xref: /openbmc/linux/fs/xfs/xfs_trans.c (revision 73eb94a0)
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_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_inum.h"
25 #include "xfs_trans.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_error.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
39 #include "xfs_bmap.h"
40 #include "xfs_quota.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_trace.h"
45 
46 kmem_zone_t	*xfs_trans_zone;
47 kmem_zone_t	*xfs_log_item_desc_zone;
48 
49 
50 /*
51  * Various log reservation values.
52  *
53  * These are based on the size of the file system block because that is what
54  * most transactions manipulate.  Each adds in an additional 128 bytes per
55  * item logged to try to account for the overhead of the transaction mechanism.
56  *
57  * Note:  Most of the reservations underestimate the number of allocation
58  * groups into which they could free extents in the xfs_bmap_finish() call.
59  * This is because the number in the worst case is quite high and quite
60  * unusual.  In order to fix this we need to change xfs_bmap_finish() to free
61  * extents in only a single AG at a time.  This will require changes to the
62  * EFI code as well, however, so that the EFI for the extents not freed is
63  * logged again in each transaction.  See SGI PV #261917.
64  *
65  * Reservation functions here avoid a huge stack in xfs_trans_init due to
66  * register overflow from temporaries in the calculations.
67  */
68 
69 
70 /*
71  * In a write transaction we can allocate a maximum of 2
72  * extents.  This gives:
73  *    the inode getting the new extents: inode size
74  *    the inode's bmap btree: max depth * block size
75  *    the agfs of the ags from which the extents are allocated: 2 * sector
76  *    the superblock free block counter: sector size
77  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
78  * And the bmap_finish transaction can free bmap blocks in a join:
79  *    the agfs of the ags containing the blocks: 2 * sector size
80  *    the agfls of the ags containing the blocks: 2 * sector size
81  *    the super block free block counter: sector size
82  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
83  */
84 STATIC uint
85 xfs_calc_write_reservation(
86 	struct xfs_mount	*mp)
87 {
88 	return XFS_DQUOT_LOGRES(mp) +
89 		MAX((mp->m_sb.sb_inodesize +
90 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
91 		     2 * mp->m_sb.sb_sectsize +
92 		     mp->m_sb.sb_sectsize +
93 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
94 		     128 * (4 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
95 			    XFS_ALLOCFREE_LOG_COUNT(mp, 2))),
96 		    (2 * mp->m_sb.sb_sectsize +
97 		     2 * mp->m_sb.sb_sectsize +
98 		     mp->m_sb.sb_sectsize +
99 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
100 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
101 }
102 
103 /*
104  * In truncating a file we free up to two extents at once.  We can modify:
105  *    the inode being truncated: inode size
106  *    the inode's bmap btree: (max depth + 1) * block size
107  * And the bmap_finish transaction can free the blocks and bmap blocks:
108  *    the agf for each of the ags: 4 * sector size
109  *    the agfl for each of the ags: 4 * sector size
110  *    the super block to reflect the freed blocks: sector size
111  *    worst case split in allocation btrees per extent assuming 4 extents:
112  *		4 exts * 2 trees * (2 * max depth - 1) * block size
113  *    the inode btree: max depth * blocksize
114  *    the allocation btrees: 2 trees * (max depth - 1) * block size
115  */
116 STATIC uint
117 xfs_calc_itruncate_reservation(
118 	struct xfs_mount	*mp)
119 {
120 	return XFS_DQUOT_LOGRES(mp) +
121 		MAX((mp->m_sb.sb_inodesize +
122 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1) +
123 		     128 * (2 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
124 		    (4 * mp->m_sb.sb_sectsize +
125 		     4 * mp->m_sb.sb_sectsize +
126 		     mp->m_sb.sb_sectsize +
127 		     XFS_ALLOCFREE_LOG_RES(mp, 4) +
128 		     128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)) +
129 		     128 * 5 +
130 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
131 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
132 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
133 }
134 
135 /*
136  * In renaming a files we can modify:
137  *    the four inodes involved: 4 * inode size
138  *    the two directory btrees: 2 * (max depth + v2) * dir block size
139  *    the two directory bmap btrees: 2 * max depth * block size
140  * And the bmap_finish transaction can free dir and bmap blocks (two sets
141  *	of bmap blocks) giving:
142  *    the agf for the ags in which the blocks live: 3 * sector size
143  *    the agfl for the ags in which the blocks live: 3 * sector size
144  *    the superblock for the free block count: sector size
145  *    the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
146  */
147 STATIC uint
148 xfs_calc_rename_reservation(
149 	struct xfs_mount	*mp)
150 {
151 	return XFS_DQUOT_LOGRES(mp) +
152 		MAX((4 * mp->m_sb.sb_inodesize +
153 		     2 * XFS_DIROP_LOG_RES(mp) +
154 		     128 * (4 + 2 * XFS_DIROP_LOG_COUNT(mp))),
155 		    (3 * mp->m_sb.sb_sectsize +
156 		     3 * mp->m_sb.sb_sectsize +
157 		     mp->m_sb.sb_sectsize +
158 		     XFS_ALLOCFREE_LOG_RES(mp, 3) +
159 		     128 * (7 + XFS_ALLOCFREE_LOG_COUNT(mp, 3))));
160 }
161 
162 /*
163  * For creating a link to an inode:
164  *    the parent directory inode: inode size
165  *    the linked inode: inode size
166  *    the directory btree could split: (max depth + v2) * dir block size
167  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
168  * And the bmap_finish transaction can free some bmap blocks giving:
169  *    the agf for the ag in which the blocks live: sector size
170  *    the agfl for the ag in which the blocks live: sector size
171  *    the superblock for the free block count: sector size
172  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
173  */
174 STATIC uint
175 xfs_calc_link_reservation(
176 	struct xfs_mount	*mp)
177 {
178 	return XFS_DQUOT_LOGRES(mp) +
179 		MAX((mp->m_sb.sb_inodesize +
180 		     mp->m_sb.sb_inodesize +
181 		     XFS_DIROP_LOG_RES(mp) +
182 		     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
183 		    (mp->m_sb.sb_sectsize +
184 		     mp->m_sb.sb_sectsize +
185 		     mp->m_sb.sb_sectsize +
186 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
187 		     128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
188 }
189 
190 /*
191  * For removing a directory entry we can modify:
192  *    the parent directory inode: inode size
193  *    the removed inode: inode size
194  *    the directory btree could join: (max depth + v2) * dir block size
195  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
196  * And the bmap_finish transaction can free the dir and bmap blocks giving:
197  *    the agf for the ag in which the blocks live: 2 * sector size
198  *    the agfl for the ag in which the blocks live: 2 * sector size
199  *    the superblock for the free block count: sector size
200  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
201  */
202 STATIC uint
203 xfs_calc_remove_reservation(
204 	struct xfs_mount	*mp)
205 {
206 	return XFS_DQUOT_LOGRES(mp) +
207 		MAX((mp->m_sb.sb_inodesize +
208 		     mp->m_sb.sb_inodesize +
209 		     XFS_DIROP_LOG_RES(mp) +
210 		     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
211 		    (2 * mp->m_sb.sb_sectsize +
212 		     2 * mp->m_sb.sb_sectsize +
213 		     mp->m_sb.sb_sectsize +
214 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
215 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
216 }
217 
218 /*
219  * For symlink we can modify:
220  *    the parent directory inode: inode size
221  *    the new inode: inode size
222  *    the inode btree entry: 1 block
223  *    the directory btree: (max depth + v2) * dir block size
224  *    the directory inode's bmap btree: (max depth + v2) * block size
225  *    the blocks for the symlink: 1 kB
226  * Or in the first xact we allocate some inodes giving:
227  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
228  *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
229  *    the inode btree: max depth * blocksize
230  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
231  */
232 STATIC uint
233 xfs_calc_symlink_reservation(
234 	struct xfs_mount	*mp)
235 {
236 	return XFS_DQUOT_LOGRES(mp) +
237 		MAX((mp->m_sb.sb_inodesize +
238 		     mp->m_sb.sb_inodesize +
239 		     XFS_FSB_TO_B(mp, 1) +
240 		     XFS_DIROP_LOG_RES(mp) +
241 		     1024 +
242 		     128 * (4 + XFS_DIROP_LOG_COUNT(mp))),
243 		    (2 * mp->m_sb.sb_sectsize +
244 		     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
245 		     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
246 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
247 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
248 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
249 }
250 
251 /*
252  * For create we can modify:
253  *    the parent directory inode: inode size
254  *    the new inode: inode size
255  *    the inode btree entry: block size
256  *    the superblock for the nlink flag: sector size
257  *    the directory btree: (max depth + v2) * dir block size
258  *    the directory inode's bmap btree: (max depth + v2) * block size
259  * Or in the first xact we allocate some inodes giving:
260  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
261  *    the superblock for the nlink flag: sector size
262  *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
263  *    the inode btree: max depth * blocksize
264  *    the allocation btrees: 2 trees * (max depth - 1) * block size
265  */
266 STATIC uint
267 xfs_calc_create_reservation(
268 	struct xfs_mount	*mp)
269 {
270 	return XFS_DQUOT_LOGRES(mp) +
271 		MAX((mp->m_sb.sb_inodesize +
272 		     mp->m_sb.sb_inodesize +
273 		     mp->m_sb.sb_sectsize +
274 		     XFS_FSB_TO_B(mp, 1) +
275 		     XFS_DIROP_LOG_RES(mp) +
276 		     128 * (3 + XFS_DIROP_LOG_COUNT(mp))),
277 		    (3 * mp->m_sb.sb_sectsize +
278 		     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
279 		     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
280 		     XFS_ALLOCFREE_LOG_RES(mp, 1) +
281 		     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
282 			    XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
283 }
284 
285 /*
286  * Making a new directory is the same as creating a new file.
287  */
288 STATIC uint
289 xfs_calc_mkdir_reservation(
290 	struct xfs_mount	*mp)
291 {
292 	return xfs_calc_create_reservation(mp);
293 }
294 
295 /*
296  * In freeing an inode we can modify:
297  *    the inode being freed: inode size
298  *    the super block free inode counter: sector size
299  *    the agi hash list and counters: sector size
300  *    the inode btree entry: block size
301  *    the on disk inode before ours in the agi hash list: inode cluster size
302  *    the inode btree: max depth * blocksize
303  *    the allocation btrees: 2 trees * (max depth - 1) * block size
304  */
305 STATIC uint
306 xfs_calc_ifree_reservation(
307 	struct xfs_mount	*mp)
308 {
309 	return XFS_DQUOT_LOGRES(mp) +
310 		mp->m_sb.sb_inodesize +
311 		mp->m_sb.sb_sectsize +
312 		mp->m_sb.sb_sectsize +
313 		XFS_FSB_TO_B(mp, 1) +
314 		MAX((__uint16_t)XFS_FSB_TO_B(mp, 1),
315 		    XFS_INODE_CLUSTER_SIZE(mp)) +
316 		128 * 5 +
317 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
318 		128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
319 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
320 }
321 
322 /*
323  * When only changing the inode we log the inode and possibly the superblock
324  * We also add a bit of slop for the transaction stuff.
325  */
326 STATIC uint
327 xfs_calc_ichange_reservation(
328 	struct xfs_mount	*mp)
329 {
330 	return XFS_DQUOT_LOGRES(mp) +
331 		mp->m_sb.sb_inodesize +
332 		mp->m_sb.sb_sectsize +
333 		512;
334 
335 }
336 
337 /*
338  * Growing the data section of the filesystem.
339  *	superblock
340  *	agi and agf
341  *	allocation btrees
342  */
343 STATIC uint
344 xfs_calc_growdata_reservation(
345 	struct xfs_mount	*mp)
346 {
347 	return mp->m_sb.sb_sectsize * 3 +
348 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
349 		128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1));
350 }
351 
352 /*
353  * Growing the rt section of the filesystem.
354  * In the first set of transactions (ALLOC) we allocate space to the
355  * bitmap or summary files.
356  *	superblock: sector size
357  *	agf of the ag from which the extent is allocated: sector size
358  *	bmap btree for bitmap/summary inode: max depth * blocksize
359  *	bitmap/summary inode: inode size
360  *	allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
361  */
362 STATIC uint
363 xfs_calc_growrtalloc_reservation(
364 	struct xfs_mount	*mp)
365 {
366 	return 2 * mp->m_sb.sb_sectsize +
367 		XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
368 		mp->m_sb.sb_inodesize +
369 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
370 		128 * (3 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
371 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
372 }
373 
374 /*
375  * Growing the rt section of the filesystem.
376  * In the second set of transactions (ZERO) we zero the new metadata blocks.
377  *	one bitmap/summary block: blocksize
378  */
379 STATIC uint
380 xfs_calc_growrtzero_reservation(
381 	struct xfs_mount	*mp)
382 {
383 	return mp->m_sb.sb_blocksize + 128;
384 }
385 
386 /*
387  * Growing the rt section of the filesystem.
388  * In the third set of transactions (FREE) we update metadata without
389  * allocating any new blocks.
390  *	superblock: sector size
391  *	bitmap inode: inode size
392  *	summary inode: inode size
393  *	one bitmap block: blocksize
394  *	summary blocks: new summary size
395  */
396 STATIC uint
397 xfs_calc_growrtfree_reservation(
398 	struct xfs_mount	*mp)
399 {
400 	return mp->m_sb.sb_sectsize +
401 		2 * mp->m_sb.sb_inodesize +
402 		mp->m_sb.sb_blocksize +
403 		mp->m_rsumsize +
404 		128 * 5;
405 }
406 
407 /*
408  * Logging the inode modification timestamp on a synchronous write.
409  *	inode
410  */
411 STATIC uint
412 xfs_calc_swrite_reservation(
413 	struct xfs_mount	*mp)
414 {
415 	return mp->m_sb.sb_inodesize + 128;
416 }
417 
418 /*
419  * Logging the inode mode bits when writing a setuid/setgid file
420  *	inode
421  */
422 STATIC uint
423 xfs_calc_writeid_reservation(xfs_mount_t *mp)
424 {
425 	return mp->m_sb.sb_inodesize + 128;
426 }
427 
428 /*
429  * Converting the inode from non-attributed to attributed.
430  *	the inode being converted: inode size
431  *	agf block and superblock (for block allocation)
432  *	the new block (directory sized)
433  *	bmap blocks for the new directory block
434  *	allocation btrees
435  */
436 STATIC uint
437 xfs_calc_addafork_reservation(
438 	struct xfs_mount	*mp)
439 {
440 	return XFS_DQUOT_LOGRES(mp) +
441 		mp->m_sb.sb_inodesize +
442 		mp->m_sb.sb_sectsize * 2 +
443 		mp->m_dirblksize +
444 		XFS_FSB_TO_B(mp, XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1) +
445 		XFS_ALLOCFREE_LOG_RES(mp, 1) +
446 		128 * (4 + XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1 +
447 		       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
448 }
449 
450 /*
451  * Removing the attribute fork of a file
452  *    the inode being truncated: inode size
453  *    the inode's bmap btree: max depth * block size
454  * And the bmap_finish transaction can free the blocks and bmap blocks:
455  *    the agf for each of the ags: 4 * sector size
456  *    the agfl for each of the ags: 4 * sector size
457  *    the super block to reflect the freed blocks: sector size
458  *    worst case split in allocation btrees per extent assuming 4 extents:
459  *		4 exts * 2 trees * (2 * max depth - 1) * block size
460  */
461 STATIC uint
462 xfs_calc_attrinval_reservation(
463 	struct xfs_mount	*mp)
464 {
465 	return MAX((mp->m_sb.sb_inodesize +
466 		    XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
467 		    128 * (1 + XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))),
468 		   (4 * mp->m_sb.sb_sectsize +
469 		    4 * mp->m_sb.sb_sectsize +
470 		    mp->m_sb.sb_sectsize +
471 		    XFS_ALLOCFREE_LOG_RES(mp, 4) +
472 		    128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4))));
473 }
474 
475 /*
476  * Setting an attribute.
477  *	the inode getting the attribute
478  *	the superblock for allocations
479  *	the agfs extents are allocated from
480  *	the attribute btree * max depth
481  *	the inode allocation btree
482  * Since attribute transaction space is dependent on the size of the attribute,
483  * the calculation is done partially at mount time and partially at runtime.
484  */
485 STATIC uint
486 xfs_calc_attrset_reservation(
487 	struct xfs_mount	*mp)
488 {
489 	return XFS_DQUOT_LOGRES(mp) +
490 		mp->m_sb.sb_inodesize +
491 		mp->m_sb.sb_sectsize +
492 		XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
493 		128 * (2 + XFS_DA_NODE_MAXDEPTH);
494 }
495 
496 /*
497  * Removing an attribute.
498  *    the inode: inode size
499  *    the attribute btree could join: max depth * block size
500  *    the inode bmap btree could join or split: max depth * block size
501  * And the bmap_finish transaction can free the attr blocks freed giving:
502  *    the agf for the ag in which the blocks live: 2 * sector size
503  *    the agfl for the ag in which the blocks live: 2 * sector size
504  *    the superblock for the free block count: sector size
505  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
506  */
507 STATIC uint
508 xfs_calc_attrrm_reservation(
509 	struct xfs_mount	*mp)
510 {
511 	return XFS_DQUOT_LOGRES(mp) +
512 		MAX((mp->m_sb.sb_inodesize +
513 		     XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
514 		     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
515 		     128 * (1 + XFS_DA_NODE_MAXDEPTH +
516 			    XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
517 		    (2 * mp->m_sb.sb_sectsize +
518 		     2 * mp->m_sb.sb_sectsize +
519 		     mp->m_sb.sb_sectsize +
520 		     XFS_ALLOCFREE_LOG_RES(mp, 2) +
521 		     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
522 }
523 
524 /*
525  * Clearing a bad agino number in an agi hash bucket.
526  */
527 STATIC uint
528 xfs_calc_clear_agi_bucket_reservation(
529 	struct xfs_mount	*mp)
530 {
531 	return mp->m_sb.sb_sectsize + 128;
532 }
533 
534 /*
535  * Initialize the precomputed transaction reservation values
536  * in the mount structure.
537  */
538 void
539 xfs_trans_init(
540 	struct xfs_mount	*mp)
541 {
542 	struct xfs_trans_reservations *resp = &mp->m_reservations;
543 
544 	resp->tr_write = xfs_calc_write_reservation(mp);
545 	resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
546 	resp->tr_rename = xfs_calc_rename_reservation(mp);
547 	resp->tr_link = xfs_calc_link_reservation(mp);
548 	resp->tr_remove = xfs_calc_remove_reservation(mp);
549 	resp->tr_symlink = xfs_calc_symlink_reservation(mp);
550 	resp->tr_create = xfs_calc_create_reservation(mp);
551 	resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
552 	resp->tr_ifree = xfs_calc_ifree_reservation(mp);
553 	resp->tr_ichange = xfs_calc_ichange_reservation(mp);
554 	resp->tr_growdata = xfs_calc_growdata_reservation(mp);
555 	resp->tr_swrite = xfs_calc_swrite_reservation(mp);
556 	resp->tr_writeid = xfs_calc_writeid_reservation(mp);
557 	resp->tr_addafork = xfs_calc_addafork_reservation(mp);
558 	resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
559 	resp->tr_attrset = xfs_calc_attrset_reservation(mp);
560 	resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
561 	resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
562 	resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
563 	resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
564 	resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
565 }
566 
567 /*
568  * This routine is called to allocate a transaction structure.
569  * The type parameter indicates the type of the transaction.  These
570  * are enumerated in xfs_trans.h.
571  *
572  * Dynamically allocate the transaction structure from the transaction
573  * zone, initialize it, and return it to the caller.
574  */
575 xfs_trans_t *
576 xfs_trans_alloc(
577 	xfs_mount_t	*mp,
578 	uint		type)
579 {
580 	xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
581 	return _xfs_trans_alloc(mp, type, KM_SLEEP);
582 }
583 
584 xfs_trans_t *
585 _xfs_trans_alloc(
586 	xfs_mount_t	*mp,
587 	uint		type,
588 	uint		memflags)
589 {
590 	xfs_trans_t	*tp;
591 
592 	atomic_inc(&mp->m_active_trans);
593 
594 	tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
595 	tp->t_magic = XFS_TRANS_MAGIC;
596 	tp->t_type = type;
597 	tp->t_mountp = mp;
598 	INIT_LIST_HEAD(&tp->t_items);
599 	INIT_LIST_HEAD(&tp->t_busy);
600 	return tp;
601 }
602 
603 /*
604  * Free the transaction structure.  If there is more clean up
605  * to do when the structure is freed, add it here.
606  */
607 STATIC void
608 xfs_trans_free(
609 	struct xfs_trans	*tp)
610 {
611 	xfs_alloc_busy_sort(&tp->t_busy);
612 	xfs_alloc_busy_clear(tp->t_mountp, &tp->t_busy, false);
613 
614 	atomic_dec(&tp->t_mountp->m_active_trans);
615 	xfs_trans_free_dqinfo(tp);
616 	kmem_zone_free(xfs_trans_zone, tp);
617 }
618 
619 /*
620  * This is called to create a new transaction which will share the
621  * permanent log reservation of the given transaction.  The remaining
622  * unused block and rt extent reservations are also inherited.  This
623  * implies that the original transaction is no longer allowed to allocate
624  * blocks.  Locks and log items, however, are no inherited.  They must
625  * be added to the new transaction explicitly.
626  */
627 xfs_trans_t *
628 xfs_trans_dup(
629 	xfs_trans_t	*tp)
630 {
631 	xfs_trans_t	*ntp;
632 
633 	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
634 
635 	/*
636 	 * Initialize the new transaction structure.
637 	 */
638 	ntp->t_magic = XFS_TRANS_MAGIC;
639 	ntp->t_type = tp->t_type;
640 	ntp->t_mountp = tp->t_mountp;
641 	INIT_LIST_HEAD(&ntp->t_items);
642 	INIT_LIST_HEAD(&ntp->t_busy);
643 
644 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
645 	ASSERT(tp->t_ticket != NULL);
646 
647 	ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
648 	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
649 	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
650 	tp->t_blk_res = tp->t_blk_res_used;
651 	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
652 	tp->t_rtx_res = tp->t_rtx_res_used;
653 	ntp->t_pflags = tp->t_pflags;
654 
655 	xfs_trans_dup_dqinfo(tp, ntp);
656 
657 	atomic_inc(&tp->t_mountp->m_active_trans);
658 	return ntp;
659 }
660 
661 /*
662  * This is called to reserve free disk blocks and log space for the
663  * given transaction.  This must be done before allocating any resources
664  * within the transaction.
665  *
666  * This will return ENOSPC if there are not enough blocks available.
667  * It will sleep waiting for available log space.
668  * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
669  * is used by long running transactions.  If any one of the reservations
670  * fails then they will all be backed out.
671  *
672  * This does not do quota reservations. That typically is done by the
673  * caller afterwards.
674  */
675 int
676 xfs_trans_reserve(
677 	xfs_trans_t	*tp,
678 	uint		blocks,
679 	uint		logspace,
680 	uint		rtextents,
681 	uint		flags,
682 	uint		logcount)
683 {
684 	int		error = 0;
685 	int		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
686 
687 	/* Mark this thread as being in a transaction */
688 	current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
689 
690 	/*
691 	 * Attempt to reserve the needed disk blocks by decrementing
692 	 * the number needed from the number available.  This will
693 	 * fail if the count would go below zero.
694 	 */
695 	if (blocks > 0) {
696 		error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
697 					  -((int64_t)blocks), rsvd);
698 		if (error != 0) {
699 			current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
700 			return (XFS_ERROR(ENOSPC));
701 		}
702 		tp->t_blk_res += blocks;
703 	}
704 
705 	/*
706 	 * Reserve the log space needed for this transaction.
707 	 */
708 	if (logspace > 0) {
709 		bool	permanent = false;
710 
711 		ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace);
712 		ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount);
713 
714 		if (flags & XFS_TRANS_PERM_LOG_RES) {
715 			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
716 			permanent = true;
717 		} else {
718 			ASSERT(tp->t_ticket == NULL);
719 			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
720 		}
721 
722 		if (tp->t_ticket != NULL) {
723 			ASSERT(flags & XFS_TRANS_PERM_LOG_RES);
724 			error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
725 		} else {
726 			error = xfs_log_reserve(tp->t_mountp, logspace,
727 						logcount, &tp->t_ticket,
728 						XFS_TRANSACTION, permanent,
729 						tp->t_type);
730 		}
731 
732 		if (error)
733 			goto undo_blocks;
734 
735 		tp->t_log_res = logspace;
736 		tp->t_log_count = logcount;
737 	}
738 
739 	/*
740 	 * Attempt to reserve the needed realtime extents by decrementing
741 	 * the number needed from the number available.  This will
742 	 * fail if the count would go below zero.
743 	 */
744 	if (rtextents > 0) {
745 		error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
746 					  -((int64_t)rtextents), rsvd);
747 		if (error) {
748 			error = XFS_ERROR(ENOSPC);
749 			goto undo_log;
750 		}
751 		tp->t_rtx_res += rtextents;
752 	}
753 
754 	return 0;
755 
756 	/*
757 	 * Error cases jump to one of these labels to undo any
758 	 * reservations which have already been performed.
759 	 */
760 undo_log:
761 	if (logspace > 0) {
762 		int		log_flags;
763 
764 		if (flags & XFS_TRANS_PERM_LOG_RES) {
765 			log_flags = XFS_LOG_REL_PERM_RESERV;
766 		} else {
767 			log_flags = 0;
768 		}
769 		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
770 		tp->t_ticket = NULL;
771 		tp->t_log_res = 0;
772 		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
773 	}
774 
775 undo_blocks:
776 	if (blocks > 0) {
777 		xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
778 					 (int64_t)blocks, rsvd);
779 		tp->t_blk_res = 0;
780 	}
781 
782 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
783 
784 	return error;
785 }
786 
787 /*
788  * Record the indicated change to the given field for application
789  * to the file system's superblock when the transaction commits.
790  * For now, just store the change in the transaction structure.
791  *
792  * Mark the transaction structure to indicate that the superblock
793  * needs to be updated before committing.
794  *
795  * Because we may not be keeping track of allocated/free inodes and
796  * used filesystem blocks in the superblock, we do not mark the
797  * superblock dirty in this transaction if we modify these fields.
798  * We still need to update the transaction deltas so that they get
799  * applied to the incore superblock, but we don't want them to
800  * cause the superblock to get locked and logged if these are the
801  * only fields in the superblock that the transaction modifies.
802  */
803 void
804 xfs_trans_mod_sb(
805 	xfs_trans_t	*tp,
806 	uint		field,
807 	int64_t		delta)
808 {
809 	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
810 	xfs_mount_t	*mp = tp->t_mountp;
811 
812 	switch (field) {
813 	case XFS_TRANS_SB_ICOUNT:
814 		tp->t_icount_delta += delta;
815 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
816 			flags &= ~XFS_TRANS_SB_DIRTY;
817 		break;
818 	case XFS_TRANS_SB_IFREE:
819 		tp->t_ifree_delta += delta;
820 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
821 			flags &= ~XFS_TRANS_SB_DIRTY;
822 		break;
823 	case XFS_TRANS_SB_FDBLOCKS:
824 		/*
825 		 * Track the number of blocks allocated in the
826 		 * transaction.  Make sure it does not exceed the
827 		 * number reserved.
828 		 */
829 		if (delta < 0) {
830 			tp->t_blk_res_used += (uint)-delta;
831 			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
832 		}
833 		tp->t_fdblocks_delta += delta;
834 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
835 			flags &= ~XFS_TRANS_SB_DIRTY;
836 		break;
837 	case XFS_TRANS_SB_RES_FDBLOCKS:
838 		/*
839 		 * The allocation has already been applied to the
840 		 * in-core superblock's counter.  This should only
841 		 * be applied to the on-disk superblock.
842 		 */
843 		ASSERT(delta < 0);
844 		tp->t_res_fdblocks_delta += delta;
845 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
846 			flags &= ~XFS_TRANS_SB_DIRTY;
847 		break;
848 	case XFS_TRANS_SB_FREXTENTS:
849 		/*
850 		 * Track the number of blocks allocated in the
851 		 * transaction.  Make sure it does not exceed the
852 		 * number reserved.
853 		 */
854 		if (delta < 0) {
855 			tp->t_rtx_res_used += (uint)-delta;
856 			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
857 		}
858 		tp->t_frextents_delta += delta;
859 		break;
860 	case XFS_TRANS_SB_RES_FREXTENTS:
861 		/*
862 		 * The allocation has already been applied to the
863 		 * in-core superblock's counter.  This should only
864 		 * be applied to the on-disk superblock.
865 		 */
866 		ASSERT(delta < 0);
867 		tp->t_res_frextents_delta += delta;
868 		break;
869 	case XFS_TRANS_SB_DBLOCKS:
870 		ASSERT(delta > 0);
871 		tp->t_dblocks_delta += delta;
872 		break;
873 	case XFS_TRANS_SB_AGCOUNT:
874 		ASSERT(delta > 0);
875 		tp->t_agcount_delta += delta;
876 		break;
877 	case XFS_TRANS_SB_IMAXPCT:
878 		tp->t_imaxpct_delta += delta;
879 		break;
880 	case XFS_TRANS_SB_REXTSIZE:
881 		tp->t_rextsize_delta += delta;
882 		break;
883 	case XFS_TRANS_SB_RBMBLOCKS:
884 		tp->t_rbmblocks_delta += delta;
885 		break;
886 	case XFS_TRANS_SB_RBLOCKS:
887 		tp->t_rblocks_delta += delta;
888 		break;
889 	case XFS_TRANS_SB_REXTENTS:
890 		tp->t_rextents_delta += delta;
891 		break;
892 	case XFS_TRANS_SB_REXTSLOG:
893 		tp->t_rextslog_delta += delta;
894 		break;
895 	default:
896 		ASSERT(0);
897 		return;
898 	}
899 
900 	tp->t_flags |= flags;
901 }
902 
903 /*
904  * xfs_trans_apply_sb_deltas() is called from the commit code
905  * to bring the superblock buffer into the current transaction
906  * and modify it as requested by earlier calls to xfs_trans_mod_sb().
907  *
908  * For now we just look at each field allowed to change and change
909  * it if necessary.
910  */
911 STATIC void
912 xfs_trans_apply_sb_deltas(
913 	xfs_trans_t	*tp)
914 {
915 	xfs_dsb_t	*sbp;
916 	xfs_buf_t	*bp;
917 	int		whole = 0;
918 
919 	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
920 	sbp = XFS_BUF_TO_SBP(bp);
921 
922 	/*
923 	 * Check that superblock mods match the mods made to AGF counters.
924 	 */
925 	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
926 	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
927 		tp->t_ag_btree_delta));
928 
929 	/*
930 	 * Only update the superblock counters if we are logging them
931 	 */
932 	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
933 		if (tp->t_icount_delta)
934 			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
935 		if (tp->t_ifree_delta)
936 			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
937 		if (tp->t_fdblocks_delta)
938 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
939 		if (tp->t_res_fdblocks_delta)
940 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
941 	}
942 
943 	if (tp->t_frextents_delta)
944 		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
945 	if (tp->t_res_frextents_delta)
946 		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
947 
948 	if (tp->t_dblocks_delta) {
949 		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
950 		whole = 1;
951 	}
952 	if (tp->t_agcount_delta) {
953 		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
954 		whole = 1;
955 	}
956 	if (tp->t_imaxpct_delta) {
957 		sbp->sb_imax_pct += tp->t_imaxpct_delta;
958 		whole = 1;
959 	}
960 	if (tp->t_rextsize_delta) {
961 		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
962 		whole = 1;
963 	}
964 	if (tp->t_rbmblocks_delta) {
965 		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
966 		whole = 1;
967 	}
968 	if (tp->t_rblocks_delta) {
969 		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
970 		whole = 1;
971 	}
972 	if (tp->t_rextents_delta) {
973 		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
974 		whole = 1;
975 	}
976 	if (tp->t_rextslog_delta) {
977 		sbp->sb_rextslog += tp->t_rextslog_delta;
978 		whole = 1;
979 	}
980 
981 	if (whole)
982 		/*
983 		 * Log the whole thing, the fields are noncontiguous.
984 		 */
985 		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
986 	else
987 		/*
988 		 * Since all the modifiable fields are contiguous, we
989 		 * can get away with this.
990 		 */
991 		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
992 				  offsetof(xfs_dsb_t, sb_frextents) +
993 				  sizeof(sbp->sb_frextents) - 1);
994 }
995 
996 /*
997  * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
998  * and apply superblock counter changes to the in-core superblock.  The
999  * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
1000  * applied to the in-core superblock.  The idea is that that has already been
1001  * done.
1002  *
1003  * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
1004  * However, we have to ensure that we only modify each superblock field only
1005  * once because the application of the delta values may not be atomic. That can
1006  * lead to ENOSPC races occurring if we have two separate modifcations of the
1007  * free space counter to put back the entire reservation and then take away
1008  * what we used.
1009  *
1010  * If we are not logging superblock counters, then the inode allocated/free and
1011  * used block counts are not updated in the on disk superblock. In this case,
1012  * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1013  * still need to update the incore superblock with the changes.
1014  */
1015 void
1016 xfs_trans_unreserve_and_mod_sb(
1017 	xfs_trans_t	*tp)
1018 {
1019 	xfs_mod_sb_t	msb[9];	/* If you add cases, add entries */
1020 	xfs_mod_sb_t	*msbp;
1021 	xfs_mount_t	*mp = tp->t_mountp;
1022 	/* REFERENCED */
1023 	int		error;
1024 	int		rsvd;
1025 	int64_t		blkdelta = 0;
1026 	int64_t		rtxdelta = 0;
1027 	int64_t		idelta = 0;
1028 	int64_t		ifreedelta = 0;
1029 
1030 	msbp = msb;
1031 	rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1032 
1033 	/* calculate deltas */
1034 	if (tp->t_blk_res > 0)
1035 		blkdelta = tp->t_blk_res;
1036 	if ((tp->t_fdblocks_delta != 0) &&
1037 	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1038 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1039 	        blkdelta += tp->t_fdblocks_delta;
1040 
1041 	if (tp->t_rtx_res > 0)
1042 		rtxdelta = tp->t_rtx_res;
1043 	if ((tp->t_frextents_delta != 0) &&
1044 	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
1045 		rtxdelta += tp->t_frextents_delta;
1046 
1047 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1048 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1049 		idelta = tp->t_icount_delta;
1050 		ifreedelta = tp->t_ifree_delta;
1051 	}
1052 
1053 	/* apply the per-cpu counters */
1054 	if (blkdelta) {
1055 		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1056 						 blkdelta, rsvd);
1057 		if (error)
1058 			goto out;
1059 	}
1060 
1061 	if (idelta) {
1062 		error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1063 						 idelta, rsvd);
1064 		if (error)
1065 			goto out_undo_fdblocks;
1066 	}
1067 
1068 	if (ifreedelta) {
1069 		error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1070 						 ifreedelta, rsvd);
1071 		if (error)
1072 			goto out_undo_icount;
1073 	}
1074 
1075 	/* apply remaining deltas */
1076 	if (rtxdelta != 0) {
1077 		msbp->msb_field = XFS_SBS_FREXTENTS;
1078 		msbp->msb_delta = rtxdelta;
1079 		msbp++;
1080 	}
1081 
1082 	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
1083 		if (tp->t_dblocks_delta != 0) {
1084 			msbp->msb_field = XFS_SBS_DBLOCKS;
1085 			msbp->msb_delta = tp->t_dblocks_delta;
1086 			msbp++;
1087 		}
1088 		if (tp->t_agcount_delta != 0) {
1089 			msbp->msb_field = XFS_SBS_AGCOUNT;
1090 			msbp->msb_delta = tp->t_agcount_delta;
1091 			msbp++;
1092 		}
1093 		if (tp->t_imaxpct_delta != 0) {
1094 			msbp->msb_field = XFS_SBS_IMAX_PCT;
1095 			msbp->msb_delta = tp->t_imaxpct_delta;
1096 			msbp++;
1097 		}
1098 		if (tp->t_rextsize_delta != 0) {
1099 			msbp->msb_field = XFS_SBS_REXTSIZE;
1100 			msbp->msb_delta = tp->t_rextsize_delta;
1101 			msbp++;
1102 		}
1103 		if (tp->t_rbmblocks_delta != 0) {
1104 			msbp->msb_field = XFS_SBS_RBMBLOCKS;
1105 			msbp->msb_delta = tp->t_rbmblocks_delta;
1106 			msbp++;
1107 		}
1108 		if (tp->t_rblocks_delta != 0) {
1109 			msbp->msb_field = XFS_SBS_RBLOCKS;
1110 			msbp->msb_delta = tp->t_rblocks_delta;
1111 			msbp++;
1112 		}
1113 		if (tp->t_rextents_delta != 0) {
1114 			msbp->msb_field = XFS_SBS_REXTENTS;
1115 			msbp->msb_delta = tp->t_rextents_delta;
1116 			msbp++;
1117 		}
1118 		if (tp->t_rextslog_delta != 0) {
1119 			msbp->msb_field = XFS_SBS_REXTSLOG;
1120 			msbp->msb_delta = tp->t_rextslog_delta;
1121 			msbp++;
1122 		}
1123 	}
1124 
1125 	/*
1126 	 * If we need to change anything, do it.
1127 	 */
1128 	if (msbp > msb) {
1129 		error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1130 			(uint)(msbp - msb), rsvd);
1131 		if (error)
1132 			goto out_undo_ifreecount;
1133 	}
1134 
1135 	return;
1136 
1137 out_undo_ifreecount:
1138 	if (ifreedelta)
1139 		xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1140 out_undo_icount:
1141 	if (idelta)
1142 		xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1143 out_undo_fdblocks:
1144 	if (blkdelta)
1145 		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1146 out:
1147 	ASSERT(error == 0);
1148 	return;
1149 }
1150 
1151 /*
1152  * Add the given log item to the transaction's list of log items.
1153  *
1154  * The log item will now point to its new descriptor with its li_desc field.
1155  */
1156 void
1157 xfs_trans_add_item(
1158 	struct xfs_trans	*tp,
1159 	struct xfs_log_item	*lip)
1160 {
1161 	struct xfs_log_item_desc *lidp;
1162 
1163 	ASSERT(lip->li_mountp == tp->t_mountp);
1164 	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
1165 
1166 	lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
1167 
1168 	lidp->lid_item = lip;
1169 	lidp->lid_flags = 0;
1170 	list_add_tail(&lidp->lid_trans, &tp->t_items);
1171 
1172 	lip->li_desc = lidp;
1173 }
1174 
1175 STATIC void
1176 xfs_trans_free_item_desc(
1177 	struct xfs_log_item_desc *lidp)
1178 {
1179 	list_del_init(&lidp->lid_trans);
1180 	kmem_zone_free(xfs_log_item_desc_zone, lidp);
1181 }
1182 
1183 /*
1184  * Unlink and free the given descriptor.
1185  */
1186 void
1187 xfs_trans_del_item(
1188 	struct xfs_log_item	*lip)
1189 {
1190 	xfs_trans_free_item_desc(lip->li_desc);
1191 	lip->li_desc = NULL;
1192 }
1193 
1194 /*
1195  * Unlock all of the items of a transaction and free all the descriptors
1196  * of that transaction.
1197  */
1198 void
1199 xfs_trans_free_items(
1200 	struct xfs_trans	*tp,
1201 	xfs_lsn_t		commit_lsn,
1202 	int			flags)
1203 {
1204 	struct xfs_log_item_desc *lidp, *next;
1205 
1206 	list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1207 		struct xfs_log_item	*lip = lidp->lid_item;
1208 
1209 		lip->li_desc = NULL;
1210 
1211 		if (commit_lsn != NULLCOMMITLSN)
1212 			IOP_COMMITTING(lip, commit_lsn);
1213 		if (flags & XFS_TRANS_ABORT)
1214 			lip->li_flags |= XFS_LI_ABORTED;
1215 		IOP_UNLOCK(lip);
1216 
1217 		xfs_trans_free_item_desc(lidp);
1218 	}
1219 }
1220 
1221 static inline void
1222 xfs_log_item_batch_insert(
1223 	struct xfs_ail		*ailp,
1224 	struct xfs_ail_cursor	*cur,
1225 	struct xfs_log_item	**log_items,
1226 	int			nr_items,
1227 	xfs_lsn_t		commit_lsn)
1228 {
1229 	int	i;
1230 
1231 	spin_lock(&ailp->xa_lock);
1232 	/* xfs_trans_ail_update_bulk drops ailp->xa_lock */
1233 	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
1234 
1235 	for (i = 0; i < nr_items; i++)
1236 		IOP_UNPIN(log_items[i], 0);
1237 }
1238 
1239 /*
1240  * Bulk operation version of xfs_trans_committed that takes a log vector of
1241  * items to insert into the AIL. This uses bulk AIL insertion techniques to
1242  * minimise lock traffic.
1243  *
1244  * If we are called with the aborted flag set, it is because a log write during
1245  * a CIL checkpoint commit has failed. In this case, all the items in the
1246  * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
1247  * means that checkpoint commit abort handling is treated exactly the same
1248  * as an iclog write error even though we haven't started any IO yet. Hence in
1249  * this case all we need to do is IOP_COMMITTED processing, followed by an
1250  * IOP_UNPIN(aborted) call.
1251  *
1252  * The AIL cursor is used to optimise the insert process. If commit_lsn is not
1253  * at the end of the AIL, the insert cursor avoids the need to walk
1254  * the AIL to find the insertion point on every xfs_log_item_batch_insert()
1255  * call. This saves a lot of needless list walking and is a net win, even
1256  * though it slightly increases that amount of AIL lock traffic to set it up
1257  * and tear it down.
1258  */
1259 void
1260 xfs_trans_committed_bulk(
1261 	struct xfs_ail		*ailp,
1262 	struct xfs_log_vec	*log_vector,
1263 	xfs_lsn_t		commit_lsn,
1264 	int			aborted)
1265 {
1266 #define LOG_ITEM_BATCH_SIZE	32
1267 	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
1268 	struct xfs_log_vec	*lv;
1269 	struct xfs_ail_cursor	cur;
1270 	int			i = 0;
1271 
1272 	spin_lock(&ailp->xa_lock);
1273 	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
1274 	spin_unlock(&ailp->xa_lock);
1275 
1276 	/* unpin all the log items */
1277 	for (lv = log_vector; lv; lv = lv->lv_next ) {
1278 		struct xfs_log_item	*lip = lv->lv_item;
1279 		xfs_lsn_t		item_lsn;
1280 
1281 		if (aborted)
1282 			lip->li_flags |= XFS_LI_ABORTED;
1283 		item_lsn = IOP_COMMITTED(lip, commit_lsn);
1284 
1285 		/* item_lsn of -1 means the item needs no further processing */
1286 		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1287 			continue;
1288 
1289 		/*
1290 		 * if we are aborting the operation, no point in inserting the
1291 		 * object into the AIL as we are in a shutdown situation.
1292 		 */
1293 		if (aborted) {
1294 			ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
1295 			IOP_UNPIN(lip, 1);
1296 			continue;
1297 		}
1298 
1299 		if (item_lsn != commit_lsn) {
1300 
1301 			/*
1302 			 * Not a bulk update option due to unusual item_lsn.
1303 			 * Push into AIL immediately, rechecking the lsn once
1304 			 * we have the ail lock. Then unpin the item. This does
1305 			 * not affect the AIL cursor the bulk insert path is
1306 			 * using.
1307 			 */
1308 			spin_lock(&ailp->xa_lock);
1309 			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1310 				xfs_trans_ail_update(ailp, lip, item_lsn);
1311 			else
1312 				spin_unlock(&ailp->xa_lock);
1313 			IOP_UNPIN(lip, 0);
1314 			continue;
1315 		}
1316 
1317 		/* Item is a candidate for bulk AIL insert.  */
1318 		log_items[i++] = lv->lv_item;
1319 		if (i >= LOG_ITEM_BATCH_SIZE) {
1320 			xfs_log_item_batch_insert(ailp, &cur, log_items,
1321 					LOG_ITEM_BATCH_SIZE, commit_lsn);
1322 			i = 0;
1323 		}
1324 	}
1325 
1326 	/* make sure we insert the remainder! */
1327 	if (i)
1328 		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
1329 
1330 	spin_lock(&ailp->xa_lock);
1331 	xfs_trans_ail_cursor_done(ailp, &cur);
1332 	spin_unlock(&ailp->xa_lock);
1333 }
1334 
1335 /*
1336  * Commit the given transaction to the log.
1337  *
1338  * XFS disk error handling mechanism is not based on a typical
1339  * transaction abort mechanism. Logically after the filesystem
1340  * gets marked 'SHUTDOWN', we can't let any new transactions
1341  * be durable - ie. committed to disk - because some metadata might
1342  * be inconsistent. In such cases, this returns an error, and the
1343  * caller may assume that all locked objects joined to the transaction
1344  * have already been unlocked as if the commit had succeeded.
1345  * Do not reference the transaction structure after this call.
1346  */
1347 int
1348 xfs_trans_commit(
1349 	struct xfs_trans	*tp,
1350 	uint			flags)
1351 {
1352 	struct xfs_mount	*mp = tp->t_mountp;
1353 	xfs_lsn_t		commit_lsn = -1;
1354 	int			error = 0;
1355 	int			log_flags = 0;
1356 	int			sync = tp->t_flags & XFS_TRANS_SYNC;
1357 
1358 	/*
1359 	 * Determine whether this commit is releasing a permanent
1360 	 * log reservation or not.
1361 	 */
1362 	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1363 		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1364 		log_flags = XFS_LOG_REL_PERM_RESERV;
1365 	}
1366 
1367 	/*
1368 	 * If there is nothing to be logged by the transaction,
1369 	 * then unlock all of the items associated with the
1370 	 * transaction and free the transaction structure.
1371 	 * Also make sure to return any reserved blocks to
1372 	 * the free pool.
1373 	 */
1374 	if (!(tp->t_flags & XFS_TRANS_DIRTY))
1375 		goto out_unreserve;
1376 
1377 	if (XFS_FORCED_SHUTDOWN(mp)) {
1378 		error = XFS_ERROR(EIO);
1379 		goto out_unreserve;
1380 	}
1381 
1382 	ASSERT(tp->t_ticket != NULL);
1383 
1384 	/*
1385 	 * If we need to update the superblock, then do it now.
1386 	 */
1387 	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1388 		xfs_trans_apply_sb_deltas(tp);
1389 	xfs_trans_apply_dquot_deltas(tp);
1390 
1391 	error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags);
1392 	if (error == ENOMEM) {
1393 		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1394 		error = XFS_ERROR(EIO);
1395 		goto out_unreserve;
1396 	}
1397 
1398 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1399 	xfs_trans_free(tp);
1400 
1401 	/*
1402 	 * If the transaction needs to be synchronous, then force the
1403 	 * log out now and wait for it.
1404 	 */
1405 	if (sync) {
1406 		if (!error) {
1407 			error = _xfs_log_force_lsn(mp, commit_lsn,
1408 				      XFS_LOG_SYNC, NULL);
1409 		}
1410 		XFS_STATS_INC(xs_trans_sync);
1411 	} else {
1412 		XFS_STATS_INC(xs_trans_async);
1413 	}
1414 
1415 	return error;
1416 
1417 out_unreserve:
1418 	xfs_trans_unreserve_and_mod_sb(tp);
1419 
1420 	/*
1421 	 * It is indeed possible for the transaction to be not dirty but
1422 	 * the dqinfo portion to be.  All that means is that we have some
1423 	 * (non-persistent) quota reservations that need to be unreserved.
1424 	 */
1425 	xfs_trans_unreserve_and_mod_dquots(tp);
1426 	if (tp->t_ticket) {
1427 		commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1428 		if (commit_lsn == -1 && !error)
1429 			error = XFS_ERROR(EIO);
1430 	}
1431 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1432 	xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
1433 	xfs_trans_free(tp);
1434 
1435 	XFS_STATS_INC(xs_trans_empty);
1436 	return error;
1437 }
1438 
1439 /*
1440  * Unlock all of the transaction's items and free the transaction.
1441  * The transaction must not have modified any of its items, because
1442  * there is no way to restore them to their previous state.
1443  *
1444  * If the transaction has made a log reservation, make sure to release
1445  * it as well.
1446  */
1447 void
1448 xfs_trans_cancel(
1449 	xfs_trans_t		*tp,
1450 	int			flags)
1451 {
1452 	int			log_flags;
1453 	xfs_mount_t		*mp = tp->t_mountp;
1454 
1455 	/*
1456 	 * See if the caller is being too lazy to figure out if
1457 	 * the transaction really needs an abort.
1458 	 */
1459 	if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1460 		flags &= ~XFS_TRANS_ABORT;
1461 	/*
1462 	 * See if the caller is relying on us to shut down the
1463 	 * filesystem.  This happens in paths where we detect
1464 	 * corruption and decide to give up.
1465 	 */
1466 	if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1467 		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1468 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1469 	}
1470 #ifdef DEBUG
1471 	if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1472 		struct xfs_log_item_desc *lidp;
1473 
1474 		list_for_each_entry(lidp, &tp->t_items, lid_trans)
1475 			ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1476 	}
1477 #endif
1478 	xfs_trans_unreserve_and_mod_sb(tp);
1479 	xfs_trans_unreserve_and_mod_dquots(tp);
1480 
1481 	if (tp->t_ticket) {
1482 		if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1483 			ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1484 			log_flags = XFS_LOG_REL_PERM_RESERV;
1485 		} else {
1486 			log_flags = 0;
1487 		}
1488 		xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1489 	}
1490 
1491 	/* mark this thread as no longer being in a transaction */
1492 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1493 
1494 	xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1495 	xfs_trans_free(tp);
1496 }
1497 
1498 /*
1499  * Roll from one trans in the sequence of PERMANENT transactions to
1500  * the next: permanent transactions are only flushed out when
1501  * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1502  * as possible to let chunks of it go to the log. So we commit the
1503  * chunk we've been working on and get a new transaction to continue.
1504  */
1505 int
1506 xfs_trans_roll(
1507 	struct xfs_trans	**tpp,
1508 	struct xfs_inode	*dp)
1509 {
1510 	struct xfs_trans	*trans;
1511 	unsigned int		logres, count;
1512 	int			error;
1513 
1514 	/*
1515 	 * Ensure that the inode is always logged.
1516 	 */
1517 	trans = *tpp;
1518 	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1519 
1520 	/*
1521 	 * Copy the critical parameters from one trans to the next.
1522 	 */
1523 	logres = trans->t_log_res;
1524 	count = trans->t_log_count;
1525 	*tpp = xfs_trans_dup(trans);
1526 
1527 	/*
1528 	 * Commit the current transaction.
1529 	 * If this commit failed, then it'd just unlock those items that
1530 	 * are not marked ihold. That also means that a filesystem shutdown
1531 	 * is in progress. The caller takes the responsibility to cancel
1532 	 * the duplicate transaction that gets returned.
1533 	 */
1534 	error = xfs_trans_commit(trans, 0);
1535 	if (error)
1536 		return (error);
1537 
1538 	trans = *tpp;
1539 
1540 	/*
1541 	 * transaction commit worked ok so we can drop the extra ticket
1542 	 * reference that we gained in xfs_trans_dup()
1543 	 */
1544 	xfs_log_ticket_put(trans->t_ticket);
1545 
1546 
1547 	/*
1548 	 * Reserve space in the log for th next transaction.
1549 	 * This also pushes items in the "AIL", the list of logged items,
1550 	 * out to disk if they are taking up space at the tail of the log
1551 	 * that we want to use.  This requires that either nothing be locked
1552 	 * across this call, or that anything that is locked be logged in
1553 	 * the prior and the next transactions.
1554 	 */
1555 	error = xfs_trans_reserve(trans, 0, logres, 0,
1556 				  XFS_TRANS_PERM_LOG_RES, count);
1557 	/*
1558 	 *  Ensure that the inode is in the new transaction and locked.
1559 	 */
1560 	if (error)
1561 		return error;
1562 
1563 	xfs_trans_ijoin(trans, dp, 0);
1564 	return 0;
1565 }
1566