blob: b7a5fe7c52c895776f125d53851d40c48795dce7 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
32#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_dinode.h"
34#include "xfs_inode.h"
35#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_ialloc.h"
38#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_bmap.h"
Christoph Hellwigef14f0c2009-06-10 17:07:47 +020040#include "xfs_acl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_attr.h"
42#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_quota.h"
45#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110046#include "xfs_rtalloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100049#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100050#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000051#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Christoph Hellwig993386c2007-08-28 16:12:30 +100053int
Linus Torvalds1da177e2005-04-16 15:20:36 -070054xfs_setattr(
Christoph Hellwig0f285c82008-07-18 17:13:28 +100055 struct xfs_inode *ip,
56 struct iattr *iattr,
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +110057 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
Christoph Hellwig993386c2007-08-28 16:12:30 +100059 xfs_mount_t *mp = ip->i_mount;
David Chinnere4f75292008-08-13 16:00:45 +100060 struct inode *inode = VFS_I(ip);
Christoph Hellwig0f285c82008-07-18 17:13:28 +100061 int mask = iattr->ia_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 int code;
64 uint lock_flags;
65 uint commit_flags=0;
66 uid_t uid=0, iuid=0;
67 gid_t gid=0, igid=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
Dean Roehrich5fcbab32005-05-05 13:27:19 -070069 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Christoph Hellwigcca28fb2010-06-24 11:57:09 +100071 trace_xfs_setattr(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Christoph Hellwigbd186aa2007-08-30 17:21:12 +100073 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 return XFS_ERROR(EROFS);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (XFS_FORCED_SHUTDOWN(mp))
77 return XFS_ERROR(EIO);
78
Christoph Hellwigc4cd7472008-12-09 04:47:34 -050079 code = -inode_change_ok(inode, iattr);
80 if (code)
81 return code;
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 olddquot1 = olddquot2 = NULL;
84 udqp = gdqp = NULL;
85
86 /*
87 * If disk quotas is on, we make sure that the dquots do exist on disk,
88 * before we start any other transactions. Trying to do this later
89 * is messy. We don't care to take a readlock to look at the ids
90 * in inode here, because we can't hold it across the trans_reserve.
91 * If the IDs do change before we take the ilock, we're covered
92 * because the i_*dquot fields will get updated anyway.
93 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +100094 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 uint qflags = 0;
96
Christoph Hellwig0f285c82008-07-18 17:13:28 +100097 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
98 uid = iattr->ia_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 qflags |= XFS_QMOPT_UQUOTA;
100 } else {
101 uid = ip->i_d.di_uid;
102 }
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000103 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
104 gid = iattr->ia_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 qflags |= XFS_QMOPT_GQUOTA;
106 } else {
107 gid = ip->i_d.di_gid;
108 }
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 /*
111 * We take a reference when we initialize udqp and gdqp,
112 * so it is important that we never blindly double trip on
113 * the same variable. See xfs_create() for an example.
114 */
115 ASSERT(udqp == NULL);
116 ASSERT(gdqp == NULL);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000117 code = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000118 qflags, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100120 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 }
122
123 /*
124 * For the other attributes, we acquire the inode lock and
125 * first do an error checking pass.
126 */
127 tp = NULL;
128 lock_flags = XFS_ILOCK_EXCL;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000129 if (flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700130 need_iolock = 0;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000131 if (!(mask & ATTR_SIZE)) {
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000132 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
133 commit_flags = 0;
134 code = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp),
135 0, 0, 0);
136 if (code) {
137 lock_flags = 0;
138 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 }
140 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (need_iolock)
142 lock_flags |= XFS_IOLOCK_EXCL;
143 }
144
145 xfs_ilock(ip, lock_flags);
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 /*
148 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000150 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 /*
152 * These IDs could have changed since we last looked at them.
153 * But, we're assured that if the ownership did change
154 * while we didn't have the inode locked, inode's dquot(s)
155 * would have changed also.
156 */
157 iuid = ip->i_d.di_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 igid = ip->i_d.di_gid;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000159 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
160 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 /*
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000163 * Do a quota reservation only if uid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * going to change.
165 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200166 if (XFS_IS_QUOTA_RUNNING(mp) &&
167 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
168 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 ASSERT(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200170 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 capable(CAP_FOWNER) ?
172 XFS_QMOPT_FORCE_RES : 0);
173 if (code) /* out of quota */
174 goto error_return;
175 }
176 }
177
178 /*
179 * Truncate file. Must have write permission and not be a directory.
180 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000181 if (mask & ATTR_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 /* Short circuit the truncate case for zero length files */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000183 if (iattr->ia_size == 0 &&
184 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
186 lock_flags &= ~XFS_ILOCK_EXCL;
Dave Chinnerdcd79a12010-09-28 12:27:25 +1000187 if (mask & ATTR_CTIME) {
188 inode->i_mtime = inode->i_ctime =
189 current_fs_time(inode->i_sb);
190 xfs_mark_inode_dirty_sync(ip);
191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 code = 0;
193 goto error_return;
194 }
195
Christoph Hellwig42173f62008-04-22 17:33:25 +1000196 if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 code = XFS_ERROR(EISDIR);
198 goto error_return;
Christoph Hellwig42173f62008-04-22 17:33:25 +1000199 } else if (!S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 code = XFS_ERROR(EINVAL);
201 goto error_return;
202 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 /*
205 * Make sure that the dquots are attached to the inode.
206 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200207 code = xfs_qm_dqattach_locked(ip, 0);
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500208 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500211 /*
212 * Now we can make the changes. Before we join the inode
213 * to the transaction, if ATTR_SIZE is set then take care of
214 * the part of the truncation that must be done without the
215 * inode lock. This needs to be done before joining the inode
216 * to the transaction, because the inode cannot be unlocked
217 * once it is a part of the transaction.
218 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000219 if (iattr->ia_size > ip->i_size) {
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000220 /*
221 * Do the first part of growing a file: zero any data
222 * in the last block that is beyond the old EOF. We
223 * need to do this before the inode is joined to the
224 * transaction to modify the i_size.
225 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000226 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400227 if (code)
228 goto error_return;
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400231 lock_flags &= ~XFS_ILOCK_EXCL;
David Chinnerc32676e2007-07-19 16:28:58 +1000232
233 /*
234 * We are going to log the inode size change in this
235 * transaction so any previous writes that are beyond the on
236 * disk EOF and the new EOF that have not been written out need
237 * to be written here. If we do not write the data out, we
238 * expose ourselves to the null files problem.
239 *
240 * Only flush from the on disk size to the smaller of the in
241 * memory file size or the new size as that's the range we
242 * really care about here and prevents waiting for other data
243 * not within the range we care about here.
244 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400245 if (ip->i_size != ip->i_d.di_size &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000246 iattr->ia_size > ip->i_d.di_size) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000247 code = xfs_flush_pages(ip,
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000248 ip->i_d.di_size, iattr->ia_size,
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000249 XBF_ASYNC, FI_NONE);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400250 if (code)
251 goto error_return;
David Chinnerc32676e2007-07-19 16:28:58 +1000252 }
253
254 /* wait for all I/O to complete */
Christoph Hellwig25e41b32008-12-03 12:20:39 +0100255 xfs_ioend_wait(ip);
David Chinnerc32676e2007-07-19 16:28:58 +1000256
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400257 code = -block_truncate_page(inode->i_mapping, iattr->ia_size,
258 xfs_get_blocks);
259 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 goto error_return;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400263 code = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
264 XFS_TRANS_PERM_LOG_RES,
265 XFS_ITRUNCATE_LOG_COUNT);
266 if (code)
267 goto error_return;
268
269 truncate_setsize(inode, iattr->ia_size);
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400272 lock_flags |= XFS_ILOCK_EXCL;
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Christoph Hellwig898621d2010-06-24 11:36:58 +1000276 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
David Chinner44d814c2008-03-06 13:45:29 +1100278 /*
279 * Only change the c/mtime if we are changing the size
280 * or we are explicitly asked to change it. This handles
281 * the semantic difference between truncate() and ftruncate()
282 * as implemented in the VFS.
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000283 *
284 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
285 * is a special case where we need to update the times despite
286 * not having these flags set. For all other operations the
287 * VFS set these flags explicitly if it wants a timestamp
288 * update.
David Chinner44d814c2008-03-06 13:45:29 +1100289 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000290 if (iattr->ia_size != ip->i_size &&
291 (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
292 iattr->ia_ctime = iattr->ia_mtime =
293 current_fs_time(inode->i_sb);
294 mask |= ATTR_CTIME | ATTR_MTIME;
295 }
David Chinner44d814c2008-03-06 13:45:29 +1100296
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000297 if (iattr->ia_size > ip->i_size) {
298 ip->i_d.di_size = iattr->ia_size;
299 ip->i_size = iattr->ia_size;
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000300 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000301 } else if (iattr->ia_size <= ip->i_size ||
302 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 /*
304 * signal a sync transaction unless
305 * we're truncating an already unlinked
306 * file on a wsync filesystem
307 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000308 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 XFS_DATA_FORK,
310 ((ip->i_d.di_nlink != 0 ||
311 !(mp->m_flags & XFS_MOUNT_WSYNC))
312 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000313 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000315 /*
316 * Truncated "down", so we're removing references
317 * to old data here - if we now delay flushing for
318 * a long time, we expose ourselves unduly to the
319 * notorious NULL files problem. So, we mark this
320 * vnode and flush it when the file is closed, and
321 * do not wait the usual (long) time for writeout.
322 */
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000323 xfs_iflags_set(ip, XFS_ITRUNCATED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500325 } else if (tp) {
Christoph Hellwig898621d2010-06-24 11:36:58 +1000326 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328
329 /*
330 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000332 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /*
334 * CAP_FSETID overrides the following restrictions:
335 *
336 * The set-user-ID and set-group-ID bits of a file will be
337 * cleared upon successful return from chown()
338 */
339 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
340 !capable(CAP_FSETID)) {
341 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
342 }
343
344 /*
345 * Change the ownerships and register quota modifications
346 * in the transaction.
347 */
348 if (iuid != uid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200349 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000350 ASSERT(mask & ATTR_UID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 ASSERT(udqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200352 olddquot1 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 &ip->i_udquot, udqp);
354 }
355 ip->i_d.di_uid = uid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000356 inode->i_uid = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
358 if (igid != gid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200359 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000360 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000361 ASSERT(mask & ATTR_GID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 ASSERT(gdqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200363 olddquot2 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 &ip->i_gdquot, gdqp);
365 }
366 ip->i_d.di_gid = gid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000367 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500371 /*
372 * Change file access modes.
373 */
374 if (mask & ATTR_MODE) {
375 umode_t mode = iattr->ia_mode;
376
377 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
378 mode &= ~S_ISGID;
379
380 ip->i_d.di_mode &= S_IFMT;
381 ip->i_d.di_mode |= mode & ~S_IFMT;
382
383 inode->i_mode &= S_IFMT;
384 inode->i_mode |= mode & ~S_IFMT;
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 /*
388 * Change file access or modified times.
389 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000390 if (mask & ATTR_ATIME) {
391 inode->i_atime = iattr->ia_atime;
392 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
393 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
394 ip->i_update_core = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000396 if (mask & ATTR_CTIME) {
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000397 inode->i_ctime = iattr->ia_ctime;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000398 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
399 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 ip->i_update_core = 1;
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000401 }
402 if (mask & ATTR_MTIME) {
403 inode->i_mtime = iattr->ia_mtime;
404 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
405 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
406 ip->i_update_core = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408
409 /*
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000410 * And finally, log the inode core if any attribute in it
411 * has been changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000413 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE|
414 ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
415 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 XFS_STATS_INC(xs_ig_attrchg);
418
419 /*
420 * If this is a synchronous mount, make sure that the
421 * transaction goes to disk before returning to the user.
422 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000423 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 * One for the truncate and one for the timestamps since we
425 * don't want to change the timestamps unless we're sure the
426 * truncate worked. Truncates are less than 1% of the laddis
427 * mix so this probably isn't worth the trouble to optimize.
428 */
429 code = 0;
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000430 if (mp->m_flags & XFS_MOUNT_WSYNC)
431 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000433 code = xfs_trans_commit(tp, commit_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 xfs_iunlock(ip, lock_flags);
436
437 /*
438 * Release any dquot(s) the inode had kept before chown.
439 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200440 xfs_qm_dqrele(olddquot1);
441 xfs_qm_dqrele(olddquot2);
442 xfs_qm_dqrele(udqp);
443 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200445 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 return code;
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200447
448 /*
449 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
450 * update. We could avoid this with linked transactions
451 * and passing down the transaction pointer all the way
452 * to attr_set. No previous user of the generic
453 * Posix ACL code seems to care about this issue either.
454 */
455 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
456 code = -xfs_acl_chmod(inode);
457 if (code)
458 return XFS_ERROR(code);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 return 0;
462
463 abort_return:
464 commit_flags |= XFS_TRANS_ABORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 error_return:
Christoph Hellwig7d095252009-06-08 15:33:32 +0200466 xfs_qm_dqrele(udqp);
467 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (tp) {
469 xfs_trans_cancel(tp, commit_flags);
470 }
471 if (lock_flags != 0) {
472 xfs_iunlock(ip, lock_flags);
473 }
474 return code;
475}
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000478 * The maximum pathlen is 1024 bytes. Since the minimum file system
479 * blocksize is 512 bytes, we can get a max of 2 extents back from
480 * bmapi.
481 */
482#define SYMLINK_MAPS 2
483
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000484STATIC int
485xfs_readlink_bmap(
486 xfs_inode_t *ip,
487 char *link)
488{
489 xfs_mount_t *mp = ip->i_mount;
490 int pathlen = ip->i_d.di_size;
491 int nmaps = SYMLINK_MAPS;
492 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
493 xfs_daddr_t d;
494 int byte_cnt;
495 int n;
496 xfs_buf_t *bp;
497 int error = 0;
498
499 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000500 mval, &nmaps, NULL);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000501 if (error)
502 goto out;
503
504 for (n = 0; n < nmaps; n++) {
505 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
506 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
507
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000508 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
509 XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000510 error = XFS_BUF_GETERROR(bp);
511 if (error) {
512 xfs_ioerror_alert("xfs_readlink",
513 ip->i_mount, bp, XFS_BUF_ADDR(bp));
514 xfs_buf_relse(bp);
515 goto out;
516 }
517 if (pathlen < byte_cnt)
518 byte_cnt = pathlen;
519 pathlen -= byte_cnt;
520
521 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
522 xfs_buf_relse(bp);
523 }
524
525 link[ip->i_d.di_size] = '\0';
526 error = 0;
527
528 out:
529 return error;
530}
531
Christoph Hellwig993386c2007-08-28 16:12:30 +1000532int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000534 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000535 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000537 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 int pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000541 trace_xfs_readlink(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
543 if (XFS_FORCED_SHUTDOWN(mp))
544 return XFS_ERROR(EIO);
545
546 xfs_ilock(ip, XFS_ILOCK_SHARED);
547
548 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000549 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000551 pathlen = ip->i_d.di_size;
552 if (!pathlen)
553 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000556 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
557 link[pathlen] = '\0';
558 } else {
559 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
561
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000562 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return error;
565}
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000568 * Flags for xfs_free_eofblocks
569 */
570#define XFS_FREE_EOF_TRYLOCK (1<<0)
571
572/*
David Chinner92dfe8d2007-05-24 15:22:19 +1000573 * This is called by xfs_inactive to free any blocks beyond eof
574 * when the link count isn't zero and by xfs_dm_punch_hole() when
575 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 */
Eric Sandeend96f8f82009-07-02 00:09:33 -0500577STATIC int
David Chinner92dfe8d2007-05-24 15:22:19 +1000578xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +1000580 xfs_inode_t *ip,
581 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
583 xfs_trans_t *tp;
584 int error;
585 xfs_fileoff_t end_fsb;
586 xfs_fileoff_t last_fsb;
587 xfs_filblks_t map_len;
588 int nimaps;
589 xfs_bmbt_irec_t imap;
590
591 /*
592 * Figure out if there are any blocks beyond the end
593 * of the file. If not, then there is nothing to do.
594 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000595 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000597 if (last_fsb <= end_fsb)
Jesper Juhl014c2542006-01-15 02:37:08 +0100598 return 0;
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000599 map_len = last_fsb - end_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 nimaps = 1;
602 xfs_ilock(ip, XFS_ILOCK_SHARED);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000603 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000604 NULL, 0, &imap, &nimaps, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 xfs_iunlock(ip, XFS_ILOCK_SHARED);
606
607 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +1100608 (imap.br_startblock != HOLESTARTBLOCK ||
609 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 /*
611 * Attach the dquots to the inode up front.
612 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200613 error = xfs_qm_dqattach(ip, 0);
614 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100615 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 /*
618 * There are blocks after the end of file.
619 * Free them up now by truncating the file to
620 * its current size.
621 */
622 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
623
624 /*
625 * Do the xfs_itruncate_start() call before
626 * reserving any log space because
627 * itruncate_start will call into the buffer
628 * cache and we can't
629 * do that within a transaction.
630 */
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000631 if (flags & XFS_FREE_EOF_TRYLOCK) {
632 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
633 xfs_trans_cancel(tp, 0);
634 return 0;
635 }
636 } else {
David Chinner92dfe8d2007-05-24 15:22:19 +1000637 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000638 }
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000639 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000640 ip->i_size);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000641 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +1000642 xfs_trans_cancel(tp, 0);
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000643 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000644 return error;
645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 error = xfs_trans_reserve(tp, 0,
648 XFS_ITRUNCATE_LOG_RES(mp),
649 0, XFS_TRANS_PERM_LOG_RES,
650 XFS_ITRUNCATE_LOG_COUNT);
651 if (error) {
652 ASSERT(XFS_FORCED_SHUTDOWN(mp));
653 xfs_trans_cancel(tp, 0);
654 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100655 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 }
657
658 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +1000659 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 error = xfs_itruncate_finish(&tp, ip,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000662 ip->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 XFS_DATA_FORK,
664 0);
665 /*
666 * If we get an error at this point we
667 * simply don't bother truncating the file.
668 */
669 if (error) {
670 xfs_trans_cancel(tp,
671 (XFS_TRANS_RELEASE_LOG_RES |
672 XFS_TRANS_ABORT));
673 } else {
674 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000675 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000677 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100679 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680}
681
682/*
683 * Free a symlink that has blocks associated with it.
684 */
685STATIC int
686xfs_inactive_symlink_rmt(
687 xfs_inode_t *ip,
688 xfs_trans_t **tpp)
689{
690 xfs_buf_t *bp;
691 int committed;
692 int done;
693 int error;
694 xfs_fsblock_t first_block;
695 xfs_bmap_free_t free_list;
696 int i;
697 xfs_mount_t *mp;
698 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
699 int nmaps;
700 xfs_trans_t *ntp;
701 int size;
702 xfs_trans_t *tp;
703
704 tp = *tpp;
705 mp = ip->i_mount;
706 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
707 /*
708 * We're freeing a symlink that has some
709 * blocks allocated to it. Free the
710 * blocks here. We know that we've got
711 * either 1 or 2 extents and that we can
712 * free them all in one bunmapi call.
713 */
714 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
715 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
716 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
717 ASSERT(XFS_FORCED_SHUTDOWN(mp));
718 xfs_trans_cancel(tp, 0);
719 *tpp = NULL;
720 return error;
721 }
722 /*
723 * Lock the inode, fix the size, and join it to the transaction.
724 * Hold it so in the normal path, we still have it locked for
725 * the second transaction. In the error paths we need it
726 * held so the cancel won't rele it, see below.
727 */
728 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
729 size = (int)ip->i_d.di_size;
730 ip->i_d.di_size = 0;
Christoph Hellwig898621d2010-06-24 11:36:58 +1000731 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
733 /*
734 * Find the block(s) so we can inval and unmap them.
735 */
736 done = 0;
Eric Sandeen9d87c312009-01-14 23:22:07 -0600737 xfs_bmap_init(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -0800738 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
740 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000741 &free_list)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 goto error0;
743 /*
744 * Invalidate the block(s).
745 */
746 for (i = 0; i < nmaps; i++) {
747 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
748 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
749 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
750 xfs_trans_binval(tp, bp);
751 }
752 /*
753 * Unmap the dead block(s) to the free_list.
754 */
755 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000756 &first_block, &free_list, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 goto error1;
758 ASSERT(done);
759 /*
760 * Commit the first transaction. This logs the EFI and the inode.
761 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100762 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 goto error1;
764 /*
765 * The transaction must have been committed, since there were
766 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
767 * The new tp has the extent freeing and EFDs.
768 */
769 ASSERT(committed);
770 /*
771 * The first xact was committed, so add the inode to the new one.
772 * Mark it dirty so it will be logged and moved forward in the log as
773 * part of every commit.
774 */
Christoph Hellwig898621d2010-06-24 11:36:58 +1000775 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
777 /*
778 * Get a new, empty transaction to return to our caller.
779 */
780 ntp = xfs_trans_dup(tp);
781 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000782 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * If we get an error on the commit here or on the reserve below,
784 * we need to unlock the inode since the new transaction doesn't
785 * have the inode attached.
786 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000787 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 tp = ntp;
789 if (error) {
790 ASSERT(XFS_FORCED_SHUTDOWN(mp));
791 goto error0;
792 }
793 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100794 * transaction commit worked ok so we can drop the extra ticket
795 * reference that we gained in xfs_trans_dup()
796 */
797 xfs_log_ticket_put(tp->t_ticket);
798
799 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * Remove the memory for extent descriptions (just bookkeeping).
801 */
802 if (ip->i_df.if_bytes)
803 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
804 ASSERT(ip->i_df.if_bytes == 0);
805 /*
806 * Put an itruncate log reservation in the new transaction
807 * for our caller.
808 */
809 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
810 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
811 ASSERT(XFS_FORCED_SHUTDOWN(mp));
812 goto error0;
813 }
814 /*
815 * Return with the inode locked but not joined to the transaction.
816 */
817 *tpp = tp;
818 return 0;
819
820 error1:
821 xfs_bmap_cancel(&free_list);
822 error0:
823 /*
824 * Have to come here with the inode locked and either
825 * (held and in the transaction) or (not in the transaction).
826 * If the inode isn't held then cancel would iput it, but
827 * that's wrong since this is inactive and the vnode ref
828 * count is 0 already.
829 * Cancel won't do anything to the inode if held, but it still
830 * needs to be locked until the cancel is done, if it was
831 * joined to the transaction.
832 */
833 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
834 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
835 *tpp = NULL;
836 return error;
837
838}
839
840STATIC int
841xfs_inactive_symlink_local(
842 xfs_inode_t *ip,
843 xfs_trans_t **tpp)
844{
845 int error;
846
847 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
848 /*
849 * We're freeing a symlink which fit into
850 * the inode. Just free the memory used
851 * to hold the old symlink.
852 */
853 error = xfs_trans_reserve(*tpp, 0,
854 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
855 0, XFS_TRANS_PERM_LOG_RES,
856 XFS_ITRUNCATE_LOG_COUNT);
857
858 if (error) {
859 xfs_trans_cancel(*tpp, 0);
860 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +0100861 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
863 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
864
865 /*
866 * Zero length symlinks _can_ exist.
867 */
868 if (ip->i_df.if_bytes > 0) {
869 xfs_idata_realloc(ip,
870 -(ip->i_df.if_bytes),
871 XFS_DATA_FORK);
872 ASSERT(ip->i_df.if_bytes == 0);
873 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100874 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877STATIC int
878xfs_inactive_attrs(
879 xfs_inode_t *ip,
880 xfs_trans_t **tpp)
881{
882 xfs_trans_t *tp;
883 int error;
884 xfs_mount_t *mp;
885
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000886 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 tp = *tpp;
888 mp = ip->i_mount;
889 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +1000890 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +1000892 if (error)
893 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +1000896 if (error)
897 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
900 error = xfs_trans_reserve(tp, 0,
901 XFS_IFREE_LOG_RES(mp),
902 0, XFS_TRANS_PERM_LOG_RES,
903 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +1000904 if (error)
905 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +1000908 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
910
911 ASSERT(ip->i_d.di_anextents == 0);
912
913 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +0100914 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +1000915
916error_cancel:
917 ASSERT(XFS_FORCED_SHUTDOWN(mp));
918 xfs_trans_cancel(tp, 0);
919error_unlock:
920 *tpp = NULL;
921 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
922 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
Christoph Hellwig993386c2007-08-28 16:12:30 +1000925int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000927 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
Christoph Hellwig993386c2007-08-28 16:12:30 +1000929 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 int error;
931
Christoph Hellwig42173f62008-04-22 17:33:25 +1000932 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000936 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return 0;
938
David Chinner2a82b8b2007-07-11 11:09:12 +1000939 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000940 int truncated;
941
David Chinner2a82b8b2007-07-11 11:09:12 +1000942 /*
943 * If we are using filestreams, and we have an unlinked
944 * file that we are processing the last close on, then nothing
945 * will be able to reopen and write to this file. Purge this
946 * inode from the filestreams cache so that it doesn't delay
947 * teardown of the inode.
948 */
949 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
950 xfs_filestream_deassociate(ip);
951
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000952 /*
953 * If we previously truncated this file and removed old data
954 * in the process, we want to initiate "early" writeout on
955 * the last close. This is an attempt to combat the notorious
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300956 * NULL files problem which is particularly noticeable from a
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000957 * truncate down, buffered (re-)write (delalloc), followed by
958 * a crash. What we are effectively doing here is
959 * significantly reducing the time window where we'd otherwise
960 * be exposed to that problem.
961 */
Christoph Hellwig09262b42007-08-29 11:44:50 +1000962 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Christoph Hellwigdf80c932008-08-13 16:22:09 +1000963 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000964 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000965 }
966
Dave Chinner6e857562010-12-23 12:02:31 +1100967 if (ip->i_d.di_nlink == 0)
968 return 0;
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000969
Dave Chinner6e857562010-12-23 12:02:31 +1100970 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
971 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
972 ip->i_delayed_blks > 0)) &&
973 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
974 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
975
976 /*
977 * If we can't get the iolock just skip truncating the blocks
978 * past EOF because we could deadlock with the mmap_sem
979 * otherwise. We'll get another chance to drop them once the
980 * last reference to the inode is dropped, so we'll never leak
981 * blocks permanently.
982 *
983 * Further, check if the inode is being opened, written and
984 * closed frequently and we have delayed allocation blocks
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300985 * outstanding (e.g. streaming writes from the NFS server),
Dave Chinner6e857562010-12-23 12:02:31 +1100986 * truncating the blocks past EOF will cause fragmentation to
987 * occur.
988 *
989 * In this case don't do the truncation, either, but we have to
990 * be careful how we detect this case. Blocks beyond EOF show
991 * up as i_delayed_blks even when the inode is clean, so we
992 * need to truncate them away first before checking for a dirty
993 * release. Hence on the first dirty close we will still remove
994 * the speculative allocation, but after that we will leave it
995 * in place.
996 */
997 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
998 return 0;
999
1000 error = xfs_free_eofblocks(mp, ip,
1001 XFS_FREE_EOF_TRYLOCK);
1002 if (error)
1003 return error;
1004
1005 /* delalloc blocks after truncation means it really is dirty */
1006 if (ip->i_delayed_blks)
1007 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 return 0;
1010}
1011
1012/*
1013 * xfs_inactive
1014 *
1015 * This is called when the vnode reference count for the vnode
1016 * goes to zero. If the file has been unlinked, then it must
1017 * now be truncated. Also, we clear all of the read-ahead state
1018 * kept for the inode here since the file is now closed.
1019 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001020int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001022 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Nathan Scott67fcaa72006-06-09 17:00:52 +10001024 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 xfs_fsblock_t first_block;
1026 int committed;
1027 xfs_trans_t *tp;
1028 xfs_mount_t *mp;
1029 int error;
1030 int truncate;
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 /*
1033 * If the inode is already free, then there can be nothing
1034 * to clean up here.
1035 */
Christoph Hellwigcb4c8cc2009-03-16 08:25:25 +01001036 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 ASSERT(ip->i_df.if_real_bytes == 0);
1038 ASSERT(ip->i_df.if_broot_bytes == 0);
1039 return VN_INACTIVE_CACHE;
1040 }
1041
1042 /*
1043 * Only do a truncate if it's a regular file with
1044 * some actual space in it. It's OK to look at the
1045 * inode's fields without the lock because we're the
1046 * only one with a reference to the inode.
1047 */
1048 truncate = ((ip->i_d.di_nlink == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001049 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1050 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1052
1053 mp = ip->i_mount;
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 error = 0;
1056
1057 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001058 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto out;
1060
1061 if (ip->i_d.di_nlink != 0) {
1062 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001063 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001064 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001065 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1066 (!(ip->i_d.di_flags &
1067 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1068 (ip->i_delayed_blks != 0)))) {
Christoph Hellwigc56c9632009-10-19 04:03:46 +00001069 error = xfs_free_eofblocks(mp, ip, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +10001070 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001071 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073 goto out;
1074 }
1075
1076 ASSERT(ip->i_d.di_nlink == 0);
1077
Christoph Hellwig7d095252009-06-08 15:33:32 +02001078 error = xfs_qm_dqattach(ip, 0);
1079 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001080 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
1082 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1083 if (truncate) {
1084 /*
1085 * Do the xfs_itruncate_start() call before
1086 * reserving any log space because itruncate_start
1087 * will call into the buffer cache and we can't
1088 * do that within a transaction.
1089 */
1090 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1091
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001092 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1093 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001094 xfs_trans_cancel(tp, 0);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001095 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1096 return VN_INACTIVE_CACHE;
1097 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
1099 error = xfs_trans_reserve(tp, 0,
1100 XFS_ITRUNCATE_LOG_RES(mp),
1101 0, XFS_TRANS_PERM_LOG_RES,
1102 XFS_ITRUNCATE_LOG_COUNT);
1103 if (error) {
1104 /* Don't call itruncate_cleanup */
1105 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1106 xfs_trans_cancel(tp, 0);
1107 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001108 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 }
1110
1111 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001112 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 /*
1115 * normally, we have to run xfs_itruncate_finish sync.
1116 * But if filesystem is wsync and we're in the inactive
1117 * path, then we know that nlink == 0, and that the
1118 * xaction that made nlink == 0 is permanently committed
1119 * since xfs_remove runs as a synchronous transaction.
1120 */
1121 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1122 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1123
1124 if (error) {
1125 xfs_trans_cancel(tp,
1126 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1127 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001128 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1131
1132 /*
1133 * If we get an error while cleaning up a
1134 * symlink we bail out.
1135 */
1136 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1137 xfs_inactive_symlink_rmt(ip, &tp) :
1138 xfs_inactive_symlink_local(ip, &tp);
1139
1140 if (error) {
1141 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001142 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 }
1144
Christoph Hellwig898621d2010-06-24 11:36:58 +10001145 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 } else {
1147 error = xfs_trans_reserve(tp, 0,
1148 XFS_IFREE_LOG_RES(mp),
1149 0, XFS_TRANS_PERM_LOG_RES,
1150 XFS_INACTIVE_LOG_COUNT);
1151 if (error) {
1152 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1153 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001154 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
1156
1157 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001158 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
1160
1161 /*
1162 * If there are attributes associated with the file
1163 * then blow them away now. The code calls a routine
1164 * that recursively deconstructs the attribute fork.
1165 * We need to just commit the current transaction
1166 * because we can't use it for xfs_attr_inactive().
1167 */
1168 if (ip->i_d.di_anextents > 0) {
1169 error = xfs_inactive_attrs(ip, &tp);
1170 /*
1171 * If we got an error, the transaction is already
1172 * cancelled, and the inode is unlocked. Just get out.
1173 */
1174 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001175 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 } else if (ip->i_afp) {
1177 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1178 }
1179
1180 /*
1181 * Free the inode.
1182 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001183 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 error = xfs_ifree(tp, ip, &free_list);
1185 if (error) {
1186 /*
1187 * If we fail to free the inode, shut down. The cancel
1188 * might do that, we need to make sure. Otherwise the
1189 * inode might be lost for a long time or forever.
1190 */
1191 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001192 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1193 __func__, error);
Nathan Scott7d04a332006-06-09 14:58:38 +10001194 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
1196 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1197 } else {
1198 /*
1199 * Credit the quota account(s). The inode is gone.
1200 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001201 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
1203 /*
David Chinner78e9da72008-04-10 12:24:17 +10001204 * Just ignore errors at this point. There is nothing we can
1205 * do except to try to keep going. Make sure it's not a silent
1206 * error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 */
David Chinner78e9da72008-04-10 12:24:17 +10001208 error = xfs_bmap_finish(&tp, &free_list, &committed);
1209 if (error)
Dave Chinner53487782011-03-07 10:05:35 +11001210 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1211 __func__, error);
David Chinner78e9da72008-04-10 12:24:17 +10001212 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1213 if (error)
Dave Chinner53487782011-03-07 10:05:35 +11001214 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1215 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
Christoph Hellwig7d095252009-06-08 15:33:32 +02001217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 /*
1219 * Release the dquots held by inode, if any.
1220 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001221 xfs_qm_dqdetach(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1223
1224 out:
1225 return VN_INACTIVE_CACHE;
1226}
1227
Barry Naujok384f3ce2008-05-21 16:58:22 +10001228/*
1229 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1230 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1231 * ci_name->name will point to a the actual name (caller must free) or
1232 * will be set to NULL if an exact match is found.
1233 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001234int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001236 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001237 struct xfs_name *name,
Barry Naujok384f3ce2008-05-21 16:58:22 +10001238 xfs_inode_t **ipp,
1239 struct xfs_name *ci_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001241 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 int error;
1243 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001245 trace_xfs_lookup(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1248 return XFS_ERROR(EIO);
1249
1250 lock_mode = xfs_ilock_map_shared(dp);
Barry Naujok384f3ce2008-05-21 16:58:22 +10001251 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 xfs_iunlock_map_shared(dp, lock_mode);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001253
1254 if (error)
1255 goto out;
1256
Dave Chinner7b6259e2010-06-24 11:35:17 +10001257 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001258 if (error)
Barry Naujok384f3ce2008-05-21 16:58:22 +10001259 goto out_free_name;
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001260
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001261 return 0;
1262
Barry Naujok384f3ce2008-05-21 16:58:22 +10001263out_free_name:
1264 if (ci_name)
1265 kmem_free(ci_name->name);
1266out:
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001267 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return error;
1269}
1270
Christoph Hellwig993386c2007-08-28 16:12:30 +10001271int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001273 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001274 struct xfs_name *name,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001275 mode_t mode,
1276 xfs_dev_t rdev,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001277 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001279 int is_dir = S_ISDIR(mode);
1280 struct xfs_mount *mp = dp->i_mount;
1281 struct xfs_inode *ip = NULL;
1282 struct xfs_trans *tp = NULL;
Barry Naujok556b8b12008-04-10 12:22:07 +10001283 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 xfs_bmap_free_t free_list;
1285 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001286 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 uint cancel_flags;
1288 int committed;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001289 prid_t prid;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001290 struct xfs_dquot *udqp = NULL;
1291 struct xfs_dquot *gdqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 uint resblks;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001293 uint log_res;
1294 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001296 trace_xfs_create(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001298 if (XFS_FORCED_SHUTDOWN(mp))
1299 return XFS_ERROR(EIO);
1300
Nathan Scott365ca832005-06-21 15:39:12 +10001301 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001302 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001304 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306 /*
1307 * Make sure that we have allocated dquot(s) on disk.
1308 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001309 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001310 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001312 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001314 if (is_dir) {
1315 rdev = 0;
1316 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1317 log_res = XFS_MKDIR_LOG_RES(mp);
1318 log_count = XFS_MKDIR_LOG_COUNT;
1319 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1320 } else {
1321 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1322 log_res = XFS_CREATE_LOG_RES(mp);
1323 log_count = XFS_CREATE_LOG_COUNT;
1324 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 /*
1330 * Initially assume that the file does not exist and
1331 * reserve the resources for that case. If that is not
1332 * the case we'll drop the one we have and get a more
1333 * appropriate transaction later.
1334 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001335 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1336 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 if (error == ENOSPC) {
Dave Chinner153fec42009-04-06 18:48:30 +02001338 /* flush outstanding delalloc blocks and retry */
1339 xfs_flush_inodes(dp);
Christoph Hellwig4734d402009-09-09 18:19:02 -05001340 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1341 XFS_TRANS_PERM_LOG_RES, log_count);
Dave Chinner153fec42009-04-06 18:48:30 +02001342 }
1343 if (error == ENOSPC) {
1344 /* No space at all so try a "no-allocation" reservation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 resblks = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001346 error = xfs_trans_reserve(tp, 0, log_res, 0,
1347 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349 if (error) {
1350 cancel_flags = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001351 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001354 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001355 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001357 /*
1358 * Check for directory link count overflow.
1359 */
1360 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1361 error = XFS_ERROR(EMLINK);
1362 goto out_trans_cancel;
1363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001365 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 /*
1368 * Reserve disk quota and the inode.
1369 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001370 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001372 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Barry Naujok556b8b12008-04-10 12:22:07 +10001374 error = xfs_dir_canenter(tp, dp, name, resblks);
1375 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001376 goto out_trans_cancel;
1377
1378 /*
1379 * A newly created regular or special file just has one directory
1380 * entry pointing to them, but a directory also the "." entry
1381 * pointing to itself.
1382 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001383 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001384 prid, resblks > 0, &ip, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 if (error) {
1386 if (error == ENOSPC)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001387 goto out_trans_cancel;
1388 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +10001392 * Now we join the directory inode to the transaction. We do not do it
1393 * earlier because xfs_dir_ialloc might commit the previous transaction
1394 * (and release all the locks). An error from here on will result in
1395 * the transaction cancel unlocking dp so don't do it explicitly in the
1396 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 */
Christoph Hellwig898621d2010-06-24 11:36:58 +10001398 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001399 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Barry Naujok556b8b12008-04-10 12:22:07 +10001401 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001402 &first_block, &free_list, resblks ?
1403 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (error) {
1405 ASSERT(error != ENOSPC);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001406 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001408 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1410
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001411 if (is_dir) {
1412 error = xfs_dir_init(tp, ip, dp);
1413 if (error)
1414 goto out_bmap_cancel;
1415
1416 error = xfs_bumplink(tp, dp);
1417 if (error)
1418 goto out_bmap_cancel;
1419 }
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 /*
1422 * If this is a synchronous mount, make sure that the
1423 * create transaction goes to disk before returning to
1424 * the user.
1425 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001426 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 /*
1430 * Attach the dquot(s) to the inodes and modify them incore.
1431 * These ids of the inode couldn't have changed since the new
1432 * inode has been locked ever since it was created.
1433 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001434 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001436 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001437 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001438 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001440 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001441 if (error)
1442 goto out_release_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Christoph Hellwig7d095252009-06-08 15:33:32 +02001444 xfs_qm_dqrele(udqp);
1445 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001447 *ipp = ip;
Christoph Hellwig288699f2010-06-23 18:11:15 +10001448 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001450 out_bmap_cancel:
1451 xfs_bmap_cancel(&free_list);
1452 out_trans_abort:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001454 out_trans_cancel:
1455 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001456 out_release_inode:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 /*
1458 * Wait until after the current transaction is aborted to
1459 * release the inode. This prevents recursive transactions
1460 * and deadlocks from xfs_inactive.
1461 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001462 if (ip)
1463 IRELE(ip);
1464
1465 xfs_qm_dqrele(udqp);
1466 xfs_qm_dqrele(gdqp);
1467
1468 if (unlock_dp_on_error)
1469 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1470 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
1472
1473#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474int xfs_locked_n;
1475int xfs_small_retries;
1476int xfs_middle_retries;
1477int xfs_lots_retries;
1478int xfs_lock_delays;
1479#endif
1480
1481/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001482 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1483 * a different value
1484 */
1485static inline int
1486xfs_lock_inumorder(int lock_mode, int subclass)
1487{
1488 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001489 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001490 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001491 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001492
1493 return lock_mode;
1494}
1495
1496/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 * The following routine will lock n inodes in exclusive mode.
1498 * We assume the caller calls us with the inodes in i_ino order.
1499 *
1500 * We need to detect deadlock where an inode that we lock
1501 * is in the AIL and we start waiting for another inode that is locked
1502 * by a thread in a long running transaction (such as truncate). This can
1503 * result in deadlock since the long running trans might need to wait
1504 * for the inode we just locked in order to push the tail and free space
1505 * in the log.
1506 */
1507void
1508xfs_lock_inodes(
1509 xfs_inode_t **ips,
1510 int inodes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 uint lock_mode)
1512{
1513 int attempts = 0, i, j, try_lock;
1514 xfs_log_item_t *lp;
1515
1516 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1517
Christoph Hellwigcfa853e2008-04-22 17:34:06 +10001518 try_lock = 0;
1519 i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521again:
1522 for (; i < inodes; i++) {
1523 ASSERT(ips[i]);
1524
1525 if (i && (ips[i] == ips[i-1])) /* Already locked */
1526 continue;
1527
1528 /*
1529 * If try_lock is not set yet, make sure all locked inodes
1530 * are not in the AIL.
1531 * If any are, set try_lock to be used later.
1532 */
1533
1534 if (!try_lock) {
1535 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1536 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1537 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1538 try_lock++;
1539 }
1540 }
1541 }
1542
1543 /*
1544 * If any of the previous locks we have locked is in the AIL,
1545 * we must TRY to get the second and subsequent locks. If
1546 * we can't get any, we must release all we have
1547 * and try again.
1548 */
1549
1550 if (try_lock) {
1551 /* try_lock must be 0 if i is 0. */
1552 /*
1553 * try_lock means we have an inode locked
1554 * that is in the AIL.
1555 */
1556 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001557 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 attempts++;
1559
1560 /*
1561 * Unlock all previous guys and try again.
1562 * xfs_iunlock will try to push the tail
1563 * if the inode is in the AIL.
1564 */
1565
1566 for(j = i - 1; j >= 0; j--) {
1567
1568 /*
1569 * Check to see if we've already
1570 * unlocked this one.
1571 * Not the first one going back,
1572 * and the inode ptr is the same.
1573 */
1574 if ((j != (i - 1)) && ips[j] ==
1575 ips[j+1])
1576 continue;
1577
1578 xfs_iunlock(ips[j], lock_mode);
1579 }
1580
1581 if ((attempts % 5) == 0) {
1582 delay(1); /* Don't just spin the CPU */
1583#ifdef DEBUG
1584 xfs_lock_delays++;
1585#endif
1586 }
1587 i = 0;
1588 try_lock = 0;
1589 goto again;
1590 }
1591 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001592 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 }
1594 }
1595
1596#ifdef DEBUG
1597 if (attempts) {
1598 if (attempts < 5) xfs_small_retries++;
1599 else if (attempts < 100) xfs_middle_retries++;
1600 else xfs_lots_retries++;
1601 } else {
1602 xfs_locked_n++;
1603 }
1604#endif
1605}
1606
David Chinnerf9114eb2008-09-17 16:51:21 +10001607/*
1608 * xfs_lock_two_inodes() can only be used to lock one type of lock
1609 * at a time - the iolock or the ilock, but not both at once. If
1610 * we lock both at once, lockdep will report false positives saying
1611 * we have violated locking orders.
1612 */
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001613void
1614xfs_lock_two_inodes(
1615 xfs_inode_t *ip0,
1616 xfs_inode_t *ip1,
1617 uint lock_mode)
1618{
1619 xfs_inode_t *temp;
1620 int attempts = 0;
1621 xfs_log_item_t *lp;
1622
David Chinnerf9114eb2008-09-17 16:51:21 +10001623 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1624 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001625 ASSERT(ip0->i_ino != ip1->i_ino);
1626
1627 if (ip0->i_ino > ip1->i_ino) {
1628 temp = ip0;
1629 ip0 = ip1;
1630 ip1 = temp;
1631 }
1632
1633 again:
1634 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1635
1636 /*
1637 * If the first lock we have locked is in the AIL, we must TRY to get
1638 * the second lock. If we can't get it, we must release the first one
1639 * and try again.
1640 */
1641 lp = (xfs_log_item_t *)ip0->i_itemp;
1642 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1643 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1644 xfs_iunlock(ip0, lock_mode);
1645 if ((++attempts % 5) == 0)
1646 delay(1); /* Don't just spin the CPU */
1647 goto again;
1648 }
1649 } else {
1650 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1651 }
1652}
1653
Christoph Hellwig993386c2007-08-28 16:12:30 +10001654int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001656 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001657 struct xfs_name *name,
1658 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001660 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 xfs_trans_t *tp = NULL;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001662 int is_dir = S_ISDIR(ip->i_d.di_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 int error = 0;
1664 xfs_bmap_free_t free_list;
1665 xfs_fsblock_t first_block;
1666 int cancel_flags;
1667 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 int link_zero;
1669 uint resblks;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001670 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001672 trace_xfs_remove(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (XFS_FORCED_SHUTDOWN(mp))
1675 return XFS_ERROR(EIO);
1676
Christoph Hellwig7d095252009-06-08 15:33:32 +02001677 error = xfs_qm_dqattach(dp, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001678 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Christoph Hellwig7d095252009-06-08 15:33:32 +02001681 error = xfs_qm_dqattach(ip, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001682 if (error)
1683 goto std_return;
1684
1685 if (is_dir) {
1686 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1687 log_count = XFS_DEFAULT_LOG_COUNT;
1688 } else {
1689 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1690 log_count = XFS_REMOVE_LOG_COUNT;
1691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001693
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 /*
1695 * We try to get the real space reservation first,
1696 * allowing for directory btree deletion(s) implying
1697 * possible bmap insert(s). If we can't get the space
1698 * reservation then we use 0 instead, and avoid the bmap
1699 * btree insert(s) in the directory code by, if the bmap
1700 * insert tries to happen, instead trimming the LAST
1701 * block from the directory.
1702 */
1703 resblks = XFS_REMOVE_SPACE_RES(mp);
1704 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001705 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 if (error == ENOSPC) {
1707 resblks = 0;
1708 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001709 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711 if (error) {
1712 ASSERT(error != ENOSPC);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001713 cancel_flags = 0;
1714 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
1716
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001717 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Christoph Hellwig898621d2010-06-24 11:36:58 +10001719 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1720 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 /*
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001723 * If we're removing a directory perform some additional validation.
1724 */
1725 if (is_dir) {
1726 ASSERT(ip->i_d.di_nlink >= 2);
1727 if (ip->i_d.di_nlink != 2) {
1728 error = XFS_ERROR(ENOTEMPTY);
1729 goto out_trans_cancel;
1730 }
1731 if (!xfs_dir_isempty(ip)) {
1732 error = XFS_ERROR(ENOTEMPTY);
1733 goto out_trans_cancel;
1734 }
1735 }
1736
Eric Sandeen9d87c312009-01-14 23:22:07 -06001737 xfs_bmap_init(&free_list, &first_block);
Barry Naujok556b8b12008-04-10 12:22:07 +10001738 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
Christoph Hellwigd4377d82008-04-22 17:33:46 +10001739 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 if (error) {
1741 ASSERT(error != ENOENT);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001742 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001744 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001746 if (is_dir) {
1747 /*
1748 * Drop the link from ip's "..".
1749 */
1750 error = xfs_droplink(tp, dp);
1751 if (error)
1752 goto out_bmap_cancel;
1753
1754 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001755 * Drop the "." link from ip to self.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001756 */
1757 error = xfs_droplink(tp, ip);
1758 if (error)
1759 goto out_bmap_cancel;
1760 } else {
1761 /*
1762 * When removing a non-directory we need to log the parent
Dave Chinner26c52952008-11-28 14:23:37 +11001763 * inode here. For a directory this is done implicitly
1764 * by the xfs_droplink call for the ".." entry.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001765 */
1766 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 }
1768
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001769 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001770 * Drop the link from dp to ip.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001771 */
1772 error = xfs_droplink(tp, ip);
1773 if (error)
1774 goto out_bmap_cancel;
1775
1776 /*
1777 * Determine if this is the last link while
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 * we are in the transaction.
1779 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001780 link_zero = (ip->i_d.di_nlink == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 * If this is a synchronous mount, make sure that the
1784 * remove transaction goes to disk before returning to
1785 * the user.
1786 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001787 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001790 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001791 if (error)
1792 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001794 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig5df78e72008-04-22 17:34:24 +10001795 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10001799 * If we are using filestreams, kill the stream association.
1800 * If the file is still open it may get a new one but that
1801 * will get killed on last close in xfs_close() so we don't
1802 * have to worry about that.
1803 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001804 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
David Chinner2a82b8b2007-07-11 11:09:12 +10001805 xfs_filestream_deassociate(ip);
1806
Christoph Hellwig288699f2010-06-23 18:11:15 +10001807 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001809 out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 xfs_bmap_cancel(&free_list);
1811 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001812 out_trans_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001814 std_return:
1815 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816}
1817
Christoph Hellwig993386c2007-08-28 16:12:30 +10001818int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001820 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001821 xfs_inode_t *sip,
Barry Naujok556b8b12008-04-10 12:22:07 +10001822 struct xfs_name *target_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001824 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 int error;
1827 xfs_bmap_free_t free_list;
1828 xfs_fsblock_t first_block;
1829 int cancel_flags;
1830 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 int resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001833 trace_xfs_link(tdp, target_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001835 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (XFS_FORCED_SHUTDOWN(mp))
1838 return XFS_ERROR(EIO);
1839
Christoph Hellwig7d095252009-06-08 15:33:32 +02001840 error = xfs_qm_dqattach(sip, 0);
Christoph Hellwigcb3f35b2009-02-04 09:34:20 +01001841 if (error)
1842 goto std_return;
1843
Christoph Hellwig7d095252009-06-08 15:33:32 +02001844 error = xfs_qm_dqattach(tdp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 if (error)
1846 goto std_return;
1847
1848 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1849 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10001850 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1852 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1853 if (error == ENOSPC) {
1854 resblks = 0;
1855 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1856 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1857 }
1858 if (error) {
1859 cancel_flags = 0;
1860 goto error_return;
1861 }
1862
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001863 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Christoph Hellwig898621d2010-06-24 11:36:58 +10001865 xfs_trans_ijoin_ref(tp, sip, XFS_ILOCK_EXCL);
1866 xfs_trans_ijoin_ref(tp, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
1868 /*
1869 * If the source has too many links, we can't make any more to it.
1870 */
1871 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
1872 error = XFS_ERROR(EMLINK);
1873 goto error_return;
1874 }
1875
Nathan Scott365ca832005-06-21 15:39:12 +10001876 /*
1877 * If we are using project inheritance, we only allow hard link
1878 * creation in our tree when the project IDs are the same; else
1879 * the tree quota mechanism could be circumvented.
1880 */
1881 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001882 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10001883 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10001884 goto error_return;
1885 }
1886
Barry Naujok556b8b12008-04-10 12:22:07 +10001887 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1888 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 goto error_return;
1890
Eric Sandeen9d87c312009-01-14 23:22:07 -06001891 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Barry Naujok556b8b12008-04-10 12:22:07 +10001893 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1894 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 if (error)
1896 goto abort_return;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001897 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1899
1900 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001901 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 /*
1905 * If this is a synchronous mount, make sure that the
1906 * link transaction goes to disk before returning to
1907 * the user.
1908 */
1909 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1910 xfs_trans_set_sync(tp);
1911 }
1912
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001913 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 if (error) {
1915 xfs_bmap_cancel(&free_list);
1916 goto abort_return;
1917 }
1918
Christoph Hellwig288699f2010-06-23 18:11:15 +10001919 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 abort_return:
1922 cancel_flags |= XFS_TRANS_ABORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 error_return:
1924 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001925 std_return:
1926 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001928
Christoph Hellwig993386c2007-08-28 16:12:30 +10001929int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001931 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001932 struct xfs_name *link_name,
1933 const char *target_path,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001934 mode_t mode,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001935 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001937 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 xfs_inode_t *ip;
1940 int error;
1941 int pathlen;
1942 xfs_bmap_free_t free_list;
1943 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001944 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 uint cancel_flags;
1946 int committed;
1947 xfs_fileoff_t first_fsb;
1948 xfs_filblks_t fs_blocks;
1949 int nmaps;
1950 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1951 xfs_daddr_t d;
Barry Naujok556b8b12008-04-10 12:22:07 +10001952 const char *cur_chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 int byte_cnt;
1954 int n;
1955 xfs_buf_t *bp;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001956 prid_t prid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 struct xfs_dquot *udqp, *gdqp;
1958 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Christoph Hellwig3937be52008-03-06 13:46:19 +11001960 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 error = 0;
1962 ip = NULL;
1963 tp = NULL;
1964
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001965 trace_xfs_symlink(dp, link_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 if (XFS_FORCED_SHUTDOWN(mp))
1968 return XFS_ERROR(EIO);
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 /*
1971 * Check component lengths of the target path name.
1972 */
1973 pathlen = strlen(target_path);
1974 if (pathlen >= MAXPATHLEN) /* total string too long */
1975 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001978 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001979 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001981 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983 /*
1984 * Make sure that we have allocated dquot(s) on disk.
1985 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001986 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1988 if (error)
1989 goto std_return;
1990
1991 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1992 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1993 /*
1994 * The symlink will fit into the inode data fork?
1995 * There can't be any attributes so we get the whole variable part.
1996 */
1997 if (pathlen <= XFS_LITINO(mp))
1998 fs_blocks = 0;
1999 else
2000 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
Barry Naujok556b8b12008-04-10 12:22:07 +10002001 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2003 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2004 if (error == ENOSPC && fs_blocks == 0) {
2005 resblks = 0;
2006 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2007 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2008 }
2009 if (error) {
2010 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 goto error_return;
2012 }
2013
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002014 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002015 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
2017 /*
2018 * Check whether the directory allows new symlinks or not.
2019 */
2020 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2021 error = XFS_ERROR(EPERM);
2022 goto error_return;
2023 }
2024
2025 /*
2026 * Reserve disk quota : blocks and inode.
2027 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002028 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (error)
2030 goto error_return;
2031
2032 /*
2033 * Check for ability to enter directory entry, if no space reserved.
2034 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002035 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2036 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 goto error_return;
2038 /*
2039 * Initialize the bmap freelist prior to calling either
2040 * bmapi or the directory create code.
2041 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002042 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 /*
2045 * Allocate an inode for the symlink.
2046 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00002047 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
2048 prid, resblks > 0, &ip, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 if (error) {
2050 if (error == ENOSPC)
2051 goto error_return;
2052 goto error1;
2053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Christoph Hellwig993386c2007-08-28 16:12:30 +10002055 /*
2056 * An error after we've joined dp to the transaction will result in the
2057 * transaction cancel unlocking dp so don't do it explicitly in the
2058 * error path.
2059 */
Christoph Hellwig898621d2010-06-24 11:36:58 +10002060 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002061 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
2063 /*
2064 * Also attach the dquot(s) to it, if applicable.
2065 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002066 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
2068 if (resblks)
2069 resblks -= XFS_IALLOC_SPACE_RES(mp);
2070 /*
2071 * If the symlink will fit into the inode, write it inline.
2072 */
2073 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2074 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2075 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2076 ip->i_d.di_size = pathlen;
2077
2078 /*
2079 * The inode was initially created in extent format.
2080 */
2081 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2082 ip->i_df.if_flags |= XFS_IFINLINE;
2083
2084 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2085 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2086
2087 } else {
2088 first_fsb = 0;
2089 nmaps = SYMLINK_MAPS;
2090
2091 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2092 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2093 &first_block, resblks, mval, &nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002094 &free_list);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00002095 if (error)
2096 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 if (resblks)
2099 resblks -= fs_blocks;
2100 ip->i_d.di_size = pathlen;
2101 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2102
2103 cur_chunk = target_path;
2104 for (n = 0; n < nmaps; n++) {
2105 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2106 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2107 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2108 BTOBB(byte_cnt), 0);
2109 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2110 if (pathlen < byte_cnt) {
2111 byte_cnt = pathlen;
2112 }
2113 pathlen -= byte_cnt;
2114
2115 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2116 cur_chunk += byte_cnt;
2117
2118 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2119 }
2120 }
2121
2122 /*
2123 * Create the directory entry for the symlink.
2124 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002125 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2126 &first_block, &free_list, resblks);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002127 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00002128 goto error2;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10002129 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2131
2132 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 * If this is a synchronous mount, make sure that the
2134 * symlink transaction goes to disk before returning to
2135 * the user.
2136 */
2137 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2138 xfs_trans_set_sync(tp);
2139 }
2140
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002141 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (error) {
2143 goto error2;
2144 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002145 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002146 xfs_qm_dqrele(udqp);
2147 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
Christoph Hellwig288699f2010-06-23 18:11:15 +10002149 *ipp = ip;
2150 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
2152 error2:
2153 IRELE(ip);
2154 error1:
2155 xfs_bmap_cancel(&free_list);
2156 cancel_flags |= XFS_TRANS_ABORT;
2157 error_return:
2158 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002159 xfs_qm_dqrele(udqp);
2160 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
Christoph Hellwig993386c2007-08-28 16:12:30 +10002162 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002164 std_return:
2165 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166}
2167
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168int
Christoph Hellwig993386c2007-08-28 16:12:30 +10002169xfs_set_dmattrs(
2170 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10002172 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002174 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 int error;
2177
2178 if (!capable(CAP_SYS_ADMIN))
2179 return XFS_ERROR(EPERM);
2180
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 if (XFS_FORCED_SHUTDOWN(mp))
2182 return XFS_ERROR(EIO);
2183
2184 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2185 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2186 if (error) {
2187 xfs_trans_cancel(tp, 0);
2188 return error;
2189 }
2190 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10002191 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
Christoph Hellwig613d7042007-10-11 17:44:08 +10002193 ip->i_d.di_dmevmask = evmask;
2194 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
2196 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002197 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199 return error;
2200}
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202/*
2203 * xfs_alloc_file_space()
2204 * This routine allocates disk space for the given file.
2205 *
2206 * If alloc_type == 0, this request is for an ALLOCSP type
2207 * request which will change the file size. In this case, no
2208 * DMAPI event will be generated by the call. A TRUNCATE event
2209 * will be generated later by xfs_setattr.
2210 *
2211 * If alloc_type != 0, this request is for a RESVSP type
2212 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2213 * lower block boundary byte address is less than the file's
2214 * length.
2215 *
2216 * RETURNS:
2217 * 0 on success
2218 * errno on error
2219 *
2220 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002221STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222xfs_alloc_file_space(
2223 xfs_inode_t *ip,
2224 xfs_off_t offset,
2225 xfs_off_t len,
2226 int alloc_type,
2227 int attr_flags)
2228{
Nathan Scottdd9f4382006-01-11 15:28:28 +11002229 xfs_mount_t *mp = ip->i_mount;
2230 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 xfs_filblks_t allocated_fsb;
2232 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002233 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002235 xfs_fsblock_t firstfsb;
2236 int nimaps;
2237 int bmapi_flag;
2238 int quota_flag;
2239 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002241 xfs_bmbt_irec_t imaps[1], *imapp;
2242 xfs_bmap_free_t free_list;
2243 uint qblocks, resblks, resrtextents;
2244 int committed;
2245 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10002247 trace_xfs_alloc_file_space(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249 if (XFS_FORCED_SHUTDOWN(mp))
2250 return XFS_ERROR(EIO);
2251
Christoph Hellwig7d095252009-06-08 15:33:32 +02002252 error = xfs_qm_dqattach(ip, 0);
2253 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 return error;
2255
2256 if (len <= 0)
2257 return XFS_ERROR(EINVAL);
2258
David Chinner957d0eb2007-06-18 16:50:37 +10002259 rt = XFS_IS_REALTIME_INODE(ip);
2260 extsz = xfs_get_extsz_hint(ip);
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11002264 nimaps = 1;
Dave Chinner44722352010-08-24 12:02:11 +10002265 bmapi_flag = XFS_BMAPI_WRITE | alloc_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2267 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002270 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002273 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Nathan Scottdd9f4382006-01-11 15:28:28 +11002275 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11002276 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11002277 */
2278 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002280 do_div(s, extsz);
2281 s *= extsz;
2282 e = startoffset_fsb + allocatesize_fsb;
2283 if ((temp = do_mod(startoffset_fsb, extsz)))
2284 e += temp;
2285 if ((temp = do_mod(e, extsz)))
2286 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002288 s = 0;
2289 e = allocatesize_fsb;
2290 }
2291
Dave Chinner72656c42010-09-03 12:19:33 +10002292 /*
2293 * The transaction reservation is limited to a 32-bit block
2294 * count, hence we need to limit the number of blocks we are
2295 * trying to reserve to avoid an overflow. We can't allocate
2296 * more than @nimaps extents, and an extent is limited on disk
2297 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
2298 */
2299 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
Nathan Scottdd9f4382006-01-11 15:28:28 +11002300 if (unlikely(rt)) {
Dave Chinner72656c42010-09-03 12:19:33 +10002301 resrtextents = qblocks = resblks;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002302 resrtextents /= mp->m_sb.sb_rextsize;
2303 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2304 quota_flag = XFS_QMOPT_RES_RTBLKS;
2305 } else {
2306 resrtextents = 0;
Dave Chinner72656c42010-09-03 12:19:33 +10002307 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002308 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 }
2310
2311 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002312 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 */
2314 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002315 error = xfs_trans_reserve(tp, resblks,
2316 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 XFS_TRANS_PERM_LOG_RES,
2318 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002320 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 */
2322 if (error) {
2323 /*
2324 * Free the transaction structure.
2325 */
2326 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2327 xfs_trans_cancel(tp, 0);
2328 break;
2329 }
2330 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002331 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2332 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 if (error)
2334 goto error1;
2335
Christoph Hellwig898621d2010-06-24 11:36:58 +10002336 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
2338 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002339 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002341 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002342 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11002343 allocatesize_fsb, bmapi_flag,
2344 &firstfsb, 0, imapp, &nimaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002345 &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 if (error) {
2347 goto error0;
2348 }
2349
2350 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002351 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002353 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 if (error) {
2355 goto error0;
2356 }
2357
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002358 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2360 if (error) {
2361 break;
2362 }
2363
2364 allocated_fsb = imapp->br_blockcount;
2365
Nathan Scottdd9f4382006-01-11 15:28:28 +11002366 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 error = XFS_ERROR(ENOSPC);
2368 break;
2369 }
2370
2371 startoffset_fsb += allocated_fsb;
2372 allocatesize_fsb -= allocated_fsb;
2373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374
2375 return error;
2376
Nathan Scottdd9f4382006-01-11 15:28:28 +11002377error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 xfs_bmap_cancel(&free_list);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002379 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002380
2381error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2383 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002384 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385}
2386
2387/*
2388 * Zero file bytes between startoff and endoff inclusive.
2389 * The iolock is held exclusive and no blocks are buffered.
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002390 *
2391 * This function is used by xfs_free_file_space() to zero
2392 * partial blocks when the range to free is not block aligned.
2393 * When unreserving space with boundaries that are not block
2394 * aligned we round up the start and round down the end
2395 * boundaries and then use this function to zero the parts of
2396 * the blocks that got dropped during the rounding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 */
2398STATIC int
2399xfs_zero_remaining_bytes(
2400 xfs_inode_t *ip,
2401 xfs_off_t startoff,
2402 xfs_off_t endoff)
2403{
2404 xfs_bmbt_irec_t imap;
2405 xfs_fileoff_t offset_fsb;
2406 xfs_off_t lastoffset;
2407 xfs_off_t offset;
2408 xfs_buf_t *bp;
2409 xfs_mount_t *mp = ip->i_mount;
2410 int nimap;
2411 int error = 0;
2412
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002413 /*
2414 * Avoid doing I/O beyond eof - it's not necessary
2415 * since nothing can read beyond eof. The space will
2416 * be zeroed when the file is extended anyway.
2417 */
2418 if (startoff >= ip->i_size)
2419 return 0;
2420
2421 if (endoff > ip->i_size)
2422 endoff = ip->i_size;
2423
Dave Chinner686865f2010-09-24 20:07:47 +10002424 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
2425 mp->m_rtdev_targp : mp->m_ddev_targp,
2426 mp->m_sb.sb_blocksize, XBF_DONT_BLOCK);
Lachlan McIlroyc6422612008-12-05 13:16:15 +11002427 if (!bp)
2428 return XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
2430 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2431 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2432 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002433 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002434 NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 if (error || nimap < 1)
2436 break;
2437 ASSERT(imap.br_blockcount >= 1);
2438 ASSERT(imap.br_startoff == offset_fsb);
2439 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2440 if (lastoffset > endoff)
2441 lastoffset = endoff;
2442 if (imap.br_startblock == HOLESTARTBLOCK)
2443 continue;
2444 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2445 if (imap.br_state == XFS_EXT_UNWRITTEN)
2446 continue;
2447 XFS_BUF_UNDONE(bp);
2448 XFS_BUF_UNWRITE(bp);
2449 XFS_BUF_READ(bp);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002450 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002452 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002453 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2455 mp, bp, XFS_BUF_ADDR(bp));
2456 break;
2457 }
2458 memset(XFS_BUF_PTR(bp) +
2459 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2460 0, lastoffset - offset + 1);
2461 XFS_BUF_UNDONE(bp);
2462 XFS_BUF_UNREAD(bp);
2463 XFS_BUF_WRITE(bp);
2464 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002465 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002466 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2468 mp, bp, XFS_BUF_ADDR(bp));
2469 break;
2470 }
2471 }
2472 xfs_buf_free(bp);
2473 return error;
2474}
2475
2476/*
2477 * xfs_free_file_space()
2478 * This routine frees disk space for the given file.
2479 *
2480 * This routine is only called by xfs_change_file_space
2481 * for an UNRESVSP type call.
2482 *
2483 * RETURNS:
2484 * 0 on success
2485 * errno on error
2486 *
2487 */
2488STATIC int
2489xfs_free_file_space(
2490 xfs_inode_t *ip,
2491 xfs_off_t offset,
2492 xfs_off_t len,
2493 int attr_flags)
2494{
2495 int committed;
2496 int done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 xfs_fileoff_t endoffset_fsb;
2498 int error;
2499 xfs_fsblock_t firstfsb;
2500 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 xfs_bmbt_irec_t imap;
2502 xfs_off_t ioffset;
2503 xfs_extlen_t mod=0;
2504 xfs_mount_t *mp;
2505 int nimap;
2506 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10002507 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 int rt;
2509 xfs_fileoff_t startoffset_fsb;
2510 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002511 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 mp = ip->i_mount;
2514
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10002515 trace_xfs_free_file_space(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002516
Christoph Hellwig7d095252009-06-08 15:33:32 +02002517 error = xfs_qm_dqattach(ip, 0);
2518 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 return error;
2520
2521 error = 0;
2522 if (len <= 0) /* if nothing being freed */
2523 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11002524 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002526 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002528 if (attr_flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002529 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11002530 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig25e41b32008-12-03 12:20:39 +01002532 /* wait for the completion of any pending DIOs */
2533 xfs_ioend_wait(ip);
Yingping Lu9fa80462006-03-22 12:44:35 +11002534 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002535
Tim Shimmine6a4b372007-11-23 16:30:42 +11002536 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002538
Christoph Hellwigdf80c932008-08-13 16:22:09 +10002539 if (VN_CACHED(VFS_I(ip)) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11002540 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10002541 if (error)
2542 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002543 }
2544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 /*
2546 * Need to zero the stuff we're not freeing, on disk.
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002547 * If it's a realtime file & can't use unwritten extents then we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2549 * will take care of it for us.
2550 */
Eric Sandeen62118702008-03-06 13:44:28 +11002551 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002553 error = xfs_bmapi(NULL, ip, startoffset_fsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002554 1, 0, NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (error)
2556 goto out_unlock_iolock;
2557 ASSERT(nimap == 0 || nimap == 1);
2558 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2559 xfs_daddr_t block;
2560
2561 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2562 block = imap.br_startblock;
2563 mod = do_div(block, mp->m_sb.sb_rextsize);
2564 if (mod)
2565 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2566 }
2567 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002568 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002569 1, 0, NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 if (error)
2571 goto out_unlock_iolock;
2572 ASSERT(nimap == 0 || nimap == 1);
2573 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2574 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2575 mod++;
2576 if (mod && (mod != mp->m_sb.sb_rextsize))
2577 endoffset_fsb -= mod;
2578 }
2579 }
2580 if ((done = (endoffset_fsb <= startoffset_fsb)))
2581 /*
2582 * One contiguous piece to clear
2583 */
2584 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2585 else {
2586 /*
2587 * Some full blocks, possibly two pieces to clear
2588 */
2589 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2590 error = xfs_zero_remaining_bytes(ip, offset,
2591 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2592 if (!error &&
2593 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2594 error = xfs_zero_remaining_bytes(ip,
2595 XFS_FSB_TO_B(mp, endoffset_fsb),
2596 offset + len - 1);
2597 }
2598
2599 /*
2600 * free file space until done or until there is an error
2601 */
2602 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2603 while (!error && !done) {
2604
2605 /*
David Chinner91ebecc2007-07-19 16:28:30 +10002606 * allocate and setup the transaction. Allow this
2607 * transaction to dip into the reserve blocks to ensure
2608 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 */
2610 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10002611 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 error = xfs_trans_reserve(tp,
2613 resblks,
2614 XFS_WRITE_LOG_RES(mp),
2615 0,
2616 XFS_TRANS_PERM_LOG_RES,
2617 XFS_WRITE_LOG_COUNT);
2618
2619 /*
2620 * check for running out of space
2621 */
2622 if (error) {
2623 /*
2624 * Free the transaction structure.
2625 */
2626 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2627 xfs_trans_cancel(tp, 0);
2628 break;
2629 }
2630 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002631 error = xfs_trans_reserve_quota(tp, mp,
2632 ip->i_udquot, ip->i_gdquot,
2633 resblks, 0, XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if (error)
2635 goto error1;
2636
Christoph Hellwig898621d2010-06-24 11:36:58 +10002637 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
2639 /*
2640 * issue the bunmapi() call to free the blocks
2641 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002642 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002643 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 endoffset_fsb - startoffset_fsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002645 0, 2, &firstfsb, &free_list, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 if (error) {
2647 goto error0;
2648 }
2649
2650 /*
2651 * complete the transaction
2652 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002653 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 if (error) {
2655 goto error0;
2656 }
2657
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002658 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2660 }
2661
2662 out_unlock_iolock:
2663 if (need_iolock)
2664 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2665 return error;
2666
2667 error0:
2668 xfs_bmap_cancel(&free_list);
2669 error1:
2670 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2671 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2672 XFS_ILOCK_EXCL);
2673 return error;
2674}
2675
2676/*
2677 * xfs_change_file_space()
2678 * This routine allocates or frees disk space for the given file.
2679 * The user specified parameters are checked for alignment and size
2680 * limitations.
2681 *
2682 * RETURNS:
2683 * 0 on success
2684 * errno on error
2685 *
2686 */
2687int
2688xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002689 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 int cmd,
2691 xfs_flock64_t *bf,
2692 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 int attr_flags)
2694{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002695 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 int clrprealloc;
2697 int error;
2698 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 int setprealloc;
2700 xfs_off_t startoffset;
2701 xfs_off_t llen;
2702 xfs_trans_t *tp;
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002703 struct iattr iattr;
Dave Chinner44722352010-08-24 12:02:11 +10002704 int prealloc_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705
Christoph Hellwig993386c2007-08-28 16:12:30 +10002706 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 return XFS_ERROR(EINVAL);
2708
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 switch (bf->l_whence) {
2710 case 0: /*SEEK_SET*/
2711 break;
2712 case 1: /*SEEK_CUR*/
2713 bf->l_start += offset;
2714 break;
2715 case 2: /*SEEK_END*/
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002716 bf->l_start += ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 break;
2718 default:
2719 return XFS_ERROR(EINVAL);
2720 }
2721
2722 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2723
2724 if ( (bf->l_start < 0)
2725 || (bf->l_start > XFS_MAXIOFFSET(mp))
2726 || (bf->l_start + llen < 0)
2727 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2728 return XFS_ERROR(EINVAL);
2729
2730 bf->l_whence = 0;
2731
2732 startoffset = bf->l_start;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002733 fsize = ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
2735 /*
2736 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2737 * file space.
2738 * These calls do NOT zero the data space allocated to the file,
2739 * nor do they change the file size.
2740 *
2741 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2742 * space.
2743 * These calls cause the new file data to be zeroed and the file
2744 * size to be changed.
2745 */
2746 setprealloc = clrprealloc = 0;
Dave Chinner44722352010-08-24 12:02:11 +10002747 prealloc_type = XFS_BMAPI_PREALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748
2749 switch (cmd) {
Dave Chinner44722352010-08-24 12:02:11 +10002750 case XFS_IOC_ZERO_RANGE:
2751 prealloc_type |= XFS_BMAPI_CONVERT;
2752 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2753 /* FALLTHRU */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 case XFS_IOC_RESVSP:
2755 case XFS_IOC_RESVSP64:
2756 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
Dave Chinner44722352010-08-24 12:02:11 +10002757 prealloc_type, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 if (error)
2759 return error;
2760 setprealloc = 1;
2761 break;
2762
2763 case XFS_IOC_UNRESVSP:
2764 case XFS_IOC_UNRESVSP64:
2765 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2766 attr_flags)))
2767 return error;
2768 break;
2769
2770 case XFS_IOC_ALLOCSP:
2771 case XFS_IOC_ALLOCSP64:
2772 case XFS_IOC_FREESP:
2773 case XFS_IOC_FREESP64:
2774 if (startoffset > fsize) {
2775 error = xfs_alloc_file_space(ip, fsize,
2776 startoffset - fsize, 0, attr_flags);
2777 if (error)
2778 break;
2779 }
2780
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002781 iattr.ia_valid = ATTR_SIZE;
2782 iattr.ia_size = startoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +11002784 error = xfs_setattr(ip, &iattr, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
2786 if (error)
2787 return error;
2788
2789 clrprealloc = 1;
2790 break;
2791
2792 default:
2793 ASSERT(0);
2794 return XFS_ERROR(EINVAL);
2795 }
2796
2797 /*
2798 * update the inode timestamp, mode, and prealloc flag bits
2799 */
2800 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2801
2802 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2803 0, 0, 0))) {
2804 /* ASSERT(0); */
2805 xfs_trans_cancel(tp, 0);
2806 return error;
2807 }
2808
2809 xfs_ilock(ip, XFS_ILOCK_EXCL);
2810
Christoph Hellwig898621d2010-06-24 11:36:58 +10002811 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002813 if ((attr_flags & XFS_ATTR_DMI) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 ip->i_d.di_mode &= ~S_ISUID;
2815
2816 /*
2817 * Note that we don't have to worry about mandatory
2818 * file locking being disabled here because we only
2819 * clear the S_ISGID bit if the Group execute bit is
2820 * on, but if it was on then mandatory locking wouldn't
2821 * have been enabled.
2822 */
2823 if (ip->i_d.di_mode & S_IXGRP)
2824 ip->i_d.di_mode &= ~S_ISGID;
2825
Dave Chinnerdcd79a12010-09-28 12:27:25 +10002826 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 }
2828 if (setprealloc)
2829 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2830 else if (clrprealloc)
2831 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2832
2833 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Dave Chinner82878892011-03-26 09:13:08 +11002834 if (attr_flags & XFS_ATTR_SYNC)
2835 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002837 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838
2839 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2840
2841 return error;
2842}