xref: /linux/fs/xfs/xfs_trans.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * Copyright (C) 2010 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_types.h"
22 #include "xfs_log.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_error.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_alloc.h"
37 #include "xfs_extent_busy.h"
38 #include "xfs_bmap.h"
39 #include "xfs_quota.h"
40 #include "xfs_trans_priv.h"
41 #include "xfs_trans_space.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_trace.h"
44 
45 kmem_zone_t	*xfs_trans_zone;
46 kmem_zone_t	*xfs_log_item_desc_zone;
47 
48 
49 /*
50  * Various log reservation values.
51  *
52  * These are based on the size of the file system block because that is what
53  * most transactions manipulate.  Each adds in an additional 128 bytes per
54  * item logged to try to account for the overhead of the transaction mechanism.
55  *
56  * Note:  Most of the reservations underestimate the number of allocation
57  * groups into which they could free extents in the xfs_bmap_finish() call.
58  * This is because the number in the worst case is quite high and quite
59  * unusual.  In order to fix this we need to change xfs_bmap_finish() to free
60  * extents in only a single AG at a time.  This will require changes to the
61  * EFI code as well, however, so that the EFI for the extents not freed is
62  * logged again in each transaction.  See SGI PV #261917.
63  *
64  * Reservation functions here avoid a huge stack in xfs_trans_init due to
65  * register overflow from temporaries in the calculations.
66  */
67 
68 
69 /*
70  * In a write transaction we can allocate a maximum of 2
71  * extents.  This gives:
72  *    the inode getting the new extents: inode size
73  *    the inode's bmap btree: max depth * block size
74  *    the agfs of the ags from which the extents are allocated: 2 * sector
75  *    the superblock free block counter: sector size
76  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
77  * And the bmap_finish transaction can free bmap blocks in a join:
78  *    the agfs of the ags containing the blocks: 2 * sector size
79  *    the agfls of the ags containing the blocks: 2 * sector size
80  *    the super block free block counter: sector size
81  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
82  */
83 STATIC uint
84 xfs_calc_write_reservation(
85 	struct xfs_mount	*mp)
86 {
87 	return XFS_DQUOT_LOGRES(mp) +
88 		MAX((mp->m_sb.sb_inodesize +
89 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
90 		     2 * mp->m_sb.sb_sectsize +
91 		     mp->m_sb.sb_sectsize +
92 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
93 		     128 * (4 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
94 			    XFS_ALLOCFREE_LOG_COUNT(mp, 2))),
95 		    (2 * mp->m_sb.sb_sectsize +
96 		     2 * mp->m_sb.sb_sectsize +
97 		     mp->m_sb.sb_sectsize +
98 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
99 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
100 }
101 
102 /*
103  * In truncating a file we free up to two extents at once.  We can modify:
104  *    the inode being truncated: inode size
105  *    the inode's bmap btree: (max depth + 1) * block size
106  * And the bmap_finish transaction can free the blocks and bmap blocks:
107  *    the agf for each of the ags: 4 * sector size
108  *    the agfl for each of the ags: 4 * sector size
109  *    the super block to reflect the freed blocks: sector size
110  *    worst case split in allocation btrees per extent assuming 4 extents:
111  *		4 exts * 2 trees * (2 * max depth - 1) * block size
112  *    the inode btree: max depth * blocksize
113  *    the allocation btrees: 2 trees * (max depth - 1) * block size
114  */
115 STATIC uint
116 xfs_calc_itruncate_reservation(
117 	struct xfs_mount	*mp)
118 {
119 	return XFS_DQUOT_LOGRES(mp) +
120 		MAX((mp->m_sb.sb_inodesize +
121 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1) +
122 		     128 * (2 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
123 		    (4 * mp->m_sb.sb_sectsize +
124 		     4 * mp->m_sb.sb_sectsize +
125 		     mp->m_sb.sb_sectsize +
126 		     XFS_ALLOCFREE_LOG_RES(mp, 4) +
127 		     128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)) +
128 		     128 * 5 +
129 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
130 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
131 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
132 }
133 
134 /*
135  * In renaming a files we can modify:
136  *    the four inodes involved: 4 * inode size
137  *    the two directory btrees: 2 * (max depth + v2) * dir block size
138  *    the two directory bmap btrees: 2 * max depth * block size
139  * And the bmap_finish transaction can free dir and bmap blocks (two sets
140  *	of bmap blocks) giving:
141  *    the agf for the ags in which the blocks live: 3 * sector size
142  *    the agfl for the ags in which the blocks live: 3 * sector size
143  *    the superblock for the free block count: sector size
144  *    the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
145  */
146 STATIC uint
147 xfs_calc_rename_reservation(
148 	struct xfs_mount	*mp)
149 {
150 	return XFS_DQUOT_LOGRES(mp) +
151 		MAX((4 * mp->m_sb.sb_inodesize +
152 		     2 * XFS_DIROP_LOG_RES(mp) +
153 		     128 * (4 + 2 * XFS_DIROP_LOG_COUNT(mp))),
154 		    (3 * mp->m_sb.sb_sectsize +
155 		     3 * mp->m_sb.sb_sectsize +
156 		     mp->m_sb.sb_sectsize +
157 		     XFS_ALLOCFREE_LOG_RES(mp, 3) +
158 		     128 * (7 + XFS_ALLOCFREE_LOG_COUNT(mp, 3))));
159 }
160 
161 /*
162  * For creating a link to an inode:
163  *    the parent directory inode: inode size
164  *    the linked inode: inode size
165  *    the directory btree could split: (max depth + v2) * dir block size
166  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
167  * And the bmap_finish transaction can free some bmap blocks giving:
168  *    the agf for the ag in which the blocks live: sector size
169  *    the agfl for the ag in which the blocks live: sector size
170  *    the superblock for the free block count: sector size
171  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
172  */
173 STATIC uint
174 xfs_calc_link_reservation(
175 	struct xfs_mount	*mp)
176 {
177 	return XFS_DQUOT_LOGRES(mp) +
178 		MAX((mp->m_sb.sb_inodesize +
179 		     mp->m_sb.sb_inodesize +
180 		     XFS_DIROP_LOG_RES(mp) +
181 		     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
182 		    (mp->m_sb.sb_sectsize +
183 		     mp->m_sb.sb_sectsize +
184 		     mp->m_sb.sb_sectsize +
185 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
186 		     128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
187 }
188 
189 /*
190  * For removing a directory entry we can modify:
191  *    the parent directory inode: inode size
192  *    the removed inode: inode size
193  *    the directory btree could join: (max depth + v2) * dir block size
194  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
195  * And the bmap_finish transaction can free the dir and bmap blocks giving:
196  *    the agf for the ag in which the blocks live: 2 * sector size
197  *    the agfl for the ag in which the blocks live: 2 * sector size
198  *    the superblock for the free block count: sector size
199  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
200  */
201 STATIC uint
202 xfs_calc_remove_reservation(
203 	struct xfs_mount	*mp)
204 {
205 	return XFS_DQUOT_LOGRES(mp) +
206 		MAX((mp->m_sb.sb_inodesize +
207 		     mp->m_sb.sb_inodesize +
208 		     XFS_DIROP_LOG_RES(mp) +
209 		     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
210 		    (2 * mp->m_sb.sb_sectsize +
211 		     2 * mp->m_sb.sb_sectsize +
212 		     mp->m_sb.sb_sectsize +
213 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
214 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
215 }
216 
217 /*
218  * For symlink we can modify:
219  *    the parent directory inode: inode size
220  *    the new inode: inode size
221  *    the inode btree entry: 1 block
222  *    the directory btree: (max depth + v2) * dir block size
223  *    the directory inode's bmap btree: (max depth + v2) * block size
224  *    the blocks for the symlink: 1 kB
225  * Or in the first xact we allocate some inodes giving:
226  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
227  *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
228  *    the inode btree: max depth * blocksize
229  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
230  */
231 STATIC uint
232 xfs_calc_symlink_reservation(
233 	struct xfs_mount	*mp)
234 {
235 	return XFS_DQUOT_LOGRES(mp) +
236 		MAX((mp->m_sb.sb_inodesize +
237 		     mp->m_sb.sb_inodesize +
238 		     XFS_FSB_TO_B(mp, 1) +
239 		     XFS_DIROP_LOG_RES(mp) +
240 		     1024 +
241 		     128 * (4 + XFS_DIROP_LOG_COUNT(mp))),
242 		    (2 * mp->m_sb.sb_sectsize +
243 		     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
244 		     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
245 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
246 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
247 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
248 }
249 
250 /*
251  * For create we can modify:
252  *    the parent directory inode: inode size
253  *    the new inode: inode size
254  *    the inode btree entry: block size
255  *    the superblock for the nlink flag: sector size
256  *    the directory btree: (max depth + v2) * dir block size
257  *    the directory inode's bmap btree: (max depth + v2) * block size
258  * Or in the first xact we allocate some inodes giving:
259  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
260  *    the superblock for the nlink flag: sector size
261  *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
262  *    the inode btree: max depth * blocksize
263  *    the allocation btrees: 2 trees * (max depth - 1) * block size
264  */
265 STATIC uint
266 xfs_calc_create_reservation(
267 	struct xfs_mount	*mp)
268 {
269 	return XFS_DQUOT_LOGRES(mp) +
270 		MAX((mp->m_sb.sb_inodesize +
271 		     mp->m_sb.sb_inodesize +
272 		     mp->m_sb.sb_sectsize +
273 		     XFS_FSB_TO_B(mp, 1) +
274 		     XFS_DIROP_LOG_RES(mp) +
275 		     128 * (3 + XFS_DIROP_LOG_COUNT(mp))),
276 		    (3 * mp->m_sb.sb_sectsize +
277 		     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
278 		     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
279 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
280 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
281 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
282 }
283 
284 /*
285  * Making a new directory is the same as creating a new file.
286  */
287 STATIC uint
288 xfs_calc_mkdir_reservation(
289 	struct xfs_mount	*mp)
290 {
291 	return xfs_calc_create_reservation(mp);
292 }
293 
294 /*
295  * In freeing an inode we can modify:
296  *    the inode being freed: inode size
297  *    the super block free inode counter: sector size
298  *    the agi hash list and counters: sector size
299  *    the inode btree entry: block size
300  *    the on disk inode before ours in the agi hash list: inode cluster size
301  *    the inode btree: max depth * blocksize
302  *    the allocation btrees: 2 trees * (max depth - 1) * block size
303  */
304 STATIC uint
305 xfs_calc_ifree_reservation(
306 	struct xfs_mount	*mp)
307 {
308 	return XFS_DQUOT_LOGRES(mp) +
309 		mp->m_sb.sb_inodesize +
310 		mp->m_sb.sb_sectsize +
311 		mp->m_sb.sb_sectsize +
312 		XFS_FSB_TO_B(mp, 1) +
313 		MAX((__uint16_t)XFS_FSB_TO_B(mp, 1),
314 		    XFS_INODE_CLUSTER_SIZE(mp)) +
315 		128 * 5 +
316 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
317 		128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
318 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
319 }
320 
321 /*
322  * When only changing the inode we log the inode and possibly the superblock
323  * We also add a bit of slop for the transaction stuff.
324  */
325 STATIC uint
326 xfs_calc_ichange_reservation(
327 	struct xfs_mount	*mp)
328 {
329 	return XFS_DQUOT_LOGRES(mp) +
330 		mp->m_sb.sb_inodesize +
331 		mp->m_sb.sb_sectsize +
332 		512;
333 
334 }
335 
336 /*
337  * Growing the data section of the filesystem.
338  *	superblock
339  *	agi and agf
340  *	allocation btrees
341  */
342 STATIC uint
343 xfs_calc_growdata_reservation(
344 	struct xfs_mount	*mp)
345 {
346 	return mp->m_sb.sb_sectsize * 3 +
347 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
348 		128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1));
349 }
350 
351 /*
352  * Growing the rt section of the filesystem.
353  * In the first set of transactions (ALLOC) we allocate space to the
354  * bitmap or summary files.
355  *	superblock: sector size
356  *	agf of the ag from which the extent is allocated: sector size
357  *	bmap btree for bitmap/summary inode: max depth * blocksize
358  *	bitmap/summary inode: inode size
359  *	allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
360  */
361 STATIC uint
362 xfs_calc_growrtalloc_reservation(
363 	struct xfs_mount	*mp)
364 {
365 	return 2 * mp->m_sb.sb_sectsize +
366 		XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
367 		mp->m_sb.sb_inodesize +
368 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
369 		128 * (3 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
370 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
371 }
372 
373 /*
374  * Growing the rt section of the filesystem.
375  * In the second set of transactions (ZERO) we zero the new metadata blocks.
376  *	one bitmap/summary block: blocksize
377  */
378 STATIC uint
379 xfs_calc_growrtzero_reservation(
380 	struct xfs_mount	*mp)
381 {
382 	return mp->m_sb.sb_blocksize + 128;
383 }
384 
385 /*
386  * Growing the rt section of the filesystem.
387  * In the third set of transactions (FREE) we update metadata without
388  * allocating any new blocks.
389  *	superblock: sector size
390  *	bitmap inode: inode size
391  *	summary inode: inode size
392  *	one bitmap block: blocksize
393  *	summary blocks: new summary size
394  */
395 STATIC uint
396 xfs_calc_growrtfree_reservation(
397 	struct xfs_mount	*mp)
398 {
399 	return mp->m_sb.sb_sectsize +
400 		2 * mp->m_sb.sb_inodesize +
401 		mp->m_sb.sb_blocksize +
402 		mp->m_rsumsize +
403 		128 * 5;
404 }
405 
406 /*
407  * Logging the inode modification timestamp on a synchronous write.
408  *	inode
409  */
410 STATIC uint
411 xfs_calc_swrite_reservation(
412 	struct xfs_mount	*mp)
413 {
414 	return mp->m_sb.sb_inodesize + 128;
415 }
416 
417 /*
418  * Logging the inode mode bits when writing a setuid/setgid file
419  *	inode
420  */
421 STATIC uint
422 xfs_calc_writeid_reservation(xfs_mount_t *mp)
423 {
424 	return mp->m_sb.sb_inodesize + 128;
425 }
426 
427 /*
428  * Converting the inode from non-attributed to attributed.
429  *	the inode being converted: inode size
430  *	agf block and superblock (for block allocation)
431  *	the new block (directory sized)
432  *	bmap blocks for the new directory block
433  *	allocation btrees
434  */
435 STATIC uint
436 xfs_calc_addafork_reservation(
437 	struct xfs_mount	*mp)
438 {
439 	return XFS_DQUOT_LOGRES(mp) +
440 		mp->m_sb.sb_inodesize +
441 		mp->m_sb.sb_sectsize * 2 +
442 		mp->m_dirblksize +
443 		XFS_FSB_TO_B(mp, XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1) +
444 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
445 		128 * (4 + XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1 +
446 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
447 }
448 
449 /*
450  * Removing the attribute fork of a file
451  *    the inode being truncated: inode size
452  *    the inode's bmap btree: max depth * block size
453  * And the bmap_finish transaction can free the blocks and bmap blocks:
454  *    the agf for each of the ags: 4 * sector size
455  *    the agfl for each of the ags: 4 * sector size
456  *    the super block to reflect the freed blocks: sector size
457  *    worst case split in allocation btrees per extent assuming 4 extents:
458  *		4 exts * 2 trees * (2 * max depth - 1) * block size
459  */
460 STATIC uint
461 xfs_calc_attrinval_reservation(
462 	struct xfs_mount	*mp)
463 {
464 	return MAX((mp->m_sb.sb_inodesize +
465 		    XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
466 		    128 * (1 + XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))),
467 		   (4 * mp->m_sb.sb_sectsize +
468 		    4 * mp->m_sb.sb_sectsize +
469 		    mp->m_sb.sb_sectsize +
470 		    XFS_ALLOCFREE_LOG_RES(mp, 4) +
471 		    128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4))));
472 }
473 
474 /*
475  * Setting an attribute.
476  *	the inode getting the attribute
477  *	the superblock for allocations
478  *	the agfs extents are allocated from
479  *	the attribute btree * max depth
480  *	the inode allocation btree
481  * Since attribute transaction space is dependent on the size of the attribute,
482  * the calculation is done partially at mount time and partially at runtime.
483  */
484 STATIC uint
485 xfs_calc_attrset_reservation(
486 	struct xfs_mount	*mp)
487 {
488 	return XFS_DQUOT_LOGRES(mp) +
489 		mp->m_sb.sb_inodesize +
490 		mp->m_sb.sb_sectsize +
491 		XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
492 		128 * (2 + XFS_DA_NODE_MAXDEPTH);
493 }
494 
495 /*
496  * Removing an attribute.
497  *    the inode: inode size
498  *    the attribute btree could join: max depth * block size
499  *    the inode bmap btree could join or split: max depth * block size
500  * And the bmap_finish transaction can free the attr blocks freed giving:
501  *    the agf for the ag in which the blocks live: 2 * sector size
502  *    the agfl for the ag in which the blocks live: 2 * sector size
503  *    the superblock for the free block count: sector size
504  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
505  */
506 STATIC uint
507 xfs_calc_attrrm_reservation(
508 	struct xfs_mount	*mp)
509 {
510 	return XFS_DQUOT_LOGRES(mp) +
511 		MAX((mp->m_sb.sb_inodesize +
512 		     XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
513 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
514 		     128 * (1 + XFS_DA_NODE_MAXDEPTH +
515 			    XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
516 		    (2 * mp->m_sb.sb_sectsize +
517 		     2 * mp->m_sb.sb_sectsize +
518 		     mp->m_sb.sb_sectsize +
519 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
520 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
521 }
522 
523 /*
524  * Clearing a bad agino number in an agi hash bucket.
525  */
526 STATIC uint
527 xfs_calc_clear_agi_bucket_reservation(
528 	struct xfs_mount	*mp)
529 {
530 	return mp->m_sb.sb_sectsize + 128;
531 }
532 
533 /*
534  * Initialize the precomputed transaction reservation values
535  * in the mount structure.
536  */
537 void
538 xfs_trans_init(
539 	struct xfs_mount	*mp)
540 {
541 	struct xfs_trans_reservations *resp = &mp->m_reservations;
542 
543 	resp->tr_write = xfs_calc_write_reservation(mp);
544 	resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
545 	resp->tr_rename = xfs_calc_rename_reservation(mp);
546 	resp->tr_link = xfs_calc_link_reservation(mp);
547 	resp->tr_remove = xfs_calc_remove_reservation(mp);
548 	resp->tr_symlink = xfs_calc_symlink_reservation(mp);
549 	resp->tr_create = xfs_calc_create_reservation(mp);
550 	resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
551 	resp->tr_ifree = xfs_calc_ifree_reservation(mp);
552 	resp->tr_ichange = xfs_calc_ichange_reservation(mp);
553 	resp->tr_growdata = xfs_calc_growdata_reservation(mp);
554 	resp->tr_swrite = xfs_calc_swrite_reservation(mp);
555 	resp->tr_writeid = xfs_calc_writeid_reservation(mp);
556 	resp->tr_addafork = xfs_calc_addafork_reservation(mp);
557 	resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
558 	resp->tr_attrset = xfs_calc_attrset_reservation(mp);
559 	resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
560 	resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
561 	resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
562 	resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
563 	resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
564 }
565 
566 /*
567  * This routine is called to allocate a transaction structure.
568  * The type parameter indicates the type of the transaction.  These
569  * are enumerated in xfs_trans.h.
570  *
571  * Dynamically allocate the transaction structure from the transaction
572  * zone, initialize it, and return it to the caller.
573  */
574 xfs_trans_t *
575 xfs_trans_alloc(
576 	xfs_mount_t	*mp,
577 	uint		type)
578 {
579 	xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
580 	return _xfs_trans_alloc(mp, type, KM_SLEEP);
581 }
582 
583 xfs_trans_t *
584 _xfs_trans_alloc(
585 	xfs_mount_t	*mp,
586 	uint		type,
587 	xfs_km_flags_t	memflags)
588 {
589 	xfs_trans_t	*tp;
590 
591 	atomic_inc(&mp->m_active_trans);
592 
593 	tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
594 	tp->t_magic = XFS_TRANS_MAGIC;
595 	tp->t_type = type;
596 	tp->t_mountp = mp;
597 	INIT_LIST_HEAD(&tp->t_items);
598 	INIT_LIST_HEAD(&tp->t_busy);
599 	return tp;
600 }
601 
602 /*
603  * Free the transaction structure.  If there is more clean up
604  * to do when the structure is freed, add it here.
605  */
606 STATIC void
607 xfs_trans_free(
608 	struct xfs_trans	*tp)
609 {
610 	xfs_extent_busy_sort(&tp->t_busy);
611 	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
612 
613 	atomic_dec(&tp->t_mountp->m_active_trans);
614 	xfs_trans_free_dqinfo(tp);
615 	kmem_zone_free(xfs_trans_zone, tp);
616 }
617 
618 /*
619  * This is called to create a new transaction which will share the
620  * permanent log reservation of the given transaction.  The remaining
621  * unused block and rt extent reservations are also inherited.  This
622  * implies that the original transaction is no longer allowed to allocate
623  * blocks.  Locks and log items, however, are no inherited.  They must
624  * be added to the new transaction explicitly.
625  */
626 xfs_trans_t *
627 xfs_trans_dup(
628 	xfs_trans_t	*tp)
629 {
630 	xfs_trans_t	*ntp;
631 
632 	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
633 
634 	/*
635 	 * Initialize the new transaction structure.
636 	 */
637 	ntp->t_magic = XFS_TRANS_MAGIC;
638 	ntp->t_type = tp->t_type;
639 	ntp->t_mountp = tp->t_mountp;
640 	INIT_LIST_HEAD(&ntp->t_items);
641 	INIT_LIST_HEAD(&ntp->t_busy);
642 
643 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
644 	ASSERT(tp->t_ticket != NULL);
645 
646 	ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
647 	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
648 	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
649 	tp->t_blk_res = tp->t_blk_res_used;
650 	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
651 	tp->t_rtx_res = tp->t_rtx_res_used;
652 	ntp->t_pflags = tp->t_pflags;
653 
654 	xfs_trans_dup_dqinfo(tp, ntp);
655 
656 	atomic_inc(&tp->t_mountp->m_active_trans);
657 	return ntp;
658 }
659 
660 /*
661  * This is called to reserve free disk blocks and log space for the
662  * given transaction.  This must be done before allocating any resources
663  * within the transaction.
664  *
665  * This will return ENOSPC if there are not enough blocks available.
666  * It will sleep waiting for available log space.
667  * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
668  * is used by long running transactions.  If any one of the reservations
669  * fails then they will all be backed out.
670  *
671  * This does not do quota reservations. That typically is done by the
672  * caller afterwards.
673  */
674 int
675 xfs_trans_reserve(
676 	xfs_trans_t	*tp,
677 	uint		blocks,
678 	uint		logspace,
679 	uint		rtextents,
680 	uint		flags,
681 	uint		logcount)
682 {
683 	int		error = 0;
684 	int		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
685 
686 	/* Mark this thread as being in a transaction */
687 	current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
688 
689 	/*
690 	 * Attempt to reserve the needed disk blocks by decrementing
691 	 * the number needed from the number available.  This will
692 	 * fail if the count would go below zero.
693 	 */
694 	if (blocks > 0) {
695 		error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
696 					  -((int64_t)blocks), rsvd);
697 		if (error != 0) {
698 			current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
699 			return (XFS_ERROR(ENOSPC));
700 		}
701 		tp->t_blk_res += blocks;
702 	}
703 
704 	/*
705 	 * Reserve the log space needed for this transaction.
706 	 */
707 	if (logspace > 0) {
708 		bool	permanent = false;
709 
710 		ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace);
711 		ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount);
712 
713 		if (flags & XFS_TRANS_PERM_LOG_RES) {
714 			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
715 			permanent = true;
716 		} else {
717 			ASSERT(tp->t_ticket == NULL);
718 			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
719 		}
720 
721 		if (tp->t_ticket != NULL) {
722 			ASSERT(flags & XFS_TRANS_PERM_LOG_RES);
723 			error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
724 		} else {
725 			error = xfs_log_reserve(tp->t_mountp, logspace,
726 						logcount, &tp->t_ticket,
727 						XFS_TRANSACTION, permanent,
728 						tp->t_type);
729 		}
730 
731 		if (error)
732 			goto undo_blocks;
733 
734 		tp->t_log_res = logspace;
735 		tp->t_log_count = logcount;
736 	}
737 
738 	/*
739 	 * Attempt to reserve the needed realtime extents by decrementing
740 	 * the number needed from the number available.  This will
741 	 * fail if the count would go below zero.
742 	 */
743 	if (rtextents > 0) {
744 		error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
745 					  -((int64_t)rtextents), rsvd);
746 		if (error) {
747 			error = XFS_ERROR(ENOSPC);
748 			goto undo_log;
749 		}
750 		tp->t_rtx_res += rtextents;
751 	}
752 
753 	return 0;
754 
755 	/*
756 	 * Error cases jump to one of these labels to undo any
757 	 * reservations which have already been performed.
758 	 */
759 undo_log:
760 	if (logspace > 0) {
761 		int		log_flags;
762 
763 		if (flags & XFS_TRANS_PERM_LOG_RES) {
764 			log_flags = XFS_LOG_REL_PERM_RESERV;
765 		} else {
766 			log_flags = 0;
767 		}
768 		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
769 		tp->t_ticket = NULL;
770 		tp->t_log_res = 0;
771 		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
772 	}
773 
774 undo_blocks:
775 	if (blocks > 0) {
776 		xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
777 					 (int64_t)blocks, rsvd);
778 		tp->t_blk_res = 0;
779 	}
780 
781 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
782 
783 	return error;
784 }
785 
786 /*
787  * Record the indicated change to the given field for application
788  * to the file system's superblock when the transaction commits.
789  * For now, just store the change in the transaction structure.
790  *
791  * Mark the transaction structure to indicate that the superblock
792  * needs to be updated before committing.
793  *
794  * Because we may not be keeping track of allocated/free inodes and
795  * used filesystem blocks in the superblock, we do not mark the
796  * superblock dirty in this transaction if we modify these fields.
797  * We still need to update the transaction deltas so that they get
798  * applied to the incore superblock, but we don't want them to
799  * cause the superblock to get locked and logged if these are the
800  * only fields in the superblock that the transaction modifies.
801  */
802 void
803 xfs_trans_mod_sb(
804 	xfs_trans_t	*tp,
805 	uint		field,
806 	int64_t		delta)
807 {
808 	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
809 	xfs_mount_t	*mp = tp->t_mountp;
810 
811 	switch (field) {
812 	case XFS_TRANS_SB_ICOUNT:
813 		tp->t_icount_delta += delta;
814 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
815 			flags &= ~XFS_TRANS_SB_DIRTY;
816 		break;
817 	case XFS_TRANS_SB_IFREE:
818 		tp->t_ifree_delta += delta;
819 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
820 			flags &= ~XFS_TRANS_SB_DIRTY;
821 		break;
822 	case XFS_TRANS_SB_FDBLOCKS:
823 		/*
824 		 * Track the number of blocks allocated in the
825 		 * transaction.  Make sure it does not exceed the
826 		 * number reserved.
827 		 */
828 		if (delta < 0) {
829 			tp->t_blk_res_used += (uint)-delta;
830 			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
831 		}
832 		tp->t_fdblocks_delta += delta;
833 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
834 			flags &= ~XFS_TRANS_SB_DIRTY;
835 		break;
836 	case XFS_TRANS_SB_RES_FDBLOCKS:
837 		/*
838 		 * The allocation has already been applied to the
839 		 * in-core superblock's counter.  This should only
840 		 * be applied to the on-disk superblock.
841 		 */
842 		ASSERT(delta < 0);
843 		tp->t_res_fdblocks_delta += delta;
844 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
845 			flags &= ~XFS_TRANS_SB_DIRTY;
846 		break;
847 	case XFS_TRANS_SB_FREXTENTS:
848 		/*
849 		 * Track the number of blocks allocated in the
850 		 * transaction.  Make sure it does not exceed the
851 		 * number reserved.
852 		 */
853 		if (delta < 0) {
854 			tp->t_rtx_res_used += (uint)-delta;
855 			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
856 		}
857 		tp->t_frextents_delta += delta;
858 		break;
859 	case XFS_TRANS_SB_RES_FREXTENTS:
860 		/*
861 		 * The allocation has already been applied to the
862 		 * in-core superblock's counter.  This should only
863 		 * be applied to the on-disk superblock.
864 		 */
865 		ASSERT(delta < 0);
866 		tp->t_res_frextents_delta += delta;
867 		break;
868 	case XFS_TRANS_SB_DBLOCKS:
869 		ASSERT(delta > 0);
870 		tp->t_dblocks_delta += delta;
871 		break;
872 	case XFS_TRANS_SB_AGCOUNT:
873 		ASSERT(delta > 0);
874 		tp->t_agcount_delta += delta;
875 		break;
876 	case XFS_TRANS_SB_IMAXPCT:
877 		tp->t_imaxpct_delta += delta;
878 		break;
879 	case XFS_TRANS_SB_REXTSIZE:
880 		tp->t_rextsize_delta += delta;
881 		break;
882 	case XFS_TRANS_SB_RBMBLOCKS:
883 		tp->t_rbmblocks_delta += delta;
884 		break;
885 	case XFS_TRANS_SB_RBLOCKS:
886 		tp->t_rblocks_delta += delta;
887 		break;
888 	case XFS_TRANS_SB_REXTENTS:
889 		tp->t_rextents_delta += delta;
890 		break;
891 	case XFS_TRANS_SB_REXTSLOG:
892 		tp->t_rextslog_delta += delta;
893 		break;
894 	default:
895 		ASSERT(0);
896 		return;
897 	}
898 
899 	tp->t_flags |= flags;
900 }
901 
902 /*
903  * xfs_trans_apply_sb_deltas() is called from the commit code
904  * to bring the superblock buffer into the current transaction
905  * and modify it as requested by earlier calls to xfs_trans_mod_sb().
906  *
907  * For now we just look at each field allowed to change and change
908  * it if necessary.
909  */
910 STATIC void
911 xfs_trans_apply_sb_deltas(
912 	xfs_trans_t	*tp)
913 {
914 	xfs_dsb_t	*sbp;
915 	xfs_buf_t	*bp;
916 	int		whole = 0;
917 
918 	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
919 	sbp = XFS_BUF_TO_SBP(bp);
920 
921 	/*
922 	 * Check that superblock mods match the mods made to AGF counters.
923 	 */
924 	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
925 	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
926 		tp->t_ag_btree_delta));
927 
928 	/*
929 	 * Only update the superblock counters if we are logging them
930 	 */
931 	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
932 		if (tp->t_icount_delta)
933 			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
934 		if (tp->t_ifree_delta)
935 			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
936 		if (tp->t_fdblocks_delta)
937 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
938 		if (tp->t_res_fdblocks_delta)
939 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
940 	}
941 
942 	if (tp->t_frextents_delta)
943 		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
944 	if (tp->t_res_frextents_delta)
945 		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
946 
947 	if (tp->t_dblocks_delta) {
948 		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
949 		whole = 1;
950 	}
951 	if (tp->t_agcount_delta) {
952 		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
953 		whole = 1;
954 	}
955 	if (tp->t_imaxpct_delta) {
956 		sbp->sb_imax_pct += tp->t_imaxpct_delta;
957 		whole = 1;
958 	}
959 	if (tp->t_rextsize_delta) {
960 		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
961 		whole = 1;
962 	}
963 	if (tp->t_rbmblocks_delta) {
964 		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
965 		whole = 1;
966 	}
967 	if (tp->t_rblocks_delta) {
968 		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
969 		whole = 1;
970 	}
971 	if (tp->t_rextents_delta) {
972 		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
973 		whole = 1;
974 	}
975 	if (tp->t_rextslog_delta) {
976 		sbp->sb_rextslog += tp->t_rextslog_delta;
977 		whole = 1;
978 	}
979 
980 	if (whole)
981 		/*
982 		 * Log the whole thing, the fields are noncontiguous.
983 		 */
984 		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
985 	else
986 		/*
987 		 * Since all the modifiable fields are contiguous, we
988 		 * can get away with this.
989 		 */
990 		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
991 				  offsetof(xfs_dsb_t, sb_frextents) +
992 				  sizeof(sbp->sb_frextents) - 1);
993 }
994 
995 /*
996  * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
997  * and apply superblock counter changes to the in-core superblock.  The
998  * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
999  * applied to the in-core superblock.  The idea is that that has already been
1000  * done.
1001  *
1002  * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
1003  * However, we have to ensure that we only modify each superblock field only
1004  * once because the application of the delta values may not be atomic. That can
1005  * lead to ENOSPC races occurring if we have two separate modifcations of the
1006  * free space counter to put back the entire reservation and then take away
1007  * what we used.
1008  *
1009  * If we are not logging superblock counters, then the inode allocated/free and
1010  * used block counts are not updated in the on disk superblock. In this case,
1011  * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1012  * still need to update the incore superblock with the changes.
1013  */
1014 void
1015 xfs_trans_unreserve_and_mod_sb(
1016 	xfs_trans_t	*tp)
1017 {
1018 	xfs_mod_sb_t	msb[9];	/* If you add cases, add entries */
1019 	xfs_mod_sb_t	*msbp;
1020 	xfs_mount_t	*mp = tp->t_mountp;
1021 	/* REFERENCED */
1022 	int		error;
1023 	int		rsvd;
1024 	int64_t		blkdelta = 0;
1025 	int64_t		rtxdelta = 0;
1026 	int64_t		idelta = 0;
1027 	int64_t		ifreedelta = 0;
1028 
1029 	msbp = msb;
1030 	rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1031 
1032 	/* calculate deltas */
1033 	if (tp->t_blk_res > 0)
1034 		blkdelta = tp->t_blk_res;
1035 	if ((tp->t_fdblocks_delta != 0) &&
1036 	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1037 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1038 	        blkdelta += tp->t_fdblocks_delta;
1039 
1040 	if (tp->t_rtx_res > 0)
1041 		rtxdelta = tp->t_rtx_res;
1042 	if ((tp->t_frextents_delta != 0) &&
1043 	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
1044 		rtxdelta += tp->t_frextents_delta;
1045 
1046 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1047 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1048 		idelta = tp->t_icount_delta;
1049 		ifreedelta = tp->t_ifree_delta;
1050 	}
1051 
1052 	/* apply the per-cpu counters */
1053 	if (blkdelta) {
1054 		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1055 						 blkdelta, rsvd);
1056 		if (error)
1057 			goto out;
1058 	}
1059 
1060 	if (idelta) {
1061 		error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1062 						 idelta, rsvd);
1063 		if (error)
1064 			goto out_undo_fdblocks;
1065 	}
1066 
1067 	if (ifreedelta) {
1068 		error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1069 						 ifreedelta, rsvd);
1070 		if (error)
1071 			goto out_undo_icount;
1072 	}
1073 
1074 	/* apply remaining deltas */
1075 	if (rtxdelta != 0) {
1076 		msbp->msb_field = XFS_SBS_FREXTENTS;
1077 		msbp->msb_delta = rtxdelta;
1078 		msbp++;
1079 	}
1080 
1081 	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
1082 		if (tp->t_dblocks_delta != 0) {
1083 			msbp->msb_field = XFS_SBS_DBLOCKS;
1084 			msbp->msb_delta = tp->t_dblocks_delta;
1085 			msbp++;
1086 		}
1087 		if (tp->t_agcount_delta != 0) {
1088 			msbp->msb_field = XFS_SBS_AGCOUNT;
1089 			msbp->msb_delta = tp->t_agcount_delta;
1090 			msbp++;
1091 		}
1092 		if (tp->t_imaxpct_delta != 0) {
1093 			msbp->msb_field = XFS_SBS_IMAX_PCT;
1094 			msbp->msb_delta = tp->t_imaxpct_delta;
1095 			msbp++;
1096 		}
1097 		if (tp->t_rextsize_delta != 0) {
1098 			msbp->msb_field = XFS_SBS_REXTSIZE;
1099 			msbp->msb_delta = tp->t_rextsize_delta;
1100 			msbp++;
1101 		}
1102 		if (tp->t_rbmblocks_delta != 0) {
1103 			msbp->msb_field = XFS_SBS_RBMBLOCKS;
1104 			msbp->msb_delta = tp->t_rbmblocks_delta;
1105 			msbp++;
1106 		}
1107 		if (tp->t_rblocks_delta != 0) {
1108 			msbp->msb_field = XFS_SBS_RBLOCKS;
1109 			msbp->msb_delta = tp->t_rblocks_delta;
1110 			msbp++;
1111 		}
1112 		if (tp->t_rextents_delta != 0) {
1113 			msbp->msb_field = XFS_SBS_REXTENTS;
1114 			msbp->msb_delta = tp->t_rextents_delta;
1115 			msbp++;
1116 		}
1117 		if (tp->t_rextslog_delta != 0) {
1118 			msbp->msb_field = XFS_SBS_REXTSLOG;
1119 			msbp->msb_delta = tp->t_rextslog_delta;
1120 			msbp++;
1121 		}
1122 	}
1123 
1124 	/*
1125 	 * If we need to change anything, do it.
1126 	 */
1127 	if (msbp > msb) {
1128 		error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1129 			(uint)(msbp - msb), rsvd);
1130 		if (error)
1131 			goto out_undo_ifreecount;
1132 	}
1133 
1134 	return;
1135 
1136 out_undo_ifreecount:
1137 	if (ifreedelta)
1138 		xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1139 out_undo_icount:
1140 	if (idelta)
1141 		xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1142 out_undo_fdblocks:
1143 	if (blkdelta)
1144 		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1145 out:
1146 	ASSERT(error == 0);
1147 	return;
1148 }
1149 
1150 /*
1151  * Add the given log item to the transaction's list of log items.
1152  *
1153  * The log item will now point to its new descriptor with its li_desc field.
1154  */
1155 void
1156 xfs_trans_add_item(
1157 	struct xfs_trans	*tp,
1158 	struct xfs_log_item	*lip)
1159 {
1160 	struct xfs_log_item_desc *lidp;
1161 
1162 	ASSERT(lip->li_mountp == tp->t_mountp);
1163 	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
1164 
1165 	lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
1166 
1167 	lidp->lid_item = lip;
1168 	lidp->lid_flags = 0;
1169 	list_add_tail(&lidp->lid_trans, &tp->t_items);
1170 
1171 	lip->li_desc = lidp;
1172 }
1173 
1174 STATIC void
1175 xfs_trans_free_item_desc(
1176 	struct xfs_log_item_desc *lidp)
1177 {
1178 	list_del_init(&lidp->lid_trans);
1179 	kmem_zone_free(xfs_log_item_desc_zone, lidp);
1180 }
1181 
1182 /*
1183  * Unlink and free the given descriptor.
1184  */
1185 void
1186 xfs_trans_del_item(
1187 	struct xfs_log_item	*lip)
1188 {
1189 	xfs_trans_free_item_desc(lip->li_desc);
1190 	lip->li_desc = NULL;
1191 }
1192 
1193 /*
1194  * Unlock all of the items of a transaction and free all the descriptors
1195  * of that transaction.
1196  */
1197 void
1198 xfs_trans_free_items(
1199 	struct xfs_trans	*tp,
1200 	xfs_lsn_t		commit_lsn,
1201 	int			flags)
1202 {
1203 	struct xfs_log_item_desc *lidp, *next;
1204 
1205 	list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1206 		struct xfs_log_item	*lip = lidp->lid_item;
1207 
1208 		lip->li_desc = NULL;
1209 
1210 		if (commit_lsn != NULLCOMMITLSN)
1211 			IOP_COMMITTING(lip, commit_lsn);
1212 		if (flags & XFS_TRANS_ABORT)
1213 			lip->li_flags |= XFS_LI_ABORTED;
1214 		IOP_UNLOCK(lip);
1215 
1216 		xfs_trans_free_item_desc(lidp);
1217 	}
1218 }
1219 
1220 static inline void
1221 xfs_log_item_batch_insert(
1222 	struct xfs_ail		*ailp,
1223 	struct xfs_ail_cursor	*cur,
1224 	struct xfs_log_item	**log_items,
1225 	int			nr_items,
1226 	xfs_lsn_t		commit_lsn)
1227 {
1228 	int	i;
1229 
1230 	spin_lock(&ailp->xa_lock);
1231 	/* xfs_trans_ail_update_bulk drops ailp->xa_lock */
1232 	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
1233 
1234 	for (i = 0; i < nr_items; i++)
1235 		IOP_UNPIN(log_items[i], 0);
1236 }
1237 
1238 /*
1239  * Bulk operation version of xfs_trans_committed that takes a log vector of
1240  * items to insert into the AIL. This uses bulk AIL insertion techniques to
1241  * minimise lock traffic.
1242  *
1243  * If we are called with the aborted flag set, it is because a log write during
1244  * a CIL checkpoint commit has failed. In this case, all the items in the
1245  * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
1246  * means that checkpoint commit abort handling is treated exactly the same
1247  * as an iclog write error even though we haven't started any IO yet. Hence in
1248  * this case all we need to do is IOP_COMMITTED processing, followed by an
1249  * IOP_UNPIN(aborted) call.
1250  *
1251  * The AIL cursor is used to optimise the insert process. If commit_lsn is not
1252  * at the end of the AIL, the insert cursor avoids the need to walk
1253  * the AIL to find the insertion point on every xfs_log_item_batch_insert()
1254  * call. This saves a lot of needless list walking and is a net win, even
1255  * though it slightly increases that amount of AIL lock traffic to set it up
1256  * and tear it down.
1257  */
1258 void
1259 xfs_trans_committed_bulk(
1260 	struct xfs_ail		*ailp,
1261 	struct xfs_log_vec	*log_vector,
1262 	xfs_lsn_t		commit_lsn,
1263 	int			aborted)
1264 {
1265 #define LOG_ITEM_BATCH_SIZE	32
1266 	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
1267 	struct xfs_log_vec	*lv;
1268 	struct xfs_ail_cursor	cur;
1269 	int			i = 0;
1270 
1271 	spin_lock(&ailp->xa_lock);
1272 	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
1273 	spin_unlock(&ailp->xa_lock);
1274 
1275 	/* unpin all the log items */
1276 	for (lv = log_vector; lv; lv = lv->lv_next ) {
1277 		struct xfs_log_item	*lip = lv->lv_item;
1278 		xfs_lsn_t		item_lsn;
1279 
1280 		if (aborted)
1281 			lip->li_flags |= XFS_LI_ABORTED;
1282 		item_lsn = IOP_COMMITTED(lip, commit_lsn);
1283 
1284 		/* item_lsn of -1 means the item needs no further processing */
1285 		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1286 			continue;
1287 
1288 		/*
1289 		 * if we are aborting the operation, no point in inserting the
1290 		 * object into the AIL as we are in a shutdown situation.
1291 		 */
1292 		if (aborted) {
1293 			ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
1294 			IOP_UNPIN(lip, 1);
1295 			continue;
1296 		}
1297 
1298 		if (item_lsn != commit_lsn) {
1299 
1300 			/*
1301 			 * Not a bulk update option due to unusual item_lsn.
1302 			 * Push into AIL immediately, rechecking the lsn once
1303 			 * we have the ail lock. Then unpin the item. This does
1304 			 * not affect the AIL cursor the bulk insert path is
1305 			 * using.
1306 			 */
1307 			spin_lock(&ailp->xa_lock);
1308 			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1309 				xfs_trans_ail_update(ailp, lip, item_lsn);
1310 			else
1311 				spin_unlock(&ailp->xa_lock);
1312 			IOP_UNPIN(lip, 0);
1313 			continue;
1314 		}
1315 
1316 		/* Item is a candidate for bulk AIL insert.  */
1317 		log_items[i++] = lv->lv_item;
1318 		if (i >= LOG_ITEM_BATCH_SIZE) {
1319 			xfs_log_item_batch_insert(ailp, &cur, log_items,
1320 					LOG_ITEM_BATCH_SIZE, commit_lsn);
1321 			i = 0;
1322 		}
1323 	}
1324 
1325 	/* make sure we insert the remainder! */
1326 	if (i)
1327 		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
1328 
1329 	spin_lock(&ailp->xa_lock);
1330 	xfs_trans_ail_cursor_done(ailp, &cur);
1331 	spin_unlock(&ailp->xa_lock);
1332 }
1333 
1334 /*
1335  * Commit the given transaction to the log.
1336  *
1337  * XFS disk error handling mechanism is not based on a typical
1338  * transaction abort mechanism. Logically after the filesystem
1339  * gets marked 'SHUTDOWN', we can't let any new transactions
1340  * be durable - ie. committed to disk - because some metadata might
1341  * be inconsistent. In such cases, this returns an error, and the
1342  * caller may assume that all locked objects joined to the transaction
1343  * have already been unlocked as if the commit had succeeded.
1344  * Do not reference the transaction structure after this call.
1345  */
1346 int
1347 xfs_trans_commit(
1348 	struct xfs_trans	*tp,
1349 	uint			flags)
1350 {
1351 	struct xfs_mount	*mp = tp->t_mountp;
1352 	xfs_lsn_t		commit_lsn = -1;
1353 	int			error = 0;
1354 	int			log_flags = 0;
1355 	int			sync = tp->t_flags & XFS_TRANS_SYNC;
1356 
1357 	/*
1358 	 * Determine whether this commit is releasing a permanent
1359 	 * log reservation or not.
1360 	 */
1361 	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1362 		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1363 		log_flags = XFS_LOG_REL_PERM_RESERV;
1364 	}
1365 
1366 	/*
1367 	 * If there is nothing to be logged by the transaction,
1368 	 * then unlock all of the items associated with the
1369 	 * transaction and free the transaction structure.
1370 	 * Also make sure to return any reserved blocks to
1371 	 * the free pool.
1372 	 */
1373 	if (!(tp->t_flags & XFS_TRANS_DIRTY))
1374 		goto out_unreserve;
1375 
1376 	if (XFS_FORCED_SHUTDOWN(mp)) {
1377 		error = XFS_ERROR(EIO);
1378 		goto out_unreserve;
1379 	}
1380 
1381 	ASSERT(tp->t_ticket != NULL);
1382 
1383 	/*
1384 	 * If we need to update the superblock, then do it now.
1385 	 */
1386 	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1387 		xfs_trans_apply_sb_deltas(tp);
1388 	xfs_trans_apply_dquot_deltas(tp);
1389 
1390 	error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags);
1391 	if (error == ENOMEM) {
1392 		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1393 		error = XFS_ERROR(EIO);
1394 		goto out_unreserve;
1395 	}
1396 
1397 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1398 	xfs_trans_free(tp);
1399 
1400 	/*
1401 	 * If the transaction needs to be synchronous, then force the
1402 	 * log out now and wait for it.
1403 	 */
1404 	if (sync) {
1405 		if (!error) {
1406 			error = _xfs_log_force_lsn(mp, commit_lsn,
1407 				      XFS_LOG_SYNC, NULL);
1408 		}
1409 		XFS_STATS_INC(xs_trans_sync);
1410 	} else {
1411 		XFS_STATS_INC(xs_trans_async);
1412 	}
1413 
1414 	return error;
1415 
1416 out_unreserve:
1417 	xfs_trans_unreserve_and_mod_sb(tp);
1418 
1419 	/*
1420 	 * It is indeed possible for the transaction to be not dirty but
1421 	 * the dqinfo portion to be.  All that means is that we have some
1422 	 * (non-persistent) quota reservations that need to be unreserved.
1423 	 */
1424 	xfs_trans_unreserve_and_mod_dquots(tp);
1425 	if (tp->t_ticket) {
1426 		commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1427 		if (commit_lsn == -1 && !error)
1428 			error = XFS_ERROR(EIO);
1429 	}
1430 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1431 	xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
1432 	xfs_trans_free(tp);
1433 
1434 	XFS_STATS_INC(xs_trans_empty);
1435 	return error;
1436 }
1437 
1438 /*
1439  * Unlock all of the transaction's items and free the transaction.
1440  * The transaction must not have modified any of its items, because
1441  * there is no way to restore them to their previous state.
1442  *
1443  * If the transaction has made a log reservation, make sure to release
1444  * it as well.
1445  */
1446 void
1447 xfs_trans_cancel(
1448 	xfs_trans_t		*tp,
1449 	int			flags)
1450 {
1451 	int			log_flags;
1452 	xfs_mount_t		*mp = tp->t_mountp;
1453 
1454 	/*
1455 	 * See if the caller is being too lazy to figure out if
1456 	 * the transaction really needs an abort.
1457 	 */
1458 	if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1459 		flags &= ~XFS_TRANS_ABORT;
1460 	/*
1461 	 * See if the caller is relying on us to shut down the
1462 	 * filesystem.  This happens in paths where we detect
1463 	 * corruption and decide to give up.
1464 	 */
1465 	if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1466 		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1467 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1468 	}
1469 #ifdef DEBUG
1470 	if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1471 		struct xfs_log_item_desc *lidp;
1472 
1473 		list_for_each_entry(lidp, &tp->t_items, lid_trans)
1474 			ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1475 	}
1476 #endif
1477 	xfs_trans_unreserve_and_mod_sb(tp);
1478 	xfs_trans_unreserve_and_mod_dquots(tp);
1479 
1480 	if (tp->t_ticket) {
1481 		if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1482 			ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1483 			log_flags = XFS_LOG_REL_PERM_RESERV;
1484 		} else {
1485 			log_flags = 0;
1486 		}
1487 		xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1488 	}
1489 
1490 	/* mark this thread as no longer being in a transaction */
1491 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1492 
1493 	xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1494 	xfs_trans_free(tp);
1495 }
1496 
1497 /*
1498  * Roll from one trans in the sequence of PERMANENT transactions to
1499  * the next: permanent transactions are only flushed out when
1500  * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1501  * as possible to let chunks of it go to the log. So we commit the
1502  * chunk we've been working on and get a new transaction to continue.
1503  */
1504 int
1505 xfs_trans_roll(
1506 	struct xfs_trans	**tpp,
1507 	struct xfs_inode	*dp)
1508 {
1509 	struct xfs_trans	*trans;
1510 	unsigned int		logres, count;
1511 	int			error;
1512 
1513 	/*
1514 	 * Ensure that the inode is always logged.
1515 	 */
1516 	trans = *tpp;
1517 	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1518 
1519 	/*
1520 	 * Copy the critical parameters from one trans to the next.
1521 	 */
1522 	logres = trans->t_log_res;
1523 	count = trans->t_log_count;
1524 	*tpp = xfs_trans_dup(trans);
1525 
1526 	/*
1527 	 * Commit the current transaction.
1528 	 * If this commit failed, then it'd just unlock those items that
1529 	 * are not marked ihold. That also means that a filesystem shutdown
1530 	 * is in progress. The caller takes the responsibility to cancel
1531 	 * the duplicate transaction that gets returned.
1532 	 */
1533 	error = xfs_trans_commit(trans, 0);
1534 	if (error)
1535 		return (error);
1536 
1537 	trans = *tpp;
1538 
1539 	/*
1540 	 * transaction commit worked ok so we can drop the extra ticket
1541 	 * reference that we gained in xfs_trans_dup()
1542 	 */
1543 	xfs_log_ticket_put(trans->t_ticket);
1544 
1545 
1546 	/*
1547 	 * Reserve space in the log for th next transaction.
1548 	 * This also pushes items in the "AIL", the list of logged items,
1549 	 * out to disk if they are taking up space at the tail of the log
1550 	 * that we want to use.  This requires that either nothing be locked
1551 	 * across this call, or that anything that is locked be logged in
1552 	 * the prior and the next transactions.
1553 	 */
1554 	error = xfs_trans_reserve(trans, 0, logres, 0,
1555 				  XFS_TRANS_PERM_LOG_RES, count);
1556 	/*
1557 	 *  Ensure that the inode is in the new transaction and locked.
1558 	 */
1559 	if (error)
1560 		return error;
1561 
1562 	xfs_trans_ijoin(trans, dp, 0);
1563 	return 0;
1564 }
1565