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" |
1da177e4 LT |
47 | |
48 | ||
49 | STATIC void xfs_trans_apply_sb_deltas(xfs_trans_t *); | |
1da177e4 LT |
50 | STATIC void xfs_trans_uncommit(xfs_trans_t *, uint); |
51 | STATIC void xfs_trans_committed(xfs_trans_t *, int); | |
52 | STATIC void xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int); | |
53 | STATIC void xfs_trans_free(xfs_trans_t *); | |
54 | ||
8f794055 | 55 | kmem_zone_t *xfs_trans_zone; |
1da177e4 LT |
56 | |
57 | ||
8f794055 NS |
58 | /* |
59 | * Reservation functions here avoid a huge stack in xfs_trans_init | |
60 | * due to register overflow from temporaries in the calculations. | |
61 | */ | |
62 | ||
63 | STATIC uint | |
64 | xfs_calc_write_reservation(xfs_mount_t *mp) | |
65 | { | |
66 | return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
67 | } | |
68 | ||
69 | STATIC uint | |
70 | xfs_calc_itruncate_reservation(xfs_mount_t *mp) | |
71 | { | |
72 | return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
73 | } | |
74 | ||
75 | STATIC uint | |
76 | xfs_calc_rename_reservation(xfs_mount_t *mp) | |
77 | { | |
78 | return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
79 | } | |
80 | ||
81 | STATIC uint | |
82 | xfs_calc_link_reservation(xfs_mount_t *mp) | |
83 | { | |
2ddd5928 | 84 | return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); |
8f794055 NS |
85 | } |
86 | ||
87 | STATIC uint | |
88 | xfs_calc_remove_reservation(xfs_mount_t *mp) | |
89 | { | |
90 | return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
91 | } | |
92 | ||
93 | STATIC uint | |
94 | xfs_calc_symlink_reservation(xfs_mount_t *mp) | |
95 | { | |
96 | return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
97 | } | |
98 | ||
99 | STATIC uint | |
100 | xfs_calc_create_reservation(xfs_mount_t *mp) | |
101 | { | |
102 | return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
103 | } | |
104 | ||
105 | STATIC uint | |
106 | xfs_calc_mkdir_reservation(xfs_mount_t *mp) | |
107 | { | |
108 | return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
109 | } | |
110 | ||
111 | STATIC uint | |
112 | xfs_calc_ifree_reservation(xfs_mount_t *mp) | |
113 | { | |
114 | return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
115 | } | |
116 | ||
117 | STATIC uint | |
118 | xfs_calc_ichange_reservation(xfs_mount_t *mp) | |
119 | { | |
120 | return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
121 | } | |
122 | ||
123 | STATIC uint | |
124 | xfs_calc_growdata_reservation(xfs_mount_t *mp) | |
125 | { | |
126 | return XFS_CALC_GROWDATA_LOG_RES(mp); | |
127 | } | |
128 | ||
129 | STATIC uint | |
130 | xfs_calc_growrtalloc_reservation(xfs_mount_t *mp) | |
131 | { | |
132 | return XFS_CALC_GROWRTALLOC_LOG_RES(mp); | |
133 | } | |
134 | ||
135 | STATIC uint | |
136 | xfs_calc_growrtzero_reservation(xfs_mount_t *mp) | |
137 | { | |
138 | return XFS_CALC_GROWRTZERO_LOG_RES(mp); | |
139 | } | |
140 | ||
141 | STATIC uint | |
142 | xfs_calc_growrtfree_reservation(xfs_mount_t *mp) | |
143 | { | |
144 | return XFS_CALC_GROWRTFREE_LOG_RES(mp); | |
145 | } | |
146 | ||
147 | STATIC uint | |
148 | xfs_calc_swrite_reservation(xfs_mount_t *mp) | |
149 | { | |
150 | return XFS_CALC_SWRITE_LOG_RES(mp); | |
151 | } | |
152 | ||
153 | STATIC uint | |
154 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
155 | { | |
156 | return XFS_CALC_WRITEID_LOG_RES(mp); | |
157 | } | |
158 | ||
159 | STATIC uint | |
160 | xfs_calc_addafork_reservation(xfs_mount_t *mp) | |
161 | { | |
162 | return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
163 | } | |
164 | ||
165 | STATIC uint | |
166 | xfs_calc_attrinval_reservation(xfs_mount_t *mp) | |
167 | { | |
168 | return XFS_CALC_ATTRINVAL_LOG_RES(mp); | |
169 | } | |
170 | ||
171 | STATIC uint | |
172 | xfs_calc_attrset_reservation(xfs_mount_t *mp) | |
173 | { | |
174 | return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
175 | } | |
176 | ||
177 | STATIC uint | |
178 | xfs_calc_attrrm_reservation(xfs_mount_t *mp) | |
179 | { | |
180 | return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
181 | } | |
182 | ||
183 | STATIC uint | |
184 | xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp) | |
185 | { | |
186 | return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp); | |
187 | } | |
188 | ||
1da177e4 LT |
189 | /* |
190 | * Initialize the precomputed transaction reservation values | |
191 | * in the mount structure. | |
192 | */ | |
193 | void | |
194 | xfs_trans_init( | |
195 | xfs_mount_t *mp) | |
196 | { | |
197 | xfs_trans_reservations_t *resp; | |
198 | ||
199 | resp = &(mp->m_reservations); | |
8f794055 NS |
200 | resp->tr_write = xfs_calc_write_reservation(mp); |
201 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
202 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
203 | resp->tr_link = xfs_calc_link_reservation(mp); | |
204 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
205 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
206 | resp->tr_create = xfs_calc_create_reservation(mp); | |
207 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
208 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
209 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
210 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
211 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
212 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
213 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
214 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
215 | resp->tr_attrset = xfs_calc_attrset_reservation(mp); | |
216 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); | |
217 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
218 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
219 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
220 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
1da177e4 LT |
221 | } |
222 | ||
223 | /* | |
224 | * This routine is called to allocate a transaction structure. | |
225 | * The type parameter indicates the type of the transaction. These | |
226 | * are enumerated in xfs_trans.h. | |
227 | * | |
228 | * Dynamically allocate the transaction structure from the transaction | |
229 | * zone, initialize it, and return it to the caller. | |
230 | */ | |
231 | xfs_trans_t * | |
232 | xfs_trans_alloc( | |
233 | xfs_mount_t *mp, | |
234 | uint type) | |
235 | { | |
b267ce99 | 236 | xfs_wait_for_freeze(mp, SB_FREEZE_TRANS); |
80641dc6 | 237 | return _xfs_trans_alloc(mp, type, KM_SLEEP); |
1da177e4 LT |
238 | } |
239 | ||
240 | xfs_trans_t * | |
241 | _xfs_trans_alloc( | |
242 | xfs_mount_t *mp, | |
80641dc6 CH |
243 | uint type, |
244 | uint memflags) | |
1da177e4 LT |
245 | { |
246 | xfs_trans_t *tp; | |
247 | ||
34327e13 | 248 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 249 | |
80641dc6 | 250 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
251 | tp->t_magic = XFS_TRANS_MAGIC; |
252 | tp->t_type = type; | |
253 | tp->t_mountp = mp; | |
254 | tp->t_items_free = XFS_LIC_NUM_SLOTS; | |
255 | tp->t_busy_free = XFS_LBC_NUM_SLOTS; | |
39dab9d7 | 256 | xfs_lic_init(&(tp->t_items)); |
1da177e4 | 257 | XFS_LBC_INIT(&(tp->t_busy)); |
34327e13 | 258 | return tp; |
1da177e4 LT |
259 | } |
260 | ||
261 | /* | |
262 | * This is called to create a new transaction which will share the | |
263 | * permanent log reservation of the given transaction. The remaining | |
264 | * unused block and rt extent reservations are also inherited. This | |
265 | * implies that the original transaction is no longer allowed to allocate | |
266 | * blocks. Locks and log items, however, are no inherited. They must | |
267 | * be added to the new transaction explicitly. | |
268 | */ | |
269 | xfs_trans_t * | |
270 | xfs_trans_dup( | |
271 | xfs_trans_t *tp) | |
272 | { | |
273 | xfs_trans_t *ntp; | |
274 | ||
275 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
276 | ||
277 | /* | |
278 | * Initialize the new transaction structure. | |
279 | */ | |
280 | ntp->t_magic = XFS_TRANS_MAGIC; | |
281 | ntp->t_type = tp->t_type; | |
282 | ntp->t_mountp = tp->t_mountp; | |
283 | ntp->t_items_free = XFS_LIC_NUM_SLOTS; | |
284 | ntp->t_busy_free = XFS_LBC_NUM_SLOTS; | |
39dab9d7 | 285 | xfs_lic_init(&(ntp->t_items)); |
1da177e4 LT |
286 | XFS_LBC_INIT(&(ntp->t_busy)); |
287 | ||
288 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 289 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 290 | |
1da177e4 | 291 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE); |
cc09c0dc | 292 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
293 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
294 | tp->t_blk_res = tp->t_blk_res_used; | |
295 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
296 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 297 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 298 | |
7d095257 | 299 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
300 | |
301 | atomic_inc(&tp->t_mountp->m_active_trans); | |
302 | return ntp; | |
303 | } | |
304 | ||
305 | /* | |
306 | * This is called to reserve free disk blocks and log space for the | |
307 | * given transaction. This must be done before allocating any resources | |
308 | * within the transaction. | |
309 | * | |
310 | * This will return ENOSPC if there are not enough blocks available. | |
311 | * It will sleep waiting for available log space. | |
312 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
313 | * is used by long running transactions. If any one of the reservations | |
314 | * fails then they will all be backed out. | |
315 | * | |
316 | * This does not do quota reservations. That typically is done by the | |
317 | * caller afterwards. | |
318 | */ | |
319 | int | |
320 | xfs_trans_reserve( | |
321 | xfs_trans_t *tp, | |
322 | uint blocks, | |
323 | uint logspace, | |
324 | uint rtextents, | |
325 | uint flags, | |
326 | uint logcount) | |
327 | { | |
328 | int log_flags; | |
59c1b082 NS |
329 | int error = 0; |
330 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
331 | |
332 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 333 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
334 | |
335 | /* | |
336 | * Attempt to reserve the needed disk blocks by decrementing | |
337 | * the number needed from the number available. This will | |
338 | * fail if the count would go below zero. | |
339 | */ | |
340 | if (blocks > 0) { | |
341 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 342 | -((int64_t)blocks), rsvd); |
1da177e4 | 343 | if (error != 0) { |
59c1b082 | 344 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
345 | return (XFS_ERROR(ENOSPC)); |
346 | } | |
347 | tp->t_blk_res += blocks; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Reserve the log space needed for this transaction. | |
352 | */ | |
353 | if (logspace > 0) { | |
354 | ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace)); | |
355 | ASSERT((tp->t_log_count == 0) || | |
356 | (tp->t_log_count == logcount)); | |
357 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
358 | log_flags = XFS_LOG_PERM_RESERV; | |
359 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; | |
360 | } else { | |
361 | ASSERT(tp->t_ticket == NULL); | |
362 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
363 | log_flags = 0; | |
364 | } | |
365 | ||
366 | error = xfs_log_reserve(tp->t_mountp, logspace, logcount, | |
367 | &tp->t_ticket, | |
7e9c6396 | 368 | XFS_TRANSACTION, log_flags, tp->t_type); |
1da177e4 LT |
369 | if (error) { |
370 | goto undo_blocks; | |
371 | } | |
372 | tp->t_log_res = logspace; | |
373 | tp->t_log_count = logcount; | |
374 | } | |
375 | ||
376 | /* | |
377 | * Attempt to reserve the needed realtime extents by decrementing | |
378 | * the number needed from the number available. This will | |
379 | * fail if the count would go below zero. | |
380 | */ | |
381 | if (rtextents > 0) { | |
382 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 383 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
384 | if (error) { |
385 | error = XFS_ERROR(ENOSPC); | |
386 | goto undo_log; | |
387 | } | |
388 | tp->t_rtx_res += rtextents; | |
389 | } | |
390 | ||
391 | return 0; | |
392 | ||
393 | /* | |
394 | * Error cases jump to one of these labels to undo any | |
395 | * reservations which have already been performed. | |
396 | */ | |
397 | undo_log: | |
398 | if (logspace > 0) { | |
399 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
400 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
401 | } else { | |
402 | log_flags = 0; | |
403 | } | |
404 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
405 | tp->t_ticket = NULL; | |
406 | tp->t_log_res = 0; | |
407 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
408 | } | |
409 | ||
410 | undo_blocks: | |
411 | if (blocks > 0) { | |
412 | (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 413 | (int64_t)blocks, rsvd); |
1da177e4 LT |
414 | tp->t_blk_res = 0; |
415 | } | |
416 | ||
59c1b082 | 417 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 418 | |
59c1b082 | 419 | return error; |
1da177e4 LT |
420 | } |
421 | ||
422 | ||
1da177e4 LT |
423 | /* |
424 | * Record the indicated change to the given field for application | |
425 | * to the file system's superblock when the transaction commits. | |
426 | * For now, just store the change in the transaction structure. | |
427 | * | |
428 | * Mark the transaction structure to indicate that the superblock | |
429 | * needs to be updated before committing. | |
92821e2b DC |
430 | * |
431 | * Because we may not be keeping track of allocated/free inodes and | |
432 | * used filesystem blocks in the superblock, we do not mark the | |
433 | * superblock dirty in this transaction if we modify these fields. | |
434 | * We still need to update the transaction deltas so that they get | |
435 | * applied to the incore superblock, but we don't want them to | |
436 | * cause the superblock to get locked and logged if these are the | |
437 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
438 | */ |
439 | void | |
440 | xfs_trans_mod_sb( | |
441 | xfs_trans_t *tp, | |
442 | uint field, | |
20f4ebf2 | 443 | int64_t delta) |
1da177e4 | 444 | { |
92821e2b DC |
445 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
446 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
447 | |
448 | switch (field) { | |
449 | case XFS_TRANS_SB_ICOUNT: | |
450 | tp->t_icount_delta += delta; | |
92821e2b DC |
451 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
452 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
453 | break; |
454 | case XFS_TRANS_SB_IFREE: | |
455 | tp->t_ifree_delta += delta; | |
92821e2b DC |
456 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
457 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
458 | break; |
459 | case XFS_TRANS_SB_FDBLOCKS: | |
460 | /* | |
461 | * Track the number of blocks allocated in the | |
462 | * transaction. Make sure it does not exceed the | |
463 | * number reserved. | |
464 | */ | |
465 | if (delta < 0) { | |
466 | tp->t_blk_res_used += (uint)-delta; | |
467 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
468 | } | |
469 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
470 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
471 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
472 | break; |
473 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
474 | /* | |
475 | * The allocation has already been applied to the | |
476 | * in-core superblock's counter. This should only | |
477 | * be applied to the on-disk superblock. | |
478 | */ | |
479 | ASSERT(delta < 0); | |
480 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
481 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
482 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
483 | break; |
484 | case XFS_TRANS_SB_FREXTENTS: | |
485 | /* | |
486 | * Track the number of blocks allocated in the | |
487 | * transaction. Make sure it does not exceed the | |
488 | * number reserved. | |
489 | */ | |
490 | if (delta < 0) { | |
491 | tp->t_rtx_res_used += (uint)-delta; | |
492 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
493 | } | |
494 | tp->t_frextents_delta += delta; | |
495 | break; | |
496 | case XFS_TRANS_SB_RES_FREXTENTS: | |
497 | /* | |
498 | * The allocation has already been applied to the | |
c41564b5 | 499 | * in-core superblock's counter. This should only |
1da177e4 LT |
500 | * be applied to the on-disk superblock. |
501 | */ | |
502 | ASSERT(delta < 0); | |
503 | tp->t_res_frextents_delta += delta; | |
504 | break; | |
505 | case XFS_TRANS_SB_DBLOCKS: | |
506 | ASSERT(delta > 0); | |
507 | tp->t_dblocks_delta += delta; | |
508 | break; | |
509 | case XFS_TRANS_SB_AGCOUNT: | |
510 | ASSERT(delta > 0); | |
511 | tp->t_agcount_delta += delta; | |
512 | break; | |
513 | case XFS_TRANS_SB_IMAXPCT: | |
514 | tp->t_imaxpct_delta += delta; | |
515 | break; | |
516 | case XFS_TRANS_SB_REXTSIZE: | |
517 | tp->t_rextsize_delta += delta; | |
518 | break; | |
519 | case XFS_TRANS_SB_RBMBLOCKS: | |
520 | tp->t_rbmblocks_delta += delta; | |
521 | break; | |
522 | case XFS_TRANS_SB_RBLOCKS: | |
523 | tp->t_rblocks_delta += delta; | |
524 | break; | |
525 | case XFS_TRANS_SB_REXTENTS: | |
526 | tp->t_rextents_delta += delta; | |
527 | break; | |
528 | case XFS_TRANS_SB_REXTSLOG: | |
529 | tp->t_rextslog_delta += delta; | |
530 | break; | |
531 | default: | |
532 | ASSERT(0); | |
533 | return; | |
534 | } | |
535 | ||
210c6f1c | 536 | tp->t_flags |= flags; |
1da177e4 LT |
537 | } |
538 | ||
539 | /* | |
540 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
541 | * to bring the superblock buffer into the current transaction | |
542 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
543 | * | |
544 | * For now we just look at each field allowed to change and change | |
545 | * it if necessary. | |
546 | */ | |
547 | STATIC void | |
548 | xfs_trans_apply_sb_deltas( | |
549 | xfs_trans_t *tp) | |
550 | { | |
2bdf7cd0 | 551 | xfs_dsb_t *sbp; |
1da177e4 LT |
552 | xfs_buf_t *bp; |
553 | int whole = 0; | |
554 | ||
555 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
556 | sbp = XFS_BUF_TO_SBP(bp); | |
557 | ||
558 | /* | |
559 | * Check that superblock mods match the mods made to AGF counters. | |
560 | */ | |
561 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
562 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
563 | tp->t_ag_btree_delta)); | |
564 | ||
92821e2b DC |
565 | /* |
566 | * Only update the superblock counters if we are logging them | |
567 | */ | |
568 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 569 | if (tp->t_icount_delta) |
413d57c9 | 570 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 571 | if (tp->t_ifree_delta) |
413d57c9 | 572 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 573 | if (tp->t_fdblocks_delta) |
413d57c9 | 574 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 575 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 576 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
577 | } |
578 | ||
2bdf7cd0 | 579 | if (tp->t_frextents_delta) |
413d57c9 | 580 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 581 | if (tp->t_res_frextents_delta) |
413d57c9 | 582 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
583 | |
584 | if (tp->t_dblocks_delta) { | |
413d57c9 | 585 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
586 | whole = 1; |
587 | } | |
2bdf7cd0 | 588 | if (tp->t_agcount_delta) { |
413d57c9 | 589 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
590 | whole = 1; |
591 | } | |
2bdf7cd0 CH |
592 | if (tp->t_imaxpct_delta) { |
593 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
594 | whole = 1; |
595 | } | |
2bdf7cd0 | 596 | if (tp->t_rextsize_delta) { |
413d57c9 | 597 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
598 | whole = 1; |
599 | } | |
2bdf7cd0 | 600 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 601 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
602 | whole = 1; |
603 | } | |
2bdf7cd0 | 604 | if (tp->t_rblocks_delta) { |
413d57c9 | 605 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
606 | whole = 1; |
607 | } | |
2bdf7cd0 | 608 | if (tp->t_rextents_delta) { |
413d57c9 | 609 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
610 | whole = 1; |
611 | } | |
2bdf7cd0 CH |
612 | if (tp->t_rextslog_delta) { |
613 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
614 | whole = 1; |
615 | } | |
616 | ||
617 | if (whole) | |
618 | /* | |
c41564b5 | 619 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 620 | */ |
2bdf7cd0 | 621 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
622 | else |
623 | /* | |
624 | * Since all the modifiable fields are contiguous, we | |
625 | * can get away with this. | |
626 | */ | |
2bdf7cd0 CH |
627 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
628 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 629 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
630 | } |
631 | ||
632 | /* | |
45c34141 DC |
633 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
634 | * and apply superblock counter changes to the in-core superblock. The | |
635 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
636 | * applied to the in-core superblock. The idea is that that has already been | |
637 | * done. | |
1da177e4 LT |
638 | * |
639 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
640 | * However, we have to ensure that we only modify each superblock field only |
641 | * once because the application of the delta values may not be atomic. That can | |
642 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
643 | * free space counter to put back the entire reservation and then take away | |
644 | * what we used. | |
645 | * | |
646 | * If we are not logging superblock counters, then the inode allocated/free and | |
647 | * used block counts are not updated in the on disk superblock. In this case, | |
648 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
649 | * still need to update the incore superblock with the changes. | |
1da177e4 | 650 | */ |
ba0f32d4 | 651 | STATIC void |
1da177e4 LT |
652 | xfs_trans_unreserve_and_mod_sb( |
653 | xfs_trans_t *tp) | |
654 | { | |
655 | xfs_mod_sb_t msb[14]; /* If you add cases, add entries */ | |
656 | xfs_mod_sb_t *msbp; | |
92821e2b | 657 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
658 | /* REFERENCED */ |
659 | int error; | |
660 | int rsvd; | |
45c34141 DC |
661 | int64_t blkdelta = 0; |
662 | int64_t rtxdelta = 0; | |
1da177e4 LT |
663 | |
664 | msbp = msb; | |
665 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
666 | ||
45c34141 DC |
667 | /* calculate free blocks delta */ |
668 | if (tp->t_blk_res > 0) | |
669 | blkdelta = tp->t_blk_res; | |
670 | ||
671 | if ((tp->t_fdblocks_delta != 0) && | |
672 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
673 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
674 | blkdelta += tp->t_fdblocks_delta; | |
675 | ||
676 | if (blkdelta != 0) { | |
1da177e4 | 677 | msbp->msb_field = XFS_SBS_FDBLOCKS; |
45c34141 | 678 | msbp->msb_delta = blkdelta; |
1da177e4 LT |
679 | msbp++; |
680 | } | |
681 | ||
45c34141 DC |
682 | /* calculate free realtime extents delta */ |
683 | if (tp->t_rtx_res > 0) | |
684 | rtxdelta = tp->t_rtx_res; | |
685 | ||
686 | if ((tp->t_frextents_delta != 0) && | |
687 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
688 | rtxdelta += tp->t_frextents_delta; | |
689 | ||
690 | if (rtxdelta != 0) { | |
1da177e4 | 691 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 692 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
693 | msbp++; |
694 | } | |
695 | ||
45c34141 DC |
696 | /* apply remaining deltas */ |
697 | ||
92821e2b DC |
698 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
699 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1da177e4 LT |
700 | if (tp->t_icount_delta != 0) { |
701 | msbp->msb_field = XFS_SBS_ICOUNT; | |
20f4ebf2 | 702 | msbp->msb_delta = tp->t_icount_delta; |
1da177e4 LT |
703 | msbp++; |
704 | } | |
705 | if (tp->t_ifree_delta != 0) { | |
706 | msbp->msb_field = XFS_SBS_IFREE; | |
20f4ebf2 | 707 | msbp->msb_delta = tp->t_ifree_delta; |
1da177e4 LT |
708 | msbp++; |
709 | } | |
92821e2b DC |
710 | } |
711 | ||
712 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { | |
1da177e4 LT |
713 | if (tp->t_dblocks_delta != 0) { |
714 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 715 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
716 | msbp++; |
717 | } | |
718 | if (tp->t_agcount_delta != 0) { | |
719 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 720 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
721 | msbp++; |
722 | } | |
723 | if (tp->t_imaxpct_delta != 0) { | |
724 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 725 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
726 | msbp++; |
727 | } | |
728 | if (tp->t_rextsize_delta != 0) { | |
729 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 730 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
731 | msbp++; |
732 | } | |
733 | if (tp->t_rbmblocks_delta != 0) { | |
734 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 735 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
736 | msbp++; |
737 | } | |
738 | if (tp->t_rblocks_delta != 0) { | |
739 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 740 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
741 | msbp++; |
742 | } | |
743 | if (tp->t_rextents_delta != 0) { | |
744 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 745 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
746 | msbp++; |
747 | } | |
748 | if (tp->t_rextslog_delta != 0) { | |
749 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 750 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
751 | msbp++; |
752 | } | |
753 | } | |
754 | ||
755 | /* | |
756 | * If we need to change anything, do it. | |
757 | */ | |
758 | if (msbp > msb) { | |
759 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
760 | (uint)(msbp - msb), rsvd); | |
761 | ASSERT(error == 0); | |
762 | } | |
763 | } | |
764 | ||
1da177e4 | 765 | /* |
0924378a DC |
766 | * Total up the number of log iovecs needed to commit this |
767 | * transaction. The transaction itself needs one for the | |
768 | * transaction header. Ask each dirty item in turn how many | |
769 | * it needs to get the total. | |
1da177e4 | 770 | */ |
0924378a DC |
771 | static uint |
772 | xfs_trans_count_vecs( | |
773 | xfs_trans_t *tp) | |
1da177e4 | 774 | { |
0924378a DC |
775 | int nvecs; |
776 | xfs_log_item_desc_t *lidp; | |
1da177e4 | 777 | |
0924378a DC |
778 | nvecs = 1; |
779 | lidp = xfs_trans_first_item(tp); | |
780 | ASSERT(lidp != NULL); | |
1da177e4 | 781 | |
0924378a DC |
782 | /* In the non-debug case we need to start bailing out if we |
783 | * didn't find a log_item here, return zero and let trans_commit | |
784 | * deal with it. | |
1da177e4 | 785 | */ |
0924378a DC |
786 | if (lidp == NULL) |
787 | return 0; | |
788 | ||
789 | while (lidp != NULL) { | |
790 | /* | |
791 | * Skip items which aren't dirty in this transaction. | |
792 | */ | |
793 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
794 | lidp = xfs_trans_next_item(tp, lidp); | |
795 | continue; | |
796 | } | |
797 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
798 | nvecs += lidp->lid_size; | |
799 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 800 | } |
0924378a DC |
801 | |
802 | return nvecs; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Fill in the vector with pointers to data to be logged | |
807 | * by this transaction. The transaction header takes | |
808 | * the first vector, and then each dirty item takes the | |
809 | * number of vectors it indicated it needed in xfs_trans_count_vecs(). | |
810 | * | |
811 | * As each item fills in the entries it needs, also pin the item | |
812 | * so that it cannot be flushed out until the log write completes. | |
813 | */ | |
814 | static void | |
815 | xfs_trans_fill_vecs( | |
816 | struct xfs_trans *tp, | |
817 | struct xfs_log_iovec *log_vector) | |
818 | { | |
819 | xfs_log_item_desc_t *lidp; | |
820 | struct xfs_log_iovec *vecp; | |
821 | uint nitems; | |
1da177e4 LT |
822 | |
823 | /* | |
0924378a DC |
824 | * Skip over the entry for the transaction header, we'll |
825 | * fill that in at the end. | |
1da177e4 | 826 | */ |
0924378a DC |
827 | vecp = log_vector + 1; |
828 | ||
829 | nitems = 0; | |
830 | lidp = xfs_trans_first_item(tp); | |
831 | ASSERT(lidp); | |
832 | while (lidp) { | |
833 | /* Skip items which aren't dirty in this transaction. */ | |
834 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
835 | lidp = xfs_trans_next_item(tp, lidp); | |
836 | continue; | |
837 | } | |
838 | ||
1da177e4 | 839 | /* |
0924378a DC |
840 | * The item may be marked dirty but not log anything. This can |
841 | * be used to get called when a transaction is committed. | |
1da177e4 | 842 | */ |
0924378a DC |
843 | if (lidp->lid_size) |
844 | nitems++; | |
845 | IOP_FORMAT(lidp->lid_item, vecp); | |
846 | vecp += lidp->lid_size; | |
847 | IOP_PIN(lidp->lid_item); | |
848 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 849 | } |
1da177e4 LT |
850 | |
851 | /* | |
0924378a DC |
852 | * Now that we've counted the number of items in this transaction, fill |
853 | * in the transaction header. Note that the transaction header does not | |
854 | * have a log item. | |
1da177e4 | 855 | */ |
0924378a DC |
856 | tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC; |
857 | tp->t_header.th_type = tp->t_type; | |
858 | tp->t_header.th_num_items = nitems; | |
859 | log_vector->i_addr = (xfs_caddr_t)&tp->t_header; | |
860 | log_vector->i_len = sizeof(xfs_trans_header_t); | |
861 | log_vector->i_type = XLOG_REG_TYPE_TRANSHDR; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Format the transaction direct to the iclog. This isolates the physical | |
866 | * transaction commit operation from the logical operation and hence allows | |
867 | * other methods to be introduced without affecting the existing commit path. | |
868 | */ | |
869 | static int | |
870 | xfs_trans_commit_iclog( | |
871 | struct xfs_mount *mp, | |
872 | struct xfs_trans *tp, | |
873 | xfs_lsn_t *commit_lsn, | |
874 | int flags) | |
875 | { | |
876 | int shutdown; | |
877 | int error; | |
878 | int log_flags = 0; | |
879 | struct xlog_in_core *commit_iclog; | |
880 | #define XFS_TRANS_LOGVEC_COUNT 16 | |
881 | struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT]; | |
882 | struct xfs_log_iovec *log_vector; | |
883 | uint nvec; | |
884 | ||
1da177e4 LT |
885 | |
886 | /* | |
887 | * Ask each log item how many log_vector entries it will | |
888 | * need so we can figure out how many to allocate. | |
889 | * Try to avoid the kmem_alloc() call in the common case | |
890 | * by using a vector from the stack when it fits. | |
891 | */ | |
892 | nvec = xfs_trans_count_vecs(tp); | |
1da177e4 | 893 | if (nvec == 0) { |
0924378a | 894 | return ENOMEM; /* triggers a shutdown! */ |
cfcbbbd0 | 895 | } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) { |
1da177e4 LT |
896 | log_vector = log_vector_fast; |
897 | } else { | |
898 | log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec * | |
899 | sizeof(xfs_log_iovec_t), | |
900 | KM_SLEEP); | |
901 | } | |
902 | ||
903 | /* | |
904 | * Fill in the log_vector and pin the logged items, and | |
905 | * then write the transaction to the log. | |
906 | */ | |
907 | xfs_trans_fill_vecs(tp, log_vector); | |
908 | ||
0924378a DC |
909 | if (flags & XFS_TRANS_RELEASE_LOG_RES) |
910 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
911 | ||
cfcbbbd0 | 912 | error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn)); |
1da177e4 | 913 | |
1da177e4 | 914 | /* |
cfcbbbd0 NS |
915 | * The transaction is committed incore here, and can go out to disk |
916 | * at any time after this call. However, all the items associated | |
917 | * with the transaction are still locked and pinned in memory. | |
1da177e4 | 918 | */ |
0924378a | 919 | *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); |
1da177e4 | 920 | |
0924378a DC |
921 | tp->t_commit_lsn = *commit_lsn; |
922 | if (nvec > XFS_TRANS_LOGVEC_COUNT) | |
f0e2d93c | 923 | kmem_free(log_vector); |
1da177e4 | 924 | |
1da177e4 LT |
925 | /* |
926 | * If we got a log write error. Unpin the logitems that we | |
927 | * had pinned, clean up, free trans structure, and return error. | |
928 | */ | |
0924378a | 929 | if (error || *commit_lsn == -1) { |
59c1b082 | 930 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
931 | xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT); |
932 | return XFS_ERROR(EIO); | |
933 | } | |
934 | ||
935 | /* | |
936 | * Once the transaction has committed, unused | |
937 | * reservations need to be released and changes to | |
938 | * the superblock need to be reflected in the in-core | |
939 | * version. Do that now. | |
940 | */ | |
941 | xfs_trans_unreserve_and_mod_sb(tp); | |
942 | ||
1da177e4 LT |
943 | /* |
944 | * Tell the LM to call the transaction completion routine | |
945 | * when the log write with LSN commit_lsn completes (e.g. | |
946 | * when the transaction commit really hits the on-disk log). | |
947 | * After this call we cannot reference tp, because the call | |
948 | * can happen at any time and the call will free the transaction | |
949 | * structure pointed to by tp. The only case where we call | |
950 | * the completion routine (xfs_trans_committed) directly is | |
951 | * if the log is turned off on a debug kernel or we're | |
952 | * running in simulation mode (the log is explicitly turned | |
953 | * off). | |
954 | */ | |
955 | tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed; | |
956 | tp->t_logcb.cb_arg = tp; | |
957 | ||
958 | /* | |
959 | * We need to pass the iclog buffer which was used for the | |
960 | * transaction commit record into this function, and attach | |
961 | * the callback to it. The callback must be attached before | |
962 | * the items are unlocked to avoid racing with other threads | |
963 | * waiting for an item to unlock. | |
964 | */ | |
965 | shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb)); | |
966 | ||
967 | /* | |
968 | * Mark this thread as no longer being in a transaction | |
969 | */ | |
59c1b082 | 970 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
971 | |
972 | /* | |
973 | * Once all the items of the transaction have been copied | |
974 | * to the in core log and the callback is attached, the | |
975 | * items can be unlocked. | |
976 | * | |
977 | * This will free descriptors pointing to items which were | |
978 | * not logged since there is nothing more to do with them. | |
979 | * For items which were logged, we will keep pointers to them | |
980 | * so they can be unpinned after the transaction commits to disk. | |
981 | * This will also stamp each modified meta-data item with | |
982 | * the commit lsn of this transaction for dependency tracking | |
983 | * purposes. | |
984 | */ | |
0924378a | 985 | xfs_trans_unlock_items(tp, *commit_lsn); |
1da177e4 LT |
986 | |
987 | /* | |
988 | * If we detected a log error earlier, finish committing | |
989 | * the transaction now (unpin log items, etc). | |
990 | * | |
991 | * Order is critical here, to avoid using the transaction | |
992 | * pointer after its been freed (by xfs_trans_committed | |
993 | * either here now, or as a callback). We cannot do this | |
994 | * step inside xfs_log_notify as was done earlier because | |
995 | * of this issue. | |
996 | */ | |
997 | if (shutdown) | |
998 | xfs_trans_committed(tp, XFS_LI_ABORTED); | |
999 | ||
1000 | /* | |
1001 | * Now that the xfs_trans_committed callback has been attached, | |
1002 | * and the items are released we can finally allow the iclog to | |
1003 | * go to disk. | |
1004 | */ | |
0924378a DC |
1005 | return xfs_log_release_iclog(mp, commit_iclog); |
1006 | } | |
1007 | ||
1008 | ||
1009 | /* | |
1010 | * xfs_trans_commit | |
1011 | * | |
1012 | * Commit the given transaction to the log a/synchronously. | |
1013 | * | |
1014 | * XFS disk error handling mechanism is not based on a typical | |
1015 | * transaction abort mechanism. Logically after the filesystem | |
1016 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1017 | * be durable - ie. committed to disk - because some metadata might | |
1018 | * be inconsistent. In such cases, this returns an error, and the | |
1019 | * caller may assume that all locked objects joined to the transaction | |
1020 | * have already been unlocked as if the commit had succeeded. | |
1021 | * Do not reference the transaction structure after this call. | |
1022 | */ | |
1023 | /*ARGSUSED*/ | |
1024 | int | |
1025 | _xfs_trans_commit( | |
1026 | xfs_trans_t *tp, | |
1027 | uint flags, | |
1028 | int *log_flushed) | |
1029 | { | |
1030 | xfs_mount_t *mp = tp->t_mountp; | |
1031 | xfs_lsn_t commit_lsn = -1; | |
1032 | int error; | |
1033 | int log_flags = 0; | |
1034 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
1035 | int shutdown; | |
1036 | ||
1037 | /* | |
1038 | * Determine whether this commit is releasing a permanent | |
1039 | * log reservation or not. | |
1040 | */ | |
1041 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1042 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1043 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * If there is nothing to be logged by the transaction, | |
1048 | * then unlock all of the items associated with the | |
1049 | * transaction and free the transaction structure. | |
1050 | * Also make sure to return any reserved blocks to | |
1051 | * the free pool. | |
1052 | */ | |
1053 | shut_us_down: | |
1054 | shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0; | |
1055 | if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) { | |
1056 | xfs_trans_unreserve_and_mod_sb(tp); | |
1057 | /* | |
1058 | * It is indeed possible for the transaction to be | |
1059 | * not dirty but the dqinfo portion to be. All that | |
1060 | * means is that we have some (non-persistent) quota | |
1061 | * reservations that need to be unreserved. | |
1062 | */ | |
1063 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1064 | if (tp->t_ticket) { | |
1065 | commit_lsn = xfs_log_done(mp, tp->t_ticket, | |
1066 | NULL, log_flags); | |
1067 | if (commit_lsn == -1 && !shutdown) | |
1068 | shutdown = XFS_ERROR(EIO); | |
1069 | } | |
1070 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
1071 | xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0); | |
1072 | xfs_trans_free_busy(tp); | |
1073 | xfs_trans_free(tp); | |
1074 | XFS_STATS_INC(xs_trans_empty); | |
1075 | return (shutdown); | |
1076 | } | |
1077 | ASSERT(tp->t_ticket != NULL); | |
1078 | ||
1079 | /* | |
1080 | * If we need to update the superblock, then do it now. | |
1081 | */ | |
1082 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1083 | xfs_trans_apply_sb_deltas(tp); | |
1084 | xfs_trans_apply_dquot_deltas(tp); | |
1085 | ||
1086 | error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags); | |
1087 | if (error == ENOMEM) { | |
1088 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
1089 | goto shut_us_down; | |
1090 | } | |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * If the transaction needs to be synchronous, then force the | |
1094 | * log out now and wait for it. | |
1095 | */ | |
1096 | if (sync) { | |
f538d4da | 1097 | if (!error) { |
a14a348b CH |
1098 | error = _xfs_log_force_lsn(mp, commit_lsn, |
1099 | XFS_LOG_SYNC, log_flushed); | |
f538d4da | 1100 | } |
1da177e4 LT |
1101 | XFS_STATS_INC(xs_trans_sync); |
1102 | } else { | |
1103 | XFS_STATS_INC(xs_trans_async); | |
1104 | } | |
1105 | ||
1106 | return (error); | |
1107 | } | |
1108 | ||
1da177e4 LT |
1109 | /* |
1110 | * Called from the trans_commit code when we notice that | |
1111 | * the filesystem is in the middle of a forced shutdown. | |
1112 | */ | |
1113 | STATIC void | |
1114 | xfs_trans_uncommit( | |
1115 | xfs_trans_t *tp, | |
1116 | uint flags) | |
1117 | { | |
1118 | xfs_log_item_desc_t *lidp; | |
1119 | ||
1120 | for (lidp = xfs_trans_first_item(tp); | |
1121 | lidp != NULL; | |
1122 | lidp = xfs_trans_next_item(tp, lidp)) { | |
1123 | /* | |
1124 | * Unpin all but those that aren't dirty. | |
1125 | */ | |
1126 | if (lidp->lid_flags & XFS_LID_DIRTY) | |
1127 | IOP_UNPIN_REMOVE(lidp->lid_item, tp); | |
1128 | } | |
1129 | ||
1130 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1131 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1132 | |
1133 | xfs_trans_free_items(tp, flags); | |
1134 | xfs_trans_free_busy(tp); | |
1135 | xfs_trans_free(tp); | |
1136 | } | |
1137 | ||
1da177e4 LT |
1138 | /* |
1139 | * Unlock all of the transaction's items and free the transaction. | |
1140 | * The transaction must not have modified any of its items, because | |
1141 | * there is no way to restore them to their previous state. | |
1142 | * | |
1143 | * If the transaction has made a log reservation, make sure to release | |
1144 | * it as well. | |
1145 | */ | |
1146 | void | |
1147 | xfs_trans_cancel( | |
1148 | xfs_trans_t *tp, | |
1149 | int flags) | |
1150 | { | |
1151 | int log_flags; | |
1152 | #ifdef DEBUG | |
1153 | xfs_log_item_chunk_t *licp; | |
1154 | xfs_log_item_desc_t *lidp; | |
1155 | xfs_log_item_t *lip; | |
1156 | int i; | |
1157 | #endif | |
0733af21 | 1158 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1159 | |
1160 | /* | |
1161 | * See if the caller is being too lazy to figure out if | |
1162 | * the transaction really needs an abort. | |
1163 | */ | |
1164 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1165 | flags &= ~XFS_TRANS_ABORT; | |
1166 | /* | |
1167 | * See if the caller is relying on us to shut down the | |
1168 | * filesystem. This happens in paths where we detect | |
1169 | * corruption and decide to give up. | |
1170 | */ | |
60a204f0 | 1171 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1172 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1173 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1174 | } |
1da177e4 LT |
1175 | #ifdef DEBUG |
1176 | if (!(flags & XFS_TRANS_ABORT)) { | |
1177 | licp = &(tp->t_items); | |
1178 | while (licp != NULL) { | |
1179 | lidp = licp->lic_descs; | |
1180 | for (i = 0; i < licp->lic_unused; i++, lidp++) { | |
39dab9d7 | 1181 | if (xfs_lic_isfree(licp, i)) { |
1da177e4 LT |
1182 | continue; |
1183 | } | |
1184 | ||
1185 | lip = lidp->lid_item; | |
0733af21 | 1186 | if (!XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 LT |
1187 | ASSERT(!(lip->li_type == XFS_LI_EFD)); |
1188 | } | |
1189 | licp = licp->lic_next; | |
1190 | } | |
1191 | } | |
1192 | #endif | |
1193 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1194 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1195 | |
1196 | if (tp->t_ticket) { | |
1197 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1198 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1199 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1200 | } else { | |
1201 | log_flags = 0; | |
1202 | } | |
0733af21 | 1203 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1207 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1208 | |
1209 | xfs_trans_free_items(tp, flags); | |
1210 | xfs_trans_free_busy(tp); | |
1211 | xfs_trans_free(tp); | |
1212 | } | |
1213 | ||
1214 | ||
1215 | /* | |
1216 | * Free the transaction structure. If there is more clean up | |
1217 | * to do when the structure is freed, add it here. | |
1218 | */ | |
1219 | STATIC void | |
1220 | xfs_trans_free( | |
1221 | xfs_trans_t *tp) | |
1222 | { | |
1223 | atomic_dec(&tp->t_mountp->m_active_trans); | |
7d095257 | 1224 | xfs_trans_free_dqinfo(tp); |
1da177e4 LT |
1225 | kmem_zone_free(xfs_trans_zone, tp); |
1226 | } | |
1227 | ||
322ff6b8 NS |
1228 | /* |
1229 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1230 | * the next: permanent transactions are only flushed out when | |
1231 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1232 | * as possible to let chunks of it go to the log. So we commit the | |
1233 | * chunk we've been working on and get a new transaction to continue. | |
1234 | */ | |
1235 | int | |
1236 | xfs_trans_roll( | |
1237 | struct xfs_trans **tpp, | |
1238 | struct xfs_inode *dp) | |
1239 | { | |
1240 | struct xfs_trans *trans; | |
1241 | unsigned int logres, count; | |
1242 | int error; | |
1243 | ||
1244 | /* | |
1245 | * Ensure that the inode is always logged. | |
1246 | */ | |
1247 | trans = *tpp; | |
1248 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1249 | ||
1250 | /* | |
1251 | * Copy the critical parameters from one trans to the next. | |
1252 | */ | |
1253 | logres = trans->t_log_res; | |
1254 | count = trans->t_log_count; | |
1255 | *tpp = xfs_trans_dup(trans); | |
1256 | ||
1257 | /* | |
1258 | * Commit the current transaction. | |
1259 | * If this commit failed, then it'd just unlock those items that | |
1260 | * are not marked ihold. That also means that a filesystem shutdown | |
1261 | * is in progress. The caller takes the responsibility to cancel | |
1262 | * the duplicate transaction that gets returned. | |
1263 | */ | |
1264 | error = xfs_trans_commit(trans, 0); | |
1265 | if (error) | |
1266 | return (error); | |
1267 | ||
1268 | trans = *tpp; | |
1269 | ||
cc09c0dc DC |
1270 | /* |
1271 | * transaction commit worked ok so we can drop the extra ticket | |
1272 | * reference that we gained in xfs_trans_dup() | |
1273 | */ | |
1274 | xfs_log_ticket_put(trans->t_ticket); | |
1275 | ||
1276 | ||
322ff6b8 NS |
1277 | /* |
1278 | * Reserve space in the log for th next transaction. | |
1279 | * This also pushes items in the "AIL", the list of logged items, | |
1280 | * out to disk if they are taking up space at the tail of the log | |
1281 | * that we want to use. This requires that either nothing be locked | |
1282 | * across this call, or that anything that is locked be logged in | |
1283 | * the prior and the next transactions. | |
1284 | */ | |
1285 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1286 | XFS_TRANS_PERM_LOG_RES, count); | |
1287 | /* | |
1288 | * Ensure that the inode is in the new transaction and locked. | |
1289 | */ | |
1290 | if (error) | |
1291 | return error; | |
1292 | ||
1293 | xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL); | |
1294 | xfs_trans_ihold(trans, dp); | |
1295 | return 0; | |
1296 | } | |
1da177e4 LT |
1297 | |
1298 | /* | |
1299 | * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item(). | |
1300 | * | |
1301 | * This is typically called by the LM when a transaction has been fully | |
1302 | * committed to disk. It needs to unpin the items which have | |
1303 | * been logged by the transaction and update their positions | |
1304 | * in the AIL if necessary. | |
1305 | * This also gets called when the transactions didn't get written out | |
1306 | * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then. | |
1307 | * | |
1308 | * Call xfs_trans_chunk_committed() to process the items in | |
1309 | * each chunk. | |
1310 | */ | |
1311 | STATIC void | |
1312 | xfs_trans_committed( | |
1313 | xfs_trans_t *tp, | |
1314 | int abortflag) | |
1315 | { | |
1316 | xfs_log_item_chunk_t *licp; | |
1317 | xfs_log_item_chunk_t *next_licp; | |
1318 | xfs_log_busy_chunk_t *lbcp; | |
1319 | xfs_log_busy_slot_t *lbsp; | |
1320 | int i; | |
1321 | ||
1322 | /* | |
1323 | * Call the transaction's completion callback if there | |
1324 | * is one. | |
1325 | */ | |
1326 | if (tp->t_callback != NULL) { | |
1327 | tp->t_callback(tp, tp->t_callarg); | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Special case the chunk embedded in the transaction. | |
1332 | */ | |
1333 | licp = &(tp->t_items); | |
39dab9d7 | 1334 | if (!(xfs_lic_are_all_free(licp))) { |
1da177e4 LT |
1335 | xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag); |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Process the items in each chunk in turn. | |
1340 | */ | |
1341 | licp = licp->lic_next; | |
1342 | while (licp != NULL) { | |
39dab9d7 | 1343 | ASSERT(!xfs_lic_are_all_free(licp)); |
1da177e4 LT |
1344 | xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag); |
1345 | next_licp = licp->lic_next; | |
f0e2d93c | 1346 | kmem_free(licp); |
1da177e4 LT |
1347 | licp = next_licp; |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * Clear all the per-AG busy list items listed in this transaction | |
1352 | */ | |
1353 | lbcp = &tp->t_busy; | |
1354 | while (lbcp != NULL) { | |
1355 | for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) { | |
1356 | if (!XFS_LBC_ISFREE(lbcp, i)) { | |
1357 | xfs_alloc_clear_busy(tp, lbsp->lbc_ag, | |
1358 | lbsp->lbc_idx); | |
1359 | } | |
1360 | } | |
1361 | lbcp = lbcp->lbc_next; | |
1362 | } | |
1363 | xfs_trans_free_busy(tp); | |
1364 | ||
1365 | /* | |
1366 | * That's it for the transaction structure. Free it. | |
1367 | */ | |
1368 | xfs_trans_free(tp); | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * This is called to perform the commit processing for each | |
1373 | * item described by the given chunk. | |
1374 | * | |
1375 | * The commit processing consists of unlocking items which were | |
1376 | * held locked with the SYNC_UNLOCK attribute, calling the committed | |
1377 | * routine of each logged item, updating the item's position in the AIL | |
1378 | * if necessary, and unpinning each item. If the committed routine | |
1379 | * returns -1, then do nothing further with the item because it | |
1380 | * may have been freed. | |
1381 | * | |
1382 | * Since items are unlocked when they are copied to the incore | |
1383 | * log, it is possible for two transactions to be completing | |
1384 | * and manipulating the same item simultaneously. The AIL lock | |
1385 | * will protect the lsn field of each item. The value of this | |
1386 | * field can never go backwards. | |
1387 | * | |
1388 | * We unpin the items after repositioning them in the AIL, because | |
1389 | * otherwise they could be immediately flushed and we'd have to race | |
1390 | * with the flusher trying to pull the item from the AIL as we add it. | |
1391 | */ | |
1392 | STATIC void | |
1393 | xfs_trans_chunk_committed( | |
1394 | xfs_log_item_chunk_t *licp, | |
1395 | xfs_lsn_t lsn, | |
1396 | int aborted) | |
1397 | { | |
1398 | xfs_log_item_desc_t *lidp; | |
1399 | xfs_log_item_t *lip; | |
1400 | xfs_lsn_t item_lsn; | |
1da177e4 | 1401 | int i; |
1da177e4 LT |
1402 | |
1403 | lidp = licp->lic_descs; | |
1404 | for (i = 0; i < licp->lic_unused; i++, lidp++) { | |
fc1829f3 DC |
1405 | struct xfs_ail *ailp; |
1406 | ||
39dab9d7 | 1407 | if (xfs_lic_isfree(licp, i)) { |
1da177e4 LT |
1408 | continue; |
1409 | } | |
1410 | ||
1411 | lip = lidp->lid_item; | |
1412 | if (aborted) | |
1413 | lip->li_flags |= XFS_LI_ABORTED; | |
1414 | ||
1415 | /* | |
1416 | * Send in the ABORTED flag to the COMMITTED routine | |
1417 | * so that it knows whether the transaction was aborted | |
1418 | * or not. | |
1419 | */ | |
1420 | item_lsn = IOP_COMMITTED(lip, lsn); | |
1421 | ||
1422 | /* | |
1423 | * If the committed routine returns -1, make | |
1424 | * no more references to the item. | |
1425 | */ | |
1426 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) { | |
1427 | continue; | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * If the returned lsn is greater than what it | |
1432 | * contained before, update the location of the | |
1433 | * item in the AIL. If it is not, then do nothing. | |
1434 | * Items can never move backwards in the AIL. | |
1435 | * | |
1436 | * While the new lsn should usually be greater, it | |
1437 | * is possible that a later transaction completing | |
1438 | * simultaneously with an earlier one using the | |
1439 | * same item could complete first with a higher lsn. | |
1440 | * This would cause the earlier transaction to fail | |
1441 | * the test below. | |
1442 | */ | |
fc1829f3 DC |
1443 | ailp = lip->li_ailp; |
1444 | spin_lock(&ailp->xa_lock); | |
1da177e4 LT |
1445 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) { |
1446 | /* | |
1447 | * This will set the item's lsn to item_lsn | |
1448 | * and update the position of the item in | |
1449 | * the AIL. | |
1450 | * | |
783a2f65 | 1451 | * xfs_trans_ail_update() drops the AIL lock. |
1da177e4 | 1452 | */ |
783a2f65 | 1453 | xfs_trans_ail_update(ailp, lip, item_lsn); |
1da177e4 | 1454 | } else { |
fc1829f3 | 1455 | spin_unlock(&ailp->xa_lock); |
1da177e4 LT |
1456 | } |
1457 | ||
1458 | /* | |
1459 | * Now that we've repositioned the item in the AIL, | |
1460 | * unpin it so it can be flushed. Pass information | |
1461 | * about buffer stale state down from the log item | |
1462 | * flags, if anyone else stales the buffer we do not | |
1463 | * want to pay any attention to it. | |
1464 | */ | |
8e123850 | 1465 | IOP_UNPIN(lip); |
1da177e4 LT |
1466 | } |
1467 | } |