blob: 329b06aba1c2c59ab1b973a3a956a3af23c68abf [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.
Alex Elderb2ce3972011-07-11 09:51:44 -0500571 *
572 * Dynamically allocate the transaction structure from the transaction
573 * zone, initialize it, and return it to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 */
Alex Elderb2ce3972011-07-11 09:51:44 -0500575xfs_trans_t *
576xfs_trans_alloc(
577 xfs_mount_t *mp,
578 uint type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Alex Elderb2ce3972011-07-11 09:51:44 -0500580 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
581 return _xfs_trans_alloc(mp, type, KM_SLEEP);
582}
583
584xfs_trans_t *
585_xfs_trans_alloc(
586 xfs_mount_t *mp,
587 uint type,
588 uint memflags)
589{
590 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
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{
Christoph Hellwig8a072a42011-04-24 19:06:17 +0000611 xfs_alloc_busy_sort(&tp->t_busy);
Christoph Hellwige84661a2011-05-20 13:45:32 +0000612 xfs_alloc_busy_clear(tp->t_mountp, &tp->t_busy, false);
Dave Chinnered3b4d62010-05-21 12:07:08 +1000613
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +1100614 atomic_dec(&tp->t_mountp->m_active_trans);
615 xfs_trans_free_dqinfo(tp);
616 kmem_zone_free(xfs_trans_zone, tp);
617}
618
619/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 * This is called to create a new transaction which will share the
621 * permanent log reservation of the given transaction. The remaining
622 * unused block and rt extent reservations are also inherited. This
623 * implies that the original transaction is no longer allowed to allocate
624 * blocks. Locks and log items, however, are no inherited. They must
625 * be added to the new transaction explicitly.
626 */
627xfs_trans_t *
628xfs_trans_dup(
629 xfs_trans_t *tp)
630{
631 xfs_trans_t *ntp;
632
633 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
634
635 /*
636 * Initialize the new transaction structure.
637 */
638 ntp->t_magic = XFS_TRANS_MAGIC;
639 ntp->t_type = tp->t_type;
640 ntp->t_mountp = tp->t_mountp;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000641 INIT_LIST_HEAD(&ntp->t_items);
Dave Chinnered3b4d62010-05-21 12:07:08 +1000642 INIT_LIST_HEAD(&ntp->t_busy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 ASSERT(tp->t_ticket != NULL);
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100648 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
650 tp->t_blk_res = tp->t_blk_res_used;
651 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
652 tp->t_rtx_res = tp->t_rtx_res_used;
Nathan Scott59c1b082006-06-09 14:59:13 +1000653 ntp->t_pflags = tp->t_pflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Christoph Hellwig7d095252009-06-08 15:33:32 +0200655 xfs_trans_dup_dqinfo(tp, ntp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 atomic_inc(&tp->t_mountp->m_active_trans);
658 return ntp;
659}
660
661/*
662 * This is called to reserve free disk blocks and log space for the
663 * given transaction. This must be done before allocating any resources
664 * within the transaction.
665 *
666 * This will return ENOSPC if there are not enough blocks available.
667 * It will sleep waiting for available log space.
668 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
669 * is used by long running transactions. If any one of the reservations
670 * fails then they will all be backed out.
671 *
672 * This does not do quota reservations. That typically is done by the
673 * caller afterwards.
674 */
675int
676xfs_trans_reserve(
677 xfs_trans_t *tp,
678 uint blocks,
679 uint logspace,
680 uint rtextents,
681 uint flags,
682 uint logcount)
683{
684 int log_flags;
Nathan Scott59c1b082006-06-09 14:59:13 +1000685 int error = 0;
686 int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
688 /* Mark this thread as being in a transaction */
Nathan Scott59c1b082006-06-09 14:59:13 +1000689 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 /*
692 * Attempt to reserve the needed disk blocks by decrementing
693 * the number needed from the number available. This will
694 * fail if the count would go below zero.
695 */
696 if (blocks > 0) {
Christoph Hellwig96540c72010-09-30 02:25:55 +0000697 error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100698 -((int64_t)blocks), rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (error != 0) {
Nathan Scott59c1b082006-06-09 14:59:13 +1000700 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return (XFS_ERROR(ENOSPC));
702 }
703 tp->t_blk_res += blocks;
704 }
705
706 /*
707 * Reserve the log space needed for this transaction.
708 */
709 if (logspace > 0) {
710 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
711 ASSERT((tp->t_log_count == 0) ||
712 (tp->t_log_count == logcount));
713 if (flags & XFS_TRANS_PERM_LOG_RES) {
714 log_flags = XFS_LOG_PERM_RESERV;
715 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
716 } else {
717 ASSERT(tp->t_ticket == NULL);
718 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
719 log_flags = 0;
720 }
721
722 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
723 &tp->t_ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000724 XFS_TRANSACTION, log_flags, tp->t_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (error) {
726 goto undo_blocks;
727 }
728 tp->t_log_res = logspace;
729 tp->t_log_count = logcount;
730 }
731
732 /*
733 * Attempt to reserve the needed realtime extents by decrementing
734 * the number needed from the number available. This will
735 * fail if the count would go below zero.
736 */
737 if (rtextents > 0) {
738 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100739 -((int64_t)rtextents), rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 if (error) {
741 error = XFS_ERROR(ENOSPC);
742 goto undo_log;
743 }
744 tp->t_rtx_res += rtextents;
745 }
746
747 return 0;
748
749 /*
750 * Error cases jump to one of these labels to undo any
751 * reservations which have already been performed.
752 */
753undo_log:
754 if (logspace > 0) {
755 if (flags & XFS_TRANS_PERM_LOG_RES) {
756 log_flags = XFS_LOG_REL_PERM_RESERV;
757 } else {
758 log_flags = 0;
759 }
760 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
761 tp->t_ticket = NULL;
762 tp->t_log_res = 0;
763 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
764 }
765
766undo_blocks:
767 if (blocks > 0) {
Christoph Hellwig96540c72010-09-30 02:25:55 +0000768 xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
David Chinner20f4ebf2007-02-10 18:36:10 +1100769 (int64_t)blocks, rsvd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 tp->t_blk_res = 0;
771 }
772
Nathan Scott59c1b082006-06-09 14:59:13 +1000773 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Nathan Scott59c1b082006-06-09 14:59:13 +1000775 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 * Record the indicated change to the given field for application
780 * to the file system's superblock when the transaction commits.
781 * For now, just store the change in the transaction structure.
782 *
783 * Mark the transaction structure to indicate that the superblock
784 * needs to be updated before committing.
David Chinner92821e22007-05-24 15:26:31 +1000785 *
786 * Because we may not be keeping track of allocated/free inodes and
787 * used filesystem blocks in the superblock, we do not mark the
788 * superblock dirty in this transaction if we modify these fields.
789 * We still need to update the transaction deltas so that they get
790 * applied to the incore superblock, but we don't want them to
791 * cause the superblock to get locked and logged if these are the
792 * only fields in the superblock that the transaction modifies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 */
794void
795xfs_trans_mod_sb(
796 xfs_trans_t *tp,
797 uint field,
David Chinner20f4ebf2007-02-10 18:36:10 +1100798 int64_t delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
David Chinner92821e22007-05-24 15:26:31 +1000800 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
801 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
803 switch (field) {
804 case XFS_TRANS_SB_ICOUNT:
805 tp->t_icount_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000806 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
807 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 break;
809 case XFS_TRANS_SB_IFREE:
810 tp->t_ifree_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000811 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
812 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 break;
814 case XFS_TRANS_SB_FDBLOCKS:
815 /*
816 * Track the number of blocks allocated in the
817 * transaction. Make sure it does not exceed the
818 * number reserved.
819 */
820 if (delta < 0) {
821 tp->t_blk_res_used += (uint)-delta;
822 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
823 }
824 tp->t_fdblocks_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000825 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
826 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 break;
828 case XFS_TRANS_SB_RES_FDBLOCKS:
829 /*
830 * The allocation has already been applied to the
831 * in-core superblock's counter. This should only
832 * be applied to the on-disk superblock.
833 */
834 ASSERT(delta < 0);
835 tp->t_res_fdblocks_delta += delta;
David Chinner92821e22007-05-24 15:26:31 +1000836 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
837 flags &= ~XFS_TRANS_SB_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 break;
839 case XFS_TRANS_SB_FREXTENTS:
840 /*
841 * Track the number of blocks allocated in the
842 * transaction. Make sure it does not exceed the
843 * number reserved.
844 */
845 if (delta < 0) {
846 tp->t_rtx_res_used += (uint)-delta;
847 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
848 }
849 tp->t_frextents_delta += delta;
850 break;
851 case XFS_TRANS_SB_RES_FREXTENTS:
852 /*
853 * The allocation has already been applied to the
Nathan Scottc41564b2006-03-29 08:55:14 +1000854 * in-core superblock's counter. This should only
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 * be applied to the on-disk superblock.
856 */
857 ASSERT(delta < 0);
858 tp->t_res_frextents_delta += delta;
859 break;
860 case XFS_TRANS_SB_DBLOCKS:
861 ASSERT(delta > 0);
862 tp->t_dblocks_delta += delta;
863 break;
864 case XFS_TRANS_SB_AGCOUNT:
865 ASSERT(delta > 0);
866 tp->t_agcount_delta += delta;
867 break;
868 case XFS_TRANS_SB_IMAXPCT:
869 tp->t_imaxpct_delta += delta;
870 break;
871 case XFS_TRANS_SB_REXTSIZE:
872 tp->t_rextsize_delta += delta;
873 break;
874 case XFS_TRANS_SB_RBMBLOCKS:
875 tp->t_rbmblocks_delta += delta;
876 break;
877 case XFS_TRANS_SB_RBLOCKS:
878 tp->t_rblocks_delta += delta;
879 break;
880 case XFS_TRANS_SB_REXTENTS:
881 tp->t_rextents_delta += delta;
882 break;
883 case XFS_TRANS_SB_REXTSLOG:
884 tp->t_rextslog_delta += delta;
885 break;
886 default:
887 ASSERT(0);
888 return;
889 }
890
David Chinner210c6f12007-05-24 15:26:51 +1000891 tp->t_flags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
894/*
895 * xfs_trans_apply_sb_deltas() is called from the commit code
896 * to bring the superblock buffer into the current transaction
897 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
898 *
899 * For now we just look at each field allowed to change and change
900 * it if necessary.
901 */
902STATIC void
903xfs_trans_apply_sb_deltas(
904 xfs_trans_t *tp)
905{
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000906 xfs_dsb_t *sbp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 xfs_buf_t *bp;
908 int whole = 0;
909
910 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
911 sbp = XFS_BUF_TO_SBP(bp);
912
913 /*
914 * Check that superblock mods match the mods made to AGF counters.
915 */
916 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
917 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
918 tp->t_ag_btree_delta));
919
David Chinner92821e22007-05-24 15:26:31 +1000920 /*
921 * Only update the superblock counters if we are logging them
922 */
923 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000924 if (tp->t_icount_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800925 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000926 if (tp->t_ifree_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800927 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000928 if (tp->t_fdblocks_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800929 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000930 if (tp->t_res_fdblocks_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800931 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 }
933
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000934 if (tp->t_frextents_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800935 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000936 if (tp->t_res_frextents_delta)
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800937 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000938
939 if (tp->t_dblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800940 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 whole = 1;
942 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000943 if (tp->t_agcount_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800944 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 whole = 1;
946 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000947 if (tp->t_imaxpct_delta) {
948 sbp->sb_imax_pct += tp->t_imaxpct_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 whole = 1;
950 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000951 if (tp->t_rextsize_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800952 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 whole = 1;
954 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000955 if (tp->t_rbmblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800956 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 whole = 1;
958 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000959 if (tp->t_rblocks_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800960 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 whole = 1;
962 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000963 if (tp->t_rextents_delta) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800964 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 whole = 1;
966 }
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000967 if (tp->t_rextslog_delta) {
968 sbp->sb_rextslog += tp->t_rextslog_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 whole = 1;
970 }
971
972 if (whole)
973 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000974 * Log the whole thing, the fields are noncontiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000976 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 else
978 /*
979 * Since all the modifiable fields are contiguous, we
980 * can get away with this.
981 */
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +1000982 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
983 offsetof(xfs_dsb_t, sb_frextents) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 sizeof(sbp->sb_frextents) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985}
986
987/*
David Chinner45c34142007-06-18 16:49:44 +1000988 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
989 * and apply superblock counter changes to the in-core superblock. The
990 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
991 * applied to the in-core superblock. The idea is that that has already been
992 * done.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 *
994 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
David Chinner45c34142007-06-18 16:49:44 +1000995 * However, we have to ensure that we only modify each superblock field only
996 * once because the application of the delta values may not be atomic. That can
997 * lead to ENOSPC races occurring if we have two separate modifcations of the
998 * free space counter to put back the entire reservation and then take away
999 * what we used.
1000 *
1001 * If we are not logging superblock counters, then the inode allocated/free and
1002 * used block counts are not updated in the on disk superblock. In this case,
1003 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1004 * still need to update the incore superblock with the changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001006void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007xfs_trans_unreserve_and_mod_sb(
1008 xfs_trans_t *tp)
1009{
Christoph Hellwig1b040712010-09-30 02:25:56 +00001010 xfs_mod_sb_t msb[9]; /* If you add cases, add entries */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 xfs_mod_sb_t *msbp;
David Chinner92821e22007-05-24 15:26:31 +10001012 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /* REFERENCED */
1014 int error;
1015 int rsvd;
David Chinner45c34142007-06-18 16:49:44 +10001016 int64_t blkdelta = 0;
1017 int64_t rtxdelta = 0;
Christoph Hellwig1b040712010-09-30 02:25:56 +00001018 int64_t idelta = 0;
1019 int64_t ifreedelta = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 msbp = msb;
1022 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1023
Christoph Hellwig1b040712010-09-30 02:25:56 +00001024 /* calculate deltas */
David Chinner45c34142007-06-18 16:49:44 +10001025 if (tp->t_blk_res > 0)
1026 blkdelta = tp->t_blk_res;
David Chinner45c34142007-06-18 16:49:44 +10001027 if ((tp->t_fdblocks_delta != 0) &&
1028 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1029 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1030 blkdelta += tp->t_fdblocks_delta;
1031
David Chinner45c34142007-06-18 16:49:44 +10001032 if (tp->t_rtx_res > 0)
1033 rtxdelta = tp->t_rtx_res;
David Chinner45c34142007-06-18 16:49:44 +10001034 if ((tp->t_frextents_delta != 0) &&
1035 (tp->t_flags & XFS_TRANS_SB_DIRTY))
1036 rtxdelta += tp->t_frextents_delta;
1037
Christoph Hellwig1b040712010-09-30 02:25:56 +00001038 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1039 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1040 idelta = tp->t_icount_delta;
1041 ifreedelta = tp->t_ifree_delta;
1042 }
1043
1044 /* apply the per-cpu counters */
1045 if (blkdelta) {
1046 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1047 blkdelta, rsvd);
1048 if (error)
1049 goto out;
1050 }
1051
1052 if (idelta) {
1053 error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1054 idelta, rsvd);
1055 if (error)
1056 goto out_undo_fdblocks;
1057 }
1058
1059 if (ifreedelta) {
1060 error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1061 ifreedelta, rsvd);
1062 if (error)
1063 goto out_undo_icount;
1064 }
1065
1066 /* apply remaining deltas */
David Chinner45c34142007-06-18 16:49:44 +10001067 if (rtxdelta != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 msbp->msb_field = XFS_SBS_FREXTENTS;
David Chinner45c34142007-06-18 16:49:44 +10001069 msbp->msb_delta = rtxdelta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 msbp++;
1071 }
1072
David Chinner92821e22007-05-24 15:26:31 +10001073 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (tp->t_dblocks_delta != 0) {
1075 msbp->msb_field = XFS_SBS_DBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001076 msbp->msb_delta = tp->t_dblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 msbp++;
1078 }
1079 if (tp->t_agcount_delta != 0) {
1080 msbp->msb_field = XFS_SBS_AGCOUNT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001081 msbp->msb_delta = tp->t_agcount_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 msbp++;
1083 }
1084 if (tp->t_imaxpct_delta != 0) {
1085 msbp->msb_field = XFS_SBS_IMAX_PCT;
David Chinner20f4ebf2007-02-10 18:36:10 +11001086 msbp->msb_delta = tp->t_imaxpct_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 msbp++;
1088 }
1089 if (tp->t_rextsize_delta != 0) {
1090 msbp->msb_field = XFS_SBS_REXTSIZE;
David Chinner20f4ebf2007-02-10 18:36:10 +11001091 msbp->msb_delta = tp->t_rextsize_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 msbp++;
1093 }
1094 if (tp->t_rbmblocks_delta != 0) {
1095 msbp->msb_field = XFS_SBS_RBMBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001096 msbp->msb_delta = tp->t_rbmblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 msbp++;
1098 }
1099 if (tp->t_rblocks_delta != 0) {
1100 msbp->msb_field = XFS_SBS_RBLOCKS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001101 msbp->msb_delta = tp->t_rblocks_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 msbp++;
1103 }
1104 if (tp->t_rextents_delta != 0) {
1105 msbp->msb_field = XFS_SBS_REXTENTS;
David Chinner20f4ebf2007-02-10 18:36:10 +11001106 msbp->msb_delta = tp->t_rextents_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 msbp++;
1108 }
1109 if (tp->t_rextslog_delta != 0) {
1110 msbp->msb_field = XFS_SBS_REXTSLOG;
David Chinner20f4ebf2007-02-10 18:36:10 +11001111 msbp->msb_delta = tp->t_rextslog_delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 msbp++;
1113 }
1114 }
1115
1116 /*
1117 * If we need to change anything, do it.
1118 */
1119 if (msbp > msb) {
1120 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1121 (uint)(msbp - msb), rsvd);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001122 if (error)
1123 goto out_undo_ifreecount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
Christoph Hellwig1b040712010-09-30 02:25:56 +00001125
1126 return;
1127
1128out_undo_ifreecount:
1129 if (ifreedelta)
1130 xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1131out_undo_icount:
1132 if (idelta)
1133 xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1134out_undo_fdblocks:
1135 if (blkdelta)
1136 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1137out:
Jesper Juhl1884bd82010-12-25 20:14:53 +00001138 ASSERT(error == 0);
Christoph Hellwig1b040712010-09-30 02:25:56 +00001139 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
1141
Dave Chinner09243782010-03-08 11:28:28 +11001142/*
Christoph Hellwige98c4142010-06-23 18:11:15 +10001143 * Add the given log item to the transaction's list of log items.
1144 *
1145 * The log item will now point to its new descriptor with its li_desc field.
1146 */
1147void
1148xfs_trans_add_item(
1149 struct xfs_trans *tp,
1150 struct xfs_log_item *lip)
1151{
1152 struct xfs_log_item_desc *lidp;
1153
1154 ASSERT(lip->li_mountp = tp->t_mountp);
1155 ASSERT(lip->li_ailp = tp->t_mountp->m_ail);
1156
Dave Chinner43869702010-07-20 17:53:44 +10001157 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
Christoph Hellwige98c4142010-06-23 18:11:15 +10001158
1159 lidp->lid_item = lip;
1160 lidp->lid_flags = 0;
Christoph Hellwige98c4142010-06-23 18:11:15 +10001161 list_add_tail(&lidp->lid_trans, &tp->t_items);
1162
1163 lip->li_desc = lidp;
1164}
1165
1166STATIC void
1167xfs_trans_free_item_desc(
1168 struct xfs_log_item_desc *lidp)
1169{
1170 list_del_init(&lidp->lid_trans);
1171 kmem_zone_free(xfs_log_item_desc_zone, lidp);
1172}
1173
1174/*
1175 * Unlink and free the given descriptor.
1176 */
1177void
1178xfs_trans_del_item(
1179 struct xfs_log_item *lip)
1180{
1181 xfs_trans_free_item_desc(lip->li_desc);
1182 lip->li_desc = NULL;
1183}
1184
1185/*
1186 * Unlock all of the items of a transaction and free all the descriptors
1187 * of that transaction.
1188 */
Dave Chinnerd17c7012010-08-24 11:42:52 +10001189void
Christoph Hellwige98c4142010-06-23 18:11:15 +10001190xfs_trans_free_items(
1191 struct xfs_trans *tp,
1192 xfs_lsn_t commit_lsn,
1193 int flags)
1194{
1195 struct xfs_log_item_desc *lidp, *next;
1196
1197 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1198 struct xfs_log_item *lip = lidp->lid_item;
1199
1200 lip->li_desc = NULL;
1201
1202 if (commit_lsn != NULLCOMMITLSN)
1203 IOP_COMMITTING(lip, commit_lsn);
1204 if (flags & XFS_TRANS_ABORT)
1205 lip->li_flags |= XFS_LI_ABORTED;
1206 IOP_UNLOCK(lip);
1207
1208 xfs_trans_free_item_desc(lidp);
1209 }
1210}
1211
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001212static inline void
1213xfs_log_item_batch_insert(
1214 struct xfs_ail *ailp,
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001215 struct xfs_ail_cursor *cur,
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001216 struct xfs_log_item **log_items,
1217 int nr_items,
1218 xfs_lsn_t commit_lsn)
1219{
1220 int i;
1221
1222 spin_lock(&ailp->xa_lock);
1223 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001224 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001225
1226 for (i = 0; i < nr_items; i++)
1227 IOP_UNPIN(log_items[i], 0);
1228}
1229
1230/*
1231 * Bulk operation version of xfs_trans_committed that takes a log vector of
1232 * items to insert into the AIL. This uses bulk AIL insertion techniques to
1233 * minimise lock traffic.
Dave Chinnere34a3142011-01-27 12:13:35 +11001234 *
1235 * If we are called with the aborted flag set, it is because a log write during
1236 * a CIL checkpoint commit has failed. In this case, all the items in the
1237 * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
1238 * means that checkpoint commit abort handling is treated exactly the same
1239 * as an iclog write error even though we haven't started any IO yet. Hence in
1240 * this case all we need to do is IOP_COMMITTED processing, followed by an
1241 * IOP_UNPIN(aborted) call.
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001242 *
1243 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
1244 * at the end of the AIL, the insert cursor avoids the need to walk
1245 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
1246 * call. This saves a lot of needless list walking and is a net win, even
1247 * though it slightly increases that amount of AIL lock traffic to set it up
1248 * and tear it down.
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001249 */
1250void
1251xfs_trans_committed_bulk(
1252 struct xfs_ail *ailp,
1253 struct xfs_log_vec *log_vector,
1254 xfs_lsn_t commit_lsn,
1255 int aborted)
1256{
1257#define LOG_ITEM_BATCH_SIZE 32
1258 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
1259 struct xfs_log_vec *lv;
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001260 struct xfs_ail_cursor cur;
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001261 int i = 0;
1262
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001263 spin_lock(&ailp->xa_lock);
1264 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
1265 spin_unlock(&ailp->xa_lock);
1266
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001267 /* unpin all the log items */
1268 for (lv = log_vector; lv; lv = lv->lv_next ) {
1269 struct xfs_log_item *lip = lv->lv_item;
1270 xfs_lsn_t item_lsn;
1271
1272 if (aborted)
1273 lip->li_flags |= XFS_LI_ABORTED;
1274 item_lsn = IOP_COMMITTED(lip, commit_lsn);
1275
Dave Chinner1316d4d2011-07-04 05:27:36 +00001276 /* item_lsn of -1 means the item needs no further processing */
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001277 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1278 continue;
1279
Dave Chinnere34a3142011-01-27 12:13:35 +11001280 /*
1281 * if we are aborting the operation, no point in inserting the
1282 * object into the AIL as we are in a shutdown situation.
1283 */
1284 if (aborted) {
1285 ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
1286 IOP_UNPIN(lip, 1);
1287 continue;
1288 }
1289
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001290 if (item_lsn != commit_lsn) {
1291
1292 /*
1293 * Not a bulk update option due to unusual item_lsn.
1294 * Push into AIL immediately, rechecking the lsn once
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001295 * we have the ail lock. Then unpin the item. This does
1296 * not affect the AIL cursor the bulk insert path is
1297 * using.
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001298 */
1299 spin_lock(&ailp->xa_lock);
1300 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1301 xfs_trans_ail_update(ailp, lip, item_lsn);
1302 else
1303 spin_unlock(&ailp->xa_lock);
1304 IOP_UNPIN(lip, 0);
1305 continue;
1306 }
1307
1308 /* Item is a candidate for bulk AIL insert. */
1309 log_items[i++] = lv->lv_item;
1310 if (i >= LOG_ITEM_BATCH_SIZE) {
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001311 xfs_log_item_batch_insert(ailp, &cur, log_items,
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001312 LOG_ITEM_BATCH_SIZE, commit_lsn);
1313 i = 0;
1314 }
1315 }
1316
1317 /* make sure we insert the remainder! */
1318 if (i)
Dave Chinner1d8c95a2011-07-18 03:40:16 +00001319 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
1320
1321 spin_lock(&ailp->xa_lock);
1322 xfs_trans_ail_cursor_done(ailp, &cur);
1323 spin_unlock(&ailp->xa_lock);
Dave Chinner0e57f6a2010-12-20 12:02:19 +11001324}
1325
Dave Chinnerb1c1b5b2010-03-23 10:11:05 +11001326/*
Christoph Hellwigb1037052011-09-19 14:55:51 +00001327 * Commit the given transaction to the log.
Dave Chinner09243782010-03-08 11:28:28 +11001328 *
1329 * XFS disk error handling mechanism is not based on a typical
1330 * transaction abort mechanism. Logically after the filesystem
1331 * gets marked 'SHUTDOWN', we can't let any new transactions
1332 * be durable - ie. committed to disk - because some metadata might
1333 * be inconsistent. In such cases, this returns an error, and the
1334 * caller may assume that all locked objects joined to the transaction
1335 * have already been unlocked as if the commit had succeeded.
1336 * Do not reference the transaction structure after this call.
1337 */
Dave Chinner09243782010-03-08 11:28:28 +11001338int
Christoph Hellwigb1037052011-09-19 14:55:51 +00001339xfs_trans_commit(
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001340 struct xfs_trans *tp,
Christoph Hellwigb1037052011-09-19 14:55:51 +00001341 uint flags)
Dave Chinner09243782010-03-08 11:28:28 +11001342{
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001343 struct xfs_mount *mp = tp->t_mountp;
Dave Chinner09243782010-03-08 11:28:28 +11001344 xfs_lsn_t commit_lsn = -1;
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001345 int error = 0;
Dave Chinner09243782010-03-08 11:28:28 +11001346 int log_flags = 0;
1347 int sync = tp->t_flags & XFS_TRANS_SYNC;
Dave Chinner09243782010-03-08 11:28:28 +11001348
1349 /*
1350 * Determine whether this commit is releasing a permanent
1351 * log reservation or not.
1352 */
1353 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1354 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1355 log_flags = XFS_LOG_REL_PERM_RESERV;
1356 }
1357
1358 /*
1359 * If there is nothing to be logged by the transaction,
1360 * then unlock all of the items associated with the
1361 * transaction and free the transaction structure.
1362 * Also make sure to return any reserved blocks to
1363 * the free pool.
1364 */
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001365 if (!(tp->t_flags & XFS_TRANS_DIRTY))
1366 goto out_unreserve;
1367
1368 if (XFS_FORCED_SHUTDOWN(mp)) {
1369 error = XFS_ERROR(EIO);
1370 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001371 }
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001372
Dave Chinner09243782010-03-08 11:28:28 +11001373 ASSERT(tp->t_ticket != NULL);
1374
1375 /*
1376 * If we need to update the superblock, then do it now.
1377 */
1378 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1379 xfs_trans_apply_sb_deltas(tp);
1380 xfs_trans_apply_dquot_deltas(tp);
1381
Christoph Hellwig0244b962011-12-06 21:58:08 +00001382 error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags);
Dave Chinner09243782010-03-08 11:28:28 +11001383 if (error == ENOMEM) {
1384 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001385 error = XFS_ERROR(EIO);
1386 goto out_unreserve;
Dave Chinner09243782010-03-08 11:28:28 +11001387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Christoph Hellwig0244b962011-12-06 21:58:08 +00001389 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1390 xfs_trans_free(tp);
1391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 /*
1393 * If the transaction needs to be synchronous, then force the
1394 * log out now and wait for it.
1395 */
1396 if (sync) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001397 if (!error) {
Christoph Hellwiga14a3482010-01-19 09:56:46 +00001398 error = _xfs_log_force_lsn(mp, commit_lsn,
Christoph Hellwigb1037052011-09-19 14:55:51 +00001399 XFS_LOG_SYNC, NULL);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 XFS_STATS_INC(xs_trans_sync);
1402 } else {
1403 XFS_STATS_INC(xs_trans_async);
1404 }
1405
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001406 return error;
1407
1408out_unreserve:
1409 xfs_trans_unreserve_and_mod_sb(tp);
1410
1411 /*
1412 * It is indeed possible for the transaction to be not dirty but
1413 * the dqinfo portion to be. All that means is that we have some
1414 * (non-persistent) quota reservations that need to be unreserved.
1415 */
1416 xfs_trans_unreserve_and_mod_dquots(tp);
1417 if (tp->t_ticket) {
1418 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1419 if (commit_lsn == -1 && !error)
1420 error = XFS_ERROR(EIO);
1421 }
1422 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner71e330b2010-05-21 14:37:18 +10001423 xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
Christoph Hellwiga3ccd2c2010-03-15 12:52:49 +11001424 xfs_trans_free(tp);
1425
1426 XFS_STATS_INC(xs_trans_empty);
1427 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 * Unlock all of the transaction's items and free the transaction.
1432 * The transaction must not have modified any of its items, because
1433 * there is no way to restore them to their previous state.
1434 *
1435 * If the transaction has made a log reservation, make sure to release
1436 * it as well.
1437 */
1438void
1439xfs_trans_cancel(
1440 xfs_trans_t *tp,
1441 int flags)
1442{
1443 int log_flags;
Ryan Hankins0733af22006-01-11 15:36:44 +11001444 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
1446 /*
1447 * See if the caller is being too lazy to figure out if
1448 * the transaction really needs an abort.
1449 */
1450 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1451 flags &= ~XFS_TRANS_ABORT;
1452 /*
1453 * See if the caller is relying on us to shut down the
1454 * filesystem. This happens in paths where we detect
1455 * corruption and decide to give up.
1456 */
Nathan Scott60a204f2006-01-11 15:37:00 +11001457 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
Ryan Hankins0733af22006-01-11 15:36:44 +11001458 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001459 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Nathan Scott60a204f2006-01-11 15:37:00 +11001460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461#ifdef DEBUG
Christoph Hellwige98c4142010-06-23 18:11:15 +10001462 if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1463 struct xfs_log_item_desc *lidp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Christoph Hellwige98c4142010-06-23 18:11:15 +10001465 list_for_each_entry(lidp, &tp->t_items, lid_trans)
1466 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468#endif
1469 xfs_trans_unreserve_and_mod_sb(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +02001470 xfs_trans_unreserve_and_mod_dquots(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472 if (tp->t_ticket) {
1473 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1474 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1475 log_flags = XFS_LOG_REL_PERM_RESERV;
1476 } else {
1477 log_flags = 0;
1478 }
Ryan Hankins0733af22006-01-11 15:36:44 +11001479 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
1481
1482 /* mark this thread as no longer being in a transaction */
Nathan Scott59c1b082006-06-09 14:59:13 +10001483 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Dave Chinner71e330b2010-05-21 14:37:18 +10001485 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 xfs_trans_free(tp);
1487}
1488
Niv Sardi322ff6b2008-08-13 16:05:49 +10001489/*
1490 * Roll from one trans in the sequence of PERMANENT transactions to
1491 * the next: permanent transactions are only flushed out when
1492 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1493 * as possible to let chunks of it go to the log. So we commit the
1494 * chunk we've been working on and get a new transaction to continue.
1495 */
1496int
1497xfs_trans_roll(
1498 struct xfs_trans **tpp,
1499 struct xfs_inode *dp)
1500{
1501 struct xfs_trans *trans;
1502 unsigned int logres, count;
1503 int error;
1504
1505 /*
1506 * Ensure that the inode is always logged.
1507 */
1508 trans = *tpp;
1509 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1510
1511 /*
1512 * Copy the critical parameters from one trans to the next.
1513 */
1514 logres = trans->t_log_res;
1515 count = trans->t_log_count;
1516 *tpp = xfs_trans_dup(trans);
1517
1518 /*
1519 * Commit the current transaction.
1520 * If this commit failed, then it'd just unlock those items that
1521 * are not marked ihold. That also means that a filesystem shutdown
1522 * is in progress. The caller takes the responsibility to cancel
1523 * the duplicate transaction that gets returned.
1524 */
1525 error = xfs_trans_commit(trans, 0);
1526 if (error)
1527 return (error);
1528
1529 trans = *tpp;
1530
1531 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001532 * transaction commit worked ok so we can drop the extra ticket
1533 * reference that we gained in xfs_trans_dup()
1534 */
1535 xfs_log_ticket_put(trans->t_ticket);
1536
1537
1538 /*
Niv Sardi322ff6b2008-08-13 16:05:49 +10001539 * Reserve space in the log for th next transaction.
1540 * This also pushes items in the "AIL", the list of logged items,
1541 * out to disk if they are taking up space at the tail of the log
1542 * that we want to use. This requires that either nothing be locked
1543 * across this call, or that anything that is locked be logged in
1544 * the prior and the next transactions.
1545 */
1546 error = xfs_trans_reserve(trans, 0, logres, 0,
1547 XFS_TRANS_PERM_LOG_RES, count);
1548 /*
1549 * Ensure that the inode is in the new transaction and locked.
1550 */
1551 if (error)
1552 return error;
1553
Christoph Hellwigddc34152011-09-19 15:00:54 +00001554 xfs_trans_ijoin(trans, dp, 0);
Niv Sardi322ff6b2008-08-13 16:05:49 +10001555 return 0;
1556}