blob: fdca7416c754636a26dae9a312003932f7306d9f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
Christoph Hellwige98c4142010-06-23 18:11:15 +10003 * Copyright (C) 2010 Red Hat, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11004 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
Nathan Scott7b718762005-11-02 14:58:39 +11006 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * published by the Free Software Foundation.
9 *
Nathan Scott7b718762005-11-02 14:58:39 +110010 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
Nathan Scott7b718762005-11-02 14:58:39 +110015 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
29#include "xfs_error.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_dinode.h"
35#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_btree.h"
37#include "xfs_ialloc.h"
38#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_quota.h"
Nathan Scotta844f452005-11-02 14:38:42 +110041#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_trans_space.h"
Niv Sardi322ff6b2008-08-13 16:05:49 +100043#include "xfs_inode_item.h"
Dave Chinnered3b4d62010-05-21 12:07:08 +100044#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nathan Scott8f794052006-03-14 13:32:41 +110046kmem_zone_t *xfs_trans_zone;
Christoph Hellwige98c4142010-06-23 18:11:15 +100047kmem_zone_t *xfs_log_item_desc_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Christoph Hellwig025101d2010-05-04 13:53:48 +000049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
Christoph Hellwig025101d2010-05-04 13:53:48 +000051 * 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
Nathan Scott8f794052006-03-14 13:32:41 +110083 */
Nathan Scott8f794052006-03-14 13:32:41 +110084STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +000085xfs_calc_write_reservation(
86 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +110087{
Christoph Hellwig025101d2010-05-04 13:53:48 +000088 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100101}
102
Christoph Hellwig025101d2010-05-04 13:53:48 +0000103/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100116STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000117xfs_calc_itruncate_reservation(
118 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100119{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000120 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100133}
134
Christoph Hellwig025101d2010-05-04 13:53:48 +0000135/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100147STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000148xfs_calc_rename_reservation(
149 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100150{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000151 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100160}
161
Christoph Hellwig025101d2010-05-04 13:53:48 +0000162/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100174STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000175xfs_calc_link_reservation(
176 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100177{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000178 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100188}
189
Christoph Hellwig025101d2010-05-04 13:53:48 +0000190/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100202STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000203xfs_calc_remove_reservation(
204 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100205{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000206 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100216}
217
Christoph Hellwig025101d2010-05-04 13:53:48 +0000218/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100232STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000233xfs_calc_symlink_reservation(
234 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100235{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000236 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100249}
250
Christoph Hellwig025101d2010-05-04 13:53:48 +0000251/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100266STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000267xfs_calc_create_reservation(
268 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100269{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000270 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100283}
284
Christoph Hellwig025101d2010-05-04 13:53:48 +0000285/*
286 * Making a new directory is the same as creating a new file.
287 */
Nathan Scott8f794052006-03-14 13:32:41 +1100288STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000289xfs_calc_mkdir_reservation(
290 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100291{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000292 return xfs_calc_create_reservation(mp);
Nathan Scott8f794052006-03-14 13:32:41 +1100293}
294
Christoph Hellwig025101d2010-05-04 13:53:48 +0000295/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100305STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000306xfs_calc_ifree_reservation(
307 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100308{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000309 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));
Nathan Scott8f794052006-03-14 13:32:41 +1100320}
321
Christoph Hellwig025101d2010-05-04 13:53:48 +0000322/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100326STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000327xfs_calc_ichange_reservation(
328 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100329{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000330 return XFS_DQUOT_LOGRES(mp) +
331 mp->m_sb.sb_inodesize +
332 mp->m_sb.sb_sectsize +
333 512;
334
Nathan Scott8f794052006-03-14 13:32:41 +1100335}
336
Christoph Hellwig025101d2010-05-04 13:53:48 +0000337/*
338 * Growing the data section of the filesystem.
339 * superblock
340 * agi and agf
341 * allocation btrees
342 */
Nathan Scott8f794052006-03-14 13:32:41 +1100343STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000344xfs_calc_growdata_reservation(
345 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100346{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000347 return mp->m_sb.sb_sectsize * 3 +
348 XFS_ALLOCFREE_LOG_RES(mp, 1) +
349 128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1));
Nathan Scott8f794052006-03-14 13:32:41 +1100350}
351
Christoph Hellwig025101d2010-05-04 13:53:48 +0000352/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100362STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000363xfs_calc_growrtalloc_reservation(
364 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100365{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000366 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));
Nathan Scott8f794052006-03-14 13:32:41 +1100372}
373
Christoph Hellwig025101d2010-05-04 13:53:48 +0000374/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100379STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000380xfs_calc_growrtzero_reservation(
381 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100382{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000383 return mp->m_sb.sb_blocksize + 128;
Nathan Scott8f794052006-03-14 13:32:41 +1100384}
385
Christoph Hellwig025101d2010-05-04 13:53:48 +0000386/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100396STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000397xfs_calc_growrtfree_reservation(
398 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100399{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000400 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;
Nathan Scott8f794052006-03-14 13:32:41 +1100405}
406
Christoph Hellwig025101d2010-05-04 13:53:48 +0000407/*
408 * Logging the inode modification timestamp on a synchronous write.
409 * inode
410 */
Nathan Scott8f794052006-03-14 13:32:41 +1100411STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000412xfs_calc_swrite_reservation(
413 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100414{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000415 return mp->m_sb.sb_inodesize + 128;
Nathan Scott8f794052006-03-14 13:32:41 +1100416}
417
Christoph Hellwig025101d2010-05-04 13:53:48 +0000418/*
419 * Logging the inode mode bits when writing a setuid/setgid file
420 * inode
421 */
Nathan Scott8f794052006-03-14 13:32:41 +1100422STATIC uint
423xfs_calc_writeid_reservation(xfs_mount_t *mp)
424{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000425 return mp->m_sb.sb_inodesize + 128;
Nathan Scott8f794052006-03-14 13:32:41 +1100426}
427
Christoph Hellwig025101d2010-05-04 13:53:48 +0000428/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100436STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000437xfs_calc_addafork_reservation(
438 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100439{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000440 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));
Nathan Scott8f794052006-03-14 13:32:41 +1100448}
449
Christoph Hellwig025101d2010-05-04 13:53:48 +0000450/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100461STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000462xfs_calc_attrinval_reservation(
463 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100464{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000465 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100473}
474
Christoph Hellwig025101d2010-05-04 13:53:48 +0000475/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100485STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000486xfs_calc_attrset_reservation(
487 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100488{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000489 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);
Nathan Scott8f794052006-03-14 13:32:41 +1100494}
495
Christoph Hellwig025101d2010-05-04 13:53:48 +0000496/*
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 */
Nathan Scott8f794052006-03-14 13:32:41 +1100507STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000508xfs_calc_attrrm_reservation(
509 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100510{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000511 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))));
Nathan Scott8f794052006-03-14 13:32:41 +1100522}
523
Christoph Hellwig025101d2010-05-04 13:53:48 +0000524/*
525 * Clearing a bad agino number in an agi hash bucket.
526 */
Nathan Scott8f794052006-03-14 13:32:41 +1100527STATIC uint
Christoph Hellwig025101d2010-05-04 13:53:48 +0000528xfs_calc_clear_agi_bucket_reservation(
529 struct xfs_mount *mp)
Nathan Scott8f794052006-03-14 13:32:41 +1100530{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000531 return mp->m_sb.sb_sectsize + 128;
Nathan Scott8f794052006-03-14 13:32:41 +1100532}
533
534/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 * Initialize the precomputed transaction reservation values
536 * in the mount structure.
537 */
538void
539xfs_trans_init(
Christoph Hellwig025101d2010-05-04 13:53:48 +0000540 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Christoph Hellwig025101d2010-05-04 13:53:48 +0000542 struct xfs_trans_reservations *resp = &mp->m_reservations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Nathan Scott8f794052006-03-14 13:32:41 +1100544 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
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 */
575xfs_trans_t *
576xfs_trans_alloc(
577 xfs_mount_t *mp,
578 uint type)
579{
Christoph Hellwigb267ce92007-08-30 17:21:30 +1000580 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000581 return _xfs_trans_alloc(mp, type, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
584xfs_trans_t *
585_xfs_trans_alloc(
586 xfs_mount_t *mp,
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000587 uint type,
588 uint memflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
590 xfs_trans_t *tp;
591
Nathan Scott34327e12006-06-09 17:11:55 +1000592 atomic_inc(&mp->m_active_trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Christoph Hellwig80641dc2009-10-19 04:00:03 +0000594 tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 tp->t_magic = XFS_TRANS_MAGIC;
596 tp->t_type = type;
597 tp->t_mountp = mp;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000598 INIT_LIST_HEAD(&tp->t_items);
Dave Chinnered3b4d62010-05-21 12:07:08 +1000599 INIT_LIST_HEAD(&tp->t_busy);
Nathan Scott34327e12006-06-09 17:11:55 +1000600 return tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
603/*
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +1100604 * Free the transaction structure. If there is more clean up
605 * to do when the structure is freed, add it here.
606 */
607STATIC void
608xfs_trans_free(
Dave Chinnered3b4d62010-05-21 12:07:08 +1000609 struct xfs_trans *tp)
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +1100610{
Dave Chinnered3b4d62010-05-21 12:07:08 +1000611 struct xfs_busy_extent *busyp, *n;
612
613 list_for_each_entry_safe(busyp, n, &tp->t_busy, list)
614 xfs_alloc_busy_clear(tp->t_mountp, busyp);
615
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +1100616 atomic_dec(&tp->t_mountp->m_active_trans);
617 xfs_trans_free_dqinfo(tp);
618 kmem_zone_free(xfs_trans_zone, tp);
619}
620
621/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 * This is called to create a new transaction which will share the
623 * permanent log reservation of the given transaction. The remaining
624 * unused block and rt extent reservations are also inherited. This
625 * implies that the original transaction is no longer allowed to allocate
626 * blocks. Locks and log items, however, are no inherited. They must
627 * be added to the new transaction explicitly.
628 */
629xfs_trans_t *
630xfs_trans_dup(
631 xfs_trans_t *tp)
632{
633 xfs_trans_t *ntp;
634
635 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
636
637 /*
638 * Initialize the new transaction structure.
639 */
640 ntp->t_magic = XFS_TRANS_MAGIC;
641 ntp->t_type = tp->t_type;
642 ntp->t_mountp = tp->t_mountp;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000643 INIT_LIST_HEAD(&ntp->t_items);
Dave Chinnered3b4d62010-05-21 12:07:08 +1000644 INIT_LIST_HEAD(&ntp->t_busy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 ASSERT(tp->t_ticket != NULL);
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100650 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
652 tp->t_blk_res = tp->t_blk_res_used;
653 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
654 tp->t_rtx_res = tp->t_rtx_res_used;
Nathan Scott59c1b082006-06-09 14:59:13 +1000655 ntp->t_pflags = tp->t_pflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Christoph Hellwig7d095252009-06-08 15:33:32 +0200657 xfs_trans_dup_dqinfo(tp, ntp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 atomic_inc(&tp->t_mountp->m_active_trans);
660 return ntp;
661}
662
663/*
664 * This is called to reserve free disk blocks and log space for the
665 * given transaction. This must be done before allocating any resources
666 * within the transaction.
667 *
668 * This will return ENOSPC if there are not enough blocks available.
669 * It will sleep waiting for available log space.
670 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
671 * is used by long running transactions. If any one of the reservations
672 * fails then they will all be backed out.
673 *
674 * This does not do quota reservations. That typically is done by the
675 * caller afterwards.
676 */
677int
678xfs_trans_reserve(
679 xfs_trans_t *tp,
680 uint blocks,
681 uint logspace,
682 uint rtextents,
683 uint flags,
684 uint logcount)
685{
686 int log_flags;
Nathan Scott59c1b082006-06-09 14:59:13 +1000687 int error = 0;
688 int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 /* Mark this thread as being in a transaction */
Nathan Scott59c1b082006-06-09 14:59:13 +1000691 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 /*
694 * Attempt to reserve the needed disk blocks by decrementing
695 * the number needed from the number available. This will
696 * fail if the count would go below zero.
697 */
698 if (blocks > 0) {
699 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100700 -((int64_t)blocks), rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (error != 0) {
Nathan Scott59c1b082006-06-09 14:59:13 +1000702 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return (XFS_ERROR(ENOSPC));
704 }
705 tp->t_blk_res += blocks;
706 }
707
708 /*
709 * Reserve the log space needed for this transaction.
710 */
711 if (logspace > 0) {
712 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
713 ASSERT((tp->t_log_count == 0) ||
714 (tp->t_log_count == logcount));
715 if (flags & XFS_TRANS_PERM_LOG_RES) {
716 log_flags = XFS_LOG_PERM_RESERV;
717 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
718 } else {
719 ASSERT(tp->t_ticket == NULL);
720 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
721 log_flags = 0;
722 }
723
724 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
725 &tp->t_ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000726 XFS_TRANSACTION, log_flags, tp->t_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if (error) {
728 goto undo_blocks;
729 }
730 tp->t_log_res = logspace;
731 tp->t_log_count = logcount;
732 }
733
734 /*
735 * Attempt to reserve the needed realtime extents by decrementing
736 * the number needed from the number available. This will
737 * fail if the count would go below zero.
738 */
739 if (rtextents > 0) {
740 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100741 -((int64_t)rtextents), rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (error) {
743 error = XFS_ERROR(ENOSPC);
744 goto undo_log;
745 }
746 tp->t_rtx_res += rtextents;
747 }
748
749 return 0;
750
751 /*
752 * Error cases jump to one of these labels to undo any
753 * reservations which have already been performed.
754 */
755undo_log:
756 if (logspace > 0) {
757 if (flags & XFS_TRANS_PERM_LOG_RES) {
758 log_flags = XFS_LOG_REL_PERM_RESERV;
759 } else {
760 log_flags = 0;
761 }
762 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
763 tp->t_ticket = NULL;
764 tp->t_log_res = 0;
765 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
766 }
767
768undo_blocks:
769 if (blocks > 0) {
770 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100771 (int64_t)blocks, rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 tp->t_blk_res = 0;
773 }
774
Nathan Scott59c1b082006-06-09 14:59:13 +1000775 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Nathan Scott59c1b082006-06-09 14:59:13 +1000777 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 * Record the indicated change to the given field for application
782 * to the file system's superblock when the transaction commits.
783 * For now, just store the change in the transaction structure.
784 *
785 * Mark the transaction structure to indicate that the superblock
786 * needs to be updated before committing.
David Chinner92821e22007-05-24 15:26:31 +1000787 *
788 * Because we may not be keeping track of allocated/free inodes and
789 * used filesystem blocks in the superblock, we do not mark the
790 * superblock dirty in this transaction if we modify these fields.
791 * We still need to update the transaction deltas so that they get
792 * applied to the incore superblock, but we don't want them to
793 * cause the superblock to get locked and logged if these are the
794 * only fields in the superblock that the transaction modifies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 */
796void
797xfs_trans_mod_sb(
798 xfs_trans_t *tp,
799 uint field,
David Chinner20f4ebf2007-02-10 18:36:10 +1100800 int64_t delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
David Chinner92821e22007-05-24 15:26:31 +1000802 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
803 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805 switch (field) {
806 case XFS_TRANS_SB_ICOUNT:
807 tp->t_icount_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000808 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
809 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 break;
811 case XFS_TRANS_SB_IFREE:
812 tp->t_ifree_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000813 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
814 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 break;
816 case XFS_TRANS_SB_FDBLOCKS:
817 /*
818 * Track the number of blocks allocated in the
819 * transaction. Make sure it does not exceed the
820 * number reserved.
821 */
822 if (delta < 0) {
823 tp->t_blk_res_used += (uint)-delta;
824 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
825 }
826 tp->t_fdblocks_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000827 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
828 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 break;
830 case XFS_TRANS_SB_RES_FDBLOCKS:
831 /*
832 * The allocation has already been applied to the
833 * in-core superblock's counter. This should only
834 * be applied to the on-disk superblock.
835 */
836 ASSERT(delta < 0);
837 tp->t_res_fdblocks_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000838 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
839 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 break;
841 case XFS_TRANS_SB_FREXTENTS:
842 /*
843 * Track the number of blocks allocated in the
844 * transaction. Make sure it does not exceed the
845 * number reserved.
846 */
847 if (delta < 0) {
848 tp->t_rtx_res_used += (uint)-delta;
849 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
850 }
851 tp->t_frextents_delta += delta;
852 break;
853 case XFS_TRANS_SB_RES_FREXTENTS:
854 /*
855 * The allocation has already been applied to the
Nathan Scottc41564b2006-03-29 08:55:14 +1000856 * in-core superblock's counter. This should only
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 * be applied to the on-disk superblock.
858 */
859 ASSERT(delta < 0);
860 tp->t_res_frextents_delta += delta;
861 break;
862 case XFS_TRANS_SB_DBLOCKS:
863 ASSERT(delta > 0);
864 tp->t_dblocks_delta += delta;
865 break;
866 case XFS_TRANS_SB_AGCOUNT:
867 ASSERT(delta > 0);
868 tp->t_agcount_delta += delta;
869 break;
870 case XFS_TRANS_SB_IMAXPCT:
871 tp->t_imaxpct_delta += delta;
872 break;
873 case XFS_TRANS_SB_REXTSIZE:
874 tp->t_rextsize_delta += delta;
875 break;
876 case XFS_TRANS_SB_RBMBLOCKS:
877 tp->t_rbmblocks_delta += delta;
878 break;
879 case XFS_TRANS_SB_RBLOCKS:
880 tp->t_rblocks_delta += delta;
881 break;
882 case XFS_TRANS_SB_REXTENTS:
883 tp->t_rextents_delta += delta;
884 break;
885 case XFS_TRANS_SB_REXTSLOG:
886 tp->t_rextslog_delta += delta;
887 break;
888 default:
889 ASSERT(0);
890 return;
891 }
892
David Chinner210c6f12007-05-24 15:26:51 +1000893 tp->t_flags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895
896/*
897 * xfs_trans_apply_sb_deltas() is called from the commit code
898 * to bring the superblock buffer into the current transaction
899 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
900 *
901 * For now we just look at each field allowed to change and change
902 * it if necessary.
903 */
904STATIC void
905xfs_trans_apply_sb_deltas(
906 xfs_trans_t *tp)
907{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000908 xfs_dsb_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 xfs_buf_t *bp;
910 int whole = 0;
911
912 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
913 sbp = XFS_BUF_TO_SBP(bp);
914
915 /*
916 * Check that superblock mods match the mods made to AGF counters.
917 */
918 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
919 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
920 tp->t_ag_btree_delta));
921
David Chinner92821e22007-05-24 15:26:31 +1000922 /*
923 * Only update the superblock counters if we are logging them
924 */
925 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000926 if (tp->t_icount_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800927 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000928 if (tp->t_ifree_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800929 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000930 if (tp->t_fdblocks_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800931 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000932 if (tp->t_res_fdblocks_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800933 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 }
935
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000936 if (tp->t_frextents_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800937 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000938 if (tp->t_res_frextents_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800939 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000940
941 if (tp->t_dblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800942 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 whole = 1;
944 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000945 if (tp->t_agcount_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800946 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 whole = 1;
948 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000949 if (tp->t_imaxpct_delta) {
950 sbp->sb_imax_pct += tp->t_imaxpct_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 whole = 1;
952 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000953 if (tp->t_rextsize_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800954 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 whole = 1;
956 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000957 if (tp->t_rbmblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800958 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 whole = 1;
960 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000961 if (tp->t_rblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800962 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 whole = 1;
964 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000965 if (tp->t_rextents_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800966 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 whole = 1;
968 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000969 if (tp->t_rextslog_delta) {
970 sbp->sb_rextslog += tp->t_rextslog_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 whole = 1;
972 }
973
974 if (whole)
975 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000976 * Log the whole thing, the fields are noncontiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000978 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 else
980 /*
981 * Since all the modifiable fields are contiguous, we
982 * can get away with this.
983 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000984 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
985 offsetof(xfs_dsb_t, sb_frextents) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 sizeof(sbp->sb_frextents) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
988
989/*
David Chinner45c34142007-06-18 16:49:44 +1000990 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
991 * and apply superblock counter changes to the in-core superblock. The
992 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
993 * applied to the in-core superblock. The idea is that that has already been
994 * done.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 *
996 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
David Chinner45c34142007-06-18 16:49:44 +1000997 * However, we have to ensure that we only modify each superblock field only
998 * once because the application of the delta values may not be atomic. That can
999 * lead to ENOSPC races occurring if we have two separate modifcations of the
1000 * free space counter to put back the entire reservation and then take away
1001 * what we used.
1002 *
1003 * If we are not logging superblock counters, then the inode allocated/free and
1004 * used block counts are not updated in the on disk superblock. In this case,
1005 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1006 * still need to update the incore superblock with the changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001008void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009xfs_trans_unreserve_and_mod_sb(
1010 xfs_trans_t *tp)
1011{
1012 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
1013 xfs_mod_sb_t *msbp;
David Chinner92821e22007-05-24 15:26:31 +10001014 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* REFERENCED */
1016 int error;
1017 int rsvd;
David Chinner45c34142007-06-18 16:49:44 +10001018 int64_t blkdelta = 0;
1019 int64_t rtxdelta = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 msbp = msb;
1022 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1023
David Chinner45c34142007-06-18 16:49:44 +10001024 /* calculate free blocks delta */
1025 if (tp->t_blk_res > 0)
1026 blkdelta = tp->t_blk_res;
1027
1028 if ((tp->t_fdblocks_delta != 0) &&
1029 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1030 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1031 blkdelta += tp->t_fdblocks_delta;
1032
1033 if (blkdelta != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 msbp->msb_field = XFS_SBS_FDBLOCKS;
David Chinner45c34142007-06-18 16:49:44 +10001035 msbp->msb_delta = blkdelta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 msbp++;
1037 }
1038
David Chinner45c34142007-06-18 16:49:44 +10001039 /* calculate free realtime extents delta */
1040 if (tp->t_rtx_res > 0)
1041 rtxdelta = tp->t_rtx_res;
1042
1043 if ((tp->t_frextents_delta != 0) &&
1044 (tp->t_flags & XFS_TRANS_SB_DIRTY))
1045 rtxdelta += tp->t_frextents_delta;
1046
1047 if (rtxdelta != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 msbp->msb_field = XFS_SBS_FREXTENTS;
David Chinner45c34142007-06-18 16:49:44 +10001049 msbp->msb_delta = rtxdelta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 msbp++;
1051 }
1052
David Chinner45c34142007-06-18 16:49:44 +10001053 /* apply remaining deltas */
1054
David Chinner92821e22007-05-24 15:26:31 +10001055 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1056 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (tp->t_icount_delta != 0) {
1058 msbp->msb_field = XFS_SBS_ICOUNT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001059 msbp->msb_delta = tp->t_icount_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 msbp++;
1061 }
1062 if (tp->t_ifree_delta != 0) {
1063 msbp->msb_field = XFS_SBS_IFREE;
David Chinner20f4ebf2007-02-10 18:36:10 +11001064 msbp->msb_delta = tp->t_ifree_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 msbp++;
1066 }
David Chinner92821e22007-05-24 15:26:31 +10001067 }
1068
1069 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 if (tp->t_dblocks_delta != 0) {
1071 msbp->msb_field = XFS_SBS_DBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001072 msbp->msb_delta = tp->t_dblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 msbp++;
1074 }
1075 if (tp->t_agcount_delta != 0) {
1076 msbp->msb_field = XFS_SBS_AGCOUNT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001077 msbp->msb_delta = tp->t_agcount_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 msbp++;
1079 }
1080 if (tp->t_imaxpct_delta != 0) {
1081 msbp->msb_field = XFS_SBS_IMAX_PCT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001082 msbp->msb_delta = tp->t_imaxpct_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 msbp++;
1084 }
1085 if (tp->t_rextsize_delta != 0) {
1086 msbp->msb_field = XFS_SBS_REXTSIZE;
David Chinner20f4ebf2007-02-10 18:36:10 +11001087 msbp->msb_delta = tp->t_rextsize_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 msbp++;
1089 }
1090 if (tp->t_rbmblocks_delta != 0) {
1091 msbp->msb_field = XFS_SBS_RBMBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001092 msbp->msb_delta = tp->t_rbmblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 msbp++;
1094 }
1095 if (tp->t_rblocks_delta != 0) {
1096 msbp->msb_field = XFS_SBS_RBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001097 msbp->msb_delta = tp->t_rblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 msbp++;
1099 }
1100 if (tp->t_rextents_delta != 0) {
1101 msbp->msb_field = XFS_SBS_REXTENTS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001102 msbp->msb_delta = tp->t_rextents_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 msbp++;
1104 }
1105 if (tp->t_rextslog_delta != 0) {
1106 msbp->msb_field = XFS_SBS_REXTSLOG;
David Chinner20f4ebf2007-02-10 18:36:10 +11001107 msbp->msb_delta = tp->t_rextslog_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 msbp++;
1109 }
1110 }
1111
1112 /*
1113 * If we need to change anything, do it.
1114 */
1115 if (msbp > msb) {
1116 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1117 (uint)(msbp - msb), rsvd);
1118 ASSERT(error == 0);
1119 }
1120}
1121
Dave Chinner09243782010-03-08 11:28:28 +11001122/*
Christoph Hellwige98c4142010-06-23 18:11:15 +10001123 * Add the given log item to the transaction's list of log items.
1124 *
1125 * The log item will now point to its new descriptor with its li_desc field.
1126 */
1127void
1128xfs_trans_add_item(
1129 struct xfs_trans *tp,
1130 struct xfs_log_item *lip)
1131{
1132 struct xfs_log_item_desc *lidp;
1133
1134 ASSERT(lip->li_mountp = tp->t_mountp);
1135 ASSERT(lip->li_ailp = tp->t_mountp->m_ail);
1136
Dave Chinner43869702010-07-20 17:53:44 +10001137 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001138
1139 lidp->lid_item = lip;
1140 lidp->lid_flags = 0;
1141 lidp->lid_size = 0;
1142 list_add_tail(&lidp->lid_trans, &tp->t_items);
1143
1144 lip->li_desc = lidp;
1145}
1146
1147STATIC void
1148xfs_trans_free_item_desc(
1149 struct xfs_log_item_desc *lidp)
1150{
1151 list_del_init(&lidp->lid_trans);
1152 kmem_zone_free(xfs_log_item_desc_zone, lidp);
1153}
1154
1155/*
1156 * Unlink and free the given descriptor.
1157 */
1158void
1159xfs_trans_del_item(
1160 struct xfs_log_item *lip)
1161{
1162 xfs_trans_free_item_desc(lip->li_desc);
1163 lip->li_desc = NULL;
1164}
1165
1166/*
1167 * Unlock all of the items of a transaction and free all the descriptors
1168 * of that transaction.
1169 */
1170STATIC void
1171xfs_trans_free_items(
1172 struct xfs_trans *tp,
1173 xfs_lsn_t commit_lsn,
1174 int flags)
1175{
1176 struct xfs_log_item_desc *lidp, *next;
1177
1178 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1179 struct xfs_log_item *lip = lidp->lid_item;
1180
1181 lip->li_desc = NULL;
1182
1183 if (commit_lsn != NULLCOMMITLSN)
1184 IOP_COMMITTING(lip, commit_lsn);
1185 if (flags & XFS_TRANS_ABORT)
1186 lip->li_flags |= XFS_LI_ABORTED;
1187 IOP_UNLOCK(lip);
1188
1189 xfs_trans_free_item_desc(lidp);
1190 }
1191}
1192
1193/*
1194 * Unlock the items associated with a transaction.
1195 *
1196 * Items which were not logged should be freed. Those which were logged must
1197 * still be tracked so they can be unpinned when the transaction commits.
1198 */
1199STATIC void
1200xfs_trans_unlock_items(
1201 struct xfs_trans *tp,
1202 xfs_lsn_t commit_lsn)
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 IOP_UNLOCK(lip);
1214
1215 /*
1216 * Free the descriptor if the item is not dirty
1217 * within this transaction.
1218 */
1219 if (!(lidp->lid_flags & XFS_LID_DIRTY))
1220 xfs_trans_free_item_desc(lidp);
1221 }
1222}
1223
1224/*
Dave Chinner09243782010-03-08 11:28:28 +11001225 * Total up the number of log iovecs needed to commit this
1226 * transaction. The transaction itself needs one for the
1227 * transaction header. Ask each dirty item in turn how many
1228 * it needs to get the total.
1229 */
1230static uint
1231xfs_trans_count_vecs(
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001232 struct xfs_trans *tp)
Dave Chinner09243782010-03-08 11:28:28 +11001233{
1234 int nvecs;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001235 struct xfs_log_item_desc *lidp;
Dave Chinner09243782010-03-08 11:28:28 +11001236
1237 nvecs = 1;
Dave Chinner09243782010-03-08 11:28:28 +11001238
1239 /* In the non-debug case we need to start bailing out if we
1240 * didn't find a log_item here, return zero and let trans_commit
1241 * deal with it.
1242 */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001243 if (list_empty(&tp->t_items)) {
1244 ASSERT(0);
Dave Chinner09243782010-03-08 11:28:28 +11001245 return 0;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001246 }
Dave Chinner09243782010-03-08 11:28:28 +11001247
Christoph Hellwige98c4142010-06-23 18:11:15 +10001248 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner09243782010-03-08 11:28:28 +11001249 /*
1250 * Skip items which aren't dirty in this transaction.
1251 */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001252 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner09243782010-03-08 11:28:28 +11001253 continue;
Dave Chinner09243782010-03-08 11:28:28 +11001254 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1255 nvecs += lidp->lid_size;
Dave Chinner09243782010-03-08 11:28:28 +11001256 }
1257
1258 return nvecs;
1259}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261/*
Dave Chinner09243782010-03-08 11:28:28 +11001262 * Fill in the vector with pointers to data to be logged
1263 * by this transaction. The transaction header takes
1264 * the first vector, and then each dirty item takes the
1265 * number of vectors it indicated it needed in xfs_trans_count_vecs().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 *
Dave Chinner09243782010-03-08 11:28:28 +11001267 * As each item fills in the entries it needs, also pin the item
1268 * so that it cannot be flushed out until the log write completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 */
Dave Chinner09243782010-03-08 11:28:28 +11001270static void
1271xfs_trans_fill_vecs(
1272 struct xfs_trans *tp,
1273 struct xfs_log_iovec *log_vector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001275 struct xfs_log_item_desc *lidp;
Dave Chinner09243782010-03-08 11:28:28 +11001276 struct xfs_log_iovec *vecp;
1277 uint nitems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 /*
Dave Chinner09243782010-03-08 11:28:28 +11001280 * Skip over the entry for the transaction header, we'll
1281 * fill that in at the end.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 */
Dave Chinner09243782010-03-08 11:28:28 +11001283 vecp = log_vector + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Dave Chinner09243782010-03-08 11:28:28 +11001285 nitems = 0;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001286 ASSERT(!list_empty(&tp->t_items));
1287 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner09243782010-03-08 11:28:28 +11001288 /* Skip items which aren't dirty in this transaction. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001289 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner09243782010-03-08 11:28:28 +11001290 continue;
Dave Chinner09243782010-03-08 11:28:28 +11001291
1292 /*
1293 * The item may be marked dirty but not log anything. This can
1294 * be used to get called when a transaction is committed.
1295 */
1296 if (lidp->lid_size)
1297 nitems++;
1298 IOP_FORMAT(lidp->lid_item, vecp);
1299 vecp += lidp->lid_size;
1300 IOP_PIN(lidp->lid_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 /*
Dave Chinner09243782010-03-08 11:28:28 +11001304 * Now that we've counted the number of items in this transaction, fill
1305 * in the transaction header. Note that the transaction header does not
1306 * have a log item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 */
Dave Chinner09243782010-03-08 11:28:28 +11001308 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1309 tp->t_header.th_type = tp->t_type;
1310 tp->t_header.th_num_items = nitems;
1311 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1312 log_vector->i_len = sizeof(xfs_trans_header_t);
1313 log_vector->i_type = XLOG_REG_TYPE_TRANSHDR;
1314}
1315
1316/*
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001317 * The committed item processing consists of calling the committed routine of
1318 * each logged item, updating the item's position in the AIL if necessary, and
1319 * unpinning each item. If the committed routine returns -1, then do nothing
1320 * further with the item because it may have been freed.
1321 *
1322 * Since items are unlocked when they are copied to the incore log, it is
1323 * possible for two transactions to be completing and manipulating the same
1324 * item simultaneously. The AIL lock will protect the lsn field of each item.
1325 * The value of this field can never go backwards.
1326 *
1327 * We unpin the items after repositioning them in the AIL, because otherwise
1328 * they could be immediately flushed and we'd have to race with the flusher
1329 * trying to pull the item from the AIL as we add it.
1330 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001331void
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001332xfs_trans_item_committed(
1333 struct xfs_log_item *lip,
1334 xfs_lsn_t commit_lsn,
1335 int aborted)
1336{
1337 xfs_lsn_t item_lsn;
1338 struct xfs_ail *ailp;
1339
1340 if (aborted)
1341 lip->li_flags |= XFS_LI_ABORTED;
1342 item_lsn = IOP_COMMITTED(lip, commit_lsn);
1343
1344 /* If the committed routine returns -1, item has been freed. */
1345 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1346 return;
1347
1348 /*
1349 * If the returned lsn is greater than what it contained before, update
1350 * the location of the item in the AIL. If it is not, then do nothing.
1351 * Items can never move backwards in the AIL.
1352 *
1353 * While the new lsn should usually be greater, it is possible that a
1354 * later transaction completing simultaneously with an earlier one
1355 * using the same item could complete first with a higher lsn. This
1356 * would cause the earlier transaction to fail the test below.
1357 */
1358 ailp = lip->li_ailp;
1359 spin_lock(&ailp->xa_lock);
1360 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1361 /*
1362 * This will set the item's lsn to item_lsn and update the
1363 * position of the item in the AIL.
1364 *
1365 * xfs_trans_ail_update() drops the AIL lock.
1366 */
1367 xfs_trans_ail_update(ailp, lip, item_lsn);
1368 } else {
1369 spin_unlock(&ailp->xa_lock);
1370 }
1371
1372 /*
1373 * Now that we've repositioned the item in the AIL, unpin it so it can
1374 * be flushed. Pass information about buffer stale state down from the
1375 * log item flags, if anyone else stales the buffer we do not want to
1376 * pay any attention to it.
1377 */
Christoph Hellwig9412e312010-06-23 18:11:15 +10001378 IOP_UNPIN(lip, 0);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001379}
1380
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001381/*
1382 * This is typically called by the LM when a transaction has been fully
1383 * committed to disk. It needs to unpin the items which have
1384 * been logged by the transaction and update their positions
1385 * in the AIL if necessary.
1386 *
1387 * This also gets called when the transactions didn't get written out
1388 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1389 */
1390STATIC void
1391xfs_trans_committed(
1392 struct xfs_trans *tp,
1393 int abortflag)
1394{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001395 struct xfs_log_item_desc *lidp, *next;
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001396
1397 /* Call the transaction's completion callback if there is one. */
1398 if (tp->t_callback != NULL)
1399 tp->t_callback(tp, tp->t_callarg);
1400
Christoph Hellwige98c4142010-06-23 18:11:15 +10001401 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001402 xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001403 xfs_trans_free_item_desc(lidp);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001404 }
1405
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001406 xfs_trans_free(tp);
1407}
1408
1409/*
1410 * Called from the trans_commit code when we notice that
1411 * the filesystem is in the middle of a forced shutdown.
1412 */
1413STATIC void
1414xfs_trans_uncommit(
1415 struct xfs_trans *tp,
1416 uint flags)
1417{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001418 struct xfs_log_item_desc *lidp;
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001419
Christoph Hellwige98c4142010-06-23 18:11:15 +10001420 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001421 /*
1422 * Unpin all but those that aren't dirty.
1423 */
1424 if (lidp->lid_flags & XFS_LID_DIRTY)
Christoph Hellwig9412e312010-06-23 18:11:15 +10001425 IOP_UNPIN(lidp->lid_item, 1);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001426 }
1427
1428 xfs_trans_unreserve_and_mod_sb(tp);
1429 xfs_trans_unreserve_and_mod_dquots(tp);
1430
Dave Chinner71e330b2010-05-21 14:37:18 +10001431 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001432 xfs_trans_free(tp);
1433}
1434
1435/*
Dave Chinner09243782010-03-08 11:28:28 +11001436 * Format the transaction direct to the iclog. This isolates the physical
1437 * transaction commit operation from the logical operation and hence allows
1438 * other methods to be introduced without affecting the existing commit path.
1439 */
1440static int
1441xfs_trans_commit_iclog(
1442 struct xfs_mount *mp,
1443 struct xfs_trans *tp,
1444 xfs_lsn_t *commit_lsn,
1445 int flags)
1446{
1447 int shutdown;
1448 int error;
1449 int log_flags = 0;
1450 struct xlog_in_core *commit_iclog;
1451#define XFS_TRANS_LOGVEC_COUNT 16
1452 struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
1453 struct xfs_log_iovec *log_vector;
1454 uint nvec;
1455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 /*
1458 * Ask each log item how many log_vector entries it will
1459 * need so we can figure out how many to allocate.
1460 * Try to avoid the kmem_alloc() call in the common case
1461 * by using a vector from the stack when it fits.
1462 */
1463 nvec = xfs_trans_count_vecs(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (nvec == 0) {
Dave Chinner09243782010-03-08 11:28:28 +11001465 return ENOMEM; /* triggers a shutdown! */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001466 } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 log_vector = log_vector_fast;
1468 } else {
1469 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
1470 sizeof(xfs_log_iovec_t),
1471 KM_SLEEP);
1472 }
1473
1474 /*
1475 * Fill in the log_vector and pin the logged items, and
1476 * then write the transaction to the log.
1477 */
1478 xfs_trans_fill_vecs(tp, log_vector);
1479
Dave Chinner09243782010-03-08 11:28:28 +11001480 if (flags & XFS_TRANS_RELEASE_LOG_RES)
1481 log_flags = XFS_LOG_REL_PERM_RESERV;
1482
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001483 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 /*
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001486 * The transaction is committed incore here, and can go out to disk
1487 * at any time after this call. However, all the items associated
1488 * with the transaction are still locked and pinned in memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 */
Dave Chinner09243782010-03-08 11:28:28 +11001490 *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
Dave Chinner09243782010-03-08 11:28:28 +11001492 tp->t_commit_lsn = *commit_lsn;
Dave Chinnered3b4d62010-05-21 12:07:08 +10001493 trace_xfs_trans_commit_lsn(tp);
1494
Dave Chinner09243782010-03-08 11:28:28 +11001495 if (nvec > XFS_TRANS_LOGVEC_COUNT)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001496 kmem_free(log_vector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 /*
1499 * If we got a log write error. Unpin the logitems that we
1500 * had pinned, clean up, free trans structure, and return error.
1501 */
Dave Chinner09243782010-03-08 11:28:28 +11001502 if (error || *commit_lsn == -1) {
Nathan Scott59c1b082006-06-09 14:59:13 +10001503 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
1505 return XFS_ERROR(EIO);
1506 }
1507
1508 /*
1509 * Once the transaction has committed, unused
1510 * reservations need to be released and changes to
1511 * the superblock need to be reflected in the in-core
1512 * version. Do that now.
1513 */
1514 xfs_trans_unreserve_and_mod_sb(tp);
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /*
1517 * Tell the LM to call the transaction completion routine
1518 * when the log write with LSN commit_lsn completes (e.g.
1519 * when the transaction commit really hits the on-disk log).
1520 * After this call we cannot reference tp, because the call
1521 * can happen at any time and the call will free the transaction
1522 * structure pointed to by tp. The only case where we call
1523 * the completion routine (xfs_trans_committed) directly is
1524 * if the log is turned off on a debug kernel or we're
1525 * running in simulation mode (the log is explicitly turned
1526 * off).
1527 */
1528 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
1529 tp->t_logcb.cb_arg = tp;
1530
1531 /*
1532 * We need to pass the iclog buffer which was used for the
1533 * transaction commit record into this function, and attach
1534 * the callback to it. The callback must be attached before
1535 * the items are unlocked to avoid racing with other threads
1536 * waiting for an item to unlock.
1537 */
1538 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
1539
1540 /*
1541 * Mark this thread as no longer being in a transaction
1542 */
Nathan Scott59c1b082006-06-09 14:59:13 +10001543 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 /*
1546 * Once all the items of the transaction have been copied
1547 * to the in core log and the callback is attached, the
1548 * items can be unlocked.
1549 *
1550 * This will free descriptors pointing to items which were
1551 * not logged since there is nothing more to do with them.
1552 * For items which were logged, we will keep pointers to them
1553 * so they can be unpinned after the transaction commits to disk.
1554 * This will also stamp each modified meta-data item with
1555 * the commit lsn of this transaction for dependency tracking
1556 * purposes.
1557 */
Dave Chinner09243782010-03-08 11:28:28 +11001558 xfs_trans_unlock_items(tp, *commit_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560 /*
1561 * If we detected a log error earlier, finish committing
1562 * the transaction now (unpin log items, etc).
1563 *
1564 * Order is critical here, to avoid using the transaction
1565 * pointer after its been freed (by xfs_trans_committed
1566 * either here now, or as a callback). We cannot do this
1567 * step inside xfs_log_notify as was done earlier because
1568 * of this issue.
1569 */
1570 if (shutdown)
1571 xfs_trans_committed(tp, XFS_LI_ABORTED);
1572
1573 /*
1574 * Now that the xfs_trans_committed callback has been attached,
1575 * and the items are released we can finally allow the iclog to
1576 * go to disk.
1577 */
Dave Chinner09243782010-03-08 11:28:28 +11001578 return xfs_log_release_iclog(mp, commit_iclog);
1579}
1580
Dave Chinner71e330b2010-05-21 14:37:18 +10001581/*
1582 * Walk the log items and allocate log vector structures for
1583 * each item large enough to fit all the vectors they require.
1584 * Note that this format differs from the old log vector format in
1585 * that there is no transaction header in these log vectors.
1586 */
1587STATIC struct xfs_log_vec *
1588xfs_trans_alloc_log_vecs(
1589 xfs_trans_t *tp)
1590{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001591 struct xfs_log_item_desc *lidp;
Dave Chinner71e330b2010-05-21 14:37:18 +10001592 struct xfs_log_vec *lv = NULL;
1593 struct xfs_log_vec *ret_lv = NULL;
1594
Dave Chinner71e330b2010-05-21 14:37:18 +10001595
1596 /* Bail out if we didn't find a log item. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001597 if (list_empty(&tp->t_items)) {
Dave Chinner71e330b2010-05-21 14:37:18 +10001598 ASSERT(0);
1599 return NULL;
1600 }
1601
Christoph Hellwige98c4142010-06-23 18:11:15 +10001602 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner71e330b2010-05-21 14:37:18 +10001603 struct xfs_log_vec *new_lv;
1604
1605 /* Skip items which aren't dirty in this transaction. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001606 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner71e330b2010-05-21 14:37:18 +10001607 continue;
Dave Chinner71e330b2010-05-21 14:37:18 +10001608
1609 /* Skip items that do not have any vectors for writing */
1610 lidp->lid_size = IOP_SIZE(lidp->lid_item);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001611 if (!lidp->lid_size)
Dave Chinner71e330b2010-05-21 14:37:18 +10001612 continue;
Dave Chinner71e330b2010-05-21 14:37:18 +10001613
1614 new_lv = kmem_zalloc(sizeof(*new_lv) +
1615 lidp->lid_size * sizeof(struct xfs_log_iovec),
1616 KM_SLEEP);
1617
1618 /* The allocated iovec region lies beyond the log vector. */
1619 new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1];
1620 new_lv->lv_niovecs = lidp->lid_size;
1621 new_lv->lv_item = lidp->lid_item;
1622 if (!ret_lv)
1623 ret_lv = new_lv;
1624 else
1625 lv->lv_next = new_lv;
1626 lv = new_lv;
Dave Chinner71e330b2010-05-21 14:37:18 +10001627 }
1628
1629 return ret_lv;
1630}
1631
1632static int
1633xfs_trans_commit_cil(
1634 struct xfs_mount *mp,
1635 struct xfs_trans *tp,
1636 xfs_lsn_t *commit_lsn,
1637 int flags)
1638{
1639 struct xfs_log_vec *log_vector;
1640 int error;
1641
1642 /*
1643 * Get each log item to allocate a vector structure for
1644 * the log item to to pass to the log write code. The
1645 * CIL commit code will format the vector and save it away.
1646 */
1647 log_vector = xfs_trans_alloc_log_vecs(tp);
1648 if (!log_vector)
1649 return ENOMEM;
1650
1651 error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags);
1652 if (error)
1653 return error;
1654
1655 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1656
1657 /* xfs_trans_free_items() unlocks them first */
1658 xfs_trans_free_items(tp, *commit_lsn, 0);
1659 xfs_trans_free(tp);
1660 return 0;
1661}
Dave Chinner09243782010-03-08 11:28:28 +11001662
1663/*
1664 * xfs_trans_commit
1665 *
1666 * Commit the given transaction to the log a/synchronously.
1667 *
1668 * XFS disk error handling mechanism is not based on a typical
1669 * transaction abort mechanism. Logically after the filesystem
1670 * gets marked 'SHUTDOWN', we can't let any new transactions
1671 * be durable - ie. committed to disk - because some metadata might
1672 * be inconsistent. In such cases, this returns an error, and the
1673 * caller may assume that all locked objects joined to the transaction
1674 * have already been unlocked as if the commit had succeeded.
1675 * Do not reference the transaction structure after this call.
1676 */
Dave Chinner09243782010-03-08 11:28:28 +11001677int
1678_xfs_trans_commit(
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001679 struct xfs_trans *tp,
1680 uint flags,
1681 int *log_flushed)
Dave Chinner09243782010-03-08 11:28:28 +11001682{
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001683 struct xfs_mount *mp = tp->t_mountp;
Dave Chinner09243782010-03-08 11:28:28 +11001684 xfs_lsn_t commit_lsn = -1;
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001685 int error = 0;
Dave Chinner09243782010-03-08 11:28:28 +11001686 int log_flags = 0;
1687 int sync = tp->t_flags & XFS_TRANS_SYNC;
Dave Chinner09243782010-03-08 11:28:28 +11001688
1689 /*
1690 * Determine whether this commit is releasing a permanent
1691 * log reservation or not.
1692 */
1693 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1694 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1695 log_flags = XFS_LOG_REL_PERM_RESERV;
1696 }
1697
1698 /*
1699 * If there is nothing to be logged by the transaction,
1700 * then unlock all of the items associated with the
1701 * transaction and free the transaction structure.
1702 * Also make sure to return any reserved blocks to
1703 * the free pool.
1704 */
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001705 if (!(tp->t_flags & XFS_TRANS_DIRTY))
1706 goto out_unreserve;
1707
1708 if (XFS_FORCED_SHUTDOWN(mp)) {
1709 error = XFS_ERROR(EIO);
1710 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001711 }
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001712
Dave Chinner09243782010-03-08 11:28:28 +11001713 ASSERT(tp->t_ticket != NULL);
1714
1715 /*
1716 * If we need to update the superblock, then do it now.
1717 */
1718 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1719 xfs_trans_apply_sb_deltas(tp);
1720 xfs_trans_apply_dquot_deltas(tp);
1721
Dave Chinner71e330b2010-05-21 14:37:18 +10001722 if (mp->m_flags & XFS_MOUNT_DELAYLOG)
1723 error = xfs_trans_commit_cil(mp, tp, &commit_lsn, flags);
1724 else
1725 error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags);
1726
Dave Chinner09243782010-03-08 11:28:28 +11001727 if (error == ENOMEM) {
1728 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001729 error = XFS_ERROR(EIO);
1730 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 /*
1734 * If the transaction needs to be synchronous, then force the
1735 * log out now and wait for it.
1736 */
1737 if (sync) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001738 if (!error) {
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001739 error = _xfs_log_force_lsn(mp, commit_lsn,
1740 XFS_LOG_SYNC, log_flushed);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 XFS_STATS_INC(xs_trans_sync);
1743 } else {
1744 XFS_STATS_INC(xs_trans_async);
1745 }
1746
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001747 return error;
1748
1749out_unreserve:
1750 xfs_trans_unreserve_and_mod_sb(tp);
1751
1752 /*
1753 * It is indeed possible for the transaction to be not dirty but
1754 * the dqinfo portion to be. All that means is that we have some
1755 * (non-persistent) quota reservations that need to be unreserved.
1756 */
1757 xfs_trans_unreserve_and_mod_dquots(tp);
1758 if (tp->t_ticket) {
1759 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1760 if (commit_lsn == -1 && !error)
1761 error = XFS_ERROR(EIO);
1762 }
1763 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner71e330b2010-05-21 14:37:18 +10001764 xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001765 xfs_trans_free(tp);
1766
1767 XFS_STATS_INC(xs_trans_empty);
1768 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769}
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 * Unlock all of the transaction's items and free the transaction.
1773 * The transaction must not have modified any of its items, because
1774 * there is no way to restore them to their previous state.
1775 *
1776 * If the transaction has made a log reservation, make sure to release
1777 * it as well.
1778 */
1779void
1780xfs_trans_cancel(
1781 xfs_trans_t *tp,
1782 int flags)
1783{
1784 int log_flags;
Ryan Hankins0733af22006-01-11 15:36:44 +11001785 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787 /*
1788 * See if the caller is being too lazy to figure out if
1789 * the transaction really needs an abort.
1790 */
1791 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1792 flags &= ~XFS_TRANS_ABORT;
1793 /*
1794 * See if the caller is relying on us to shut down the
1795 * filesystem. This happens in paths where we detect
1796 * corruption and decide to give up.
1797 */
Nathan Scott60a204f2006-01-11 15:37:00 +11001798 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
Ryan Hankins0733af22006-01-11 15:36:44 +11001799 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001800 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Nathan Scott60a204f2006-01-11 15:37:00 +11001801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802#ifdef DEBUG
Christoph Hellwige98c4142010-06-23 18:11:15 +10001803 if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1804 struct xfs_log_item_desc *lidp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
Christoph Hellwige98c4142010-06-23 18:11:15 +10001806 list_for_each_entry(lidp, &tp->t_items, lid_trans)
1807 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
1809#endif
1810 xfs_trans_unreserve_and_mod_sb(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001811 xfs_trans_unreserve_and_mod_dquots(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
1813 if (tp->t_ticket) {
1814 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1815 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1816 log_flags = XFS_LOG_REL_PERM_RESERV;
1817 } else {
1818 log_flags = 0;
1819 }
Ryan Hankins0733af22006-01-11 15:36:44 +11001820 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822
1823 /* mark this thread as no longer being in a transaction */
Nathan Scott59c1b082006-06-09 14:59:13 +10001824 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Dave Chinner71e330b2010-05-21 14:37:18 +10001826 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 xfs_trans_free(tp);
1828}
1829
Niv Sardi322ff6b2008-08-13 16:05:49 +10001830/*
1831 * Roll from one trans in the sequence of PERMANENT transactions to
1832 * the next: permanent transactions are only flushed out when
1833 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1834 * as possible to let chunks of it go to the log. So we commit the
1835 * chunk we've been working on and get a new transaction to continue.
1836 */
1837int
1838xfs_trans_roll(
1839 struct xfs_trans **tpp,
1840 struct xfs_inode *dp)
1841{
1842 struct xfs_trans *trans;
1843 unsigned int logres, count;
1844 int error;
1845
1846 /*
1847 * Ensure that the inode is always logged.
1848 */
1849 trans = *tpp;
1850 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1851
1852 /*
1853 * Copy the critical parameters from one trans to the next.
1854 */
1855 logres = trans->t_log_res;
1856 count = trans->t_log_count;
1857 *tpp = xfs_trans_dup(trans);
1858
1859 /*
1860 * Commit the current transaction.
1861 * If this commit failed, then it'd just unlock those items that
1862 * are not marked ihold. That also means that a filesystem shutdown
1863 * is in progress. The caller takes the responsibility to cancel
1864 * the duplicate transaction that gets returned.
1865 */
1866 error = xfs_trans_commit(trans, 0);
1867 if (error)
1868 return (error);
1869
1870 trans = *tpp;
1871
1872 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001873 * transaction commit worked ok so we can drop the extra ticket
1874 * reference that we gained in xfs_trans_dup()
1875 */
1876 xfs_log_ticket_put(trans->t_ticket);
1877
1878
1879 /*
Niv Sardi322ff6b2008-08-13 16:05:49 +10001880 * Reserve space in the log for th next transaction.
1881 * This also pushes items in the "AIL", the list of logged items,
1882 * out to disk if they are taking up space at the tail of the log
1883 * that we want to use. This requires that either nothing be locked
1884 * across this call, or that anything that is locked be logged in
1885 * the prior and the next transactions.
1886 */
1887 error = xfs_trans_reserve(trans, 0, logres, 0,
1888 XFS_TRANS_PERM_LOG_RES, count);
1889 /*
1890 * Ensure that the inode is in the new transaction and locked.
1891 */
1892 if (error)
1893 return error;
1894
Christoph Hellwig898621d2010-06-24 11:36:58 +10001895 xfs_trans_ijoin(trans, dp);
Niv Sardi322ff6b2008-08-13 16:05:49 +10001896 return 0;
1897}