blob: 7d9c514fd23198db8f2ad99fd977082710d46c3c [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_nlink = nlink;
Dwight Engen7aab1b22013-08-15 14:08:01 -0400807 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
808 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000809 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000811 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 ip->i_d.di_gid = pip->i_d.di_gid;
Al Viroabbede12011-07-26 02:31:30 -0400813 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 ip->i_d.di_mode |= S_ISGID;
815 }
816 }
817
818 /*
819 * If the group ID of the new file does not match the effective group
820 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
821 * (and only if the irix_sgid_inherit compatibility variable is set).
822 */
823 if ((irix_sgid_inherit) &&
824 (ip->i_d.di_mode & S_ISGID) &&
Dwight Engen7aab1b22013-08-15 14:08:01 -0400825 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 ip->i_d.di_mode &= ~S_ISGID;
827 }
828
829 ip->i_d.di_size = 0;
830 ip->i_d.di_nextents = 0;
831 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000832
Dave Chinnere076b0f2014-10-02 09:18:13 +1000833 tv = current_fs_time(mp->m_super);
Dave Chinner39878482016-02-09 16:54:58 +1100834 inode->i_mtime = tv;
835 inode->i_atime = tv;
836 inode->i_ctime = tv;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000837
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 /*
839 * di_gen will have been taken care of in xfs_iread.
840 */
841 ip->i_d.di_extsize = 0;
842 ip->i_d.di_dmevmask = 0;
843 ip->i_d.di_dmstate = 0;
844 ip->i_d.di_flags = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100845
846 if (ip->i_d.di_version == 3) {
Christoph Hellwig93848a92013-04-03 16:11:17 +1100847 ip->i_d.di_changecount = 1;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100848 ip->i_d.di_flags2 = 0;
Dave Chinner39878482016-02-09 16:54:58 +1100849 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
850 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100851 }
852
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 flags = XFS_ILOG_CORE;
855 switch (mode & S_IFMT) {
856 case S_IFIFO:
857 case S_IFCHR:
858 case S_IFBLK:
859 case S_IFSOCK:
860 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
861 ip->i_df.if_u2.if_rdev = rdev;
862 ip->i_df.if_flags = 0;
863 flags |= XFS_ILOG_DEV;
864 break;
865 case S_IFREG:
866 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +1000867 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Dave Chinner58f88ca2016-01-04 16:44:15 +1100868 uint64_t di_flags2 = 0;
869 uint di_flags = 0;
Nathan Scott365ca832005-06-21 15:39:12 +1000870
Al Viroabbede12011-07-26 02:31:30 -0400871 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000872 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
873 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100874 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
875 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
876 ip->i_d.di_extsize = pip->i_d.di_extsize;
877 }
Dave Chinner9336e3a2014-10-02 09:18:40 +1000878 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
879 di_flags |= XFS_DIFLAG_PROJINHERIT;
Al Viroabbede12011-07-26 02:31:30 -0400880 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000881 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +1000882 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100883 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
884 di_flags |= XFS_DIFLAG_EXTSIZE;
885 ip->i_d.di_extsize = pip->i_d.di_extsize;
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 }
888 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
889 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +1000890 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
892 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +1000893 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
895 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +1000896 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
898 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +1000899 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000900 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
901 xfs_inherit_nodefrag)
902 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000903 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
904 di_flags |= XFS_DIFLAG_FILESTREAM;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100905 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
906 di_flags2 |= XFS_DIFLAG2_DAX;
907
Nathan Scott365ca832005-06-21 15:39:12 +1000908 ip->i_d.di_flags |= di_flags;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100909 ip->i_d.di_flags2 |= di_flags2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
911 /* FALLTHROUGH */
912 case S_IFLNK:
913 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
914 ip->i_df.if_flags = XFS_IFEXTENTS;
915 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
916 ip->i_df.if_u1.if_extents = NULL;
917 break;
918 default:
919 ASSERT(0);
920 }
921 /*
922 * Attribute fork settings for new inode.
923 */
924 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
925 ip->i_d.di_anextents = 0;
926
927 /*
928 * Log the new values stuffed into the inode.
929 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000930 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 xfs_trans_log_inode(tp, ip, flags);
932
Dave Chinner58c90472015-02-23 22:38:08 +1100933 /* now that we have an i_mode we can setup the inode structure */
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000934 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936 *ipp = ip;
937 return 0;
938}
939
Dave Chinnere546cb72013-08-12 20:49:47 +1000940/*
941 * Allocates a new inode from disk and return a pointer to the
942 * incore copy. This routine will internally commit the current
943 * transaction and allocate a new one if the Space Manager needed
944 * to do an allocation to replenish the inode free-list.
945 *
946 * This routine is designed to be called from xfs_create and
947 * xfs_create_dir.
948 *
949 */
950int
951xfs_dir_ialloc(
952 xfs_trans_t **tpp, /* input: current transaction;
953 output: may be a new transaction. */
954 xfs_inode_t *dp, /* directory within whose allocate
955 the inode. */
956 umode_t mode,
957 xfs_nlink_t nlink,
958 xfs_dev_t rdev,
959 prid_t prid, /* project id */
960 int okalloc, /* ok to allocate new space */
961 xfs_inode_t **ipp, /* pointer to inode; it will be
962 locked. */
963 int *committed)
964
965{
966 xfs_trans_t *tp;
Dave Chinnere546cb72013-08-12 20:49:47 +1000967 xfs_inode_t *ip;
968 xfs_buf_t *ialloc_context = NULL;
969 int code;
Dave Chinnere546cb72013-08-12 20:49:47 +1000970 void *dqinfo;
971 uint tflags;
972
973 tp = *tpp;
974 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
975
976 /*
977 * xfs_ialloc will return a pointer to an incore inode if
978 * the Space Manager has an available inode on the free
979 * list. Otherwise, it will do an allocation and replenish
980 * the freelist. Since we can only do one allocation per
981 * transaction without deadlocks, we will need to commit the
982 * current transaction and start a new one. We will then
983 * need to call xfs_ialloc again to get the inode.
984 *
985 * If xfs_ialloc did an allocation to replenish the freelist,
986 * it returns the bp containing the head of the freelist as
987 * ialloc_context. We will hold a lock on it across the
988 * transaction commit so that no other process can steal
989 * the inode(s) that we've just allocated.
990 */
991 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
992 &ialloc_context, &ip);
993
994 /*
995 * Return an error if we were unable to allocate a new inode.
996 * This should only happen if we run out of space on disk or
997 * encounter a disk error.
998 */
999 if (code) {
1000 *ipp = NULL;
1001 return code;
1002 }
1003 if (!ialloc_context && !ip) {
1004 *ipp = NULL;
Dave Chinner24513372014-06-25 14:58:08 +10001005 return -ENOSPC;
Dave Chinnere546cb72013-08-12 20:49:47 +10001006 }
1007
1008 /*
1009 * If the AGI buffer is non-NULL, then we were unable to get an
1010 * inode in one operation. We need to commit the current
1011 * transaction and call xfs_ialloc() again. It is guaranteed
1012 * to succeed the second time.
1013 */
1014 if (ialloc_context) {
1015 /*
1016 * Normally, xfs_trans_commit releases all the locks.
1017 * We call bhold to hang on to the ialloc_context across
1018 * the commit. Holding this buffer prevents any other
1019 * processes from doing any allocations in this
1020 * allocation group.
1021 */
1022 xfs_trans_bhold(tp, ialloc_context);
Dave Chinnere546cb72013-08-12 20:49:47 +10001023
1024 /*
1025 * We want the quota changes to be associated with the next
1026 * transaction, NOT this one. So, detach the dqinfo from this
1027 * and attach it to the next transaction.
1028 */
1029 dqinfo = NULL;
1030 tflags = 0;
1031 if (tp->t_dqinfo) {
1032 dqinfo = (void *)tp->t_dqinfo;
1033 tp->t_dqinfo = NULL;
1034 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1035 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1036 }
1037
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001038 code = xfs_trans_roll(&tp, 0);
1039 if (committed != NULL)
Dave Chinnere546cb72013-08-12 20:49:47 +10001040 *committed = 1;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001041
Dave Chinnere546cb72013-08-12 20:49:47 +10001042 /*
1043 * Re-attach the quota info that we detached from prev trx.
1044 */
1045 if (dqinfo) {
1046 tp->t_dqinfo = dqinfo;
1047 tp->t_flags |= tflags;
1048 }
1049
1050 if (code) {
1051 xfs_buf_relse(ialloc_context);
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001052 *tpp = tp;
Dave Chinnere546cb72013-08-12 20:49:47 +10001053 *ipp = NULL;
1054 return code;
1055 }
1056 xfs_trans_bjoin(tp, ialloc_context);
1057
1058 /*
1059 * Call ialloc again. Since we've locked out all
1060 * other allocations in this allocation group,
1061 * this call should always succeed.
1062 */
1063 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1064 okalloc, &ialloc_context, &ip);
1065
1066 /*
1067 * If we get an error at this point, return to the caller
1068 * so that the current transaction can be aborted.
1069 */
1070 if (code) {
1071 *tpp = tp;
1072 *ipp = NULL;
1073 return code;
1074 }
1075 ASSERT(!ialloc_context && ip);
1076
1077 } else {
1078 if (committed != NULL)
1079 *committed = 0;
1080 }
1081
1082 *ipp = ip;
1083 *tpp = tp;
1084
1085 return 0;
1086}
1087
1088/*
1089 * Decrement the link count on an inode & log the change.
1090 * If this causes the link count to go to zero, initiate the
1091 * logging activity required to truncate a file.
1092 */
1093int /* error */
1094xfs_droplink(
1095 xfs_trans_t *tp,
1096 xfs_inode_t *ip)
1097{
1098 int error;
1099
1100 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1101
1102 ASSERT (ip->i_d.di_nlink > 0);
1103 ip->i_d.di_nlink--;
1104 drop_nlink(VFS_I(ip));
1105 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1106
1107 error = 0;
1108 if (ip->i_d.di_nlink == 0) {
1109 /*
1110 * We're dropping the last link to this file.
1111 * Move the on-disk inode to the AGI unlinked list.
1112 * From xfs_inactive() we will pull the inode from
1113 * the list and free it.
1114 */
1115 error = xfs_iunlink(tp, ip);
1116 }
1117 return error;
1118}
1119
1120/*
Dave Chinnere546cb72013-08-12 20:49:47 +10001121 * Increment the link count on an inode & log the change.
1122 */
1123int
1124xfs_bumplink(
1125 xfs_trans_t *tp,
1126 xfs_inode_t *ip)
1127{
1128 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1129
Dave Chinner263997a2014-05-20 07:46:40 +10001130 ASSERT(ip->i_d.di_version > 1);
Zhi Yong Wuab297432013-12-18 08:22:41 +08001131 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
Dave Chinnere546cb72013-08-12 20:49:47 +10001132 ip->i_d.di_nlink++;
1133 inc_nlink(VFS_I(ip));
Dave Chinnere546cb72013-08-12 20:49:47 +10001134 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1135 return 0;
1136}
1137
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001138int
1139xfs_create(
1140 xfs_inode_t *dp,
1141 struct xfs_name *name,
1142 umode_t mode,
1143 xfs_dev_t rdev,
1144 xfs_inode_t **ipp)
1145{
1146 int is_dir = S_ISDIR(mode);
1147 struct xfs_mount *mp = dp->i_mount;
1148 struct xfs_inode *ip = NULL;
1149 struct xfs_trans *tp = NULL;
1150 int error;
1151 xfs_bmap_free_t free_list;
1152 xfs_fsblock_t first_block;
1153 bool unlock_dp_on_error = false;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001154 prid_t prid;
1155 struct xfs_dquot *udqp = NULL;
1156 struct xfs_dquot *gdqp = NULL;
1157 struct xfs_dquot *pdqp = NULL;
Brian Foster062647a2014-11-28 14:00:16 +11001158 struct xfs_trans_res *tres;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001159 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001160
1161 trace_xfs_create(dp, name);
1162
1163 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001164 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001165
Zhi Yong Wu163467d2013-12-18 08:22:39 +08001166 prid = xfs_get_initial_prid(dp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001167
1168 /*
1169 * Make sure that we have allocated dquot(s) on disk.
1170 */
Dwight Engen7aab1b22013-08-15 14:08:01 -04001171 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1172 xfs_kgid_to_gid(current_fsgid()), prid,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001173 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1174 &udqp, &gdqp, &pdqp);
1175 if (error)
1176 return error;
1177
1178 if (is_dir) {
1179 rdev = 0;
1180 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001181 tres = &M_RES(mp)->tr_mkdir;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001182 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1183 } else {
1184 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001185 tres = &M_RES(mp)->tr_create;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001186 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1187 }
1188
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001189 /*
1190 * Initially assume that the file does not exist and
1191 * reserve the resources for that case. If that is not
1192 * the case we'll drop the one we have and get a more
1193 * appropriate transaction later.
1194 */
Brian Foster062647a2014-11-28 14:00:16 +11001195 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001196 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001197 /* flush outstanding delalloc blocks and retry */
1198 xfs_flush_inodes(mp);
Brian Foster062647a2014-11-28 14:00:16 +11001199 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001200 }
Dave Chinner24513372014-06-25 14:58:08 +10001201 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001202 /* No space at all so try a "no-allocation" reservation */
1203 resblks = 0;
Brian Foster062647a2014-11-28 14:00:16 +11001204 error = xfs_trans_reserve(tp, tres, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001205 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001206 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001207 goto out_trans_cancel;
Christoph Hellwig4906e212015-06-04 13:47:56 +10001208
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001209
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001210 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1211 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001212 unlock_dp_on_error = true;
1213
1214 xfs_bmap_init(&free_list, &first_block);
1215
1216 /*
1217 * Reserve disk quota and the inode.
1218 */
1219 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1220 pdqp, resblks, 1, 0);
1221 if (error)
1222 goto out_trans_cancel;
1223
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001224 if (!resblks) {
1225 error = xfs_dir_canenter(tp, dp, name);
1226 if (error)
1227 goto out_trans_cancel;
1228 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001229
1230 /*
1231 * A newly created regular or special file just has one directory
1232 * entry pointing to them, but a directory also the "." entry
1233 * pointing to itself.
1234 */
1235 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001236 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001237 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001238 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001239
1240 /*
1241 * Now we join the directory inode to the transaction. We do not do it
1242 * earlier because xfs_dir_ialloc might commit the previous transaction
1243 * (and release all the locks). An error from here on will result in
1244 * the transaction cancel unlocking dp so don't do it explicitly in the
1245 * error path.
1246 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001247 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001248 unlock_dp_on_error = false;
1249
1250 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1251 &first_block, &free_list, resblks ?
1252 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1253 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001254 ASSERT(error != -ENOSPC);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001255 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001256 }
1257 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1258 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1259
1260 if (is_dir) {
1261 error = xfs_dir_init(tp, ip, dp);
1262 if (error)
1263 goto out_bmap_cancel;
1264
1265 error = xfs_bumplink(tp, dp);
1266 if (error)
1267 goto out_bmap_cancel;
1268 }
1269
1270 /*
1271 * If this is a synchronous mount, make sure that the
1272 * create transaction goes to disk before returning to
1273 * the user.
1274 */
1275 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1276 xfs_trans_set_sync(tp);
1277
1278 /*
1279 * Attach the dquot(s) to the inodes and modify them incore.
1280 * These ids of the inode couldn't have changed since the new
1281 * inode has been locked ever since it was created.
1282 */
1283 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1284
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001285 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001286 if (error)
1287 goto out_bmap_cancel;
1288
Christoph Hellwig70393312015-06-04 13:48:08 +10001289 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001290 if (error)
1291 goto out_release_inode;
1292
1293 xfs_qm_dqrele(udqp);
1294 xfs_qm_dqrele(gdqp);
1295 xfs_qm_dqrele(pdqp);
1296
1297 *ipp = ip;
1298 return 0;
1299
1300 out_bmap_cancel:
1301 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001302 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001303 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001304 out_release_inode:
1305 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001306 * Wait until after the current transaction is aborted to finish the
1307 * setup of the inode and release the inode. This prevents recursive
1308 * transactions and deadlocks from xfs_inactive.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001309 */
Dave Chinner58c90472015-02-23 22:38:08 +11001310 if (ip) {
1311 xfs_finish_inode_setup(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001312 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001313 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001314
1315 xfs_qm_dqrele(udqp);
1316 xfs_qm_dqrele(gdqp);
1317 xfs_qm_dqrele(pdqp);
1318
1319 if (unlock_dp_on_error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001320 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001321 return error;
1322}
1323
1324int
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001325xfs_create_tmpfile(
1326 struct xfs_inode *dp,
1327 struct dentry *dentry,
Brian Foster330033d2014-04-17 08:15:30 +10001328 umode_t mode,
1329 struct xfs_inode **ipp)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001330{
1331 struct xfs_mount *mp = dp->i_mount;
1332 struct xfs_inode *ip = NULL;
1333 struct xfs_trans *tp = NULL;
1334 int error;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001335 prid_t prid;
1336 struct xfs_dquot *udqp = NULL;
1337 struct xfs_dquot *gdqp = NULL;
1338 struct xfs_dquot *pdqp = NULL;
1339 struct xfs_trans_res *tres;
1340 uint resblks;
1341
1342 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001343 return -EIO;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001344
1345 prid = xfs_get_initial_prid(dp);
1346
1347 /*
1348 * Make sure that we have allocated dquot(s) on disk.
1349 */
1350 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1351 xfs_kgid_to_gid(current_fsgid()), prid,
1352 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1353 &udqp, &gdqp, &pdqp);
1354 if (error)
1355 return error;
1356
1357 resblks = XFS_IALLOC_SPACE_RES(mp);
1358 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1359
1360 tres = &M_RES(mp)->tr_create_tmpfile;
1361 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001362 if (error == -ENOSPC) {
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001363 /* No space at all so try a "no-allocation" reservation */
1364 resblks = 0;
1365 error = xfs_trans_reserve(tp, tres, 0, 0);
1366 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001367 if (error)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001368 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001369
1370 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1371 pdqp, resblks, 1, 0);
1372 if (error)
1373 goto out_trans_cancel;
1374
1375 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1376 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001377 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001378 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001379
1380 if (mp->m_flags & XFS_MOUNT_WSYNC)
1381 xfs_trans_set_sync(tp);
1382
1383 /*
1384 * Attach the dquot(s) to the inodes and modify them incore.
1385 * These ids of the inode couldn't have changed since the new
1386 * inode has been locked ever since it was created.
1387 */
1388 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1389
1390 ip->i_d.di_nlink--;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001391 error = xfs_iunlink(tp, ip);
1392 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001393 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001394
Christoph Hellwig70393312015-06-04 13:48:08 +10001395 error = xfs_trans_commit(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001396 if (error)
1397 goto out_release_inode;
1398
1399 xfs_qm_dqrele(udqp);
1400 xfs_qm_dqrele(gdqp);
1401 xfs_qm_dqrele(pdqp);
1402
Brian Foster330033d2014-04-17 08:15:30 +10001403 *ipp = ip;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001404 return 0;
1405
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001406 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001407 xfs_trans_cancel(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001408 out_release_inode:
1409 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001410 * Wait until after the current transaction is aborted to finish the
1411 * setup of the inode and release the inode. This prevents recursive
1412 * transactions and deadlocks from xfs_inactive.
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001413 */
Dave Chinner58c90472015-02-23 22:38:08 +11001414 if (ip) {
1415 xfs_finish_inode_setup(ip);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001416 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001417 }
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001418
1419 xfs_qm_dqrele(udqp);
1420 xfs_qm_dqrele(gdqp);
1421 xfs_qm_dqrele(pdqp);
1422
1423 return error;
1424}
1425
1426int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001427xfs_link(
1428 xfs_inode_t *tdp,
1429 xfs_inode_t *sip,
1430 struct xfs_name *target_name)
1431{
1432 xfs_mount_t *mp = tdp->i_mount;
1433 xfs_trans_t *tp;
1434 int error;
1435 xfs_bmap_free_t free_list;
1436 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001437 int resblks;
1438
1439 trace_xfs_link(tdp, target_name);
1440
1441 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1442
1443 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001444 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001445
1446 error = xfs_qm_dqattach(sip, 0);
1447 if (error)
1448 goto std_return;
1449
1450 error = xfs_qm_dqattach(tdp, 0);
1451 if (error)
1452 goto std_return;
1453
1454 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001455 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001456 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001457 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001458 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001459 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001460 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001461 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001462 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001463
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001464 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001465 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1466
1467 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001468 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001469
1470 /*
1471 * If we are using project inheritance, we only allow hard link
1472 * creation in our tree when the project IDs are the same; else
1473 * the tree quota mechanism could be circumvented.
1474 */
1475 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1476 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10001477 error = -EXDEV;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001478 goto error_return;
1479 }
1480
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001481 if (!resblks) {
1482 error = xfs_dir_canenter(tp, tdp, target_name);
1483 if (error)
1484 goto error_return;
1485 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001486
1487 xfs_bmap_init(&free_list, &first_block);
1488
Zhi Yong Wuab297432013-12-18 08:22:41 +08001489 if (sip->i_d.di_nlink == 0) {
1490 error = xfs_iunlink_remove(tp, sip);
1491 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001492 goto error_return;
Zhi Yong Wuab297432013-12-18 08:22:41 +08001493 }
1494
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001495 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1496 &first_block, &free_list, resblks);
1497 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001498 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001499 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1500 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1501
1502 error = xfs_bumplink(tp, sip);
1503 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001504 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001505
1506 /*
1507 * If this is a synchronous mount, make sure that the
1508 * link transaction goes to disk before returning to
1509 * the user.
1510 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001511 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001512 xfs_trans_set_sync(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001513
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001514 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001515 if (error) {
1516 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001517 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001518 }
1519
Christoph Hellwig70393312015-06-04 13:48:08 +10001520 return xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001521
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001522 error_return:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001523 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001524 std_return:
1525 return error;
1526}
1527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001529 * Free up the underlying blocks past new_size. The new size must be smaller
1530 * than the current size. This routine can be used both for the attribute and
1531 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 *
David Chinnerf6485052008-04-17 16:50:04 +10001533 * The transaction passed to this routine must have made a permanent log
1534 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1535 * given transaction and start new ones, so make sure everything involved in
1536 * the transaction is tidy before calling here. Some transaction will be
1537 * returned to the caller to be committed. The incoming transaction must
1538 * already include the inode, and both inode locks must be held exclusively.
1539 * The inode must also be "held" within the transaction. On return the inode
1540 * will be "held" within the returned transaction. This routine does NOT
1541 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 *
David Chinnerf6485052008-04-17 16:50:04 +10001543 * If we get an error, we must return with the inode locked and linked into the
1544 * current transaction. This keeps things simple for the higher level code,
1545 * because it always knows that the inode is locked and held in the transaction
1546 * that returns to it whether errors occur or not. We don't mark the inode
1547 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 */
1549int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001550xfs_itruncate_extents(
1551 struct xfs_trans **tpp,
1552 struct xfs_inode *ip,
1553 int whichfork,
1554 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001556 struct xfs_mount *mp = ip->i_mount;
1557 struct xfs_trans *tp = *tpp;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001558 xfs_bmap_free_t free_list;
1559 xfs_fsblock_t first_block;
1560 xfs_fileoff_t first_unmap_block;
1561 xfs_fileoff_t last_block;
1562 xfs_filblks_t unmap_len;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001563 int error = 0;
1564 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Christoph Hellwig0b561852012-07-04 11:13:31 -04001566 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1567 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1568 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001569 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001570 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001572 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001573 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001575 trace_xfs_itruncate_extents_start(ip, new_size);
1576
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 /*
1578 * Since it is possible for space to become allocated beyond
1579 * the end of the file (in a crash where the space is allocated
1580 * but the inode size is not yet updated), simply remove any
1581 * blocks which show up between the new EOF and the maximum
1582 * possible file size. If the first block to be removed is
1583 * beyond the maximum file size (ie it is the same as last_block),
1584 * then there is nothing to do.
1585 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001586 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001587 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001588 if (first_unmap_block == last_block)
1589 return 0;
1590
1591 ASSERT(first_unmap_block < last_block);
1592 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001594 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001595 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001596 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001597 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001599 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001600 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001601 if (error)
1602 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
1604 /*
1605 * Duplicate the transaction that has the permanent
1606 * reservation and commit the old transaction.
1607 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001608 error = xfs_bmap_finish(&tp, &free_list, ip);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001609 if (error)
1610 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001612 error = xfs_trans_roll(&tp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001613 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001614 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001616
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001617 /*
1618 * Always re-log the inode so that our permanent transaction can keep
1619 * on rolling it forward in the log.
1620 */
1621 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1622
1623 trace_xfs_itruncate_extents_end(ip, new_size);
1624
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001625out:
1626 *tpp = tp;
1627 return error;
1628out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001630 * If the bunmapi call encounters an error, return to the caller where
1631 * the transaction can be properly aborted. We just need to make sure
1632 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001634 xfs_bmap_cancel(&free_list);
1635 goto out;
1636}
1637
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001638int
1639xfs_release(
1640 xfs_inode_t *ip)
1641{
1642 xfs_mount_t *mp = ip->i_mount;
1643 int error;
1644
1645 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1646 return 0;
1647
1648 /* If this is a read-only mount, don't do this (would generate I/O) */
1649 if (mp->m_flags & XFS_MOUNT_RDONLY)
1650 return 0;
1651
1652 if (!XFS_FORCED_SHUTDOWN(mp)) {
1653 int truncated;
1654
1655 /*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001656 * If we previously truncated this file and removed old data
1657 * in the process, we want to initiate "early" writeout on
1658 * the last close. This is an attempt to combat the notorious
1659 * NULL files problem which is particularly noticeable from a
1660 * truncate down, buffered (re-)write (delalloc), followed by
1661 * a crash. What we are effectively doing here is
1662 * significantly reducing the time window where we'd otherwise
1663 * be exposed to that problem.
1664 */
1665 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1666 if (truncated) {
1667 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
Dave Chinnereac152b2014-08-04 13:22:49 +10001668 if (ip->i_delayed_blks > 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001669 error = filemap_flush(VFS_I(ip)->i_mapping);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001670 if (error)
1671 return error;
1672 }
1673 }
1674 }
1675
1676 if (ip->i_d.di_nlink == 0)
1677 return 0;
1678
1679 if (xfs_can_free_eofblocks(ip, false)) {
1680
1681 /*
1682 * If we can't get the iolock just skip truncating the blocks
1683 * past EOF because we could deadlock with the mmap_sem
1684 * otherwise. We'll get another chance to drop them once the
1685 * last reference to the inode is dropped, so we'll never leak
1686 * blocks permanently.
1687 *
1688 * Further, check if the inode is being opened, written and
1689 * closed frequently and we have delayed allocation blocks
1690 * outstanding (e.g. streaming writes from the NFS server),
1691 * truncating the blocks past EOF will cause fragmentation to
1692 * occur.
1693 *
1694 * In this case don't do the truncation, either, but we have to
1695 * be careful how we detect this case. Blocks beyond EOF show
1696 * up as i_delayed_blks even when the inode is clean, so we
1697 * need to truncate them away first before checking for a dirty
1698 * release. Hence on the first dirty close we will still remove
1699 * the speculative allocation, but after that we will leave it
1700 * in place.
1701 */
1702 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1703 return 0;
1704
1705 error = xfs_free_eofblocks(mp, ip, true);
Dave Chinner24513372014-06-25 14:58:08 +10001706 if (error && error != -EAGAIN)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001707 return error;
1708
1709 /* delalloc blocks after truncation means it really is dirty */
1710 if (ip->i_delayed_blks)
1711 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1712 }
1713 return 0;
1714}
1715
1716/*
Brian Fosterf7be2d72013-09-20 11:06:10 -04001717 * xfs_inactive_truncate
1718 *
1719 * Called to perform a truncate when an inode becomes unlinked.
1720 */
1721STATIC int
1722xfs_inactive_truncate(
1723 struct xfs_inode *ip)
1724{
1725 struct xfs_mount *mp = ip->i_mount;
1726 struct xfs_trans *tp;
1727 int error;
1728
1729 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1730 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1731 if (error) {
1732 ASSERT(XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +10001733 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001734 return error;
1735 }
1736
1737 xfs_ilock(ip, XFS_ILOCK_EXCL);
1738 xfs_trans_ijoin(tp, ip, 0);
1739
1740 /*
1741 * Log the inode size first to prevent stale data exposure in the event
1742 * of a system crash before the truncate completes. See the related
1743 * comment in xfs_setattr_size() for details.
1744 */
1745 ip->i_d.di_size = 0;
1746 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1747
1748 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1749 if (error)
1750 goto error_trans_cancel;
1751
1752 ASSERT(ip->i_d.di_nextents == 0);
1753
Christoph Hellwig70393312015-06-04 13:48:08 +10001754 error = xfs_trans_commit(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001755 if (error)
1756 goto error_unlock;
1757
1758 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1759 return 0;
1760
1761error_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001762 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001763error_unlock:
1764 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1765 return error;
1766}
1767
1768/*
Brian Foster88877d22013-09-20 11:06:11 -04001769 * xfs_inactive_ifree()
1770 *
1771 * Perform the inode free when an inode is unlinked.
1772 */
1773STATIC int
1774xfs_inactive_ifree(
1775 struct xfs_inode *ip)
1776{
1777 xfs_bmap_free_t free_list;
1778 xfs_fsblock_t first_block;
Brian Foster88877d22013-09-20 11:06:11 -04001779 struct xfs_mount *mp = ip->i_mount;
1780 struct xfs_trans *tp;
1781 int error;
1782
1783 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
Brian Foster9d43b182014-04-24 16:00:52 +10001784
1785 /*
1786 * The ifree transaction might need to allocate blocks for record
1787 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1788 * allow ifree to dip into the reserved block pool if necessary.
1789 *
1790 * Freeing large sets of inodes generally means freeing inode chunks,
1791 * directory and file data blocks, so this should be relatively safe.
1792 * Only under severe circumstances should it be possible to free enough
1793 * inodes to exhaust the reserve block pool via finobt expansion while
1794 * at the same time not creating free space in the filesystem.
1795 *
1796 * Send a warning if the reservation does happen to fail, as the inode
1797 * now remains allocated and sits on the unlinked list until the fs is
1798 * repaired.
1799 */
1800 tp->t_flags |= XFS_TRANS_RESERVE;
1801 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1802 XFS_IFREE_SPACE_RES(mp), 0);
Brian Foster88877d22013-09-20 11:06:11 -04001803 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001804 if (error == -ENOSPC) {
Brian Foster9d43b182014-04-24 16:00:52 +10001805 xfs_warn_ratelimited(mp,
1806 "Failed to remove inode(s) from unlinked list. "
1807 "Please free space, unmount and run xfs_repair.");
1808 } else {
1809 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1810 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001811 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001812 return error;
1813 }
1814
1815 xfs_ilock(ip, XFS_ILOCK_EXCL);
1816 xfs_trans_ijoin(tp, ip, 0);
1817
1818 xfs_bmap_init(&free_list, &first_block);
1819 error = xfs_ifree(tp, ip, &free_list);
1820 if (error) {
1821 /*
1822 * If we fail to free the inode, shut down. The cancel
1823 * might do that, we need to make sure. Otherwise the
1824 * inode might be lost for a long time or forever.
1825 */
1826 if (!XFS_FORCED_SHUTDOWN(mp)) {
1827 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1828 __func__, error);
1829 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1830 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001831 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001832 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1833 return error;
1834 }
1835
1836 /*
1837 * Credit the quota account(s). The inode is gone.
1838 */
1839 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1840
1841 /*
Brian Fosterd4a97a02015-08-19 10:01:40 +10001842 * Just ignore errors at this point. There is nothing we can do except
1843 * to try to keep going. Make sure it's not a silent error.
Brian Foster88877d22013-09-20 11:06:11 -04001844 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001845 error = xfs_bmap_finish(&tp, &free_list, NULL);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001846 if (error) {
Brian Foster88877d22013-09-20 11:06:11 -04001847 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1848 __func__, error);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001849 xfs_bmap_cancel(&free_list);
1850 }
Christoph Hellwig70393312015-06-04 13:48:08 +10001851 error = xfs_trans_commit(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001852 if (error)
1853 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1854 __func__, error);
1855
1856 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1857 return 0;
1858}
1859
1860/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001861 * xfs_inactive
1862 *
1863 * This is called when the vnode reference count for the vnode
1864 * goes to zero. If the file has been unlinked, then it must
1865 * now be truncated. Also, we clear all of the read-ahead state
1866 * kept for the inode here since the file is now closed.
1867 */
Brian Foster74564fb2013-09-20 11:06:12 -04001868void
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001869xfs_inactive(
1870 xfs_inode_t *ip)
1871{
Jie Liu3d3c8b52013-08-12 20:49:59 +10001872 struct xfs_mount *mp;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001873 int error;
1874 int truncate = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001875
1876 /*
1877 * If the inode is already free, then there can be nothing
1878 * to clean up here.
1879 */
Ben Myersd9487092013-09-10 18:11:22 -05001880 if (ip->i_d.di_mode == 0) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001881 ASSERT(ip->i_df.if_real_bytes == 0);
1882 ASSERT(ip->i_df.if_broot_bytes == 0);
Brian Foster74564fb2013-09-20 11:06:12 -04001883 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001884 }
1885
1886 mp = ip->i_mount;
1887
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001888 /* If this is a read-only mount, don't do this (would generate I/O) */
1889 if (mp->m_flags & XFS_MOUNT_RDONLY)
Brian Foster74564fb2013-09-20 11:06:12 -04001890 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001891
1892 if (ip->i_d.di_nlink != 0) {
1893 /*
1894 * force is true because we are evicting an inode from the
1895 * cache. Post-eof blocks must be freed, lest we end up with
1896 * broken free space accounting.
1897 */
Brian Foster74564fb2013-09-20 11:06:12 -04001898 if (xfs_can_free_eofblocks(ip, true))
1899 xfs_free_eofblocks(mp, ip, false);
1900
1901 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001902 }
1903
1904 if (S_ISREG(ip->i_d.di_mode) &&
1905 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1906 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1907 truncate = 1;
1908
1909 error = xfs_qm_dqattach(ip, 0);
1910 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001911 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001912
Brian Fosterf7be2d72013-09-20 11:06:10 -04001913 if (S_ISLNK(ip->i_d.di_mode))
Brian Foster36b21dd2013-09-20 11:06:09 -04001914 error = xfs_inactive_symlink(ip);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001915 else if (truncate)
1916 error = xfs_inactive_truncate(ip);
1917 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001918 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001919
1920 /*
1921 * If there are attributes associated with the file then blow them away
1922 * now. The code calls a routine that recursively deconstructs the
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001923 * attribute fork. If also blows away the in-core attribute fork.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001924 */
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001925 if (XFS_IFORK_Q(ip)) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001926 error = xfs_attr_inactive(ip);
1927 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001928 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001929 }
1930
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001931 ASSERT(!ip->i_afp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001932 ASSERT(ip->i_d.di_anextents == 0);
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001933 ASSERT(ip->i_d.di_forkoff == 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001934
1935 /*
1936 * Free the inode.
1937 */
Brian Foster88877d22013-09-20 11:06:11 -04001938 error = xfs_inactive_ifree(ip);
1939 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001940 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001941
1942 /*
1943 * Release the dquots held by inode, if any.
1944 */
1945 xfs_qm_dqdetach(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001946}
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948/*
1949 * This is called when the inode's link count goes to 0.
1950 * We place the on-disk inode on a list in the AGI. It
1951 * will be pulled from this list when the inode is freed.
1952 */
1953int
1954xfs_iunlink(
1955 xfs_trans_t *tp,
1956 xfs_inode_t *ip)
1957{
1958 xfs_mount_t *mp;
1959 xfs_agi_t *agi;
1960 xfs_dinode_t *dip;
1961 xfs_buf_t *agibp;
1962 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 xfs_agino_t agino;
1964 short bucket_index;
1965 int offset;
1966 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968 ASSERT(ip->i_d.di_nlink == 0);
1969 ASSERT(ip->i_d.di_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 mp = tp->t_mountp;
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 /*
1974 * Get the agi buffer first. It ensures lock ordering
1975 * on the list.
1976 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001977 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001978 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001981
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 /*
1983 * Get the index into the agi hash table for the
1984 * list this inode will go on.
1985 */
1986 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1987 ASSERT(agino != 0);
1988 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1989 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001990 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001992 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 /*
1994 * There is already another inode in the bucket we need
1995 * to add ourselves to. Add us at the front of the list.
1996 * Here we put the head pointer into our next pointer,
1997 * and then we fall through to point the head at us.
1998 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001999 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2000 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002001 if (error)
2002 return error;
2003
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002004 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002006 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002008
2009 /* need to recalc the inode CRC if appropriate */
2010 xfs_dinode_calc_crc(mp, dip);
2011
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 xfs_trans_inode_buf(tp, ibp);
2013 xfs_trans_log_buf(tp, ibp, offset,
2014 (offset + sizeof(xfs_agino_t) - 1));
2015 xfs_inobp_check(mp, ibp);
2016 }
2017
2018 /*
2019 * Point the bucket head pointer at the inode being inserted.
2020 */
2021 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002022 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 offset = offsetof(xfs_agi_t, agi_unlinked) +
2024 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002025 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 xfs_trans_log_buf(tp, agibp, offset,
2027 (offset + sizeof(xfs_agino_t) - 1));
2028 return 0;
2029}
2030
2031/*
2032 * Pull the on-disk inode from the AGI unlinked list.
2033 */
2034STATIC int
2035xfs_iunlink_remove(
2036 xfs_trans_t *tp,
2037 xfs_inode_t *ip)
2038{
2039 xfs_ino_t next_ino;
2040 xfs_mount_t *mp;
2041 xfs_agi_t *agi;
2042 xfs_dinode_t *dip;
2043 xfs_buf_t *agibp;
2044 xfs_buf_t *ibp;
2045 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 xfs_agino_t agino;
2047 xfs_agino_t next_agino;
2048 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002049 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002051 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
2057 /*
2058 * Get the agi buffer first. It ensures lock ordering
2059 * on the list.
2060 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002061 error = xfs_read_agi(mp, tp, agno, &agibp);
2062 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002064
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 /*
2068 * Get the index into the agi hash table for the
2069 * list this inode will go on.
2070 */
2071 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2072 ASSERT(agino != 0);
2073 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002074 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 ASSERT(agi->agi_unlinked[bucket_index]);
2076
Christoph Hellwig16259e72005-11-02 15:11:25 +11002077 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002079 * We're at the head of the list. Get the inode's on-disk
2080 * buffer to see if there is anyone after us on the list.
2081 * Only modify our next pointer if it is not already NULLAGINO.
2082 * This saves us the overhead of dealing with the buffer when
2083 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002085 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2086 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002088 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002089 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return error;
2091 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002092 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 ASSERT(next_agino != 0);
2094 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002095 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002096 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002098
2099 /* need to recalc the inode CRC if appropriate */
2100 xfs_dinode_calc_crc(mp, dip);
2101
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 xfs_trans_inode_buf(tp, ibp);
2103 xfs_trans_log_buf(tp, ibp, offset,
2104 (offset + sizeof(xfs_agino_t) - 1));
2105 xfs_inobp_check(mp, ibp);
2106 } else {
2107 xfs_trans_brelse(tp, ibp);
2108 }
2109 /*
2110 * Point the bucket head pointer at the next inode.
2111 */
2112 ASSERT(next_agino != 0);
2113 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002114 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 offset = offsetof(xfs_agi_t, agi_unlinked) +
2116 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002117 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 xfs_trans_log_buf(tp, agibp, offset,
2119 (offset + sizeof(xfs_agino_t) - 1));
2120 } else {
2121 /*
2122 * We need to search the list for the inode being freed.
2123 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002124 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 last_ibp = NULL;
2126 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002127 struct xfs_imap imap;
2128
2129 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002131
2132 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002134
2135 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002137 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002138 "%s: xfs_imap returned error %d.",
2139 __func__, error);
2140 return error;
2141 }
2142
2143 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2144 &last_ibp, 0, 0);
2145 if (error) {
2146 xfs_warn(mp,
2147 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002148 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 return error;
2150 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002151
2152 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002153 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 ASSERT(next_agino != NULLAGINO);
2155 ASSERT(next_agino != 0);
2156 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002159 * Now last_ibp points to the buffer previous to us on the
2160 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002162 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2163 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002165 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002166 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 return error;
2168 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002169 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 ASSERT(next_agino != 0);
2171 ASSERT(next_agino != agino);
2172 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002173 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002174 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002176
2177 /* need to recalc the inode CRC if appropriate */
2178 xfs_dinode_calc_crc(mp, dip);
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 xfs_trans_inode_buf(tp, ibp);
2181 xfs_trans_log_buf(tp, ibp, offset,
2182 (offset + sizeof(xfs_agino_t) - 1));
2183 xfs_inobp_check(mp, ibp);
2184 } else {
2185 xfs_trans_brelse(tp, ibp);
2186 }
2187 /*
2188 * Point the previous inode on the list to the next inode.
2189 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002190 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 ASSERT(next_agino != 0);
2192 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002193
2194 /* need to recalc the inode CRC if appropriate */
2195 xfs_dinode_calc_crc(mp, last_dip);
2196
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 xfs_trans_inode_buf(tp, last_ibp);
2198 xfs_trans_log_buf(tp, last_ibp, offset,
2199 (offset + sizeof(xfs_agino_t) - 1));
2200 xfs_inobp_check(mp, last_ibp);
2201 }
2202 return 0;
2203}
2204
Dave Chinner5b3eed72010-08-24 11:42:41 +10002205/*
Zhi Yong Wu0b8182d2013-08-12 03:14:59 +00002206 * A big issue when freeing the inode cluster is that we _cannot_ skip any
Dave Chinner5b3eed72010-08-24 11:42:41 +10002207 * inodes that are in memory - they all must be marked stale and attached to
2208 * the cluster buffer.
2209 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002210STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211xfs_ifree_cluster(
Brian Foster09b56602015-05-29 09:26:03 +10002212 xfs_inode_t *free_ip,
2213 xfs_trans_t *tp,
2214 struct xfs_icluster *xic)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215{
2216 xfs_mount_t *mp = free_ip->i_mount;
2217 int blks_per_cluster;
Jie Liu982e9392013-12-13 15:51:49 +11002218 int inodes_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 int nbufs;
Dave Chinner5b257b42010-06-03 16:22:29 +10002220 int i, j;
Brian Foster3cdaa182015-06-04 13:03:34 +10002221 int ioffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 xfs_daddr_t blkno;
2223 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10002224 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 xfs_inode_log_item_t *iip;
2226 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00002227 struct xfs_perag *pag;
Brian Foster09b56602015-05-29 09:26:03 +10002228 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Brian Foster09b56602015-05-29 09:26:03 +10002230 inum = xic->first_ino;
Dave Chinner5017e972010-01-11 11:47:40 +00002231 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Jie Liu982e9392013-12-13 15:51:49 +11002232 blks_per_cluster = xfs_icluster_size_fsb(mp);
2233 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2234 nbufs = mp->m_ialloc_blks / blks_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
Jie Liu982e9392013-12-13 15:51:49 +11002236 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
Brian Foster09b56602015-05-29 09:26:03 +10002237 /*
2238 * The allocation bitmap tells us which inodes of the chunk were
2239 * physically allocated. Skip the cluster if an inode falls into
2240 * a sparse region.
2241 */
Brian Foster3cdaa182015-06-04 13:03:34 +10002242 ioffset = inum - xic->first_ino;
2243 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2244 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
Brian Foster09b56602015-05-29 09:26:03 +10002245 continue;
2246 }
2247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2249 XFS_INO_TO_AGBNO(mp, inum));
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10002252 * We obtain and lock the backing buffer first in the process
2253 * here, as we have to ensure that any dirty inode that we
2254 * can't get the flush lock on is attached to the buffer.
2255 * If we scan the in-memory inodes first, then buffer IO can
2256 * complete before we get a lock on it, and hence we may fail
2257 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002259 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11002260 mp->m_bsize * blks_per_cluster,
2261 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002263 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002264 return -ENOMEM;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002265
2266 /*
2267 * This buffer may not have been correctly initialised as we
2268 * didn't read it from disk. That's not important because we are
2269 * only using to mark the buffer as stale in the log, and to
2270 * attach stale cached inodes on it. That means it will never be
2271 * dispatched for IO. If it is, we want to know about it, and we
2272 * want it to fail. We can acheive this by adding a write
2273 * verifier to the buffer.
2274 */
Dave Chinner1813dd62012-11-14 17:54:40 +11002275 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002276
Dave Chinner5b257b42010-06-03 16:22:29 +10002277 /*
2278 * Walk the inodes already attached to the buffer and mark them
2279 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10002280 * in-memory inode walk can't lock them. By marking them all
2281 * stale first, we will not attempt to lock them in the loop
2282 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10002283 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002284 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 while (lip) {
2286 if (lip->li_type == XFS_LI_INODE) {
2287 iip = (xfs_inode_log_item_t *)lip;
2288 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002289 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002290 xfs_trans_ail_copy_lsn(mp->m_ail,
2291 &iip->ili_flush_lsn,
2292 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002293 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295 lip = lip->li_bio_list;
2296 }
2297
Dave Chinner5b3eed72010-08-24 11:42:41 +10002298
Dave Chinner5b257b42010-06-03 16:22:29 +10002299 /*
2300 * For each inode in memory attempt to add it to the inode
2301 * buffer and set it up for being staled on buffer IO
2302 * completion. This is safe as we've locked out tail pushing
2303 * and flushing by locking the buffer.
2304 *
2305 * We have already marked every inode that was part of a
2306 * transaction stale above, which means there is no point in
2307 * even trying to lock them.
2308 */
Jie Liu982e9392013-12-13 15:51:49 +11002309 for (i = 0; i < inodes_per_cluster; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10002310retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002311 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002312 ip = radix_tree_lookup(&pag->pag_ici_root,
2313 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002315 /* Inode not in memory, nothing to do */
2316 if (!ip) {
2317 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002318 continue;
2319 }
2320
Dave Chinner5b3eed72010-08-24 11:42:41 +10002321 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002322 * because this is an RCU protected lookup, we could
2323 * find a recently freed or even reallocated inode
2324 * during the lookup. We need to check under the
2325 * i_flags_lock for a valid inode here. Skip it if it
2326 * is not valid, the wrong inode or stale.
2327 */
2328 spin_lock(&ip->i_flags_lock);
2329 if (ip->i_ino != inum + i ||
2330 __xfs_iflags_test(ip, XFS_ISTALE)) {
2331 spin_unlock(&ip->i_flags_lock);
2332 rcu_read_unlock();
2333 continue;
2334 }
2335 spin_unlock(&ip->i_flags_lock);
2336
2337 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002338 * Don't try to lock/unlock the current inode, but we
2339 * _cannot_ skip the other inodes that we did not find
2340 * in the list attached to the buffer and are not
2341 * already marked stale. If we can't lock it, back off
2342 * and retry.
2343 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002344 if (ip != free_ip &&
2345 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002346 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002347 delay(1);
2348 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002349 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002350 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002351
Dave Chinner5b3eed72010-08-24 11:42:41 +10002352 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002353 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002354
Dave Chinner5b3eed72010-08-24 11:42:41 +10002355 /*
2356 * we don't need to attach clean inodes or those only
2357 * with unlogged changes (which we throw away, anyway).
2358 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002359 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002360 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002361 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 xfs_ifunlock(ip);
2363 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2364 continue;
2365 }
2366
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002367 iip->ili_last_fields = iip->ili_fields;
2368 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11002369 iip->ili_fsync_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002371 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2372 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002374 xfs_buf_attach_iodone(bp, xfs_istale_done,
2375 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002376
2377 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 }
2380
Dave Chinner5b3eed72010-08-24 11:42:41 +10002381 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 xfs_trans_binval(tp, bp);
2383 }
2384
Dave Chinner5017e972010-01-11 11:47:40 +00002385 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002386 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387}
2388
2389/*
2390 * This is called to return an inode to the inode free list.
2391 * The inode should already be truncated to 0 length and have
2392 * no pages associated with it. This routine also assumes that
2393 * the inode is already a part of the transaction.
2394 *
2395 * The on-disk copy of the inode will have been added to the list
2396 * of unlinked inodes in the AGI. We need to remove the inode from
2397 * that list atomically with respect to freeing it here.
2398 */
2399int
2400xfs_ifree(
2401 xfs_trans_t *tp,
2402 xfs_inode_t *ip,
2403 xfs_bmap_free_t *flist)
2404{
2405 int error;
Brian Foster09b56602015-05-29 09:26:03 +10002406 struct xfs_icluster xic = { 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002408 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 ASSERT(ip->i_d.di_nlink == 0);
2410 ASSERT(ip->i_d.di_nextents == 0);
2411 ASSERT(ip->i_d.di_anextents == 0);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002412 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 ASSERT(ip->i_d.di_nblocks == 0);
2414
2415 /*
2416 * Pull the on-disk inode from the AGI unlinked list.
2417 */
2418 error = xfs_iunlink_remove(tp, ip);
Dave Chinner1baaed82013-06-27 16:04:50 +10002419 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Brian Foster09b56602015-05-29 09:26:03 +10002422 error = xfs_difree(tp, ip->i_ino, flist, &xic);
Dave Chinner1baaed82013-06-27 16:04:50 +10002423 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return error;
Dave Chinner1baaed82013-06-27 16:04:50 +10002425
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 ip->i_d.di_mode = 0; /* mark incore inode as free */
2427 ip->i_d.di_flags = 0;
2428 ip->i_d.di_dmevmask = 0;
2429 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2431 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2432 /*
2433 * Bump the generation count so no one will be confused
2434 * by reincarnations of this inode.
2435 */
2436 ip->i_d.di_gen++;
2437 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2438
Brian Foster09b56602015-05-29 09:26:03 +10002439 if (xic.deleted)
2440 error = xfs_ifree_cluster(ip, tp, &xic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002442 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443}
2444
2445/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002446 * This is called to unpin an inode. The caller must have the inode locked
2447 * in at least shared mode so that the buffer cannot be subsequently pinned
2448 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002450static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002451xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002452 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002453{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002454 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002455
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002456 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2457
David Chinnera3f74ff2008-03-06 13:43:42 +11002458 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002459 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002460
David Chinnera3f74ff2008-03-06 13:43:42 +11002461}
2462
Christoph Hellwigf392e632011-12-18 20:00:10 +00002463static void
2464__xfs_iunpin_wait(
2465 struct xfs_inode *ip)
2466{
2467 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2468 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2469
2470 xfs_iunpin(ip);
2471
2472 do {
2473 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2474 if (xfs_ipincount(ip))
2475 io_schedule();
2476 } while (xfs_ipincount(ip));
2477 finish_wait(wq, &wait.wait);
2478}
2479
Dave Chinner777df5a2010-02-06 12:37:26 +11002480void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002482 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002484 if (xfs_ipincount(ip))
2485 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Dave Chinner27320362013-10-29 22:11:44 +11002488/*
2489 * Removing an inode from the namespace involves removing the directory entry
2490 * and dropping the link count on the inode. Removing the directory entry can
2491 * result in locking an AGF (directory blocks were freed) and removing a link
2492 * count can result in placing the inode on an unlinked list which results in
2493 * locking an AGI.
2494 *
2495 * The big problem here is that we have an ordering constraint on AGF and AGI
2496 * locking - inode allocation locks the AGI, then can allocate a new extent for
2497 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2498 * removes the inode from the unlinked list, requiring that we lock the AGI
2499 * first, and then freeing the inode can result in an inode chunk being freed
2500 * and hence freeing disk space requiring that we lock an AGF.
2501 *
2502 * Hence the ordering that is imposed by other parts of the code is AGI before
2503 * AGF. This means we cannot remove the directory entry before we drop the inode
2504 * reference count and put it on the unlinked list as this results in a lock
2505 * order of AGF then AGI, and this can deadlock against inode allocation and
2506 * freeing. Therefore we must drop the link counts before we remove the
2507 * directory entry.
2508 *
2509 * This is still safe from a transactional point of view - it is not until we
2510 * get to xfs_bmap_finish() that we have the possibility of multiple
2511 * transactions in this operation. Hence as long as we remove the directory
2512 * entry and drop the link count in the first transaction of the remove
2513 * operation, there are no transactional constraints on the ordering here.
2514 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002515int
2516xfs_remove(
2517 xfs_inode_t *dp,
2518 struct xfs_name *name,
2519 xfs_inode_t *ip)
2520{
2521 xfs_mount_t *mp = dp->i_mount;
2522 xfs_trans_t *tp = NULL;
2523 int is_dir = S_ISDIR(ip->i_d.di_mode);
2524 int error = 0;
2525 xfs_bmap_free_t free_list;
2526 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002527 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002528
2529 trace_xfs_remove(dp, name);
2530
2531 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10002532 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002533
2534 error = xfs_qm_dqattach(dp, 0);
2535 if (error)
2536 goto std_return;
2537
2538 error = xfs_qm_dqattach(ip, 0);
2539 if (error)
2540 goto std_return;
2541
Dave Chinner32296f82014-12-04 09:43:17 +11002542 if (is_dir)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002543 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
Dave Chinner32296f82014-12-04 09:43:17 +11002544 else
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002545 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002546
2547 /*
2548 * We try to get the real space reservation first,
2549 * allowing for directory btree deletion(s) implying
2550 * possible bmap insert(s). If we can't get the space
2551 * reservation then we use 0 instead, and avoid the bmap
2552 * btree insert(s) in the directory code by, if the bmap
2553 * insert tries to happen, instead trimming the LAST
2554 * block from the directory.
2555 */
2556 resblks = XFS_REMOVE_SPACE_RES(mp);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002557 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002558 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002559 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002560 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002561 }
2562 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002563 ASSERT(error != -ENOSPC);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002564 goto out_trans_cancel;
2565 }
2566
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002567 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002568 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2569
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002570 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002571 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2572
2573 /*
2574 * If we're removing a directory perform some additional validation.
2575 */
2576 if (is_dir) {
2577 ASSERT(ip->i_d.di_nlink >= 2);
2578 if (ip->i_d.di_nlink != 2) {
Dave Chinner24513372014-06-25 14:58:08 +10002579 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002580 goto out_trans_cancel;
2581 }
2582 if (!xfs_dir_isempty(ip)) {
Dave Chinner24513372014-06-25 14:58:08 +10002583 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002584 goto out_trans_cancel;
2585 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002586
Dave Chinner27320362013-10-29 22:11:44 +11002587 /* Drop the link from ip's "..". */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002588 error = xfs_droplink(tp, dp);
2589 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002590 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002591
Dave Chinner27320362013-10-29 22:11:44 +11002592 /* Drop the "." link from ip to self. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002593 error = xfs_droplink(tp, ip);
2594 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002595 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002596 } else {
2597 /*
2598 * When removing a non-directory we need to log the parent
2599 * inode here. For a directory this is done implicitly
2600 * by the xfs_droplink call for the ".." entry.
2601 */
2602 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2603 }
Dave Chinner27320362013-10-29 22:11:44 +11002604 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002605
Dave Chinner27320362013-10-29 22:11:44 +11002606 /* Drop the link from dp to ip. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002607 error = xfs_droplink(tp, ip);
2608 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002609 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002610
Dave Chinner27320362013-10-29 22:11:44 +11002611 xfs_bmap_init(&free_list, &first_block);
2612 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2613 &first_block, &free_list, resblks);
2614 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002615 ASSERT(error != -ENOENT);
Dave Chinner27320362013-10-29 22:11:44 +11002616 goto out_bmap_cancel;
2617 }
2618
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002619 /*
2620 * If this is a synchronous mount, make sure that the
2621 * remove transaction goes to disk before returning to
2622 * the user.
2623 */
2624 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2625 xfs_trans_set_sync(tp);
2626
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002627 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002628 if (error)
2629 goto out_bmap_cancel;
2630
Christoph Hellwig70393312015-06-04 13:48:08 +10002631 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002632 if (error)
2633 goto std_return;
2634
Christoph Hellwig2cd2ef62014-04-23 07:11:51 +10002635 if (is_dir && xfs_inode_is_filestream(ip))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002636 xfs_filestream_deassociate(ip);
2637
2638 return 0;
2639
2640 out_bmap_cancel:
2641 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002642 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10002643 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002644 std_return:
2645 return error;
2646}
2647
Dave Chinnerf6bba202013-08-12 20:49:46 +10002648/*
2649 * Enter all inodes for a rename transaction into a sorted array.
2650 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002651#define __XFS_SORT_INODES 5
Dave Chinnerf6bba202013-08-12 20:49:46 +10002652STATIC void
2653xfs_sort_for_rename(
Dave Chinner95afcf52015-03-25 14:03:32 +11002654 struct xfs_inode *dp1, /* in: old (source) directory inode */
2655 struct xfs_inode *dp2, /* in: new (target) directory inode */
2656 struct xfs_inode *ip1, /* in: inode of old entry */
2657 struct xfs_inode *ip2, /* in: inode of new entry */
2658 struct xfs_inode *wip, /* in: whiteout inode */
2659 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2660 int *num_inodes) /* in/out: inodes in array */
Dave Chinnerf6bba202013-08-12 20:49:46 +10002661{
Dave Chinnerf6bba202013-08-12 20:49:46 +10002662 int i, j;
2663
Dave Chinner95afcf52015-03-25 14:03:32 +11002664 ASSERT(*num_inodes == __XFS_SORT_INODES);
2665 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2666
Dave Chinnerf6bba202013-08-12 20:49:46 +10002667 /*
2668 * i_tab contains a list of pointers to inodes. We initialize
2669 * the table here & we'll sort it. We will then use it to
2670 * order the acquisition of the inode locks.
2671 *
2672 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2673 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002674 i = 0;
2675 i_tab[i++] = dp1;
2676 i_tab[i++] = dp2;
2677 i_tab[i++] = ip1;
2678 if (ip2)
2679 i_tab[i++] = ip2;
2680 if (wip)
2681 i_tab[i++] = wip;
2682 *num_inodes = i;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002683
2684 /*
2685 * Sort the elements via bubble sort. (Remember, there are at
Dave Chinner95afcf52015-03-25 14:03:32 +11002686 * most 5 elements to sort, so this is adequate.)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002687 */
2688 for (i = 0; i < *num_inodes; i++) {
2689 for (j = 1; j < *num_inodes; j++) {
2690 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
Dave Chinner95afcf52015-03-25 14:03:32 +11002691 struct xfs_inode *temp = i_tab[j];
Dave Chinnerf6bba202013-08-12 20:49:46 +10002692 i_tab[j] = i_tab[j-1];
2693 i_tab[j-1] = temp;
2694 }
2695 }
2696 }
2697}
2698
Dave Chinner310606b2015-03-25 14:06:07 +11002699static int
2700xfs_finish_rename(
2701 struct xfs_trans *tp,
2702 struct xfs_bmap_free *free_list)
2703{
Dave Chinner310606b2015-03-25 14:06:07 +11002704 int error;
2705
2706 /*
2707 * If this is a synchronous mount, make sure that the rename transaction
2708 * goes to disk before returning to the user.
2709 */
2710 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2711 xfs_trans_set_sync(tp);
2712
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002713 error = xfs_bmap_finish(&tp, free_list, NULL);
Dave Chinner310606b2015-03-25 14:06:07 +11002714 if (error) {
2715 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002716 xfs_trans_cancel(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002717 return error;
2718 }
2719
Christoph Hellwig70393312015-06-04 13:48:08 +10002720 return xfs_trans_commit(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002721}
2722
Dave Chinnerf6bba202013-08-12 20:49:46 +10002723/*
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002724 * xfs_cross_rename()
2725 *
2726 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2727 */
2728STATIC int
2729xfs_cross_rename(
2730 struct xfs_trans *tp,
2731 struct xfs_inode *dp1,
2732 struct xfs_name *name1,
2733 struct xfs_inode *ip1,
2734 struct xfs_inode *dp2,
2735 struct xfs_name *name2,
2736 struct xfs_inode *ip2,
2737 struct xfs_bmap_free *free_list,
2738 xfs_fsblock_t *first_block,
2739 int spaceres)
2740{
2741 int error = 0;
2742 int ip1_flags = 0;
2743 int ip2_flags = 0;
2744 int dp2_flags = 0;
2745
2746 /* Swap inode number for dirent in first parent */
2747 error = xfs_dir_replace(tp, dp1, name1,
2748 ip2->i_ino,
2749 first_block, free_list, spaceres);
2750 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002751 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002752
2753 /* Swap inode number for dirent in second parent */
2754 error = xfs_dir_replace(tp, dp2, name2,
2755 ip1->i_ino,
2756 first_block, free_list, spaceres);
2757 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002758 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002759
2760 /*
2761 * If we're renaming one or more directories across different parents,
2762 * update the respective ".." entries (and link counts) to match the new
2763 * parents.
2764 */
2765 if (dp1 != dp2) {
2766 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2767
2768 if (S_ISDIR(ip2->i_d.di_mode)) {
2769 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2770 dp1->i_ino, first_block,
2771 free_list, spaceres);
2772 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002773 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002774
2775 /* transfer ip2 ".." reference to dp1 */
2776 if (!S_ISDIR(ip1->i_d.di_mode)) {
2777 error = xfs_droplink(tp, dp2);
2778 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002779 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002780 error = xfs_bumplink(tp, dp1);
2781 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002782 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002783 }
2784
2785 /*
2786 * Although ip1 isn't changed here, userspace needs
2787 * to be warned about the change, so that applications
2788 * relying on it (like backup ones), will properly
2789 * notify the change
2790 */
2791 ip1_flags |= XFS_ICHGTIME_CHG;
2792 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2793 }
2794
2795 if (S_ISDIR(ip1->i_d.di_mode)) {
2796 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2797 dp2->i_ino, first_block,
2798 free_list, spaceres);
2799 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002800 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002801
2802 /* transfer ip1 ".." reference to dp2 */
2803 if (!S_ISDIR(ip2->i_d.di_mode)) {
2804 error = xfs_droplink(tp, dp1);
2805 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002806 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002807 error = xfs_bumplink(tp, dp2);
2808 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002809 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002810 }
2811
2812 /*
2813 * Although ip2 isn't changed here, userspace needs
2814 * to be warned about the change, so that applications
2815 * relying on it (like backup ones), will properly
2816 * notify the change
2817 */
2818 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2819 ip2_flags |= XFS_ICHGTIME_CHG;
2820 }
2821 }
2822
2823 if (ip1_flags) {
2824 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2825 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2826 }
2827 if (ip2_flags) {
2828 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2829 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2830 }
2831 if (dp2_flags) {
2832 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2833 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2834 }
2835 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2836 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
Dave Chinnereeacd322015-03-25 14:08:07 +11002837 return xfs_finish_rename(tp, free_list);
2838
2839out_trans_abort:
2840 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002841 xfs_trans_cancel(tp);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002842 return error;
2843}
2844
2845/*
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002846 * xfs_rename_alloc_whiteout()
2847 *
2848 * Return a referenced, unlinked, unlocked inode that that can be used as a
2849 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2850 * crash between allocating the inode and linking it into the rename transaction
2851 * recovery will free the inode and we won't leak it.
2852 */
2853static int
2854xfs_rename_alloc_whiteout(
2855 struct xfs_inode *dp,
2856 struct xfs_inode **wip)
2857{
2858 struct xfs_inode *tmpfile;
2859 int error;
2860
2861 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2862 if (error)
2863 return error;
2864
Brian Foster22419ac2015-05-29 08:14:55 +10002865 /*
2866 * Prepare the tmpfile inode as if it were created through the VFS.
2867 * Otherwise, the link increment paths will complain about nlink 0->1.
2868 * Drop the link count as done by d_tmpfile(), complete the inode setup
2869 * and flag it as linkable.
2870 */
2871 drop_nlink(VFS_I(tmpfile));
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002872 xfs_finish_inode_setup(tmpfile);
2873 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2874
2875 *wip = tmpfile;
2876 return 0;
2877}
2878
2879/*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002880 * xfs_rename
2881 */
2882int
2883xfs_rename(
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002884 struct xfs_inode *src_dp,
2885 struct xfs_name *src_name,
2886 struct xfs_inode *src_ip,
2887 struct xfs_inode *target_dp,
2888 struct xfs_name *target_name,
2889 struct xfs_inode *target_ip,
2890 unsigned int flags)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002891{
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002892 struct xfs_mount *mp = src_dp->i_mount;
2893 struct xfs_trans *tp;
2894 struct xfs_bmap_free free_list;
2895 xfs_fsblock_t first_block;
2896 struct xfs_inode *wip = NULL; /* whiteout inode */
2897 struct xfs_inode *inodes[__XFS_SORT_INODES];
2898 int num_inodes = __XFS_SORT_INODES;
Dave Chinner2b936812015-03-25 15:12:30 +11002899 bool new_parent = (src_dp != target_dp);
2900 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002901 int spaceres;
2902 int error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002903
2904 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2905
Dave Chinnereeacd322015-03-25 14:08:07 +11002906 if ((flags & RENAME_EXCHANGE) && !target_ip)
2907 return -EINVAL;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002908
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002909 /*
2910 * If we are doing a whiteout operation, allocate the whiteout inode
2911 * we will be placing at the target and ensure the type is set
2912 * appropriately.
2913 */
2914 if (flags & RENAME_WHITEOUT) {
2915 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2916 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2917 if (error)
2918 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002919
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002920 /* setup target dirent info as whiteout */
2921 src_name->type = XFS_DIR3_FT_CHRDEV;
2922 }
2923
2924 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
Dave Chinnerf6bba202013-08-12 20:49:46 +10002925 inodes, &num_inodes);
2926
Dave Chinnerf6bba202013-08-12 20:49:46 +10002927 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002928 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002929 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002930 if (error == -ENOSPC) {
Dave Chinnerf6bba202013-08-12 20:49:46 +10002931 spaceres = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002932 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002933 }
Dave Chinner445883e2015-03-25 14:05:43 +11002934 if (error)
2935 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002936
2937 /*
2938 * Attach the dquots to the inodes
2939 */
2940 error = xfs_qm_vop_rename_dqattach(inodes);
Dave Chinner445883e2015-03-25 14:05:43 +11002941 if (error)
2942 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002943
2944 /*
2945 * Lock all the participating inodes. Depending upon whether
2946 * the target_name exists in the target directory, and
2947 * whether the target directory is the same as the source
2948 * directory, we can lock from 2 to 4 inodes.
2949 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002950 if (!new_parent)
2951 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2952 else
2953 xfs_lock_two_inodes(src_dp, target_dp,
2954 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2955
Dave Chinnerf6bba202013-08-12 20:49:46 +10002956 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2957
2958 /*
2959 * Join all the inodes to the transaction. From this point on,
2960 * we can rely on either trans_commit or trans_cancel to unlock
2961 * them.
2962 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002963 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002964 if (new_parent)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002965 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002966 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2967 if (target_ip)
2968 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002969 if (wip)
2970 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002971
2972 /*
2973 * If we are using project inheritance, we only allow renames
2974 * into our tree when the project IDs are the same; else the
2975 * tree quota mechanism would be circumvented.
2976 */
2977 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2978 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10002979 error = -EXDEV;
Dave Chinner445883e2015-03-25 14:05:43 +11002980 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002981 }
2982
Dave Chinner445883e2015-03-25 14:05:43 +11002983 xfs_bmap_init(&free_list, &first_block);
2984
Dave Chinnereeacd322015-03-25 14:08:07 +11002985 /* RENAME_EXCHANGE is unique from here on. */
2986 if (flags & RENAME_EXCHANGE)
2987 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2988 target_dp, target_name, target_ip,
2989 &free_list, &first_block, spaceres);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002990
2991 /*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002992 * Set up the target.
2993 */
2994 if (target_ip == NULL) {
2995 /*
2996 * If there's no space reservation, check the entry will
2997 * fit before actually inserting it.
2998 */
Eric Sandeen94f3cad2014-09-09 11:57:52 +10002999 if (!spaceres) {
3000 error = xfs_dir_canenter(tp, target_dp, target_name);
3001 if (error)
Dave Chinner445883e2015-03-25 14:05:43 +11003002 goto out_trans_cancel;
Eric Sandeen94f3cad2014-09-09 11:57:52 +10003003 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003004 /*
3005 * If target does not exist and the rename crosses
3006 * directories, adjust the target directory link count
3007 * to account for the ".." reference from the new entry.
3008 */
3009 error = xfs_dir_createname(tp, target_dp, target_name,
3010 src_ip->i_ino, &first_block,
3011 &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003012 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003013 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003014
3015 xfs_trans_ichgtime(tp, target_dp,
3016 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3017
3018 if (new_parent && src_is_directory) {
3019 error = xfs_bumplink(tp, target_dp);
3020 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003021 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003022 }
3023 } else { /* target_ip != NULL */
3024 /*
3025 * If target exists and it's a directory, check that both
3026 * target and source are directories and that target can be
3027 * destroyed, or that neither is a directory.
3028 */
3029 if (S_ISDIR(target_ip->i_d.di_mode)) {
3030 /*
3031 * Make sure target dir is empty.
3032 */
3033 if (!(xfs_dir_isempty(target_ip)) ||
3034 (target_ip->i_d.di_nlink > 2)) {
Dave Chinner24513372014-06-25 14:58:08 +10003035 error = -EEXIST;
Dave Chinner445883e2015-03-25 14:05:43 +11003036 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003037 }
3038 }
3039
3040 /*
3041 * Link the source inode under the target name.
3042 * If the source inode is a directory and we are moving
3043 * it across directories, its ".." entry will be
3044 * inconsistent until we replace that down below.
3045 *
3046 * In case there is already an entry with the same
3047 * name at the destination directory, remove it first.
3048 */
3049 error = xfs_dir_replace(tp, target_dp, target_name,
3050 src_ip->i_ino,
3051 &first_block, &free_list, spaceres);
3052 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003053 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003054
3055 xfs_trans_ichgtime(tp, target_dp,
3056 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3057
3058 /*
3059 * Decrement the link count on the target since the target
3060 * dir no longer points to it.
3061 */
3062 error = xfs_droplink(tp, target_ip);
3063 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003064 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003065
3066 if (src_is_directory) {
3067 /*
3068 * Drop the link from the old "." entry.
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 } /* target_ip != NULL */
3075
3076 /*
3077 * Remove the source.
3078 */
3079 if (new_parent && src_is_directory) {
3080 /*
3081 * Rewrite the ".." entry to point to the new
3082 * directory.
3083 */
3084 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3085 target_dp->i_ino,
3086 &first_block, &free_list, spaceres);
Dave Chinner24513372014-06-25 14:58:08 +10003087 ASSERT(error != -EEXIST);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003088 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003089 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003090 }
3091
3092 /*
3093 * We always want to hit the ctime on the source inode.
3094 *
3095 * This isn't strictly required by the standards since the source
3096 * inode isn't really being changed, but old unix file systems did
3097 * it and some incremental backup programs won't work without it.
3098 */
3099 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3100 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3101
3102 /*
3103 * Adjust the link count on src_dp. This is necessary when
3104 * renaming a directory, either within one parent when
3105 * the target existed, or across two parent directories.
3106 */
3107 if (src_is_directory && (new_parent || target_ip != NULL)) {
3108
3109 /*
3110 * Decrement link count on src_directory since the
3111 * entry that's moved no longer points to it.
3112 */
3113 error = xfs_droplink(tp, src_dp);
3114 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003115 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003116 }
3117
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003118 /*
3119 * For whiteouts, we only need to update the source dirent with the
3120 * inode number of the whiteout inode rather than removing it
3121 * altogether.
3122 */
3123 if (wip) {
3124 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
Dave Chinnerf6bba202013-08-12 20:49:46 +10003125 &first_block, &free_list, spaceres);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003126 } else
3127 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3128 &first_block, &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003129 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003130 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003131
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003132 /*
3133 * For whiteouts, we need to bump the link count on the whiteout inode.
3134 * This means that failures all the way up to this point leave the inode
3135 * on the unlinked list and so cleanup is a simple matter of dropping
3136 * the remaining reference to it. If we fail here after bumping the link
3137 * count, we're shutting down the filesystem so we'll never see the
3138 * intermediate state on disk.
3139 */
3140 if (wip) {
Brian Foster22419ac2015-05-29 08:14:55 +10003141 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003142 error = xfs_bumplink(tp, wip);
3143 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003144 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003145 error = xfs_iunlink_remove(tp, wip);
3146 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003147 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003148 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3149
3150 /*
3151 * Now we have a real link, clear the "I'm a tmpfile" state
3152 * flag from the inode so it doesn't accidentally get misused in
3153 * future.
3154 */
3155 VFS_I(wip)->i_state &= ~I_LINKABLE;
3156 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003157
3158 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3159 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3160 if (new_parent)
3161 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3162
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003163 error = xfs_finish_rename(tp, &free_list);
3164 if (wip)
3165 IRELE(wip);
3166 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003167
Dave Chinner445883e2015-03-25 14:05:43 +11003168out_bmap_cancel:
Dave Chinnerf6bba202013-08-12 20:49:46 +10003169 xfs_bmap_cancel(&free_list);
Dave Chinner445883e2015-03-25 14:05:43 +11003170out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003171 xfs_trans_cancel(tp);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003172 if (wip)
3173 IRELE(wip);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003174 return error;
3175}
3176
David Chinnerbad55842008-03-06 13:43:49 +11003177STATIC int
3178xfs_iflush_cluster(
3179 xfs_inode_t *ip,
3180 xfs_buf_t *bp)
3181{
3182 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00003183 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11003184 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10003185 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11003186 int ilist_size;
3187 xfs_inode_t **ilist;
3188 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11003189 int nr_found;
3190 int clcount = 0;
3191 int bufwasdelwri;
3192 int i;
3193
Dave Chinner5017e972010-01-11 11:47:40 +00003194 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11003195
Jie Liu0f49efd2013-12-13 15:51:48 +11003196 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
David Chinnerc8f5f122008-05-20 11:30:15 +10003197 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10003198 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11003199 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00003200 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11003201
Jie Liu0f49efd2013-12-13 15:51:48 +11003202 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
David Chinnerbad55842008-03-06 13:43:49 +11003203 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003204 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11003205 /* really need a gang lookup range call here */
3206 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10003207 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11003208 if (nr_found == 0)
3209 goto out_free;
3210
3211 for (i = 0; i < nr_found; i++) {
3212 iq = ilist[i];
3213 if (iq == ip)
3214 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003215
3216 /*
3217 * because this is an RCU protected lookup, we could find a
3218 * recently freed or even reallocated inode during the lookup.
3219 * We need to check under the i_flags_lock for a valid inode
3220 * here. Skip it if it is not valid or the wrong inode.
3221 */
3222 spin_lock(&ip->i_flags_lock);
3223 if (!ip->i_ino ||
3224 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3225 spin_unlock(&ip->i_flags_lock);
3226 continue;
3227 }
3228 spin_unlock(&ip->i_flags_lock);
3229
David Chinnerbad55842008-03-06 13:43:49 +11003230 /*
3231 * Do an un-protected check to see if the inode is dirty and
3232 * is a candidate for flushing. These checks will be repeated
3233 * later after the appropriate locks are acquired.
3234 */
David Chinner33540402008-03-06 13:43:59 +11003235 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11003236 continue;
David Chinnerbad55842008-03-06 13:43:49 +11003237
3238 /*
3239 * Try to get locks. If any are unavailable or it is pinned,
3240 * then this inode cannot be flushed and is skipped.
3241 */
3242
3243 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3244 continue;
3245 if (!xfs_iflock_nowait(iq)) {
3246 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3247 continue;
3248 }
3249 if (xfs_ipincount(iq)) {
3250 xfs_ifunlock(iq);
3251 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3252 continue;
3253 }
3254
3255 /*
3256 * arriving here means that this inode can be flushed. First
3257 * re-check that it's dirty before flushing.
3258 */
David Chinner33540402008-03-06 13:43:59 +11003259 if (!xfs_inode_clean(iq)) {
3260 int error;
David Chinnerbad55842008-03-06 13:43:49 +11003261 error = xfs_iflush_int(iq, bp);
3262 if (error) {
3263 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3264 goto cluster_corrupt_out;
3265 }
3266 clcount++;
3267 } else {
3268 xfs_ifunlock(iq);
3269 }
3270 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3271 }
3272
3273 if (clcount) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003274 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3275 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
David Chinnerbad55842008-03-06 13:43:49 +11003276 }
3277
3278out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003279 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003280 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003281out_put:
3282 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11003283 return 0;
3284
3285
3286cluster_corrupt_out:
3287 /*
3288 * Corruption detected in the clustering loop. Invalidate the
3289 * inode buffer and shut down the filesystem.
3290 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003291 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11003292 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003293 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11003294 * brelse can handle it with no problems. If not, shut down the
3295 * filesystem before releasing the buffer.
3296 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003297 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11003298 if (bufwasdelwri)
3299 xfs_buf_relse(bp);
3300
3301 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3302
3303 if (!bufwasdelwri) {
3304 /*
3305 * Just like incore_relse: if we have b_iodone functions,
3306 * mark the buffer as an error and call them. Otherwise
3307 * mark it as stale and brelse.
3308 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003309 if (bp->b_iodone) {
David Chinnerbad55842008-03-06 13:43:49 +11003310 XFS_BUF_UNDONE(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003311 xfs_buf_stale(bp);
Dave Chinner24513372014-06-25 14:58:08 +10003312 xfs_buf_ioerror(bp, -EIO);
Dave Chinnere8aaba92014-10-02 09:04:22 +10003313 xfs_buf_ioend(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003314 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003315 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003316 xfs_buf_relse(bp);
3317 }
3318 }
3319
3320 /*
3321 * Unlocks the flush lock
3322 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003323 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003324 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003325 xfs_perag_put(pag);
Dave Chinner24513372014-06-25 14:58:08 +10003326 return -EFSCORRUPTED;
David Chinnerbad55842008-03-06 13:43:49 +11003327}
3328
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10003330 * Flush dirty inode metadata into the backing buffer.
3331 *
3332 * The caller must have the inode lock and the inode flush lock held. The
3333 * inode lock will still be held upon return to the caller, and the inode
3334 * flush lock will be released after the inode has reached the disk.
3335 *
3336 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 */
3338int
3339xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10003340 struct xfs_inode *ip,
3341 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342{
Christoph Hellwig4c468192012-04-23 15:58:36 +10003343 struct xfs_mount *mp = ip->i_mount;
3344 struct xfs_buf *bp;
3345 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003348 XFS_STATS_INC(mp, xs_iflush_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003350 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003351 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003353 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354
Christoph Hellwig4c468192012-04-23 15:58:36 +10003355 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 xfs_iunpin_wait(ip);
3358
3359 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00003360 * For stale inodes we cannot rely on the backing buffer remaining
3361 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04003362 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00003363 * inodes below. We have to check this after ensuring the inode is
3364 * unpinned so that it is safe to reclaim the stale inode after the
3365 * flush call.
3366 */
3367 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3368 xfs_ifunlock(ip);
3369 return 0;
3370 }
3371
3372 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 * This may have been unpinned because the filesystem is shutting
3374 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003375 * to disk, because the log record didn't make it to disk.
3376 *
3377 * We also have to remove the log item from the AIL in this case,
3378 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 */
3380 if (XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10003381 error = -EIO;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003382 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 }
3384
3385 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003386 * Get the buffer containing the on-disk inode.
3387 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003388 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3389 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003390 if (error || !bp) {
3391 xfs_ifunlock(ip);
3392 return error;
3393 }
3394
3395 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 * First flush out the inode that xfs_iflush was called with.
3397 */
3398 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003399 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401
3402 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003403 * If the buffer is pinned then push on the log now so we won't
3404 * get stuck waiting in the write for too long.
3405 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00003406 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003407 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003408
3409 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 * inode clustering:
3411 * see if other inodes can be gathered into this write
3412 */
David Chinnerbad55842008-03-06 13:43:49 +11003413 error = xfs_iflush_cluster(ip, bp);
3414 if (error)
3415 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416
Christoph Hellwig4c468192012-04-23 15:58:36 +10003417 *bpp = bp;
3418 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
3420corrupt_out:
3421 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003422 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423cluster_corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003424 error = -EFSCORRUPTED;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003425abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 /*
3427 * Unlocks the flush lock
3428 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003429 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003430 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431}
3432
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433STATIC int
3434xfs_iflush_int(
Christoph Hellwig93848a92013-04-03 16:11:17 +11003435 struct xfs_inode *ip,
3436 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437{
Christoph Hellwig93848a92013-04-03 16:11:17 +11003438 struct xfs_inode_log_item *iip = ip->i_itemp;
3439 struct xfs_dinode *dip;
3440 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003442 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003443 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003445 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Christoph Hellwig93848a92013-04-03 16:11:17 +11003446 ASSERT(iip != NULL && iip->ili_fields != 0);
Dave Chinner263997a2014-05-20 07:46:40 +10003447 ASSERT(ip->i_d.di_version > 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 /* set *dip = inode's place in the buffer */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10003450 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003452 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003454 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3455 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3456 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 goto corrupt_out;
3458 }
Al Viroabbede12011-07-26 02:31:30 -04003459 if (S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 if (XFS_TEST_ERROR(
3461 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3462 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3463 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003464 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3465 "%s: Bad regular inode %Lu, ptr 0x%p",
3466 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 goto corrupt_out;
3468 }
Al Viroabbede12011-07-26 02:31:30 -04003469 } else if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470 if (XFS_TEST_ERROR(
3471 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3472 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3473 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3474 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003475 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3476 "%s: Bad directory inode %Lu, ptr 0x%p",
3477 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 goto corrupt_out;
3479 }
3480 }
3481 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3482 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3483 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003484 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3485 "%s: detected corrupt incore inode %Lu, "
3486 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3487 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11003489 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 goto corrupt_out;
3491 }
3492 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3493 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003494 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3495 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3496 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 goto corrupt_out;
3498 }
Dave Chinnere60896d2013-07-24 15:47:30 +10003499
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 /*
Dave Chinner263997a2014-05-20 07:46:40 +10003501 * Inode item log recovery for v2 inodes are dependent on the
Dave Chinnere60896d2013-07-24 15:47:30 +10003502 * di_flushiter count for correct sequencing. We bump the flush
3503 * iteration count so we can detect flushes which postdate a log record
3504 * during recovery. This is redundant as we now log every change and
3505 * hence this can't happen but we need to still do it to ensure
3506 * backwards compatibility with old kernels that predate logging all
3507 * inode changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 */
Dave Chinnere60896d2013-07-24 15:47:30 +10003509 if (ip->i_d.di_version < 3)
3510 ip->i_d.di_flushiter++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511
3512 /*
Dave Chinner39878482016-02-09 16:54:58 +11003513 * Copy the dirty parts of the inode into the on-disk inode. We always
3514 * copy out the core of the inode, because if the inode is dirty at all
3515 * the core must be.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516 */
Dave Chinner93f958f2016-02-09 16:54:58 +11003517 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
3519 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3520 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3521 ip->i_d.di_flushiter = 0;
3522
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003523 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10003524 if (XFS_IFORK_Q(ip))
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003525 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 xfs_inobp_check(mp, bp);
3527
3528 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003529 * We've recorded everything logged in the inode, so we'd like to clear
3530 * the ili_fields bits so we don't log and flush things unnecessarily.
3531 * However, we can't stop logging all this information until the data
3532 * we've copied into the disk buffer is written to disk. If we did we
3533 * might overwrite the copy of the inode in the log with all the data
3534 * after re-logging only part of it, and in the face of a crash we
3535 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003537 * What we do is move the bits to the ili_last_fields field. When
3538 * logging the inode, these bits are moved back to the ili_fields field.
3539 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3540 * know that the information those bits represent is permanently on
3541 * disk. As long as the flush completes before the inode is logged
3542 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003544 * We can play with the ili_fields bits here, because the inode lock
3545 * must be held exclusively in order to set bits there and the flush
3546 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3547 * done routine can tell whether or not to look in the AIL. Also, store
3548 * the current LSN of the inode so that we can tell whether the item has
3549 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3550 * need the AIL lock, because it is a 64 bit value that cannot be read
3551 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003553 iip->ili_last_fields = iip->ili_fields;
3554 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11003555 iip->ili_fsync_fields = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +11003556 iip->ili_logged = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557
Christoph Hellwig93848a92013-04-03 16:11:17 +11003558 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3559 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Christoph Hellwig93848a92013-04-03 16:11:17 +11003561 /*
3562 * Attach the function xfs_iflush_done to the inode's
3563 * buffer. This will remove the inode from the AIL
3564 * and unlock the inode's flush lock when the inode is
3565 * completely written to disk.
3566 */
3567 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568
Christoph Hellwig93848a92013-04-03 16:11:17 +11003569 /* generate the checksum. */
3570 xfs_dinode_calc_crc(mp, dip);
3571
3572 ASSERT(bp->b_fspriv != NULL);
3573 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 return 0;
3575
3576corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003577 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578}