xref: /linux/fs/xfs/xfs_trans.c (revision 148f9bb87745ed45f7a11b2cbd3bc0f017d5d257)
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_qm.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_log_priv.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trace.h"
47 
48 kmem_zone_t	*xfs_trans_zone;
49 kmem_zone_t	*xfs_log_item_desc_zone;
50 
51 /*
52  * A buffer has a format structure overhead in the log in addition
53  * to the data, so we need to take this into account when reserving
54  * space in a transaction for a buffer.  Round the space required up
55  * to a multiple of 128 bytes so that we don't change the historical
56  * reservation that has been used for this overhead.
57  */
58 STATIC uint
59 xfs_buf_log_overhead(void)
60 {
61 	return round_up(sizeof(struct xlog_op_header) +
62 			sizeof(struct xfs_buf_log_format), 128);
63 }
64 
65 /*
66  * Calculate out transaction log reservation per item in bytes.
67  *
68  * The nbufs argument is used to indicate the number of items that
69  * will be changed in a transaction.  size is used to tell how many
70  * bytes should be reserved per item.
71  */
72 STATIC uint
73 xfs_calc_buf_res(
74 	uint		nbufs,
75 	uint		size)
76 {
77 	return nbufs * (size + xfs_buf_log_overhead());
78 }
79 
80 /*
81  * Various log reservation values.
82  *
83  * These are based on the size of the file system block because that is what
84  * most transactions manipulate.  Each adds in an additional 128 bytes per
85  * item logged to try to account for the overhead of the transaction mechanism.
86  *
87  * Note:  Most of the reservations underestimate the number of allocation
88  * groups into which they could free extents in the xfs_bmap_finish() call.
89  * This is because the number in the worst case is quite high and quite
90  * unusual.  In order to fix this we need to change xfs_bmap_finish() to free
91  * extents in only a single AG at a time.  This will require changes to the
92  * EFI code as well, however, so that the EFI for the extents not freed is
93  * logged again in each transaction.  See SGI PV #261917.
94  *
95  * Reservation functions here avoid a huge stack in xfs_trans_init due to
96  * register overflow from temporaries in the calculations.
97  */
98 
99 
100 /*
101  * In a write transaction we can allocate a maximum of 2
102  * extents.  This gives:
103  *    the inode getting the new extents: inode size
104  *    the inode's bmap btree: max depth * block size
105  *    the agfs of the ags from which the extents are allocated: 2 * sector
106  *    the superblock free block counter: sector size
107  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
108  * And the bmap_finish transaction can free bmap blocks in a join:
109  *    the agfs of the ags containing the blocks: 2 * sector size
110  *    the agfls of the ags containing the blocks: 2 * sector size
111  *    the super block free block counter: sector size
112  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
113  */
114 STATIC uint
115 xfs_calc_write_reservation(
116 	struct xfs_mount	*mp)
117 {
118 	return XFS_DQUOT_LOGRES(mp) +
119 		MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
120 		     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
121 				      XFS_FSB_TO_B(mp, 1)) +
122 		     xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
123 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2),
124 				      XFS_FSB_TO_B(mp, 1))),
125 		    (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
126 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2),
127 				      XFS_FSB_TO_B(mp, 1))));
128 }
129 
130 /*
131  * In truncating a file we free up to two extents at once.  We can modify:
132  *    the inode being truncated: inode size
133  *    the inode's bmap btree: (max depth + 1) * block size
134  * And the bmap_finish transaction can free the blocks and bmap blocks:
135  *    the agf for each of the ags: 4 * sector size
136  *    the agfl for each of the ags: 4 * sector size
137  *    the super block to reflect the freed blocks: sector size
138  *    worst case split in allocation btrees per extent assuming 4 extents:
139  *		4 exts * 2 trees * (2 * max depth - 1) * block size
140  *    the inode btree: max depth * blocksize
141  *    the allocation btrees: 2 trees * (max depth - 1) * block size
142  */
143 STATIC uint
144 xfs_calc_itruncate_reservation(
145 	struct xfs_mount	*mp)
146 {
147 	return XFS_DQUOT_LOGRES(mp) +
148 		MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
149 		     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1,
150 				      XFS_FSB_TO_B(mp, 1))),
151 		    (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
152 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 4),
153 				      XFS_FSB_TO_B(mp, 1)) +
154 		    xfs_calc_buf_res(5, 0) +
155 		    xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
156 				     XFS_FSB_TO_B(mp, 1)) +
157 		    xfs_calc_buf_res(2 + XFS_IALLOC_BLOCKS(mp) +
158 				     mp->m_in_maxlevels, 0)));
159 }
160 
161 /*
162  * In renaming a files we can modify:
163  *    the four inodes involved: 4 * inode size
164  *    the two directory btrees: 2 * (max depth + v2) * dir block size
165  *    the two directory bmap btrees: 2 * max depth * block size
166  * And the bmap_finish transaction can free dir and bmap blocks (two sets
167  *	of bmap blocks) giving:
168  *    the agf for the ags in which the blocks live: 3 * sector size
169  *    the agfl for the ags in which the blocks live: 3 * sector size
170  *    the superblock for the free block count: sector size
171  *    the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
172  */
173 STATIC uint
174 xfs_calc_rename_reservation(
175 	struct xfs_mount	*mp)
176 {
177 	return XFS_DQUOT_LOGRES(mp) +
178 		MAX((xfs_calc_buf_res(4, mp->m_sb.sb_inodesize) +
179 		     xfs_calc_buf_res(2 * XFS_DIROP_LOG_COUNT(mp),
180 				      XFS_FSB_TO_B(mp, 1))),
181 		    (xfs_calc_buf_res(7, mp->m_sb.sb_sectsize) +
182 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 3),
183 				      XFS_FSB_TO_B(mp, 1))));
184 }
185 
186 /*
187  * For creating a link to an inode:
188  *    the parent directory inode: inode size
189  *    the linked inode: inode size
190  *    the directory btree could split: (max depth + v2) * dir block size
191  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
192  * And the bmap_finish transaction can free some bmap blocks giving:
193  *    the agf for the ag in which the blocks live: sector size
194  *    the agfl for the ag in which the blocks live: sector size
195  *    the superblock for the free block count: sector size
196  *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
197  */
198 STATIC uint
199 xfs_calc_link_reservation(
200 	struct xfs_mount	*mp)
201 {
202 	return XFS_DQUOT_LOGRES(mp) +
203 		MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) +
204 		     xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
205 				      XFS_FSB_TO_B(mp, 1))),
206 		    (xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
207 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
208 				      XFS_FSB_TO_B(mp, 1))));
209 }
210 
211 /*
212  * For removing a directory entry we can modify:
213  *    the parent directory inode: inode size
214  *    the removed inode: inode size
215  *    the directory btree could join: (max depth + v2) * dir block size
216  *    the directory bmap btree could join or split: (max depth + v2) * blocksize
217  * And the bmap_finish transaction can free the dir and bmap blocks giving:
218  *    the agf for the ag in which the blocks live: 2 * sector size
219  *    the agfl for the ag in which the blocks live: 2 * sector size
220  *    the superblock for the free block count: sector size
221  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
222  */
223 STATIC uint
224 xfs_calc_remove_reservation(
225 	struct xfs_mount	*mp)
226 {
227 	return XFS_DQUOT_LOGRES(mp) +
228 		MAX((xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) +
229 		     xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp),
230 				      XFS_FSB_TO_B(mp, 1))),
231 		    (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
232 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2),
233 				      XFS_FSB_TO_B(mp, 1))));
234 }
235 
236 /*
237  * For create, break it in to the two cases that the transaction
238  * covers. We start with the modify case - allocation done by modification
239  * of the state of existing inodes - and the allocation case.
240  */
241 
242 /*
243  * For create we can modify:
244  *    the parent directory inode: inode size
245  *    the new inode: inode size
246  *    the inode btree entry: block size
247  *    the superblock for the nlink flag: sector size
248  *    the directory btree: (max depth + v2) * dir block size
249  *    the directory inode's bmap btree: (max depth + v2) * block size
250  */
251 STATIC uint
252 xfs_calc_create_resv_modify(
253 	struct xfs_mount	*mp)
254 {
255 	return xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) +
256 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
257 		(uint)XFS_FSB_TO_B(mp, 1) +
258 		xfs_calc_buf_res(XFS_DIROP_LOG_COUNT(mp), XFS_FSB_TO_B(mp, 1));
259 }
260 
261 /*
262  * For create we can allocate some inodes giving:
263  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
264  *    the superblock for the nlink flag: sector size
265  *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
266  *    the inode btree: max depth * blocksize
267  *    the allocation btrees: 2 trees * (max depth - 1) * block size
268  */
269 STATIC uint
270 xfs_calc_create_resv_alloc(
271 	struct xfs_mount	*mp)
272 {
273 	return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
274 		mp->m_sb.sb_sectsize +
275 		xfs_calc_buf_res(XFS_IALLOC_BLOCKS(mp), XFS_FSB_TO_B(mp, 1)) +
276 		xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1)) +
277 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
278 				 XFS_FSB_TO_B(mp, 1));
279 }
280 
281 STATIC uint
282 __xfs_calc_create_reservation(
283 	struct xfs_mount	*mp)
284 {
285 	return XFS_DQUOT_LOGRES(mp) +
286 		MAX(xfs_calc_create_resv_alloc(mp),
287 		    xfs_calc_create_resv_modify(mp));
288 }
289 
290 /*
291  * For icreate we can allocate some inodes giving:
292  *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
293  *    the superblock for the nlink flag: sector size
294  *    the inode btree: max depth * blocksize
295  *    the allocation btrees: 2 trees * (max depth - 1) * block size
296  */
297 STATIC uint
298 xfs_calc_icreate_resv_alloc(
299 	struct xfs_mount	*mp)
300 {
301 	return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
302 		mp->m_sb.sb_sectsize +
303 		xfs_calc_buf_res(mp->m_in_maxlevels, XFS_FSB_TO_B(mp, 1)) +
304 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
305 				 XFS_FSB_TO_B(mp, 1));
306 }
307 
308 STATIC uint
309 xfs_calc_icreate_reservation(xfs_mount_t *mp)
310 {
311 	return XFS_DQUOT_LOGRES(mp) +
312 		MAX(xfs_calc_icreate_resv_alloc(mp),
313 		    xfs_calc_create_resv_modify(mp));
314 }
315 
316 STATIC uint
317 xfs_calc_create_reservation(
318 	struct xfs_mount	*mp)
319 {
320 	if (xfs_sb_version_hascrc(&mp->m_sb))
321 		return xfs_calc_icreate_reservation(mp);
322 	return __xfs_calc_create_reservation(mp);
323 
324 }
325 
326 /*
327  * Making a new directory is the same as creating a new file.
328  */
329 STATIC uint
330 xfs_calc_mkdir_reservation(
331 	struct xfs_mount	*mp)
332 {
333 	return xfs_calc_create_reservation(mp);
334 }
335 
336 
337 /*
338  * Making a new symplink is the same as creating a new file, but
339  * with the added blocks for remote symlink data which can be up to 1kB in
340  * length (MAXPATHLEN).
341  */
342 STATIC uint
343 xfs_calc_symlink_reservation(
344 	struct xfs_mount	*mp)
345 {
346 	return xfs_calc_create_reservation(mp) +
347 	       xfs_calc_buf_res(1, MAXPATHLEN);
348 }
349 
350 /*
351  * In freeing an inode we can modify:
352  *    the inode being freed: inode size
353  *    the super block free inode counter: sector size
354  *    the agi hash list and counters: sector size
355  *    the inode btree entry: block size
356  *    the on disk inode before ours in the agi hash list: inode cluster size
357  *    the inode btree: max depth * blocksize
358  *    the allocation btrees: 2 trees * (max depth - 1) * block size
359  */
360 STATIC uint
361 xfs_calc_ifree_reservation(
362 	struct xfs_mount	*mp)
363 {
364 	return XFS_DQUOT_LOGRES(mp) +
365 		xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
366 		xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
367 		xfs_calc_buf_res(1, XFS_FSB_TO_B(mp, 1)) +
368 		MAX((__uint16_t)XFS_FSB_TO_B(mp, 1),
369 		    XFS_INODE_CLUSTER_SIZE(mp)) +
370 		xfs_calc_buf_res(1, 0) +
371 		xfs_calc_buf_res(2 + XFS_IALLOC_BLOCKS(mp) +
372 				 mp->m_in_maxlevels, 0) +
373 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
374 				 XFS_FSB_TO_B(mp, 1));
375 }
376 
377 /*
378  * When only changing the inode we log the inode and possibly the superblock
379  * We also add a bit of slop for the transaction stuff.
380  */
381 STATIC uint
382 xfs_calc_ichange_reservation(
383 	struct xfs_mount	*mp)
384 {
385 	return XFS_DQUOT_LOGRES(mp) +
386 		mp->m_sb.sb_inodesize +
387 		mp->m_sb.sb_sectsize +
388 		512;
389 
390 }
391 
392 /*
393  * Growing the data section of the filesystem.
394  *	superblock
395  *	agi and agf
396  *	allocation btrees
397  */
398 STATIC uint
399 xfs_calc_growdata_reservation(
400 	struct xfs_mount	*mp)
401 {
402 	return xfs_calc_buf_res(3, mp->m_sb.sb_sectsize) +
403 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
404 				 XFS_FSB_TO_B(mp, 1));
405 }
406 
407 /*
408  * Growing the rt section of the filesystem.
409  * In the first set of transactions (ALLOC) we allocate space to the
410  * bitmap or summary files.
411  *	superblock: sector size
412  *	agf of the ag from which the extent is allocated: sector size
413  *	bmap btree for bitmap/summary inode: max depth * blocksize
414  *	bitmap/summary inode: inode size
415  *	allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
416  */
417 STATIC uint
418 xfs_calc_growrtalloc_reservation(
419 	struct xfs_mount	*mp)
420 {
421 	return xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
422 		xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK),
423 				 XFS_FSB_TO_B(mp, 1)) +
424 		xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
425 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
426 				 XFS_FSB_TO_B(mp, 1));
427 }
428 
429 /*
430  * Growing the rt section of the filesystem.
431  * In the second set of transactions (ZERO) we zero the new metadata blocks.
432  *	one bitmap/summary block: blocksize
433  */
434 STATIC uint
435 xfs_calc_growrtzero_reservation(
436 	struct xfs_mount	*mp)
437 {
438 	return xfs_calc_buf_res(1, mp->m_sb.sb_blocksize);
439 }
440 
441 /*
442  * Growing the rt section of the filesystem.
443  * In the third set of transactions (FREE) we update metadata without
444  * allocating any new blocks.
445  *	superblock: sector size
446  *	bitmap inode: inode size
447  *	summary inode: inode size
448  *	one bitmap block: blocksize
449  *	summary blocks: new summary size
450  */
451 STATIC uint
452 xfs_calc_growrtfree_reservation(
453 	struct xfs_mount	*mp)
454 {
455 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
456 		xfs_calc_buf_res(2, mp->m_sb.sb_inodesize) +
457 		xfs_calc_buf_res(1, mp->m_sb.sb_blocksize) +
458 		xfs_calc_buf_res(1, mp->m_rsumsize);
459 }
460 
461 /*
462  * Logging the inode modification timestamp on a synchronous write.
463  *	inode
464  */
465 STATIC uint
466 xfs_calc_swrite_reservation(
467 	struct xfs_mount	*mp)
468 {
469 	return xfs_calc_buf_res(1, mp->m_sb.sb_inodesize);
470 }
471 
472 /*
473  * Logging the inode mode bits when writing a setuid/setgid file
474  *	inode
475  */
476 STATIC uint
477 xfs_calc_writeid_reservation(xfs_mount_t *mp)
478 {
479 	return xfs_calc_buf_res(1, mp->m_sb.sb_inodesize);
480 }
481 
482 /*
483  * Converting the inode from non-attributed to attributed.
484  *	the inode being converted: inode size
485  *	agf block and superblock (for block allocation)
486  *	the new block (directory sized)
487  *	bmap blocks for the new directory block
488  *	allocation btrees
489  */
490 STATIC uint
491 xfs_calc_addafork_reservation(
492 	struct xfs_mount	*mp)
493 {
494 	return XFS_DQUOT_LOGRES(mp) +
495 		xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
496 		xfs_calc_buf_res(2, mp->m_sb.sb_sectsize) +
497 		xfs_calc_buf_res(1, mp->m_dirblksize) +
498 		xfs_calc_buf_res(XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1,
499 				 XFS_FSB_TO_B(mp, 1)) +
500 		xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 1),
501 				 XFS_FSB_TO_B(mp, 1));
502 }
503 
504 /*
505  * Removing the attribute fork of a file
506  *    the inode being truncated: inode size
507  *    the inode's bmap btree: max depth * block size
508  * And the bmap_finish transaction can free the blocks and bmap blocks:
509  *    the agf for each of the ags: 4 * sector size
510  *    the agfl for each of the ags: 4 * sector size
511  *    the super block to reflect the freed blocks: sector size
512  *    worst case split in allocation btrees per extent assuming 4 extents:
513  *		4 exts * 2 trees * (2 * max depth - 1) * block size
514  */
515 STATIC uint
516 xfs_calc_attrinval_reservation(
517 	struct xfs_mount	*mp)
518 {
519 	return MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
520 		    xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
521 				     XFS_FSB_TO_B(mp, 1))),
522 		   (xfs_calc_buf_res(9, mp->m_sb.sb_sectsize) +
523 		    xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 4),
524 				     XFS_FSB_TO_B(mp, 1))));
525 }
526 
527 /*
528  * Setting an attribute at mount time.
529  *	the inode getting the attribute
530  *	the superblock for allocations
531  *	the agfs extents are allocated from
532  *	the attribute btree * max depth
533  *	the inode allocation btree
534  * Since attribute transaction space is dependent on the size of the attribute,
535  * the calculation is done partially at mount time and partially at runtime(see
536  * below).
537  */
538 STATIC uint
539 xfs_calc_attrsetm_reservation(
540 	struct xfs_mount	*mp)
541 {
542 	return XFS_DQUOT_LOGRES(mp) +
543 		xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
544 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
545 		xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH, XFS_FSB_TO_B(mp, 1));
546 }
547 
548 /*
549  * Setting an attribute at runtime, transaction space unit per block.
550  * 	the superblock for allocations: sector size
551  *	the inode bmap btree could join or split: max depth * block size
552  * Since the runtime attribute transaction space is dependent on the total
553  * blocks needed for the 1st bmap, here we calculate out the space unit for
554  * one block so that the caller could figure out the total space according
555  * to the attibute extent length in blocks by: ext * XFS_ATTRSETRT_LOG_RES(mp).
556  */
557 STATIC uint
558 xfs_calc_attrsetrt_reservation(
559 	struct xfs_mount	*mp)
560 {
561 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize) +
562 		xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK),
563 				 XFS_FSB_TO_B(mp, 1));
564 }
565 
566 /*
567  * Removing an attribute.
568  *    the inode: inode size
569  *    the attribute btree could join: max depth * block size
570  *    the inode bmap btree could join or split: max depth * block size
571  * And the bmap_finish transaction can free the attr blocks freed giving:
572  *    the agf for the ag in which the blocks live: 2 * sector size
573  *    the agfl for the ag in which the blocks live: 2 * sector size
574  *    the superblock for the free block count: sector size
575  *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
576  */
577 STATIC uint
578 xfs_calc_attrrm_reservation(
579 	struct xfs_mount	*mp)
580 {
581 	return XFS_DQUOT_LOGRES(mp) +
582 		MAX((xfs_calc_buf_res(1, mp->m_sb.sb_inodesize) +
583 		     xfs_calc_buf_res(XFS_DA_NODE_MAXDEPTH,
584 				      XFS_FSB_TO_B(mp, 1)) +
585 		     (uint)XFS_FSB_TO_B(mp,
586 					XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
587 		     xfs_calc_buf_res(XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK), 0)),
588 		    (xfs_calc_buf_res(5, mp->m_sb.sb_sectsize) +
589 		     xfs_calc_buf_res(XFS_ALLOCFREE_LOG_COUNT(mp, 2),
590 				      XFS_FSB_TO_B(mp, 1))));
591 }
592 
593 /*
594  * Clearing a bad agino number in an agi hash bucket.
595  */
596 STATIC uint
597 xfs_calc_clear_agi_bucket_reservation(
598 	struct xfs_mount	*mp)
599 {
600 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
601 }
602 
603 /*
604  * Clearing the quotaflags in the superblock.
605  *	the super block for changing quota flags: sector size
606  */
607 STATIC uint
608 xfs_calc_qm_sbchange_reservation(
609 	struct xfs_mount	*mp)
610 {
611 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
612 }
613 
614 /*
615  * Adjusting quota limits.
616  *    the xfs_disk_dquot_t: sizeof(struct xfs_disk_dquot)
617  */
618 STATIC uint
619 xfs_calc_qm_setqlim_reservation(
620 	struct xfs_mount	*mp)
621 {
622 	return xfs_calc_buf_res(1, sizeof(struct xfs_disk_dquot));
623 }
624 
625 /*
626  * Allocating quota on disk if needed.
627  *	the write transaction log space: XFS_WRITE_LOG_RES(mp)
628  *	the unit of quota allocation: one system block size
629  */
630 STATIC uint
631 xfs_calc_qm_dqalloc_reservation(
632 	struct xfs_mount	*mp)
633 {
634 	return XFS_WRITE_LOG_RES(mp) +
635 		xfs_calc_buf_res(1,
636 			XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB) - 1);
637 }
638 
639 /*
640  * Turning off quotas.
641  *    the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2
642  *    the superblock for the quota flags: sector size
643  */
644 STATIC uint
645 xfs_calc_qm_quotaoff_reservation(
646 	struct xfs_mount	*mp)
647 {
648 	return sizeof(struct xfs_qoff_logitem) * 2 +
649 		xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
650 }
651 
652 /*
653  * End of turning off quotas.
654  *    the xfs_qoff_logitem_t: sizeof(struct xfs_qoff_logitem) * 2
655  */
656 STATIC uint
657 xfs_calc_qm_quotaoff_end_reservation(
658 	struct xfs_mount	*mp)
659 {
660 	return sizeof(struct xfs_qoff_logitem) * 2;
661 }
662 
663 /*
664  * Syncing the incore super block changes to disk.
665  *     the super block to reflect the changes: sector size
666  */
667 STATIC uint
668 xfs_calc_sb_reservation(
669 	struct xfs_mount	*mp)
670 {
671 	return xfs_calc_buf_res(1, mp->m_sb.sb_sectsize);
672 }
673 
674 /*
675  * Initialize the precomputed transaction reservation values
676  * in the mount structure.
677  */
678 void
679 xfs_trans_init(
680 	struct xfs_mount	*mp)
681 {
682 	struct xfs_trans_reservations *resp = &mp->m_reservations;
683 
684 	resp->tr_write = xfs_calc_write_reservation(mp);
685 	resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
686 	resp->tr_rename = xfs_calc_rename_reservation(mp);
687 	resp->tr_link = xfs_calc_link_reservation(mp);
688 	resp->tr_remove = xfs_calc_remove_reservation(mp);
689 	resp->tr_symlink = xfs_calc_symlink_reservation(mp);
690 	resp->tr_create = xfs_calc_create_reservation(mp);
691 	resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
692 	resp->tr_ifree = xfs_calc_ifree_reservation(mp);
693 	resp->tr_ichange = xfs_calc_ichange_reservation(mp);
694 	resp->tr_growdata = xfs_calc_growdata_reservation(mp);
695 	resp->tr_swrite = xfs_calc_swrite_reservation(mp);
696 	resp->tr_writeid = xfs_calc_writeid_reservation(mp);
697 	resp->tr_addafork = xfs_calc_addafork_reservation(mp);
698 	resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
699 	resp->tr_attrsetm = xfs_calc_attrsetm_reservation(mp);
700 	resp->tr_attrsetrt = xfs_calc_attrsetrt_reservation(mp);
701 	resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
702 	resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
703 	resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
704 	resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
705 	resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
706 	resp->tr_qm_sbchange = xfs_calc_qm_sbchange_reservation(mp);
707 	resp->tr_qm_setqlim = xfs_calc_qm_setqlim_reservation(mp);
708 	resp->tr_qm_dqalloc = xfs_calc_qm_dqalloc_reservation(mp);
709 	resp->tr_qm_quotaoff = xfs_calc_qm_quotaoff_reservation(mp);
710 	resp->tr_qm_equotaoff = xfs_calc_qm_quotaoff_end_reservation(mp);
711 	resp->tr_sb = xfs_calc_sb_reservation(mp);
712 }
713 
714 /*
715  * This routine is called to allocate a transaction structure.
716  * The type parameter indicates the type of the transaction.  These
717  * are enumerated in xfs_trans.h.
718  *
719  * Dynamically allocate the transaction structure from the transaction
720  * zone, initialize it, and return it to the caller.
721  */
722 xfs_trans_t *
723 xfs_trans_alloc(
724 	xfs_mount_t	*mp,
725 	uint		type)
726 {
727 	xfs_trans_t     *tp;
728 
729 	sb_start_intwrite(mp->m_super);
730 	tp = _xfs_trans_alloc(mp, type, KM_SLEEP);
731 	tp->t_flags |= XFS_TRANS_FREEZE_PROT;
732 	return tp;
733 }
734 
735 xfs_trans_t *
736 _xfs_trans_alloc(
737 	xfs_mount_t	*mp,
738 	uint		type,
739 	xfs_km_flags_t	memflags)
740 {
741 	xfs_trans_t	*tp;
742 
743 	WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
744 	atomic_inc(&mp->m_active_trans);
745 
746 	tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
747 	tp->t_magic = XFS_TRANS_MAGIC;
748 	tp->t_type = type;
749 	tp->t_mountp = mp;
750 	INIT_LIST_HEAD(&tp->t_items);
751 	INIT_LIST_HEAD(&tp->t_busy);
752 	return tp;
753 }
754 
755 /*
756  * Free the transaction structure.  If there is more clean up
757  * to do when the structure is freed, add it here.
758  */
759 STATIC void
760 xfs_trans_free(
761 	struct xfs_trans	*tp)
762 {
763 	xfs_extent_busy_sort(&tp->t_busy);
764 	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
765 
766 	atomic_dec(&tp->t_mountp->m_active_trans);
767 	if (tp->t_flags & XFS_TRANS_FREEZE_PROT)
768 		sb_end_intwrite(tp->t_mountp->m_super);
769 	xfs_trans_free_dqinfo(tp);
770 	kmem_zone_free(xfs_trans_zone, tp);
771 }
772 
773 /*
774  * This is called to create a new transaction which will share the
775  * permanent log reservation of the given transaction.  The remaining
776  * unused block and rt extent reservations are also inherited.  This
777  * implies that the original transaction is no longer allowed to allocate
778  * blocks.  Locks and log items, however, are no inherited.  They must
779  * be added to the new transaction explicitly.
780  */
781 xfs_trans_t *
782 xfs_trans_dup(
783 	xfs_trans_t	*tp)
784 {
785 	xfs_trans_t	*ntp;
786 
787 	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
788 
789 	/*
790 	 * Initialize the new transaction structure.
791 	 */
792 	ntp->t_magic = XFS_TRANS_MAGIC;
793 	ntp->t_type = tp->t_type;
794 	ntp->t_mountp = tp->t_mountp;
795 	INIT_LIST_HEAD(&ntp->t_items);
796 	INIT_LIST_HEAD(&ntp->t_busy);
797 
798 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
799 	ASSERT(tp->t_ticket != NULL);
800 
801 	ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
802 		       (tp->t_flags & XFS_TRANS_RESERVE) |
803 		       (tp->t_flags & XFS_TRANS_FREEZE_PROT);
804 	/* We gave our writer reference to the new transaction */
805 	tp->t_flags &= ~XFS_TRANS_FREEZE_PROT;
806 	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
807 	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
808 	tp->t_blk_res = tp->t_blk_res_used;
809 	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
810 	tp->t_rtx_res = tp->t_rtx_res_used;
811 	ntp->t_pflags = tp->t_pflags;
812 
813 	xfs_trans_dup_dqinfo(tp, ntp);
814 
815 	atomic_inc(&tp->t_mountp->m_active_trans);
816 	return ntp;
817 }
818 
819 /*
820  * This is called to reserve free disk blocks and log space for the
821  * given transaction.  This must be done before allocating any resources
822  * within the transaction.
823  *
824  * This will return ENOSPC if there are not enough blocks available.
825  * It will sleep waiting for available log space.
826  * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
827  * is used by long running transactions.  If any one of the reservations
828  * fails then they will all be backed out.
829  *
830  * This does not do quota reservations. That typically is done by the
831  * caller afterwards.
832  */
833 int
834 xfs_trans_reserve(
835 	xfs_trans_t	*tp,
836 	uint		blocks,
837 	uint		logspace,
838 	uint		rtextents,
839 	uint		flags,
840 	uint		logcount)
841 {
842 	int		error = 0;
843 	int		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
844 
845 	/* Mark this thread as being in a transaction */
846 	current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
847 
848 	/*
849 	 * Attempt to reserve the needed disk blocks by decrementing
850 	 * the number needed from the number available.  This will
851 	 * fail if the count would go below zero.
852 	 */
853 	if (blocks > 0) {
854 		error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
855 					  -((int64_t)blocks), rsvd);
856 		if (error != 0) {
857 			current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
858 			return (XFS_ERROR(ENOSPC));
859 		}
860 		tp->t_blk_res += blocks;
861 	}
862 
863 	/*
864 	 * Reserve the log space needed for this transaction.
865 	 */
866 	if (logspace > 0) {
867 		bool	permanent = false;
868 
869 		ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace);
870 		ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount);
871 
872 		if (flags & XFS_TRANS_PERM_LOG_RES) {
873 			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
874 			permanent = true;
875 		} else {
876 			ASSERT(tp->t_ticket == NULL);
877 			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
878 		}
879 
880 		if (tp->t_ticket != NULL) {
881 			ASSERT(flags & XFS_TRANS_PERM_LOG_RES);
882 			error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
883 		} else {
884 			error = xfs_log_reserve(tp->t_mountp, logspace,
885 						logcount, &tp->t_ticket,
886 						XFS_TRANSACTION, permanent,
887 						tp->t_type);
888 		}
889 
890 		if (error)
891 			goto undo_blocks;
892 
893 		tp->t_log_res = logspace;
894 		tp->t_log_count = logcount;
895 	}
896 
897 	/*
898 	 * Attempt to reserve the needed realtime extents by decrementing
899 	 * the number needed from the number available.  This will
900 	 * fail if the count would go below zero.
901 	 */
902 	if (rtextents > 0) {
903 		error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
904 					  -((int64_t)rtextents), rsvd);
905 		if (error) {
906 			error = XFS_ERROR(ENOSPC);
907 			goto undo_log;
908 		}
909 		tp->t_rtx_res += rtextents;
910 	}
911 
912 	return 0;
913 
914 	/*
915 	 * Error cases jump to one of these labels to undo any
916 	 * reservations which have already been performed.
917 	 */
918 undo_log:
919 	if (logspace > 0) {
920 		int		log_flags;
921 
922 		if (flags & XFS_TRANS_PERM_LOG_RES) {
923 			log_flags = XFS_LOG_REL_PERM_RESERV;
924 		} else {
925 			log_flags = 0;
926 		}
927 		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
928 		tp->t_ticket = NULL;
929 		tp->t_log_res = 0;
930 		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
931 	}
932 
933 undo_blocks:
934 	if (blocks > 0) {
935 		xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
936 					 (int64_t)blocks, rsvd);
937 		tp->t_blk_res = 0;
938 	}
939 
940 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
941 
942 	return error;
943 }
944 
945 /*
946  * Record the indicated change to the given field for application
947  * to the file system's superblock when the transaction commits.
948  * For now, just store the change in the transaction structure.
949  *
950  * Mark the transaction structure to indicate that the superblock
951  * needs to be updated before committing.
952  *
953  * Because we may not be keeping track of allocated/free inodes and
954  * used filesystem blocks in the superblock, we do not mark the
955  * superblock dirty in this transaction if we modify these fields.
956  * We still need to update the transaction deltas so that they get
957  * applied to the incore superblock, but we don't want them to
958  * cause the superblock to get locked and logged if these are the
959  * only fields in the superblock that the transaction modifies.
960  */
961 void
962 xfs_trans_mod_sb(
963 	xfs_trans_t	*tp,
964 	uint		field,
965 	int64_t		delta)
966 {
967 	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
968 	xfs_mount_t	*mp = tp->t_mountp;
969 
970 	switch (field) {
971 	case XFS_TRANS_SB_ICOUNT:
972 		tp->t_icount_delta += delta;
973 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
974 			flags &= ~XFS_TRANS_SB_DIRTY;
975 		break;
976 	case XFS_TRANS_SB_IFREE:
977 		tp->t_ifree_delta += delta;
978 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
979 			flags &= ~XFS_TRANS_SB_DIRTY;
980 		break;
981 	case XFS_TRANS_SB_FDBLOCKS:
982 		/*
983 		 * Track the number of blocks allocated in the
984 		 * transaction.  Make sure it does not exceed the
985 		 * number reserved.
986 		 */
987 		if (delta < 0) {
988 			tp->t_blk_res_used += (uint)-delta;
989 			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
990 		}
991 		tp->t_fdblocks_delta += delta;
992 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
993 			flags &= ~XFS_TRANS_SB_DIRTY;
994 		break;
995 	case XFS_TRANS_SB_RES_FDBLOCKS:
996 		/*
997 		 * The allocation has already been applied to the
998 		 * in-core superblock's counter.  This should only
999 		 * be applied to the on-disk superblock.
1000 		 */
1001 		ASSERT(delta < 0);
1002 		tp->t_res_fdblocks_delta += delta;
1003 		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
1004 			flags &= ~XFS_TRANS_SB_DIRTY;
1005 		break;
1006 	case XFS_TRANS_SB_FREXTENTS:
1007 		/*
1008 		 * Track the number of blocks allocated in the
1009 		 * transaction.  Make sure it does not exceed the
1010 		 * number reserved.
1011 		 */
1012 		if (delta < 0) {
1013 			tp->t_rtx_res_used += (uint)-delta;
1014 			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
1015 		}
1016 		tp->t_frextents_delta += delta;
1017 		break;
1018 	case XFS_TRANS_SB_RES_FREXTENTS:
1019 		/*
1020 		 * The allocation has already been applied to the
1021 		 * in-core superblock's counter.  This should only
1022 		 * be applied to the on-disk superblock.
1023 		 */
1024 		ASSERT(delta < 0);
1025 		tp->t_res_frextents_delta += delta;
1026 		break;
1027 	case XFS_TRANS_SB_DBLOCKS:
1028 		ASSERT(delta > 0);
1029 		tp->t_dblocks_delta += delta;
1030 		break;
1031 	case XFS_TRANS_SB_AGCOUNT:
1032 		ASSERT(delta > 0);
1033 		tp->t_agcount_delta += delta;
1034 		break;
1035 	case XFS_TRANS_SB_IMAXPCT:
1036 		tp->t_imaxpct_delta += delta;
1037 		break;
1038 	case XFS_TRANS_SB_REXTSIZE:
1039 		tp->t_rextsize_delta += delta;
1040 		break;
1041 	case XFS_TRANS_SB_RBMBLOCKS:
1042 		tp->t_rbmblocks_delta += delta;
1043 		break;
1044 	case XFS_TRANS_SB_RBLOCKS:
1045 		tp->t_rblocks_delta += delta;
1046 		break;
1047 	case XFS_TRANS_SB_REXTENTS:
1048 		tp->t_rextents_delta += delta;
1049 		break;
1050 	case XFS_TRANS_SB_REXTSLOG:
1051 		tp->t_rextslog_delta += delta;
1052 		break;
1053 	default:
1054 		ASSERT(0);
1055 		return;
1056 	}
1057 
1058 	tp->t_flags |= flags;
1059 }
1060 
1061 /*
1062  * xfs_trans_apply_sb_deltas() is called from the commit code
1063  * to bring the superblock buffer into the current transaction
1064  * and modify it as requested by earlier calls to xfs_trans_mod_sb().
1065  *
1066  * For now we just look at each field allowed to change and change
1067  * it if necessary.
1068  */
1069 STATIC void
1070 xfs_trans_apply_sb_deltas(
1071 	xfs_trans_t	*tp)
1072 {
1073 	xfs_dsb_t	*sbp;
1074 	xfs_buf_t	*bp;
1075 	int		whole = 0;
1076 
1077 	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
1078 	sbp = XFS_BUF_TO_SBP(bp);
1079 
1080 	/*
1081 	 * Check that superblock mods match the mods made to AGF counters.
1082 	 */
1083 	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
1084 	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
1085 		tp->t_ag_btree_delta));
1086 
1087 	/*
1088 	 * Only update the superblock counters if we are logging them
1089 	 */
1090 	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
1091 		if (tp->t_icount_delta)
1092 			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
1093 		if (tp->t_ifree_delta)
1094 			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
1095 		if (tp->t_fdblocks_delta)
1096 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
1097 		if (tp->t_res_fdblocks_delta)
1098 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
1099 	}
1100 
1101 	if (tp->t_frextents_delta)
1102 		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
1103 	if (tp->t_res_frextents_delta)
1104 		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
1105 
1106 	if (tp->t_dblocks_delta) {
1107 		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
1108 		whole = 1;
1109 	}
1110 	if (tp->t_agcount_delta) {
1111 		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
1112 		whole = 1;
1113 	}
1114 	if (tp->t_imaxpct_delta) {
1115 		sbp->sb_imax_pct += tp->t_imaxpct_delta;
1116 		whole = 1;
1117 	}
1118 	if (tp->t_rextsize_delta) {
1119 		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
1120 		whole = 1;
1121 	}
1122 	if (tp->t_rbmblocks_delta) {
1123 		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
1124 		whole = 1;
1125 	}
1126 	if (tp->t_rblocks_delta) {
1127 		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
1128 		whole = 1;
1129 	}
1130 	if (tp->t_rextents_delta) {
1131 		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
1132 		whole = 1;
1133 	}
1134 	if (tp->t_rextslog_delta) {
1135 		sbp->sb_rextslog += tp->t_rextslog_delta;
1136 		whole = 1;
1137 	}
1138 
1139 	if (whole)
1140 		/*
1141 		 * Log the whole thing, the fields are noncontiguous.
1142 		 */
1143 		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
1144 	else
1145 		/*
1146 		 * Since all the modifiable fields are contiguous, we
1147 		 * can get away with this.
1148 		 */
1149 		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
1150 				  offsetof(xfs_dsb_t, sb_frextents) +
1151 				  sizeof(sbp->sb_frextents) - 1);
1152 }
1153 
1154 /*
1155  * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
1156  * and apply superblock counter changes to the in-core superblock.  The
1157  * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
1158  * applied to the in-core superblock.  The idea is that that has already been
1159  * done.
1160  *
1161  * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
1162  * However, we have to ensure that we only modify each superblock field only
1163  * once because the application of the delta values may not be atomic. That can
1164  * lead to ENOSPC races occurring if we have two separate modifcations of the
1165  * free space counter to put back the entire reservation and then take away
1166  * what we used.
1167  *
1168  * If we are not logging superblock counters, then the inode allocated/free and
1169  * used block counts are not updated in the on disk superblock. In this case,
1170  * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1171  * still need to update the incore superblock with the changes.
1172  */
1173 void
1174 xfs_trans_unreserve_and_mod_sb(
1175 	xfs_trans_t	*tp)
1176 {
1177 	xfs_mod_sb_t	msb[9];	/* If you add cases, add entries */
1178 	xfs_mod_sb_t	*msbp;
1179 	xfs_mount_t	*mp = tp->t_mountp;
1180 	/* REFERENCED */
1181 	int		error;
1182 	int		rsvd;
1183 	int64_t		blkdelta = 0;
1184 	int64_t		rtxdelta = 0;
1185 	int64_t		idelta = 0;
1186 	int64_t		ifreedelta = 0;
1187 
1188 	msbp = msb;
1189 	rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1190 
1191 	/* calculate deltas */
1192 	if (tp->t_blk_res > 0)
1193 		blkdelta = tp->t_blk_res;
1194 	if ((tp->t_fdblocks_delta != 0) &&
1195 	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1196 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1197 	        blkdelta += tp->t_fdblocks_delta;
1198 
1199 	if (tp->t_rtx_res > 0)
1200 		rtxdelta = tp->t_rtx_res;
1201 	if ((tp->t_frextents_delta != 0) &&
1202 	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
1203 		rtxdelta += tp->t_frextents_delta;
1204 
1205 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1206 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1207 		idelta = tp->t_icount_delta;
1208 		ifreedelta = tp->t_ifree_delta;
1209 	}
1210 
1211 	/* apply the per-cpu counters */
1212 	if (blkdelta) {
1213 		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1214 						 blkdelta, rsvd);
1215 		if (error)
1216 			goto out;
1217 	}
1218 
1219 	if (idelta) {
1220 		error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1221 						 idelta, rsvd);
1222 		if (error)
1223 			goto out_undo_fdblocks;
1224 	}
1225 
1226 	if (ifreedelta) {
1227 		error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1228 						 ifreedelta, rsvd);
1229 		if (error)
1230 			goto out_undo_icount;
1231 	}
1232 
1233 	/* apply remaining deltas */
1234 	if (rtxdelta != 0) {
1235 		msbp->msb_field = XFS_SBS_FREXTENTS;
1236 		msbp->msb_delta = rtxdelta;
1237 		msbp++;
1238 	}
1239 
1240 	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
1241 		if (tp->t_dblocks_delta != 0) {
1242 			msbp->msb_field = XFS_SBS_DBLOCKS;
1243 			msbp->msb_delta = tp->t_dblocks_delta;
1244 			msbp++;
1245 		}
1246 		if (tp->t_agcount_delta != 0) {
1247 			msbp->msb_field = XFS_SBS_AGCOUNT;
1248 			msbp->msb_delta = tp->t_agcount_delta;
1249 			msbp++;
1250 		}
1251 		if (tp->t_imaxpct_delta != 0) {
1252 			msbp->msb_field = XFS_SBS_IMAX_PCT;
1253 			msbp->msb_delta = tp->t_imaxpct_delta;
1254 			msbp++;
1255 		}
1256 		if (tp->t_rextsize_delta != 0) {
1257 			msbp->msb_field = XFS_SBS_REXTSIZE;
1258 			msbp->msb_delta = tp->t_rextsize_delta;
1259 			msbp++;
1260 		}
1261 		if (tp->t_rbmblocks_delta != 0) {
1262 			msbp->msb_field = XFS_SBS_RBMBLOCKS;
1263 			msbp->msb_delta = tp->t_rbmblocks_delta;
1264 			msbp++;
1265 		}
1266 		if (tp->t_rblocks_delta != 0) {
1267 			msbp->msb_field = XFS_SBS_RBLOCKS;
1268 			msbp->msb_delta = tp->t_rblocks_delta;
1269 			msbp++;
1270 		}
1271 		if (tp->t_rextents_delta != 0) {
1272 			msbp->msb_field = XFS_SBS_REXTENTS;
1273 			msbp->msb_delta = tp->t_rextents_delta;
1274 			msbp++;
1275 		}
1276 		if (tp->t_rextslog_delta != 0) {
1277 			msbp->msb_field = XFS_SBS_REXTSLOG;
1278 			msbp->msb_delta = tp->t_rextslog_delta;
1279 			msbp++;
1280 		}
1281 	}
1282 
1283 	/*
1284 	 * If we need to change anything, do it.
1285 	 */
1286 	if (msbp > msb) {
1287 		error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1288 			(uint)(msbp - msb), rsvd);
1289 		if (error)
1290 			goto out_undo_ifreecount;
1291 	}
1292 
1293 	return;
1294 
1295 out_undo_ifreecount:
1296 	if (ifreedelta)
1297 		xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1298 out_undo_icount:
1299 	if (idelta)
1300 		xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1301 out_undo_fdblocks:
1302 	if (blkdelta)
1303 		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1304 out:
1305 	ASSERT(error == 0);
1306 	return;
1307 }
1308 
1309 /*
1310  * Add the given log item to the transaction's list of log items.
1311  *
1312  * The log item will now point to its new descriptor with its li_desc field.
1313  */
1314 void
1315 xfs_trans_add_item(
1316 	struct xfs_trans	*tp,
1317 	struct xfs_log_item	*lip)
1318 {
1319 	struct xfs_log_item_desc *lidp;
1320 
1321 	ASSERT(lip->li_mountp == tp->t_mountp);
1322 	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
1323 
1324 	lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
1325 
1326 	lidp->lid_item = lip;
1327 	lidp->lid_flags = 0;
1328 	list_add_tail(&lidp->lid_trans, &tp->t_items);
1329 
1330 	lip->li_desc = lidp;
1331 }
1332 
1333 STATIC void
1334 xfs_trans_free_item_desc(
1335 	struct xfs_log_item_desc *lidp)
1336 {
1337 	list_del_init(&lidp->lid_trans);
1338 	kmem_zone_free(xfs_log_item_desc_zone, lidp);
1339 }
1340 
1341 /*
1342  * Unlink and free the given descriptor.
1343  */
1344 void
1345 xfs_trans_del_item(
1346 	struct xfs_log_item	*lip)
1347 {
1348 	xfs_trans_free_item_desc(lip->li_desc);
1349 	lip->li_desc = NULL;
1350 }
1351 
1352 /*
1353  * Unlock all of the items of a transaction and free all the descriptors
1354  * of that transaction.
1355  */
1356 void
1357 xfs_trans_free_items(
1358 	struct xfs_trans	*tp,
1359 	xfs_lsn_t		commit_lsn,
1360 	int			flags)
1361 {
1362 	struct xfs_log_item_desc *lidp, *next;
1363 
1364 	list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1365 		struct xfs_log_item	*lip = lidp->lid_item;
1366 
1367 		lip->li_desc = NULL;
1368 
1369 		if (commit_lsn != NULLCOMMITLSN)
1370 			IOP_COMMITTING(lip, commit_lsn);
1371 		if (flags & XFS_TRANS_ABORT)
1372 			lip->li_flags |= XFS_LI_ABORTED;
1373 		IOP_UNLOCK(lip);
1374 
1375 		xfs_trans_free_item_desc(lidp);
1376 	}
1377 }
1378 
1379 static inline void
1380 xfs_log_item_batch_insert(
1381 	struct xfs_ail		*ailp,
1382 	struct xfs_ail_cursor	*cur,
1383 	struct xfs_log_item	**log_items,
1384 	int			nr_items,
1385 	xfs_lsn_t		commit_lsn)
1386 {
1387 	int	i;
1388 
1389 	spin_lock(&ailp->xa_lock);
1390 	/* xfs_trans_ail_update_bulk drops ailp->xa_lock */
1391 	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
1392 
1393 	for (i = 0; i < nr_items; i++)
1394 		IOP_UNPIN(log_items[i], 0);
1395 }
1396 
1397 /*
1398  * Bulk operation version of xfs_trans_committed that takes a log vector of
1399  * items to insert into the AIL. This uses bulk AIL insertion techniques to
1400  * minimise lock traffic.
1401  *
1402  * If we are called with the aborted flag set, it is because a log write during
1403  * a CIL checkpoint commit has failed. In this case, all the items in the
1404  * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
1405  * means that checkpoint commit abort handling is treated exactly the same
1406  * as an iclog write error even though we haven't started any IO yet. Hence in
1407  * this case all we need to do is IOP_COMMITTED processing, followed by an
1408  * IOP_UNPIN(aborted) call.
1409  *
1410  * The AIL cursor is used to optimise the insert process. If commit_lsn is not
1411  * at the end of the AIL, the insert cursor avoids the need to walk
1412  * the AIL to find the insertion point on every xfs_log_item_batch_insert()
1413  * call. This saves a lot of needless list walking and is a net win, even
1414  * though it slightly increases that amount of AIL lock traffic to set it up
1415  * and tear it down.
1416  */
1417 void
1418 xfs_trans_committed_bulk(
1419 	struct xfs_ail		*ailp,
1420 	struct xfs_log_vec	*log_vector,
1421 	xfs_lsn_t		commit_lsn,
1422 	int			aborted)
1423 {
1424 #define LOG_ITEM_BATCH_SIZE	32
1425 	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
1426 	struct xfs_log_vec	*lv;
1427 	struct xfs_ail_cursor	cur;
1428 	int			i = 0;
1429 
1430 	spin_lock(&ailp->xa_lock);
1431 	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
1432 	spin_unlock(&ailp->xa_lock);
1433 
1434 	/* unpin all the log items */
1435 	for (lv = log_vector; lv; lv = lv->lv_next ) {
1436 		struct xfs_log_item	*lip = lv->lv_item;
1437 		xfs_lsn_t		item_lsn;
1438 
1439 		if (aborted)
1440 			lip->li_flags |= XFS_LI_ABORTED;
1441 		item_lsn = IOP_COMMITTED(lip, commit_lsn);
1442 
1443 		/* item_lsn of -1 means the item needs no further processing */
1444 		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1445 			continue;
1446 
1447 		/*
1448 		 * if we are aborting the operation, no point in inserting the
1449 		 * object into the AIL as we are in a shutdown situation.
1450 		 */
1451 		if (aborted) {
1452 			ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
1453 			IOP_UNPIN(lip, 1);
1454 			continue;
1455 		}
1456 
1457 		if (item_lsn != commit_lsn) {
1458 
1459 			/*
1460 			 * Not a bulk update option due to unusual item_lsn.
1461 			 * Push into AIL immediately, rechecking the lsn once
1462 			 * we have the ail lock. Then unpin the item. This does
1463 			 * not affect the AIL cursor the bulk insert path is
1464 			 * using.
1465 			 */
1466 			spin_lock(&ailp->xa_lock);
1467 			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1468 				xfs_trans_ail_update(ailp, lip, item_lsn);
1469 			else
1470 				spin_unlock(&ailp->xa_lock);
1471 			IOP_UNPIN(lip, 0);
1472 			continue;
1473 		}
1474 
1475 		/* Item is a candidate for bulk AIL insert.  */
1476 		log_items[i++] = lv->lv_item;
1477 		if (i >= LOG_ITEM_BATCH_SIZE) {
1478 			xfs_log_item_batch_insert(ailp, &cur, log_items,
1479 					LOG_ITEM_BATCH_SIZE, commit_lsn);
1480 			i = 0;
1481 		}
1482 	}
1483 
1484 	/* make sure we insert the remainder! */
1485 	if (i)
1486 		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
1487 
1488 	spin_lock(&ailp->xa_lock);
1489 	xfs_trans_ail_cursor_done(ailp, &cur);
1490 	spin_unlock(&ailp->xa_lock);
1491 }
1492 
1493 /*
1494  * Commit the given transaction to the log.
1495  *
1496  * XFS disk error handling mechanism is not based on a typical
1497  * transaction abort mechanism. Logically after the filesystem
1498  * gets marked 'SHUTDOWN', we can't let any new transactions
1499  * be durable - ie. committed to disk - because some metadata might
1500  * be inconsistent. In such cases, this returns an error, and the
1501  * caller may assume that all locked objects joined to the transaction
1502  * have already been unlocked as if the commit had succeeded.
1503  * Do not reference the transaction structure after this call.
1504  */
1505 int
1506 xfs_trans_commit(
1507 	struct xfs_trans	*tp,
1508 	uint			flags)
1509 {
1510 	struct xfs_mount	*mp = tp->t_mountp;
1511 	xfs_lsn_t		commit_lsn = -1;
1512 	int			error = 0;
1513 	int			log_flags = 0;
1514 	int			sync = tp->t_flags & XFS_TRANS_SYNC;
1515 
1516 	/*
1517 	 * Determine whether this commit is releasing a permanent
1518 	 * log reservation or not.
1519 	 */
1520 	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1521 		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1522 		log_flags = XFS_LOG_REL_PERM_RESERV;
1523 	}
1524 
1525 	/*
1526 	 * If there is nothing to be logged by the transaction,
1527 	 * then unlock all of the items associated with the
1528 	 * transaction and free the transaction structure.
1529 	 * Also make sure to return any reserved blocks to
1530 	 * the free pool.
1531 	 */
1532 	if (!(tp->t_flags & XFS_TRANS_DIRTY))
1533 		goto out_unreserve;
1534 
1535 	if (XFS_FORCED_SHUTDOWN(mp)) {
1536 		error = XFS_ERROR(EIO);
1537 		goto out_unreserve;
1538 	}
1539 
1540 	ASSERT(tp->t_ticket != NULL);
1541 
1542 	/*
1543 	 * If we need to update the superblock, then do it now.
1544 	 */
1545 	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1546 		xfs_trans_apply_sb_deltas(tp);
1547 	xfs_trans_apply_dquot_deltas(tp);
1548 
1549 	error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags);
1550 	if (error == ENOMEM) {
1551 		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1552 		error = XFS_ERROR(EIO);
1553 		goto out_unreserve;
1554 	}
1555 
1556 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1557 	xfs_trans_free(tp);
1558 
1559 	/*
1560 	 * If the transaction needs to be synchronous, then force the
1561 	 * log out now and wait for it.
1562 	 */
1563 	if (sync) {
1564 		if (!error) {
1565 			error = _xfs_log_force_lsn(mp, commit_lsn,
1566 				      XFS_LOG_SYNC, NULL);
1567 		}
1568 		XFS_STATS_INC(xs_trans_sync);
1569 	} else {
1570 		XFS_STATS_INC(xs_trans_async);
1571 	}
1572 
1573 	return error;
1574 
1575 out_unreserve:
1576 	xfs_trans_unreserve_and_mod_sb(tp);
1577 
1578 	/*
1579 	 * It is indeed possible for the transaction to be not dirty but
1580 	 * the dqinfo portion to be.  All that means is that we have some
1581 	 * (non-persistent) quota reservations that need to be unreserved.
1582 	 */
1583 	xfs_trans_unreserve_and_mod_dquots(tp);
1584 	if (tp->t_ticket) {
1585 		commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1586 		if (commit_lsn == -1 && !error)
1587 			error = XFS_ERROR(EIO);
1588 	}
1589 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1590 	xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
1591 	xfs_trans_free(tp);
1592 
1593 	XFS_STATS_INC(xs_trans_empty);
1594 	return error;
1595 }
1596 
1597 /*
1598  * Unlock all of the transaction's items and free the transaction.
1599  * The transaction must not have modified any of its items, because
1600  * there is no way to restore them to their previous state.
1601  *
1602  * If the transaction has made a log reservation, make sure to release
1603  * it as well.
1604  */
1605 void
1606 xfs_trans_cancel(
1607 	xfs_trans_t		*tp,
1608 	int			flags)
1609 {
1610 	int			log_flags;
1611 	xfs_mount_t		*mp = tp->t_mountp;
1612 
1613 	/*
1614 	 * See if the caller is being too lazy to figure out if
1615 	 * the transaction really needs an abort.
1616 	 */
1617 	if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1618 		flags &= ~XFS_TRANS_ABORT;
1619 	/*
1620 	 * See if the caller is relying on us to shut down the
1621 	 * filesystem.  This happens in paths where we detect
1622 	 * corruption and decide to give up.
1623 	 */
1624 	if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1625 		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1626 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1627 	}
1628 #ifdef DEBUG
1629 	if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1630 		struct xfs_log_item_desc *lidp;
1631 
1632 		list_for_each_entry(lidp, &tp->t_items, lid_trans)
1633 			ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1634 	}
1635 #endif
1636 	xfs_trans_unreserve_and_mod_sb(tp);
1637 	xfs_trans_unreserve_and_mod_dquots(tp);
1638 
1639 	if (tp->t_ticket) {
1640 		if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1641 			ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1642 			log_flags = XFS_LOG_REL_PERM_RESERV;
1643 		} else {
1644 			log_flags = 0;
1645 		}
1646 		xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1647 	}
1648 
1649 	/* mark this thread as no longer being in a transaction */
1650 	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1651 
1652 	xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1653 	xfs_trans_free(tp);
1654 }
1655 
1656 /*
1657  * Roll from one trans in the sequence of PERMANENT transactions to
1658  * the next: permanent transactions are only flushed out when
1659  * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1660  * as possible to let chunks of it go to the log. So we commit the
1661  * chunk we've been working on and get a new transaction to continue.
1662  */
1663 int
1664 xfs_trans_roll(
1665 	struct xfs_trans	**tpp,
1666 	struct xfs_inode	*dp)
1667 {
1668 	struct xfs_trans	*trans;
1669 	unsigned int		logres, count;
1670 	int			error;
1671 
1672 	/*
1673 	 * Ensure that the inode is always logged.
1674 	 */
1675 	trans = *tpp;
1676 	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1677 
1678 	/*
1679 	 * Copy the critical parameters from one trans to the next.
1680 	 */
1681 	logres = trans->t_log_res;
1682 	count = trans->t_log_count;
1683 	*tpp = xfs_trans_dup(trans);
1684 
1685 	/*
1686 	 * Commit the current transaction.
1687 	 * If this commit failed, then it'd just unlock those items that
1688 	 * are not marked ihold. That also means that a filesystem shutdown
1689 	 * is in progress. The caller takes the responsibility to cancel
1690 	 * the duplicate transaction that gets returned.
1691 	 */
1692 	error = xfs_trans_commit(trans, 0);
1693 	if (error)
1694 		return (error);
1695 
1696 	trans = *tpp;
1697 
1698 	/*
1699 	 * transaction commit worked ok so we can drop the extra ticket
1700 	 * reference that we gained in xfs_trans_dup()
1701 	 */
1702 	xfs_log_ticket_put(trans->t_ticket);
1703 
1704 
1705 	/*
1706 	 * Reserve space in the log for th next transaction.
1707 	 * This also pushes items in the "AIL", the list of logged items,
1708 	 * out to disk if they are taking up space at the tail of the log
1709 	 * that we want to use.  This requires that either nothing be locked
1710 	 * across this call, or that anything that is locked be logged in
1711 	 * the prior and the next transactions.
1712 	 */
1713 	error = xfs_trans_reserve(trans, 0, logres, 0,
1714 				  XFS_TRANS_PERM_LOG_RES, count);
1715 	/*
1716 	 *  Ensure that the inode is in the new transaction and locked.
1717 	 */
1718 	if (error)
1719 		return error;
1720 
1721 	xfs_trans_ijoin(trans, dp, 0);
1722 	return 0;
1723 }
1724