]>
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 | 47 | |
8f794055 | 48 | kmem_zone_t *xfs_trans_zone; |
1da177e4 | 49 | |
8f794055 NS |
50 | /* |
51 | * Reservation functions here avoid a huge stack in xfs_trans_init | |
52 | * due to register overflow from temporaries in the calculations. | |
53 | */ | |
8f794055 NS |
54 | STATIC uint |
55 | xfs_calc_write_reservation(xfs_mount_t *mp) | |
56 | { | |
57 | return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
58 | } | |
59 | ||
60 | STATIC uint | |
61 | xfs_calc_itruncate_reservation(xfs_mount_t *mp) | |
62 | { | |
63 | return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
64 | } | |
65 | ||
66 | STATIC uint | |
67 | xfs_calc_rename_reservation(xfs_mount_t *mp) | |
68 | { | |
69 | return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
70 | } | |
71 | ||
72 | STATIC uint | |
73 | xfs_calc_link_reservation(xfs_mount_t *mp) | |
74 | { | |
2ddd5928 | 75 | return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); |
8f794055 NS |
76 | } |
77 | ||
78 | STATIC uint | |
79 | xfs_calc_remove_reservation(xfs_mount_t *mp) | |
80 | { | |
81 | return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
82 | } | |
83 | ||
84 | STATIC uint | |
85 | xfs_calc_symlink_reservation(xfs_mount_t *mp) | |
86 | { | |
87 | return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
88 | } | |
89 | ||
90 | STATIC uint | |
91 | xfs_calc_create_reservation(xfs_mount_t *mp) | |
92 | { | |
93 | return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
94 | } | |
95 | ||
96 | STATIC uint | |
97 | xfs_calc_mkdir_reservation(xfs_mount_t *mp) | |
98 | { | |
99 | return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
100 | } | |
101 | ||
102 | STATIC uint | |
103 | xfs_calc_ifree_reservation(xfs_mount_t *mp) | |
104 | { | |
105 | return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
106 | } | |
107 | ||
108 | STATIC uint | |
109 | xfs_calc_ichange_reservation(xfs_mount_t *mp) | |
110 | { | |
111 | return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
112 | } | |
113 | ||
114 | STATIC uint | |
115 | xfs_calc_growdata_reservation(xfs_mount_t *mp) | |
116 | { | |
117 | return XFS_CALC_GROWDATA_LOG_RES(mp); | |
118 | } | |
119 | ||
120 | STATIC uint | |
121 | xfs_calc_growrtalloc_reservation(xfs_mount_t *mp) | |
122 | { | |
123 | return XFS_CALC_GROWRTALLOC_LOG_RES(mp); | |
124 | } | |
125 | ||
126 | STATIC uint | |
127 | xfs_calc_growrtzero_reservation(xfs_mount_t *mp) | |
128 | { | |
129 | return XFS_CALC_GROWRTZERO_LOG_RES(mp); | |
130 | } | |
131 | ||
132 | STATIC uint | |
133 | xfs_calc_growrtfree_reservation(xfs_mount_t *mp) | |
134 | { | |
135 | return XFS_CALC_GROWRTFREE_LOG_RES(mp); | |
136 | } | |
137 | ||
138 | STATIC uint | |
139 | xfs_calc_swrite_reservation(xfs_mount_t *mp) | |
140 | { | |
141 | return XFS_CALC_SWRITE_LOG_RES(mp); | |
142 | } | |
143 | ||
144 | STATIC uint | |
145 | xfs_calc_writeid_reservation(xfs_mount_t *mp) | |
146 | { | |
147 | return XFS_CALC_WRITEID_LOG_RES(mp); | |
148 | } | |
149 | ||
150 | STATIC uint | |
151 | xfs_calc_addafork_reservation(xfs_mount_t *mp) | |
152 | { | |
153 | return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
154 | } | |
155 | ||
156 | STATIC uint | |
157 | xfs_calc_attrinval_reservation(xfs_mount_t *mp) | |
158 | { | |
159 | return XFS_CALC_ATTRINVAL_LOG_RES(mp); | |
160 | } | |
161 | ||
162 | STATIC uint | |
163 | xfs_calc_attrset_reservation(xfs_mount_t *mp) | |
164 | { | |
165 | return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
166 | } | |
167 | ||
168 | STATIC uint | |
169 | xfs_calc_attrrm_reservation(xfs_mount_t *mp) | |
170 | { | |
171 | return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); | |
172 | } | |
173 | ||
174 | STATIC uint | |
175 | xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp) | |
176 | { | |
177 | return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp); | |
178 | } | |
179 | ||
1da177e4 LT |
180 | /* |
181 | * Initialize the precomputed transaction reservation values | |
182 | * in the mount structure. | |
183 | */ | |
184 | void | |
185 | xfs_trans_init( | |
186 | xfs_mount_t *mp) | |
187 | { | |
188 | xfs_trans_reservations_t *resp; | |
189 | ||
190 | resp = &(mp->m_reservations); | |
8f794055 NS |
191 | resp->tr_write = xfs_calc_write_reservation(mp); |
192 | resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); | |
193 | resp->tr_rename = xfs_calc_rename_reservation(mp); | |
194 | resp->tr_link = xfs_calc_link_reservation(mp); | |
195 | resp->tr_remove = xfs_calc_remove_reservation(mp); | |
196 | resp->tr_symlink = xfs_calc_symlink_reservation(mp); | |
197 | resp->tr_create = xfs_calc_create_reservation(mp); | |
198 | resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); | |
199 | resp->tr_ifree = xfs_calc_ifree_reservation(mp); | |
200 | resp->tr_ichange = xfs_calc_ichange_reservation(mp); | |
201 | resp->tr_growdata = xfs_calc_growdata_reservation(mp); | |
202 | resp->tr_swrite = xfs_calc_swrite_reservation(mp); | |
203 | resp->tr_writeid = xfs_calc_writeid_reservation(mp); | |
204 | resp->tr_addafork = xfs_calc_addafork_reservation(mp); | |
205 | resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); | |
206 | resp->tr_attrset = xfs_calc_attrset_reservation(mp); | |
207 | resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); | |
208 | resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); | |
209 | resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); | |
210 | resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); | |
211 | resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); | |
1da177e4 LT |
212 | } |
213 | ||
214 | /* | |
215 | * This routine is called to allocate a transaction structure. | |
216 | * The type parameter indicates the type of the transaction. These | |
217 | * are enumerated in xfs_trans.h. | |
218 | * | |
219 | * Dynamically allocate the transaction structure from the transaction | |
220 | * zone, initialize it, and return it to the caller. | |
221 | */ | |
222 | xfs_trans_t * | |
223 | xfs_trans_alloc( | |
224 | xfs_mount_t *mp, | |
225 | uint type) | |
226 | { | |
b267ce99 | 227 | xfs_wait_for_freeze(mp, SB_FREEZE_TRANS); |
80641dc6 | 228 | return _xfs_trans_alloc(mp, type, KM_SLEEP); |
1da177e4 LT |
229 | } |
230 | ||
231 | xfs_trans_t * | |
232 | _xfs_trans_alloc( | |
233 | xfs_mount_t *mp, | |
80641dc6 CH |
234 | uint type, |
235 | uint memflags) | |
1da177e4 LT |
236 | { |
237 | xfs_trans_t *tp; | |
238 | ||
34327e13 | 239 | atomic_inc(&mp->m_active_trans); |
1da177e4 | 240 | |
80641dc6 | 241 | tp = kmem_zone_zalloc(xfs_trans_zone, memflags); |
1da177e4 LT |
242 | tp->t_magic = XFS_TRANS_MAGIC; |
243 | tp->t_type = type; | |
244 | tp->t_mountp = mp; | |
245 | tp->t_items_free = XFS_LIC_NUM_SLOTS; | |
246 | tp->t_busy_free = XFS_LBC_NUM_SLOTS; | |
39dab9d7 | 247 | xfs_lic_init(&(tp->t_items)); |
1da177e4 | 248 | XFS_LBC_INIT(&(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( | |
258 | xfs_trans_t *tp) | |
259 | { | |
260 | atomic_dec(&tp->t_mountp->m_active_trans); | |
261 | xfs_trans_free_dqinfo(tp); | |
262 | kmem_zone_free(xfs_trans_zone, tp); | |
263 | } | |
264 | ||
1da177e4 LT |
265 | /* |
266 | * This is called to create a new transaction which will share the | |
267 | * permanent log reservation of the given transaction. The remaining | |
268 | * unused block and rt extent reservations are also inherited. This | |
269 | * implies that the original transaction is no longer allowed to allocate | |
270 | * blocks. Locks and log items, however, are no inherited. They must | |
271 | * be added to the new transaction explicitly. | |
272 | */ | |
273 | xfs_trans_t * | |
274 | xfs_trans_dup( | |
275 | xfs_trans_t *tp) | |
276 | { | |
277 | xfs_trans_t *ntp; | |
278 | ||
279 | ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP); | |
280 | ||
281 | /* | |
282 | * Initialize the new transaction structure. | |
283 | */ | |
284 | ntp->t_magic = XFS_TRANS_MAGIC; | |
285 | ntp->t_type = tp->t_type; | |
286 | ntp->t_mountp = tp->t_mountp; | |
287 | ntp->t_items_free = XFS_LIC_NUM_SLOTS; | |
288 | ntp->t_busy_free = XFS_LBC_NUM_SLOTS; | |
39dab9d7 | 289 | xfs_lic_init(&(ntp->t_items)); |
1da177e4 LT |
290 | XFS_LBC_INIT(&(ntp->t_busy)); |
291 | ||
292 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1da177e4 | 293 | ASSERT(tp->t_ticket != NULL); |
cfcbbbd0 | 294 | |
1da177e4 | 295 | ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE); |
cc09c0dc | 296 | ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); |
1da177e4 LT |
297 | ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; |
298 | tp->t_blk_res = tp->t_blk_res_used; | |
299 | ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used; | |
300 | tp->t_rtx_res = tp->t_rtx_res_used; | |
59c1b082 | 301 | ntp->t_pflags = tp->t_pflags; |
1da177e4 | 302 | |
7d095257 | 303 | xfs_trans_dup_dqinfo(tp, ntp); |
1da177e4 LT |
304 | |
305 | atomic_inc(&tp->t_mountp->m_active_trans); | |
306 | return ntp; | |
307 | } | |
308 | ||
309 | /* | |
310 | * This is called to reserve free disk blocks and log space for the | |
311 | * given transaction. This must be done before allocating any resources | |
312 | * within the transaction. | |
313 | * | |
314 | * This will return ENOSPC if there are not enough blocks available. | |
315 | * It will sleep waiting for available log space. | |
316 | * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which | |
317 | * is used by long running transactions. If any one of the reservations | |
318 | * fails then they will all be backed out. | |
319 | * | |
320 | * This does not do quota reservations. That typically is done by the | |
321 | * caller afterwards. | |
322 | */ | |
323 | int | |
324 | xfs_trans_reserve( | |
325 | xfs_trans_t *tp, | |
326 | uint blocks, | |
327 | uint logspace, | |
328 | uint rtextents, | |
329 | uint flags, | |
330 | uint logcount) | |
331 | { | |
332 | int log_flags; | |
59c1b082 NS |
333 | int error = 0; |
334 | int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
1da177e4 LT |
335 | |
336 | /* Mark this thread as being in a transaction */ | |
59c1b082 | 337 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
338 | |
339 | /* | |
340 | * Attempt to reserve the needed disk blocks by decrementing | |
341 | * the number needed from the number available. This will | |
342 | * fail if the count would go below zero. | |
343 | */ | |
344 | if (blocks > 0) { | |
345 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 346 | -((int64_t)blocks), rsvd); |
1da177e4 | 347 | if (error != 0) { |
59c1b082 | 348 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
349 | return (XFS_ERROR(ENOSPC)); |
350 | } | |
351 | tp->t_blk_res += blocks; | |
352 | } | |
353 | ||
354 | /* | |
355 | * Reserve the log space needed for this transaction. | |
356 | */ | |
357 | if (logspace > 0) { | |
358 | ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace)); | |
359 | ASSERT((tp->t_log_count == 0) || | |
360 | (tp->t_log_count == logcount)); | |
361 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
362 | log_flags = XFS_LOG_PERM_RESERV; | |
363 | tp->t_flags |= XFS_TRANS_PERM_LOG_RES; | |
364 | } else { | |
365 | ASSERT(tp->t_ticket == NULL); | |
366 | ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); | |
367 | log_flags = 0; | |
368 | } | |
369 | ||
370 | error = xfs_log_reserve(tp->t_mountp, logspace, logcount, | |
371 | &tp->t_ticket, | |
7e9c6396 | 372 | XFS_TRANSACTION, log_flags, tp->t_type); |
1da177e4 LT |
373 | if (error) { |
374 | goto undo_blocks; | |
375 | } | |
376 | tp->t_log_res = logspace; | |
377 | tp->t_log_count = logcount; | |
378 | } | |
379 | ||
380 | /* | |
381 | * Attempt to reserve the needed realtime extents by decrementing | |
382 | * the number needed from the number available. This will | |
383 | * fail if the count would go below zero. | |
384 | */ | |
385 | if (rtextents > 0) { | |
386 | error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS, | |
20f4ebf2 | 387 | -((int64_t)rtextents), rsvd); |
1da177e4 LT |
388 | if (error) { |
389 | error = XFS_ERROR(ENOSPC); | |
390 | goto undo_log; | |
391 | } | |
392 | tp->t_rtx_res += rtextents; | |
393 | } | |
394 | ||
395 | return 0; | |
396 | ||
397 | /* | |
398 | * Error cases jump to one of these labels to undo any | |
399 | * reservations which have already been performed. | |
400 | */ | |
401 | undo_log: | |
402 | if (logspace > 0) { | |
403 | if (flags & XFS_TRANS_PERM_LOG_RES) { | |
404 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
405 | } else { | |
406 | log_flags = 0; | |
407 | } | |
408 | xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags); | |
409 | tp->t_ticket = NULL; | |
410 | tp->t_log_res = 0; | |
411 | tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES; | |
412 | } | |
413 | ||
414 | undo_blocks: | |
415 | if (blocks > 0) { | |
416 | (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, | |
20f4ebf2 | 417 | (int64_t)blocks, rsvd); |
1da177e4 LT |
418 | tp->t_blk_res = 0; |
419 | } | |
420 | ||
59c1b082 | 421 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 | 422 | |
59c1b082 | 423 | return error; |
1da177e4 LT |
424 | } |
425 | ||
426 | ||
1da177e4 LT |
427 | /* |
428 | * Record the indicated change to the given field for application | |
429 | * to the file system's superblock when the transaction commits. | |
430 | * For now, just store the change in the transaction structure. | |
431 | * | |
432 | * Mark the transaction structure to indicate that the superblock | |
433 | * needs to be updated before committing. | |
92821e2b DC |
434 | * |
435 | * Because we may not be keeping track of allocated/free inodes and | |
436 | * used filesystem blocks in the superblock, we do not mark the | |
437 | * superblock dirty in this transaction if we modify these fields. | |
438 | * We still need to update the transaction deltas so that they get | |
439 | * applied to the incore superblock, but we don't want them to | |
440 | * cause the superblock to get locked and logged if these are the | |
441 | * only fields in the superblock that the transaction modifies. | |
1da177e4 LT |
442 | */ |
443 | void | |
444 | xfs_trans_mod_sb( | |
445 | xfs_trans_t *tp, | |
446 | uint field, | |
20f4ebf2 | 447 | int64_t delta) |
1da177e4 | 448 | { |
92821e2b DC |
449 | uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY); |
450 | xfs_mount_t *mp = tp->t_mountp; | |
1da177e4 LT |
451 | |
452 | switch (field) { | |
453 | case XFS_TRANS_SB_ICOUNT: | |
454 | tp->t_icount_delta += delta; | |
92821e2b DC |
455 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
456 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
457 | break; |
458 | case XFS_TRANS_SB_IFREE: | |
459 | tp->t_ifree_delta += delta; | |
92821e2b DC |
460 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
461 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
462 | break; |
463 | case XFS_TRANS_SB_FDBLOCKS: | |
464 | /* | |
465 | * Track the number of blocks allocated in the | |
466 | * transaction. Make sure it does not exceed the | |
467 | * number reserved. | |
468 | */ | |
469 | if (delta < 0) { | |
470 | tp->t_blk_res_used += (uint)-delta; | |
471 | ASSERT(tp->t_blk_res_used <= tp->t_blk_res); | |
472 | } | |
473 | tp->t_fdblocks_delta += delta; | |
92821e2b DC |
474 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
475 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
476 | break; |
477 | case XFS_TRANS_SB_RES_FDBLOCKS: | |
478 | /* | |
479 | * The allocation has already been applied to the | |
480 | * in-core superblock's counter. This should only | |
481 | * be applied to the on-disk superblock. | |
482 | */ | |
483 | ASSERT(delta < 0); | |
484 | tp->t_res_fdblocks_delta += delta; | |
92821e2b DC |
485 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
486 | flags &= ~XFS_TRANS_SB_DIRTY; | |
1da177e4 LT |
487 | break; |
488 | case XFS_TRANS_SB_FREXTENTS: | |
489 | /* | |
490 | * Track the number of blocks allocated in the | |
491 | * transaction. Make sure it does not exceed the | |
492 | * number reserved. | |
493 | */ | |
494 | if (delta < 0) { | |
495 | tp->t_rtx_res_used += (uint)-delta; | |
496 | ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res); | |
497 | } | |
498 | tp->t_frextents_delta += delta; | |
499 | break; | |
500 | case XFS_TRANS_SB_RES_FREXTENTS: | |
501 | /* | |
502 | * The allocation has already been applied to the | |
c41564b5 | 503 | * in-core superblock's counter. This should only |
1da177e4 LT |
504 | * be applied to the on-disk superblock. |
505 | */ | |
506 | ASSERT(delta < 0); | |
507 | tp->t_res_frextents_delta += delta; | |
508 | break; | |
509 | case XFS_TRANS_SB_DBLOCKS: | |
510 | ASSERT(delta > 0); | |
511 | tp->t_dblocks_delta += delta; | |
512 | break; | |
513 | case XFS_TRANS_SB_AGCOUNT: | |
514 | ASSERT(delta > 0); | |
515 | tp->t_agcount_delta += delta; | |
516 | break; | |
517 | case XFS_TRANS_SB_IMAXPCT: | |
518 | tp->t_imaxpct_delta += delta; | |
519 | break; | |
520 | case XFS_TRANS_SB_REXTSIZE: | |
521 | tp->t_rextsize_delta += delta; | |
522 | break; | |
523 | case XFS_TRANS_SB_RBMBLOCKS: | |
524 | tp->t_rbmblocks_delta += delta; | |
525 | break; | |
526 | case XFS_TRANS_SB_RBLOCKS: | |
527 | tp->t_rblocks_delta += delta; | |
528 | break; | |
529 | case XFS_TRANS_SB_REXTENTS: | |
530 | tp->t_rextents_delta += delta; | |
531 | break; | |
532 | case XFS_TRANS_SB_REXTSLOG: | |
533 | tp->t_rextslog_delta += delta; | |
534 | break; | |
535 | default: | |
536 | ASSERT(0); | |
537 | return; | |
538 | } | |
539 | ||
210c6f1c | 540 | tp->t_flags |= flags; |
1da177e4 LT |
541 | } |
542 | ||
543 | /* | |
544 | * xfs_trans_apply_sb_deltas() is called from the commit code | |
545 | * to bring the superblock buffer into the current transaction | |
546 | * and modify it as requested by earlier calls to xfs_trans_mod_sb(). | |
547 | * | |
548 | * For now we just look at each field allowed to change and change | |
549 | * it if necessary. | |
550 | */ | |
551 | STATIC void | |
552 | xfs_trans_apply_sb_deltas( | |
553 | xfs_trans_t *tp) | |
554 | { | |
2bdf7cd0 | 555 | xfs_dsb_t *sbp; |
1da177e4 LT |
556 | xfs_buf_t *bp; |
557 | int whole = 0; | |
558 | ||
559 | bp = xfs_trans_getsb(tp, tp->t_mountp, 0); | |
560 | sbp = XFS_BUF_TO_SBP(bp); | |
561 | ||
562 | /* | |
563 | * Check that superblock mods match the mods made to AGF counters. | |
564 | */ | |
565 | ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) == | |
566 | (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta + | |
567 | tp->t_ag_btree_delta)); | |
568 | ||
92821e2b DC |
569 | /* |
570 | * Only update the superblock counters if we are logging them | |
571 | */ | |
572 | if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) { | |
2bdf7cd0 | 573 | if (tp->t_icount_delta) |
413d57c9 | 574 | be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta); |
2bdf7cd0 | 575 | if (tp->t_ifree_delta) |
413d57c9 | 576 | be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta); |
2bdf7cd0 | 577 | if (tp->t_fdblocks_delta) |
413d57c9 | 578 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta); |
2bdf7cd0 | 579 | if (tp->t_res_fdblocks_delta) |
413d57c9 | 580 | be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta); |
1da177e4 LT |
581 | } |
582 | ||
2bdf7cd0 | 583 | if (tp->t_frextents_delta) |
413d57c9 | 584 | be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta); |
2bdf7cd0 | 585 | if (tp->t_res_frextents_delta) |
413d57c9 | 586 | be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta); |
2bdf7cd0 CH |
587 | |
588 | if (tp->t_dblocks_delta) { | |
413d57c9 | 589 | be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta); |
1da177e4 LT |
590 | whole = 1; |
591 | } | |
2bdf7cd0 | 592 | if (tp->t_agcount_delta) { |
413d57c9 | 593 | be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta); |
1da177e4 LT |
594 | whole = 1; |
595 | } | |
2bdf7cd0 CH |
596 | if (tp->t_imaxpct_delta) { |
597 | sbp->sb_imax_pct += tp->t_imaxpct_delta; | |
1da177e4 LT |
598 | whole = 1; |
599 | } | |
2bdf7cd0 | 600 | if (tp->t_rextsize_delta) { |
413d57c9 | 601 | be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta); |
1da177e4 LT |
602 | whole = 1; |
603 | } | |
2bdf7cd0 | 604 | if (tp->t_rbmblocks_delta) { |
413d57c9 | 605 | be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta); |
1da177e4 LT |
606 | whole = 1; |
607 | } | |
2bdf7cd0 | 608 | if (tp->t_rblocks_delta) { |
413d57c9 | 609 | be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta); |
1da177e4 LT |
610 | whole = 1; |
611 | } | |
2bdf7cd0 | 612 | if (tp->t_rextents_delta) { |
413d57c9 | 613 | be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta); |
1da177e4 LT |
614 | whole = 1; |
615 | } | |
2bdf7cd0 CH |
616 | if (tp->t_rextslog_delta) { |
617 | sbp->sb_rextslog += tp->t_rextslog_delta; | |
1da177e4 LT |
618 | whole = 1; |
619 | } | |
620 | ||
621 | if (whole) | |
622 | /* | |
c41564b5 | 623 | * Log the whole thing, the fields are noncontiguous. |
1da177e4 | 624 | */ |
2bdf7cd0 | 625 | xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1); |
1da177e4 LT |
626 | else |
627 | /* | |
628 | * Since all the modifiable fields are contiguous, we | |
629 | * can get away with this. | |
630 | */ | |
2bdf7cd0 CH |
631 | xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount), |
632 | offsetof(xfs_dsb_t, sb_frextents) + | |
1da177e4 | 633 | sizeof(sbp->sb_frextents) - 1); |
1da177e4 LT |
634 | } |
635 | ||
636 | /* | |
45c34141 DC |
637 | * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations |
638 | * and apply superblock counter changes to the in-core superblock. The | |
639 | * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT | |
640 | * applied to the in-core superblock. The idea is that that has already been | |
641 | * done. | |
1da177e4 LT |
642 | * |
643 | * This is done efficiently with a single call to xfs_mod_incore_sb_batch(). | |
45c34141 DC |
644 | * However, we have to ensure that we only modify each superblock field only |
645 | * once because the application of the delta values may not be atomic. That can | |
646 | * lead to ENOSPC races occurring if we have two separate modifcations of the | |
647 | * free space counter to put back the entire reservation and then take away | |
648 | * what we used. | |
649 | * | |
650 | * If we are not logging superblock counters, then the inode allocated/free and | |
651 | * used block counts are not updated in the on disk superblock. In this case, | |
652 | * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we | |
653 | * still need to update the incore superblock with the changes. | |
1da177e4 | 654 | */ |
ba0f32d4 | 655 | STATIC void |
1da177e4 LT |
656 | xfs_trans_unreserve_and_mod_sb( |
657 | xfs_trans_t *tp) | |
658 | { | |
659 | xfs_mod_sb_t msb[14]; /* If you add cases, add entries */ | |
660 | xfs_mod_sb_t *msbp; | |
92821e2b | 661 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
662 | /* REFERENCED */ |
663 | int error; | |
664 | int rsvd; | |
45c34141 DC |
665 | int64_t blkdelta = 0; |
666 | int64_t rtxdelta = 0; | |
1da177e4 LT |
667 | |
668 | msbp = msb; | |
669 | rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; | |
670 | ||
45c34141 DC |
671 | /* calculate free blocks delta */ |
672 | if (tp->t_blk_res > 0) | |
673 | blkdelta = tp->t_blk_res; | |
674 | ||
675 | if ((tp->t_fdblocks_delta != 0) && | |
676 | (xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
677 | (tp->t_flags & XFS_TRANS_SB_DIRTY))) | |
678 | blkdelta += tp->t_fdblocks_delta; | |
679 | ||
680 | if (blkdelta != 0) { | |
1da177e4 | 681 | msbp->msb_field = XFS_SBS_FDBLOCKS; |
45c34141 | 682 | msbp->msb_delta = blkdelta; |
1da177e4 LT |
683 | msbp++; |
684 | } | |
685 | ||
45c34141 DC |
686 | /* calculate free realtime extents delta */ |
687 | if (tp->t_rtx_res > 0) | |
688 | rtxdelta = tp->t_rtx_res; | |
689 | ||
690 | if ((tp->t_frextents_delta != 0) && | |
691 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) | |
692 | rtxdelta += tp->t_frextents_delta; | |
693 | ||
694 | if (rtxdelta != 0) { | |
1da177e4 | 695 | msbp->msb_field = XFS_SBS_FREXTENTS; |
45c34141 | 696 | msbp->msb_delta = rtxdelta; |
1da177e4 LT |
697 | msbp++; |
698 | } | |
699 | ||
45c34141 DC |
700 | /* apply remaining deltas */ |
701 | ||
92821e2b DC |
702 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) || |
703 | (tp->t_flags & XFS_TRANS_SB_DIRTY)) { | |
1da177e4 LT |
704 | if (tp->t_icount_delta != 0) { |
705 | msbp->msb_field = XFS_SBS_ICOUNT; | |
20f4ebf2 | 706 | msbp->msb_delta = tp->t_icount_delta; |
1da177e4 LT |
707 | msbp++; |
708 | } | |
709 | if (tp->t_ifree_delta != 0) { | |
710 | msbp->msb_field = XFS_SBS_IFREE; | |
20f4ebf2 | 711 | msbp->msb_delta = tp->t_ifree_delta; |
1da177e4 LT |
712 | msbp++; |
713 | } | |
92821e2b DC |
714 | } |
715 | ||
716 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) { | |
1da177e4 LT |
717 | if (tp->t_dblocks_delta != 0) { |
718 | msbp->msb_field = XFS_SBS_DBLOCKS; | |
20f4ebf2 | 719 | msbp->msb_delta = tp->t_dblocks_delta; |
1da177e4 LT |
720 | msbp++; |
721 | } | |
722 | if (tp->t_agcount_delta != 0) { | |
723 | msbp->msb_field = XFS_SBS_AGCOUNT; | |
20f4ebf2 | 724 | msbp->msb_delta = tp->t_agcount_delta; |
1da177e4 LT |
725 | msbp++; |
726 | } | |
727 | if (tp->t_imaxpct_delta != 0) { | |
728 | msbp->msb_field = XFS_SBS_IMAX_PCT; | |
20f4ebf2 | 729 | msbp->msb_delta = tp->t_imaxpct_delta; |
1da177e4 LT |
730 | msbp++; |
731 | } | |
732 | if (tp->t_rextsize_delta != 0) { | |
733 | msbp->msb_field = XFS_SBS_REXTSIZE; | |
20f4ebf2 | 734 | msbp->msb_delta = tp->t_rextsize_delta; |
1da177e4 LT |
735 | msbp++; |
736 | } | |
737 | if (tp->t_rbmblocks_delta != 0) { | |
738 | msbp->msb_field = XFS_SBS_RBMBLOCKS; | |
20f4ebf2 | 739 | msbp->msb_delta = tp->t_rbmblocks_delta; |
1da177e4 LT |
740 | msbp++; |
741 | } | |
742 | if (tp->t_rblocks_delta != 0) { | |
743 | msbp->msb_field = XFS_SBS_RBLOCKS; | |
20f4ebf2 | 744 | msbp->msb_delta = tp->t_rblocks_delta; |
1da177e4 LT |
745 | msbp++; |
746 | } | |
747 | if (tp->t_rextents_delta != 0) { | |
748 | msbp->msb_field = XFS_SBS_REXTENTS; | |
20f4ebf2 | 749 | msbp->msb_delta = tp->t_rextents_delta; |
1da177e4 LT |
750 | msbp++; |
751 | } | |
752 | if (tp->t_rextslog_delta != 0) { | |
753 | msbp->msb_field = XFS_SBS_REXTSLOG; | |
20f4ebf2 | 754 | msbp->msb_delta = tp->t_rextslog_delta; |
1da177e4 LT |
755 | msbp++; |
756 | } | |
757 | } | |
758 | ||
759 | /* | |
760 | * If we need to change anything, do it. | |
761 | */ | |
762 | if (msbp > msb) { | |
763 | error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, | |
764 | (uint)(msbp - msb), rsvd); | |
765 | ASSERT(error == 0); | |
766 | } | |
767 | } | |
768 | ||
1da177e4 | 769 | /* |
0924378a DC |
770 | * Total up the number of log iovecs needed to commit this |
771 | * transaction. The transaction itself needs one for the | |
772 | * transaction header. Ask each dirty item in turn how many | |
773 | * it needs to get the total. | |
1da177e4 | 774 | */ |
0924378a DC |
775 | static uint |
776 | xfs_trans_count_vecs( | |
b1c1b5b6 | 777 | struct xfs_trans *tp) |
1da177e4 | 778 | { |
0924378a DC |
779 | int nvecs; |
780 | xfs_log_item_desc_t *lidp; | |
1da177e4 | 781 | |
0924378a DC |
782 | nvecs = 1; |
783 | lidp = xfs_trans_first_item(tp); | |
784 | ASSERT(lidp != NULL); | |
1da177e4 | 785 | |
0924378a DC |
786 | /* In the non-debug case we need to start bailing out if we |
787 | * didn't find a log_item here, return zero and let trans_commit | |
788 | * deal with it. | |
1da177e4 | 789 | */ |
0924378a DC |
790 | if (lidp == NULL) |
791 | return 0; | |
792 | ||
793 | while (lidp != NULL) { | |
794 | /* | |
795 | * Skip items which aren't dirty in this transaction. | |
796 | */ | |
797 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
798 | lidp = xfs_trans_next_item(tp, lidp); | |
799 | continue; | |
800 | } | |
801 | lidp->lid_size = IOP_SIZE(lidp->lid_item); | |
802 | nvecs += lidp->lid_size; | |
803 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 804 | } |
0924378a DC |
805 | |
806 | return nvecs; | |
807 | } | |
808 | ||
809 | /* | |
810 | * Fill in the vector with pointers to data to be logged | |
811 | * by this transaction. The transaction header takes | |
812 | * the first vector, and then each dirty item takes the | |
813 | * number of vectors it indicated it needed in xfs_trans_count_vecs(). | |
814 | * | |
815 | * As each item fills in the entries it needs, also pin the item | |
816 | * so that it cannot be flushed out until the log write completes. | |
817 | */ | |
818 | static void | |
819 | xfs_trans_fill_vecs( | |
820 | struct xfs_trans *tp, | |
821 | struct xfs_log_iovec *log_vector) | |
822 | { | |
823 | xfs_log_item_desc_t *lidp; | |
824 | struct xfs_log_iovec *vecp; | |
825 | uint nitems; | |
1da177e4 LT |
826 | |
827 | /* | |
0924378a DC |
828 | * Skip over the entry for the transaction header, we'll |
829 | * fill that in at the end. | |
1da177e4 | 830 | */ |
0924378a DC |
831 | vecp = log_vector + 1; |
832 | ||
833 | nitems = 0; | |
834 | lidp = xfs_trans_first_item(tp); | |
835 | ASSERT(lidp); | |
836 | while (lidp) { | |
837 | /* Skip items which aren't dirty in this transaction. */ | |
838 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) { | |
839 | lidp = xfs_trans_next_item(tp, lidp); | |
840 | continue; | |
841 | } | |
842 | ||
1da177e4 | 843 | /* |
0924378a DC |
844 | * The item may be marked dirty but not log anything. This can |
845 | * be used to get called when a transaction is committed. | |
1da177e4 | 846 | */ |
0924378a DC |
847 | if (lidp->lid_size) |
848 | nitems++; | |
849 | IOP_FORMAT(lidp->lid_item, vecp); | |
850 | vecp += lidp->lid_size; | |
851 | IOP_PIN(lidp->lid_item); | |
852 | lidp = xfs_trans_next_item(tp, lidp); | |
1da177e4 | 853 | } |
1da177e4 LT |
854 | |
855 | /* | |
0924378a DC |
856 | * Now that we've counted the number of items in this transaction, fill |
857 | * in the transaction header. Note that the transaction header does not | |
858 | * have a log item. | |
1da177e4 | 859 | */ |
0924378a DC |
860 | tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC; |
861 | tp->t_header.th_type = tp->t_type; | |
862 | tp->t_header.th_num_items = nitems; | |
863 | log_vector->i_addr = (xfs_caddr_t)&tp->t_header; | |
864 | log_vector->i_len = sizeof(xfs_trans_header_t); | |
865 | log_vector->i_type = XLOG_REG_TYPE_TRANSHDR; | |
866 | } | |
867 | ||
b1c1b5b6 DC |
868 | /* |
869 | * The committed item processing consists of calling the committed routine of | |
870 | * each logged item, updating the item's position in the AIL if necessary, and | |
871 | * unpinning each item. If the committed routine returns -1, then do nothing | |
872 | * further with the item because it may have been freed. | |
873 | * | |
874 | * Since items are unlocked when they are copied to the incore log, it is | |
875 | * possible for two transactions to be completing and manipulating the same | |
876 | * item simultaneously. The AIL lock will protect the lsn field of each item. | |
877 | * The value of this field can never go backwards. | |
878 | * | |
879 | * We unpin the items after repositioning them in the AIL, because otherwise | |
880 | * they could be immediately flushed and we'd have to race with the flusher | |
881 | * trying to pull the item from the AIL as we add it. | |
882 | */ | |
883 | static void | |
884 | xfs_trans_item_committed( | |
885 | struct xfs_log_item *lip, | |
886 | xfs_lsn_t commit_lsn, | |
887 | int aborted) | |
888 | { | |
889 | xfs_lsn_t item_lsn; | |
890 | struct xfs_ail *ailp; | |
891 | ||
892 | if (aborted) | |
893 | lip->li_flags |= XFS_LI_ABORTED; | |
894 | item_lsn = IOP_COMMITTED(lip, commit_lsn); | |
895 | ||
896 | /* If the committed routine returns -1, item has been freed. */ | |
897 | if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) | |
898 | return; | |
899 | ||
900 | /* | |
901 | * If the returned lsn is greater than what it contained before, update | |
902 | * the location of the item in the AIL. If it is not, then do nothing. | |
903 | * Items can never move backwards in the AIL. | |
904 | * | |
905 | * While the new lsn should usually be greater, it is possible that a | |
906 | * later transaction completing simultaneously with an earlier one | |
907 | * using the same item could complete first with a higher lsn. This | |
908 | * would cause the earlier transaction to fail the test below. | |
909 | */ | |
910 | ailp = lip->li_ailp; | |
911 | spin_lock(&ailp->xa_lock); | |
912 | if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) { | |
913 | /* | |
914 | * This will set the item's lsn to item_lsn and update the | |
915 | * position of the item in the AIL. | |
916 | * | |
917 | * xfs_trans_ail_update() drops the AIL lock. | |
918 | */ | |
919 | xfs_trans_ail_update(ailp, lip, item_lsn); | |
920 | } else { | |
921 | spin_unlock(&ailp->xa_lock); | |
922 | } | |
923 | ||
924 | /* | |
925 | * Now that we've repositioned the item in the AIL, unpin it so it can | |
926 | * be flushed. Pass information about buffer stale state down from the | |
927 | * log item flags, if anyone else stales the buffer we do not want to | |
928 | * pay any attention to it. | |
929 | */ | |
930 | IOP_UNPIN(lip); | |
931 | } | |
932 | ||
933 | /* Clear all the per-AG busy list items listed in this transaction */ | |
934 | static void | |
935 | xfs_trans_clear_busy_extents( | |
936 | struct xfs_trans *tp) | |
937 | { | |
938 | xfs_log_busy_chunk_t *lbcp; | |
939 | xfs_log_busy_slot_t *lbsp; | |
940 | int i; | |
941 | ||
942 | for (lbcp = &tp->t_busy; lbcp != NULL; lbcp = lbcp->lbc_next) { | |
943 | i = 0; | |
944 | for (lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) { | |
945 | if (XFS_LBC_ISFREE(lbcp, i)) | |
946 | continue; | |
947 | xfs_alloc_clear_busy(tp, lbsp->lbc_ag, lbsp->lbc_idx); | |
948 | } | |
949 | } | |
950 | xfs_trans_free_busy(tp); | |
951 | } | |
952 | ||
953 | /* | |
954 | * This is typically called by the LM when a transaction has been fully | |
955 | * committed to disk. It needs to unpin the items which have | |
956 | * been logged by the transaction and update their positions | |
957 | * in the AIL if necessary. | |
958 | * | |
959 | * This also gets called when the transactions didn't get written out | |
960 | * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then. | |
961 | */ | |
962 | STATIC void | |
963 | xfs_trans_committed( | |
964 | struct xfs_trans *tp, | |
965 | int abortflag) | |
966 | { | |
967 | xfs_log_item_desc_t *lidp; | |
968 | xfs_log_item_chunk_t *licp; | |
969 | xfs_log_item_chunk_t *next_licp; | |
970 | ||
971 | /* Call the transaction's completion callback if there is one. */ | |
972 | if (tp->t_callback != NULL) | |
973 | tp->t_callback(tp, tp->t_callarg); | |
974 | ||
975 | for (lidp = xfs_trans_first_item(tp); | |
976 | lidp != NULL; | |
977 | lidp = xfs_trans_next_item(tp, lidp)) { | |
978 | xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag); | |
979 | } | |
980 | ||
981 | /* free the item chunks, ignoring the embedded chunk */ | |
982 | for (licp = tp->t_items.lic_next; licp != NULL; licp = next_licp) { | |
983 | next_licp = licp->lic_next; | |
984 | kmem_free(licp); | |
985 | } | |
986 | ||
987 | xfs_trans_clear_busy_extents(tp); | |
988 | xfs_trans_free(tp); | |
989 | } | |
990 | ||
991 | /* | |
992 | * Called from the trans_commit code when we notice that | |
993 | * the filesystem is in the middle of a forced shutdown. | |
994 | */ | |
995 | STATIC void | |
996 | xfs_trans_uncommit( | |
997 | struct xfs_trans *tp, | |
998 | uint flags) | |
999 | { | |
1000 | xfs_log_item_desc_t *lidp; | |
1001 | ||
1002 | for (lidp = xfs_trans_first_item(tp); | |
1003 | lidp != NULL; | |
1004 | lidp = xfs_trans_next_item(tp, lidp)) { | |
1005 | /* | |
1006 | * Unpin all but those that aren't dirty. | |
1007 | */ | |
1008 | if (lidp->lid_flags & XFS_LID_DIRTY) | |
1009 | IOP_UNPIN_REMOVE(lidp->lid_item, tp); | |
1010 | } | |
1011 | ||
1012 | xfs_trans_unreserve_and_mod_sb(tp); | |
1013 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1014 | ||
1015 | xfs_trans_free_items(tp, flags); | |
1016 | xfs_trans_free_busy(tp); | |
1017 | xfs_trans_free(tp); | |
1018 | } | |
1019 | ||
0924378a DC |
1020 | /* |
1021 | * Format the transaction direct to the iclog. This isolates the physical | |
1022 | * transaction commit operation from the logical operation and hence allows | |
1023 | * other methods to be introduced without affecting the existing commit path. | |
1024 | */ | |
1025 | static int | |
1026 | xfs_trans_commit_iclog( | |
1027 | struct xfs_mount *mp, | |
1028 | struct xfs_trans *tp, | |
1029 | xfs_lsn_t *commit_lsn, | |
1030 | int flags) | |
1031 | { | |
1032 | int shutdown; | |
1033 | int error; | |
1034 | int log_flags = 0; | |
1035 | struct xlog_in_core *commit_iclog; | |
1036 | #define XFS_TRANS_LOGVEC_COUNT 16 | |
1037 | struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT]; | |
1038 | struct xfs_log_iovec *log_vector; | |
1039 | uint nvec; | |
1040 | ||
1da177e4 LT |
1041 | |
1042 | /* | |
1043 | * Ask each log item how many log_vector entries it will | |
1044 | * need so we can figure out how many to allocate. | |
1045 | * Try to avoid the kmem_alloc() call in the common case | |
1046 | * by using a vector from the stack when it fits. | |
1047 | */ | |
1048 | nvec = xfs_trans_count_vecs(tp); | |
1da177e4 | 1049 | if (nvec == 0) { |
0924378a | 1050 | return ENOMEM; /* triggers a shutdown! */ |
cfcbbbd0 | 1051 | } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) { |
1da177e4 LT |
1052 | log_vector = log_vector_fast; |
1053 | } else { | |
1054 | log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec * | |
1055 | sizeof(xfs_log_iovec_t), | |
1056 | KM_SLEEP); | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Fill in the log_vector and pin the logged items, and | |
1061 | * then write the transaction to the log. | |
1062 | */ | |
1063 | xfs_trans_fill_vecs(tp, log_vector); | |
1064 | ||
0924378a DC |
1065 | if (flags & XFS_TRANS_RELEASE_LOG_RES) |
1066 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1067 | ||
cfcbbbd0 | 1068 | error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn)); |
1da177e4 | 1069 | |
1da177e4 | 1070 | /* |
cfcbbbd0 NS |
1071 | * The transaction is committed incore here, and can go out to disk |
1072 | * at any time after this call. However, all the items associated | |
1073 | * with the transaction are still locked and pinned in memory. | |
1da177e4 | 1074 | */ |
0924378a | 1075 | *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); |
1da177e4 | 1076 | |
0924378a DC |
1077 | tp->t_commit_lsn = *commit_lsn; |
1078 | if (nvec > XFS_TRANS_LOGVEC_COUNT) | |
f0e2d93c | 1079 | kmem_free(log_vector); |
1da177e4 | 1080 | |
1da177e4 LT |
1081 | /* |
1082 | * If we got a log write error. Unpin the logitems that we | |
1083 | * had pinned, clean up, free trans structure, and return error. | |
1084 | */ | |
0924378a | 1085 | if (error || *commit_lsn == -1) { |
59c1b082 | 1086 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1087 | xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT); |
1088 | return XFS_ERROR(EIO); | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * Once the transaction has committed, unused | |
1093 | * reservations need to be released and changes to | |
1094 | * the superblock need to be reflected in the in-core | |
1095 | * version. Do that now. | |
1096 | */ | |
1097 | xfs_trans_unreserve_and_mod_sb(tp); | |
1098 | ||
1da177e4 LT |
1099 | /* |
1100 | * Tell the LM to call the transaction completion routine | |
1101 | * when the log write with LSN commit_lsn completes (e.g. | |
1102 | * when the transaction commit really hits the on-disk log). | |
1103 | * After this call we cannot reference tp, because the call | |
1104 | * can happen at any time and the call will free the transaction | |
1105 | * structure pointed to by tp. The only case where we call | |
1106 | * the completion routine (xfs_trans_committed) directly is | |
1107 | * if the log is turned off on a debug kernel or we're | |
1108 | * running in simulation mode (the log is explicitly turned | |
1109 | * off). | |
1110 | */ | |
1111 | tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed; | |
1112 | tp->t_logcb.cb_arg = tp; | |
1113 | ||
1114 | /* | |
1115 | * We need to pass the iclog buffer which was used for the | |
1116 | * transaction commit record into this function, and attach | |
1117 | * the callback to it. The callback must be attached before | |
1118 | * the items are unlocked to avoid racing with other threads | |
1119 | * waiting for an item to unlock. | |
1120 | */ | |
1121 | shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb)); | |
1122 | ||
1123 | /* | |
1124 | * Mark this thread as no longer being in a transaction | |
1125 | */ | |
59c1b082 | 1126 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1127 | |
1128 | /* | |
1129 | * Once all the items of the transaction have been copied | |
1130 | * to the in core log and the callback is attached, the | |
1131 | * items can be unlocked. | |
1132 | * | |
1133 | * This will free descriptors pointing to items which were | |
1134 | * not logged since there is nothing more to do with them. | |
1135 | * For items which were logged, we will keep pointers to them | |
1136 | * so they can be unpinned after the transaction commits to disk. | |
1137 | * This will also stamp each modified meta-data item with | |
1138 | * the commit lsn of this transaction for dependency tracking | |
1139 | * purposes. | |
1140 | */ | |
0924378a | 1141 | xfs_trans_unlock_items(tp, *commit_lsn); |
1da177e4 LT |
1142 | |
1143 | /* | |
1144 | * If we detected a log error earlier, finish committing | |
1145 | * the transaction now (unpin log items, etc). | |
1146 | * | |
1147 | * Order is critical here, to avoid using the transaction | |
1148 | * pointer after its been freed (by xfs_trans_committed | |
1149 | * either here now, or as a callback). We cannot do this | |
1150 | * step inside xfs_log_notify as was done earlier because | |
1151 | * of this issue. | |
1152 | */ | |
1153 | if (shutdown) | |
1154 | xfs_trans_committed(tp, XFS_LI_ABORTED); | |
1155 | ||
1156 | /* | |
1157 | * Now that the xfs_trans_committed callback has been attached, | |
1158 | * and the items are released we can finally allow the iclog to | |
1159 | * go to disk. | |
1160 | */ | |
0924378a DC |
1161 | return xfs_log_release_iclog(mp, commit_iclog); |
1162 | } | |
1163 | ||
1164 | ||
1165 | /* | |
1166 | * xfs_trans_commit | |
1167 | * | |
1168 | * Commit the given transaction to the log a/synchronously. | |
1169 | * | |
1170 | * XFS disk error handling mechanism is not based on a typical | |
1171 | * transaction abort mechanism. Logically after the filesystem | |
1172 | * gets marked 'SHUTDOWN', we can't let any new transactions | |
1173 | * be durable - ie. committed to disk - because some metadata might | |
1174 | * be inconsistent. In such cases, this returns an error, and the | |
1175 | * caller may assume that all locked objects joined to the transaction | |
1176 | * have already been unlocked as if the commit had succeeded. | |
1177 | * Do not reference the transaction structure after this call. | |
1178 | */ | |
0924378a DC |
1179 | int |
1180 | _xfs_trans_commit( | |
a3ccd2ca CH |
1181 | struct xfs_trans *tp, |
1182 | uint flags, | |
1183 | int *log_flushed) | |
0924378a | 1184 | { |
a3ccd2ca | 1185 | struct xfs_mount *mp = tp->t_mountp; |
0924378a | 1186 | xfs_lsn_t commit_lsn = -1; |
a3ccd2ca | 1187 | int error = 0; |
0924378a DC |
1188 | int log_flags = 0; |
1189 | int sync = tp->t_flags & XFS_TRANS_SYNC; | |
0924378a DC |
1190 | |
1191 | /* | |
1192 | * Determine whether this commit is releasing a permanent | |
1193 | * log reservation or not. | |
1194 | */ | |
1195 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1196 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1197 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * If there is nothing to be logged by the transaction, | |
1202 | * then unlock all of the items associated with the | |
1203 | * transaction and free the transaction structure. | |
1204 | * Also make sure to return any reserved blocks to | |
1205 | * the free pool. | |
1206 | */ | |
a3ccd2ca CH |
1207 | if (!(tp->t_flags & XFS_TRANS_DIRTY)) |
1208 | goto out_unreserve; | |
1209 | ||
1210 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1211 | error = XFS_ERROR(EIO); | |
1212 | goto out_unreserve; | |
0924378a | 1213 | } |
a3ccd2ca | 1214 | |
0924378a DC |
1215 | ASSERT(tp->t_ticket != NULL); |
1216 | ||
1217 | /* | |
1218 | * If we need to update the superblock, then do it now. | |
1219 | */ | |
1220 | if (tp->t_flags & XFS_TRANS_SB_DIRTY) | |
1221 | xfs_trans_apply_sb_deltas(tp); | |
1222 | xfs_trans_apply_dquot_deltas(tp); | |
1223 | ||
1224 | error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags); | |
1225 | if (error == ENOMEM) { | |
1226 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); | |
a3ccd2ca CH |
1227 | error = XFS_ERROR(EIO); |
1228 | goto out_unreserve; | |
0924378a | 1229 | } |
1da177e4 LT |
1230 | |
1231 | /* | |
1232 | * If the transaction needs to be synchronous, then force the | |
1233 | * log out now and wait for it. | |
1234 | */ | |
1235 | if (sync) { | |
f538d4da | 1236 | if (!error) { |
a14a348b CH |
1237 | error = _xfs_log_force_lsn(mp, commit_lsn, |
1238 | XFS_LOG_SYNC, log_flushed); | |
f538d4da | 1239 | } |
1da177e4 LT |
1240 | XFS_STATS_INC(xs_trans_sync); |
1241 | } else { | |
1242 | XFS_STATS_INC(xs_trans_async); | |
1243 | } | |
1244 | ||
a3ccd2ca CH |
1245 | return error; |
1246 | ||
1247 | out_unreserve: | |
1248 | xfs_trans_unreserve_and_mod_sb(tp); | |
1249 | ||
1250 | /* | |
1251 | * It is indeed possible for the transaction to be not dirty but | |
1252 | * the dqinfo portion to be. All that means is that we have some | |
1253 | * (non-persistent) quota reservations that need to be unreserved. | |
1254 | */ | |
1255 | xfs_trans_unreserve_and_mod_dquots(tp); | |
1256 | if (tp->t_ticket) { | |
1257 | commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
1258 | if (commit_lsn == -1 && !error) | |
1259 | error = XFS_ERROR(EIO); | |
1260 | } | |
1261 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
1262 | xfs_trans_free_items(tp, error ? XFS_TRANS_ABORT : 0); | |
1263 | xfs_trans_free_busy(tp); | |
1264 | xfs_trans_free(tp); | |
1265 | ||
1266 | XFS_STATS_INC(xs_trans_empty); | |
1267 | return error; | |
1da177e4 LT |
1268 | } |
1269 | ||
1da177e4 LT |
1270 | /* |
1271 | * Unlock all of the transaction's items and free the transaction. | |
1272 | * The transaction must not have modified any of its items, because | |
1273 | * there is no way to restore them to their previous state. | |
1274 | * | |
1275 | * If the transaction has made a log reservation, make sure to release | |
1276 | * it as well. | |
1277 | */ | |
1278 | void | |
1279 | xfs_trans_cancel( | |
1280 | xfs_trans_t *tp, | |
1281 | int flags) | |
1282 | { | |
1283 | int log_flags; | |
1284 | #ifdef DEBUG | |
1285 | xfs_log_item_chunk_t *licp; | |
1286 | xfs_log_item_desc_t *lidp; | |
1287 | xfs_log_item_t *lip; | |
1288 | int i; | |
1289 | #endif | |
0733af21 | 1290 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1291 | |
1292 | /* | |
1293 | * See if the caller is being too lazy to figure out if | |
1294 | * the transaction really needs an abort. | |
1295 | */ | |
1296 | if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY)) | |
1297 | flags &= ~XFS_TRANS_ABORT; | |
1298 | /* | |
1299 | * See if the caller is relying on us to shut down the | |
1300 | * filesystem. This happens in paths where we detect | |
1301 | * corruption and decide to give up. | |
1302 | */ | |
60a204f0 | 1303 | if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) { |
0733af21 | 1304 | XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp); |
7d04a335 | 1305 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
60a204f0 | 1306 | } |
1da177e4 LT |
1307 | #ifdef DEBUG |
1308 | if (!(flags & XFS_TRANS_ABORT)) { | |
1309 | licp = &(tp->t_items); | |
1310 | while (licp != NULL) { | |
1311 | lidp = licp->lic_descs; | |
1312 | for (i = 0; i < licp->lic_unused; i++, lidp++) { | |
39dab9d7 | 1313 | if (xfs_lic_isfree(licp, i)) { |
1da177e4 LT |
1314 | continue; |
1315 | } | |
1316 | ||
1317 | lip = lidp->lid_item; | |
0733af21 | 1318 | if (!XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 LT |
1319 | ASSERT(!(lip->li_type == XFS_LI_EFD)); |
1320 | } | |
1321 | licp = licp->lic_next; | |
1322 | } | |
1323 | } | |
1324 | #endif | |
1325 | xfs_trans_unreserve_and_mod_sb(tp); | |
7d095257 | 1326 | xfs_trans_unreserve_and_mod_dquots(tp); |
1da177e4 LT |
1327 | |
1328 | if (tp->t_ticket) { | |
1329 | if (flags & XFS_TRANS_RELEASE_LOG_RES) { | |
1330 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1331 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
1332 | } else { | |
1333 | log_flags = 0; | |
1334 | } | |
0733af21 | 1335 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | /* mark this thread as no longer being in a transaction */ | |
59c1b082 | 1339 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
1da177e4 LT |
1340 | |
1341 | xfs_trans_free_items(tp, flags); | |
1342 | xfs_trans_free_busy(tp); | |
1343 | xfs_trans_free(tp); | |
1344 | } | |
1345 | ||
322ff6b8 NS |
1346 | /* |
1347 | * Roll from one trans in the sequence of PERMANENT transactions to | |
1348 | * the next: permanent transactions are only flushed out when | |
1349 | * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon | |
1350 | * as possible to let chunks of it go to the log. So we commit the | |
1351 | * chunk we've been working on and get a new transaction to continue. | |
1352 | */ | |
1353 | int | |
1354 | xfs_trans_roll( | |
1355 | struct xfs_trans **tpp, | |
1356 | struct xfs_inode *dp) | |
1357 | { | |
1358 | struct xfs_trans *trans; | |
1359 | unsigned int logres, count; | |
1360 | int error; | |
1361 | ||
1362 | /* | |
1363 | * Ensure that the inode is always logged. | |
1364 | */ | |
1365 | trans = *tpp; | |
1366 | xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE); | |
1367 | ||
1368 | /* | |
1369 | * Copy the critical parameters from one trans to the next. | |
1370 | */ | |
1371 | logres = trans->t_log_res; | |
1372 | count = trans->t_log_count; | |
1373 | *tpp = xfs_trans_dup(trans); | |
1374 | ||
1375 | /* | |
1376 | * Commit the current transaction. | |
1377 | * If this commit failed, then it'd just unlock those items that | |
1378 | * are not marked ihold. That also means that a filesystem shutdown | |
1379 | * is in progress. The caller takes the responsibility to cancel | |
1380 | * the duplicate transaction that gets returned. | |
1381 | */ | |
1382 | error = xfs_trans_commit(trans, 0); | |
1383 | if (error) | |
1384 | return (error); | |
1385 | ||
1386 | trans = *tpp; | |
1387 | ||
cc09c0dc DC |
1388 | /* |
1389 | * transaction commit worked ok so we can drop the extra ticket | |
1390 | * reference that we gained in xfs_trans_dup() | |
1391 | */ | |
1392 | xfs_log_ticket_put(trans->t_ticket); | |
1393 | ||
1394 | ||
322ff6b8 NS |
1395 | /* |
1396 | * Reserve space in the log for th next transaction. | |
1397 | * This also pushes items in the "AIL", the list of logged items, | |
1398 | * out to disk if they are taking up space at the tail of the log | |
1399 | * that we want to use. This requires that either nothing be locked | |
1400 | * across this call, or that anything that is locked be logged in | |
1401 | * the prior and the next transactions. | |
1402 | */ | |
1403 | error = xfs_trans_reserve(trans, 0, logres, 0, | |
1404 | XFS_TRANS_PERM_LOG_RES, count); | |
1405 | /* | |
1406 | * Ensure that the inode is in the new transaction and locked. | |
1407 | */ | |
1408 | if (error) | |
1409 | return error; | |
1410 | ||
1411 | xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL); | |
1412 | xfs_trans_ihold(trans, dp); | |
1413 | return 0; | |
1414 | } |