blob: 33dbc4e0ad622a94dbbf8aa67b49e0eacef0cbbc [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) {
Christoph Hellwig96540c72010-09-30 02:25:55 +0000699 error = xfs_icsb_modify_counters(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) {
Christoph Hellwig96540c72010-09-30 02:25:55 +0000770 xfs_icsb_modify_counters(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{
Christoph Hellwig1b040712010-09-30 02:25:56 +00001012 xfs_mod_sb_t msb[9]; /* If you add cases, add entries */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 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;
Christoph Hellwig1b040712010-09-30 02:25:56 +00001020 int64_t idelta = 0;
1021 int64_t ifreedelta = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 msbp = msb;
1024 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1025
Christoph Hellwig1b040712010-09-30 02:25:56 +00001026 /* calculate deltas */
David Chinner45c34142007-06-18 16:49:44 +10001027 if (tp->t_blk_res > 0)
1028 blkdelta = tp->t_blk_res;
David Chinner45c34142007-06-18 16:49:44 +10001029 if ((tp->t_fdblocks_delta != 0) &&
1030 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1031 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1032 blkdelta += tp->t_fdblocks_delta;
1033
David Chinner45c34142007-06-18 16:49:44 +10001034 if (tp->t_rtx_res > 0)
1035 rtxdelta = tp->t_rtx_res;
David Chinner45c34142007-06-18 16:49:44 +10001036 if ((tp->t_frextents_delta != 0) &&
1037 (tp->t_flags & XFS_TRANS_SB_DIRTY))
1038 rtxdelta += tp->t_frextents_delta;
1039
Christoph Hellwig1b040712010-09-30 02:25:56 +00001040 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1041 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1042 idelta = tp->t_icount_delta;
1043 ifreedelta = tp->t_ifree_delta;
1044 }
1045
1046 /* apply the per-cpu counters */
1047 if (blkdelta) {
1048 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1049 blkdelta, rsvd);
1050 if (error)
1051 goto out;
1052 }
1053
1054 if (idelta) {
1055 error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1056 idelta, rsvd);
1057 if (error)
1058 goto out_undo_fdblocks;
1059 }
1060
1061 if (ifreedelta) {
1062 error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1063 ifreedelta, rsvd);
1064 if (error)
1065 goto out_undo_icount;
1066 }
1067
1068 /* apply remaining deltas */
David Chinner45c34142007-06-18 16:49:44 +10001069 if (rtxdelta != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 msbp->msb_field = XFS_SBS_FREXTENTS;
David Chinner45c34142007-06-18 16:49:44 +10001071 msbp->msb_delta = rtxdelta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 msbp++;
1073 }
1074
David Chinner92821e22007-05-24 15:26:31 +10001075 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (tp->t_dblocks_delta != 0) {
1077 msbp->msb_field = XFS_SBS_DBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001078 msbp->msb_delta = tp->t_dblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 msbp++;
1080 }
1081 if (tp->t_agcount_delta != 0) {
1082 msbp->msb_field = XFS_SBS_AGCOUNT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001083 msbp->msb_delta = tp->t_agcount_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 msbp++;
1085 }
1086 if (tp->t_imaxpct_delta != 0) {
1087 msbp->msb_field = XFS_SBS_IMAX_PCT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001088 msbp->msb_delta = tp->t_imaxpct_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 msbp++;
1090 }
1091 if (tp->t_rextsize_delta != 0) {
1092 msbp->msb_field = XFS_SBS_REXTSIZE;
David Chinner20f4ebf2007-02-10 18:36:10 +11001093 msbp->msb_delta = tp->t_rextsize_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 msbp++;
1095 }
1096 if (tp->t_rbmblocks_delta != 0) {
1097 msbp->msb_field = XFS_SBS_RBMBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001098 msbp->msb_delta = tp->t_rbmblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 msbp++;
1100 }
1101 if (tp->t_rblocks_delta != 0) {
1102 msbp->msb_field = XFS_SBS_RBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001103 msbp->msb_delta = tp->t_rblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 msbp++;
1105 }
1106 if (tp->t_rextents_delta != 0) {
1107 msbp->msb_field = XFS_SBS_REXTENTS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001108 msbp->msb_delta = tp->t_rextents_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 msbp++;
1110 }
1111 if (tp->t_rextslog_delta != 0) {
1112 msbp->msb_field = XFS_SBS_REXTSLOG;
David Chinner20f4ebf2007-02-10 18:36:10 +11001113 msbp->msb_delta = tp->t_rextslog_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 msbp++;
1115 }
1116 }
1117
1118 /*
1119 * If we need to change anything, do it.
1120 */
1121 if (msbp > msb) {
1122 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1123 (uint)(msbp - msb), rsvd);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001124 if (error)
1125 goto out_undo_ifreecount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
Christoph Hellwig1b040712010-09-30 02:25:56 +00001127
1128 return;
1129
1130out_undo_ifreecount:
1131 if (ifreedelta)
1132 xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1133out_undo_icount:
1134 if (idelta)
1135 xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1136out_undo_fdblocks:
1137 if (blkdelta)
1138 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1139out:
Jesper Juhl1884bd82010-12-25 20:14:53 +00001140 ASSERT(error == 0);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001141 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}
1143
Dave Chinner09243782010-03-08 11:28:28 +11001144/*
Christoph Hellwige98c4142010-06-23 18:11:15 +10001145 * Add the given log item to the transaction's list of log items.
1146 *
1147 * The log item will now point to its new descriptor with its li_desc field.
1148 */
1149void
1150xfs_trans_add_item(
1151 struct xfs_trans *tp,
1152 struct xfs_log_item *lip)
1153{
1154 struct xfs_log_item_desc *lidp;
1155
1156 ASSERT(lip->li_mountp = tp->t_mountp);
1157 ASSERT(lip->li_ailp = tp->t_mountp->m_ail);
1158
Dave Chinner43869702010-07-20 17:53:44 +10001159 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001160
1161 lidp->lid_item = lip;
1162 lidp->lid_flags = 0;
1163 lidp->lid_size = 0;
1164 list_add_tail(&lidp->lid_trans, &tp->t_items);
1165
1166 lip->li_desc = lidp;
1167}
1168
1169STATIC void
1170xfs_trans_free_item_desc(
1171 struct xfs_log_item_desc *lidp)
1172{
1173 list_del_init(&lidp->lid_trans);
1174 kmem_zone_free(xfs_log_item_desc_zone, lidp);
1175}
1176
1177/*
1178 * Unlink and free the given descriptor.
1179 */
1180void
1181xfs_trans_del_item(
1182 struct xfs_log_item *lip)
1183{
1184 xfs_trans_free_item_desc(lip->li_desc);
1185 lip->li_desc = NULL;
1186}
1187
1188/*
1189 * Unlock all of the items of a transaction and free all the descriptors
1190 * of that transaction.
1191 */
Dave Chinnerd17c7012010-08-24 11:42:52 +10001192void
Christoph Hellwige98c4142010-06-23 18:11:15 +10001193xfs_trans_free_items(
1194 struct xfs_trans *tp,
1195 xfs_lsn_t commit_lsn,
1196 int flags)
1197{
1198 struct xfs_log_item_desc *lidp, *next;
1199
1200 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1201 struct xfs_log_item *lip = lidp->lid_item;
1202
1203 lip->li_desc = NULL;
1204
1205 if (commit_lsn != NULLCOMMITLSN)
1206 IOP_COMMITTING(lip, commit_lsn);
1207 if (flags & XFS_TRANS_ABORT)
1208 lip->li_flags |= XFS_LI_ABORTED;
1209 IOP_UNLOCK(lip);
1210
1211 xfs_trans_free_item_desc(lidp);
1212 }
1213}
1214
1215/*
1216 * Unlock the items associated with a transaction.
1217 *
1218 * Items which were not logged should be freed. Those which were logged must
1219 * still be tracked so they can be unpinned when the transaction commits.
1220 */
1221STATIC void
1222xfs_trans_unlock_items(
1223 struct xfs_trans *tp,
1224 xfs_lsn_t commit_lsn)
1225{
1226 struct xfs_log_item_desc *lidp, *next;
1227
1228 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1229 struct xfs_log_item *lip = lidp->lid_item;
1230
1231 lip->li_desc = NULL;
1232
1233 if (commit_lsn != NULLCOMMITLSN)
1234 IOP_COMMITTING(lip, commit_lsn);
1235 IOP_UNLOCK(lip);
1236
1237 /*
1238 * Free the descriptor if the item is not dirty
1239 * within this transaction.
1240 */
1241 if (!(lidp->lid_flags & XFS_LID_DIRTY))
1242 xfs_trans_free_item_desc(lidp);
1243 }
1244}
1245
1246/*
Dave Chinner09243782010-03-08 11:28:28 +11001247 * Total up the number of log iovecs needed to commit this
1248 * transaction. The transaction itself needs one for the
1249 * transaction header. Ask each dirty item in turn how many
1250 * it needs to get the total.
1251 */
1252static uint
1253xfs_trans_count_vecs(
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001254 struct xfs_trans *tp)
Dave Chinner09243782010-03-08 11:28:28 +11001255{
1256 int nvecs;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001257 struct xfs_log_item_desc *lidp;
Dave Chinner09243782010-03-08 11:28:28 +11001258
1259 nvecs = 1;
Dave Chinner09243782010-03-08 11:28:28 +11001260
1261 /* In the non-debug case we need to start bailing out if we
1262 * didn't find a log_item here, return zero and let trans_commit
1263 * deal with it.
1264 */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001265 if (list_empty(&tp->t_items)) {
1266 ASSERT(0);
Dave Chinner09243782010-03-08 11:28:28 +11001267 return 0;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001268 }
Dave Chinner09243782010-03-08 11:28:28 +11001269
Christoph Hellwige98c4142010-06-23 18:11:15 +10001270 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner09243782010-03-08 11:28:28 +11001271 /*
1272 * Skip items which aren't dirty in this transaction.
1273 */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001274 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner09243782010-03-08 11:28:28 +11001275 continue;
Dave Chinner09243782010-03-08 11:28:28 +11001276 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1277 nvecs += lidp->lid_size;
Dave Chinner09243782010-03-08 11:28:28 +11001278 }
1279
1280 return nvecs;
1281}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283/*
Dave Chinner09243782010-03-08 11:28:28 +11001284 * Fill in the vector with pointers to data to be logged
1285 * by this transaction. The transaction header takes
1286 * the first vector, and then each dirty item takes the
1287 * number of vectors it indicated it needed in xfs_trans_count_vecs().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 *
Dave Chinner09243782010-03-08 11:28:28 +11001289 * As each item fills in the entries it needs, also pin the item
1290 * so that it cannot be flushed out until the log write completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 */
Dave Chinner09243782010-03-08 11:28:28 +11001292static void
1293xfs_trans_fill_vecs(
1294 struct xfs_trans *tp,
1295 struct xfs_log_iovec *log_vector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001297 struct xfs_log_item_desc *lidp;
Dave Chinner09243782010-03-08 11:28:28 +11001298 struct xfs_log_iovec *vecp;
1299 uint nitems;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 /*
Dave Chinner09243782010-03-08 11:28:28 +11001302 * Skip over the entry for the transaction header, we'll
1303 * fill that in at the end.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 */
Dave Chinner09243782010-03-08 11:28:28 +11001305 vecp = log_vector + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Dave Chinner09243782010-03-08 11:28:28 +11001307 nitems = 0;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001308 ASSERT(!list_empty(&tp->t_items));
1309 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner09243782010-03-08 11:28:28 +11001310 /* Skip items which aren't dirty in this transaction. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001311 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner09243782010-03-08 11:28:28 +11001312 continue;
Dave Chinner09243782010-03-08 11:28:28 +11001313
1314 /*
1315 * The item may be marked dirty but not log anything. This can
1316 * be used to get called when a transaction is committed.
1317 */
1318 if (lidp->lid_size)
1319 nitems++;
1320 IOP_FORMAT(lidp->lid_item, vecp);
1321 vecp += lidp->lid_size;
1322 IOP_PIN(lidp->lid_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 /*
Dave Chinner09243782010-03-08 11:28:28 +11001326 * Now that we've counted the number of items in this transaction, fill
1327 * in the transaction header. Note that the transaction header does not
1328 * have a log item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 */
Dave Chinner09243782010-03-08 11:28:28 +11001330 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1331 tp->t_header.th_type = tp->t_type;
1332 tp->t_header.th_num_items = nitems;
1333 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1334 log_vector->i_len = sizeof(xfs_trans_header_t);
1335 log_vector->i_type = XLOG_REG_TYPE_TRANSHDR;
1336}
1337
1338/*
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001339 * The committed item processing consists of calling the committed routine of
1340 * each logged item, updating the item's position in the AIL if necessary, and
1341 * unpinning each item. If the committed routine returns -1, then do nothing
1342 * further with the item because it may have been freed.
1343 *
1344 * Since items are unlocked when they are copied to the incore log, it is
1345 * possible for two transactions to be completing and manipulating the same
1346 * item simultaneously. The AIL lock will protect the lsn field of each item.
1347 * The value of this field can never go backwards.
1348 *
1349 * We unpin the items after repositioning them in the AIL, because otherwise
1350 * they could be immediately flushed and we'd have to race with the flusher
1351 * trying to pull the item from the AIL as we add it.
1352 */
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001353static void
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001354xfs_trans_item_committed(
1355 struct xfs_log_item *lip,
1356 xfs_lsn_t commit_lsn,
1357 int aborted)
1358{
1359 xfs_lsn_t item_lsn;
1360 struct xfs_ail *ailp;
1361
1362 if (aborted)
1363 lip->li_flags |= XFS_LI_ABORTED;
1364 item_lsn = IOP_COMMITTED(lip, commit_lsn);
1365
1366 /* If the committed routine returns -1, item has been freed. */
1367 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1368 return;
1369
1370 /*
1371 * If the returned lsn is greater than what it contained before, update
1372 * the location of the item in the AIL. If it is not, then do nothing.
1373 * Items can never move backwards in the AIL.
1374 *
1375 * While the new lsn should usually be greater, it is possible that a
1376 * later transaction completing simultaneously with an earlier one
1377 * using the same item could complete first with a higher lsn. This
1378 * would cause the earlier transaction to fail the test below.
1379 */
1380 ailp = lip->li_ailp;
1381 spin_lock(&ailp->xa_lock);
1382 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1383 /*
1384 * This will set the item's lsn to item_lsn and update the
1385 * position of the item in the AIL.
1386 *
1387 * xfs_trans_ail_update() drops the AIL lock.
1388 */
1389 xfs_trans_ail_update(ailp, lip, item_lsn);
1390 } else {
1391 spin_unlock(&ailp->xa_lock);
1392 }
1393
1394 /*
1395 * Now that we've repositioned the item in the AIL, unpin it so it can
1396 * be flushed. Pass information about buffer stale state down from the
1397 * log item flags, if anyone else stales the buffer we do not want to
1398 * pay any attention to it.
1399 */
Christoph Hellwig9412e312010-06-23 18:11:15 +10001400 IOP_UNPIN(lip, 0);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001401}
1402
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001403/*
1404 * This is typically called by the LM when a transaction has been fully
1405 * committed to disk. It needs to unpin the items which have
1406 * been logged by the transaction and update their positions
1407 * in the AIL if necessary.
1408 *
1409 * This also gets called when the transactions didn't get written out
1410 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1411 */
1412STATIC void
1413xfs_trans_committed(
Christoph Hellwig4957a442010-10-06 18:41:13 +00001414 void *arg,
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001415 int abortflag)
1416{
Christoph Hellwig4957a442010-10-06 18:41:13 +00001417 struct xfs_trans *tp = arg;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001418 struct xfs_log_item_desc *lidp, *next;
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001419
Christoph Hellwige98c4142010-06-23 18:11:15 +10001420 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001421 xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001422 xfs_trans_free_item_desc(lidp);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001423 }
1424
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001425 xfs_trans_free(tp);
1426}
1427
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001428static inline void
1429xfs_log_item_batch_insert(
1430 struct xfs_ail *ailp,
1431 struct xfs_log_item **log_items,
1432 int nr_items,
1433 xfs_lsn_t commit_lsn)
1434{
1435 int i;
1436
1437 spin_lock(&ailp->xa_lock);
1438 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
1439 xfs_trans_ail_update_bulk(ailp, log_items, nr_items, commit_lsn);
1440
1441 for (i = 0; i < nr_items; i++)
1442 IOP_UNPIN(log_items[i], 0);
1443}
1444
1445/*
1446 * Bulk operation version of xfs_trans_committed that takes a log vector of
1447 * items to insert into the AIL. This uses bulk AIL insertion techniques to
1448 * minimise lock traffic.
1449 */
1450void
1451xfs_trans_committed_bulk(
1452 struct xfs_ail *ailp,
1453 struct xfs_log_vec *log_vector,
1454 xfs_lsn_t commit_lsn,
1455 int aborted)
1456{
1457#define LOG_ITEM_BATCH_SIZE 32
1458 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
1459 struct xfs_log_vec *lv;
1460 int i = 0;
1461
1462 /* unpin all the log items */
1463 for (lv = log_vector; lv; lv = lv->lv_next ) {
1464 struct xfs_log_item *lip = lv->lv_item;
1465 xfs_lsn_t item_lsn;
1466
1467 if (aborted)
1468 lip->li_flags |= XFS_LI_ABORTED;
1469 item_lsn = IOP_COMMITTED(lip, commit_lsn);
1470
1471 /* item_lsn of -1 means the item was freed */
1472 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1473 continue;
1474
1475 if (item_lsn != commit_lsn) {
1476
1477 /*
1478 * Not a bulk update option due to unusual item_lsn.
1479 * Push into AIL immediately, rechecking the lsn once
1480 * we have the ail lock. Then unpin the item.
1481 */
1482 spin_lock(&ailp->xa_lock);
1483 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1484 xfs_trans_ail_update(ailp, lip, item_lsn);
1485 else
1486 spin_unlock(&ailp->xa_lock);
1487 IOP_UNPIN(lip, 0);
1488 continue;
1489 }
1490
1491 /* Item is a candidate for bulk AIL insert. */
1492 log_items[i++] = lv->lv_item;
1493 if (i >= LOG_ITEM_BATCH_SIZE) {
1494 xfs_log_item_batch_insert(ailp, log_items,
1495 LOG_ITEM_BATCH_SIZE, commit_lsn);
1496 i = 0;
1497 }
1498 }
1499
1500 /* make sure we insert the remainder! */
1501 if (i)
1502 xfs_log_item_batch_insert(ailp, log_items, i, commit_lsn);
1503}
1504
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001505/*
1506 * Called from the trans_commit code when we notice that
1507 * the filesystem is in the middle of a forced shutdown.
1508 */
1509STATIC void
1510xfs_trans_uncommit(
1511 struct xfs_trans *tp,
1512 uint flags)
1513{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001514 struct xfs_log_item_desc *lidp;
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001515
Christoph Hellwige98c4142010-06-23 18:11:15 +10001516 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001517 /*
1518 * Unpin all but those that aren't dirty.
1519 */
1520 if (lidp->lid_flags & XFS_LID_DIRTY)
Christoph Hellwig9412e312010-06-23 18:11:15 +10001521 IOP_UNPIN(lidp->lid_item, 1);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001522 }
1523
1524 xfs_trans_unreserve_and_mod_sb(tp);
1525 xfs_trans_unreserve_and_mod_dquots(tp);
1526
Dave Chinner71e330b2010-05-21 14:37:18 +10001527 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001528 xfs_trans_free(tp);
1529}
1530
1531/*
Dave Chinner09243782010-03-08 11:28:28 +11001532 * Format the transaction direct to the iclog. This isolates the physical
1533 * transaction commit operation from the logical operation and hence allows
1534 * other methods to be introduced without affecting the existing commit path.
1535 */
1536static int
1537xfs_trans_commit_iclog(
1538 struct xfs_mount *mp,
1539 struct xfs_trans *tp,
1540 xfs_lsn_t *commit_lsn,
1541 int flags)
1542{
1543 int shutdown;
1544 int error;
1545 int log_flags = 0;
1546 struct xlog_in_core *commit_iclog;
1547#define XFS_TRANS_LOGVEC_COUNT 16
1548 struct xfs_log_iovec log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
1549 struct xfs_log_iovec *log_vector;
1550 uint nvec;
1551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553 /*
1554 * Ask each log item how many log_vector entries it will
1555 * need so we can figure out how many to allocate.
1556 * Try to avoid the kmem_alloc() call in the common case
1557 * by using a vector from the stack when it fits.
1558 */
1559 nvec = xfs_trans_count_vecs(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 if (nvec == 0) {
Dave Chinner09243782010-03-08 11:28:28 +11001561 return ENOMEM; /* triggers a shutdown! */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001562 } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 log_vector = log_vector_fast;
1564 } else {
1565 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
1566 sizeof(xfs_log_iovec_t),
1567 KM_SLEEP);
1568 }
1569
1570 /*
1571 * Fill in the log_vector and pin the logged items, and
1572 * then write the transaction to the log.
1573 */
1574 xfs_trans_fill_vecs(tp, log_vector);
1575
Dave Chinner09243782010-03-08 11:28:28 +11001576 if (flags & XFS_TRANS_RELEASE_LOG_RES)
1577 log_flags = XFS_LOG_REL_PERM_RESERV;
1578
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001579 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /*
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001582 * The transaction is committed incore here, and can go out to disk
1583 * at any time after this call. However, all the items associated
1584 * with the transaction are still locked and pinned in memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 */
Dave Chinner09243782010-03-08 11:28:28 +11001586 *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Dave Chinner09243782010-03-08 11:28:28 +11001588 tp->t_commit_lsn = *commit_lsn;
Dave Chinnered3b4d62010-05-21 12:07:08 +10001589 trace_xfs_trans_commit_lsn(tp);
1590
Dave Chinner09243782010-03-08 11:28:28 +11001591 if (nvec > XFS_TRANS_LOGVEC_COUNT)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001592 kmem_free(log_vector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 /*
1595 * If we got a log write error. Unpin the logitems that we
1596 * had pinned, clean up, free trans structure, and return error.
1597 */
Dave Chinner09243782010-03-08 11:28:28 +11001598 if (error || *commit_lsn == -1) {
Nathan Scott59c1b082006-06-09 14:59:13 +10001599 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
1601 return XFS_ERROR(EIO);
1602 }
1603
1604 /*
1605 * Once the transaction has committed, unused
1606 * reservations need to be released and changes to
1607 * the superblock need to be reflected in the in-core
1608 * version. Do that now.
1609 */
1610 xfs_trans_unreserve_and_mod_sb(tp);
1611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 /*
1613 * Tell the LM to call the transaction completion routine
1614 * when the log write with LSN commit_lsn completes (e.g.
1615 * when the transaction commit really hits the on-disk log).
1616 * After this call we cannot reference tp, because the call
1617 * can happen at any time and the call will free the transaction
1618 * structure pointed to by tp. The only case where we call
1619 * the completion routine (xfs_trans_committed) directly is
1620 * if the log is turned off on a debug kernel or we're
1621 * running in simulation mode (the log is explicitly turned
1622 * off).
1623 */
Christoph Hellwig4957a442010-10-06 18:41:13 +00001624 tp->t_logcb.cb_func = xfs_trans_committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 tp->t_logcb.cb_arg = tp;
1626
1627 /*
1628 * We need to pass the iclog buffer which was used for the
1629 * transaction commit record into this function, and attach
1630 * the callback to it. The callback must be attached before
1631 * the items are unlocked to avoid racing with other threads
1632 * waiting for an item to unlock.
1633 */
1634 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
1635
1636 /*
1637 * Mark this thread as no longer being in a transaction
1638 */
Nathan Scott59c1b082006-06-09 14:59:13 +10001639 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
1641 /*
1642 * Once all the items of the transaction have been copied
1643 * to the in core log and the callback is attached, the
1644 * items can be unlocked.
1645 *
1646 * This will free descriptors pointing to items which were
1647 * not logged since there is nothing more to do with them.
1648 * For items which were logged, we will keep pointers to them
1649 * so they can be unpinned after the transaction commits to disk.
1650 * This will also stamp each modified meta-data item with
1651 * the commit lsn of this transaction for dependency tracking
1652 * purposes.
1653 */
Dave Chinner09243782010-03-08 11:28:28 +11001654 xfs_trans_unlock_items(tp, *commit_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656 /*
1657 * If we detected a log error earlier, finish committing
1658 * the transaction now (unpin log items, etc).
1659 *
1660 * Order is critical here, to avoid using the transaction
1661 * pointer after its been freed (by xfs_trans_committed
1662 * either here now, or as a callback). We cannot do this
1663 * step inside xfs_log_notify as was done earlier because
1664 * of this issue.
1665 */
1666 if (shutdown)
1667 xfs_trans_committed(tp, XFS_LI_ABORTED);
1668
1669 /*
1670 * Now that the xfs_trans_committed callback has been attached,
1671 * and the items are released we can finally allow the iclog to
1672 * go to disk.
1673 */
Dave Chinner09243782010-03-08 11:28:28 +11001674 return xfs_log_release_iclog(mp, commit_iclog);
1675}
1676
Dave Chinner71e330b2010-05-21 14:37:18 +10001677/*
1678 * Walk the log items and allocate log vector structures for
1679 * each item large enough to fit all the vectors they require.
1680 * Note that this format differs from the old log vector format in
1681 * that there is no transaction header in these log vectors.
1682 */
1683STATIC struct xfs_log_vec *
1684xfs_trans_alloc_log_vecs(
1685 xfs_trans_t *tp)
1686{
Christoph Hellwige98c4142010-06-23 18:11:15 +10001687 struct xfs_log_item_desc *lidp;
Dave Chinner71e330b2010-05-21 14:37:18 +10001688 struct xfs_log_vec *lv = NULL;
1689 struct xfs_log_vec *ret_lv = NULL;
1690
Dave Chinner71e330b2010-05-21 14:37:18 +10001691
1692 /* Bail out if we didn't find a log item. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001693 if (list_empty(&tp->t_items)) {
Dave Chinner71e330b2010-05-21 14:37:18 +10001694 ASSERT(0);
1695 return NULL;
1696 }
1697
Christoph Hellwige98c4142010-06-23 18:11:15 +10001698 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner71e330b2010-05-21 14:37:18 +10001699 struct xfs_log_vec *new_lv;
1700
1701 /* Skip items which aren't dirty in this transaction. */
Christoph Hellwige98c4142010-06-23 18:11:15 +10001702 if (!(lidp->lid_flags & XFS_LID_DIRTY))
Dave Chinner71e330b2010-05-21 14:37:18 +10001703 continue;
Dave Chinner71e330b2010-05-21 14:37:18 +10001704
1705 /* Skip items that do not have any vectors for writing */
1706 lidp->lid_size = IOP_SIZE(lidp->lid_item);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001707 if (!lidp->lid_size)
Dave Chinner71e330b2010-05-21 14:37:18 +10001708 continue;
Dave Chinner71e330b2010-05-21 14:37:18 +10001709
1710 new_lv = kmem_zalloc(sizeof(*new_lv) +
1711 lidp->lid_size * sizeof(struct xfs_log_iovec),
1712 KM_SLEEP);
1713
1714 /* The allocated iovec region lies beyond the log vector. */
1715 new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1];
1716 new_lv->lv_niovecs = lidp->lid_size;
1717 new_lv->lv_item = lidp->lid_item;
1718 if (!ret_lv)
1719 ret_lv = new_lv;
1720 else
1721 lv->lv_next = new_lv;
1722 lv = new_lv;
Dave Chinner71e330b2010-05-21 14:37:18 +10001723 }
1724
1725 return ret_lv;
1726}
1727
1728static int
1729xfs_trans_commit_cil(
1730 struct xfs_mount *mp,
1731 struct xfs_trans *tp,
1732 xfs_lsn_t *commit_lsn,
1733 int flags)
1734{
1735 struct xfs_log_vec *log_vector;
1736 int error;
1737
1738 /*
1739 * Get each log item to allocate a vector structure for
1740 * the log item to to pass to the log write code. The
1741 * CIL commit code will format the vector and save it away.
1742 */
1743 log_vector = xfs_trans_alloc_log_vecs(tp);
1744 if (!log_vector)
1745 return ENOMEM;
1746
1747 error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags);
1748 if (error)
1749 return error;
1750
1751 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner71e330b2010-05-21 14:37:18 +10001752 xfs_trans_free(tp);
1753 return 0;
1754}
Dave Chinner09243782010-03-08 11:28:28 +11001755
1756/*
1757 * xfs_trans_commit
1758 *
1759 * Commit the given transaction to the log a/synchronously.
1760 *
1761 * XFS disk error handling mechanism is not based on a typical
1762 * transaction abort mechanism. Logically after the filesystem
1763 * gets marked 'SHUTDOWN', we can't let any new transactions
1764 * be durable - ie. committed to disk - because some metadata might
1765 * be inconsistent. In such cases, this returns an error, and the
1766 * caller may assume that all locked objects joined to the transaction
1767 * have already been unlocked as if the commit had succeeded.
1768 * Do not reference the transaction structure after this call.
1769 */
Dave Chinner09243782010-03-08 11:28:28 +11001770int
1771_xfs_trans_commit(
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001772 struct xfs_trans *tp,
1773 uint flags,
1774 int *log_flushed)
Dave Chinner09243782010-03-08 11:28:28 +11001775{
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001776 struct xfs_mount *mp = tp->t_mountp;
Dave Chinner09243782010-03-08 11:28:28 +11001777 xfs_lsn_t commit_lsn = -1;
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001778 int error = 0;
Dave Chinner09243782010-03-08 11:28:28 +11001779 int log_flags = 0;
1780 int sync = tp->t_flags & XFS_TRANS_SYNC;
Dave Chinner09243782010-03-08 11:28:28 +11001781
1782 /*
1783 * Determine whether this commit is releasing a permanent
1784 * log reservation or not.
1785 */
1786 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1787 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1788 log_flags = XFS_LOG_REL_PERM_RESERV;
1789 }
1790
1791 /*
1792 * If there is nothing to be logged by the transaction,
1793 * then unlock all of the items associated with the
1794 * transaction and free the transaction structure.
1795 * Also make sure to return any reserved blocks to
1796 * the free pool.
1797 */
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001798 if (!(tp->t_flags & XFS_TRANS_DIRTY))
1799 goto out_unreserve;
1800
1801 if (XFS_FORCED_SHUTDOWN(mp)) {
1802 error = XFS_ERROR(EIO);
1803 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001804 }
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001805
Dave Chinner09243782010-03-08 11:28:28 +11001806 ASSERT(tp->t_ticket != NULL);
1807
1808 /*
1809 * If we need to update the superblock, then do it now.
1810 */
1811 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1812 xfs_trans_apply_sb_deltas(tp);
1813 xfs_trans_apply_dquot_deltas(tp);
1814
Dave Chinner71e330b2010-05-21 14:37:18 +10001815 if (mp->m_flags & XFS_MOUNT_DELAYLOG)
1816 error = xfs_trans_commit_cil(mp, tp, &commit_lsn, flags);
1817 else
1818 error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags);
1819
Dave Chinner09243782010-03-08 11:28:28 +11001820 if (error == ENOMEM) {
1821 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001822 error = XFS_ERROR(EIO);
1823 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
1826 /*
1827 * If the transaction needs to be synchronous, then force the
1828 * log out now and wait for it.
1829 */
1830 if (sync) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001831 if (!error) {
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001832 error = _xfs_log_force_lsn(mp, commit_lsn,
1833 XFS_LOG_SYNC, log_flushed);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 XFS_STATS_INC(xs_trans_sync);
1836 } else {
1837 XFS_STATS_INC(xs_trans_async);
1838 }
1839
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001840 return error;
1841
1842out_unreserve:
1843 xfs_trans_unreserve_and_mod_sb(tp);
1844
1845 /*
1846 * It is indeed possible for the transaction to be not dirty but
1847 * the dqinfo portion to be. All that means is that we have some
1848 * (non-persistent) quota reservations that need to be unreserved.
1849 */
1850 xfs_trans_unreserve_and_mod_dquots(tp);
1851 if (tp->t_ticket) {
1852 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1853 if (commit_lsn == -1 && !error)
1854 error = XFS_ERROR(EIO);
1855 }
1856 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner71e330b2010-05-21 14:37:18 +10001857 xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001858 xfs_trans_free(tp);
1859
1860 XFS_STATS_INC(xs_trans_empty);
1861 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862}
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 * Unlock all of the transaction's items and free the transaction.
1866 * The transaction must not have modified any of its items, because
1867 * there is no way to restore them to their previous state.
1868 *
1869 * If the transaction has made a log reservation, make sure to release
1870 * it as well.
1871 */
1872void
1873xfs_trans_cancel(
1874 xfs_trans_t *tp,
1875 int flags)
1876{
1877 int log_flags;
Ryan Hankins0733af22006-01-11 15:36:44 +11001878 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
1880 /*
1881 * See if the caller is being too lazy to figure out if
1882 * the transaction really needs an abort.
1883 */
1884 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1885 flags &= ~XFS_TRANS_ABORT;
1886 /*
1887 * See if the caller is relying on us to shut down the
1888 * filesystem. This happens in paths where we detect
1889 * corruption and decide to give up.
1890 */
Nathan Scott60a204f2006-01-11 15:37:00 +11001891 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
Ryan Hankins0733af22006-01-11 15:36:44 +11001892 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001893 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Nathan Scott60a204f2006-01-11 15:37:00 +11001894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895#ifdef DEBUG
Christoph Hellwige98c4142010-06-23 18:11:15 +10001896 if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1897 struct xfs_log_item_desc *lidp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Christoph Hellwige98c4142010-06-23 18:11:15 +10001899 list_for_each_entry(lidp, &tp->t_items, lid_trans)
1900 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
1902#endif
1903 xfs_trans_unreserve_and_mod_sb(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001904 xfs_trans_unreserve_and_mod_dquots(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
1906 if (tp->t_ticket) {
1907 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1908 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1909 log_flags = XFS_LOG_REL_PERM_RESERV;
1910 } else {
1911 log_flags = 0;
1912 }
Ryan Hankins0733af22006-01-11 15:36:44 +11001913 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
1915
1916 /* mark this thread as no longer being in a transaction */
Nathan Scott59c1b082006-06-09 14:59:13 +10001917 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Dave Chinner71e330b2010-05-21 14:37:18 +10001919 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 xfs_trans_free(tp);
1921}
1922
Niv Sardi322ff6b2008-08-13 16:05:49 +10001923/*
1924 * Roll from one trans in the sequence of PERMANENT transactions to
1925 * the next: permanent transactions are only flushed out when
1926 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1927 * as possible to let chunks of it go to the log. So we commit the
1928 * chunk we've been working on and get a new transaction to continue.
1929 */
1930int
1931xfs_trans_roll(
1932 struct xfs_trans **tpp,
1933 struct xfs_inode *dp)
1934{
1935 struct xfs_trans *trans;
1936 unsigned int logres, count;
1937 int error;
1938
1939 /*
1940 * Ensure that the inode is always logged.
1941 */
1942 trans = *tpp;
1943 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1944
1945 /*
1946 * Copy the critical parameters from one trans to the next.
1947 */
1948 logres = trans->t_log_res;
1949 count = trans->t_log_count;
1950 *tpp = xfs_trans_dup(trans);
1951
1952 /*
1953 * Commit the current transaction.
1954 * If this commit failed, then it'd just unlock those items that
1955 * are not marked ihold. That also means that a filesystem shutdown
1956 * is in progress. The caller takes the responsibility to cancel
1957 * the duplicate transaction that gets returned.
1958 */
1959 error = xfs_trans_commit(trans, 0);
1960 if (error)
1961 return (error);
1962
1963 trans = *tpp;
1964
1965 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001966 * transaction commit worked ok so we can drop the extra ticket
1967 * reference that we gained in xfs_trans_dup()
1968 */
1969 xfs_log_ticket_put(trans->t_ticket);
1970
1971
1972 /*
Niv Sardi322ff6b2008-08-13 16:05:49 +10001973 * Reserve space in the log for th next transaction.
1974 * This also pushes items in the "AIL", the list of logged items,
1975 * out to disk if they are taking up space at the tail of the log
1976 * that we want to use. This requires that either nothing be locked
1977 * across this call, or that anything that is locked be logged in
1978 * the prior and the next transactions.
1979 */
1980 error = xfs_trans_reserve(trans, 0, logres, 0,
1981 XFS_TRANS_PERM_LOG_RES, count);
1982 /*
1983 * Ensure that the inode is in the new transaction and locked.
1984 */
1985 if (error)
1986 return error;
1987
Christoph Hellwig898621d2010-06-24 11:36:58 +10001988 xfs_trans_ijoin(trans, dp);
Niv Sardi322ff6b2008-08-13 16:05:49 +10001989 return 0;
1990}