs_qoff_logitem) * 2; | |
625 | } | |
626 | ||
a7bd794a JL |
627 | /* |
628 | * Syncing the incore super block changes to disk. | |
629 | * the super block to reflect the changes: sector size | |
630 | */ | |
631 | STATIC uint | |
632 | xfs_calc_sb_reservation( | |
633 | struct xfs_mount *mp) | |
634 | { | |
635 | return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize); | |
636 | } | |
637 | ||
1da177e4 LT |
638 | /* |
639 | * Initialize the precomputed transaction reservation values | |
640 | * in the mount structure. | |
641 | */ | |
642 | void | |
643 | xfs_trans_init( | |
025101dc | 644 | struct xfs_mount *mp) |
1da177e4 | 645 | { |
025101dc | 646 | struct xfs_trans_reservations *resp = &mp->m_reservations; |
1da177e4 | 647 | |
8f794055 NS |
648 | resp->tr_write = xfs_calc_write_reservation(mp); |
649 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
650 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
651 | resp->tr_link = xfs_calc_link_reservation(mp); | |
652 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
653 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
654 | resp->tr_create = xfs_calc_create_reservation(mp); | |
655 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
656 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
657 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
658 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
659 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
660 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
661 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
662 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
a21cd503 JL |
663 | resp->tr_attrsetm = xfs_calc_attrsetm_reservation(mp); |
664 | resp->tr_attrsetrt = xfs_calc_attrsetrt_reservation(mp); | |
8f794055 NS |
665 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); |
666 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
667 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
668 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
669 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
b0c10b98 | 670 | resp->tr_qm_sbchange = xfs_calc_qm_sbchange_reservation(mp); |
f0f2df94 | 671 | resp->tr_qm_setqlim = xfs_calc_qm_setqlim_reservation(mp); |
48001044 | 672 | resp->tr_qm_dqalloc = xfs_calc_qm_dqalloc_reservation(mp); |
a1bd9557 | 673 | resp->tr_qm_quotaoff = xfs_calc_qm_quotaoff_reservation(mp); |
762d7ba6 | 674 | resp->tr_qm_equotaoff = xfs_calc_qm_quotaoff_end_reservation(mp); |
a7bd794a | 675 | resp->tr_sb = xfs_calc_sb_reservation(mp); |
1da177e4 LT |
676 | } |
677 | ||
678 | /* | |
679 | * This routine is called to allocate a transaction structure. | |
680 | * The type parameter indicates the type of the transaction. These | |
681 | * are enumerated in xfs_trans.h. | |
b2ce3974 AE |
682 | * |
683 | * Dynamically allocate the transaction structure from the transaction | |
684 | * zone, initialize it, and return it to the caller. | |
1da177e4 | 685 | */ |
b2ce3974 AE |
686 | xfs_trans_t * |
687 | xfs_trans_alloc( | |
688 | xfs_mount_t *mp, | |
689 | uint type) | |
690 | { | |
d9457dc0 JK |
691 | xfs_trans_t *tp; |
692 | ||
693 | sb_start_intwrite(mp->m_super); | |
694 | tp = _xfs_trans_alloc(mp, type, KM_SLEEP); | |
695 | tp->t_flags |= XFS_TRANS_FREEZE_PROT; | |
696 | return tp; | |
b2ce3974 AE |
697 | } |
698 | ||
699 | xfs_trans_t * | |
1da177e4 | 700 | _xfs_trans_alloc( |
b2ce3974 AE |
701 | xfs_mount_t *mp, |
702 | uint type, | |
77ba7877 | 703 | xfs_km_flags_t memflags) |
1da177e4 | 704 | { |
b2ce3974 | 705 | xfs_trans_t *tp; |
1da177e4 | 706 | |
d9457dc0 | 707 | WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE); |
34327e13 | 708 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 709 | |
80641dc6 | 710 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
711 | tp->t_magic = XFS_TRANS_MAGIC; |
712 | tp->t_type = type; | |
713 | tp->t_mountp = mp; | |
e98c414f | 714 | INIT_LIST_HEAD(&tp->t_items); |
ed3b4d6c | 715 | INIT_LIST_HEAD(&tp->t_busy); |
34327e13 | 716 | return tp; |
1da177e4 LT |
717 | } |
718 | ||
b1c1b5b6 DC |
719 | /* |
720 | * Free the transaction structure. If there is more clean up | |
721 | * to do when the structure is freed, add it here. | |
722 | */ | |
723 | STATIC void | |
724 | xfs_trans_free( | |
ed3b4d6c | 725 | struct xfs_trans *tp) |
b1c1b5b6 | 726 | { |
4ecbfe63 DC |
727 | xfs_extent_busy_sort(&tp->t_busy); |
728 | xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false); | |
ed3b4d6c | 729 | |
b1c1b5b6 | 730 | atomic_dec(&tp->t_mountp->m_active_trans); |
d9457dc0 JK |
731 | if (tp->t_flags & XFS_TRANS_FREEZE_PROT) |
732 | sb_end_intwrite(tp->t_mountp->m_super); | |
b1c1b5b6 DC |
733 | xfs_trans_free_dqinfo(tp); |
734 | kmem_zone_free(xfs_trans_zone, tp); | |
735 | } | |
736 | ||
1da177e4 LT |
737 | /* |
738 | * This is called to create a new transaction which will share the | |
739 | * permanent log reservation of the given transaction. The remaining | |
740 | * unused block and rt extent reservations are also inherited. This | |
741 | * implies that the original transaction is no longer allowed to allocate | |
742 | * blocks. Locks and log items, however, are no inherited. They must | |
743 | * be added to the new transaction explicitly. | |
744 | */ | |
745 | xfs_trans_t * | |
746 | xfs_trans_dup( | |
747 | xfs_trans_t *tp) | |
748 | { | |
749 | xfs_trans_t *ntp; | |
750 | ||
751 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
752 | ||
753 | /* | |
754 | * Initialize the new transaction structure. | |
755 | */ | |
756 | ntp->t_magic = XFS_TRANS_MAGIC; | |
757 | ntp->t_type = tp->t_type; | |
758 | ntp->t_mountp = tp->t_mountp; | |
e98c414f | 759 | INIT_LIST_HEAD(&ntp->t_items); |
ed3b4d6c | 760 | INIT_LIST_HEAD(&ntp->t_busy); |
1da177e4 LT |
761 | |
762 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 763 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 764 | |
d9457dc0 JK |
765 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | |
766 | (tp->t_flags & XFS_TRANS_RESERVE) | | |
767 | (tp->t_flags & XFS_TRANS_FREEZE_PROT); | |
768 | /* We gave our writer reference to the new transaction */ | |
769 | tp->t_flags &= ~XFS_TRANS_FREEZE_PROT; | |
cc09c0dc | 770 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
771 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
772 | tp->t_blk_res = tp->t_blk_res_used; | |
773 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
774 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 775 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 776 | |
7d095257 | 777 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
778 | |
779 | atomic_inc(&tp->t_mountp->m_active_trans); | |
780 | return ntp; | |
781 | } | |
782 | ||
783 | /* | |
784 | * This is called to reserve free disk blocks and log space for the | |
785 | * given transaction. This must be done before allocating any resources | |
786 | * within the transaction. | |
787 | * | |
788 | * This will return ENOSPC if there are not enough blocks available. | |
789 | * It will sleep waiting for available log space. | |
790 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
791 | * is used by long running transactions. If any one of the reservations | |
792 | * fails then they will all be backed out. | |
793 | * | |
794 | * This does not do quota reservations. That typically is done by the | |
795 | * caller afterwards. | |
796 | */ | |
797 | int | |
798 | xfs_trans_reserve( | |
799 | xfs_trans_t *tp, | |
800 | uint blocks, | |
801 | uint logspace, | |
802 | uint rtextents, | |
803 | uint flags, | |
804 | uint logcount) | |
805 | { | |
59c1b082 NS |
806 | int error = 0; |
807 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
808 | |
809 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 810 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
811 | |
812 | /* | |
813 | * Attempt to reserve the needed disk blocks by decrementing | |
814 | * the number needed from the number available. This will | |
815 | * fail if the count would go below zero. | |
816 | */ | |
817 | if (blocks > 0) { | |
96540c78 | 818 | error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, |
20f4ebf2 | 819 | -((int64_t)blocks), rsvd); |
1da177e4 | 820 | if (error != 0) { |
59c1b082 | 821 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
822 | return (XFS_ERROR(ENOSPC)); |
823 | } | |
824 | tp->t_blk_res += blocks; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Reserve the log space needed for this transaction. | |
829 | */ | |
830 | if (logspace > 0) { | |
9006fb91 CH |
831 | bool permanent = false; |
832 | ||
833 | ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace); | |
834 | ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount); | |
835 | ||
1da177e4 | 836 | if (flags & XFS_TRANS_PERM_LOG_RES) { |
1da177e4 | 837 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; |
9006fb91 | 838 | permanent = true; |
1da177e4 LT |
839 | } else { |
840 | ASSERT(tp->t_ticket == NULL); | |
841 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
1da177e4 LT |
842 | } |
843 | ||
9006fb91 CH |
844 | if (tp->t_ticket != NULL) { |
845 | ASSERT(flags & XFS_TRANS_PERM_LOG_RES); | |
846 | error = xfs_log_regrant(tp->t_mountp, tp->t_ticket); | |
847 | } else { | |
848 | error = xfs_log_reserve(tp->t_mountp, logspace, | |
849 | logcount, &tp->t_ticket, | |
850 | XFS_TRANSACTION, permanent, | |
851 | tp->t_type); | |
1da177e4 | 852 | } |
9006fb91 CH |
853 | |
854 | if (error) | |
855 | goto undo_blocks; | |
856 | ||
1da177e4 LT |
857 | tp->t_log_res = logspace; |
858 | tp->t_log_count = logcount; | |
859 | } | |
860 | ||
861 | /* | |
862 | * Attempt to reserve the needed realtime extents by decrementing | |
863 | * the number needed from the number available. This will | |
864 | * fail if the count would go below zero. | |
865 | */ | |
866 | if (rtextents > 0) { | |
867 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 868 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
869 | if (error) { |
870 | error = XFS_ERROR(ENOSPC); | |
871 | goto undo_log; | |
872 | } | |
873 | tp->t_rtx_res += rtextents; | |
874 | } | |
875 | ||
876 | return 0; | |
877 | ||
878 | /* | |
879 | * Error cases jump to one of these labels to undo any | |
880 | * reservations which have already been performed. | |
881 | */ | |
882 | undo_log: | |
883 | if (logspace > 0) { | |
9006fb91 CH |
884 | int log_flags; |
885 | ||
1da177e4 LT |
886 | if (flags & XFS_TRANS_PERM_LOG_RES) { |
887 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
888 | } else { | |
889 | log_flags = 0; | |
890 | } | |
891 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
892 | tp->t_ticket = NULL; | |
893 | tp->t_log_res = 0; | |
894 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
895 | } | |
896 | ||
897 | undo_blocks: | |
898 | if (blocks > 0) { | |
96540c78 | 899 | xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, |
20f4ebf2 | 900 | (int64_t)blocks, rsvd); |
1da177e4 LT |
901 | tp->t_blk_res = 0; |
902 | } | |
903 | ||
59c1b082 | 904 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 905 | |
59c1b082 | 906 | return error; |
1da177e4 LT |
907 | } |
908 | ||
1da177e4 LT |
909 | /* |
910 | * Record the indicated change to the given field for application | |
911 | * to the file system's superblock when the transaction commits. | |
912 | * For now, just store the change in the transaction structure. | |
913 | * | |
914 | * Mark the transaction structure to indicate that the superblock | |
915 | * needs to be updated before committing. | |
92821e2b DC |
916 | * |
917 | * Because we may not be keeping track of allocated/free inodes and | |
918 | * used filesystem blocks in the superblock, we do not mark the | |
919 | * superblock dirty in this transaction if we modify these fields. | |
920 | * We still need to update the transaction deltas so that they get | |
921 | * applied to the incore superblock, but we don't want them to | |
922 | * cause the superblock to get locked and logged if these are the | |
923 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
924 | */ |
925 | void | |
926 | xfs_trans_mod_sb( | |
927 | xfs_trans_t *tp, | |
928 | uint field, | |
20f4ebf2 | 929 | int64_t delta) |
1da177e4 | 930 | { |
92821e2b DC |
931 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
932 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
933 | |
934 | switch (field) { | |
935 | case XFS_TRANS_SB_ICOUNT: | |
936 | tp->t_icount_delta += delta; | |
92821e2b DC |
937 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
938 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
939 | break; |
940 | case XFS_TRANS_SB_IFREE: | |
941 | tp->t_ifree_delta += delta; | |
92821e2b DC |
942 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
943 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
944 | break; |
945 | case XFS_TRANS_SB_FDBLOCKS: | |
946 | /* | |
947 | * Track the number of blocks allocated in the | |
948 | * transaction. Make sure it does not exceed the | |
949 | * number reserved. | |
950 | */ | |
951 | if (delta < 0) { | |
952 | tp->t_blk_res_used += (uint)-delta; | |
953 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
954 | } | |
955 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
956 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
957 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
958 | break; |
959 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
960 | /* | |
961 | * The allocation has already been applied to the | |
962 | * in-core superblock's counter. This should only | |
963 | * be applied to the on-disk superblock. | |
964 | */ | |
965 | ASSERT(delta < 0); | |
966 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
967 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
968 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
969 | break; |
970 | case XFS_TRANS_SB_FREXTENTS: | |
971 | /* | |
972 | * Track the number of blocks allocated in the | |
973 | * transaction. Make sure it does not exceed the | |
974 | * number reserved. | |
975 | */ | |
976 | if (delta < 0) { | |
977 | tp->t_rtx_res_used += (uint)-delta; | |
978 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
979 | } | |
980 | tp->t_frextents_delta += delta; | |
981 | break; | |
982 | case XFS_TRANS_SB_RES_FREXTENTS: | |
983 | /* | |
984 | * The allocation has already been applied to the | |
c41564b5 | 985 | * in-core superblock's counter. This should only |
1da177e4 LT |
986 | * be applied to the on-disk superblock. |
987 | */ | |
988 | ASSERT(delta < 0); | |
989 | tp->t_res_frextents_delta += delta; | |
990 | break; | |
991 | case XFS_TRANS_SB_DBLOCKS: | |
992 | ASSERT(delta > 0); | |
993 | tp->t_dblocks_delta += delta; | |
994 | break; | |
995 | case XFS_TRANS_SB_AGCOUNT: | |
996 | ASSERT(delta > 0); | |
997 | tp->t_agcount_delta += delta; | |
998 | break; | |
999 | case XFS_TRANS_SB_IMAXPCT: | |
1000 | tp->t_imaxpct_delta += delta; | |
1001 | break; | |
1002 | case XFS_TRANS_SB_REXTSIZE: | |
1003 | tp->t_rextsize_delta += delta; | |
1004 | break; | |
1005 | case XFS_TRANS_SB_RBMBLOCKS: | |
1006 | tp->t_rbmblocks_delta += delta; | |
1007 | break; | |
1008 | case XFS_TRANS_SB_RBLOCKS: | |
1009 | tp->t_rblocks_delta += delta; | |
1010 | break; | |
1011 | case XFS_TRANS_SB_REXTENTS: | |
1012 | tp->t_rextents_delta += delta; | |
1013 | break; | |
1014 | case XFS_TRANS_SB_REXTSLOG: | |
1015 | tp->t_rextslog_delta += delta; | |
1016 | break; | |
1017 | default: | |
1018 | ASSERT(0); | |
1019 | return; | |
1020 | } | |
1021 | ||
210c6f1c | 1022 | tp->t_flags |= flags; |
1da177e4 LT |
1023 | } |
1024 | ||
1025 | /* | |
1026 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
1027 | * to bring the superblock buffer into the current transaction | |
1028 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
1029 | * | |
1030 | * For now we just look at each field allowed to change and change | |
1031 | * it if necessary. | |
1032 | */ | |
1033 | STATIC void | |
1034 | xfs_trans_apply_sb_deltas( | |
1035 | xfs_trans_t *tp) | |
1036 | { | |
2bdf7cd0 | 1037 | xfs_dsb_t *sbp; |
1da177e4 LT |
1038 | xfs_buf_t *bp; |
1039 | int whole = 0; | |
1040 | ||
1041 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
1042 | sbp = XFS_BUF_TO_SBP(bp); | |
1043 | ||
1044 | /* | |
1045 | * Check that superblock mods match the mods made to AGF counters. | |
1046 | */ | |
1047 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
1048 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
1049 | tp->t_ag_btree_delta)); | |
1050 | ||
92821e2b DC |
1051 | /* |
1052 | * Only update the superblock counters if we are logging them | |
1053 | */ | |
1054 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 1055 | if (tp->t_icount_delta) |
413d57c9 | 1056 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 1057 | if (tp->t_ifree_delta) |
413d57c9 | 1058 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 1059 | if (tp->t_fdblocks_delta) |
413d57c9 | 1060 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 1061 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 1062 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
1063 | } |
1064 | ||
2bdf7cd0 | 1065 | if (tp->t_frextents_delta) |
413d57c9 | 1066 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 1067 | if (tp->t_res_frextents_delta) |
413d57c9 | 1068 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
1069 | |
1070 | if (tp->t_dblocks_delta) { | |
413d57c9 | 1071 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
1072 | whole = 1; |
1073 | } | |
2bdf7cd0 | 1074 | if (tp->t_agcount_delta) { |
413d57c9 | 1075 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
1076 | whole = 1; |
1077 | } | |
2bdf7cd0 CH |
1078 | if (tp->t_imaxpct_delta) { |
1079 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
1080 | whole = 1; |
1081 | } | |
2bdf7cd0 | 1082 | if (tp->t_rextsize_delta) { |
413d57c9 | 1083 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
1084 | whole = 1; |
1085 | } | |
2bdf7cd0 | 1086 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 1087 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
1088 | whole = 1; |
1089 | } | |
2bdf7cd0 | 1090 | if (tp->t_rblocks_delta) { |
413d57c9 | 1091 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
1092 | whole = 1; |
1093 | } | |
2bdf7cd0 | 1094 | if (tp->t_rextents_delta) { |
413d57c9 | 1095 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
1096 | whole = 1; |
1097 | } | |
2bdf7cd0 CH |
1098 | if (tp->t_rextslog_delta) { |
1099 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
1100 | whole = 1; |
1101 | } | |
1102 | ||
1103 | if (whole) | |
1104 | /* | |
c41564b5 | 1105 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 1106 | */ |
2bdf7cd0 | 1107 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
1108 | else |
1109 | /* | |
1110 | * Since all the modifiable fields are contiguous, we | |
1111 | * can get away with this. | |
1112 | */ | |
2bdf7cd0 CH |
1113 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
1114 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 1115 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
1116 | } |
1117 | ||
1118 | /* | |
45c34141 DC |
1119 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
1120 | * and apply superblock counter changes to the in-core superblock. The | |
1121 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
1122 | * applied to the in-core superblock. The idea is that that has already been | |
1123 | * done. | |
1da177e4 LT |
1124 | * |
1125 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
1126 | * However, we have to ensure that we only modify each superblock field only |
1127 | * once because the application of the delta values may not be atomic. That can | |
1128 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
1129 | * free space counter to put back the entire reservation and then take away | |
1130 | * what we used. | |
1131 | * | |
1132 | * If we are not logging superblock counters, then the inode allocated/free and | |
1133 | * used block counts are not updated in the on disk superblock. In this case, | |
1134 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
1135 | * still need to update the incore superblock with the changes. | |
1da177e4 | 1136 | */ |
71e330b5 | 1137 | void |
1da177e4 LT |
1138 | xfs_trans_unreserve_and_mod_sb( |
1139 | xfs_trans_t *tp) | |
1140 | { | |
1b040712 | 1141 | xfs_mod_sb_t msb[9]; /* If you add cases, add entries */ |
1da177e4 | 1142 | xfs_mod_sb_t *msbp; |
92821e2b | 1143 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1144 | /* REFERENCED */ |
1145 | int error; | |
1146 | int rsvd; | |
45c34141 DC |
1147 | int64_t blkdelta = 0; |
1148 | int64_t rtxdelta = 0; | |
1b040712 CH |
1149 | int64_t idelta = 0; |
1150 | int64_t ifreedelta = 0; | |
1da177e4 LT |
1151 | |
1152 | msbp = msb; | |
1153 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1154 | ||
1b040712 | 1155 | /* calculate deltas */ |
45c34141 DC |
1156 | if (tp->t_blk_res > 0) |
1157 | blkdelta = tp->t_blk_res; | |
45c34141 DC |
1158 | if ((tp->t_fdblocks_delta != 0) && |
1159 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
1160 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
1161 | blkdelta += tp->t_fdblocks_delta; | |
1162 | ||
45c34141 DC |
1163 | if (tp->t_rtx_res > 0) |
1164 | rtxdelta = tp->t_rtx_res; | |
45c34141 DC |
1165 | if ((tp->t_frextents_delta != 0) && |
1166 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
1167 | rtxdelta += tp->t_frextents_delta; | |
1168 | ||
1b040712 CH |
1169 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
1170 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1171 | idelta = tp->t_icount_delta; | |
1172 | ifreedelta = tp->t_ifree_delta; | |
1173 | } | |
1174 | ||
1175 | /* apply the per-cpu counters */ | |
1176 | if (blkdelta) { | |
1177 | error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, | |
1178 | blkdelta, rsvd); | |
1179 | if (error) | |
1180 | goto out; | |
1181 | } | |
1182 | ||
1183 | if (idelta) { | |
1184 | error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, | |
1185 | idelta, rsvd); | |
1186 | if (error) | |
1187 | goto out_undo_fdblocks; | |
1188 | } | |
1189 | ||
1190 | if (ifreedelta) { | |
1191 | error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, | |
1192 | ifreedelta, rsvd); | |
1193 | if (error) | |
1194 | goto out_undo_icount; | |
1195 | } | |
1196 | ||
1197 | /* apply remaining deltas */ | |
45c34141 | 1198 | if (rtxdelta != 0) { |
1da177e4 | 1199 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 1200 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
1201 | msbp++; |
1202 | } | |
1203 | ||
92821e2b | 1204 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { |
1da177e4 LT |
1205 | if (tp->t_dblocks_delta != 0) { |
1206 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 1207 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
1208 | msbp++; |
1209 | } | |
1210 | if (tp->t_agcount_delta != 0) { | |
1211 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 1212 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
1213 | msbp++; |
1214 | } | |
1215 | if (tp->t_imaxpct_delta != 0) { | |
1216 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 1217 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
1218 | msbp++; |
1219 | } | |
1220 | if (tp->t_rextsize_delta != 0) { | |
1221 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 1222 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
1223 | msbp++; |
1224 | } | |
1225 | if (tp->t_rbmblocks_delta != 0) { | |
1226 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 1227 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
1228 | msbp++; |
1229 | } | |
1230 | if (tp->t_rblocks_delta != 0) { | |
1231 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 1232 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
1233 | msbp++; |
1234 | } | |
1235 | if (tp->t_rextents_delta != 0) { | |
1236 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 1237 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
1238 | msbp++; |
1239 | } | |
1240 | if (tp->t_rextslog_delta != 0) { | |
1241 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 1242 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
1243 | msbp++; |
1244 | } | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * If we need to change anything, do it. | |
1249 | */ | |
1250 | if (msbp > msb) { | |
1251 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
1252 | (uint)(msbp - msb), rsvd); | |
1b040712 CH |
1253 | if (error) |
1254 | goto out_undo_ifreecount; | |
1da177e4 | 1255 | } |
1b040712 CH |
1256 | |
1257 | return; | |
1258 | ||
1259 | out_undo_ifreecount: | |
1260 | if (ifreedelta) | |
1261 | xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd); | |
1262 | out_undo_icount: | |
1263 | if (idelta) | |
1264 | xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd); | |
1265 | out_undo_fdblocks: | |
1266 | if (blkdelta) | |
1267 | xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd); | |
1268 | out: | |
1884bd83 | 1269 | ASSERT(error == 0); |
1b040712 | 1270 | return; |
1da177e4 LT |
1271 | } |
1272 | ||
e98c414f CH |
1273 | /* |
1274 | * Add the given log item to the transaction's list of log items. | |
1275 | * | |
1276 | * The log item will now point to its new descriptor with its li_desc field. | |
1277 | */ | |
1278 | void | |
1279 | xfs_trans_add_item( | |
1280 | struct xfs_trans *tp, | |
1281 | struct xfs_log_item *lip) | |
1282 | { | |
1283 | struct xfs_log_item_desc *lidp; | |
1284 | ||
f65020a8 JJ |
1285 | ASSERT(lip->li_mountp == tp->t_mountp); |
1286 | ASSERT(lip->li_ailp == tp->t_mountp->m_ail); | |
e98c414f | 1287 | |
43869706 | 1288 | lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS); |
e98c414f CH |
1289 | |
1290 | lidp->lid_item = lip; | |
1291 | lidp->lid_flags = 0; | |
e98c414f CH |
1292 | list_add_tail(&lidp->lid_trans, &tp->t_items); |
1293 | ||
1294 | lip->li_desc = lidp; | |
1295 | } | |
1296 | ||
1297 | STATIC void | |
1298 | xfs_trans_free_item_desc( | |
1299 | struct xfs_log_item_desc *lidp) | |
1300 | { | |
1301 | list_del_init(&lidp->lid_trans); | |
1302 | kmem_zone_free(xfs_log_item_desc_zone, lidp); | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Unlink and free the given descriptor. | |
1307 | */ | |
1308 | void | |
1309 | xfs_trans_del_item( | |
1310 | struct xfs_log_item *lip) | |
1311 | { | |
1312 | xfs_trans_free_item_desc(lip->li_desc); | |
1313 | lip->li_desc = NULL; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Unlock all of the items of a transaction and free all the descriptors | |
1318 | * of that transaction. | |
1319 | */ | |
d17c701c | 1320 | void |
e98c414f CH |
1321 | xfs_trans_free_items( |
1322 | struct xfs_trans *tp, | |
1323 | xfs_lsn_t commit_lsn, | |
1324 | int flags) | |
1325 | { | |
1326 | struct xfs_log_item_desc *lidp, *next; | |
1327 | ||
1328 | list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) { | |
1329 | struct xfs_log_item *lip = lidp->lid_item; | |
1330 | ||
1331 | lip->li_desc = NULL; | |
1332 | ||
1333 | if (commit_lsn != NULLCOMMITLSN) | |
1334 | IOP_COMMITTING(lip, commit_lsn); | |
1335 | if (flags & XFS_TRANS_ABORT) | |
1336 | lip->li_flags |= XFS_LI_ABORTED; | |
1337 | IOP_UNLOCK(lip); | |
1338 | ||
1339 | xfs_trans_free_item_desc(lidp); | |
1340 | } | |
1341 | } | |
1342 | ||
0e57f6a3 DC |
1343 | static inline void |
1344 | xfs_log_item_batch_insert( | |
1345 | struct xfs_ail *ailp, | |
1d8c95a3 | 1346 | struct xfs_ail_cursor *cur, |
0e57f6a3 DC |
1347 | struct xfs_log_item **log_items, |
1348 | int nr_items, | |
1349 | xfs_lsn_t commit_lsn) | |
1350 | { | |
1351 | int i; | |
1352 | ||
1353 | spin_lock(&ailp->xa_lock); | |
1354 | /* xfs_trans_ail_update_bulk drops ailp->xa_lock */ | |
1d8c95a3 | 1355 | xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn); |
0e57f6a3 DC |
1356 | |
1357 | for (i = 0; i < nr_items; i++) | |
1358 | IOP_UNPIN(log_items[i], 0); | |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * Bulk operation version of xfs_trans_committed that takes a log vector of | |
1363 | * items to insert into the AIL. This uses bulk AIL insertion techniques to | |
1364 | * minimise lock traffic. | |
e34a314c DC |
1365 | * |
1366 | * If we are called with the aborted flag set, it is because a log write during | |
1367 | * a CIL checkpoint commit has failed. In this case, all the items in the | |
1368 | * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which | |
1369 | * means that checkpoint commit abort handling is treated exactly the same | |
1370 | * as an iclog write error even though we haven't started any IO yet. Hence in | |
1371 | * this case all we need to do is IOP_COMMITTED processing, followed by an | |
1372 | * IOP_UNPIN(aborted) call. | |
1d8c95a3 DC |
1373 | * |
1374 | * The AIL cursor is used to optimise the insert process. If commit_lsn is not | |
1375 | * at the end of the AIL, the insert cursor avoids the need to walk | |
1376 | * the AIL to find the insertion point on every xfs_log_item_batch_insert() | |
1377 | * call. This saves a lot of needless list walking and is a net win, even | |
1378 | * though it slightly increases that amount of AIL lock traffic to set it up | |
1379 | * and tear it down. | |
0e57f6a3 DC |
1380 | */ |
1381 | void | |
1382 | xfs_trans_committed_bulk( | |
1383 | struct xfs_ail *ailp, | |
1384 | struct xfs_log_vec *log_vector, | |
1385 | xfs_lsn_t commit_lsn, | |
1386 | int aborted) | |
1387 | { | |
1388 | #define LOG_ITEM_BATCH_SIZE 32 | |
1389 | struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE]; | |
1390 | struct xfs_log_vec *lv; | |
1d8c95a3 | 1391 | struct xfs_ail_cursor cur; |
0e57f6a3 DC |
1392 | int i = 0; |
1393 | ||
1d8c95a3 DC |
1394 | spin_lock(&ailp->xa_lock); |
1395 | xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn); | |
1396 | spin_unlock(&ailp->xa_lock); | |
1397 | ||
0e57f6a3 DC |
1398 | /* unpin all the log items */ |
1399 | for (lv = log_vector; lv; lv = lv->lv_next ) { | |
1400 | struct xfs_log_item *lip = lv->lv_item; | |
1401 | xfs_lsn_t item_lsn; | |
1402 | ||
1403 | if (aborted) | |
1404 | lip->li_flags |= XFS_LI_ABORTED; | |
1405 | item_lsn = IOP_COMMITTED(lip, commit_lsn); | |
1406 | ||
1316d4da | 1407 | /* item_lsn of -1 means the item needs no further processing */ |
0e57f6a3 DC |
1408 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) |
1409 | continue; | |
1410 | ||
e34a314c DC |
1411 | /* |
1412 | * if we are aborting the operation, no point in inserting the | |
1413 | * object into the AIL as we are in a shutdown situation. | |
1414 | */ | |
1415 | if (aborted) { | |
1416 | ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount)); | |
1417 | IOP_UNPIN(lip, 1); | |
1418 | continue; | |
1419 | } | |
1420 | ||
0e57f6a3 DC |
1421 | if (item_lsn != commit_lsn) { |
1422 | ||
1423 | /* | |
1424 | * Not a bulk update option due to unusual item_lsn. | |
1425 | * Push into AIL immediately, rechecking the lsn once | |
1d8c95a3 DC |
1426 | * we have the ail lock. Then unpin the item. This does |
1427 | * not affect the AIL cursor the bulk insert path is | |
1428 | * using. | |
0e57f6a3 DC |
1429 | */ |
1430 | spin_lock(&ailp->xa_lock); | |
1431 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) | |
1432 | xfs_trans_ail_update(ailp, lip, item_lsn); | |
1433 | else | |
1434 | spin_unlock(&ailp->xa_lock); | |
1435 | IOP_UNPIN(lip, 0); | |
1436 | continue; | |
1437 | } | |
1438 | ||
1439 | /* Item is a candidate for bulk AIL insert. */ | |
1440 | log_items[i++] = lv->lv_item; | |
1441 | if (i >= LOG_ITEM_BATCH_SIZE) { | |
1d8c95a3 | 1442 | xfs_log_item_batch_insert(ailp, &cur, log_items, |
0e57f6a3 DC |
1443 | LOG_ITEM_BATCH_SIZE, commit_lsn); |
1444 | i = 0; | |
1445 | } | |
1446 | } | |
1447 | ||
1448 | /* make sure we insert the remainder! */ | |
1449 | if (i) | |
1d8c95a3 DC |
1450 | xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn); |
1451 | ||
1452 | spin_lock(&ailp->xa_lock); | |
1453 | xfs_trans_ail_cursor_done(ailp, &cur); | |
1454 | spin_unlock(&ailp->xa_lock); | |
0e57f6a3 DC |
1455 | } |
1456 | ||
0924378a | 1457 | /* |
b1037058 | 1458 | * Commit the given transaction to the log. |
0924378a DC |
1459 | * |
1460 | * XFS disk error handling mechanism is not based on a typical | |
1461 | * transaction abort mechanism. Logically after the filesystem | |
1462 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1463 | * be durable - ie. committed to disk - because some metadata might | |
1464 | * be inconsistent. In such cases, this returns an error, and the | |
1465 | * caller may assume that all locked objects joined to the transaction | |
1466 | * have already been unlocked as if the commit had succeeded. | |
1467 | * Do not reference the transaction structure after this call. | |
1468 | */ | |
0924378a | 1469 | int |
b1037058 | 1470 | xfs_trans_commit( |
a3ccd2ca | 1471 | struct xfs_trans *tp, |
b1037058 | 1472 | uint flags) |
0924378a | 1473 | { |
a3ccd2ca | 1474 | struct xfs_mount *mp = tp->t_mountp; |
0924378a | 1475 | xfs_lsn_t commit_lsn = -1; |
a3ccd2ca | 1476 | int error = 0; |
0924378a DC |
1477 | int log_flags = 0; |
1478 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
0924378a DC |
1479 | |
1480 | /* | |
1481 | * Determine whether this commit is releasing a permanent | |
1482 | * log reservation or not. | |
1483 | */ | |
1484 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1485 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1486 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * If there is nothing to be logged by the transaction, | |
1491 | * then unlock all of the items associated with the | |
1492 | * transaction and free the transaction structure. | |
1493 | * Also make sure to return any reserved blocks to | |
1494 | * the free pool. | |
1495 | */ | |
a3ccd2ca CH |
1496 | if (!(tp->t_flags & XFS_TRANS_DIRTY)) |
1497 | goto out_unreserve; | |
1498 | ||
1499 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1500 | error = XFS_ERROR(EIO); | |
1501 | goto out_unreserve; | |
0924378a | 1502 | } |
a3ccd2ca | 1503 | |
0924378a DC |
1504 | ASSERT(tp->t_ticket != NULL); |
1505 | ||
1506 | /* | |
1507 | * If we need to update the superblock, then do it now. | |
1508 | */ | |
1509 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1510 | xfs_trans_apply_sb_deltas(tp); | |
1511 | xfs_trans_apply_dquot_deltas(tp); | |
1512 | ||
0244b960 | 1513 | error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags); |
0924378a DC |
1514 | if (error == ENOMEM) { |
1515 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
a3ccd2ca CH |
1516 | error = XFS_ERROR(EIO); |
1517 | goto out_unreserve; | |
0924378a | 1518 | } |
1da177e4 | 1519 | |
0244b960 CH |
1520 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1521 | xfs_trans_free(tp); | |
1522 | ||
1da177e4 LT |
1523 | /* |
1524 | * If the transaction needs to be synchronous, then force the | |
1525 | * log out now and wait for it. | |
1526 | */ | |
1527 | if (sync) { | |
f538d4da | 1528 | if (!error) { |
a14a348b | 1529 | error = _xfs_log_force_lsn(mp, commit_lsn, |
b1037058 | 1530 | XFS_LOG_SYNC, NULL); |
f538d4da | 1531 | } |
1da177e4 LT |
1532 | XFS_STATS_INC(xs_trans_sync); |
1533 | } else { | |
1534 | XFS_STATS_INC(xs_trans_async); | |
1535 | } | |
1536 | ||
a3ccd2ca CH |
1537 | return error; |
1538 | ||
1539 | out_unreserve: | |
1540 | xfs_trans_unreserve_and_mod_sb(tp); | |
1541 | ||
1542 | /* | |
1543 | * It is indeed possible for the transaction to be not dirty but | |
1544 | * the dqinfo portion to be. All that means is that we have some | |
1545 | * (non-persistent) quota reservations that need to be unreserved. | |
1546 | */ | |
1547 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1548 | if (tp->t_ticket) { | |
1549 | commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
1550 | if (commit_lsn == -1 && !error) | |
1551 | error = XFS_ERROR(EIO); | |
1552 | } | |
1553 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
71e330b5 | 1554 | xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0); |
a3ccd2ca CH |
1555 | xfs_trans_free(tp); |
1556 | ||
1557 | XFS_STATS_INC(xs_trans_empty); | |
1558 | return error; | |
1da177e4 LT |
1559 | } |
1560 | ||
1da177e4 LT |
1561 | /* |
1562 | * Unlock all of the transaction's items and free the transaction. | |
1563 | * The transaction must not have modified any of its items, because | |
1564 | * there is no way to restore them to their previous state. | |
1565 | * | |
1566 | * If the transaction has made a log reservation, make sure to release | |
1567 | * it as well. | |
1568 | */ | |
1569 | void | |
1570 | xfs_trans_cancel( | |
1571 | xfs_trans_t *tp, | |
1572 | int flags) | |
1573 | { | |
1574 | int log_flags; | |
0733af21 | 1575 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1576 | |
1577 | /* | |
1578 | * See if the caller is being too lazy to figure out if | |
1579 | * the transaction really needs an abort. | |
1580 | */ | |
1581 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1582 | flags &= ~XFS_TRANS_ABORT; | |
1583 | /* | |
1584 | * See if the caller is relying on us to shut down the | |
1585 | * filesystem. This happens in paths where we detect | |
1586 | * corruption and decide to give up. | |
1587 | */ | |
60a204f0 | 1588 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1589 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1590 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1591 | } |
1da177e4 | 1592 | #ifdef DEBUG |
e98c414f CH |
1593 | if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) { |
1594 | struct xfs_log_item_desc *lidp; | |
1595 | ||
1596 | list_for_each_entry(lidp, &tp->t_items, lid_trans) | |
1597 | ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD)); | |
1da177e4 LT |
1598 | } |
1599 | #endif | |
1600 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1601 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1602 | |
1603 | if (tp->t_ticket) { | |
1604 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1605 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1606 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1607 | } else { | |
1608 | log_flags = 0; | |
1609 | } | |
0733af21 | 1610 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1611 | } |
1612 | ||
1613 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1614 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 1615 | |
71e330b5 | 1616 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
1da177e4 LT |
1617 | xfs_trans_free(tp); |
1618 | } | |
1619 | ||
322ff6b8 NS |
1620 | /* |
1621 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1622 | * the next: permanent transactions are only flushed out when | |
1623 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1624 | * as possible to let chunks of it go to the log. So we commit the | |
1625 | * chunk we've been working on and get a new transaction to continue. | |
1626 | */ | |
1627 | int | |
1628 | xfs_trans_roll( | |
1629 | struct xfs_trans **tpp, | |
1630 | struct xfs_inode *dp) | |
1631 | { | |
1632 | struct xfs_trans *trans; | |
1633 | unsigned int logres, count; | |
1634 | int error; | |
1635 | ||
1636 | /* | |
1637 | * Ensure that the inode is always logged. | |
1638 | */ | |
1639 | trans = *tpp; | |
1640 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1641 | ||
1642 | /* | |
1643 | * Copy the critical parameters from one trans to the next. | |
1644 | */ | |
1645 | logres = trans->t_log_res; | |
1646 | count = trans->t_log_count; | |
1647 | *tpp = xfs_trans_dup(trans); | |
1648 | ||
1649 | /* | |
1650 | * Commit the current transaction. | |
1651 | * If this commit failed, then it'd just unlock those items that | |
1652 | * are not marked ihold. That also means that a filesystem shutdown | |
1653 | * is in progress. The caller takes the responsibility to cancel | |
1654 | * the duplicate transaction that gets returned. | |
1655 | */ | |
1656 | error = xfs_trans_commit(trans, 0); | |
1657 | if (error) | |
1658 | return (error); | |
1659 | ||
1660 | trans = *tpp; | |
1661 | ||
cc09c0dc DC |
1662 | /* |
1663 | * transaction commit worked ok so we can drop the extra ticket | |
1664 | * reference that we gained in xfs_trans_dup() | |
1665 | */ | |
1666 | xfs_log_ticket_put(trans->t_ticket); | |
1667 | ||
1668 | ||
322ff6b8 NS |
1669 | /* |
1670 | * Reserve space in the log for th next transaction. | |
1671 | * This also pushes items in the "AIL", the list of logged items, | |
1672 | * out to disk if they are taking up space at the tail of the log | |
1673 | * that we want to use. This requires that either nothing be locked | |
1674 | * across this call, or that anything that is locked be logged in | |
1675 | * the prior and the next transactions. | |
1676 | */ | |
1677 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1678 | XFS_TRANS_PERM_LOG_RES, count); | |
1679 | /* | |
1680 | * Ensure that the inode is in the new transaction and locked. | |
1681 | */ | |
1682 | if (error) | |
1683 | return error; | |
1684 | ||
ddc3415a | 1685 | xfs_trans_ijoin(trans, dp, 0); |
322ff6b8 NS |
1686 | return 0; |
1687 | } |