blob: b572f7e840e0b82a372225581d87c85c104dfec6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018
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_dir2.h"
29#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_dir2_sf.h"
36#include "xfs_attr_sf.h"
37#include "xfs_dinode.h"
38#include "xfs_inode.h"
39#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_itable.h"
41#include "xfs_btree.h"
42#include "xfs_ialloc.h"
43#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_bmap.h"
Christoph Hellwigef14f0c2009-06-10 17:07:47 +020045#include "xfs_acl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include "xfs_attr.h"
47#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_quota.h"
50#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110051#include "xfs_rtalloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100054#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100055#include "xfs_vnodeops.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Christoph Hellwig993386c2007-08-28 16:12:30 +100057int
Linus Torvalds1da177e2005-04-16 15:20:36 -070058xfs_setattr(
Christoph Hellwig0f285c82008-07-18 17:13:28 +100059 struct xfs_inode *ip,
60 struct iattr *iattr,
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +110061 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062{
Christoph Hellwig993386c2007-08-28 16:12:30 +100063 xfs_mount_t *mp = ip->i_mount;
David Chinnere4f75292008-08-13 16:00:45 +100064 struct inode *inode = VFS_I(ip);
Christoph Hellwig0f285c82008-07-18 17:13:28 +100065 int mask = iattr->ia_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 int code;
68 uint lock_flags;
69 uint commit_flags=0;
70 uid_t uid=0, iuid=0;
71 gid_t gid=0, igid=0;
72 int timeflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
Dean Roehrich5fcbab32005-05-05 13:27:19 -070074 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Lachlan McIlroycf441ee2008-02-07 16:42:19 +110076 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Christoph Hellwigbd186aa2007-08-30 17:21:12 +100078 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 return XFS_ERROR(EROFS);
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 if (XFS_FORCED_SHUTDOWN(mp))
82 return XFS_ERROR(EIO);
83
Christoph Hellwigc4cd7472008-12-09 04:47:34 -050084 code = -inode_change_ok(inode, iattr);
85 if (code)
86 return code;
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 olddquot1 = olddquot2 = NULL;
89 udqp = gdqp = NULL;
90
91 /*
92 * If disk quotas is on, we make sure that the dquots do exist on disk,
93 * before we start any other transactions. Trying to do this later
94 * is messy. We don't care to take a readlock to look at the ids
95 * in inode here, because we can't hold it across the trans_reserve.
96 * If the IDs do change before we take the ilock, we're covered
97 * because the i_*dquot fields will get updated anyway.
98 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +100099 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 uint qflags = 0;
101
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000102 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
103 uid = iattr->ia_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 qflags |= XFS_QMOPT_UQUOTA;
105 } else {
106 uid = ip->i_d.di_uid;
107 }
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000108 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
109 gid = iattr->ia_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 qflags |= XFS_QMOPT_GQUOTA;
111 } else {
112 gid = ip->i_d.di_gid;
113 }
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 /*
116 * We take a reference when we initialize udqp and gdqp,
117 * so it is important that we never blindly double trip on
118 * the same variable. See xfs_create() for an example.
119 */
120 ASSERT(udqp == NULL);
121 ASSERT(gdqp == NULL);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200122 code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000123 qflags, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100125 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 }
127
128 /*
129 * For the other attributes, we acquire the inode lock and
130 * first do an error checking pass.
131 */
132 tp = NULL;
133 lock_flags = XFS_ILOCK_EXCL;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000134 if (flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700135 need_iolock = 0;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000136 if (!(mask & ATTR_SIZE)) {
137 if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 (mp->m_flags & XFS_MOUNT_WSYNC)) {
139 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
140 commit_flags = 0;
141 if ((code = xfs_trans_reserve(tp, 0,
142 XFS_ICHANGE_LOG_RES(mp), 0,
143 0, 0))) {
144 lock_flags = 0;
145 goto error_return;
146 }
147 }
148 } else {
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000149 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000150 !(flags & XFS_ATTR_DMI)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100152 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000153 iattr->ia_size, 0, dmflags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (code) {
155 lock_flags = 0;
156 goto error_return;
157 }
158 }
159 if (need_iolock)
160 lock_flags |= XFS_IOLOCK_EXCL;
161 }
162
163 xfs_ilock(ip, lock_flags);
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 /*
166 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000168 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 /*
170 * These IDs could have changed since we last looked at them.
171 * But, we're assured that if the ownership did change
172 * while we didn't have the inode locked, inode's dquot(s)
173 * would have changed also.
174 */
175 iuid = ip->i_d.di_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 igid = ip->i_d.di_gid;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000177 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
178 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 /*
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000181 * Do a quota reservation only if uid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 * going to change.
183 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200184 if (XFS_IS_QUOTA_RUNNING(mp) &&
185 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
186 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 ASSERT(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200188 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 capable(CAP_FOWNER) ?
190 XFS_QMOPT_FORCE_RES : 0);
191 if (code) /* out of quota */
192 goto error_return;
193 }
194 }
195
196 /*
197 * Truncate file. Must have write permission and not be a directory.
198 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000199 if (mask & ATTR_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 /* Short circuit the truncate case for zero length files */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000201 if (iattr->ia_size == 0 &&
202 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
204 lock_flags &= ~XFS_ILOCK_EXCL;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000205 if (mask & ATTR_CTIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
207 code = 0;
208 goto error_return;
209 }
210
Christoph Hellwig42173f62008-04-22 17:33:25 +1000211 if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 code = XFS_ERROR(EISDIR);
213 goto error_return;
Christoph Hellwig42173f62008-04-22 17:33:25 +1000214 } else if (!S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 code = XFS_ERROR(EINVAL);
216 goto error_return;
217 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 /*
220 * Make sure that the dquots are attached to the inode.
221 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200222 code = xfs_qm_dqattach_locked(ip, 0);
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500223 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500226 /*
227 * Now we can make the changes. Before we join the inode
228 * to the transaction, if ATTR_SIZE is set then take care of
229 * the part of the truncation that must be done without the
230 * inode lock. This needs to be done before joining the inode
231 * to the transaction, because the inode cannot be unlocked
232 * once it is a part of the transaction.
233 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000234 if (iattr->ia_size > ip->i_size) {
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000235 /*
236 * Do the first part of growing a file: zero any data
237 * in the last block that is beyond the old EOF. We
238 * need to do this before the inode is joined to the
239 * transaction to modify the i_size.
240 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000241 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnerc32676e2007-07-19 16:28:58 +1000244
245 /*
246 * We are going to log the inode size change in this
247 * transaction so any previous writes that are beyond the on
248 * disk EOF and the new EOF that have not been written out need
249 * to be written here. If we do not write the data out, we
250 * expose ourselves to the null files problem.
251 *
252 * Only flush from the on disk size to the smaller of the in
253 * memory file size or the new size as that's the range we
254 * really care about here and prevents waiting for other data
255 * not within the range we care about here.
256 */
257 if (!code &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000258 ip->i_size != ip->i_d.di_size &&
259 iattr->ia_size > ip->i_d.di_size) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000260 code = xfs_flush_pages(ip,
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000261 ip->i_d.di_size, iattr->ia_size,
David Chinnerc32676e2007-07-19 16:28:58 +1000262 XFS_B_ASYNC, FI_NONE);
263 }
264
265 /* wait for all I/O to complete */
Christoph Hellwig25e41b32008-12-03 12:20:39 +0100266 xfs_ioend_wait(ip);
David Chinnerc32676e2007-07-19 16:28:58 +1000267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 if (!code)
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000269 code = xfs_itruncate_data(ip, iattr->ia_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 if (code) {
271 ASSERT(tp == NULL);
272 lock_flags &= ~XFS_ILOCK_EXCL;
273 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
274 goto error_return;
275 }
276 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
277 if ((code = xfs_trans_reserve(tp, 0,
278 XFS_ITRUNCATE_LOG_RES(mp), 0,
279 XFS_TRANS_PERM_LOG_RES,
280 XFS_ITRUNCATE_LOG_COUNT))) {
281 xfs_trans_cancel(tp, 0);
282 if (need_iolock)
283 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
284 return code;
285 }
286 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
287 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 xfs_trans_ijoin(tp, ip, lock_flags);
290 xfs_trans_ihold(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
David Chinner44d814c2008-03-06 13:45:29 +1100292 /*
293 * Only change the c/mtime if we are changing the size
294 * or we are explicitly asked to change it. This handles
295 * the semantic difference between truncate() and ftruncate()
296 * as implemented in the VFS.
297 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000298 if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME))
David Chinner44d814c2008-03-06 13:45:29 +1100299 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
300
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000301 if (iattr->ia_size > ip->i_size) {
302 ip->i_d.di_size = iattr->ia_size;
303 ip->i_size = iattr->ia_size;
304 if (!(flags & XFS_ATTR_DMI))
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000305 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
306 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000307 } else if (iattr->ia_size <= ip->i_size ||
308 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 /*
310 * signal a sync transaction unless
311 * we're truncating an already unlinked
312 * file on a wsync filesystem
313 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000314 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 XFS_DATA_FORK,
316 ((ip->i_d.di_nlink != 0 ||
317 !(mp->m_flags & XFS_MOUNT_WSYNC))
318 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000319 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000321 /*
322 * Truncated "down", so we're removing references
323 * to old data here - if we now delay flushing for
324 * a long time, we expose ourselves unduly to the
325 * notorious NULL files problem. So, we mark this
326 * vnode and flush it when the file is closed, and
327 * do not wait the usual (long) time for writeout.
328 */
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000329 xfs_iflags_set(ip, XFS_ITRUNCATED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500331 } else if (tp) {
332 xfs_trans_ijoin(tp, ip, lock_flags);
333 xfs_trans_ihold(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
335
336 /*
337 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000339 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 /*
341 * CAP_FSETID overrides the following restrictions:
342 *
343 * The set-user-ID and set-group-ID bits of a file will be
344 * cleared upon successful return from chown()
345 */
346 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
347 !capable(CAP_FSETID)) {
348 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
349 }
350
351 /*
352 * Change the ownerships and register quota modifications
353 * in the transaction.
354 */
355 if (iuid != uid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200356 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000357 ASSERT(mask & ATTR_UID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 ASSERT(udqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200359 olddquot1 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 &ip->i_udquot, udqp);
361 }
362 ip->i_d.di_uid = uid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000363 inode->i_uid = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
365 if (igid != gid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200366 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000367 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000368 ASSERT(mask & ATTR_GID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 ASSERT(gdqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200370 olddquot2 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 &ip->i_gdquot, gdqp);
372 }
373 ip->i_d.di_gid = gid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000374 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
377 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
378 timeflags |= XFS_ICHGTIME_CHG;
379 }
380
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500381 /*
382 * Change file access modes.
383 */
384 if (mask & ATTR_MODE) {
385 umode_t mode = iattr->ia_mode;
386
387 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
388 mode &= ~S_ISGID;
389
390 ip->i_d.di_mode &= S_IFMT;
391 ip->i_d.di_mode |= mode & ~S_IFMT;
392
393 inode->i_mode &= S_IFMT;
394 inode->i_mode |= mode & ~S_IFMT;
395
396 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
397 timeflags |= XFS_ICHGTIME_CHG;
398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400 /*
401 * Change file access or modified times.
402 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000403 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
404 if (mask & ATTR_ATIME) {
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000405 inode->i_atime = iattr->ia_atime;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000406 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
407 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 ip->i_update_core = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000410 if (mask & ATTR_MTIME) {
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000411 inode->i_mtime = iattr->ia_mtime;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000412 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
413 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 timeflags &= ~XFS_ICHGTIME_MOD;
415 timeflags |= XFS_ICHGTIME_CHG;
416 }
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000417 if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
419 }
420
421 /*
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000422 * Change file inode change time only if ATTR_CTIME set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 * AND we have been called by a DMI function.
424 */
425
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000426 if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) {
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000427 inode->i_ctime = iattr->ia_ctime;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000428 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
429 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ip->i_update_core = 1;
431 timeflags &= ~XFS_ICHGTIME_CHG;
432 }
433
434 /*
435 * Send out timestamp changes that need to be set to the
436 * current time. Not done when called by a DMI function.
437 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000438 if (timeflags && !(flags & XFS_ATTR_DMI))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 xfs_ichgtime(ip, timeflags);
440
441 XFS_STATS_INC(xs_ig_attrchg);
442
443 /*
444 * If this is a synchronous mount, make sure that the
445 * transaction goes to disk before returning to the user.
446 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000447 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 * One for the truncate and one for the timestamps since we
449 * don't want to change the timestamps unless we're sure the
450 * truncate worked. Truncates are less than 1% of the laddis
451 * mix so this probably isn't worth the trouble to optimize.
452 */
453 code = 0;
454 if (tp) {
455 if (mp->m_flags & XFS_MOUNT_WSYNC)
456 xfs_trans_set_sync(tp);
457
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000458 code = xfs_trans_commit(tp, commit_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 xfs_iunlock(ip, lock_flags);
462
463 /*
464 * Release any dquot(s) the inode had kept before chown.
465 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200466 xfs_qm_dqrele(olddquot1);
467 xfs_qm_dqrele(olddquot2);
468 xfs_qm_dqrele(udqp);
469 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200471 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 return code;
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200473
474 /*
475 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
476 * update. We could avoid this with linked transactions
477 * and passing down the transaction pointer all the way
478 * to attr_set. No previous user of the generic
479 * Posix ACL code seems to care about this issue either.
480 */
481 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
482 code = -xfs_acl_chmod(inode);
483 if (code)
484 return XFS_ERROR(code);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
486
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000487 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000488 !(flags & XFS_ATTR_DMI)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100489 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 NULL, DM_RIGHT_NULL, NULL, NULL,
491 0, 0, AT_DELAY_FLAG(flags));
492 }
493 return 0;
494
495 abort_return:
496 commit_flags |= XFS_TRANS_ABORT;
497 /* FALLTHROUGH */
498 error_return:
Christoph Hellwig7d095252009-06-08 15:33:32 +0200499 xfs_qm_dqrele(udqp);
500 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (tp) {
502 xfs_trans_cancel(tp, commit_flags);
503 }
504 if (lock_flags != 0) {
505 xfs_iunlock(ip, lock_flags);
506 }
507 return code;
508}
509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000511 * The maximum pathlen is 1024 bytes. Since the minimum file system
512 * blocksize is 512 bytes, we can get a max of 2 extents back from
513 * bmapi.
514 */
515#define SYMLINK_MAPS 2
516
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000517STATIC int
518xfs_readlink_bmap(
519 xfs_inode_t *ip,
520 char *link)
521{
522 xfs_mount_t *mp = ip->i_mount;
523 int pathlen = ip->i_d.di_size;
524 int nmaps = SYMLINK_MAPS;
525 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
526 xfs_daddr_t d;
527 int byte_cnt;
528 int n;
529 xfs_buf_t *bp;
530 int error = 0;
531
532 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
533 mval, &nmaps, NULL, NULL);
534 if (error)
535 goto out;
536
537 for (n = 0; n < nmaps; n++) {
538 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
539 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
540
Christoph Hellwig7b02ecb2009-07-18 18:15:00 -0400541 bp = xfs_buf_read_flags(mp->m_ddev_targp, d, BTOBB(byte_cnt),
542 XBF_LOCK | XBF_MAPPED |
543 XBF_DONT_BLOCK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000544 error = XFS_BUF_GETERROR(bp);
545 if (error) {
546 xfs_ioerror_alert("xfs_readlink",
547 ip->i_mount, bp, XFS_BUF_ADDR(bp));
548 xfs_buf_relse(bp);
549 goto out;
550 }
551 if (pathlen < byte_cnt)
552 byte_cnt = pathlen;
553 pathlen -= byte_cnt;
554
555 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
556 xfs_buf_relse(bp);
557 }
558
559 link[ip->i_d.di_size] = '\0';
560 error = 0;
561
562 out:
563 return error;
564}
565
Christoph Hellwig993386c2007-08-28 16:12:30 +1000566int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000568 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000569 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000571 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 int pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100575 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 if (XFS_FORCED_SHUTDOWN(mp))
578 return XFS_ERROR(EIO);
579
580 xfs_ilock(ip, XFS_ILOCK_SHARED);
581
582 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000583 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000585 pathlen = ip->i_d.di_size;
586 if (!pathlen)
587 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000590 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
591 link[pathlen] = '\0';
592 } else {
593 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000596 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 return error;
599}
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601/*
602 * xfs_fsync
603 *
David Chinner978b7232008-05-19 16:29:46 +1000604 * This is called to sync the inode and its data out to disk. We need to hold
605 * the I/O lock while flushing the data, and the inode lock while flushing the
606 * inode. The inode lock CANNOT be held while flushing the data, so acquire
607 * after we're done with that.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
Christoph Hellwig993386c2007-08-28 16:12:30 +1000609int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610xfs_fsync(
David Chinner978b7232008-05-19 16:29:46 +1000611 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 xfs_trans_t *tp;
Christoph Hellwig13e6d5c2009-08-31 21:00:31 -0300614 int error = 0;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100615 int log_flushed = 0, changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100617 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
620 return XFS_ERROR(EIO);
621
622 /*
David Chinner978b7232008-05-19 16:29:46 +1000623 * We always need to make sure that the required inode state is safe on
Christoph Hellwig13e6d5c2009-08-31 21:00:31 -0300624 * disk. The inode might be clean but we still might need to force the
David Chinner978b7232008-05-19 16:29:46 +1000625 * log because of committed transactions that haven't hit the disk yet.
626 * Likewise, there could be unflushed non-transactional changes to the
627 * inode core that have to go to disk and this requires us to issue
628 * a synchronous transaction to capture these changes correctly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 *
David Chinner978b7232008-05-19 16:29:46 +1000630 * This code relies on the assumption that if the update_* fields
631 * of the inode are clear and the inode is unpinned then it is clean
632 * and no action is required.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 */
634 xfs_ilock(ip, XFS_ILOCK_SHARED);
635
Christoph Hellwig13e6d5c2009-08-31 21:00:31 -0300636 if (!ip->i_update_core) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 /*
David Chinner978b7232008-05-19 16:29:46 +1000638 * Timestamps/size haven't changed since last inode flush or
639 * inode transaction commit. That means either nothing got
640 * written or a transaction committed which caught the updates.
641 * If the latter happened and the transaction hasn't hit the
642 * disk yet, the inode will be still be pinned. If it is,
643 * force the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 */
645
646 xfs_iunlock(ip, XFS_ILOCK_SHARED);
647
648 if (xfs_ipincount(ip)) {
David Chinner978b7232008-05-19 16:29:46 +1000649 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
650 XFS_LOG_FORCE | XFS_LOG_SYNC,
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100651 &log_flushed);
652 } else {
653 /*
David Chinner978b7232008-05-19 16:29:46 +1000654 * If the inode is not pinned and nothing has changed
655 * we don't need to flush the cache.
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100656 */
657 changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 } else {
660 /*
David Chinner978b7232008-05-19 16:29:46 +1000661 * Kick off a transaction to log the inode core to get the
662 * updates. The sync transaction will also force the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 */
664 xfs_iunlock(ip, XFS_ILOCK_SHARED);
665 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
David Chinner978b7232008-05-19 16:29:46 +1000666 error = xfs_trans_reserve(tp, 0,
667 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
668 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 xfs_trans_cancel(tp, 0);
670 return error;
671 }
672 xfs_ilock(ip, XFS_ILOCK_EXCL);
673
674 /*
David Chinner978b7232008-05-19 16:29:46 +1000675 * Note - it's possible that we might have pushed ourselves out
676 * of the way during trans_reserve which would flush the inode.
677 * But there's no guarantee that the inode buffer has actually
678 * gone out yet (it's delwri). Plus the buffer could be pinned
679 * anyway if it's part of an inode in another recent
680 * transaction. So we play it safe and fire off the
681 * transaction anyway.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 */
683 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
684 xfs_trans_ihold(tp, ip);
685 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
David Chinner978b7232008-05-19 16:29:46 +1000686 xfs_trans_set_sync(tp);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000687 error = _xfs_trans_commit(tp, 0, &log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
689 xfs_iunlock(ip, XFS_ILOCK_EXCL);
690 }
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100691
692 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
693 /*
694 * If the log write didn't issue an ordered tag we need
695 * to flush the disk cache for the data device now.
696 */
697 if (!log_flushed)
698 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
699
700 /*
701 * If this inode is on the RT dev we need to flush that
Nathan Scottc41564b2006-03-29 08:55:14 +1000702 * cache as well.
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100703 */
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100704 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100705 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
706 }
707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return error;
709}
710
711/*
David Chinner92dfe8d2007-05-24 15:22:19 +1000712 * This is called by xfs_inactive to free any blocks beyond eof
713 * when the link count isn't zero and by xfs_dm_punch_hole() when
714 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 */
Eric Sandeend96f8f82009-07-02 00:09:33 -0500716STATIC int
David Chinner92dfe8d2007-05-24 15:22:19 +1000717xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +1000719 xfs_inode_t *ip,
720 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 xfs_trans_t *tp;
723 int error;
724 xfs_fileoff_t end_fsb;
725 xfs_fileoff_t last_fsb;
726 xfs_filblks_t map_len;
727 int nimaps;
728 xfs_bmbt_irec_t imap;
David Chinner92dfe8d2007-05-24 15:22:19 +1000729 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 /*
732 * Figure out if there are any blocks beyond the end
733 * of the file. If not, then there is nothing to do.
734 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000735 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
737 map_len = last_fsb - end_fsb;
738 if (map_len <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +0100739 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741 nimaps = 1;
742 xfs_ilock(ip, XFS_ILOCK_SHARED);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000743 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +1000744 NULL, 0, &imap, &nimaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 xfs_iunlock(ip, XFS_ILOCK_SHARED);
746
747 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +1100748 (imap.br_startblock != HOLESTARTBLOCK ||
749 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 /*
751 * Attach the dquots to the inode up front.
752 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200753 error = xfs_qm_dqattach(ip, 0);
754 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100755 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 /*
758 * There are blocks after the end of file.
759 * Free them up now by truncating the file to
760 * its current size.
761 */
762 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
763
764 /*
765 * Do the xfs_itruncate_start() call before
766 * reserving any log space because
767 * itruncate_start will call into the buffer
768 * cache and we can't
769 * do that within a transaction.
770 */
David Chinner92dfe8d2007-05-24 15:22:19 +1000771 if (use_iolock)
772 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000773 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000774 ip->i_size);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000775 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +1000776 xfs_trans_cancel(tp, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +1000777 if (use_iolock)
778 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000779 return error;
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
782 error = xfs_trans_reserve(tp, 0,
783 XFS_ITRUNCATE_LOG_RES(mp),
784 0, XFS_TRANS_PERM_LOG_RES,
785 XFS_ITRUNCATE_LOG_COUNT);
786 if (error) {
787 ASSERT(XFS_FORCED_SHUTDOWN(mp));
788 xfs_trans_cancel(tp, 0);
789 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100790 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792
793 xfs_ilock(ip, XFS_ILOCK_EXCL);
794 xfs_trans_ijoin(tp, ip,
795 XFS_IOLOCK_EXCL |
796 XFS_ILOCK_EXCL);
797 xfs_trans_ihold(tp, ip);
798
799 error = xfs_itruncate_finish(&tp, ip,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000800 ip->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 XFS_DATA_FORK,
802 0);
803 /*
804 * If we get an error at this point we
805 * simply don't bother truncating the file.
806 */
807 if (error) {
808 xfs_trans_cancel(tp,
809 (XFS_TRANS_RELEASE_LOG_RES |
810 XFS_TRANS_ABORT));
811 } else {
812 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000813 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 }
David Chinner92dfe8d2007-05-24 15:22:19 +1000815 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
816 : XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100818 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
821/*
822 * Free a symlink that has blocks associated with it.
823 */
824STATIC int
825xfs_inactive_symlink_rmt(
826 xfs_inode_t *ip,
827 xfs_trans_t **tpp)
828{
829 xfs_buf_t *bp;
830 int committed;
831 int done;
832 int error;
833 xfs_fsblock_t first_block;
834 xfs_bmap_free_t free_list;
835 int i;
836 xfs_mount_t *mp;
837 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
838 int nmaps;
839 xfs_trans_t *ntp;
840 int size;
841 xfs_trans_t *tp;
842
843 tp = *tpp;
844 mp = ip->i_mount;
845 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
846 /*
847 * We're freeing a symlink that has some
848 * blocks allocated to it. Free the
849 * blocks here. We know that we've got
850 * either 1 or 2 extents and that we can
851 * free them all in one bunmapi call.
852 */
853 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
854 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
855 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
856 ASSERT(XFS_FORCED_SHUTDOWN(mp));
857 xfs_trans_cancel(tp, 0);
858 *tpp = NULL;
859 return error;
860 }
861 /*
862 * Lock the inode, fix the size, and join it to the transaction.
863 * Hold it so in the normal path, we still have it locked for
864 * the second transaction. In the error paths we need it
865 * held so the cancel won't rele it, see below.
866 */
867 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
868 size = (int)ip->i_d.di_size;
869 ip->i_d.di_size = 0;
870 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
871 xfs_trans_ihold(tp, ip);
872 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
873 /*
874 * Find the block(s) so we can inval and unmap them.
875 */
876 done = 0;
Eric Sandeen9d87c312009-01-14 23:22:07 -0600877 xfs_bmap_init(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -0800878 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
880 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +1000881 &free_list, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 goto error0;
883 /*
884 * Invalidate the block(s).
885 */
886 for (i = 0; i < nmaps; i++) {
887 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
888 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
889 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
890 xfs_trans_binval(tp, bp);
891 }
892 /*
893 * Unmap the dead block(s) to the free_list.
894 */
895 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +1000896 &first_block, &free_list, NULL, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 goto error1;
898 ASSERT(done);
899 /*
900 * Commit the first transaction. This logs the EFI and the inode.
901 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100902 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 goto error1;
904 /*
905 * The transaction must have been committed, since there were
906 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
907 * The new tp has the extent freeing and EFDs.
908 */
909 ASSERT(committed);
910 /*
911 * The first xact was committed, so add the inode to the new one.
912 * Mark it dirty so it will be logged and moved forward in the log as
913 * part of every commit.
914 */
915 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
916 xfs_trans_ihold(tp, ip);
917 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
918 /*
919 * Get a new, empty transaction to return to our caller.
920 */
921 ntp = xfs_trans_dup(tp);
922 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000923 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 * If we get an error on the commit here or on the reserve below,
925 * we need to unlock the inode since the new transaction doesn't
926 * have the inode attached.
927 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000928 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 tp = ntp;
930 if (error) {
931 ASSERT(XFS_FORCED_SHUTDOWN(mp));
932 goto error0;
933 }
934 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100935 * transaction commit worked ok so we can drop the extra ticket
936 * reference that we gained in xfs_trans_dup()
937 */
938 xfs_log_ticket_put(tp->t_ticket);
939
940 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 * Remove the memory for extent descriptions (just bookkeeping).
942 */
943 if (ip->i_df.if_bytes)
944 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
945 ASSERT(ip->i_df.if_bytes == 0);
946 /*
947 * Put an itruncate log reservation in the new transaction
948 * for our caller.
949 */
950 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
951 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
952 ASSERT(XFS_FORCED_SHUTDOWN(mp));
953 goto error0;
954 }
955 /*
956 * Return with the inode locked but not joined to the transaction.
957 */
958 *tpp = tp;
959 return 0;
960
961 error1:
962 xfs_bmap_cancel(&free_list);
963 error0:
964 /*
965 * Have to come here with the inode locked and either
966 * (held and in the transaction) or (not in the transaction).
967 * If the inode isn't held then cancel would iput it, but
968 * that's wrong since this is inactive and the vnode ref
969 * count is 0 already.
970 * Cancel won't do anything to the inode if held, but it still
971 * needs to be locked until the cancel is done, if it was
972 * joined to the transaction.
973 */
974 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
975 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
976 *tpp = NULL;
977 return error;
978
979}
980
981STATIC int
982xfs_inactive_symlink_local(
983 xfs_inode_t *ip,
984 xfs_trans_t **tpp)
985{
986 int error;
987
988 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
989 /*
990 * We're freeing a symlink which fit into
991 * the inode. Just free the memory used
992 * to hold the old symlink.
993 */
994 error = xfs_trans_reserve(*tpp, 0,
995 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
996 0, XFS_TRANS_PERM_LOG_RES,
997 XFS_ITRUNCATE_LOG_COUNT);
998
999 if (error) {
1000 xfs_trans_cancel(*tpp, 0);
1001 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001002 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1005
1006 /*
1007 * Zero length symlinks _can_ exist.
1008 */
1009 if (ip->i_df.if_bytes > 0) {
1010 xfs_idata_realloc(ip,
1011 -(ip->i_df.if_bytes),
1012 XFS_DATA_FORK);
1013 ASSERT(ip->i_df.if_bytes == 0);
1014 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001015 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018STATIC int
1019xfs_inactive_attrs(
1020 xfs_inode_t *ip,
1021 xfs_trans_t **tpp)
1022{
1023 xfs_trans_t *tp;
1024 int error;
1025 xfs_mount_t *mp;
1026
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001027 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 tp = *tpp;
1029 mp = ip->i_mount;
1030 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001031 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +10001033 if (error)
1034 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +10001037 if (error)
1038 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1041 error = xfs_trans_reserve(tp, 0,
1042 XFS_IFREE_LOG_RES(mp),
1043 0, XFS_TRANS_PERM_LOG_RES,
1044 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +10001045 if (error)
1046 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
1048 xfs_ilock(ip, XFS_ILOCK_EXCL);
1049 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1050 xfs_trans_ihold(tp, ip);
1051 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1052
1053 ASSERT(ip->i_d.di_anextents == 0);
1054
1055 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +01001056 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +10001057
1058error_cancel:
1059 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1060 xfs_trans_cancel(tp, 0);
1061error_unlock:
1062 *tpp = NULL;
1063 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1064 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065}
1066
Christoph Hellwig993386c2007-08-28 16:12:30 +10001067int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001069 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001071 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 int error;
1073
Christoph Hellwig42173f62008-04-22 17:33:25 +10001074 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001078 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return 0;
1080
David Chinner2a82b8b2007-07-11 11:09:12 +10001081 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001082 int truncated;
1083
David Chinner2a82b8b2007-07-11 11:09:12 +10001084 /*
1085 * If we are using filestreams, and we have an unlinked
1086 * file that we are processing the last close on, then nothing
1087 * will be able to reopen and write to this file. Purge this
1088 * inode from the filestreams cache so that it doesn't delay
1089 * teardown of the inode.
1090 */
1091 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1092 xfs_filestream_deassociate(ip);
1093
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001094 /*
1095 * If we previously truncated this file and removed old data
1096 * in the process, we want to initiate "early" writeout on
1097 * the last close. This is an attempt to combat the notorious
1098 * NULL files problem which is particularly noticable from a
1099 * truncate down, buffered (re-)write (delalloc), followed by
1100 * a crash. What we are effectively doing here is
1101 * significantly reducing the time window where we'd otherwise
1102 * be exposed to that problem.
1103 */
Christoph Hellwig09262b42007-08-29 11:44:50 +10001104 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001105 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001106 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001107 }
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 if (ip->i_d.di_nlink != 0) {
1110 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001111 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001112 ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001114 (!(ip->i_d.di_flags &
1115 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001116 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1117 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001118 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120 }
1121
1122 return 0;
1123}
1124
1125/*
1126 * xfs_inactive
1127 *
1128 * This is called when the vnode reference count for the vnode
1129 * goes to zero. If the file has been unlinked, then it must
1130 * now be truncated. Also, we clear all of the read-ahead state
1131 * kept for the inode here since the file is now closed.
1132 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001133int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001135 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Nathan Scott67fcaa72006-06-09 17:00:52 +10001137 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 xfs_fsblock_t first_block;
1139 int committed;
1140 xfs_trans_t *tp;
1141 xfs_mount_t *mp;
1142 int error;
1143 int truncate;
1144
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001145 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 /*
1148 * If the inode is already free, then there can be nothing
1149 * to clean up here.
1150 */
Christoph Hellwigcb4c8cc2009-03-16 08:25:25 +01001151 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 ASSERT(ip->i_df.if_real_bytes == 0);
1153 ASSERT(ip->i_df.if_broot_bytes == 0);
1154 return VN_INACTIVE_CACHE;
1155 }
1156
1157 /*
1158 * Only do a truncate if it's a regular file with
1159 * some actual space in it. It's OK to look at the
1160 * inode's fields without the lock because we're the
1161 * only one with a reference to the inode.
1162 */
1163 truncate = ((ip->i_d.di_nlink == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001164 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1165 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1167
1168 mp = ip->i_mount;
1169
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001170 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1171 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 error = 0;
1174
1175 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001176 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 goto out;
1178
1179 if (ip->i_d.di_nlink != 0) {
1180 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001181 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001182 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001183 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1184 (!(ip->i_d.di_flags &
1185 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1186 (ip->i_delayed_blks != 0)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001187 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1188 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001189 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
1191 goto out;
1192 }
1193
1194 ASSERT(ip->i_d.di_nlink == 0);
1195
Christoph Hellwig7d095252009-06-08 15:33:32 +02001196 error = xfs_qm_dqattach(ip, 0);
1197 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001198 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
1200 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1201 if (truncate) {
1202 /*
1203 * Do the xfs_itruncate_start() call before
1204 * reserving any log space because itruncate_start
1205 * will call into the buffer cache and we can't
1206 * do that within a transaction.
1207 */
1208 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1209
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001210 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1211 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001212 xfs_trans_cancel(tp, 0);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001213 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1214 return VN_INACTIVE_CACHE;
1215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
1217 error = xfs_trans_reserve(tp, 0,
1218 XFS_ITRUNCATE_LOG_RES(mp),
1219 0, XFS_TRANS_PERM_LOG_RES,
1220 XFS_ITRUNCATE_LOG_COUNT);
1221 if (error) {
1222 /* Don't call itruncate_cleanup */
1223 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1224 xfs_trans_cancel(tp, 0);
1225 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001226 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 }
1228
1229 xfs_ilock(ip, XFS_ILOCK_EXCL);
1230 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1231 xfs_trans_ihold(tp, ip);
1232
1233 /*
1234 * normally, we have to run xfs_itruncate_finish sync.
1235 * But if filesystem is wsync and we're in the inactive
1236 * path, then we know that nlink == 0, and that the
1237 * xaction that made nlink == 0 is permanently committed
1238 * since xfs_remove runs as a synchronous transaction.
1239 */
1240 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1241 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1242
1243 if (error) {
1244 xfs_trans_cancel(tp,
1245 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1246 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001247 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
1249 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1250
1251 /*
1252 * If we get an error while cleaning up a
1253 * symlink we bail out.
1254 */
1255 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1256 xfs_inactive_symlink_rmt(ip, &tp) :
1257 xfs_inactive_symlink_local(ip, &tp);
1258
1259 if (error) {
1260 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001261 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
1263
1264 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1265 xfs_trans_ihold(tp, ip);
1266 } else {
1267 error = xfs_trans_reserve(tp, 0,
1268 XFS_IFREE_LOG_RES(mp),
1269 0, XFS_TRANS_PERM_LOG_RES,
1270 XFS_INACTIVE_LOG_COUNT);
1271 if (error) {
1272 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1273 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001274 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
1276
1277 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1278 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1279 xfs_trans_ihold(tp, ip);
1280 }
1281
1282 /*
1283 * If there are attributes associated with the file
1284 * then blow them away now. The code calls a routine
1285 * that recursively deconstructs the attribute fork.
1286 * We need to just commit the current transaction
1287 * because we can't use it for xfs_attr_inactive().
1288 */
1289 if (ip->i_d.di_anextents > 0) {
1290 error = xfs_inactive_attrs(ip, &tp);
1291 /*
1292 * If we got an error, the transaction is already
1293 * cancelled, and the inode is unlocked. Just get out.
1294 */
1295 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001296 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 } else if (ip->i_afp) {
1298 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1299 }
1300
1301 /*
1302 * Free the inode.
1303 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001304 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 error = xfs_ifree(tp, ip, &free_list);
1306 if (error) {
1307 /*
1308 * If we fail to free the inode, shut down. The cancel
1309 * might do that, we need to make sure. Otherwise the
1310 * inode might be lost for a long time or forever.
1311 */
1312 if (!XFS_FORCED_SHUTDOWN(mp)) {
1313 cmn_err(CE_NOTE,
1314 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1315 error, mp->m_fsname);
Nathan Scott7d04a332006-06-09 14:58:38 +10001316 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 }
1318 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1319 } else {
1320 /*
1321 * Credit the quota account(s). The inode is gone.
1322 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001323 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 /*
David Chinner78e9da72008-04-10 12:24:17 +10001326 * Just ignore errors at this point. There is nothing we can
1327 * do except to try to keep going. Make sure it's not a silent
1328 * error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 */
David Chinner78e9da72008-04-10 12:24:17 +10001330 error = xfs_bmap_finish(&tp, &free_list, &committed);
1331 if (error)
1332 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1333 "xfs_bmap_finish() returned error %d", error);
1334 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1335 if (error)
1336 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1337 "xfs_trans_commit() returned error %d", error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
Christoph Hellwig7d095252009-06-08 15:33:32 +02001339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 /*
1341 * Release the dquots held by inode, if any.
1342 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001343 xfs_qm_dqdetach(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1345
1346 out:
1347 return VN_INACTIVE_CACHE;
1348}
1349
Barry Naujok384f3ce2008-05-21 16:58:22 +10001350/*
1351 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1352 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1353 * ci_name->name will point to a the actual name (caller must free) or
1354 * will be set to NULL if an exact match is found.
1355 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001356int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001358 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001359 struct xfs_name *name,
Barry Naujok384f3ce2008-05-21 16:58:22 +10001360 xfs_inode_t **ipp,
1361 struct xfs_name *ci_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362{
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001363 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 int error;
1365 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001367 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
1369 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1370 return XFS_ERROR(EIO);
1371
1372 lock_mode = xfs_ilock_map_shared(dp);
Barry Naujok384f3ce2008-05-21 16:58:22 +10001373 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 xfs_iunlock_map_shared(dp, lock_mode);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001375
1376 if (error)
1377 goto out;
1378
1379 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1380 if (error)
Barry Naujok384f3ce2008-05-21 16:58:22 +10001381 goto out_free_name;
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001382
1383 xfs_itrace_ref(*ipp);
1384 return 0;
1385
Barry Naujok384f3ce2008-05-21 16:58:22 +10001386out_free_name:
1387 if (ci_name)
1388 kmem_free(ci_name->name);
1389out:
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001390 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return error;
1392}
1393
Christoph Hellwig993386c2007-08-28 16:12:30 +10001394int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001396 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001397 struct xfs_name *name,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001398 mode_t mode,
1399 xfs_dev_t rdev,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001400 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 cred_t *credp)
1402{
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001403 int is_dir = S_ISDIR(mode);
1404 struct xfs_mount *mp = dp->i_mount;
1405 struct xfs_inode *ip = NULL;
1406 struct xfs_trans *tp = NULL;
Barry Naujok556b8b12008-04-10 12:22:07 +10001407 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 xfs_bmap_free_t free_list;
1409 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001410 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 uint cancel_flags;
1412 int committed;
1413 xfs_prid_t prid;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001414 struct xfs_dquot *udqp = NULL;
1415 struct xfs_dquot *gdqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 uint resblks;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001417 uint log_res;
1418 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001420 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001422 if (XFS_FORCED_SHUTDOWN(mp))
1423 return XFS_ERROR(EIO);
1424
Christoph Hellwigeb9df392007-08-16 18:42:07 +10001425 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001427 dp, DM_RIGHT_NULL, NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10001428 DM_RIGHT_NULL, name->name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001429 mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 if (error)
1432 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
1434
Nathan Scott365ca832005-06-21 15:39:12 +10001435 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1436 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 else
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001438 prid = dfltprid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 /*
1441 * Make sure that we have allocated dquot(s) on disk.
1442 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001443 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001444 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 if (error)
1446 goto std_return;
1447
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001448 if (is_dir) {
1449 rdev = 0;
1450 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1451 log_res = XFS_MKDIR_LOG_RES(mp);
1452 log_count = XFS_MKDIR_LOG_COUNT;
1453 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1454 } else {
1455 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1456 log_res = XFS_CREATE_LOG_RES(mp);
1457 log_count = XFS_CREATE_LOG_COUNT;
1458 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 /*
1464 * Initially assume that the file does not exist and
1465 * reserve the resources for that case. If that is not
1466 * the case we'll drop the one we have and get a more
1467 * appropriate transaction later.
1468 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001469 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1470 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 if (error == ENOSPC) {
Dave Chinner153fec42009-04-06 18:48:30 +02001472 /* flush outstanding delalloc blocks and retry */
1473 xfs_flush_inodes(dp);
Christoph Hellwig4734d402009-09-09 18:19:02 -05001474 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1475 XFS_TRANS_PERM_LOG_RES, log_count);
Dave Chinner153fec42009-04-06 18:48:30 +02001476 }
1477 if (error == ENOSPC) {
1478 /* No space at all so try a "no-allocation" reservation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 resblks = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001480 error = xfs_trans_reserve(tp, 0, log_res, 0,
1481 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483 if (error) {
1484 cancel_flags = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001485 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001488 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001489 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001491 /*
1492 * Check for directory link count overflow.
1493 */
1494 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1495 error = XFS_ERROR(EMLINK);
1496 goto out_trans_cancel;
1497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001499 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 /*
1502 * Reserve disk quota and the inode.
1503 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001504 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001506 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Barry Naujok556b8b12008-04-10 12:22:07 +10001508 error = xfs_dir_canenter(tp, dp, name, resblks);
1509 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001510 goto out_trans_cancel;
1511
1512 /*
1513 * A newly created regular or special file just has one directory
1514 * entry pointing to them, but a directory also the "." entry
1515 * pointing to itself.
1516 */
1517 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp,
1518 prid, resblks > 0, &ip, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 if (error) {
1520 if (error == ENOSPC)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001521 goto out_trans_cancel;
1522 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
1525 /*
1526 * At this point, we've gotten a newly allocated inode.
1527 * It is locked (and joined to the transaction).
1528 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001529 xfs_itrace_ref(ip);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001530 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +10001533 * Now we join the directory inode to the transaction. We do not do it
1534 * earlier because xfs_dir_ialloc might commit the previous transaction
1535 * (and release all the locks). An error from here on will result in
1536 * the transaction cancel unlocking dp so don't do it explicitly in the
1537 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 */
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001539 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001541 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Barry Naujok556b8b12008-04-10 12:22:07 +10001543 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001544 &first_block, &free_list, resblks ?
1545 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 if (error) {
1547 ASSERT(error != ENOSPC);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001548 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
1550 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1551 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1552
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001553 if (is_dir) {
1554 error = xfs_dir_init(tp, ip, dp);
1555 if (error)
1556 goto out_bmap_cancel;
1557
1558 error = xfs_bumplink(tp, dp);
1559 if (error)
1560 goto out_bmap_cancel;
1561 }
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 /*
1564 * If this is a synchronous mount, make sure that the
1565 * create transaction goes to disk before returning to
1566 * the user.
1567 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001568 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 /*
1572 * Attach the dquot(s) to the inodes and modify them incore.
1573 * These ids of the inode couldn't have changed since the new
1574 * inode has been locked ever since it was created.
1575 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001576 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578 /*
1579 * xfs_trans_commit normally decrements the vnode ref count
1580 * when it unlocks the inode. Since we want to return the
1581 * vnode to the caller, we bump the vnode ref count now.
1582 */
1583 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001585 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001586 if (error)
1587 goto out_abort_rele;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001589 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (error) {
1591 IRELE(ip);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001592 goto out_dqrele;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 }
1594
Christoph Hellwig7d095252009-06-08 15:33:32 +02001595 xfs_qm_dqrele(udqp);
1596 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001598 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 /* Fallthrough to std_return with error = 0 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001601 std_return:
1602 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1603 XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL,
1604 ip, DM_RIGHT_NULL, name->name, NULL, mode,
1605 error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 }
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return error;
1609
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001610 out_bmap_cancel:
1611 xfs_bmap_cancel(&free_list);
1612 out_trans_abort:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001614 out_trans_cancel:
1615 xfs_trans_cancel(tp, cancel_flags);
1616 out_dqrele:
Christoph Hellwig7d095252009-06-08 15:33:32 +02001617 xfs_qm_dqrele(udqp);
1618 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Christoph Hellwig993386c2007-08-28 16:12:30 +10001620 if (unlock_dp_on_error)
1621 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 goto std_return;
1624
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001625 out_abort_rele:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 /*
1627 * Wait until after the current transaction is aborted to
1628 * release the inode. This prevents recursive transactions
1629 * and deadlocks from xfs_inactive.
1630 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001631 xfs_bmap_cancel(&free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 cancel_flags |= XFS_TRANS_ABORT;
1633 xfs_trans_cancel(tp, cancel_flags);
1634 IRELE(ip);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001635 unlock_dp_on_error = B_FALSE;
1636 goto out_dqrele;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637}
1638
1639#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640int xfs_locked_n;
1641int xfs_small_retries;
1642int xfs_middle_retries;
1643int xfs_lots_retries;
1644int xfs_lock_delays;
1645#endif
1646
1647/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001648 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1649 * a different value
1650 */
1651static inline int
1652xfs_lock_inumorder(int lock_mode, int subclass)
1653{
1654 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001655 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001656 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001657 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001658
1659 return lock_mode;
1660}
1661
1662/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 * The following routine will lock n inodes in exclusive mode.
1664 * We assume the caller calls us with the inodes in i_ino order.
1665 *
1666 * We need to detect deadlock where an inode that we lock
1667 * is in the AIL and we start waiting for another inode that is locked
1668 * by a thread in a long running transaction (such as truncate). This can
1669 * result in deadlock since the long running trans might need to wait
1670 * for the inode we just locked in order to push the tail and free space
1671 * in the log.
1672 */
1673void
1674xfs_lock_inodes(
1675 xfs_inode_t **ips,
1676 int inodes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 uint lock_mode)
1678{
1679 int attempts = 0, i, j, try_lock;
1680 xfs_log_item_t *lp;
1681
1682 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1683
Christoph Hellwigcfa853e2008-04-22 17:34:06 +10001684 try_lock = 0;
1685 i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687again:
1688 for (; i < inodes; i++) {
1689 ASSERT(ips[i]);
1690
1691 if (i && (ips[i] == ips[i-1])) /* Already locked */
1692 continue;
1693
1694 /*
1695 * If try_lock is not set yet, make sure all locked inodes
1696 * are not in the AIL.
1697 * If any are, set try_lock to be used later.
1698 */
1699
1700 if (!try_lock) {
1701 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1702 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1703 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1704 try_lock++;
1705 }
1706 }
1707 }
1708
1709 /*
1710 * If any of the previous locks we have locked is in the AIL,
1711 * we must TRY to get the second and subsequent locks. If
1712 * we can't get any, we must release all we have
1713 * and try again.
1714 */
1715
1716 if (try_lock) {
1717 /* try_lock must be 0 if i is 0. */
1718 /*
1719 * try_lock means we have an inode locked
1720 * that is in the AIL.
1721 */
1722 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001723 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 attempts++;
1725
1726 /*
1727 * Unlock all previous guys and try again.
1728 * xfs_iunlock will try to push the tail
1729 * if the inode is in the AIL.
1730 */
1731
1732 for(j = i - 1; j >= 0; j--) {
1733
1734 /*
1735 * Check to see if we've already
1736 * unlocked this one.
1737 * Not the first one going back,
1738 * and the inode ptr is the same.
1739 */
1740 if ((j != (i - 1)) && ips[j] ==
1741 ips[j+1])
1742 continue;
1743
1744 xfs_iunlock(ips[j], lock_mode);
1745 }
1746
1747 if ((attempts % 5) == 0) {
1748 delay(1); /* Don't just spin the CPU */
1749#ifdef DEBUG
1750 xfs_lock_delays++;
1751#endif
1752 }
1753 i = 0;
1754 try_lock = 0;
1755 goto again;
1756 }
1757 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001758 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 }
1760 }
1761
1762#ifdef DEBUG
1763 if (attempts) {
1764 if (attempts < 5) xfs_small_retries++;
1765 else if (attempts < 100) xfs_middle_retries++;
1766 else xfs_lots_retries++;
1767 } else {
1768 xfs_locked_n++;
1769 }
1770#endif
1771}
1772
David Chinnerf9114eb2008-09-17 16:51:21 +10001773/*
1774 * xfs_lock_two_inodes() can only be used to lock one type of lock
1775 * at a time - the iolock or the ilock, but not both at once. If
1776 * we lock both at once, lockdep will report false positives saying
1777 * we have violated locking orders.
1778 */
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001779void
1780xfs_lock_two_inodes(
1781 xfs_inode_t *ip0,
1782 xfs_inode_t *ip1,
1783 uint lock_mode)
1784{
1785 xfs_inode_t *temp;
1786 int attempts = 0;
1787 xfs_log_item_t *lp;
1788
David Chinnerf9114eb2008-09-17 16:51:21 +10001789 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1790 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001791 ASSERT(ip0->i_ino != ip1->i_ino);
1792
1793 if (ip0->i_ino > ip1->i_ino) {
1794 temp = ip0;
1795 ip0 = ip1;
1796 ip1 = temp;
1797 }
1798
1799 again:
1800 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1801
1802 /*
1803 * If the first lock we have locked is in the AIL, we must TRY to get
1804 * the second lock. If we can't get it, we must release the first one
1805 * and try again.
1806 */
1807 lp = (xfs_log_item_t *)ip0->i_itemp;
1808 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1809 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1810 xfs_iunlock(ip0, lock_mode);
1811 if ((++attempts % 5) == 0)
1812 delay(1); /* Don't just spin the CPU */
1813 goto again;
1814 }
1815 } else {
1816 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1817 }
1818}
1819
Christoph Hellwig993386c2007-08-28 16:12:30 +10001820int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001822 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001823 struct xfs_name *name,
1824 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001826 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 xfs_trans_t *tp = NULL;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001828 int is_dir = S_ISDIR(ip->i_d.di_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 int error = 0;
1830 xfs_bmap_free_t free_list;
1831 xfs_fsblock_t first_block;
1832 int cancel_flags;
1833 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 int link_zero;
1835 uint resblks;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001836 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001838 xfs_itrace_entry(dp);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001839 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 if (XFS_FORCED_SHUTDOWN(mp))
1842 return XFS_ERROR(EIO);
1843
Christoph Hellwigeb9df392007-08-16 18:42:07 +10001844 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
Barry Naujok556b8b12008-04-10 12:22:07 +10001845 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
1846 NULL, DM_RIGHT_NULL, name->name, NULL,
1847 ip->i_d.di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 if (error)
1849 return error;
1850 }
1851
Christoph Hellwig7d095252009-06-08 15:33:32 +02001852 error = xfs_qm_dqattach(dp, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001853 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Christoph Hellwig7d095252009-06-08 15:33:32 +02001856 error = xfs_qm_dqattach(ip, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001857 if (error)
1858 goto std_return;
1859
1860 if (is_dir) {
1861 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1862 log_count = XFS_DEFAULT_LOG_COUNT;
1863 } else {
1864 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1865 log_count = XFS_REMOVE_LOG_COUNT;
1866 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 /*
1870 * We try to get the real space reservation first,
1871 * allowing for directory btree deletion(s) implying
1872 * possible bmap insert(s). If we can't get the space
1873 * reservation then we use 0 instead, and avoid the bmap
1874 * btree insert(s) in the directory code by, if the bmap
1875 * insert tries to happen, instead trimming the LAST
1876 * block from the directory.
1877 */
1878 resblks = XFS_REMOVE_SPACE_RES(mp);
1879 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001880 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (error == ENOSPC) {
1882 resblks = 0;
1883 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001884 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 }
1886 if (error) {
1887 ASSERT(error != ENOSPC);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001888 cancel_flags = 0;
1889 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
1891
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001892 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894 /*
1895 * At this point, we've gotten both the directory and the entry
1896 * inodes locked.
1897 */
Christoph Hellwig5df78e72008-04-22 17:34:24 +10001898 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig82dab942008-04-22 17:34:18 +10001900
1901 IHOLD(dp);
1902 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 /*
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001905 * If we're removing a directory perform some additional validation.
1906 */
1907 if (is_dir) {
1908 ASSERT(ip->i_d.di_nlink >= 2);
1909 if (ip->i_d.di_nlink != 2) {
1910 error = XFS_ERROR(ENOTEMPTY);
1911 goto out_trans_cancel;
1912 }
1913 if (!xfs_dir_isempty(ip)) {
1914 error = XFS_ERROR(ENOTEMPTY);
1915 goto out_trans_cancel;
1916 }
1917 }
1918
Eric Sandeen9d87c312009-01-14 23:22:07 -06001919 xfs_bmap_init(&free_list, &first_block);
Barry Naujok556b8b12008-04-10 12:22:07 +10001920 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
Christoph Hellwigd4377d82008-04-22 17:33:46 +10001921 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (error) {
1923 ASSERT(error != ENOENT);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001924 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 }
1926 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1927
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001928 if (is_dir) {
1929 /*
1930 * Drop the link from ip's "..".
1931 */
1932 error = xfs_droplink(tp, dp);
1933 if (error)
1934 goto out_bmap_cancel;
1935
1936 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001937 * Drop the "." link from ip to self.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001938 */
1939 error = xfs_droplink(tp, ip);
1940 if (error)
1941 goto out_bmap_cancel;
1942 } else {
1943 /*
1944 * When removing a non-directory we need to log the parent
Dave Chinner26c52952008-11-28 14:23:37 +11001945 * inode here. For a directory this is done implicitly
1946 * by the xfs_droplink call for the ".." entry.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001947 */
1948 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 }
1950
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001951 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001952 * Drop the link from dp to ip.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001953 */
1954 error = xfs_droplink(tp, ip);
1955 if (error)
1956 goto out_bmap_cancel;
1957
1958 /*
1959 * Determine if this is the last link while
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 * we are in the transaction.
1961 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001962 link_zero = (ip->i_d.di_nlink == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
1964 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 * If this is a synchronous mount, make sure that the
1966 * remove transaction goes to disk before returning to
1967 * the user.
1968 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001969 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001972 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001973 if (error)
1974 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001976 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig5df78e72008-04-22 17:34:24 +10001977 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
1980 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10001981 * If we are using filestreams, kill the stream association.
1982 * If the file is still open it may get a new one but that
1983 * will get killed on last close in xfs_close() so we don't
1984 * have to worry about that.
1985 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001986 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
David Chinner2a82b8b2007-07-11 11:09:12 +10001987 xfs_filestream_deassociate(ip);
1988
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001989 xfs_itrace_exit(ip);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001990 xfs_itrace_exit(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10001993 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001994 XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL,
1995 NULL, DM_RIGHT_NULL, name->name, NULL,
1996 ip->i_d.di_mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 }
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 return error;
2000
Christoph Hellwig8f112e32008-06-23 13:25:17 +10002001 out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 xfs_bmap_cancel(&free_list);
2003 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10002004 out_trans_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 xfs_trans_cancel(tp, cancel_flags);
2006 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007}
2008
Christoph Hellwig993386c2007-08-28 16:12:30 +10002009int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002011 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002012 xfs_inode_t *sip,
Barry Naujok556b8b12008-04-10 12:22:07 +10002013 struct xfs_name *target_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002015 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 int error;
2018 xfs_bmap_free_t free_list;
2019 xfs_fsblock_t first_block;
2020 int cancel_flags;
2021 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 int resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002024 xfs_itrace_entry(tdp);
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002025 xfs_itrace_entry(sip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002027 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (XFS_FORCED_SHUTDOWN(mp))
2030 return XFS_ERROR(EIO);
2031
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002032 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002034 tdp, DM_RIGHT_NULL,
2035 sip, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002036 target_name->name, NULL, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (error)
2038 return error;
2039 }
2040
2041 /* Return through std_return after this point. */
2042
Christoph Hellwig7d095252009-06-08 15:33:32 +02002043 error = xfs_qm_dqattach(sip, 0);
Christoph Hellwigcb3f35b2009-02-04 09:34:20 +01002044 if (error)
2045 goto std_return;
2046
Christoph Hellwig7d095252009-06-08 15:33:32 +02002047 error = xfs_qm_dqattach(tdp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 if (error)
2049 goto std_return;
2050
2051 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2052 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10002053 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2055 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2056 if (error == ENOSPC) {
2057 resblks = 0;
2058 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2059 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2060 }
2061 if (error) {
2062 cancel_flags = 0;
2063 goto error_return;
2064 }
2065
Christoph Hellwige1cccd92008-08-13 16:18:07 +10002066 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
2068 /*
2069 * Increment vnode ref counts since xfs_trans_commit &
2070 * xfs_trans_cancel will both unlock the inodes and
2071 * decrement the associated ref counts.
2072 */
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002073 IHOLD(sip);
2074 IHOLD(tdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2076 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2077
2078 /*
2079 * If the source has too many links, we can't make any more to it.
2080 */
2081 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2082 error = XFS_ERROR(EMLINK);
2083 goto error_return;
2084 }
2085
Nathan Scott365ca832005-06-21 15:39:12 +10002086 /*
2087 * If we are using project inheritance, we only allow hard link
2088 * creation in our tree when the project IDs are the same; else
2089 * the tree quota mechanism could be circumvented.
2090 */
2091 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2092 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10002093 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10002094 goto error_return;
2095 }
2096
Barry Naujok556b8b12008-04-10 12:22:07 +10002097 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2098 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 goto error_return;
2100
Eric Sandeen9d87c312009-01-14 23:22:07 -06002101 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Barry Naujok556b8b12008-04-10 12:22:07 +10002103 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2104 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (error)
2106 goto abort_return;
2107 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2109
2110 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002111 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 /*
2115 * If this is a synchronous mount, make sure that the
2116 * link transaction goes to disk before returning to
2117 * the user.
2118 */
2119 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2120 xfs_trans_set_sync(tp);
2121 }
2122
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002123 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 if (error) {
2125 xfs_bmap_cancel(&free_list);
2126 goto abort_return;
2127 }
2128
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002129 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002130 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
2133 /* Fall through to std_return with error = 0. */
2134std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002135 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002137 tdp, DM_RIGHT_NULL,
2138 sip, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002139 target_name->name, NULL, 0, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 }
2141 return error;
2142
2143 abort_return:
2144 cancel_flags |= XFS_TRANS_ABORT;
2145 /* FALLTHROUGH */
Jesper Juhl014c2542006-01-15 02:37:08 +01002146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 error_return:
2148 xfs_trans_cancel(tp, cancel_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 goto std_return;
2150}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002151
Christoph Hellwig993386c2007-08-28 16:12:30 +10002152int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002154 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10002155 struct xfs_name *link_name,
2156 const char *target_path,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002157 mode_t mode,
Christoph Hellwig3937be52008-03-06 13:46:19 +11002158 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 cred_t *credp)
2160{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002161 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 xfs_inode_t *ip;
2164 int error;
2165 int pathlen;
2166 xfs_bmap_free_t free_list;
2167 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10002168 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 uint cancel_flags;
2170 int committed;
2171 xfs_fileoff_t first_fsb;
2172 xfs_filblks_t fs_blocks;
2173 int nmaps;
2174 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2175 xfs_daddr_t d;
Barry Naujok556b8b12008-04-10 12:22:07 +10002176 const char *cur_chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 int byte_cnt;
2178 int n;
2179 xfs_buf_t *bp;
2180 xfs_prid_t prid;
2181 struct xfs_dquot *udqp, *gdqp;
2182 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Christoph Hellwig3937be52008-03-06 13:46:19 +11002184 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 error = 0;
2186 ip = NULL;
2187 tp = NULL;
2188
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002189 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190
2191 if (XFS_FORCED_SHUTDOWN(mp))
2192 return XFS_ERROR(EIO);
2193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 /*
2195 * Check component lengths of the target path name.
2196 */
2197 pathlen = strlen(target_path);
2198 if (pathlen >= MAXPATHLEN) /* total string too long */
2199 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002201 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002202 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002204 link_name->name, target_path, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (error)
2206 return error;
2207 }
2208
2209 /* Return through std_return after this point. */
2210
2211 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002212 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2213 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 else
2215 prid = (xfs_prid_t)dfltprid;
2216
2217 /*
2218 * Make sure that we have allocated dquot(s) on disk.
2219 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002220 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2222 if (error)
2223 goto std_return;
2224
2225 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2226 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2227 /*
2228 * The symlink will fit into the inode data fork?
2229 * There can't be any attributes so we get the whole variable part.
2230 */
2231 if (pathlen <= XFS_LITINO(mp))
2232 fs_blocks = 0;
2233 else
2234 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
Barry Naujok556b8b12008-04-10 12:22:07 +10002235 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2237 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2238 if (error == ENOSPC && fs_blocks == 0) {
2239 resblks = 0;
2240 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2241 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2242 }
2243 if (error) {
2244 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 goto error_return;
2246 }
2247
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002248 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002249 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
2251 /*
2252 * Check whether the directory allows new symlinks or not.
2253 */
2254 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2255 error = XFS_ERROR(EPERM);
2256 goto error_return;
2257 }
2258
2259 /*
2260 * Reserve disk quota : blocks and inode.
2261 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002262 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (error)
2264 goto error_return;
2265
2266 /*
2267 * Check for ability to enter directory entry, if no space reserved.
2268 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002269 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2270 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 goto error_return;
2272 /*
2273 * Initialize the bmap freelist prior to calling either
2274 * bmapi or the directory create code.
2275 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002276 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
2278 /*
2279 * Allocate an inode for the symlink.
2280 */
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002281 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 1, 0, credp, prid, resblks > 0, &ip, NULL);
2283 if (error) {
2284 if (error == ENOSPC)
2285 goto error_return;
2286 goto error1;
2287 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002288 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Christoph Hellwig993386c2007-08-28 16:12:30 +10002290 /*
2291 * An error after we've joined dp to the transaction will result in the
2292 * transaction cancel unlocking dp so don't do it explicitly in the
2293 * error path.
2294 */
Christoph Hellwig3937be52008-03-06 13:46:19 +11002295 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002297 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /*
2300 * Also attach the dquot(s) to it, if applicable.
2301 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002302 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
2304 if (resblks)
2305 resblks -= XFS_IALLOC_SPACE_RES(mp);
2306 /*
2307 * If the symlink will fit into the inode, write it inline.
2308 */
2309 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2310 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2311 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2312 ip->i_d.di_size = pathlen;
2313
2314 /*
2315 * The inode was initially created in extent format.
2316 */
2317 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2318 ip->i_df.if_flags |= XFS_IFINLINE;
2319
2320 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2321 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2322
2323 } else {
2324 first_fsb = 0;
2325 nmaps = SYMLINK_MAPS;
2326
2327 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2328 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2329 &first_block, resblks, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002330 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (error) {
2332 goto error1;
2333 }
2334
2335 if (resblks)
2336 resblks -= fs_blocks;
2337 ip->i_d.di_size = pathlen;
2338 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2339
2340 cur_chunk = target_path;
2341 for (n = 0; n < nmaps; n++) {
2342 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2343 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2344 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2345 BTOBB(byte_cnt), 0);
2346 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2347 if (pathlen < byte_cnt) {
2348 byte_cnt = pathlen;
2349 }
2350 pathlen -= byte_cnt;
2351
2352 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2353 cur_chunk += byte_cnt;
2354
2355 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2356 }
2357 }
2358
2359 /*
2360 * Create the directory entry for the symlink.
2361 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002362 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2363 &first_block, &free_list, resblks);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002364 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2367 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2368
2369 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 * If this is a synchronous mount, make sure that the
2371 * symlink transaction goes to disk before returning to
2372 * the user.
2373 */
2374 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2375 xfs_trans_set_sync(tp);
2376 }
2377
2378 /*
2379 * xfs_trans_commit normally decrements the vnode ref count
2380 * when it unlocks the inode. Since we want to return the
2381 * vnode to the caller, we bump the vnode ref count now.
2382 */
2383 IHOLD(ip);
2384
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002385 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 if (error) {
2387 goto error2;
2388 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002389 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002390 xfs_qm_dqrele(udqp);
2391 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
2393 /* Fall through to std_return with error = 0 or errno from
2394 * xfs_trans_commit */
2395std_return:
Christoph Hellwig993386c2007-08-28 16:12:30 +10002396 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002398 dp, DM_RIGHT_NULL,
2399 error ? NULL : ip,
Barry Naujok556b8b12008-04-10 12:22:07 +10002400 DM_RIGHT_NULL, link_name->name,
2401 target_path, 0, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 }
2403
Christoph Hellwig3937be52008-03-06 13:46:19 +11002404 if (!error)
2405 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 return error;
2407
2408 error2:
2409 IRELE(ip);
2410 error1:
2411 xfs_bmap_cancel(&free_list);
2412 cancel_flags |= XFS_TRANS_ABORT;
2413 error_return:
2414 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002415 xfs_qm_dqrele(udqp);
2416 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
Christoph Hellwig993386c2007-08-28 16:12:30 +10002418 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
2421 goto std_return;
2422}
2423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424int
Christoph Hellwig993386c2007-08-28 16:12:30 +10002425xfs_set_dmattrs(
2426 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10002428 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002430 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 int error;
2433
2434 if (!capable(CAP_SYS_ADMIN))
2435 return XFS_ERROR(EPERM);
2436
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (XFS_FORCED_SHUTDOWN(mp))
2438 return XFS_ERROR(EIO);
2439
2440 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2441 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2442 if (error) {
2443 xfs_trans_cancel(tp, 0);
2444 return error;
2445 }
2446 xfs_ilock(ip, XFS_ILOCK_EXCL);
2447 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2448
Christoph Hellwig613d7042007-10-11 17:44:08 +10002449 ip->i_d.di_dmevmask = evmask;
2450 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
2452 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2453 IHOLD(ip);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002454 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
2456 return error;
2457}
2458
Christoph Hellwig993386c2007-08-28 16:12:30 +10002459int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460xfs_reclaim(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002461 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002464 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Christoph Hellwigdf80c932008-08-13 16:22:09 +10002466 ASSERT(!VN_MAPPED(VFS_I(ip)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 /* bad inode, get out here ASAP */
Christoph Hellwigcb4c8cc2009-03-16 08:25:25 +01002469 if (is_bad_inode(VFS_I(ip))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 xfs_ireclaim(ip);
2471 return 0;
2472 }
2473
Christoph Hellwig25e41b32008-12-03 12:20:39 +01002474 xfs_ioend_wait(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10002476 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002478 /*
David Chinner4c606582006-11-11 18:05:00 +11002479 * If we have nothing to flush with this inode then complete the
2480 * teardown now, otherwise break the link between the xfs inode and the
2481 * linux inode and clean up the xfs inode later. This avoids flushing
2482 * the inode to disk during the delete operation itself.
2483 *
2484 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2485 * first to ensure that xfs_iunpin() will never see an xfs inode
2486 * that has a linux inode being reclaimed. Synchronisation is provided
2487 * by the i_flags_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 */
2489 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
2490 xfs_ilock(ip, XFS_ILOCK_EXCL);
2491 xfs_iflock(ip);
David Chinnerbf904242008-10-30 17:36:14 +11002492 xfs_iflags_set(ip, XFS_IRECLAIMABLE);
David Chinner1dc33182008-10-30 17:37:15 +11002493 return xfs_reclaim_inode(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 }
David Chinner11654512008-10-30 17:37:49 +11002495 xfs_inode_set_reclaim_tag(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 return 0;
2497}
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499/*
2500 * xfs_alloc_file_space()
2501 * This routine allocates disk space for the given file.
2502 *
2503 * If alloc_type == 0, this request is for an ALLOCSP type
2504 * request which will change the file size. In this case, no
2505 * DMAPI event will be generated by the call. A TRUNCATE event
2506 * will be generated later by xfs_setattr.
2507 *
2508 * If alloc_type != 0, this request is for a RESVSP type
2509 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2510 * lower block boundary byte address is less than the file's
2511 * length.
2512 *
2513 * RETURNS:
2514 * 0 on success
2515 * errno on error
2516 *
2517 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002518STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519xfs_alloc_file_space(
2520 xfs_inode_t *ip,
2521 xfs_off_t offset,
2522 xfs_off_t len,
2523 int alloc_type,
2524 int attr_flags)
2525{
Nathan Scottdd9f4382006-01-11 15:28:28 +11002526 xfs_mount_t *mp = ip->i_mount;
2527 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 xfs_filblks_t allocated_fsb;
2529 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002530 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002532 xfs_fsblock_t firstfsb;
2533 int nimaps;
2534 int bmapi_flag;
2535 int quota_flag;
2536 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002538 xfs_bmbt_irec_t imaps[1], *imapp;
2539 xfs_bmap_free_t free_list;
2540 uint qblocks, resblks, resrtextents;
2541 int committed;
2542 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002544 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
2546 if (XFS_FORCED_SHUTDOWN(mp))
2547 return XFS_ERROR(EIO);
2548
Christoph Hellwig7d095252009-06-08 15:33:32 +02002549 error = xfs_qm_dqattach(ip, 0);
2550 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 return error;
2552
2553 if (len <= 0)
2554 return XFS_ERROR(EINVAL);
2555
David Chinner957d0eb2007-06-18 16:50:37 +10002556 rt = XFS_IS_REALTIME_INODE(ip);
2557 extsz = xfs_get_extsz_hint(ip);
2558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11002561 nimaps = 1;
2562 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2564 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2565
2566 /* Generate a DMAPI event if needed. */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002567 if (alloc_type != 0 && offset < ip->i_size &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002568 (attr_flags & XFS_ATTR_DMI) == 0 &&
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002569 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 xfs_off_t end_dmi_offset;
2571
2572 end_dmi_offset = offset+len;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002573 if (end_dmi_offset > ip->i_size)
2574 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002575 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
2576 end_dmi_offset - offset, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002578 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
2580
2581 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002582 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 */
2584retry:
2585 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002586 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Nathan Scottdd9f4382006-01-11 15:28:28 +11002588 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11002589 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11002590 */
2591 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002593 do_div(s, extsz);
2594 s *= extsz;
2595 e = startoffset_fsb + allocatesize_fsb;
2596 if ((temp = do_mod(startoffset_fsb, extsz)))
2597 e += temp;
2598 if ((temp = do_mod(e, extsz)))
2599 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002601 s = 0;
2602 e = allocatesize_fsb;
2603 }
2604
2605 if (unlikely(rt)) {
2606 resrtextents = qblocks = (uint)(e - s);
2607 resrtextents /= mp->m_sb.sb_rextsize;
2608 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2609 quota_flag = XFS_QMOPT_RES_RTBLKS;
2610 } else {
2611 resrtextents = 0;
2612 resblks = qblocks = \
2613 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
2614 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 }
2616
2617 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002618 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 */
2620 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002621 error = xfs_trans_reserve(tp, resblks,
2622 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 XFS_TRANS_PERM_LOG_RES,
2624 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002626 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 */
2628 if (error) {
2629 /*
2630 * Free the transaction structure.
2631 */
2632 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2633 xfs_trans_cancel(tp, 0);
2634 break;
2635 }
2636 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002637 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2638 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 if (error)
2640 goto error1;
2641
2642 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2643 xfs_trans_ihold(tp, ip);
2644
2645 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002646 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002648 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002649 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11002650 allocatesize_fsb, bmapi_flag,
2651 &firstfsb, 0, imapp, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002652 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (error) {
2654 goto error0;
2655 }
2656
2657 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002658 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002660 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 if (error) {
2662 goto error0;
2663 }
2664
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002665 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2667 if (error) {
2668 break;
2669 }
2670
2671 allocated_fsb = imapp->br_blockcount;
2672
Nathan Scottdd9f4382006-01-11 15:28:28 +11002673 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 error = XFS_ERROR(ENOSPC);
2675 break;
2676 }
2677
2678 startoffset_fsb += allocated_fsb;
2679 allocatesize_fsb -= allocated_fsb;
2680 }
2681dmapi_enospc_check:
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002682 if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 &&
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002683 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002685 ip, DM_RIGHT_NULL,
2686 ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
2688 if (error == 0)
2689 goto retry; /* Maybe DMAPI app. has made space */
2690 /* else fall through with error from XFS_SEND_DATA */
2691 }
2692
2693 return error;
2694
Nathan Scottdd9f4382006-01-11 15:28:28 +11002695error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 xfs_bmap_cancel(&free_list);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002697 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002698
2699error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2702 goto dmapi_enospc_check;
2703}
2704
2705/*
2706 * Zero file bytes between startoff and endoff inclusive.
2707 * The iolock is held exclusive and no blocks are buffered.
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002708 *
2709 * This function is used by xfs_free_file_space() to zero
2710 * partial blocks when the range to free is not block aligned.
2711 * When unreserving space with boundaries that are not block
2712 * aligned we round up the start and round down the end
2713 * boundaries and then use this function to zero the parts of
2714 * the blocks that got dropped during the rounding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 */
2716STATIC int
2717xfs_zero_remaining_bytes(
2718 xfs_inode_t *ip,
2719 xfs_off_t startoff,
2720 xfs_off_t endoff)
2721{
2722 xfs_bmbt_irec_t imap;
2723 xfs_fileoff_t offset_fsb;
2724 xfs_off_t lastoffset;
2725 xfs_off_t offset;
2726 xfs_buf_t *bp;
2727 xfs_mount_t *mp = ip->i_mount;
2728 int nimap;
2729 int error = 0;
2730
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002731 /*
2732 * Avoid doing I/O beyond eof - it's not necessary
2733 * since nothing can read beyond eof. The space will
2734 * be zeroed when the file is extended anyway.
2735 */
2736 if (startoff >= ip->i_size)
2737 return 0;
2738
2739 if (endoff > ip->i_size)
2740 endoff = ip->i_size;
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
Eric Sandeen71ddabb2007-11-23 16:29:42 +11002743 XFS_IS_REALTIME_INODE(ip) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 mp->m_rtdev_targp : mp->m_ddev_targp);
Lachlan McIlroyc6422612008-12-05 13:16:15 +11002745 if (!bp)
2746 return XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
2748 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2749 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2750 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002751 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002752 NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 if (error || nimap < 1)
2754 break;
2755 ASSERT(imap.br_blockcount >= 1);
2756 ASSERT(imap.br_startoff == offset_fsb);
2757 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2758 if (lastoffset > endoff)
2759 lastoffset = endoff;
2760 if (imap.br_startblock == HOLESTARTBLOCK)
2761 continue;
2762 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2763 if (imap.br_state == XFS_EXT_UNWRITTEN)
2764 continue;
2765 XFS_BUF_UNDONE(bp);
2766 XFS_BUF_UNWRITE(bp);
2767 XFS_BUF_READ(bp);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002768 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 xfsbdstrat(mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002770 error = xfs_iowait(bp);
2771 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2773 mp, bp, XFS_BUF_ADDR(bp));
2774 break;
2775 }
2776 memset(XFS_BUF_PTR(bp) +
2777 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2778 0, lastoffset - offset + 1);
2779 XFS_BUF_UNDONE(bp);
2780 XFS_BUF_UNREAD(bp);
2781 XFS_BUF_WRITE(bp);
2782 xfsbdstrat(mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002783 error = xfs_iowait(bp);
2784 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2786 mp, bp, XFS_BUF_ADDR(bp));
2787 break;
2788 }
2789 }
2790 xfs_buf_free(bp);
2791 return error;
2792}
2793
2794/*
2795 * xfs_free_file_space()
2796 * This routine frees disk space for the given file.
2797 *
2798 * This routine is only called by xfs_change_file_space
2799 * for an UNRESVSP type call.
2800 *
2801 * RETURNS:
2802 * 0 on success
2803 * errno on error
2804 *
2805 */
2806STATIC int
2807xfs_free_file_space(
2808 xfs_inode_t *ip,
2809 xfs_off_t offset,
2810 xfs_off_t len,
2811 int attr_flags)
2812{
2813 int committed;
2814 int done;
2815 xfs_off_t end_dmi_offset;
2816 xfs_fileoff_t endoffset_fsb;
2817 int error;
2818 xfs_fsblock_t firstfsb;
2819 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 xfs_bmbt_irec_t imap;
2821 xfs_off_t ioffset;
2822 xfs_extlen_t mod=0;
2823 xfs_mount_t *mp;
2824 int nimap;
2825 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10002826 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 int rt;
2828 xfs_fileoff_t startoffset_fsb;
2829 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002830 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 mp = ip->i_mount;
2833
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002834 xfs_itrace_entry(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002835
Christoph Hellwig7d095252009-06-08 15:33:32 +02002836 error = xfs_qm_dqattach(ip, 0);
2837 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 return error;
2839
2840 error = 0;
2841 if (len <= 0) /* if nothing being freed */
2842 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11002843 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2845 end_dmi_offset = offset + len;
2846 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
2847
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002848 if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 &&
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002849 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002850 if (end_dmi_offset > ip->i_size)
2851 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002852 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 offset, end_dmi_offset - offset,
2854 AT_DELAY_FLAG(attr_flags), NULL);
2855 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002856 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 }
2858
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002859 if (attr_flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002860 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11002861 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig25e41b32008-12-03 12:20:39 +01002863 /* wait for the completion of any pending DIOs */
2864 xfs_ioend_wait(ip);
Yingping Lu9fa80462006-03-22 12:44:35 +11002865 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002866
Tim Shimmine6a4b372007-11-23 16:30:42 +11002867 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002869
Christoph Hellwigdf80c932008-08-13 16:22:09 +10002870 if (VN_CACHED(VFS_I(ip)) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11002871 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
2872 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10002873 if (error)
2874 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002875 }
2876
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 /*
2878 * Need to zero the stuff we're not freeing, on disk.
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002879 * If it's a realtime file & can't use unwritten extents then we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2881 * will take care of it for us.
2882 */
Eric Sandeen62118702008-03-06 13:44:28 +11002883 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002885 error = xfs_bmapi(NULL, ip, startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002886 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 if (error)
2888 goto out_unlock_iolock;
2889 ASSERT(nimap == 0 || nimap == 1);
2890 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2891 xfs_daddr_t block;
2892
2893 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2894 block = imap.br_startblock;
2895 mod = do_div(block, mp->m_sb.sb_rextsize);
2896 if (mod)
2897 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2898 }
2899 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002900 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002901 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 if (error)
2903 goto out_unlock_iolock;
2904 ASSERT(nimap == 0 || nimap == 1);
2905 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2906 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2907 mod++;
2908 if (mod && (mod != mp->m_sb.sb_rextsize))
2909 endoffset_fsb -= mod;
2910 }
2911 }
2912 if ((done = (endoffset_fsb <= startoffset_fsb)))
2913 /*
2914 * One contiguous piece to clear
2915 */
2916 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2917 else {
2918 /*
2919 * Some full blocks, possibly two pieces to clear
2920 */
2921 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2922 error = xfs_zero_remaining_bytes(ip, offset,
2923 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2924 if (!error &&
2925 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2926 error = xfs_zero_remaining_bytes(ip,
2927 XFS_FSB_TO_B(mp, endoffset_fsb),
2928 offset + len - 1);
2929 }
2930
2931 /*
2932 * free file space until done or until there is an error
2933 */
2934 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2935 while (!error && !done) {
2936
2937 /*
David Chinner91ebecc2007-07-19 16:28:30 +10002938 * allocate and setup the transaction. Allow this
2939 * transaction to dip into the reserve blocks to ensure
2940 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 */
2942 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10002943 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 error = xfs_trans_reserve(tp,
2945 resblks,
2946 XFS_WRITE_LOG_RES(mp),
2947 0,
2948 XFS_TRANS_PERM_LOG_RES,
2949 XFS_WRITE_LOG_COUNT);
2950
2951 /*
2952 * check for running out of space
2953 */
2954 if (error) {
2955 /*
2956 * Free the transaction structure.
2957 */
2958 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2959 xfs_trans_cancel(tp, 0);
2960 break;
2961 }
2962 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002963 error = xfs_trans_reserve_quota(tp, mp,
2964 ip->i_udquot, ip->i_gdquot,
2965 resblks, 0, XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 if (error)
2967 goto error1;
2968
2969 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2970 xfs_trans_ihold(tp, ip);
2971
2972 /*
2973 * issue the bunmapi() call to free the blocks
2974 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002975 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002976 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 endoffset_fsb - startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002978 0, 2, &firstfsb, &free_list, NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 if (error) {
2980 goto error0;
2981 }
2982
2983 /*
2984 * complete the transaction
2985 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002986 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 if (error) {
2988 goto error0;
2989 }
2990
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002991 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2993 }
2994
2995 out_unlock_iolock:
2996 if (need_iolock)
2997 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2998 return error;
2999
3000 error0:
3001 xfs_bmap_cancel(&free_list);
3002 error1:
3003 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3004 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3005 XFS_ILOCK_EXCL);
3006 return error;
3007}
3008
3009/*
3010 * xfs_change_file_space()
3011 * This routine allocates or frees disk space for the given file.
3012 * The user specified parameters are checked for alignment and size
3013 * limitations.
3014 *
3015 * RETURNS:
3016 * 0 on success
3017 * errno on error
3018 *
3019 */
3020int
3021xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003022 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 int cmd,
3024 xfs_flock64_t *bf,
3025 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 int attr_flags)
3027{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003028 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 int clrprealloc;
3030 int error;
3031 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 int setprealloc;
3033 xfs_off_t startoffset;
3034 xfs_off_t llen;
3035 xfs_trans_t *tp;
Christoph Hellwig0f285c82008-07-18 17:13:28 +10003036 struct iattr iattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003038 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039
Christoph Hellwig993386c2007-08-28 16:12:30 +10003040 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 return XFS_ERROR(EINVAL);
3042
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 switch (bf->l_whence) {
3044 case 0: /*SEEK_SET*/
3045 break;
3046 case 1: /*SEEK_CUR*/
3047 bf->l_start += offset;
3048 break;
3049 case 2: /*SEEK_END*/
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003050 bf->l_start += ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 break;
3052 default:
3053 return XFS_ERROR(EINVAL);
3054 }
3055
3056 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3057
3058 if ( (bf->l_start < 0)
3059 || (bf->l_start > XFS_MAXIOFFSET(mp))
3060 || (bf->l_start + llen < 0)
3061 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3062 return XFS_ERROR(EINVAL);
3063
3064 bf->l_whence = 0;
3065
3066 startoffset = bf->l_start;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003067 fsize = ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068
3069 /*
3070 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3071 * file space.
3072 * These calls do NOT zero the data space allocated to the file,
3073 * nor do they change the file size.
3074 *
3075 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3076 * space.
3077 * These calls cause the new file data to be zeroed and the file
3078 * size to be changed.
3079 */
3080 setprealloc = clrprealloc = 0;
3081
3082 switch (cmd) {
3083 case XFS_IOC_RESVSP:
3084 case XFS_IOC_RESVSP64:
3085 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3086 1, attr_flags);
3087 if (error)
3088 return error;
3089 setprealloc = 1;
3090 break;
3091
3092 case XFS_IOC_UNRESVSP:
3093 case XFS_IOC_UNRESVSP64:
3094 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3095 attr_flags)))
3096 return error;
3097 break;
3098
3099 case XFS_IOC_ALLOCSP:
3100 case XFS_IOC_ALLOCSP64:
3101 case XFS_IOC_FREESP:
3102 case XFS_IOC_FREESP64:
3103 if (startoffset > fsize) {
3104 error = xfs_alloc_file_space(ip, fsize,
3105 startoffset - fsize, 0, attr_flags);
3106 if (error)
3107 break;
3108 }
3109
Christoph Hellwig0f285c82008-07-18 17:13:28 +10003110 iattr.ia_valid = ATTR_SIZE;
3111 iattr.ia_size = startoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +11003113 error = xfs_setattr(ip, &iattr, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
3115 if (error)
3116 return error;
3117
3118 clrprealloc = 1;
3119 break;
3120
3121 default:
3122 ASSERT(0);
3123 return XFS_ERROR(EINVAL);
3124 }
3125
3126 /*
3127 * update the inode timestamp, mode, and prealloc flag bits
3128 */
3129 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3130
3131 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3132 0, 0, 0))) {
3133 /* ASSERT(0); */
3134 xfs_trans_cancel(tp, 0);
3135 return error;
3136 }
3137
3138 xfs_ilock(ip, XFS_ILOCK_EXCL);
3139
3140 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3141 xfs_trans_ihold(tp, ip);
3142
Christoph Hellwig0f285c82008-07-18 17:13:28 +10003143 if ((attr_flags & XFS_ATTR_DMI) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 ip->i_d.di_mode &= ~S_ISUID;
3145
3146 /*
3147 * Note that we don't have to worry about mandatory
3148 * file locking being disabled here because we only
3149 * clear the S_ISGID bit if the Group execute bit is
3150 * on, but if it was on then mandatory locking wouldn't
3151 * have been enabled.
3152 */
3153 if (ip->i_d.di_mode & S_IXGRP)
3154 ip->i_d.di_mode &= ~S_ISGID;
3155
3156 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3157 }
3158 if (setprealloc)
3159 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
3160 else if (clrprealloc)
3161 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
3162
3163 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3164 xfs_trans_set_sync(tp);
3165
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003166 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167
3168 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3169
3170 return error;
3171}