blob: 9ad9e355a1c0223343c3f2e6f0258fcec0238d01 [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 */
Robert P. J. Day40ebd812007-11-23 16:30:51 +110018#include <linux/log2.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110022#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110023#include "xfs_format.h"
24#include "xfs_log_format.h"
25#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110028#include "xfs_inode.h"
Dave Chinner57062782013-10-15 09:17:51 +110029#include "xfs_da_format.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100030#include "xfs_da_btree.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100031#include "xfs_dir2.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_attr_sf.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100033#include "xfs_attr.h"
Dave Chinner239880e2013-10-23 10:50:10 +110034#include "xfs_trans_space.h"
35#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_ialloc.h"
39#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100040#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100043#include "xfs_filestream.h"
Christoph Hellwig93848a92013-04-03 16:11:17 +110044#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110046#include "xfs_icache.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100047#include "xfs_symlink.h"
Dave Chinner239880e2013-10-23 10:50:10 +110048#include "xfs_trans_priv.h"
49#include "xfs_log.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110050#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Linus Torvalds1da177e2005-04-16 15:20:36 -070052kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +020055 * Used in xfs_itruncate_extents(). This is the maximum number of extents
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 * freed from a file in a single transaction.
57 */
58#define XFS_ITRUNC_MAX_EXTENTS 2
59
60STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Zhi Yong Wuab297432013-12-18 08:22:41 +080062STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
63
Dave Chinner2a0ec1d2012-04-23 15:59:02 +100064/*
65 * helper function to extract extent size hint from inode
66 */
67xfs_extlen_t
68xfs_get_extsz_hint(
69 struct xfs_inode *ip)
70{
71 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 return ip->i_d.di_extsize;
73 if (XFS_IS_REALTIME_INODE(ip))
74 return ip->i_mount->m_sb.sb_rextsize;
75 return 0;
76}
77
Dave Chinnerfa96aca2012-10-08 21:56:10 +110078/*
Christoph Hellwigefa70be2013-12-18 02:14:39 -080079 * These two are wrapper routines around the xfs_ilock() routine used to
80 * centralize some grungy code. They are used in places that wish to lock the
81 * inode solely for reading the extents. The reason these places can't just
82 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83 * bringing in of the extents from disk for a file in b-tree format. If the
84 * inode is in b-tree format, then we need to lock the inode exclusively until
85 * the extents are read in. Locking it exclusively all the time would limit
86 * our parallelism unnecessarily, though. What we do instead is check to see
87 * if the extents have been read in yet, and only lock the inode exclusively
88 * if they have not.
Dave Chinnerfa96aca2012-10-08 21:56:10 +110089 *
Christoph Hellwigefa70be2013-12-18 02:14:39 -080090 * The functions return a value which should be given to the corresponding
Christoph Hellwig01f4f322013-12-06 12:30:08 -080091 * xfs_iunlock() call.
Dave Chinnerfa96aca2012-10-08 21:56:10 +110092 */
93uint
Christoph Hellwig309ecac2013-12-06 12:30:09 -080094xfs_ilock_data_map_shared(
95 struct xfs_inode *ip)
Dave Chinnerfa96aca2012-10-08 21:56:10 +110096{
Christoph Hellwig309ecac2013-12-06 12:30:09 -080097 uint lock_mode = XFS_ILOCK_SHARED;
Dave Chinnerfa96aca2012-10-08 21:56:10 +110098
Christoph Hellwig309ecac2013-12-06 12:30:09 -080099 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100101 lock_mode = XFS_ILOCK_EXCL;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100102 xfs_ilock(ip, lock_mode);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100103 return lock_mode;
104}
105
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800106uint
107xfs_ilock_attr_map_shared(
108 struct xfs_inode *ip)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100109{
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800110 uint lock_mode = XFS_ILOCK_SHARED;
111
112 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
113 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
114 lock_mode = XFS_ILOCK_EXCL;
115 xfs_ilock(ip, lock_mode);
116 return lock_mode;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100117}
118
119/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100120 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
121 * the i_lock. This routine allows various combinations of the locks to be
122 * obtained.
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100123 *
Dave Chinner653c60b2015-02-23 21:43:37 +1100124 * The 3 locks should always be ordered so that the IO lock is obtained first,
125 * the mmap lock second and the ilock last in order to prevent deadlock.
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100126 *
Dave Chinner653c60b2015-02-23 21:43:37 +1100127 * Basic locking order:
128 *
129 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
130 *
131 * mmap_sem locking order:
132 *
133 * i_iolock -> page lock -> mmap_sem
134 * mmap_sem -> i_mmap_lock -> page_lock
135 *
136 * The difference in mmap_sem locking order mean that we cannot hold the
137 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
138 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
139 * in get_user_pages() to map the user pages into the kernel address space for
140 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
141 * page faults already hold the mmap_sem.
142 *
143 * Hence to serialise fully against both syscall and mmap based IO, we need to
144 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
145 * taken in places where we need to invalidate the page cache in a race
146 * free manner (e.g. truncate, hole punch and other extent manipulation
147 * functions).
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100148 */
149void
150xfs_ilock(
151 xfs_inode_t *ip,
152 uint lock_flags)
153{
154 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
155
156 /*
157 * You can't set both SHARED and EXCL for the same lock,
158 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
159 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
160 */
161 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
162 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100163 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
164 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100165 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
166 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000167 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100168
169 if (lock_flags & XFS_IOLOCK_EXCL)
170 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
171 else if (lock_flags & XFS_IOLOCK_SHARED)
172 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
173
Dave Chinner653c60b2015-02-23 21:43:37 +1100174 if (lock_flags & XFS_MMAPLOCK_EXCL)
175 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
176 else if (lock_flags & XFS_MMAPLOCK_SHARED)
177 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
178
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100179 if (lock_flags & XFS_ILOCK_EXCL)
180 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
181 else if (lock_flags & XFS_ILOCK_SHARED)
182 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
183}
184
185/*
186 * This is just like xfs_ilock(), except that the caller
187 * is guaranteed not to sleep. It returns 1 if it gets
188 * the requested locks and 0 otherwise. If the IO lock is
189 * obtained but the inode lock cannot be, then the IO lock
190 * is dropped before returning.
191 *
192 * ip -- the inode being locked
193 * lock_flags -- this parameter indicates the inode's locks to be
194 * to be locked. See the comment for xfs_ilock() for a list
195 * of valid values.
196 */
197int
198xfs_ilock_nowait(
199 xfs_inode_t *ip,
200 uint lock_flags)
201{
202 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
203
204 /*
205 * You can't set both SHARED and EXCL for the same lock,
206 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
207 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
208 */
209 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
210 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100211 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
212 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100213 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
214 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000215 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100216
217 if (lock_flags & XFS_IOLOCK_EXCL) {
218 if (!mrtryupdate(&ip->i_iolock))
219 goto out;
220 } else if (lock_flags & XFS_IOLOCK_SHARED) {
221 if (!mrtryaccess(&ip->i_iolock))
222 goto out;
223 }
Dave Chinner653c60b2015-02-23 21:43:37 +1100224
225 if (lock_flags & XFS_MMAPLOCK_EXCL) {
226 if (!mrtryupdate(&ip->i_mmaplock))
227 goto out_undo_iolock;
228 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
229 if (!mrtryaccess(&ip->i_mmaplock))
230 goto out_undo_iolock;
231 }
232
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100233 if (lock_flags & XFS_ILOCK_EXCL) {
234 if (!mrtryupdate(&ip->i_lock))
Dave Chinner653c60b2015-02-23 21:43:37 +1100235 goto out_undo_mmaplock;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100236 } else if (lock_flags & XFS_ILOCK_SHARED) {
237 if (!mrtryaccess(&ip->i_lock))
Dave Chinner653c60b2015-02-23 21:43:37 +1100238 goto out_undo_mmaplock;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100239 }
240 return 1;
241
Dave Chinner653c60b2015-02-23 21:43:37 +1100242out_undo_mmaplock:
243 if (lock_flags & XFS_MMAPLOCK_EXCL)
244 mrunlock_excl(&ip->i_mmaplock);
245 else if (lock_flags & XFS_MMAPLOCK_SHARED)
246 mrunlock_shared(&ip->i_mmaplock);
247out_undo_iolock:
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100248 if (lock_flags & XFS_IOLOCK_EXCL)
249 mrunlock_excl(&ip->i_iolock);
250 else if (lock_flags & XFS_IOLOCK_SHARED)
251 mrunlock_shared(&ip->i_iolock);
Dave Chinner653c60b2015-02-23 21:43:37 +1100252out:
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100253 return 0;
254}
255
256/*
257 * xfs_iunlock() is used to drop the inode locks acquired with
258 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
259 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
260 * that we know which locks to drop.
261 *
262 * ip -- the inode being unlocked
263 * lock_flags -- this parameter indicates the inode's locks to be
264 * to be unlocked. See the comment for xfs_ilock() for a list
265 * of valid values for this parameter.
266 *
267 */
268void
269xfs_iunlock(
270 xfs_inode_t *ip,
271 uint lock_flags)
272{
273 /*
274 * You can't set both SHARED and EXCL for the same lock,
275 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
276 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
277 */
278 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
279 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100280 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
281 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100282 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
283 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000284 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100285 ASSERT(lock_flags != 0);
286
287 if (lock_flags & XFS_IOLOCK_EXCL)
288 mrunlock_excl(&ip->i_iolock);
289 else if (lock_flags & XFS_IOLOCK_SHARED)
290 mrunlock_shared(&ip->i_iolock);
291
Dave Chinner653c60b2015-02-23 21:43:37 +1100292 if (lock_flags & XFS_MMAPLOCK_EXCL)
293 mrunlock_excl(&ip->i_mmaplock);
294 else if (lock_flags & XFS_MMAPLOCK_SHARED)
295 mrunlock_shared(&ip->i_mmaplock);
296
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100297 if (lock_flags & XFS_ILOCK_EXCL)
298 mrunlock_excl(&ip->i_lock);
299 else if (lock_flags & XFS_ILOCK_SHARED)
300 mrunlock_shared(&ip->i_lock);
301
302 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
303}
304
305/*
306 * give up write locks. the i/o lock cannot be held nested
307 * if it is being demoted.
308 */
309void
310xfs_ilock_demote(
311 xfs_inode_t *ip,
312 uint lock_flags)
313{
Dave Chinner653c60b2015-02-23 21:43:37 +1100314 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
315 ASSERT((lock_flags &
316 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100317
318 if (lock_flags & XFS_ILOCK_EXCL)
319 mrdemote(&ip->i_lock);
Dave Chinner653c60b2015-02-23 21:43:37 +1100320 if (lock_flags & XFS_MMAPLOCK_EXCL)
321 mrdemote(&ip->i_mmaplock);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100322 if (lock_flags & XFS_IOLOCK_EXCL)
323 mrdemote(&ip->i_iolock);
324
325 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
326}
327
Dave Chinner742ae1e2013-04-30 21:39:34 +1000328#if defined(DEBUG) || defined(XFS_WARN)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100329int
330xfs_isilocked(
331 xfs_inode_t *ip,
332 uint lock_flags)
333{
334 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
335 if (!(lock_flags & XFS_ILOCK_SHARED))
336 return !!ip->i_lock.mr_writer;
337 return rwsem_is_locked(&ip->i_lock.mr_lock);
338 }
339
Dave Chinner653c60b2015-02-23 21:43:37 +1100340 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
341 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
342 return !!ip->i_mmaplock.mr_writer;
343 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
344 }
345
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100346 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
347 if (!(lock_flags & XFS_IOLOCK_SHARED))
348 return !!ip->i_iolock.mr_writer;
349 return rwsem_is_locked(&ip->i_iolock.mr_lock);
350 }
351
352 ASSERT(0);
353 return 0;
354}
355#endif
356
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000357#ifdef DEBUG
358int xfs_locked_n;
359int xfs_small_retries;
360int xfs_middle_retries;
361int xfs_lots_retries;
362int xfs_lock_delays;
363#endif
364
Dave Chinnerb6a99472015-08-25 10:05:13 +1000365/*
366 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
367 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
368 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
369 * errors and warnings.
370 */
371#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
Dave Chinner3403ccc2015-08-20 09:27:49 +1000372static bool
373xfs_lockdep_subclass_ok(
374 int subclass)
375{
376 return subclass < MAX_LOCKDEP_SUBCLASSES;
377}
378#else
379#define xfs_lockdep_subclass_ok(subclass) (true)
380#endif
381
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000382/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100383 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
Dave Chinner0952c812015-08-19 10:32:49 +1000384 * value. This can be called for any type of inode lock combination, including
385 * parent locking. Care must be taken to ensure we don't overrun the subclass
386 * storage fields in the class mask we build.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000387 */
388static inline int
389xfs_lock_inumorder(int lock_mode, int subclass)
390{
Dave Chinner0952c812015-08-19 10:32:49 +1000391 int class = 0;
392
393 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
394 XFS_ILOCK_RTSUM)));
Dave Chinner3403ccc2015-08-20 09:27:49 +1000395 ASSERT(xfs_lockdep_subclass_ok(subclass));
Dave Chinner0952c812015-08-19 10:32:49 +1000396
Dave Chinner653c60b2015-02-23 21:43:37 +1100397 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
Dave Chinner0952c812015-08-19 10:32:49 +1000398 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
Dave Chinner3403ccc2015-08-20 09:27:49 +1000399 ASSERT(xfs_lockdep_subclass_ok(subclass +
400 XFS_IOLOCK_PARENT_VAL));
Dave Chinner0952c812015-08-19 10:32:49 +1000401 class += subclass << XFS_IOLOCK_SHIFT;
402 if (lock_mode & XFS_IOLOCK_PARENT)
403 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
Dave Chinner653c60b2015-02-23 21:43:37 +1100404 }
405
406 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
Dave Chinner0952c812015-08-19 10:32:49 +1000407 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
408 class += subclass << XFS_MMAPLOCK_SHIFT;
Dave Chinner653c60b2015-02-23 21:43:37 +1100409 }
410
Dave Chinner0952c812015-08-19 10:32:49 +1000411 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
412 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
413 class += subclass << XFS_ILOCK_SHIFT;
414 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000415
Dave Chinner0952c812015-08-19 10:32:49 +1000416 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000417}
418
419/*
Dave Chinner95afcf52015-03-25 14:03:32 +1100420 * The following routine will lock n inodes in exclusive mode. We assume the
421 * caller calls us with the inodes in i_ino order.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000422 *
Dave Chinner95afcf52015-03-25 14:03:32 +1100423 * We need to detect deadlock where an inode that we lock is in the AIL and we
424 * start waiting for another inode that is locked by a thread in a long running
425 * transaction (such as truncate). This can result in deadlock since the long
426 * running trans might need to wait for the inode we just locked in order to
427 * push the tail and free space in the log.
Dave Chinner0952c812015-08-19 10:32:49 +1000428 *
429 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
430 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
431 * lock more than one at a time, lockdep will report false positives saying we
432 * have violated locking orders.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000433 */
434void
435xfs_lock_inodes(
436 xfs_inode_t **ips,
437 int inodes,
438 uint lock_mode)
439{
440 int attempts = 0, i, j, try_lock;
441 xfs_log_item_t *lp;
442
Dave Chinner0952c812015-08-19 10:32:49 +1000443 /*
444 * Currently supports between 2 and 5 inodes with exclusive locking. We
445 * support an arbitrary depth of locking here, but absolute limits on
446 * inodes depend on the the type of locking and the limits placed by
447 * lockdep annotations in xfs_lock_inumorder. These are all checked by
448 * the asserts.
449 */
Dave Chinner95afcf52015-03-25 14:03:32 +1100450 ASSERT(ips && inodes >= 2 && inodes <= 5);
Dave Chinner0952c812015-08-19 10:32:49 +1000451 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
452 XFS_ILOCK_EXCL));
453 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
454 XFS_ILOCK_SHARED)));
455 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
456 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
457 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
458 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
459 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
460 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
461
462 if (lock_mode & XFS_IOLOCK_EXCL) {
463 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
464 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000466
467 try_lock = 0;
468 i = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000469again:
470 for (; i < inodes; i++) {
471 ASSERT(ips[i]);
472
Dave Chinner95afcf52015-03-25 14:03:32 +1100473 if (i && (ips[i] == ips[i - 1])) /* Already locked */
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000474 continue;
475
476 /*
Dave Chinner95afcf52015-03-25 14:03:32 +1100477 * If try_lock is not set yet, make sure all locked inodes are
478 * not in the AIL. If any are, set try_lock to be used later.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000479 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000480 if (!try_lock) {
481 for (j = (i - 1); j >= 0 && !try_lock; j--) {
482 lp = (xfs_log_item_t *)ips[j]->i_itemp;
Dave Chinner95afcf52015-03-25 14:03:32 +1100483 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000484 try_lock++;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000485 }
486 }
487
488 /*
489 * If any of the previous locks we have locked is in the AIL,
490 * we must TRY to get the second and subsequent locks. If
491 * we can't get any, we must release all we have
492 * and try again.
493 */
Dave Chinner95afcf52015-03-25 14:03:32 +1100494 if (!try_lock) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000495 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Dave Chinner95afcf52015-03-25 14:03:32 +1100496 continue;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000497 }
Dave Chinner95afcf52015-03-25 14:03:32 +1100498
499 /* try_lock means we have an inode locked that is in the AIL. */
500 ASSERT(i != 0);
501 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
502 continue;
503
504 /*
505 * Unlock all previous guys and try again. xfs_iunlock will try
506 * to push the tail if the inode is in the AIL.
507 */
508 attempts++;
509 for (j = i - 1; j >= 0; j--) {
510 /*
511 * Check to see if we've already unlocked this one. Not
512 * the first one going back, and the inode ptr is the
513 * same.
514 */
515 if (j != (i - 1) && ips[j] == ips[j + 1])
516 continue;
517
518 xfs_iunlock(ips[j], lock_mode);
519 }
520
521 if ((attempts % 5) == 0) {
522 delay(1); /* Don't just spin the CPU */
523#ifdef DEBUG
524 xfs_lock_delays++;
525#endif
526 }
527 i = 0;
528 try_lock = 0;
529 goto again;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000530 }
531
532#ifdef DEBUG
533 if (attempts) {
534 if (attempts < 5) xfs_small_retries++;
535 else if (attempts < 100) xfs_middle_retries++;
536 else xfs_lots_retries++;
537 } else {
538 xfs_locked_n++;
539 }
540#endif
541}
542
543/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100544 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
545 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
546 * lock more than one at a time, lockdep will report false positives saying we
547 * have violated locking orders.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000548 */
549void
550xfs_lock_two_inodes(
551 xfs_inode_t *ip0,
552 xfs_inode_t *ip1,
553 uint lock_mode)
554{
555 xfs_inode_t *temp;
556 int attempts = 0;
557 xfs_log_item_t *lp;
558
Dave Chinner653c60b2015-02-23 21:43:37 +1100559 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
560 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
561 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
563 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000565 ASSERT(ip0->i_ino != ip1->i_ino);
566
567 if (ip0->i_ino > ip1->i_ino) {
568 temp = ip0;
569 ip0 = ip1;
570 ip1 = temp;
571 }
572
573 again:
574 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
575
576 /*
577 * If the first lock we have locked is in the AIL, we must TRY to get
578 * the second lock. If we can't get it, we must release the first one
579 * and try again.
580 */
581 lp = (xfs_log_item_t *)ip0->i_itemp;
582 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
583 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
584 xfs_iunlock(ip0, lock_mode);
585 if ((++attempts % 5) == 0)
586 delay(1); /* Don't just spin the CPU */
587 goto again;
588 }
589 } else {
590 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
591 }
592}
593
594
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100595void
596__xfs_iflock(
597 struct xfs_inode *ip)
598{
599 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
600 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
601
602 do {
603 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
604 if (xfs_isiflocked(ip))
605 io_schedule();
606 } while (!xfs_iflock_nowait(ip));
607
608 finish_wait(wq, &wait.wait);
609}
610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611STATIC uint
612_xfs_dic2xflags(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100613 __uint16_t di_flags,
614 uint64_t di_flags2,
615 bool has_attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
617 uint flags = 0;
618
619 if (di_flags & XFS_DIFLAG_ANY) {
620 if (di_flags & XFS_DIFLAG_REALTIME)
Dave Chinnere7b89482016-01-04 16:44:15 +1100621 flags |= FS_XFLAG_REALTIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (di_flags & XFS_DIFLAG_PREALLOC)
Dave Chinnere7b89482016-01-04 16:44:15 +1100623 flags |= FS_XFLAG_PREALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (di_flags & XFS_DIFLAG_IMMUTABLE)
Dave Chinnere7b89482016-01-04 16:44:15 +1100625 flags |= FS_XFLAG_IMMUTABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (di_flags & XFS_DIFLAG_APPEND)
Dave Chinnere7b89482016-01-04 16:44:15 +1100627 flags |= FS_XFLAG_APPEND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (di_flags & XFS_DIFLAG_SYNC)
Dave Chinnere7b89482016-01-04 16:44:15 +1100629 flags |= FS_XFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (di_flags & XFS_DIFLAG_NOATIME)
Dave Chinnere7b89482016-01-04 16:44:15 +1100631 flags |= FS_XFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 if (di_flags & XFS_DIFLAG_NODUMP)
Dave Chinnere7b89482016-01-04 16:44:15 +1100633 flags |= FS_XFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 if (di_flags & XFS_DIFLAG_RTINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100635 flags |= FS_XFLAG_RTINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (di_flags & XFS_DIFLAG_PROJINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100637 flags |= FS_XFLAG_PROJINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
Dave Chinnere7b89482016-01-04 16:44:15 +1100639 flags |= FS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100640 if (di_flags & XFS_DIFLAG_EXTSIZE)
Dave Chinnere7b89482016-01-04 16:44:15 +1100641 flags |= FS_XFLAG_EXTSIZE;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100642 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100643 flags |= FS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000644 if (di_flags & XFS_DIFLAG_NODEFRAG)
Dave Chinnere7b89482016-01-04 16:44:15 +1100645 flags |= FS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000646 if (di_flags & XFS_DIFLAG_FILESTREAM)
Dave Chinnere7b89482016-01-04 16:44:15 +1100647 flags |= FS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649
Dave Chinner58f88ca2016-01-04 16:44:15 +1100650 if (di_flags2 & XFS_DIFLAG2_ANY) {
651 if (di_flags2 & XFS_DIFLAG2_DAX)
652 flags |= FS_XFLAG_DAX;
653 }
654
655 if (has_attr)
656 flags |= FS_XFLAG_HASATTR;
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return flags;
659}
660
661uint
662xfs_ip2xflags(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100663 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Dave Chinner58f88ca2016-01-04 16:44:15 +1100665 struct xfs_icdinode *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Dave Chinner58f88ca2016-01-04 16:44:15 +1100667 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
670uint
671xfs_dic2xflags(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100672 struct xfs_dinode *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Dave Chinner58f88ca2016-01-04 16:44:15 +1100674 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags),
675 be64_to_cpu(dip->di_flags2), XFS_DFORK_Q(dip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
681 * ci_name->name will point to a the actual name (caller must free) or
682 * will be set to NULL if an exact match is found.
683 */
684int
685xfs_lookup(
686 xfs_inode_t *dp,
687 struct xfs_name *name,
688 xfs_inode_t **ipp,
689 struct xfs_name *ci_name)
690{
691 xfs_ino_t inum;
692 int error;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000693
694 trace_xfs_lookup(dp, name);
695
696 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
Dave Chinner24513372014-06-25 14:58:08 +1000697 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000698
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000699 xfs_ilock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000700 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000701 if (error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000702 goto out_unlock;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000703
704 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
705 if (error)
706 goto out_free_name;
707
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000708 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000709 return 0;
710
711out_free_name:
712 if (ci_name)
713 kmem_free(ci_name->name);
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000714out_unlock:
715 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000716 *ipp = NULL;
717 return error;
718}
719
720/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 * Allocate an inode on disk and return a copy of its in-core version.
722 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
723 * appropriately within the inode. The uid and gid for the inode are
724 * set according to the contents of the given cred structure.
725 *
726 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300727 * has a free inode available, call xfs_iget() to obtain the in-core
728 * version of the allocated inode. Finally, fill in the inode and
729 * log its initial contents. In this case, ialloc_context would be
730 * set to NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 *
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300732 * If xfs_dialloc() does not have an available inode, it will replenish
733 * its supply by doing an allocation. Since we can only do one
734 * allocation within a transaction without deadlocks, we must commit
735 * the current transaction before returning the inode itself.
736 * In this case, therefore, we will set ialloc_context and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 * The caller should then commit the current transaction, start a new
738 * transaction, and call xfs_ialloc() again to actually get the inode.
739 *
740 * To ensure that some other process does not grab the inode that
741 * was allocated during the first call to xfs_ialloc(), this routine
742 * also returns the [locked] bp pointing to the head of the freelist
743 * as ialloc_context. The caller should hold this buffer across
744 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000745 *
746 * If we are allocating quota inodes, we do not have a parent inode
747 * to attach to or associate with (i.e. pip == NULL) because they
748 * are not linked into the directory structure - they are attached
749 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 */
751int
752xfs_ialloc(
753 xfs_trans_t *tp,
754 xfs_inode_t *pip,
Al Viro576b1d62011-07-26 02:50:15 -0400755 umode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700756 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000758 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 int okalloc,
760 xfs_buf_t **ialloc_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 xfs_inode_t **ipp)
762{
Christoph Hellwig93848a92013-04-03 16:11:17 +1100763 struct xfs_mount *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 xfs_ino_t ino;
765 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 uint flags;
767 int error;
Dave Chinnere076b0f2014-10-02 09:18:13 +1000768 struct timespec tv;
Dave Chinner39878482016-02-09 16:54:58 +1100769 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 /*
772 * Call the space management code to pick
773 * the on-disk inode to be allocated.
774 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000775 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Christoph Hellwig08358902012-07-04 10:54:47 -0400776 ialloc_context, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100777 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return error;
Christoph Hellwig08358902012-07-04 10:54:47 -0400779 if (*ialloc_context || ino == NULLFSINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *ipp = NULL;
781 return 0;
782 }
783 ASSERT(*ialloc_context == NULL);
784
785 /*
786 * Get the in-core inode with the lock held exclusively.
787 * This is because we're setting fields here we need
788 * to prevent others from looking at until we're done.
789 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100790 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000791 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +1100792 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 ASSERT(ip != NULL);
Dave Chinner39878482016-02-09 16:54:58 +1100795 inode = VFS_I(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Dave Chinner263997a2014-05-20 07:46:40 +1000797 /*
798 * We always convert v1 inodes to v2 now - we only support filesystems
799 * with >= v2 inode capability, so there is no reason for ever leaving
800 * an inode in v1 format.
801 */
802 if (ip->i_d.di_version == 1)
803 ip->i_d.di_version = 2;
804
Al Viro576b1d62011-07-26 02:50:15 -0400805 ip->i_d.di_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 ip->i_d.di_onlink = 0;
807 ip->i_d.di_nlink = nlink;
808 ASSERT(ip->i_d.di_nlink == nlink);
Dwight Engen7aab1b22013-08-15 14:08:01 -0400809 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
810 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000811 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
813
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000814 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 ip->i_d.di_gid = pip->i_d.di_gid;
Al Viroabbede12011-07-26 02:31:30 -0400816 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 ip->i_d.di_mode |= S_ISGID;
818 }
819 }
820
821 /*
822 * If the group ID of the new file does not match the effective group
823 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
824 * (and only if the irix_sgid_inherit compatibility variable is set).
825 */
826 if ((irix_sgid_inherit) &&
827 (ip->i_d.di_mode & S_ISGID) &&
Dwight Engen7aab1b22013-08-15 14:08:01 -0400828 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 ip->i_d.di_mode &= ~S_ISGID;
830 }
831
832 ip->i_d.di_size = 0;
833 ip->i_d.di_nextents = 0;
834 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000835
Dave Chinnere076b0f2014-10-02 09:18:13 +1000836 tv = current_fs_time(mp->m_super);
Dave Chinner39878482016-02-09 16:54:58 +1100837 inode->i_mtime = tv;
838 inode->i_atime = tv;
839 inode->i_ctime = tv;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 /*
842 * di_gen will have been taken care of in xfs_iread.
843 */
844 ip->i_d.di_extsize = 0;
845 ip->i_d.di_dmevmask = 0;
846 ip->i_d.di_dmstate = 0;
847 ip->i_d.di_flags = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100848
849 if (ip->i_d.di_version == 3) {
850 ASSERT(ip->i_d.di_ino == ino);
Dave Chinnerbbf155a2015-08-19 10:31:18 +1000851 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_meta_uuid));
Christoph Hellwig93848a92013-04-03 16:11:17 +1100852 ip->i_d.di_crc = 0;
853 ip->i_d.di_changecount = 1;
854 ip->i_d.di_lsn = 0;
855 ip->i_d.di_flags2 = 0;
856 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
Dave Chinner39878482016-02-09 16:54:58 +1100857 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
858 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100859 }
860
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 flags = XFS_ILOG_CORE;
863 switch (mode & S_IFMT) {
864 case S_IFIFO:
865 case S_IFCHR:
866 case S_IFBLK:
867 case S_IFSOCK:
868 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
869 ip->i_df.if_u2.if_rdev = rdev;
870 ip->i_df.if_flags = 0;
871 flags |= XFS_ILOG_DEV;
872 break;
873 case S_IFREG:
874 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +1000875 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Dave Chinner58f88ca2016-01-04 16:44:15 +1100876 uint64_t di_flags2 = 0;
877 uint di_flags = 0;
Nathan Scott365ca832005-06-21 15:39:12 +1000878
Al Viroabbede12011-07-26 02:31:30 -0400879 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000880 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
881 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100882 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
883 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
884 ip->i_d.di_extsize = pip->i_d.di_extsize;
885 }
Dave Chinner9336e3a2014-10-02 09:18:40 +1000886 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
887 di_flags |= XFS_DIFLAG_PROJINHERIT;
Al Viroabbede12011-07-26 02:31:30 -0400888 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000889 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +1000890 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100891 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
892 di_flags |= XFS_DIFLAG_EXTSIZE;
893 ip->i_d.di_extsize = pip->i_d.di_extsize;
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 }
896 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
897 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +1000898 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
900 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +1000901 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
903 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +1000904 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
906 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +1000907 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000908 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
909 xfs_inherit_nodefrag)
910 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000911 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
912 di_flags |= XFS_DIFLAG_FILESTREAM;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100913 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
914 di_flags2 |= XFS_DIFLAG2_DAX;
915
Nathan Scott365ca832005-06-21 15:39:12 +1000916 ip->i_d.di_flags |= di_flags;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100917 ip->i_d.di_flags2 |= di_flags2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 }
919 /* FALLTHROUGH */
920 case S_IFLNK:
921 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
922 ip->i_df.if_flags = XFS_IFEXTENTS;
923 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
924 ip->i_df.if_u1.if_extents = NULL;
925 break;
926 default:
927 ASSERT(0);
928 }
929 /*
930 * Attribute fork settings for new inode.
931 */
932 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
933 ip->i_d.di_anextents = 0;
934
935 /*
936 * Log the new values stuffed into the inode.
937 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000938 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 xfs_trans_log_inode(tp, ip, flags);
940
Dave Chinner58c90472015-02-23 22:38:08 +1100941 /* now that we have an i_mode we can setup the inode structure */
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000942 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 *ipp = ip;
945 return 0;
946}
947
Dave Chinnere546cb72013-08-12 20:49:47 +1000948/*
949 * Allocates a new inode from disk and return a pointer to the
950 * incore copy. This routine will internally commit the current
951 * transaction and allocate a new one if the Space Manager needed
952 * to do an allocation to replenish the inode free-list.
953 *
954 * This routine is designed to be called from xfs_create and
955 * xfs_create_dir.
956 *
957 */
958int
959xfs_dir_ialloc(
960 xfs_trans_t **tpp, /* input: current transaction;
961 output: may be a new transaction. */
962 xfs_inode_t *dp, /* directory within whose allocate
963 the inode. */
964 umode_t mode,
965 xfs_nlink_t nlink,
966 xfs_dev_t rdev,
967 prid_t prid, /* project id */
968 int okalloc, /* ok to allocate new space */
969 xfs_inode_t **ipp, /* pointer to inode; it will be
970 locked. */
971 int *committed)
972
973{
974 xfs_trans_t *tp;
Dave Chinnere546cb72013-08-12 20:49:47 +1000975 xfs_inode_t *ip;
976 xfs_buf_t *ialloc_context = NULL;
977 int code;
Dave Chinnere546cb72013-08-12 20:49:47 +1000978 void *dqinfo;
979 uint tflags;
980
981 tp = *tpp;
982 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
983
984 /*
985 * xfs_ialloc will return a pointer to an incore inode if
986 * the Space Manager has an available inode on the free
987 * list. Otherwise, it will do an allocation and replenish
988 * the freelist. Since we can only do one allocation per
989 * transaction without deadlocks, we will need to commit the
990 * current transaction and start a new one. We will then
991 * need to call xfs_ialloc again to get the inode.
992 *
993 * If xfs_ialloc did an allocation to replenish the freelist,
994 * it returns the bp containing the head of the freelist as
995 * ialloc_context. We will hold a lock on it across the
996 * transaction commit so that no other process can steal
997 * the inode(s) that we've just allocated.
998 */
999 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1000 &ialloc_context, &ip);
1001
1002 /*
1003 * Return an error if we were unable to allocate a new inode.
1004 * This should only happen if we run out of space on disk or
1005 * encounter a disk error.
1006 */
1007 if (code) {
1008 *ipp = NULL;
1009 return code;
1010 }
1011 if (!ialloc_context && !ip) {
1012 *ipp = NULL;
Dave Chinner24513372014-06-25 14:58:08 +10001013 return -ENOSPC;
Dave Chinnere546cb72013-08-12 20:49:47 +10001014 }
1015
1016 /*
1017 * If the AGI buffer is non-NULL, then we were unable to get an
1018 * inode in one operation. We need to commit the current
1019 * transaction and call xfs_ialloc() again. It is guaranteed
1020 * to succeed the second time.
1021 */
1022 if (ialloc_context) {
1023 /*
1024 * Normally, xfs_trans_commit releases all the locks.
1025 * We call bhold to hang on to the ialloc_context across
1026 * the commit. Holding this buffer prevents any other
1027 * processes from doing any allocations in this
1028 * allocation group.
1029 */
1030 xfs_trans_bhold(tp, ialloc_context);
Dave Chinnere546cb72013-08-12 20:49:47 +10001031
1032 /*
1033 * We want the quota changes to be associated with the next
1034 * transaction, NOT this one. So, detach the dqinfo from this
1035 * and attach it to the next transaction.
1036 */
1037 dqinfo = NULL;
1038 tflags = 0;
1039 if (tp->t_dqinfo) {
1040 dqinfo = (void *)tp->t_dqinfo;
1041 tp->t_dqinfo = NULL;
1042 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1043 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1044 }
1045
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001046 code = xfs_trans_roll(&tp, 0);
1047 if (committed != NULL)
Dave Chinnere546cb72013-08-12 20:49:47 +10001048 *committed = 1;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001049
Dave Chinnere546cb72013-08-12 20:49:47 +10001050 /*
1051 * Re-attach the quota info that we detached from prev trx.
1052 */
1053 if (dqinfo) {
1054 tp->t_dqinfo = dqinfo;
1055 tp->t_flags |= tflags;
1056 }
1057
1058 if (code) {
1059 xfs_buf_relse(ialloc_context);
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001060 *tpp = tp;
Dave Chinnere546cb72013-08-12 20:49:47 +10001061 *ipp = NULL;
1062 return code;
1063 }
1064 xfs_trans_bjoin(tp, ialloc_context);
1065
1066 /*
1067 * Call ialloc again. Since we've locked out all
1068 * other allocations in this allocation group,
1069 * this call should always succeed.
1070 */
1071 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1072 okalloc, &ialloc_context, &ip);
1073
1074 /*
1075 * If we get an error at this point, return to the caller
1076 * so that the current transaction can be aborted.
1077 */
1078 if (code) {
1079 *tpp = tp;
1080 *ipp = NULL;
1081 return code;
1082 }
1083 ASSERT(!ialloc_context && ip);
1084
1085 } else {
1086 if (committed != NULL)
1087 *committed = 0;
1088 }
1089
1090 *ipp = ip;
1091 *tpp = tp;
1092
1093 return 0;
1094}
1095
1096/*
1097 * Decrement the link count on an inode & log the change.
1098 * If this causes the link count to go to zero, initiate the
1099 * logging activity required to truncate a file.
1100 */
1101int /* error */
1102xfs_droplink(
1103 xfs_trans_t *tp,
1104 xfs_inode_t *ip)
1105{
1106 int error;
1107
1108 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1109
1110 ASSERT (ip->i_d.di_nlink > 0);
1111 ip->i_d.di_nlink--;
1112 drop_nlink(VFS_I(ip));
1113 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1114
1115 error = 0;
1116 if (ip->i_d.di_nlink == 0) {
1117 /*
1118 * We're dropping the last link to this file.
1119 * Move the on-disk inode to the AGI unlinked list.
1120 * From xfs_inactive() we will pull the inode from
1121 * the list and free it.
1122 */
1123 error = xfs_iunlink(tp, ip);
1124 }
1125 return error;
1126}
1127
1128/*
Dave Chinnere546cb72013-08-12 20:49:47 +10001129 * Increment the link count on an inode & log the change.
1130 */
1131int
1132xfs_bumplink(
1133 xfs_trans_t *tp,
1134 xfs_inode_t *ip)
1135{
1136 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1137
Dave Chinner263997a2014-05-20 07:46:40 +10001138 ASSERT(ip->i_d.di_version > 1);
Zhi Yong Wuab297432013-12-18 08:22:41 +08001139 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
Dave Chinnere546cb72013-08-12 20:49:47 +10001140 ip->i_d.di_nlink++;
1141 inc_nlink(VFS_I(ip));
Dave Chinnere546cb72013-08-12 20:49:47 +10001142 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1143 return 0;
1144}
1145
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001146int
1147xfs_create(
1148 xfs_inode_t *dp,
1149 struct xfs_name *name,
1150 umode_t mode,
1151 xfs_dev_t rdev,
1152 xfs_inode_t **ipp)
1153{
1154 int is_dir = S_ISDIR(mode);
1155 struct xfs_mount *mp = dp->i_mount;
1156 struct xfs_inode *ip = NULL;
1157 struct xfs_trans *tp = NULL;
1158 int error;
1159 xfs_bmap_free_t free_list;
1160 xfs_fsblock_t first_block;
1161 bool unlock_dp_on_error = false;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001162 prid_t prid;
1163 struct xfs_dquot *udqp = NULL;
1164 struct xfs_dquot *gdqp = NULL;
1165 struct xfs_dquot *pdqp = NULL;
Brian Foster062647a2014-11-28 14:00:16 +11001166 struct xfs_trans_res *tres;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001167 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001168
1169 trace_xfs_create(dp, name);
1170
1171 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001172 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001173
Zhi Yong Wu163467d2013-12-18 08:22:39 +08001174 prid = xfs_get_initial_prid(dp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001175
1176 /*
1177 * Make sure that we have allocated dquot(s) on disk.
1178 */
Dwight Engen7aab1b22013-08-15 14:08:01 -04001179 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1180 xfs_kgid_to_gid(current_fsgid()), prid,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001181 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1182 &udqp, &gdqp, &pdqp);
1183 if (error)
1184 return error;
1185
1186 if (is_dir) {
1187 rdev = 0;
1188 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001189 tres = &M_RES(mp)->tr_mkdir;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001190 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1191 } else {
1192 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001193 tres = &M_RES(mp)->tr_create;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001194 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1195 }
1196
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001197 /*
1198 * Initially assume that the file does not exist and
1199 * reserve the resources for that case. If that is not
1200 * the case we'll drop the one we have and get a more
1201 * appropriate transaction later.
1202 */
Brian Foster062647a2014-11-28 14:00:16 +11001203 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001204 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001205 /* flush outstanding delalloc blocks and retry */
1206 xfs_flush_inodes(mp);
Brian Foster062647a2014-11-28 14:00:16 +11001207 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001208 }
Dave Chinner24513372014-06-25 14:58:08 +10001209 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001210 /* No space at all so try a "no-allocation" reservation */
1211 resblks = 0;
Brian Foster062647a2014-11-28 14:00:16 +11001212 error = xfs_trans_reserve(tp, tres, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001213 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001214 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001215 goto out_trans_cancel;
Christoph Hellwig4906e212015-06-04 13:47:56 +10001216
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001217
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001218 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1219 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001220 unlock_dp_on_error = true;
1221
1222 xfs_bmap_init(&free_list, &first_block);
1223
1224 /*
1225 * Reserve disk quota and the inode.
1226 */
1227 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1228 pdqp, resblks, 1, 0);
1229 if (error)
1230 goto out_trans_cancel;
1231
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001232 if (!resblks) {
1233 error = xfs_dir_canenter(tp, dp, name);
1234 if (error)
1235 goto out_trans_cancel;
1236 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001237
1238 /*
1239 * A newly created regular or special file just has one directory
1240 * entry pointing to them, but a directory also the "." entry
1241 * pointing to itself.
1242 */
1243 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001244 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001245 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001246 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001247
1248 /*
1249 * Now we join the directory inode to the transaction. We do not do it
1250 * earlier because xfs_dir_ialloc might commit the previous transaction
1251 * (and release all the locks). An error from here on will result in
1252 * the transaction cancel unlocking dp so don't do it explicitly in the
1253 * error path.
1254 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001255 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001256 unlock_dp_on_error = false;
1257
1258 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1259 &first_block, &free_list, resblks ?
1260 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1261 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001262 ASSERT(error != -ENOSPC);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001263 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001264 }
1265 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1266 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1267
1268 if (is_dir) {
1269 error = xfs_dir_init(tp, ip, dp);
1270 if (error)
1271 goto out_bmap_cancel;
1272
1273 error = xfs_bumplink(tp, dp);
1274 if (error)
1275 goto out_bmap_cancel;
1276 }
1277
1278 /*
1279 * If this is a synchronous mount, make sure that the
1280 * create transaction goes to disk before returning to
1281 * the user.
1282 */
1283 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1284 xfs_trans_set_sync(tp);
1285
1286 /*
1287 * Attach the dquot(s) to the inodes and modify them incore.
1288 * These ids of the inode couldn't have changed since the new
1289 * inode has been locked ever since it was created.
1290 */
1291 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1292
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001293 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001294 if (error)
1295 goto out_bmap_cancel;
1296
Christoph Hellwig70393312015-06-04 13:48:08 +10001297 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001298 if (error)
1299 goto out_release_inode;
1300
1301 xfs_qm_dqrele(udqp);
1302 xfs_qm_dqrele(gdqp);
1303 xfs_qm_dqrele(pdqp);
1304
1305 *ipp = ip;
1306 return 0;
1307
1308 out_bmap_cancel:
1309 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001310 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001311 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001312 out_release_inode:
1313 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001314 * Wait until after the current transaction is aborted to finish the
1315 * setup of the inode and release the inode. This prevents recursive
1316 * transactions and deadlocks from xfs_inactive.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001317 */
Dave Chinner58c90472015-02-23 22:38:08 +11001318 if (ip) {
1319 xfs_finish_inode_setup(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001320 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001321 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001322
1323 xfs_qm_dqrele(udqp);
1324 xfs_qm_dqrele(gdqp);
1325 xfs_qm_dqrele(pdqp);
1326
1327 if (unlock_dp_on_error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001328 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001329 return error;
1330}
1331
1332int
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001333xfs_create_tmpfile(
1334 struct xfs_inode *dp,
1335 struct dentry *dentry,
Brian Foster330033d2014-04-17 08:15:30 +10001336 umode_t mode,
1337 struct xfs_inode **ipp)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001338{
1339 struct xfs_mount *mp = dp->i_mount;
1340 struct xfs_inode *ip = NULL;
1341 struct xfs_trans *tp = NULL;
1342 int error;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001343 prid_t prid;
1344 struct xfs_dquot *udqp = NULL;
1345 struct xfs_dquot *gdqp = NULL;
1346 struct xfs_dquot *pdqp = NULL;
1347 struct xfs_trans_res *tres;
1348 uint resblks;
1349
1350 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001351 return -EIO;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001352
1353 prid = xfs_get_initial_prid(dp);
1354
1355 /*
1356 * Make sure that we have allocated dquot(s) on disk.
1357 */
1358 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1359 xfs_kgid_to_gid(current_fsgid()), prid,
1360 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1361 &udqp, &gdqp, &pdqp);
1362 if (error)
1363 return error;
1364
1365 resblks = XFS_IALLOC_SPACE_RES(mp);
1366 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1367
1368 tres = &M_RES(mp)->tr_create_tmpfile;
1369 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001370 if (error == -ENOSPC) {
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001371 /* No space at all so try a "no-allocation" reservation */
1372 resblks = 0;
1373 error = xfs_trans_reserve(tp, tres, 0, 0);
1374 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001375 if (error)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001376 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001377
1378 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1379 pdqp, resblks, 1, 0);
1380 if (error)
1381 goto out_trans_cancel;
1382
1383 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1384 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001385 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001386 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001387
1388 if (mp->m_flags & XFS_MOUNT_WSYNC)
1389 xfs_trans_set_sync(tp);
1390
1391 /*
1392 * Attach the dquot(s) to the inodes and modify them incore.
1393 * These ids of the inode couldn't have changed since the new
1394 * inode has been locked ever since it was created.
1395 */
1396 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1397
1398 ip->i_d.di_nlink--;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001399 error = xfs_iunlink(tp, ip);
1400 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001401 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001402
Christoph Hellwig70393312015-06-04 13:48:08 +10001403 error = xfs_trans_commit(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001404 if (error)
1405 goto out_release_inode;
1406
1407 xfs_qm_dqrele(udqp);
1408 xfs_qm_dqrele(gdqp);
1409 xfs_qm_dqrele(pdqp);
1410
Brian Foster330033d2014-04-17 08:15:30 +10001411 *ipp = ip;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001412 return 0;
1413
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001414 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001415 xfs_trans_cancel(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001416 out_release_inode:
1417 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001418 * Wait until after the current transaction is aborted to finish the
1419 * setup of the inode and release the inode. This prevents recursive
1420 * transactions and deadlocks from xfs_inactive.
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001421 */
Dave Chinner58c90472015-02-23 22:38:08 +11001422 if (ip) {
1423 xfs_finish_inode_setup(ip);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001424 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001425 }
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001426
1427 xfs_qm_dqrele(udqp);
1428 xfs_qm_dqrele(gdqp);
1429 xfs_qm_dqrele(pdqp);
1430
1431 return error;
1432}
1433
1434int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001435xfs_link(
1436 xfs_inode_t *tdp,
1437 xfs_inode_t *sip,
1438 struct xfs_name *target_name)
1439{
1440 xfs_mount_t *mp = tdp->i_mount;
1441 xfs_trans_t *tp;
1442 int error;
1443 xfs_bmap_free_t free_list;
1444 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001445 int resblks;
1446
1447 trace_xfs_link(tdp, target_name);
1448
1449 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1450
1451 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001452 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001453
1454 error = xfs_qm_dqattach(sip, 0);
1455 if (error)
1456 goto std_return;
1457
1458 error = xfs_qm_dqattach(tdp, 0);
1459 if (error)
1460 goto std_return;
1461
1462 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001463 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001464 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001465 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001466 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001467 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001468 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001469 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001470 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001471
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001472 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001473 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1474
1475 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001476 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001477
1478 /*
1479 * If we are using project inheritance, we only allow hard link
1480 * creation in our tree when the project IDs are the same; else
1481 * the tree quota mechanism could be circumvented.
1482 */
1483 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1484 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10001485 error = -EXDEV;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001486 goto error_return;
1487 }
1488
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001489 if (!resblks) {
1490 error = xfs_dir_canenter(tp, tdp, target_name);
1491 if (error)
1492 goto error_return;
1493 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001494
1495 xfs_bmap_init(&free_list, &first_block);
1496
Zhi Yong Wuab297432013-12-18 08:22:41 +08001497 if (sip->i_d.di_nlink == 0) {
1498 error = xfs_iunlink_remove(tp, sip);
1499 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001500 goto error_return;
Zhi Yong Wuab297432013-12-18 08:22:41 +08001501 }
1502
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001503 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1504 &first_block, &free_list, resblks);
1505 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001506 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001507 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1508 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1509
1510 error = xfs_bumplink(tp, sip);
1511 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001512 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001513
1514 /*
1515 * If this is a synchronous mount, make sure that the
1516 * link transaction goes to disk before returning to
1517 * the user.
1518 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001519 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001520 xfs_trans_set_sync(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001521
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001522 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001523 if (error) {
1524 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001525 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001526 }
1527
Christoph Hellwig70393312015-06-04 13:48:08 +10001528 return xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001529
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001530 error_return:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001531 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001532 std_return:
1533 return error;
1534}
1535
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001537 * Free up the underlying blocks past new_size. The new size must be smaller
1538 * than the current size. This routine can be used both for the attribute and
1539 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 *
David Chinnerf6485052008-04-17 16:50:04 +10001541 * The transaction passed to this routine must have made a permanent log
1542 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1543 * given transaction and start new ones, so make sure everything involved in
1544 * the transaction is tidy before calling here. Some transaction will be
1545 * returned to the caller to be committed. The incoming transaction must
1546 * already include the inode, and both inode locks must be held exclusively.
1547 * The inode must also be "held" within the transaction. On return the inode
1548 * will be "held" within the returned transaction. This routine does NOT
1549 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 *
David Chinnerf6485052008-04-17 16:50:04 +10001551 * If we get an error, we must return with the inode locked and linked into the
1552 * current transaction. This keeps things simple for the higher level code,
1553 * because it always knows that the inode is locked and held in the transaction
1554 * that returns to it whether errors occur or not. We don't mark the inode
1555 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 */
1557int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001558xfs_itruncate_extents(
1559 struct xfs_trans **tpp,
1560 struct xfs_inode *ip,
1561 int whichfork,
1562 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001564 struct xfs_mount *mp = ip->i_mount;
1565 struct xfs_trans *tp = *tpp;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001566 xfs_bmap_free_t free_list;
1567 xfs_fsblock_t first_block;
1568 xfs_fileoff_t first_unmap_block;
1569 xfs_fileoff_t last_block;
1570 xfs_filblks_t unmap_len;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001571 int error = 0;
1572 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Christoph Hellwig0b561852012-07-04 11:13:31 -04001574 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1575 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1576 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001577 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001578 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001580 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001581 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001583 trace_xfs_itruncate_extents_start(ip, new_size);
1584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 /*
1586 * Since it is possible for space to become allocated beyond
1587 * the end of the file (in a crash where the space is allocated
1588 * but the inode size is not yet updated), simply remove any
1589 * blocks which show up between the new EOF and the maximum
1590 * possible file size. If the first block to be removed is
1591 * beyond the maximum file size (ie it is the same as last_block),
1592 * then there is nothing to do.
1593 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001594 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001595 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001596 if (first_unmap_block == last_block)
1597 return 0;
1598
1599 ASSERT(first_unmap_block < last_block);
1600 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001602 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001603 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001604 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001605 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001607 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001608 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001609 if (error)
1610 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612 /*
1613 * Duplicate the transaction that has the permanent
1614 * reservation and commit the old transaction.
1615 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001616 error = xfs_bmap_finish(&tp, &free_list, ip);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001617 if (error)
1618 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001620 error = xfs_trans_roll(&tp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001621 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001622 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001624
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001625 /*
1626 * Always re-log the inode so that our permanent transaction can keep
1627 * on rolling it forward in the log.
1628 */
1629 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1630
1631 trace_xfs_itruncate_extents_end(ip, new_size);
1632
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001633out:
1634 *tpp = tp;
1635 return error;
1636out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001638 * If the bunmapi call encounters an error, return to the caller where
1639 * the transaction can be properly aborted. We just need to make sure
1640 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001642 xfs_bmap_cancel(&free_list);
1643 goto out;
1644}
1645
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001646int
1647xfs_release(
1648 xfs_inode_t *ip)
1649{
1650 xfs_mount_t *mp = ip->i_mount;
1651 int error;
1652
1653 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1654 return 0;
1655
1656 /* If this is a read-only mount, don't do this (would generate I/O) */
1657 if (mp->m_flags & XFS_MOUNT_RDONLY)
1658 return 0;
1659
1660 if (!XFS_FORCED_SHUTDOWN(mp)) {
1661 int truncated;
1662
1663 /*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001664 * If we previously truncated this file and removed old data
1665 * in the process, we want to initiate "early" writeout on
1666 * the last close. This is an attempt to combat the notorious
1667 * NULL files problem which is particularly noticeable from a
1668 * truncate down, buffered (re-)write (delalloc), followed by
1669 * a crash. What we are effectively doing here is
1670 * significantly reducing the time window where we'd otherwise
1671 * be exposed to that problem.
1672 */
1673 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1674 if (truncated) {
1675 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
Dave Chinnereac152b2014-08-04 13:22:49 +10001676 if (ip->i_delayed_blks > 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001677 error = filemap_flush(VFS_I(ip)->i_mapping);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001678 if (error)
1679 return error;
1680 }
1681 }
1682 }
1683
1684 if (ip->i_d.di_nlink == 0)
1685 return 0;
1686
1687 if (xfs_can_free_eofblocks(ip, false)) {
1688
1689 /*
1690 * If we can't get the iolock just skip truncating the blocks
1691 * past EOF because we could deadlock with the mmap_sem
1692 * otherwise. We'll get another chance to drop them once the
1693 * last reference to the inode is dropped, so we'll never leak
1694 * blocks permanently.
1695 *
1696 * Further, check if the inode is being opened, written and
1697 * closed frequently and we have delayed allocation blocks
1698 * outstanding (e.g. streaming writes from the NFS server),
1699 * truncating the blocks past EOF will cause fragmentation to
1700 * occur.
1701 *
1702 * In this case don't do the truncation, either, but we have to
1703 * be careful how we detect this case. Blocks beyond EOF show
1704 * up as i_delayed_blks even when the inode is clean, so we
1705 * need to truncate them away first before checking for a dirty
1706 * release. Hence on the first dirty close we will still remove
1707 * the speculative allocation, but after that we will leave it
1708 * in place.
1709 */
1710 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1711 return 0;
1712
1713 error = xfs_free_eofblocks(mp, ip, true);
Dave Chinner24513372014-06-25 14:58:08 +10001714 if (error && error != -EAGAIN)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001715 return error;
1716
1717 /* delalloc blocks after truncation means it really is dirty */
1718 if (ip->i_delayed_blks)
1719 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1720 }
1721 return 0;
1722}
1723
1724/*
Brian Fosterf7be2d72013-09-20 11:06:10 -04001725 * xfs_inactive_truncate
1726 *
1727 * Called to perform a truncate when an inode becomes unlinked.
1728 */
1729STATIC int
1730xfs_inactive_truncate(
1731 struct xfs_inode *ip)
1732{
1733 struct xfs_mount *mp = ip->i_mount;
1734 struct xfs_trans *tp;
1735 int error;
1736
1737 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1738 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1739 if (error) {
1740 ASSERT(XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +10001741 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001742 return error;
1743 }
1744
1745 xfs_ilock(ip, XFS_ILOCK_EXCL);
1746 xfs_trans_ijoin(tp, ip, 0);
1747
1748 /*
1749 * Log the inode size first to prevent stale data exposure in the event
1750 * of a system crash before the truncate completes. See the related
1751 * comment in xfs_setattr_size() for details.
1752 */
1753 ip->i_d.di_size = 0;
1754 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1755
1756 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1757 if (error)
1758 goto error_trans_cancel;
1759
1760 ASSERT(ip->i_d.di_nextents == 0);
1761
Christoph Hellwig70393312015-06-04 13:48:08 +10001762 error = xfs_trans_commit(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001763 if (error)
1764 goto error_unlock;
1765
1766 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1767 return 0;
1768
1769error_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001770 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001771error_unlock:
1772 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1773 return error;
1774}
1775
1776/*
Brian Foster88877d22013-09-20 11:06:11 -04001777 * xfs_inactive_ifree()
1778 *
1779 * Perform the inode free when an inode is unlinked.
1780 */
1781STATIC int
1782xfs_inactive_ifree(
1783 struct xfs_inode *ip)
1784{
1785 xfs_bmap_free_t free_list;
1786 xfs_fsblock_t first_block;
Brian Foster88877d22013-09-20 11:06:11 -04001787 struct xfs_mount *mp = ip->i_mount;
1788 struct xfs_trans *tp;
1789 int error;
1790
1791 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
Brian Foster9d43b182014-04-24 16:00:52 +10001792
1793 /*
1794 * The ifree transaction might need to allocate blocks for record
1795 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1796 * allow ifree to dip into the reserved block pool if necessary.
1797 *
1798 * Freeing large sets of inodes generally means freeing inode chunks,
1799 * directory and file data blocks, so this should be relatively safe.
1800 * Only under severe circumstances should it be possible to free enough
1801 * inodes to exhaust the reserve block pool via finobt expansion while
1802 * at the same time not creating free space in the filesystem.
1803 *
1804 * Send a warning if the reservation does happen to fail, as the inode
1805 * now remains allocated and sits on the unlinked list until the fs is
1806 * repaired.
1807 */
1808 tp->t_flags |= XFS_TRANS_RESERVE;
1809 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1810 XFS_IFREE_SPACE_RES(mp), 0);
Brian Foster88877d22013-09-20 11:06:11 -04001811 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001812 if (error == -ENOSPC) {
Brian Foster9d43b182014-04-24 16:00:52 +10001813 xfs_warn_ratelimited(mp,
1814 "Failed to remove inode(s) from unlinked list. "
1815 "Please free space, unmount and run xfs_repair.");
1816 } else {
1817 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1818 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001819 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001820 return error;
1821 }
1822
1823 xfs_ilock(ip, XFS_ILOCK_EXCL);
1824 xfs_trans_ijoin(tp, ip, 0);
1825
1826 xfs_bmap_init(&free_list, &first_block);
1827 error = xfs_ifree(tp, ip, &free_list);
1828 if (error) {
1829 /*
1830 * If we fail to free the inode, shut down. The cancel
1831 * might do that, we need to make sure. Otherwise the
1832 * inode might be lost for a long time or forever.
1833 */
1834 if (!XFS_FORCED_SHUTDOWN(mp)) {
1835 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1836 __func__, error);
1837 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1838 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001839 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001840 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1841 return error;
1842 }
1843
1844 /*
1845 * Credit the quota account(s). The inode is gone.
1846 */
1847 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1848
1849 /*
Brian Fosterd4a97a02015-08-19 10:01:40 +10001850 * Just ignore errors at this point. There is nothing we can do except
1851 * to try to keep going. Make sure it's not a silent error.
Brian Foster88877d22013-09-20 11:06:11 -04001852 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001853 error = xfs_bmap_finish(&tp, &free_list, NULL);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001854 if (error) {
Brian Foster88877d22013-09-20 11:06:11 -04001855 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1856 __func__, error);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001857 xfs_bmap_cancel(&free_list);
1858 }
Christoph Hellwig70393312015-06-04 13:48:08 +10001859 error = xfs_trans_commit(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001860 if (error)
1861 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1862 __func__, error);
1863
1864 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1865 return 0;
1866}
1867
1868/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001869 * xfs_inactive
1870 *
1871 * This is called when the vnode reference count for the vnode
1872 * goes to zero. If the file has been unlinked, then it must
1873 * now be truncated. Also, we clear all of the read-ahead state
1874 * kept for the inode here since the file is now closed.
1875 */
Brian Foster74564fb2013-09-20 11:06:12 -04001876void
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001877xfs_inactive(
1878 xfs_inode_t *ip)
1879{
Jie Liu3d3c8b52013-08-12 20:49:59 +10001880 struct xfs_mount *mp;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001881 int error;
1882 int truncate = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001883
1884 /*
1885 * If the inode is already free, then there can be nothing
1886 * to clean up here.
1887 */
Ben Myersd9487092013-09-10 18:11:22 -05001888 if (ip->i_d.di_mode == 0) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001889 ASSERT(ip->i_df.if_real_bytes == 0);
1890 ASSERT(ip->i_df.if_broot_bytes == 0);
Brian Foster74564fb2013-09-20 11:06:12 -04001891 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001892 }
1893
1894 mp = ip->i_mount;
1895
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001896 /* If this is a read-only mount, don't do this (would generate I/O) */
1897 if (mp->m_flags & XFS_MOUNT_RDONLY)
Brian Foster74564fb2013-09-20 11:06:12 -04001898 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001899
1900 if (ip->i_d.di_nlink != 0) {
1901 /*
1902 * force is true because we are evicting an inode from the
1903 * cache. Post-eof blocks must be freed, lest we end up with
1904 * broken free space accounting.
1905 */
Brian Foster74564fb2013-09-20 11:06:12 -04001906 if (xfs_can_free_eofblocks(ip, true))
1907 xfs_free_eofblocks(mp, ip, false);
1908
1909 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001910 }
1911
1912 if (S_ISREG(ip->i_d.di_mode) &&
1913 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1914 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1915 truncate = 1;
1916
1917 error = xfs_qm_dqattach(ip, 0);
1918 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001919 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001920
Brian Fosterf7be2d72013-09-20 11:06:10 -04001921 if (S_ISLNK(ip->i_d.di_mode))
Brian Foster36b21dd2013-09-20 11:06:09 -04001922 error = xfs_inactive_symlink(ip);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001923 else if (truncate)
1924 error = xfs_inactive_truncate(ip);
1925 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001926 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001927
1928 /*
1929 * If there are attributes associated with the file then blow them away
1930 * now. The code calls a routine that recursively deconstructs the
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001931 * attribute fork. If also blows away the in-core attribute fork.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001932 */
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001933 if (XFS_IFORK_Q(ip)) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001934 error = xfs_attr_inactive(ip);
1935 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001936 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001937 }
1938
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001939 ASSERT(!ip->i_afp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001940 ASSERT(ip->i_d.di_anextents == 0);
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001941 ASSERT(ip->i_d.di_forkoff == 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001942
1943 /*
1944 * Free the inode.
1945 */
Brian Foster88877d22013-09-20 11:06:11 -04001946 error = xfs_inactive_ifree(ip);
1947 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001948 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001949
1950 /*
1951 * Release the dquots held by inode, if any.
1952 */
1953 xfs_qm_dqdetach(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001954}
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956/*
1957 * This is called when the inode's link count goes to 0.
1958 * We place the on-disk inode on a list in the AGI. It
1959 * will be pulled from this list when the inode is freed.
1960 */
1961int
1962xfs_iunlink(
1963 xfs_trans_t *tp,
1964 xfs_inode_t *ip)
1965{
1966 xfs_mount_t *mp;
1967 xfs_agi_t *agi;
1968 xfs_dinode_t *dip;
1969 xfs_buf_t *agibp;
1970 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 xfs_agino_t agino;
1972 short bucket_index;
1973 int offset;
1974 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
1976 ASSERT(ip->i_d.di_nlink == 0);
1977 ASSERT(ip->i_d.di_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 mp = tp->t_mountp;
1980
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 /*
1982 * Get the agi buffer first. It ensures lock ordering
1983 * on the list.
1984 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001985 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001986 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 /*
1991 * Get the index into the agi hash table for the
1992 * list this inode will go on.
1993 */
1994 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1995 ASSERT(agino != 0);
1996 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1997 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001998 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002000 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 /*
2002 * There is already another inode in the bucket we need
2003 * to add ourselves to. Add us at the front of the list.
2004 * Here we put the head pointer into our next pointer,
2005 * and then we fall through to point the head at us.
2006 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002007 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2008 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002009 if (error)
2010 return error;
2011
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002012 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002014 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002016
2017 /* need to recalc the inode CRC if appropriate */
2018 xfs_dinode_calc_crc(mp, dip);
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 xfs_trans_inode_buf(tp, ibp);
2021 xfs_trans_log_buf(tp, ibp, offset,
2022 (offset + sizeof(xfs_agino_t) - 1));
2023 xfs_inobp_check(mp, ibp);
2024 }
2025
2026 /*
2027 * Point the bucket head pointer at the inode being inserted.
2028 */
2029 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002030 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 offset = offsetof(xfs_agi_t, agi_unlinked) +
2032 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002033 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 xfs_trans_log_buf(tp, agibp, offset,
2035 (offset + sizeof(xfs_agino_t) - 1));
2036 return 0;
2037}
2038
2039/*
2040 * Pull the on-disk inode from the AGI unlinked list.
2041 */
2042STATIC int
2043xfs_iunlink_remove(
2044 xfs_trans_t *tp,
2045 xfs_inode_t *ip)
2046{
2047 xfs_ino_t next_ino;
2048 xfs_mount_t *mp;
2049 xfs_agi_t *agi;
2050 xfs_dinode_t *dip;
2051 xfs_buf_t *agibp;
2052 xfs_buf_t *ibp;
2053 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 xfs_agino_t agino;
2055 xfs_agino_t next_agino;
2056 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002057 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002059 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
2065 /*
2066 * Get the agi buffer first. It ensures lock ordering
2067 * on the list.
2068 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002069 error = xfs_read_agi(mp, tp, agno, &agibp);
2070 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 /*
2076 * Get the index into the agi hash table for the
2077 * list this inode will go on.
2078 */
2079 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2080 ASSERT(agino != 0);
2081 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002082 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 ASSERT(agi->agi_unlinked[bucket_index]);
2084
Christoph Hellwig16259e72005-11-02 15:11:25 +11002085 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002087 * We're at the head of the list. Get the inode's on-disk
2088 * buffer to see if there is anyone after us on the list.
2089 * Only modify our next pointer if it is not already NULLAGINO.
2090 * This saves us the overhead of dealing with the buffer when
2091 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002093 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2094 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002096 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002097 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return error;
2099 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002100 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 ASSERT(next_agino != 0);
2102 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002103 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002104 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002106
2107 /* need to recalc the inode CRC if appropriate */
2108 xfs_dinode_calc_crc(mp, dip);
2109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 xfs_trans_inode_buf(tp, ibp);
2111 xfs_trans_log_buf(tp, ibp, offset,
2112 (offset + sizeof(xfs_agino_t) - 1));
2113 xfs_inobp_check(mp, ibp);
2114 } else {
2115 xfs_trans_brelse(tp, ibp);
2116 }
2117 /*
2118 * Point the bucket head pointer at the next inode.
2119 */
2120 ASSERT(next_agino != 0);
2121 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002122 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 offset = offsetof(xfs_agi_t, agi_unlinked) +
2124 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002125 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 xfs_trans_log_buf(tp, agibp, offset,
2127 (offset + sizeof(xfs_agino_t) - 1));
2128 } else {
2129 /*
2130 * We need to search the list for the inode being freed.
2131 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002132 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 last_ibp = NULL;
2134 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002135 struct xfs_imap imap;
2136
2137 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002139
2140 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002142
2143 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002145 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002146 "%s: xfs_imap returned error %d.",
2147 __func__, error);
2148 return error;
2149 }
2150
2151 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2152 &last_ibp, 0, 0);
2153 if (error) {
2154 xfs_warn(mp,
2155 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002156 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 return error;
2158 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002159
2160 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002161 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 ASSERT(next_agino != NULLAGINO);
2163 ASSERT(next_agino != 0);
2164 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04002165
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002167 * Now last_ibp points to the buffer previous to us on the
2168 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002170 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2171 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002173 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002174 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 return error;
2176 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002177 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 ASSERT(next_agino != 0);
2179 ASSERT(next_agino != agino);
2180 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002181 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002182 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002184
2185 /* need to recalc the inode CRC if appropriate */
2186 xfs_dinode_calc_crc(mp, dip);
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 xfs_trans_inode_buf(tp, ibp);
2189 xfs_trans_log_buf(tp, ibp, offset,
2190 (offset + sizeof(xfs_agino_t) - 1));
2191 xfs_inobp_check(mp, ibp);
2192 } else {
2193 xfs_trans_brelse(tp, ibp);
2194 }
2195 /*
2196 * Point the previous inode on the list to the next inode.
2197 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002198 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 ASSERT(next_agino != 0);
2200 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002201
2202 /* need to recalc the inode CRC if appropriate */
2203 xfs_dinode_calc_crc(mp, last_dip);
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 xfs_trans_inode_buf(tp, last_ibp);
2206 xfs_trans_log_buf(tp, last_ibp, offset,
2207 (offset + sizeof(xfs_agino_t) - 1));
2208 xfs_inobp_check(mp, last_ibp);
2209 }
2210 return 0;
2211}
2212
Dave Chinner5b3eed72010-08-24 11:42:41 +10002213/*
Zhi Yong Wu0b8182d2013-08-12 03:14:59 +00002214 * A big issue when freeing the inode cluster is that we _cannot_ skip any
Dave Chinner5b3eed72010-08-24 11:42:41 +10002215 * inodes that are in memory - they all must be marked stale and attached to
2216 * the cluster buffer.
2217 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002218STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219xfs_ifree_cluster(
Brian Foster09b56602015-05-29 09:26:03 +10002220 xfs_inode_t *free_ip,
2221 xfs_trans_t *tp,
2222 struct xfs_icluster *xic)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
2224 xfs_mount_t *mp = free_ip->i_mount;
2225 int blks_per_cluster;
Jie Liu982e9392013-12-13 15:51:49 +11002226 int inodes_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 int nbufs;
Dave Chinner5b257b42010-06-03 16:22:29 +10002228 int i, j;
Brian Foster3cdaa182015-06-04 13:03:34 +10002229 int ioffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 xfs_daddr_t blkno;
2231 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10002232 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 xfs_inode_log_item_t *iip;
2234 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00002235 struct xfs_perag *pag;
Brian Foster09b56602015-05-29 09:26:03 +10002236 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Brian Foster09b56602015-05-29 09:26:03 +10002238 inum = xic->first_ino;
Dave Chinner5017e972010-01-11 11:47:40 +00002239 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Jie Liu982e9392013-12-13 15:51:49 +11002240 blks_per_cluster = xfs_icluster_size_fsb(mp);
2241 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2242 nbufs = mp->m_ialloc_blks / blks_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
Jie Liu982e9392013-12-13 15:51:49 +11002244 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
Brian Foster09b56602015-05-29 09:26:03 +10002245 /*
2246 * The allocation bitmap tells us which inodes of the chunk were
2247 * physically allocated. Skip the cluster if an inode falls into
2248 * a sparse region.
2249 */
Brian Foster3cdaa182015-06-04 13:03:34 +10002250 ioffset = inum - xic->first_ino;
2251 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2252 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
Brian Foster09b56602015-05-29 09:26:03 +10002253 continue;
2254 }
2255
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2257 XFS_INO_TO_AGBNO(mp, inum));
2258
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10002260 * We obtain and lock the backing buffer first in the process
2261 * here, as we have to ensure that any dirty inode that we
2262 * can't get the flush lock on is attached to the buffer.
2263 * If we scan the in-memory inodes first, then buffer IO can
2264 * complete before we get a lock on it, and hence we may fail
2265 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002267 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11002268 mp->m_bsize * blks_per_cluster,
2269 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002271 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002272 return -ENOMEM;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002273
2274 /*
2275 * This buffer may not have been correctly initialised as we
2276 * didn't read it from disk. That's not important because we are
2277 * only using to mark the buffer as stale in the log, and to
2278 * attach stale cached inodes on it. That means it will never be
2279 * dispatched for IO. If it is, we want to know about it, and we
2280 * want it to fail. We can acheive this by adding a write
2281 * verifier to the buffer.
2282 */
Dave Chinner1813dd62012-11-14 17:54:40 +11002283 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002284
Dave Chinner5b257b42010-06-03 16:22:29 +10002285 /*
2286 * Walk the inodes already attached to the buffer and mark them
2287 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10002288 * in-memory inode walk can't lock them. By marking them all
2289 * stale first, we will not attempt to lock them in the loop
2290 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10002291 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002292 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 while (lip) {
2294 if (lip->li_type == XFS_LI_INODE) {
2295 iip = (xfs_inode_log_item_t *)lip;
2296 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002297 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002298 xfs_trans_ail_copy_lsn(mp->m_ail,
2299 &iip->ili_flush_lsn,
2300 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002301 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 }
2303 lip = lip->li_bio_list;
2304 }
2305
Dave Chinner5b3eed72010-08-24 11:42:41 +10002306
Dave Chinner5b257b42010-06-03 16:22:29 +10002307 /*
2308 * For each inode in memory attempt to add it to the inode
2309 * buffer and set it up for being staled on buffer IO
2310 * completion. This is safe as we've locked out tail pushing
2311 * and flushing by locking the buffer.
2312 *
2313 * We have already marked every inode that was part of a
2314 * transaction stale above, which means there is no point in
2315 * even trying to lock them.
2316 */
Jie Liu982e9392013-12-13 15:51:49 +11002317 for (i = 0; i < inodes_per_cluster; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10002318retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002319 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002320 ip = radix_tree_lookup(&pag->pag_ici_root,
2321 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002323 /* Inode not in memory, nothing to do */
2324 if (!ip) {
2325 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002326 continue;
2327 }
2328
Dave Chinner5b3eed72010-08-24 11:42:41 +10002329 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002330 * because this is an RCU protected lookup, we could
2331 * find a recently freed or even reallocated inode
2332 * during the lookup. We need to check under the
2333 * i_flags_lock for a valid inode here. Skip it if it
2334 * is not valid, the wrong inode or stale.
2335 */
2336 spin_lock(&ip->i_flags_lock);
2337 if (ip->i_ino != inum + i ||
2338 __xfs_iflags_test(ip, XFS_ISTALE)) {
2339 spin_unlock(&ip->i_flags_lock);
2340 rcu_read_unlock();
2341 continue;
2342 }
2343 spin_unlock(&ip->i_flags_lock);
2344
2345 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002346 * Don't try to lock/unlock the current inode, but we
2347 * _cannot_ skip the other inodes that we did not find
2348 * in the list attached to the buffer and are not
2349 * already marked stale. If we can't lock it, back off
2350 * and retry.
2351 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002352 if (ip != free_ip &&
2353 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002354 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002355 delay(1);
2356 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002357 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002358 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002359
Dave Chinner5b3eed72010-08-24 11:42:41 +10002360 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002361 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002362
Dave Chinner5b3eed72010-08-24 11:42:41 +10002363 /*
2364 * we don't need to attach clean inodes or those only
2365 * with unlogged changes (which we throw away, anyway).
2366 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002367 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002368 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002369 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 xfs_ifunlock(ip);
2371 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2372 continue;
2373 }
2374
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002375 iip->ili_last_fields = iip->ili_fields;
2376 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11002377 iip->ili_fsync_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002379 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2380 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002382 xfs_buf_attach_iodone(bp, xfs_istale_done,
2383 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002384
2385 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 }
2388
Dave Chinner5b3eed72010-08-24 11:42:41 +10002389 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 xfs_trans_binval(tp, bp);
2391 }
2392
Dave Chinner5017e972010-01-11 11:47:40 +00002393 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002394 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395}
2396
2397/*
2398 * This is called to return an inode to the inode free list.
2399 * The inode should already be truncated to 0 length and have
2400 * no pages associated with it. This routine also assumes that
2401 * the inode is already a part of the transaction.
2402 *
2403 * The on-disk copy of the inode will have been added to the list
2404 * of unlinked inodes in the AGI. We need to remove the inode from
2405 * that list atomically with respect to freeing it here.
2406 */
2407int
2408xfs_ifree(
2409 xfs_trans_t *tp,
2410 xfs_inode_t *ip,
2411 xfs_bmap_free_t *flist)
2412{
2413 int error;
Brian Foster09b56602015-05-29 09:26:03 +10002414 struct xfs_icluster xic = { 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002416 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 ASSERT(ip->i_d.di_nlink == 0);
2418 ASSERT(ip->i_d.di_nextents == 0);
2419 ASSERT(ip->i_d.di_anextents == 0);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002420 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 ASSERT(ip->i_d.di_nblocks == 0);
2422
2423 /*
2424 * Pull the on-disk inode from the AGI unlinked list.
2425 */
2426 error = xfs_iunlink_remove(tp, ip);
Dave Chinner1baaed82013-06-27 16:04:50 +10002427 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Brian Foster09b56602015-05-29 09:26:03 +10002430 error = xfs_difree(tp, ip->i_ino, flist, &xic);
Dave Chinner1baaed82013-06-27 16:04:50 +10002431 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 return error;
Dave Chinner1baaed82013-06-27 16:04:50 +10002433
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 ip->i_d.di_mode = 0; /* mark incore inode as free */
2435 ip->i_d.di_flags = 0;
2436 ip->i_d.di_dmevmask = 0;
2437 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2439 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2440 /*
2441 * Bump the generation count so no one will be confused
2442 * by reincarnations of this inode.
2443 */
2444 ip->i_d.di_gen++;
2445 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2446
Brian Foster09b56602015-05-29 09:26:03 +10002447 if (xic.deleted)
2448 error = xfs_ifree_cluster(ip, tp, &xic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002450 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451}
2452
2453/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002454 * This is called to unpin an inode. The caller must have the inode locked
2455 * in at least shared mode so that the buffer cannot be subsequently pinned
2456 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002458static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002459xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002460 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002461{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002462 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002463
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002464 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2465
David Chinnera3f74ff2008-03-06 13:43:42 +11002466 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002467 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002468
David Chinnera3f74ff2008-03-06 13:43:42 +11002469}
2470
Christoph Hellwigf392e632011-12-18 20:00:10 +00002471static void
2472__xfs_iunpin_wait(
2473 struct xfs_inode *ip)
2474{
2475 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2476 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2477
2478 xfs_iunpin(ip);
2479
2480 do {
2481 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2482 if (xfs_ipincount(ip))
2483 io_schedule();
2484 } while (xfs_ipincount(ip));
2485 finish_wait(wq, &wait.wait);
2486}
2487
Dave Chinner777df5a2010-02-06 12:37:26 +11002488void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002490 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002492 if (xfs_ipincount(ip))
2493 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002494}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Dave Chinner27320362013-10-29 22:11:44 +11002496/*
2497 * Removing an inode from the namespace involves removing the directory entry
2498 * and dropping the link count on the inode. Removing the directory entry can
2499 * result in locking an AGF (directory blocks were freed) and removing a link
2500 * count can result in placing the inode on an unlinked list which results in
2501 * locking an AGI.
2502 *
2503 * The big problem here is that we have an ordering constraint on AGF and AGI
2504 * locking - inode allocation locks the AGI, then can allocate a new extent for
2505 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2506 * removes the inode from the unlinked list, requiring that we lock the AGI
2507 * first, and then freeing the inode can result in an inode chunk being freed
2508 * and hence freeing disk space requiring that we lock an AGF.
2509 *
2510 * Hence the ordering that is imposed by other parts of the code is AGI before
2511 * AGF. This means we cannot remove the directory entry before we drop the inode
2512 * reference count and put it on the unlinked list as this results in a lock
2513 * order of AGF then AGI, and this can deadlock against inode allocation and
2514 * freeing. Therefore we must drop the link counts before we remove the
2515 * directory entry.
2516 *
2517 * This is still safe from a transactional point of view - it is not until we
2518 * get to xfs_bmap_finish() that we have the possibility of multiple
2519 * transactions in this operation. Hence as long as we remove the directory
2520 * entry and drop the link count in the first transaction of the remove
2521 * operation, there are no transactional constraints on the ordering here.
2522 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002523int
2524xfs_remove(
2525 xfs_inode_t *dp,
2526 struct xfs_name *name,
2527 xfs_inode_t *ip)
2528{
2529 xfs_mount_t *mp = dp->i_mount;
2530 xfs_trans_t *tp = NULL;
2531 int is_dir = S_ISDIR(ip->i_d.di_mode);
2532 int error = 0;
2533 xfs_bmap_free_t free_list;
2534 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002535 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002536
2537 trace_xfs_remove(dp, name);
2538
2539 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10002540 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002541
2542 error = xfs_qm_dqattach(dp, 0);
2543 if (error)
2544 goto std_return;
2545
2546 error = xfs_qm_dqattach(ip, 0);
2547 if (error)
2548 goto std_return;
2549
Dave Chinner32296f82014-12-04 09:43:17 +11002550 if (is_dir)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002551 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
Dave Chinner32296f82014-12-04 09:43:17 +11002552 else
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002553 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002554
2555 /*
2556 * We try to get the real space reservation first,
2557 * allowing for directory btree deletion(s) implying
2558 * possible bmap insert(s). If we can't get the space
2559 * reservation then we use 0 instead, and avoid the bmap
2560 * btree insert(s) in the directory code by, if the bmap
2561 * insert tries to happen, instead trimming the LAST
2562 * block from the directory.
2563 */
2564 resblks = XFS_REMOVE_SPACE_RES(mp);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002565 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002566 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002567 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002568 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002569 }
2570 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002571 ASSERT(error != -ENOSPC);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002572 goto out_trans_cancel;
2573 }
2574
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002575 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002576 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2577
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002578 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002579 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2580
2581 /*
2582 * If we're removing a directory perform some additional validation.
2583 */
2584 if (is_dir) {
2585 ASSERT(ip->i_d.di_nlink >= 2);
2586 if (ip->i_d.di_nlink != 2) {
Dave Chinner24513372014-06-25 14:58:08 +10002587 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002588 goto out_trans_cancel;
2589 }
2590 if (!xfs_dir_isempty(ip)) {
Dave Chinner24513372014-06-25 14:58:08 +10002591 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002592 goto out_trans_cancel;
2593 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002594
Dave Chinner27320362013-10-29 22:11:44 +11002595 /* Drop the link from ip's "..". */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002596 error = xfs_droplink(tp, dp);
2597 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002598 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002599
Dave Chinner27320362013-10-29 22:11:44 +11002600 /* Drop the "." link from ip to self. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002601 error = xfs_droplink(tp, ip);
2602 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002603 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002604 } else {
2605 /*
2606 * When removing a non-directory we need to log the parent
2607 * inode here. For a directory this is done implicitly
2608 * by the xfs_droplink call for the ".." entry.
2609 */
2610 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2611 }
Dave Chinner27320362013-10-29 22:11:44 +11002612 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002613
Dave Chinner27320362013-10-29 22:11:44 +11002614 /* Drop the link from dp to ip. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002615 error = xfs_droplink(tp, ip);
2616 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002617 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002618
Dave Chinner27320362013-10-29 22:11:44 +11002619 xfs_bmap_init(&free_list, &first_block);
2620 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2621 &first_block, &free_list, resblks);
2622 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002623 ASSERT(error != -ENOENT);
Dave Chinner27320362013-10-29 22:11:44 +11002624 goto out_bmap_cancel;
2625 }
2626
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002627 /*
2628 * If this is a synchronous mount, make sure that the
2629 * remove transaction goes to disk before returning to
2630 * the user.
2631 */
2632 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2633 xfs_trans_set_sync(tp);
2634
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002635 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002636 if (error)
2637 goto out_bmap_cancel;
2638
Christoph Hellwig70393312015-06-04 13:48:08 +10002639 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002640 if (error)
2641 goto std_return;
2642
Christoph Hellwig2cd2ef62014-04-23 07:11:51 +10002643 if (is_dir && xfs_inode_is_filestream(ip))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002644 xfs_filestream_deassociate(ip);
2645
2646 return 0;
2647
2648 out_bmap_cancel:
2649 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002650 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10002651 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002652 std_return:
2653 return error;
2654}
2655
Dave Chinnerf6bba202013-08-12 20:49:46 +10002656/*
2657 * Enter all inodes for a rename transaction into a sorted array.
2658 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002659#define __XFS_SORT_INODES 5
Dave Chinnerf6bba202013-08-12 20:49:46 +10002660STATIC void
2661xfs_sort_for_rename(
Dave Chinner95afcf52015-03-25 14:03:32 +11002662 struct xfs_inode *dp1, /* in: old (source) directory inode */
2663 struct xfs_inode *dp2, /* in: new (target) directory inode */
2664 struct xfs_inode *ip1, /* in: inode of old entry */
2665 struct xfs_inode *ip2, /* in: inode of new entry */
2666 struct xfs_inode *wip, /* in: whiteout inode */
2667 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2668 int *num_inodes) /* in/out: inodes in array */
Dave Chinnerf6bba202013-08-12 20:49:46 +10002669{
Dave Chinnerf6bba202013-08-12 20:49:46 +10002670 int i, j;
2671
Dave Chinner95afcf52015-03-25 14:03:32 +11002672 ASSERT(*num_inodes == __XFS_SORT_INODES);
2673 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2674
Dave Chinnerf6bba202013-08-12 20:49:46 +10002675 /*
2676 * i_tab contains a list of pointers to inodes. We initialize
2677 * the table here & we'll sort it. We will then use it to
2678 * order the acquisition of the inode locks.
2679 *
2680 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2681 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002682 i = 0;
2683 i_tab[i++] = dp1;
2684 i_tab[i++] = dp2;
2685 i_tab[i++] = ip1;
2686 if (ip2)
2687 i_tab[i++] = ip2;
2688 if (wip)
2689 i_tab[i++] = wip;
2690 *num_inodes = i;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002691
2692 /*
2693 * Sort the elements via bubble sort. (Remember, there are at
Dave Chinner95afcf52015-03-25 14:03:32 +11002694 * most 5 elements to sort, so this is adequate.)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002695 */
2696 for (i = 0; i < *num_inodes; i++) {
2697 for (j = 1; j < *num_inodes; j++) {
2698 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
Dave Chinner95afcf52015-03-25 14:03:32 +11002699 struct xfs_inode *temp = i_tab[j];
Dave Chinnerf6bba202013-08-12 20:49:46 +10002700 i_tab[j] = i_tab[j-1];
2701 i_tab[j-1] = temp;
2702 }
2703 }
2704 }
2705}
2706
Dave Chinner310606b2015-03-25 14:06:07 +11002707static int
2708xfs_finish_rename(
2709 struct xfs_trans *tp,
2710 struct xfs_bmap_free *free_list)
2711{
Dave Chinner310606b2015-03-25 14:06:07 +11002712 int error;
2713
2714 /*
2715 * If this is a synchronous mount, make sure that the rename transaction
2716 * goes to disk before returning to the user.
2717 */
2718 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2719 xfs_trans_set_sync(tp);
2720
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002721 error = xfs_bmap_finish(&tp, free_list, NULL);
Dave Chinner310606b2015-03-25 14:06:07 +11002722 if (error) {
2723 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002724 xfs_trans_cancel(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002725 return error;
2726 }
2727
Christoph Hellwig70393312015-06-04 13:48:08 +10002728 return xfs_trans_commit(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002729}
2730
Dave Chinnerf6bba202013-08-12 20:49:46 +10002731/*
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002732 * xfs_cross_rename()
2733 *
2734 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2735 */
2736STATIC int
2737xfs_cross_rename(
2738 struct xfs_trans *tp,
2739 struct xfs_inode *dp1,
2740 struct xfs_name *name1,
2741 struct xfs_inode *ip1,
2742 struct xfs_inode *dp2,
2743 struct xfs_name *name2,
2744 struct xfs_inode *ip2,
2745 struct xfs_bmap_free *free_list,
2746 xfs_fsblock_t *first_block,
2747 int spaceres)
2748{
2749 int error = 0;
2750 int ip1_flags = 0;
2751 int ip2_flags = 0;
2752 int dp2_flags = 0;
2753
2754 /* Swap inode number for dirent in first parent */
2755 error = xfs_dir_replace(tp, dp1, name1,
2756 ip2->i_ino,
2757 first_block, free_list, spaceres);
2758 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002759 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002760
2761 /* Swap inode number for dirent in second parent */
2762 error = xfs_dir_replace(tp, dp2, name2,
2763 ip1->i_ino,
2764 first_block, free_list, spaceres);
2765 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002766 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002767
2768 /*
2769 * If we're renaming one or more directories across different parents,
2770 * update the respective ".." entries (and link counts) to match the new
2771 * parents.
2772 */
2773 if (dp1 != dp2) {
2774 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2775
2776 if (S_ISDIR(ip2->i_d.di_mode)) {
2777 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2778 dp1->i_ino, first_block,
2779 free_list, spaceres);
2780 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002781 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002782
2783 /* transfer ip2 ".." reference to dp1 */
2784 if (!S_ISDIR(ip1->i_d.di_mode)) {
2785 error = xfs_droplink(tp, dp2);
2786 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002787 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002788 error = xfs_bumplink(tp, dp1);
2789 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002790 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002791 }
2792
2793 /*
2794 * Although ip1 isn't changed here, userspace needs
2795 * to be warned about the change, so that applications
2796 * relying on it (like backup ones), will properly
2797 * notify the change
2798 */
2799 ip1_flags |= XFS_ICHGTIME_CHG;
2800 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2801 }
2802
2803 if (S_ISDIR(ip1->i_d.di_mode)) {
2804 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2805 dp2->i_ino, first_block,
2806 free_list, spaceres);
2807 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002808 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002809
2810 /* transfer ip1 ".." reference to dp2 */
2811 if (!S_ISDIR(ip2->i_d.di_mode)) {
2812 error = xfs_droplink(tp, dp1);
2813 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002814 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002815 error = xfs_bumplink(tp, dp2);
2816 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002817 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002818 }
2819
2820 /*
2821 * Although ip2 isn't changed here, userspace needs
2822 * to be warned about the change, so that applications
2823 * relying on it (like backup ones), will properly
2824 * notify the change
2825 */
2826 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2827 ip2_flags |= XFS_ICHGTIME_CHG;
2828 }
2829 }
2830
2831 if (ip1_flags) {
2832 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2833 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2834 }
2835 if (ip2_flags) {
2836 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2837 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2838 }
2839 if (dp2_flags) {
2840 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2841 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2842 }
2843 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2844 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
Dave Chinnereeacd322015-03-25 14:08:07 +11002845 return xfs_finish_rename(tp, free_list);
2846
2847out_trans_abort:
2848 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002849 xfs_trans_cancel(tp);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002850 return error;
2851}
2852
2853/*
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002854 * xfs_rename_alloc_whiteout()
2855 *
2856 * Return a referenced, unlinked, unlocked inode that that can be used as a
2857 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2858 * crash between allocating the inode and linking it into the rename transaction
2859 * recovery will free the inode and we won't leak it.
2860 */
2861static int
2862xfs_rename_alloc_whiteout(
2863 struct xfs_inode *dp,
2864 struct xfs_inode **wip)
2865{
2866 struct xfs_inode *tmpfile;
2867 int error;
2868
2869 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2870 if (error)
2871 return error;
2872
Brian Foster22419ac2015-05-29 08:14:55 +10002873 /*
2874 * Prepare the tmpfile inode as if it were created through the VFS.
2875 * Otherwise, the link increment paths will complain about nlink 0->1.
2876 * Drop the link count as done by d_tmpfile(), complete the inode setup
2877 * and flag it as linkable.
2878 */
2879 drop_nlink(VFS_I(tmpfile));
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002880 xfs_finish_inode_setup(tmpfile);
2881 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2882
2883 *wip = tmpfile;
2884 return 0;
2885}
2886
2887/*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002888 * xfs_rename
2889 */
2890int
2891xfs_rename(
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002892 struct xfs_inode *src_dp,
2893 struct xfs_name *src_name,
2894 struct xfs_inode *src_ip,
2895 struct xfs_inode *target_dp,
2896 struct xfs_name *target_name,
2897 struct xfs_inode *target_ip,
2898 unsigned int flags)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002899{
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002900 struct xfs_mount *mp = src_dp->i_mount;
2901 struct xfs_trans *tp;
2902 struct xfs_bmap_free free_list;
2903 xfs_fsblock_t first_block;
2904 struct xfs_inode *wip = NULL; /* whiteout inode */
2905 struct xfs_inode *inodes[__XFS_SORT_INODES];
2906 int num_inodes = __XFS_SORT_INODES;
Dave Chinner2b936812015-03-25 15:12:30 +11002907 bool new_parent = (src_dp != target_dp);
2908 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002909 int spaceres;
2910 int error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002911
2912 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2913
Dave Chinnereeacd322015-03-25 14:08:07 +11002914 if ((flags & RENAME_EXCHANGE) && !target_ip)
2915 return -EINVAL;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002916
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002917 /*
2918 * If we are doing a whiteout operation, allocate the whiteout inode
2919 * we will be placing at the target and ensure the type is set
2920 * appropriately.
2921 */
2922 if (flags & RENAME_WHITEOUT) {
2923 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2924 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2925 if (error)
2926 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002927
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002928 /* setup target dirent info as whiteout */
2929 src_name->type = XFS_DIR3_FT_CHRDEV;
2930 }
2931
2932 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
Dave Chinnerf6bba202013-08-12 20:49:46 +10002933 inodes, &num_inodes);
2934
Dave Chinnerf6bba202013-08-12 20:49:46 +10002935 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002936 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002937 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002938 if (error == -ENOSPC) {
Dave Chinnerf6bba202013-08-12 20:49:46 +10002939 spaceres = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002940 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002941 }
Dave Chinner445883e2015-03-25 14:05:43 +11002942 if (error)
2943 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002944
2945 /*
2946 * Attach the dquots to the inodes
2947 */
2948 error = xfs_qm_vop_rename_dqattach(inodes);
Dave Chinner445883e2015-03-25 14:05:43 +11002949 if (error)
2950 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002951
2952 /*
2953 * Lock all the participating inodes. Depending upon whether
2954 * the target_name exists in the target directory, and
2955 * whether the target directory is the same as the source
2956 * directory, we can lock from 2 to 4 inodes.
2957 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002958 if (!new_parent)
2959 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2960 else
2961 xfs_lock_two_inodes(src_dp, target_dp,
2962 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2963
Dave Chinnerf6bba202013-08-12 20:49:46 +10002964 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2965
2966 /*
2967 * Join all the inodes to the transaction. From this point on,
2968 * we can rely on either trans_commit or trans_cancel to unlock
2969 * them.
2970 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002971 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002972 if (new_parent)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002973 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002974 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2975 if (target_ip)
2976 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002977 if (wip)
2978 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002979
2980 /*
2981 * If we are using project inheritance, we only allow renames
2982 * into our tree when the project IDs are the same; else the
2983 * tree quota mechanism would be circumvented.
2984 */
2985 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2986 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10002987 error = -EXDEV;
Dave Chinner445883e2015-03-25 14:05:43 +11002988 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002989 }
2990
Dave Chinner445883e2015-03-25 14:05:43 +11002991 xfs_bmap_init(&free_list, &first_block);
2992
Dave Chinnereeacd322015-03-25 14:08:07 +11002993 /* RENAME_EXCHANGE is unique from here on. */
2994 if (flags & RENAME_EXCHANGE)
2995 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2996 target_dp, target_name, target_ip,
2997 &free_list, &first_block, spaceres);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002998
2999 /*
Dave Chinnerf6bba202013-08-12 20:49:46 +10003000 * Set up the target.
3001 */
3002 if (target_ip == NULL) {
3003 /*
3004 * If there's no space reservation, check the entry will
3005 * fit before actually inserting it.
3006 */
Eric Sandeen94f3cad2014-09-09 11:57:52 +10003007 if (!spaceres) {
3008 error = xfs_dir_canenter(tp, target_dp, target_name);
3009 if (error)
Dave Chinner445883e2015-03-25 14:05:43 +11003010 goto out_trans_cancel;
Eric Sandeen94f3cad2014-09-09 11:57:52 +10003011 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003012 /*
3013 * If target does not exist and the rename crosses
3014 * directories, adjust the target directory link count
3015 * to account for the ".." reference from the new entry.
3016 */
3017 error = xfs_dir_createname(tp, target_dp, target_name,
3018 src_ip->i_ino, &first_block,
3019 &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003020 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003021 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003022
3023 xfs_trans_ichgtime(tp, target_dp,
3024 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3025
3026 if (new_parent && src_is_directory) {
3027 error = xfs_bumplink(tp, target_dp);
3028 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003029 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003030 }
3031 } else { /* target_ip != NULL */
3032 /*
3033 * If target exists and it's a directory, check that both
3034 * target and source are directories and that target can be
3035 * destroyed, or that neither is a directory.
3036 */
3037 if (S_ISDIR(target_ip->i_d.di_mode)) {
3038 /*
3039 * Make sure target dir is empty.
3040 */
3041 if (!(xfs_dir_isempty(target_ip)) ||
3042 (target_ip->i_d.di_nlink > 2)) {
Dave Chinner24513372014-06-25 14:58:08 +10003043 error = -EEXIST;
Dave Chinner445883e2015-03-25 14:05:43 +11003044 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003045 }
3046 }
3047
3048 /*
3049 * Link the source inode under the target name.
3050 * If the source inode is a directory and we are moving
3051 * it across directories, its ".." entry will be
3052 * inconsistent until we replace that down below.
3053 *
3054 * In case there is already an entry with the same
3055 * name at the destination directory, remove it first.
3056 */
3057 error = xfs_dir_replace(tp, target_dp, target_name,
3058 src_ip->i_ino,
3059 &first_block, &free_list, spaceres);
3060 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003061 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003062
3063 xfs_trans_ichgtime(tp, target_dp,
3064 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3065
3066 /*
3067 * Decrement the link count on the target since the target
3068 * dir no longer points to it.
3069 */
3070 error = xfs_droplink(tp, target_ip);
3071 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003072 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003073
3074 if (src_is_directory) {
3075 /*
3076 * Drop the link from the old "." entry.
3077 */
3078 error = xfs_droplink(tp, target_ip);
3079 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003080 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003081 }
3082 } /* target_ip != NULL */
3083
3084 /*
3085 * Remove the source.
3086 */
3087 if (new_parent && src_is_directory) {
3088 /*
3089 * Rewrite the ".." entry to point to the new
3090 * directory.
3091 */
3092 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3093 target_dp->i_ino,
3094 &first_block, &free_list, spaceres);
Dave Chinner24513372014-06-25 14:58:08 +10003095 ASSERT(error != -EEXIST);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003096 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003097 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003098 }
3099
3100 /*
3101 * We always want to hit the ctime on the source inode.
3102 *
3103 * This isn't strictly required by the standards since the source
3104 * inode isn't really being changed, but old unix file systems did
3105 * it and some incremental backup programs won't work without it.
3106 */
3107 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3108 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3109
3110 /*
3111 * Adjust the link count on src_dp. This is necessary when
3112 * renaming a directory, either within one parent when
3113 * the target existed, or across two parent directories.
3114 */
3115 if (src_is_directory && (new_parent || target_ip != NULL)) {
3116
3117 /*
3118 * Decrement link count on src_directory since the
3119 * entry that's moved no longer points to it.
3120 */
3121 error = xfs_droplink(tp, src_dp);
3122 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003123 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003124 }
3125
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003126 /*
3127 * For whiteouts, we only need to update the source dirent with the
3128 * inode number of the whiteout inode rather than removing it
3129 * altogether.
3130 */
3131 if (wip) {
3132 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
Dave Chinnerf6bba202013-08-12 20:49:46 +10003133 &first_block, &free_list, spaceres);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003134 } else
3135 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3136 &first_block, &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003137 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003138 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003139
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003140 /*
3141 * For whiteouts, we need to bump the link count on the whiteout inode.
3142 * This means that failures all the way up to this point leave the inode
3143 * on the unlinked list and so cleanup is a simple matter of dropping
3144 * the remaining reference to it. If we fail here after bumping the link
3145 * count, we're shutting down the filesystem so we'll never see the
3146 * intermediate state on disk.
3147 */
3148 if (wip) {
Brian Foster22419ac2015-05-29 08:14:55 +10003149 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003150 error = xfs_bumplink(tp, wip);
3151 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003152 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003153 error = xfs_iunlink_remove(tp, wip);
3154 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003155 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003156 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3157
3158 /*
3159 * Now we have a real link, clear the "I'm a tmpfile" state
3160 * flag from the inode so it doesn't accidentally get misused in
3161 * future.
3162 */
3163 VFS_I(wip)->i_state &= ~I_LINKABLE;
3164 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003165
3166 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3167 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3168 if (new_parent)
3169 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3170
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003171 error = xfs_finish_rename(tp, &free_list);
3172 if (wip)
3173 IRELE(wip);
3174 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003175
Dave Chinner445883e2015-03-25 14:05:43 +11003176out_bmap_cancel:
Dave Chinnerf6bba202013-08-12 20:49:46 +10003177 xfs_bmap_cancel(&free_list);
Dave Chinner445883e2015-03-25 14:05:43 +11003178out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003179 xfs_trans_cancel(tp);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003180 if (wip)
3181 IRELE(wip);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003182 return error;
3183}
3184
David Chinnerbad55842008-03-06 13:43:49 +11003185STATIC int
3186xfs_iflush_cluster(
3187 xfs_inode_t *ip,
3188 xfs_buf_t *bp)
3189{
3190 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00003191 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11003192 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10003193 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11003194 int ilist_size;
3195 xfs_inode_t **ilist;
3196 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11003197 int nr_found;
3198 int clcount = 0;
3199 int bufwasdelwri;
3200 int i;
3201
Dave Chinner5017e972010-01-11 11:47:40 +00003202 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11003203
Jie Liu0f49efd2013-12-13 15:51:48 +11003204 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
David Chinnerc8f5f122008-05-20 11:30:15 +10003205 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10003206 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11003207 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00003208 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11003209
Jie Liu0f49efd2013-12-13 15:51:48 +11003210 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
David Chinnerbad55842008-03-06 13:43:49 +11003211 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003212 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11003213 /* really need a gang lookup range call here */
3214 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10003215 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11003216 if (nr_found == 0)
3217 goto out_free;
3218
3219 for (i = 0; i < nr_found; i++) {
3220 iq = ilist[i];
3221 if (iq == ip)
3222 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003223
3224 /*
3225 * because this is an RCU protected lookup, we could find a
3226 * recently freed or even reallocated inode during the lookup.
3227 * We need to check under the i_flags_lock for a valid inode
3228 * here. Skip it if it is not valid or the wrong inode.
3229 */
3230 spin_lock(&ip->i_flags_lock);
3231 if (!ip->i_ino ||
3232 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3233 spin_unlock(&ip->i_flags_lock);
3234 continue;
3235 }
3236 spin_unlock(&ip->i_flags_lock);
3237
David Chinnerbad55842008-03-06 13:43:49 +11003238 /*
3239 * Do an un-protected check to see if the inode is dirty and
3240 * is a candidate for flushing. These checks will be repeated
3241 * later after the appropriate locks are acquired.
3242 */
David Chinner33540402008-03-06 13:43:59 +11003243 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11003244 continue;
David Chinnerbad55842008-03-06 13:43:49 +11003245
3246 /*
3247 * Try to get locks. If any are unavailable or it is pinned,
3248 * then this inode cannot be flushed and is skipped.
3249 */
3250
3251 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3252 continue;
3253 if (!xfs_iflock_nowait(iq)) {
3254 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3255 continue;
3256 }
3257 if (xfs_ipincount(iq)) {
3258 xfs_ifunlock(iq);
3259 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3260 continue;
3261 }
3262
3263 /*
3264 * arriving here means that this inode can be flushed. First
3265 * re-check that it's dirty before flushing.
3266 */
David Chinner33540402008-03-06 13:43:59 +11003267 if (!xfs_inode_clean(iq)) {
3268 int error;
David Chinnerbad55842008-03-06 13:43:49 +11003269 error = xfs_iflush_int(iq, bp);
3270 if (error) {
3271 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3272 goto cluster_corrupt_out;
3273 }
3274 clcount++;
3275 } else {
3276 xfs_ifunlock(iq);
3277 }
3278 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3279 }
3280
3281 if (clcount) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003282 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3283 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
David Chinnerbad55842008-03-06 13:43:49 +11003284 }
3285
3286out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003287 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003288 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003289out_put:
3290 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11003291 return 0;
3292
3293
3294cluster_corrupt_out:
3295 /*
3296 * Corruption detected in the clustering loop. Invalidate the
3297 * inode buffer and shut down the filesystem.
3298 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003299 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11003300 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003301 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11003302 * brelse can handle it with no problems. If not, shut down the
3303 * filesystem before releasing the buffer.
3304 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003305 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11003306 if (bufwasdelwri)
3307 xfs_buf_relse(bp);
3308
3309 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3310
3311 if (!bufwasdelwri) {
3312 /*
3313 * Just like incore_relse: if we have b_iodone functions,
3314 * mark the buffer as an error and call them. Otherwise
3315 * mark it as stale and brelse.
3316 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003317 if (bp->b_iodone) {
David Chinnerbad55842008-03-06 13:43:49 +11003318 XFS_BUF_UNDONE(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003319 xfs_buf_stale(bp);
Dave Chinner24513372014-06-25 14:58:08 +10003320 xfs_buf_ioerror(bp, -EIO);
Dave Chinnere8aaba92014-10-02 09:04:22 +10003321 xfs_buf_ioend(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003322 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003323 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003324 xfs_buf_relse(bp);
3325 }
3326 }
3327
3328 /*
3329 * Unlocks the flush lock
3330 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003331 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003332 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003333 xfs_perag_put(pag);
Dave Chinner24513372014-06-25 14:58:08 +10003334 return -EFSCORRUPTED;
David Chinnerbad55842008-03-06 13:43:49 +11003335}
3336
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10003338 * Flush dirty inode metadata into the backing buffer.
3339 *
3340 * The caller must have the inode lock and the inode flush lock held. The
3341 * inode lock will still be held upon return to the caller, and the inode
3342 * flush lock will be released after the inode has reached the disk.
3343 *
3344 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 */
3346int
3347xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10003348 struct xfs_inode *ip,
3349 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350{
Christoph Hellwig4c468192012-04-23 15:58:36 +10003351 struct xfs_mount *mp = ip->i_mount;
3352 struct xfs_buf *bp;
3353 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003356 XFS_STATS_INC(mp, xs_iflush_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003358 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003359 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003361 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362
Christoph Hellwig4c468192012-04-23 15:58:36 +10003363 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 xfs_iunpin_wait(ip);
3366
3367 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00003368 * For stale inodes we cannot rely on the backing buffer remaining
3369 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04003370 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00003371 * inodes below. We have to check this after ensuring the inode is
3372 * unpinned so that it is safe to reclaim the stale inode after the
3373 * flush call.
3374 */
3375 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3376 xfs_ifunlock(ip);
3377 return 0;
3378 }
3379
3380 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 * This may have been unpinned because the filesystem is shutting
3382 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003383 * to disk, because the log record didn't make it to disk.
3384 *
3385 * We also have to remove the log item from the AIL in this case,
3386 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 */
3388 if (XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10003389 error = -EIO;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003390 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 }
3392
3393 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003394 * Get the buffer containing the on-disk inode.
3395 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003396 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3397 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003398 if (error || !bp) {
3399 xfs_ifunlock(ip);
3400 return error;
3401 }
3402
3403 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 * First flush out the inode that xfs_iflush was called with.
3405 */
3406 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003407 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409
3410 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003411 * If the buffer is pinned then push on the log now so we won't
3412 * get stuck waiting in the write for too long.
3413 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00003414 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003415 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003416
3417 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 * inode clustering:
3419 * see if other inodes can be gathered into this write
3420 */
David Chinnerbad55842008-03-06 13:43:49 +11003421 error = xfs_iflush_cluster(ip, bp);
3422 if (error)
3423 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424
Christoph Hellwig4c468192012-04-23 15:58:36 +10003425 *bpp = bp;
3426 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
3428corrupt_out:
3429 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003430 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431cluster_corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003432 error = -EFSCORRUPTED;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003433abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 /*
3435 * Unlocks the flush lock
3436 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003437 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003438 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439}
3440
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441STATIC int
3442xfs_iflush_int(
Christoph Hellwig93848a92013-04-03 16:11:17 +11003443 struct xfs_inode *ip,
3444 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445{
Christoph Hellwig93848a92013-04-03 16:11:17 +11003446 struct xfs_inode_log_item *iip = ip->i_itemp;
3447 struct xfs_dinode *dip;
3448 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003450 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003451 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003453 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Christoph Hellwig93848a92013-04-03 16:11:17 +11003454 ASSERT(iip != NULL && iip->ili_fields != 0);
Dave Chinner263997a2014-05-20 07:46:40 +10003455 ASSERT(ip->i_d.di_version > 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 /* set *dip = inode's place in the buffer */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10003458 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003460 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003462 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3463 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3464 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465 goto corrupt_out;
3466 }
3467 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3468 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003469 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3470 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3471 __func__, ip->i_ino, ip, ip->i_d.di_magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 goto corrupt_out;
3473 }
Al Viroabbede12011-07-26 02:31:30 -04003474 if (S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 if (XFS_TEST_ERROR(
3476 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3477 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3478 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003479 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3480 "%s: Bad regular inode %Lu, ptr 0x%p",
3481 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 goto corrupt_out;
3483 }
Al Viroabbede12011-07-26 02:31:30 -04003484 } else if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 if (XFS_TEST_ERROR(
3486 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3487 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3488 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3489 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003490 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3491 "%s: Bad directory inode %Lu, ptr 0x%p",
3492 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 goto corrupt_out;
3494 }
3495 }
3496 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3497 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3498 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003499 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3500 "%s: detected corrupt incore inode %Lu, "
3501 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3502 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11003504 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 goto corrupt_out;
3506 }
3507 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3508 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003509 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3510 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3511 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 goto corrupt_out;
3513 }
Dave Chinnere60896d2013-07-24 15:47:30 +10003514
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 /*
Dave Chinner263997a2014-05-20 07:46:40 +10003516 * Inode item log recovery for v2 inodes are dependent on the
Dave Chinnere60896d2013-07-24 15:47:30 +10003517 * di_flushiter count for correct sequencing. We bump the flush
3518 * iteration count so we can detect flushes which postdate a log record
3519 * during recovery. This is redundant as we now log every change and
3520 * hence this can't happen but we need to still do it to ensure
3521 * backwards compatibility with old kernels that predate logging all
3522 * inode changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 */
Dave Chinnere60896d2013-07-24 15:47:30 +10003524 if (ip->i_d.di_version < 3)
3525 ip->i_d.di_flushiter++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526
3527 /*
Dave Chinner39878482016-02-09 16:54:58 +11003528 * Copy the dirty parts of the inode into the on-disk inode. We always
3529 * copy out the core of the inode, because if the inode is dirty at all
3530 * the core must be.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 */
Dave Chinner39878482016-02-09 16:54:58 +11003532 xfs_inode_to_disk(ip, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533
3534 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3535 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3536 ip->i_d.di_flushiter = 0;
3537
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003538 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10003539 if (XFS_IFORK_Q(ip))
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003540 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 xfs_inobp_check(mp, bp);
3542
3543 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003544 * We've recorded everything logged in the inode, so we'd like to clear
3545 * the ili_fields bits so we don't log and flush things unnecessarily.
3546 * However, we can't stop logging all this information until the data
3547 * we've copied into the disk buffer is written to disk. If we did we
3548 * might overwrite the copy of the inode in the log with all the data
3549 * after re-logging only part of it, and in the face of a crash we
3550 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003552 * What we do is move the bits to the ili_last_fields field. When
3553 * logging the inode, these bits are moved back to the ili_fields field.
3554 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3555 * know that the information those bits represent is permanently on
3556 * disk. As long as the flush completes before the inode is logged
3557 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003559 * We can play with the ili_fields bits here, because the inode lock
3560 * must be held exclusively in order to set bits there and the flush
3561 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3562 * done routine can tell whether or not to look in the AIL. Also, store
3563 * the current LSN of the inode so that we can tell whether the item has
3564 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3565 * need the AIL lock, because it is a 64 bit value that cannot be read
3566 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003568 iip->ili_last_fields = iip->ili_fields;
3569 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11003570 iip->ili_fsync_fields = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +11003571 iip->ili_logged = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
Christoph Hellwig93848a92013-04-03 16:11:17 +11003573 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3574 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575
Christoph Hellwig93848a92013-04-03 16:11:17 +11003576 /*
3577 * Attach the function xfs_iflush_done to the inode's
3578 * buffer. This will remove the inode from the AIL
3579 * and unlock the inode's flush lock when the inode is
3580 * completely written to disk.
3581 */
3582 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
Christoph Hellwig93848a92013-04-03 16:11:17 +11003584 /* update the lsn in the on disk inode if required */
3585 if (ip->i_d.di_version == 3)
3586 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587
Christoph Hellwig93848a92013-04-03 16:11:17 +11003588 /* generate the checksum. */
3589 xfs_dinode_calc_crc(mp, dip);
3590
3591 ASSERT(bp->b_fspriv != NULL);
3592 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 return 0;
3594
3595corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003596 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597}