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