]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
30 | #include "xfs_error.h" | |
a844f451 | 31 | #include "xfs_da_btree.h" |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_dir2_sf.h" |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 LT |
37 | #include "xfs_dinode.h" |
38 | #include "xfs_inode.h" | |
a844f451 NS |
39 | #include "xfs_btree.h" |
40 | #include "xfs_ialloc.h" | |
41 | #include "xfs_alloc.h" | |
1da177e4 | 42 | #include "xfs_bmap.h" |
1da177e4 | 43 | #include "xfs_quota.h" |
a844f451 | 44 | #include "xfs_trans_priv.h" |
1da177e4 | 45 | #include "xfs_trans_space.h" |
322ff6b8 | 46 | #include "xfs_inode_item.h" |
ed3b4d6c | 47 | #include "xfs_trace.h" |
1da177e4 | 48 | |
8f794055 | 49 | kmem_zone_t *xfs_trans_zone; |
1da177e4 | 50 | |
8f794055 NS |
51 | /* |
52 | * Reservation functions here avoid a huge stack in xfs_trans_init | |
53 | * due to register overflow from temporaries in the calculations. | |
54 | */ | |
8f794055 NS |
55 | STATIC uint |
56 | xfs_calc_write_reservation(xfs_mount_t *mp) | |
57 | { | |
58 | return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
59 | } | |
60 | ||
61 | STATIC uint | |
62 | xfs_calc_itruncate_reservation(xfs_mount_t *mp) | |
63 | { | |
64 | return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
65 | } | |
66 | ||
67 | STATIC uint | |
68 | xfs_calc_rename_reservation(xfs_mount_t *mp) | |
69 | { | |
70 | return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
71 | } | |
72 | ||
73 | STATIC uint | |
74 | xfs_calc_link_reservation(xfs_mount_t *mp) | |
75 | { | |
2ddd5928 | 76 | return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); |
8f794055 NS |
77 | } |
78 | ||
79 | STATIC uint | |
80 | xfs_calc_remove_reservation(xfs_mount_t *mp) | |
81 | { | |
82 | return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
83 | } | |
84 | ||
85 | STATIC uint | |
86 | xfs_calc_symlink_reservation(xfs_mount_t *mp) | |
87 | { | |
88 | return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
89 | } | |
90 | ||
91 | STATIC uint | |
92 | xfs_calc_create_reservation(xfs_mount_t *mp) | |
93 | { | |
94 | return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
95 | } | |
96 | ||
97 | STATIC uint | |
98 | xfs_calc_mkdir_reservation(xfs_mount_t *mp) | |
99 | { | |
100 | return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
101 | } | |
102 | ||
103 | STATIC uint | |
104 | xfs_calc_ifree_reservation(xfs_mount_t *mp) | |
105 | { | |
106 | return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
107 | } | |
108 | ||
109 | STATIC uint | |
110 | xfs_calc_ichange_reservation(xfs_mount_t *mp) | |
111 | { | |
112 | return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
113 | } | |
114 | ||
115 | STATIC uint | |
116 | xfs_calc_growdata_reservation(xfs_mount_t *mp) | |
117 | { | |
118 | return XFS_CALC_GROWDATA_LOG_RES(mp); | |
119 | } | |
120 | ||
121 | STATIC uint | |
122 | xfs_calc_growrtalloc_reservation(xfs_mount_t *mp) | |
123 | { | |
124 | return XFS_CALC_GROWRTALLOC_LOG_RES(mp); | |
125 | } | |
126 | ||
127 | STATIC uint | |
128 | xfs_calc_growrtzero_reservation(xfs_mount_t *mp) | |
129 | { | |
130 | return XFS_CALC_GROWRTZERO_LOG_RES(mp); | |
131 | } | |
132 | ||
133 | STATIC uint | |
134 | xfs_calc_growrtfree_reservation(xfs_mount_t *mp) | |
135 | { | |
136 | return XFS_CALC_GROWRTFREE_LOG_RES(mp); | |
137 | } | |
138 | ||
139 | STATIC uint | |
140 | xfs_calc_swrite_reservation(xfs_mount_t *mp) | |
141 | { | |
142 | return XFS_CALC_SWRITE_LOG_RES(mp); | |
143 | } | |
144 | ||
145 | STATIC uint | |
146 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
147 | { | |
148 | return XFS_CALC_WRITEID_LOG_RES(mp); | |
149 | } | |
150 | ||
151 | STATIC uint | |
152 | xfs_calc_addafork_reservation(xfs_mount_t *mp) | |
153 | { | |
154 | return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
155 | } | |
156 | ||
157 | STATIC uint | |
158 | xfs_calc_attrinval_reservation(xfs_mount_t *mp) | |
159 | { | |
160 | return XFS_CALC_ATTRINVAL_LOG_RES(mp); | |
161 | } | |
162 | ||
163 | STATIC uint | |
164 | xfs_calc_attrset_reservation(xfs_mount_t *mp) | |
165 | { | |
166 | return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
167 | } | |
168 | ||
169 | STATIC uint | |
170 | xfs_calc_attrrm_reservation(xfs_mount_t *mp) | |
171 | { | |
172 | return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
173 | } | |
174 | ||
175 | STATIC uint | |
176 | xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp) | |
177 | { | |
178 | return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp); | |
179 | } | |
180 | ||
1da177e4 LT |
181 | /* |
182 | * Initialize the precomputed transaction reservation values | |
183 | * in the mount structure. | |
184 | */ | |
185 | void | |
186 | xfs_trans_init( | |
187 | xfs_mount_t *mp) | |
188 | { | |
189 | xfs_trans_reservations_t *resp; | |
190 | ||
191 | resp = &(mp->m_reservations); | |
8f794055 NS |
192 | resp->tr_write = xfs_calc_write_reservation(mp); |
193 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
194 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
195 | resp->tr_link = xfs_calc_link_reservation(mp); | |
196 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
197 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
198 | resp->tr_create = xfs_calc_create_reservation(mp); | |
199 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
200 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
201 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
202 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
203 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
204 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
205 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
206 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
207 | resp->tr_attrset = xfs_calc_attrset_reservation(mp); | |
208 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); | |
209 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
210 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
211 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
212 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
1da177e4 LT |
213 | } |
214 | ||
215 | /* | |
216 | * This routine is called to allocate a transaction structure. | |
217 | * The type parameter indicates the type of the transaction. These | |
218 | * are enumerated in xfs_trans.h. | |
219 | * | |
220 | * Dynamically allocate the transaction structure from the transaction | |
221 | * zone, initialize it, and return it to the caller. | |
222 | */ | |
223 | xfs_trans_t * | |
224 | xfs_trans_alloc( | |
225 | xfs_mount_t *mp, | |
226 | uint type) | |
227 | { | |
b267ce99 | 228 | xfs_wait_for_freeze(mp, SB_FREEZE_TRANS); |
80641dc6 | 229 | return _xfs_trans_alloc(mp, type, KM_SLEEP); |
1da177e4 LT |
230 | } |
231 | ||
232 | xfs_trans_t * | |
233 | _xfs_trans_alloc( | |
234 | xfs_mount_t *mp, | |
80641dc6 CH |
235 | uint type, |
236 | uint memflags) | |
1da177e4 LT |
237 | { |
238 | xfs_trans_t *tp; | |
239 | ||
34327e13 | 240 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 241 | |
80641dc6 | 242 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
243 | tp->t_magic = XFS_TRANS_MAGIC; |
244 | tp->t_type = type; | |
245 | tp->t_mountp = mp; | |
246 | tp->t_items_free = XFS_LIC_NUM_SLOTS; | |
39dab9d7 | 247 | xfs_lic_init(&(tp->t_items)); |
ed3b4d6c | 248 | INIT_LIST_HEAD(&tp->t_busy); |
34327e13 | 249 | return tp; |
1da177e4 LT |
250 | } |
251 | ||
b1c1b5b6 DC |
252 | /* |
253 | * Free the transaction structure. If there is more clean up | |
254 | * to do when the structure is freed, add it here. | |
255 | */ | |
256 | STATIC void | |
257 | xfs_trans_free( | |
ed3b4d6c | 258 | struct xfs_trans *tp) |
b1c1b5b6 | 259 | { |
ed3b4d6c DC |
260 | struct xfs_busy_extent *busyp, *n; |
261 | ||
262 | list_for_each_entry_safe(busyp, n, &tp->t_busy, list) | |
263 | xfs_alloc_busy_clear(tp->t_mountp, busyp); | |
264 | ||
b1c1b5b6 DC |
265 | atomic_dec(&tp->t_mountp->m_active_trans); |
266 | xfs_trans_free_dqinfo(tp); | |
267 | kmem_zone_free(xfs_trans_zone, tp); | |
268 | } | |
269 | ||
1da177e4 LT |
270 | /* |
271 | * This is called to create a new transaction which will share the | |
272 | * permanent log reservation of the given transaction. The remaining | |
273 | * unused block and rt extent reservations are also inherited. This | |
274 | * implies that the original transaction is no longer allowed to allocate | |
275 | * blocks. Locks and log items, however, are no inherited. They must | |
276 | * be added to the new transaction explicitly. | |
277 | */ | |
278 | xfs_trans_t * | |
279 | xfs_trans_dup( | |
280 | xfs_trans_t *tp) | |
281 | { | |
282 | xfs_trans_t *ntp; | |
283 | ||
284 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
285 | ||
286 | /* | |
287 | * Initialize the new transaction structure. | |
288 | */ | |
289 | ntp->t_magic = XFS_TRANS_MAGIC; | |
290 | ntp->t_type = tp->t_type; | |
291 | ntp->t_mountp = tp->t_mountp; | |
292 | ntp->t_items_free = XFS_LIC_NUM_SLOTS; | |
39dab9d7 | 293 | xfs_lic_init(&(ntp->t_items)); |
ed3b4d6c | 294 | INIT_LIST_HEAD(&ntp->t_busy); |
1da177e4 LT |
295 | |
296 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 297 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 298 | |
1da177e4 | 299 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE); |
cc09c0dc | 300 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
301 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
302 | tp->t_blk_res = tp->t_blk_res_used; | |
303 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
304 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 305 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 306 | |
7d095257 | 307 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
308 | |
309 | atomic_inc(&tp->t_mountp->m_active_trans); | |
310 | return ntp; | |
311 | } | |
312 | ||
313 | /* | |
314 | * This is called to reserve free disk blocks and log space for the | |
315 | * given transaction. This must be done before allocating any resources | |
316 | * within the transaction. | |
317 | * | |
318 | * This will return ENOSPC if there are not enough blocks available. | |
319 | * It will sleep waiting for available log space. | |
320 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
321 | * is used by long running transactions. If any one of the reservations | |
322 | * fails then they will all be backed out. | |
323 | * | |
324 | * This does not do quota reservations. That typically is done by the | |
325 | * caller afterwards. | |
326 | */ | |
327 | int | |
328 | xfs_trans_reserve( | |
329 | xfs_trans_t *tp, | |
330 | uint blocks, | |
331 | uint logspace, | |
332 | uint rtextents, | |
333 | uint flags, | |
334 | uint logcount) | |
335 | { | |
336 | int log_flags; | |
59c1b082 NS |
337 | int error = 0; |
338 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
339 | |
340 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 341 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
342 | |
343 | /* | |
344 | * Attempt to reserve the needed disk blocks by decrementing | |
345 | * the number needed from the number available. This will | |
346 | * fail if the count would go below zero. | |
347 | */ | |
348 | if (blocks > 0) { | |
349 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 350 | -((int64_t)blocks), rsvd); |
1da177e4 | 351 | if (error != 0) { |
59c1b082 | 352 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
353 | return (XFS_ERROR(ENOSPC)); |
354 | } | |
355 | tp->t_blk_res += blocks; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Reserve the log space needed for this transaction. | |
360 | */ | |
361 | if (logspace > 0) { | |
362 | ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace)); | |
363 | ASSERT((tp->t_log_count == 0) || | |
364 | (tp->t_log_count == logcount)); | |
365 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
366 | log_flags = XFS_LOG_PERM_RESERV; | |
367 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; | |
368 | } else { | |
369 | ASSERT(tp->t_ticket == NULL); | |
370 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
371 | log_flags = 0; | |
372 | } | |
373 | ||
374 | error = xfs_log_reserve(tp->t_mountp, logspace, logcount, | |
375 | &tp->t_ticket, | |
7e9c6396 | 376 | XFS_TRANSACTION, log_flags, tp->t_type); |
1da177e4 LT |
377 | if (error) { |
378 | goto undo_blocks; | |
379 | } | |
380 | tp->t_log_res = logspace; | |
381 | tp->t_log_count = logcount; | |
382 | } | |
383 | ||
384 | /* | |
385 | * Attempt to reserve the needed realtime extents by decrementing | |
386 | * the number needed from the number available. This will | |
387 | * fail if the count would go below zero. | |
388 | */ | |
389 | if (rtextents > 0) { | |
390 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 391 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
392 | if (error) { |
393 | error = XFS_ERROR(ENOSPC); | |
394 | goto undo_log; | |
395 | } | |
396 | tp->t_rtx_res += rtextents; | |
397 | } | |
398 | ||
399 | return 0; | |
400 | ||
401 | /* | |
402 | * Error cases jump to one of these labels to undo any | |
403 | * reservations which have already been performed. | |
404 | */ | |
405 | undo_log: | |
406 | if (logspace > 0) { | |
407 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
408 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
409 | } else { | |
410 | log_flags = 0; | |
411 | } | |
412 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
413 | tp->t_ticket = NULL; | |
414 | tp->t_log_res = 0; | |
415 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
416 | } | |
417 | ||
418 | undo_blocks: | |
419 | if (blocks > 0) { | |
420 | (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 421 | (int64_t)blocks, rsvd); |
1da177e4 LT |
422 | tp->t_blk_res = 0; |
423 | } | |
424 | ||
59c1b082 | 425 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 426 | |
59c1b082 | 427 | return error; |
1da177e4 LT |
428 | } |
429 | ||
1da177e4 LT |
430 | /* |
431 | * Record the indicated change to the given field for application | |
432 | * to the file system's superblock when the transaction commits. | |
433 | * For now, just store the change in the transaction structure. | |
434 | * | |
435 | * Mark the transaction structure to indicate that the superblock | |
436 | * needs to be updated before committing. | |
92821e2b DC |
437 | * |
438 | * Because we may not be keeping track of allocated/free inodes and | |
439 | * used filesystem blocks in the superblock, we do not mark the | |
440 | * superblock dirty in this transaction if we modify these fields. | |
441 | * We still need to update the transaction deltas so that they get | |
442 | * applied to the incore superblock, but we don't want them to | |
443 | * cause the superblock to get locked and logged if these are the | |
444 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
445 | */ |
446 | void | |
447 | xfs_trans_mod_sb( | |
448 | xfs_trans_t *tp, | |
449 | uint field, | |
20f4ebf2 | 450 | int64_t delta) |
1da177e4 | 451 | { |
92821e2b DC |
452 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
453 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
454 | |
455 | switch (field) { | |
456 | case XFS_TRANS_SB_ICOUNT: | |
457 | tp->t_icount_delta += delta; | |
92821e2b DC |
458 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
459 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
460 | break; |
461 | case XFS_TRANS_SB_IFREE: | |
462 | tp->t_ifree_delta += delta; | |
92821e2b DC |
463 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
464 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
465 | break; |
466 | case XFS_TRANS_SB_FDBLOCKS: | |
467 | /* | |
468 | * Track the number of blocks allocated in the | |
469 | * transaction. Make sure it does not exceed the | |
470 | * number reserved. | |
471 | */ | |
472 | if (delta < 0) { | |
473 | tp->t_blk_res_used += (uint)-delta; | |
474 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
475 | } | |
476 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
477 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
478 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
479 | break; |
480 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
481 | /* | |
482 | * The allocation has already been applied to the | |
483 | * in-core superblock's counter. This should only | |
484 | * be applied to the on-disk superblock. | |
485 | */ | |
486 | ASSERT(delta < 0); | |
487 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
488 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
489 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
490 | break; |
491 | case XFS_TRANS_SB_FREXTENTS: | |
492 | /* | |
493 | * Track the number of blocks allocated in the | |
494 | * transaction. Make sure it does not exceed the | |
495 | * number reserved. | |
496 | */ | |
497 | if (delta < 0) { | |
498 | tp->t_rtx_res_used += (uint)-delta; | |
499 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
500 | } | |
501 | tp->t_frextents_delta += delta; | |
502 | break; | |
503 | case XFS_TRANS_SB_RES_FREXTENTS: | |
504 | /* | |
505 | * The allocation has already been applied to the | |
c41564b5 | 506 | * in-core superblock's counter. This should only |
1da177e4 LT |
507 | * be applied to the on-disk superblock. |
508 | */ | |
509 | ASSERT(delta < 0); | |
510 | tp->t_res_frextents_delta += delta; | |
511 | break; | |
512 | case XFS_TRANS_SB_DBLOCKS: | |
513 | ASSERT(delta > 0); | |
514 | tp->t_dblocks_delta += delta; | |
515 | break; | |
516 | case XFS_TRANS_SB_AGCOUNT: | |
517 | ASSERT(delta > 0); | |
518 | tp->t_agcount_delta += delta; | |
519 | break; | |
520 | case XFS_TRANS_SB_IMAXPCT: | |
521 | tp->t_imaxpct_delta += delta; | |
522 | break; | |
523 | case XFS_TRANS_SB_REXTSIZE: | |
524 | tp->t_rextsize_delta += delta; | |
525 | break; | |
526 | case XFS_TRANS_SB_RBMBLOCKS: | |
527 | tp->t_rbmblocks_delta += delta; | |
528 | break; | |
529 | case XFS_TRANS_SB_RBLOCKS: | |
530 | tp->t_rblocks_delta += delta; | |
531 | break; | |
532 | case XFS_TRANS_SB_REXTENTS: | |
533 | tp->t_rextents_delta += delta; | |
534 | break; | |
535 | case XFS_TRANS_SB_REXTSLOG: | |
536 | tp->t_rextslog_delta += delta; | |
537 | break; | |
538 | default: | |
539 | ASSERT(0); | |
540 | return; | |
541 | } | |
542 | ||
210c6f1c | 543 | tp->t_flags |= flags; |
1da177e4 LT |
544 | } |
545 | ||
546 | /* | |
547 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
548 | * to bring the superblock buffer into the current transaction | |
549 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
550 | * | |
551 | * For now we just look at each field allowed to change and change | |
552 | * it if necessary. | |
553 | */ | |
554 | STATIC void | |
555 | xfs_trans_apply_sb_deltas( | |
556 | xfs_trans_t *tp) | |
557 | { | |
2bdf7cd0 | 558 | xfs_dsb_t *sbp; |
1da177e4 LT |
559 | xfs_buf_t *bp; |
560 | int whole = 0; | |
561 | ||
562 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
563 | sbp = XFS_BUF_TO_SBP(bp); | |
564 | ||
565 | /* | |
566 | * Check that superblock mods match the mods made to AGF counters. | |
567 | */ | |
568 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
569 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
570 | tp->t_ag_btree_delta)); | |
571 | ||
92821e2b DC |
572 | /* |
573 | * Only update the superblock counters if we are logging them | |
574 | */ | |
575 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 576 | if (tp->t_icount_delta) |
413d57c9 | 577 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 578 | if (tp->t_ifree_delta) |
413d57c9 | 579 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 580 | if (tp->t_fdblocks_delta) |
413d57c9 | 581 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 582 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 583 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
584 | } |
585 | ||
2bdf7cd0 | 586 | if (tp->t_frextents_delta) |
413d57c9 | 587 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 588 | if (tp->t_res_frextents_delta) |
413d57c9 | 589 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
590 | |
591 | if (tp->t_dblocks_delta) { | |
413d57c9 | 592 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
593 | whole = 1; |
594 | } | |
2bdf7cd0 | 595 | if (tp->t_agcount_delta) { |
413d57c9 | 596 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
597 | whole = 1; |
598 | } | |
2bdf7cd0 CH |
599 | if (tp->t_imaxpct_delta) { |
600 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
601 | whole = 1; |
602 | } | |
2bdf7cd0 | 603 | if (tp->t_rextsize_delta) { |
413d57c9 | 604 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
605 | whole = 1; |
606 | } | |
2bdf7cd0 | 607 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 608 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
609 | whole = 1; |
610 | } | |
2bdf7cd0 | 611 | if (tp->t_rblocks_delta) { |
413d57c9 | 612 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
613 | whole = 1; |
614 | } | |
2bdf7cd0 | 615 | if (tp->t_rextents_delta) { |
413d57c9 | 616 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
617 | whole = 1; |
618 | } | |
2bdf7cd0 CH |
619 | if (tp->t_rextslog_delta) { |
620 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
621 | whole = 1; |
622 | } | |
623 | ||
624 | if (whole) | |
625 | /* | |
c41564b5 | 626 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 627 | */ |
2bdf7cd0 | 628 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
629 | else |
630 | /* | |
631 | * Since all the modifiable fields are contiguous, we | |
632 | * can get away with this. | |
633 | */ | |
2bdf7cd0 CH |
634 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
635 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 636 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
637 | } |
638 | ||
639 | /* | |
45c34141 DC |
640 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
641 | * and apply superblock counter changes to the in-core superblock. The | |
642 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
643 | * applied to the in-core superblock. The idea is that that has already been | |
644 | * done. | |
1da177e4 LT |
645 | * |
646 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
647 | * However, we have to ensure that we only modify each superblock field only |
648 | * once because the application of the delta values may not be atomic. That can | |
649 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
650 | * free space counter to put back the entire reservation and then take away | |
651 | * what we used. | |
652 | * | |
653 | * If we are not logging superblock counters, then the inode allocated/free and | |
654 | * used block counts are not updated in the on disk superblock. In this case, | |
655 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
656 | * still need to update the incore superblock with the changes. | |
1da177e4 | 657 | */ |
71e330b5 | 658 | void |
1da177e4 LT |
659 | xfs_trans_unreserve_and_mod_sb( |
660 | xfs_trans_t *tp) | |
661 | { | |
662 | xfs_mod_sb_t msb[14]; /* If you add cases, add entries */ | |
663 | xfs_mod_sb_t *msbp; | |
92821e2b | 664 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
665 | /* REFERENCED */ |
666 | int error; | |
667 | int rsvd; | |
45c34141 DC |
668 | int64_t blkdelta = 0; |
669 | int64_t rtxdelta = 0; | |
1da177e4 LT |
670 | |
671 | msbp = msb; | |
672 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
673 | ||
45c34141 DC |
674 | /* calculate free blocks delta */ |
675 | if (tp->t_blk_res > 0) | |
676 | blkdelta = tp->t_blk_res; | |
677 | ||
678 | if ((tp->t_fdblocks_delta != 0) && | |
679 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
680 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
681 | blkdelta += tp->t_fdblocks_delta; | |
682 | ||
683 | if (blkdelta != 0) { | |
1da177e4 | 684 | msbp->msb_field = XFS_SBS_FDBLOCKS; |
45c34141 | 685 | msbp->msb_delta = blkdelta; |
1da177e4 LT |
686 | msbp++; |
687 | } | |
688 | ||
45c34141 DC |
689 | /* calculate free realtime extents delta */ |
690 | if (tp->t_rtx_res > 0) | |
691 | rtxdelta = tp->t_rtx_res; | |
692 | ||
693 | if ((tp->t_frextents_delta != 0) && | |
694 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
695 | rtxdelta += tp->t_frextents_delta; | |
696 | ||
697 | if (rtxdelta != 0) { | |
1da177e4 | 698 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 699 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
700 | msbp++; |
701 | } | |
702 | ||
45c34141 DC |
703 | /* apply remaining deltas */ |
704 | ||
92821e2b DC |
705 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
706 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1da177e4 LT |
707 | if (tp->t_icount_delta != 0) { |
708 | msbp->msb_field = XFS_SBS_ICOUNT; | |
20f4ebf2 | 709 | msbp->msb_delta = tp->t_icount_delta; |
1da177e4 LT |
710 | msbp++; |
711 | } | |
712 | if (tp->t_ifree_delta != 0) { | |
713 | msbp->msb_field = XFS_SBS_IFREE; | |
20f4ebf2 | 714 | msbp->msb_delta = tp->t_ifree_delta; |
1da177e4 LT |
715 | msbp++; |
716 | } | |
92821e2b DC |
717 | } |
718 | ||
719 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { | |
1da177e4 LT |
720 | if (tp->t_dblocks_delta != 0) { |
721 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 722 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
723 | msbp++; |
724 | } | |
725 | if (tp->t_agcount_delta != 0) { | |
726 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 727 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
728 | msbp++; |
729 | } | |
730 | if (tp->t_imaxpct_delta != 0) { | |
731 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 732 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
733 | msbp++; |
734 | } | |
735 | if (tp->t_rextsize_delta != 0) { | |
736 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 737 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
738 | msbp++; |
739 | } | |
740 | if (tp->t_rbmblocks_delta != 0) { | |
741 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 742 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
743 | msbp++; |
744 | } | |
745 | if (tp->t_rblocks_delta != 0) { | |
746 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 747 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
748 | msbp++; |
749 | } | |
750 | if (tp->t_rextents_delta != 0) { | |
751 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 752 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
753 | msbp++; |
754 | } | |
755 | if (tp->t_rextslog_delta != 0) { | |
756 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 757 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
758 | msbp++; |
759 | } | |
760 | } | |
761 | ||
762 | /* | |
763 | * If we need to change anything, do it. | |
764 | */ | |
765 | if (msbp > msb) { | |
766 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
767 | (uint)(msbp - msb), rsvd); | |
768 | ASSERT(error == 0); | |
769 | } | |
770 | } | |
771 | ||
1da177e4 | 772 | /* |
0924378a DC |
773 | * Total up the number of log iovecs needed to commit this |
774 | * transaction. The transaction itself needs one for the | |
775 | * transaction header. Ask each dirty item in turn how many | |
776 | * it needs to get the total. | |
1da177e4 | 777 | */ |
0924378a DC |
778 | static uint |
779 | xfs_trans_count_vecs( | |
b1c1b5b6 | 780 | struct xfs_trans *tp) |
1da177e4 | 781 | { |
0924378a DC |
782 | int nvecs; |
783 | xfs_log_item_desc_t *lidp; | |
1da177e4 | 784 | |
0924378a DC |
785 | nvecs = 1; |
786 | lidp = xfs_trans_first_item(tp); | |
787 | ASSERT(lidp != NULL); | |
1da177e4 | 788 | |
0924378a DC |
789 | /* In the non-debug case we need to start bailing out if we |
790 | * didn't find a log_item here, return zero and let trans_commit | |
791 | * deal with it. | |
1da177e4 | 792 | */ |
0924378a DC |
793 | if (lidp == NULL) |
794 | return 0; | |
795 | ||
796 | while (lidp != NULL) { | |
797 | /* | |
798 | * Skip items which aren't dirty in this transaction. | |
799 | */ | |
800 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
801 | lidp = xfs_trans_next_item(tp, lidp); | |
802 | continue; | |
803 | } | |
804 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
805 | nvecs += lidp->lid_size; | |
806 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 807 | } |
0924378a DC |
808 | |
809 | return nvecs; | |
810 | } | |
811 | ||
812 | /* | |
813 | * Fill in the vector with pointers to data to be logged | |
814 | * by this transaction. The transaction header takes | |
815 | * the first vector, and then each dirty item takes the | |
816 | * number of vectors it indicated it needed in xfs_trans_count_vecs(). | |
817 | * | |
818 | * As each item fills in the entries it needs, also pin the item | |
819 | * so that it cannot be flushed out until the log write completes. | |
820 | */ | |
821 | static void | |
822 | xfs_trans_fill_vecs( | |
823 | struct xfs_trans *tp, | |
824 | struct xfs_log_iovec *log_vector) | |
825 | { | |
826 | xfs_log_item_desc_t *lidp; | |
827 | struct xfs_log_iovec *vecp; | |
828 | uint nitems; | |
1da177e4 LT |
829 | |
830 | /* | |
0924378a DC |
831 | * Skip over the entry for the transaction header, we'll |
832 | * fill that in at the end. | |
1da177e4 | 833 | */ |
0924378a DC |
834 | vecp = log_vector + 1; |
835 | ||
836 | nitems = 0; | |
837 | lidp = xfs_trans_first_item(tp); | |
838 | ASSERT(lidp); | |
839 | while (lidp) { | |
840 | /* Skip items which aren't dirty in this transaction. */ | |
841 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
842 | lidp = xfs_trans_next_item(tp, lidp); | |
843 | continue; | |
844 | } | |
845 | ||
1da177e4 | 846 | /* |
0924378a DC |
847 | * The item may be marked dirty but not log anything. This can |
848 | * be used to get called when a transaction is committed. | |
1da177e4 | 849 | */ |
0924378a DC |
850 | if (lidp->lid_size) |
851 | nitems++; | |
852 | IOP_FORMAT(lidp->lid_item, vecp); | |
853 | vecp += lidp->lid_size; | |
854 | IOP_PIN(lidp->lid_item); | |
855 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 856 | } |
1da177e4 LT |
857 | |
858 | /* | |
0924378a DC |
859 | * Now that we've counted the number of items in this transaction, fill |
860 | * in the transaction header. Note that the transaction header does not | |
861 | * have a log item. | |
1da177e4 | 862 | */ |
0924378a DC |
863 | tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC; |
864 | tp->t_header.th_type = tp->t_type; | |
865 | tp->t_header.th_num_items = nitems; | |
866 | log_vector->i_addr = (xfs_caddr_t)&tp->t_header; | |
867 | log_vector->i_len = sizeof(xfs_trans_header_t); | |
868 | log_vector->i_type = XLOG_REG_TYPE_TRANSHDR; | |
869 | } | |
870 | ||
b1c1b5b6 DC |
871 | /* |
872 | * The committed item processing consists of calling the committed routine of | |
873 | * each logged item, updating the item's position in the AIL if necessary, and | |
874 | * unpinning each item. If the committed routine returns -1, then do nothing | |
875 | * further with the item because it may have been freed. | |
876 | * | |
877 | * Since items are unlocked when they are copied to the incore log, it is | |
878 | * possible for two transactions to be completing and manipulating the same | |
879 | * item simultaneously. The AIL lock will protect the lsn field of each item. | |
880 | * The value of this field can never go backwards. | |
881 | * | |
882 | * We unpin the items after repositioning them in the AIL, because otherwise | |
883 | * they could be immediately flushed and we'd have to race with the flusher | |
884 | * trying to pull the item from the AIL as we add it. | |
885 | */ | |
71e330b5 | 886 | void |
b1c1b5b6 DC |
887 | xfs_trans_item_committed( |
888 | struct xfs_log_item *lip, | |
889 | xfs_lsn_t commit_lsn, | |
890 | int aborted) | |
891 | { | |
892 | xfs_lsn_t item_lsn; | |
893 | struct xfs_ail *ailp; | |
894 | ||
895 | if (aborted) | |
896 | lip->li_flags |= XFS_LI_ABORTED; | |
897 | item_lsn = IOP_COMMITTED(lip, commit_lsn); | |
898 | ||
899 | /* If the committed routine returns -1, item has been freed. */ | |
900 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) | |
901 | return; | |
902 | ||
903 | /* | |
904 | * If the returned lsn is greater than what it contained before, update | |
905 | * the location of the item in the AIL. If it is not, then do nothing. | |
906 | * Items can never move backwards in the AIL. | |
907 | * | |
908 | * While the new lsn should usually be greater, it is possible that a | |
909 | * later transaction completing simultaneously with an earlier one | |
910 | * using the same item could complete first with a higher lsn. This | |
911 | * would cause the earlier transaction to fail the test below. | |
912 | */ | |
913 | ailp = lip->li_ailp; | |
914 | spin_lock(&ailp->xa_lock); | |
915 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) { | |
916 | /* | |
917 | * This will set the item's lsn to item_lsn and update the | |
918 | * position of the item in the AIL. | |
919 | * | |
920 | * xfs_trans_ail_update() drops the AIL lock. | |
921 | */ | |
922 | xfs_trans_ail_update(ailp, lip, item_lsn); | |
923 | } else { | |
924 | spin_unlock(&ailp->xa_lock); | |
925 | } | |
926 | ||
927 | /* | |
928 | * Now that we've repositioned the item in the AIL, unpin it so it can | |
929 | * be flushed. Pass information about buffer stale state down from the | |
930 | * log item flags, if anyone else stales the buffer we do not want to | |
931 | * pay any attention to it. | |
932 | */ | |
933 | IOP_UNPIN(lip); | |
934 | } | |
935 | ||
b1c1b5b6 DC |
936 | /* |
937 | * This is typically called by the LM when a transaction has been fully | |
938 | * committed to disk. It needs to unpin the items which have | |
939 | * been logged by the transaction and update their positions | |
940 | * in the AIL if necessary. | |
941 | * | |
942 | * This also gets called when the transactions didn't get written out | |
943 | * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then. | |
944 | */ | |
945 | STATIC void | |
946 | xfs_trans_committed( | |
947 | struct xfs_trans *tp, | |
948 | int abortflag) | |
949 | { | |
950 | xfs_log_item_desc_t *lidp; | |
951 | xfs_log_item_chunk_t *licp; | |
952 | xfs_log_item_chunk_t *next_licp; | |
953 | ||
954 | /* Call the transaction's completion callback if there is one. */ | |
955 | if (tp->t_callback != NULL) | |
956 | tp->t_callback(tp, tp->t_callarg); | |
957 | ||
958 | for (lidp = xfs_trans_first_item(tp); | |
959 | lidp != NULL; | |
960 | lidp = xfs_trans_next_item(tp, lidp)) { | |
961 | xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag); | |
962 | } | |
963 | ||
964 | /* free the item chunks, ignoring the embedded chunk */ | |
965 | for (licp = tp->t_items.lic_next; licp != NULL; licp = next_licp) { | |
966 | next_licp = licp->lic_next; | |
967 | kmem_free(licp); | |
968 | } | |
969 | ||
b1c1b5b6 DC |
970 | xfs_trans_free(tp); |
971 | } | |
972 | ||
973 | /* | |
974 | * Called from the trans_commit code when we notice that | |
975 | * the filesystem is in the middle of a forced shutdown. | |
976 | */ | |
977 | STATIC void | |
978 | xfs_trans_uncommit( | |
979 | struct xfs_trans *tp, | |
980 | uint flags) | |
981 | { | |
982 | xfs_log_item_desc_t *lidp; | |
983 | ||
984 | for (lidp = xfs_trans_first_item(tp); | |
985 | lidp != NULL; | |
986 | lidp = xfs_trans_next_item(tp, lidp)) { | |
987 | /* | |
988 | * Unpin all but those that aren't dirty. | |
989 | */ | |
990 | if (lidp->lid_flags & XFS_LID_DIRTY) | |
991 | IOP_UNPIN_REMOVE(lidp->lid_item, tp); | |
992 | } | |
993 | ||
994 | xfs_trans_unreserve_and_mod_sb(tp); | |
995 | xfs_trans_unreserve_and_mod_dquots(tp); | |
996 | ||
71e330b5 | 997 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
b1c1b5b6 DC |
998 | xfs_trans_free(tp); |
999 | } | |
1000 | ||
0924378a DC |
1001 | /* |
1002 | * Format the transaction direct to the iclog. This isolates the physical | |
1003 | * transaction commit operation from the logical operation and hence allows | |
1004 | * other methods to be introduced without affecting the existing commit path. | |
1005 | */ | |
1006 | static int | |
1007 | xfs_trans_commit_iclog( | |
1008 | struct xfs_mount *mp, | |
1009 | struct xfs_trans *tp, | |
1010 | xfs_lsn_t *commit_lsn, | |
1011 | int flags) | |
1012 | { | |
1013 | int shutdown; | |
1014 | int error; | |
1015 | int log_flags = 0; | |
1016 | struct xlog_in_core *commit_iclog; | |
1017 | #define XFS_TRANS_LOGVEC_COUNT 16 | |
1018 | struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT]; | |
1019 | struct xfs_log_iovec *log_vector; | |
1020 | uint nvec; | |
1021 | ||
1da177e4 LT |
1022 | |
1023 | /* | |
1024 | * Ask each log item how many log_vector entries it will | |
1025 | * need so we can figure out how many to allocate. | |
1026 | * Try to avoid the kmem_alloc() call in the common case | |
1027 | * by using a vector from the stack when it fits. | |
1028 | */ | |
1029 | nvec = xfs_trans_count_vecs(tp); | |
1da177e4 | 1030 | if (nvec == 0) { |
0924378a | 1031 | return ENOMEM; /* triggers a shutdown! */ |
cfcbbbd0 | 1032 | } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) { |
1da177e4 LT |
1033 | log_vector = log_vector_fast; |
1034 | } else { | |
1035 | log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec * | |
1036 | sizeof(xfs_log_iovec_t), | |
1037 | KM_SLEEP); | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Fill in the log_vector and pin the logged items, and | |
1042 | * then write the transaction to the log. | |
1043 | */ | |
1044 | xfs_trans_fill_vecs(tp, log_vector); | |
1045 | ||
0924378a DC |
1046 | if (flags & XFS_TRANS_RELEASE_LOG_RES) |
1047 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1048 | ||
cfcbbbd0 | 1049 | error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn)); |
1da177e4 | 1050 | |
1da177e4 | 1051 | /* |
cfcbbbd0 NS |
1052 | * The transaction is committed incore here, and can go out to disk |
1053 | * at any time after this call. However, all the items associated | |
1054 | * with the transaction are still locked and pinned in memory. | |
1da177e4 | 1055 | */ |
0924378a | 1056 | *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); |
1da177e4 | 1057 | |
0924378a | 1058 | tp->t_commit_lsn = *commit_lsn; |
ed3b4d6c DC |
1059 | trace_xfs_trans_commit_lsn(tp); |
1060 | ||
0924378a | 1061 | if (nvec > XFS_TRANS_LOGVEC_COUNT) |
f0e2d93c | 1062 | kmem_free(log_vector); |
1da177e4 | 1063 | |
1da177e4 LT |
1064 | /* |
1065 | * If we got a log write error. Unpin the logitems that we | |
1066 | * had pinned, clean up, free trans structure, and return error. | |
1067 | */ | |
0924378a | 1068 | if (error || *commit_lsn == -1) { |
59c1b082 | 1069 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1070 | xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT); |
1071 | return XFS_ERROR(EIO); | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * Once the transaction has committed, unused | |
1076 | * reservations need to be released and changes to | |
1077 | * the superblock need to be reflected in the in-core | |
1078 | * version. Do that now. | |
1079 | */ | |
1080 | xfs_trans_unreserve_and_mod_sb(tp); | |
1081 | ||
1da177e4 LT |
1082 | /* |
1083 | * Tell the LM to call the transaction completion routine | |
1084 | * when the log write with LSN commit_lsn completes (e.g. | |
1085 | * when the transaction commit really hits the on-disk log). | |
1086 | * After this call we cannot reference tp, because the call | |
1087 | * can happen at any time and the call will free the transaction | |
1088 | * structure pointed to by tp. The only case where we call | |
1089 | * the completion routine (xfs_trans_committed) directly is | |
1090 | * if the log is turned off on a debug kernel or we're | |
1091 | * running in simulation mode (the log is explicitly turned | |
1092 | * off). | |
1093 | */ | |
1094 | tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed; | |
1095 | tp->t_logcb.cb_arg = tp; | |
1096 | ||
1097 | /* | |
1098 | * We need to pass the iclog buffer which was used for the | |
1099 | * transaction commit record into this function, and attach | |
1100 | * the callback to it. The callback must be attached before | |
1101 | * the items are unlocked to avoid racing with other threads | |
1102 | * waiting for an item to unlock. | |
1103 | */ | |
1104 | shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb)); | |
1105 | ||
1106 | /* | |
1107 | * Mark this thread as no longer being in a transaction | |
1108 | */ | |
59c1b082 | 1109 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1110 | |
1111 | /* | |
1112 | * Once all the items of the transaction have been copied | |
1113 | * to the in core log and the callback is attached, the | |
1114 | * items can be unlocked. | |
1115 | * | |
1116 | * This will free descriptors pointing to items which were | |
1117 | * not logged since there is nothing more to do with them. | |
1118 | * For items which were logged, we will keep pointers to them | |
1119 | * so they can be unpinned after the transaction commits to disk. | |
1120 | * This will also stamp each modified meta-data item with | |
1121 | * the commit lsn of this transaction for dependency tracking | |
1122 | * purposes. | |
1123 | */ | |
0924378a | 1124 | xfs_trans_unlock_items(tp, *commit_lsn); |
1da177e4 LT |
1125 | |
1126 | /* | |
1127 | * If we detected a log error earlier, finish committing | |
1128 | * the transaction now (unpin log items, etc). | |
1129 | * | |
1130 | * Order is critical here, to avoid using the transaction | |
1131 | * pointer after its been freed (by xfs_trans_committed | |
1132 | * either here now, or as a callback). We cannot do this | |
1133 | * step inside xfs_log_notify as was done earlier because | |
1134 | * of this issue. | |
1135 | */ | |
1136 | if (shutdown) | |
1137 | xfs_trans_committed(tp, XFS_LI_ABORTED); | |
1138 | ||
1139 | /* | |
1140 | * Now that the xfs_trans_committed callback has been attached, | |
1141 | * and the items are released we can finally allow the iclog to | |
1142 | * go to disk. | |
1143 | */ | |
0924378a DC |
1144 | return xfs_log_release_iclog(mp, commit_iclog); |
1145 | } | |
1146 | ||
71e330b5 DC |
1147 | /* |
1148 | * Walk the log items and allocate log vector structures for | |
1149 | * each item large enough to fit all the vectors they require. | |
1150 | * Note that this format differs from the old log vector format in | |
1151 | * that there is no transaction header in these log vectors. | |
1152 | */ | |
1153 | STATIC struct xfs_log_vec * | |
1154 | xfs_trans_alloc_log_vecs( | |
1155 | xfs_trans_t *tp) | |
1156 | { | |
1157 | xfs_log_item_desc_t *lidp; | |
1158 | struct xfs_log_vec *lv = NULL; | |
1159 | struct xfs_log_vec *ret_lv = NULL; | |
1160 | ||
1161 | lidp = xfs_trans_first_item(tp); | |
1162 | ||
1163 | /* Bail out if we didn't find a log item. */ | |
1164 | if (!lidp) { | |
1165 | ASSERT(0); | |
1166 | return NULL; | |
1167 | } | |
1168 | ||
1169 | while (lidp != NULL) { | |
1170 | struct xfs_log_vec *new_lv; | |
1171 | ||
1172 | /* Skip items which aren't dirty in this transaction. */ | |
1173 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
1174 | lidp = xfs_trans_next_item(tp, lidp); | |
1175 | continue; | |
1176 | } | |
1177 | ||
1178 | /* Skip items that do not have any vectors for writing */ | |
1179 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
1180 | if (!lidp->lid_size) { | |
1181 | lidp = xfs_trans_next_item(tp, lidp); | |
1182 | continue; | |
1183 | } | |
1184 | ||
1185 | new_lv = kmem_zalloc(sizeof(*new_lv) + | |
1186 | lidp->lid_size * sizeof(struct xfs_log_iovec), | |
1187 | KM_SLEEP); | |
1188 | ||
1189 | /* The allocated iovec region lies beyond the log vector. */ | |
1190 | new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1]; | |
1191 | new_lv->lv_niovecs = lidp->lid_size; | |
1192 | new_lv->lv_item = lidp->lid_item; | |
1193 | if (!ret_lv) | |
1194 | ret_lv = new_lv; | |
1195 | else | |
1196 | lv->lv_next = new_lv; | |
1197 | lv = new_lv; | |
1198 | lidp = xfs_trans_next_item(tp, lidp); | |
1199 | } | |
1200 | ||
1201 | return ret_lv; | |
1202 | } | |
1203 | ||
1204 | static int | |
1205 | xfs_trans_commit_cil( | |
1206 | struct xfs_mount *mp, | |
1207 | struct xfs_trans *tp, | |
1208 | xfs_lsn_t *commit_lsn, | |
1209 | int flags) | |
1210 | { | |
1211 | struct xfs_log_vec *log_vector; | |
1212 | int error; | |
1213 | ||
1214 | /* | |
1215 | * Get each log item to allocate a vector structure for | |
1216 | * the log item to to pass to the log write code. The | |
1217 | * CIL commit code will format the vector and save it away. | |
1218 | */ | |
1219 | log_vector = xfs_trans_alloc_log_vecs(tp); | |
1220 | if (!log_vector) | |
1221 | return ENOMEM; | |
1222 | ||
1223 | error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags); | |
1224 | if (error) | |
1225 | return error; | |
1226 | ||
1227 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
1228 | ||
1229 | /* xfs_trans_free_items() unlocks them first */ | |
1230 | xfs_trans_free_items(tp, *commit_lsn, 0); | |
1231 | xfs_trans_free(tp); | |
1232 | return 0; | |
1233 | } | |
0924378a DC |
1234 | |
1235 | /* | |
1236 | * xfs_trans_commit | |
1237 | * | |
1238 | * Commit the given transaction to the log a/synchronously. | |
1239 | * | |
1240 | * XFS disk error handling mechanism is not based on a typical | |
1241 | * transaction abort mechanism. Logically after the filesystem | |
1242 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1243 | * be durable - ie. committed to disk - because some metadata might | |
1244 | * be inconsistent. In such cases, this returns an error, and the | |
1245 | * caller may assume that all locked objects joined to the transaction | |
1246 | * have already been unlocked as if the commit had succeeded. | |
1247 | * Do not reference the transaction structure after this call. | |
1248 | */ | |
0924378a DC |
1249 | int |
1250 | _xfs_trans_commit( | |
a3ccd2ca CH |
1251 | struct xfs_trans *tp, |
1252 | uint flags, | |
1253 | int *log_flushed) | |
0924378a | 1254 | { |
a3ccd2ca | 1255 | struct xfs_mount *mp = tp->t_mountp; |
0924378a | 1256 | xfs_lsn_t commit_lsn = -1; |
a3ccd2ca | 1257 | int error = 0; |
0924378a DC |
1258 | int log_flags = 0; |
1259 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
0924378a DC |
1260 | |
1261 | /* | |
1262 | * Determine whether this commit is releasing a permanent | |
1263 | * log reservation or not. | |
1264 | */ | |
1265 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1266 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1267 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * If there is nothing to be logged by the transaction, | |
1272 | * then unlock all of the items associated with the | |
1273 | * transaction and free the transaction structure. | |
1274 | * Also make sure to return any reserved blocks to | |
1275 | * the free pool. | |
1276 | */ | |
a3ccd2ca CH |
1277 | if (!(tp->t_flags & XFS_TRANS_DIRTY)) |
1278 | goto out_unreserve; | |
1279 | ||
1280 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1281 | error = XFS_ERROR(EIO); | |
1282 | goto out_unreserve; | |
0924378a | 1283 | } |
a3ccd2ca | 1284 | |
0924378a DC |
1285 | ASSERT(tp->t_ticket != NULL); |
1286 | ||
1287 | /* | |
1288 | * If we need to update the superblock, then do it now. | |
1289 | */ | |
1290 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1291 | xfs_trans_apply_sb_deltas(tp); | |
1292 | xfs_trans_apply_dquot_deltas(tp); | |
1293 | ||
71e330b5 DC |
1294 | if (mp->m_flags & XFS_MOUNT_DELAYLOG) |
1295 | error = xfs_trans_commit_cil(mp, tp, &commit_lsn, flags); | |
1296 | else | |
1297 | error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags); | |
1298 | ||
0924378a DC |
1299 | if (error == ENOMEM) { |
1300 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
a3ccd2ca CH |
1301 | error = XFS_ERROR(EIO); |
1302 | goto out_unreserve; | |
0924378a | 1303 | } |
1da177e4 LT |
1304 | |
1305 | /* | |
1306 | * If the transaction needs to be synchronous, then force the | |
1307 | * log out now and wait for it. | |
1308 | */ | |
1309 | if (sync) { | |
f538d4da | 1310 | if (!error) { |
a14a348b CH |
1311 | error = _xfs_log_force_lsn(mp, commit_lsn, |
1312 | XFS_LOG_SYNC, log_flushed); | |
f538d4da | 1313 | } |
1da177e4 LT |
1314 | XFS_STATS_INC(xs_trans_sync); |
1315 | } else { | |
1316 | XFS_STATS_INC(xs_trans_async); | |
1317 | } | |
1318 | ||
a3ccd2ca CH |
1319 | return error; |
1320 | ||
1321 | out_unreserve: | |
1322 | xfs_trans_unreserve_and_mod_sb(tp); | |
1323 | ||
1324 | /* | |
1325 | * It is indeed possible for the transaction to be not dirty but | |
1326 | * the dqinfo portion to be. All that means is that we have some | |
1327 | * (non-persistent) quota reservations that need to be unreserved. | |
1328 | */ | |
1329 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1330 | if (tp->t_ticket) { | |
1331 | commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
1332 | if (commit_lsn == -1 && !error) | |
1333 | error = XFS_ERROR(EIO); | |
1334 | } | |
1335 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
71e330b5 | 1336 | xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0); |
a3ccd2ca CH |
1337 | xfs_trans_free(tp); |
1338 | ||
1339 | XFS_STATS_INC(xs_trans_empty); | |
1340 | return error; | |
1da177e4 LT |
1341 | } |
1342 | ||
1da177e4 LT |
1343 | /* |
1344 | * Unlock all of the transaction's items and free the transaction. | |
1345 | * The transaction must not have modified any of its items, because | |
1346 | * there is no way to restore them to their previous state. | |
1347 | * | |
1348 | * If the transaction has made a log reservation, make sure to release | |
1349 | * it as well. | |
1350 | */ | |
1351 | void | |
1352 | xfs_trans_cancel( | |
1353 | xfs_trans_t *tp, | |
1354 | int flags) | |
1355 | { | |
1356 | int log_flags; | |
1357 | #ifdef DEBUG | |
1358 | xfs_log_item_chunk_t *licp; | |
1359 | xfs_log_item_desc_t *lidp; | |
1360 | xfs_log_item_t *lip; | |
1361 | int i; | |
1362 | #endif | |
0733af21 | 1363 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1364 | |
1365 | /* | |
1366 | * See if the caller is being too lazy to figure out if | |
1367 | * the transaction really needs an abort. | |
1368 | */ | |
1369 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1370 | flags &= ~XFS_TRANS_ABORT; | |
1371 | /* | |
1372 | * See if the caller is relying on us to shut down the | |
1373 | * filesystem. This happens in paths where we detect | |
1374 | * corruption and decide to give up. | |
1375 | */ | |
60a204f0 | 1376 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1377 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1378 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1379 | } |
1da177e4 LT |
1380 | #ifdef DEBUG |
1381 | if (!(flags & XFS_TRANS_ABORT)) { | |
1382 | licp = &(tp->t_items); | |
1383 | while (licp != NULL) { | |
1384 | lidp = licp->lic_descs; | |
1385 | for (i = 0; i < licp->lic_unused; i++, lidp++) { | |
39dab9d7 | 1386 | if (xfs_lic_isfree(licp, i)) { |
1da177e4 LT |
1387 | continue; |
1388 | } | |
1389 | ||
1390 | lip = lidp->lid_item; | |
0733af21 | 1391 | if (!XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 LT |
1392 | ASSERT(!(lip->li_type == XFS_LI_EFD)); |
1393 | } | |
1394 | licp = licp->lic_next; | |
1395 | } | |
1396 | } | |
1397 | #endif | |
1398 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1399 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1400 | |
1401 | if (tp->t_ticket) { | |
1402 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1403 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1404 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1405 | } else { | |
1406 | log_flags = 0; | |
1407 | } | |
0733af21 | 1408 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1412 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 1413 | |
71e330b5 | 1414 | xfs_trans_free_items(tp, NULLCOMMITLSN, flags); |
1da177e4 LT |
1415 | xfs_trans_free(tp); |
1416 | } | |
1417 | ||
322ff6b8 NS |
1418 | /* |
1419 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1420 | * the next: permanent transactions are only flushed out when | |
1421 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1422 | * as possible to let chunks of it go to the log. So we commit the | |
1423 | * chunk we've been working on and get a new transaction to continue. | |
1424 | */ | |
1425 | int | |
1426 | xfs_trans_roll( | |
1427 | struct xfs_trans **tpp, | |
1428 | struct xfs_inode *dp) | |
1429 | { | |
1430 | struct xfs_trans *trans; | |
1431 | unsigned int logres, count; | |
1432 | int error; | |
1433 | ||
1434 | /* | |
1435 | * Ensure that the inode is always logged. | |
1436 | */ | |
1437 | trans = *tpp; | |
1438 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1439 | ||
1440 | /* | |
1441 | * Copy the critical parameters from one trans to the next. | |
1442 | */ | |
1443 | logres = trans->t_log_res; | |
1444 | count = trans->t_log_count; | |
1445 | *tpp = xfs_trans_dup(trans); | |
1446 | ||
1447 | /* | |
1448 | * Commit the current transaction. | |
1449 | * If this commit failed, then it'd just unlock those items that | |
1450 | * are not marked ihold. That also means that a filesystem shutdown | |
1451 | * is in progress. The caller takes the responsibility to cancel | |
1452 | * the duplicate transaction that gets returned. | |
1453 | */ | |
1454 | error = xfs_trans_commit(trans, 0); | |
1455 | if (error) | |
1456 | return (error); | |
1457 | ||
1458 | trans = *tpp; | |
1459 | ||
cc09c0dc DC |
1460 | /* |
1461 | * transaction commit worked ok so we can drop the extra ticket | |
1462 | * reference that we gained in xfs_trans_dup() | |
1463 | */ | |
1464 | xfs_log_ticket_put(trans->t_ticket); | |
1465 | ||
1466 | ||
322ff6b8 NS |
1467 | /* |
1468 | * Reserve space in the log for th next transaction. | |
1469 | * This also pushes items in the "AIL", the list of logged items, | |
1470 | * out to disk if they are taking up space at the tail of the log | |
1471 | * that we want to use. This requires that either nothing be locked | |
1472 | * across this call, or that anything that is locked be logged in | |
1473 | * the prior and the next transactions. | |
1474 | */ | |
1475 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1476 | XFS_TRANS_PERM_LOG_RES, count); | |
1477 | /* | |
1478 | * Ensure that the inode is in the new transaction and locked. | |
1479 | */ | |
1480 | if (error) | |
1481 | return error; | |
1482 | ||
1483 | xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL); | |
1484 | xfs_trans_ihold(trans, dp); | |
1485 | return 0; | |
1486 | } |