Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
40ebd81d RD |
18 | #include <linux/log2.h> |
19 | ||
1da177e4 | 20 | #include "xfs.h" |
a844f451 | 21 | #include "xfs_fs.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef DC |
23 | #include "xfs_format.h" |
24 | #include "xfs_log_format.h" | |
25 | #include "xfs_trans_resv.h" | |
1da177e4 | 26 | #include "xfs_sb.h" |
1da177e4 | 27 | #include "xfs_mount.h" |
3ab78df2 | 28 | #include "xfs_defer.h" |
a4fbe6ab | 29 | #include "xfs_inode.h" |
57062787 | 30 | #include "xfs_da_format.h" |
c24b5dfa | 31 | #include "xfs_da_btree.h" |
c24b5dfa | 32 | #include "xfs_dir2.h" |
a844f451 | 33 | #include "xfs_attr_sf.h" |
c24b5dfa | 34 | #include "xfs_attr.h" |
239880ef DC |
35 | #include "xfs_trans_space.h" |
36 | #include "xfs_trans.h" | |
1da177e4 | 37 | #include "xfs_buf_item.h" |
a844f451 | 38 | #include "xfs_inode_item.h" |
a844f451 NS |
39 | #include "xfs_ialloc.h" |
40 | #include "xfs_bmap.h" | |
68988114 | 41 | #include "xfs_bmap_util.h" |
e9e899a2 | 42 | #include "xfs_errortag.h" |
1da177e4 | 43 | #include "xfs_error.h" |
1da177e4 | 44 | #include "xfs_quota.h" |
2a82b8be | 45 | #include "xfs_filestream.h" |
93848a99 | 46 | #include "xfs_cksum.h" |
0b1b213f | 47 | #include "xfs_trace.h" |
33479e05 | 48 | #include "xfs_icache.h" |
c24b5dfa | 49 | #include "xfs_symlink.h" |
239880ef DC |
50 | #include "xfs_trans_priv.h" |
51 | #include "xfs_log.h" | |
a4fbe6ab | 52 | #include "xfs_bmap_btree.h" |
aa8968f2 | 53 | #include "xfs_reflink.h" |
005c5db8 | 54 | #include "xfs_dir2_priv.h" |
1da177e4 | 55 | |
1da177e4 | 56 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
57 | |
58 | /* | |
8f04c47a | 59 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
60 | * freed from a file in a single transaction. |
61 | */ | |
62 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
63 | ||
54d7b5c1 DC |
64 | STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); |
65 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); | |
66 | STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); | |
ab297431 | 67 | |
2a0ec1d9 DC |
68 | /* |
69 | * helper function to extract extent size hint from inode | |
70 | */ | |
71 | xfs_extlen_t | |
72 | xfs_get_extsz_hint( | |
73 | struct xfs_inode *ip) | |
74 | { | |
75 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
76 | return ip->i_d.di_extsize; | |
77 | if (XFS_IS_REALTIME_INODE(ip)) | |
78 | return ip->i_mount->m_sb.sb_rextsize; | |
79 | return 0; | |
80 | } | |
81 | ||
f7ca3522 DW |
82 | /* |
83 | * Helper function to extract CoW extent size hint from inode. | |
84 | * Between the extent size hint and the CoW extent size hint, we | |
e153aa79 DW |
85 | * return the greater of the two. If the value is zero (automatic), |
86 | * use the default size. | |
f7ca3522 DW |
87 | */ |
88 | xfs_extlen_t | |
89 | xfs_get_cowextsz_hint( | |
90 | struct xfs_inode *ip) | |
91 | { | |
92 | xfs_extlen_t a, b; | |
93 | ||
94 | a = 0; | |
95 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) | |
96 | a = ip->i_d.di_cowextsize; | |
97 | b = xfs_get_extsz_hint(ip); | |
98 | ||
e153aa79 DW |
99 | a = max(a, b); |
100 | if (a == 0) | |
101 | return XFS_DEFAULT_COWEXTSZ_HINT; | |
102 | return a; | |
f7ca3522 DW |
103 | } |
104 | ||
fa96acad | 105 | /* |
efa70be1 CH |
106 | * These two are wrapper routines around the xfs_ilock() routine used to |
107 | * centralize some grungy code. They are used in places that wish to lock the | |
108 | * inode solely for reading the extents. The reason these places can't just | |
109 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
110 | * bringing in of the extents from disk for a file in b-tree format. If the | |
111 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
112 | * the extents are read in. Locking it exclusively all the time would limit | |
113 | * our parallelism unnecessarily, though. What we do instead is check to see | |
114 | * if the extents have been read in yet, and only lock the inode exclusively | |
115 | * if they have not. | |
fa96acad | 116 | * |
efa70be1 | 117 | * The functions return a value which should be given to the corresponding |
01f4f327 | 118 | * xfs_iunlock() call. |
fa96acad DC |
119 | */ |
120 | uint | |
309ecac8 CH |
121 | xfs_ilock_data_map_shared( |
122 | struct xfs_inode *ip) | |
fa96acad | 123 | { |
309ecac8 | 124 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 125 | |
309ecac8 CH |
126 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && |
127 | (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) | |
fa96acad | 128 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 129 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
130 | return lock_mode; |
131 | } | |
132 | ||
efa70be1 CH |
133 | uint |
134 | xfs_ilock_attr_map_shared( | |
135 | struct xfs_inode *ip) | |
fa96acad | 136 | { |
efa70be1 CH |
137 | uint lock_mode = XFS_ILOCK_SHARED; |
138 | ||
139 | if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && | |
140 | (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) | |
141 | lock_mode = XFS_ILOCK_EXCL; | |
142 | xfs_ilock(ip, lock_mode); | |
143 | return lock_mode; | |
fa96acad DC |
144 | } |
145 | ||
146 | /* | |
65523218 CH |
147 | * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 |
148 | * multi-reader locks: i_mmap_lock and the i_lock. This routine allows | |
149 | * various combinations of the locks to be obtained. | |
fa96acad | 150 | * |
653c60b6 DC |
151 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
152 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 153 | * |
653c60b6 DC |
154 | * Basic locking order: |
155 | * | |
65523218 | 156 | * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock |
653c60b6 DC |
157 | * |
158 | * mmap_sem locking order: | |
159 | * | |
65523218 | 160 | * i_rwsem -> page lock -> mmap_sem |
653c60b6 DC |
161 | * mmap_sem -> i_mmap_lock -> page_lock |
162 | * | |
163 | * The difference in mmap_sem locking order mean that we cannot hold the | |
164 | * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can | |
165 | * fault in pages during copy in/out (for buffered IO) or require the mmap_sem | |
166 | * in get_user_pages() to map the user pages into the kernel address space for | |
65523218 | 167 | * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because |
653c60b6 DC |
168 | * page faults already hold the mmap_sem. |
169 | * | |
170 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
65523218 | 171 | * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both |
653c60b6 DC |
172 | * taken in places where we need to invalidate the page cache in a race |
173 | * free manner (e.g. truncate, hole punch and other extent manipulation | |
174 | * functions). | |
fa96acad DC |
175 | */ |
176 | void | |
177 | xfs_ilock( | |
178 | xfs_inode_t *ip, | |
179 | uint lock_flags) | |
180 | { | |
181 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
182 | ||
183 | /* | |
184 | * You can't set both SHARED and EXCL for the same lock, | |
185 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
186 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
187 | */ | |
188 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
189 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
190 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
191 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
192 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
193 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 194 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad | 195 | |
65523218 CH |
196 | if (lock_flags & XFS_IOLOCK_EXCL) { |
197 | down_write_nested(&VFS_I(ip)->i_rwsem, | |
198 | XFS_IOLOCK_DEP(lock_flags)); | |
199 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
200 | down_read_nested(&VFS_I(ip)->i_rwsem, | |
201 | XFS_IOLOCK_DEP(lock_flags)); | |
202 | } | |
fa96acad | 203 | |
653c60b6 DC |
204 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
205 | mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
206 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
207 | mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
208 | ||
fa96acad DC |
209 | if (lock_flags & XFS_ILOCK_EXCL) |
210 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
211 | else if (lock_flags & XFS_ILOCK_SHARED) | |
212 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
213 | } | |
214 | ||
215 | /* | |
216 | * This is just like xfs_ilock(), except that the caller | |
217 | * is guaranteed not to sleep. It returns 1 if it gets | |
218 | * the requested locks and 0 otherwise. If the IO lock is | |
219 | * obtained but the inode lock cannot be, then the IO lock | |
220 | * is dropped before returning. | |
221 | * | |
222 | * ip -- the inode being locked | |
223 | * lock_flags -- this parameter indicates the inode's locks to be | |
224 | * to be locked. See the comment for xfs_ilock() for a list | |
225 | * of valid values. | |
226 | */ | |
227 | int | |
228 | xfs_ilock_nowait( | |
229 | xfs_inode_t *ip, | |
230 | uint lock_flags) | |
231 | { | |
232 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
233 | ||
234 | /* | |
235 | * You can't set both SHARED and EXCL for the same lock, | |
236 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
237 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
238 | */ | |
239 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
240 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
241 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
242 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
243 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
244 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 245 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
246 | |
247 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
65523218 | 248 | if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
249 | goto out; |
250 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
65523218 | 251 | if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
252 | goto out; |
253 | } | |
653c60b6 DC |
254 | |
255 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
256 | if (!mrtryupdate(&ip->i_mmaplock)) | |
257 | goto out_undo_iolock; | |
258 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
259 | if (!mrtryaccess(&ip->i_mmaplock)) | |
260 | goto out_undo_iolock; | |
261 | } | |
262 | ||
fa96acad DC |
263 | if (lock_flags & XFS_ILOCK_EXCL) { |
264 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 265 | goto out_undo_mmaplock; |
fa96acad DC |
266 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
267 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 268 | goto out_undo_mmaplock; |
fa96acad DC |
269 | } |
270 | return 1; | |
271 | ||
653c60b6 DC |
272 | out_undo_mmaplock: |
273 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
274 | mrunlock_excl(&ip->i_mmaplock); | |
275 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
276 | mrunlock_shared(&ip->i_mmaplock); | |
277 | out_undo_iolock: | |
fa96acad | 278 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 279 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 280 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 281 | up_read(&VFS_I(ip)->i_rwsem); |
653c60b6 | 282 | out: |
fa96acad DC |
283 | return 0; |
284 | } | |
285 | ||
286 | /* | |
287 | * xfs_iunlock() is used to drop the inode locks acquired with | |
288 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
289 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
290 | * that we know which locks to drop. | |
291 | * | |
292 | * ip -- the inode being unlocked | |
293 | * lock_flags -- this parameter indicates the inode's locks to be | |
294 | * to be unlocked. See the comment for xfs_ilock() for a list | |
295 | * of valid values for this parameter. | |
296 | * | |
297 | */ | |
298 | void | |
299 | xfs_iunlock( | |
300 | xfs_inode_t *ip, | |
301 | uint lock_flags) | |
302 | { | |
303 | /* | |
304 | * You can't set both SHARED and EXCL for the same lock, | |
305 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
306 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
307 | */ | |
308 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
309 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
310 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
311 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
312 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
313 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 314 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
315 | ASSERT(lock_flags != 0); |
316 | ||
317 | if (lock_flags & XFS_IOLOCK_EXCL) | |
65523218 | 318 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 319 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 320 | up_read(&VFS_I(ip)->i_rwsem); |
fa96acad | 321 | |
653c60b6 DC |
322 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
323 | mrunlock_excl(&ip->i_mmaplock); | |
324 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
325 | mrunlock_shared(&ip->i_mmaplock); | |
326 | ||
fa96acad DC |
327 | if (lock_flags & XFS_ILOCK_EXCL) |
328 | mrunlock_excl(&ip->i_lock); | |
329 | else if (lock_flags & XFS_ILOCK_SHARED) | |
330 | mrunlock_shared(&ip->i_lock); | |
331 | ||
332 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
333 | } | |
334 | ||
335 | /* | |
336 | * give up write locks. the i/o lock cannot be held nested | |
337 | * if it is being demoted. | |
338 | */ | |
339 | void | |
340 | xfs_ilock_demote( | |
341 | xfs_inode_t *ip, | |
342 | uint lock_flags) | |
343 | { | |
653c60b6 DC |
344 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
345 | ASSERT((lock_flags & | |
346 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
347 | |
348 | if (lock_flags & XFS_ILOCK_EXCL) | |
349 | mrdemote(&ip->i_lock); | |
653c60b6 DC |
350 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
351 | mrdemote(&ip->i_mmaplock); | |
fa96acad | 352 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 353 | downgrade_write(&VFS_I(ip)->i_rwsem); |
fa96acad DC |
354 | |
355 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
356 | } | |
357 | ||
742ae1e3 | 358 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
359 | int |
360 | xfs_isilocked( | |
361 | xfs_inode_t *ip, | |
362 | uint lock_flags) | |
363 | { | |
364 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
365 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
366 | return !!ip->i_lock.mr_writer; | |
367 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
368 | } | |
369 | ||
653c60b6 DC |
370 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
371 | if (!(lock_flags & XFS_MMAPLOCK_SHARED)) | |
372 | return !!ip->i_mmaplock.mr_writer; | |
373 | return rwsem_is_locked(&ip->i_mmaplock.mr_lock); | |
374 | } | |
375 | ||
fa96acad DC |
376 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
377 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
65523218 CH |
378 | return !debug_locks || |
379 | lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); | |
380 | return rwsem_is_locked(&VFS_I(ip)->i_rwsem); | |
fa96acad DC |
381 | } |
382 | ||
383 | ASSERT(0); | |
384 | return 0; | |
385 | } | |
386 | #endif | |
387 | ||
b6a9947e DC |
388 | /* |
389 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
390 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
391 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
392 | * errors and warnings. | |
393 | */ | |
394 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
395 | static bool |
396 | xfs_lockdep_subclass_ok( | |
397 | int subclass) | |
398 | { | |
399 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
400 | } | |
401 | #else | |
402 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
403 | #endif | |
404 | ||
c24b5dfa | 405 | /* |
653c60b6 | 406 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
407 | * value. This can be called for any type of inode lock combination, including |
408 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
409 | * storage fields in the class mask we build. | |
c24b5dfa DC |
410 | */ |
411 | static inline int | |
412 | xfs_lock_inumorder(int lock_mode, int subclass) | |
413 | { | |
0952c818 DC |
414 | int class = 0; |
415 | ||
416 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
417 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 418 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 419 | |
653c60b6 | 420 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 421 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
0952c818 | 422 | class += subclass << XFS_IOLOCK_SHIFT; |
653c60b6 DC |
423 | } |
424 | ||
425 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
426 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
427 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
428 | } |
429 | ||
0952c818 DC |
430 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
431 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
432 | class += subclass << XFS_ILOCK_SHIFT; | |
433 | } | |
c24b5dfa | 434 | |
0952c818 | 435 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
436 | } |
437 | ||
438 | /* | |
95afcf5c DC |
439 | * The following routine will lock n inodes in exclusive mode. We assume the |
440 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 441 | * |
95afcf5c DC |
442 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
443 | * start waiting for another inode that is locked by a thread in a long running | |
444 | * transaction (such as truncate). This can result in deadlock since the long | |
445 | * running trans might need to wait for the inode we just locked in order to | |
446 | * push the tail and free space in the log. | |
0952c818 DC |
447 | * |
448 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
449 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
450 | * lock more than one at a time, lockdep will report false positives saying we | |
451 | * have violated locking orders. | |
c24b5dfa | 452 | */ |
0d5a75e9 | 453 | static void |
c24b5dfa DC |
454 | xfs_lock_inodes( |
455 | xfs_inode_t **ips, | |
456 | int inodes, | |
457 | uint lock_mode) | |
458 | { | |
459 | int attempts = 0, i, j, try_lock; | |
460 | xfs_log_item_t *lp; | |
461 | ||
0952c818 DC |
462 | /* |
463 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
464 | * support an arbitrary depth of locking here, but absolute limits on | |
465 | * inodes depend on the the type of locking and the limits placed by | |
466 | * lockdep annotations in xfs_lock_inumorder. These are all checked by | |
467 | * the asserts. | |
468 | */ | |
95afcf5c | 469 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
470 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
471 | XFS_ILOCK_EXCL)); | |
472 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
473 | XFS_ILOCK_SHARED))); | |
0952c818 DC |
474 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || |
475 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
476 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
477 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
478 | ||
479 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
480 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
481 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
482 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa DC |
483 | |
484 | try_lock = 0; | |
485 | i = 0; | |
c24b5dfa DC |
486 | again: |
487 | for (; i < inodes; i++) { | |
488 | ASSERT(ips[i]); | |
489 | ||
95afcf5c | 490 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
491 | continue; |
492 | ||
493 | /* | |
95afcf5c DC |
494 | * If try_lock is not set yet, make sure all locked inodes are |
495 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 496 | */ |
c24b5dfa DC |
497 | if (!try_lock) { |
498 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
499 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
95afcf5c | 500 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) |
c24b5dfa | 501 | try_lock++; |
c24b5dfa DC |
502 | } |
503 | } | |
504 | ||
505 | /* | |
506 | * If any of the previous locks we have locked is in the AIL, | |
507 | * we must TRY to get the second and subsequent locks. If | |
508 | * we can't get any, we must release all we have | |
509 | * and try again. | |
510 | */ | |
95afcf5c DC |
511 | if (!try_lock) { |
512 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
513 | continue; | |
514 | } | |
515 | ||
516 | /* try_lock means we have an inode locked that is in the AIL. */ | |
517 | ASSERT(i != 0); | |
518 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
519 | continue; | |
c24b5dfa | 520 | |
95afcf5c DC |
521 | /* |
522 | * Unlock all previous guys and try again. xfs_iunlock will try | |
523 | * to push the tail if the inode is in the AIL. | |
524 | */ | |
525 | attempts++; | |
526 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 527 | /* |
95afcf5c DC |
528 | * Check to see if we've already unlocked this one. Not |
529 | * the first one going back, and the inode ptr is the | |
530 | * same. | |
c24b5dfa | 531 | */ |
95afcf5c DC |
532 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
533 | continue; | |
c24b5dfa | 534 | |
95afcf5c DC |
535 | xfs_iunlock(ips[j], lock_mode); |
536 | } | |
c24b5dfa | 537 | |
95afcf5c DC |
538 | if ((attempts % 5) == 0) { |
539 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 540 | } |
95afcf5c DC |
541 | i = 0; |
542 | try_lock = 0; | |
543 | goto again; | |
c24b5dfa | 544 | } |
c24b5dfa DC |
545 | } |
546 | ||
547 | /* | |
653c60b6 DC |
548 | * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - |
549 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
550 | * lock more than one at a time, lockdep will report false positives saying we | |
551 | * have violated locking orders. | |
c24b5dfa DC |
552 | */ |
553 | void | |
554 | xfs_lock_two_inodes( | |
555 | xfs_inode_t *ip0, | |
556 | xfs_inode_t *ip1, | |
557 | uint lock_mode) | |
558 | { | |
559 | xfs_inode_t *temp; | |
560 | int attempts = 0; | |
561 | xfs_log_item_t *lp; | |
562 | ||
65523218 CH |
563 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); |
564 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) | |
653c60b6 DC |
565 | ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); |
566 | ||
c24b5dfa DC |
567 | ASSERT(ip0->i_ino != ip1->i_ino); |
568 | ||
569 | if (ip0->i_ino > ip1->i_ino) { | |
570 | temp = ip0; | |
571 | ip0 = ip1; | |
572 | ip1 = temp; | |
573 | } | |
574 | ||
575 | again: | |
576 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
577 | ||
578 | /* | |
579 | * If the first lock we have locked is in the AIL, we must TRY to get | |
580 | * the second lock. If we can't get it, we must release the first one | |
581 | * and try again. | |
582 | */ | |
583 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
584 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
585 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
586 | xfs_iunlock(ip0, lock_mode); | |
587 | if ((++attempts % 5) == 0) | |
588 | delay(1); /* Don't just spin the CPU */ | |
589 | goto again; | |
590 | } | |
591 | } else { | |
592 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
593 | } | |
594 | } | |
595 | ||
596 | ||
fa96acad DC |
597 | void |
598 | __xfs_iflock( | |
599 | struct xfs_inode *ip) | |
600 | { | |
601 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
602 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
603 | ||
604 | do { | |
21417136 | 605 | prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
fa96acad DC |
606 | if (xfs_isiflocked(ip)) |
607 | io_schedule(); | |
608 | } while (!xfs_iflock_nowait(ip)); | |
609 | ||
21417136 | 610 | finish_wait(wq, &wait.wq_entry); |
fa96acad DC |
611 | } |
612 | ||
1da177e4 LT |
613 | STATIC uint |
614 | _xfs_dic2xflags( | |
c8ce540d | 615 | uint16_t di_flags, |
58f88ca2 DC |
616 | uint64_t di_flags2, |
617 | bool has_attr) | |
1da177e4 LT |
618 | { |
619 | uint flags = 0; | |
620 | ||
621 | if (di_flags & XFS_DIFLAG_ANY) { | |
622 | if (di_flags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 623 | flags |= FS_XFLAG_REALTIME; |
1da177e4 | 624 | if (di_flags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 625 | flags |= FS_XFLAG_PREALLOC; |
1da177e4 | 626 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 627 | flags |= FS_XFLAG_IMMUTABLE; |
1da177e4 | 628 | if (di_flags & XFS_DIFLAG_APPEND) |
e7b89481 | 629 | flags |= FS_XFLAG_APPEND; |
1da177e4 | 630 | if (di_flags & XFS_DIFLAG_SYNC) |
e7b89481 | 631 | flags |= FS_XFLAG_SYNC; |
1da177e4 | 632 | if (di_flags & XFS_DIFLAG_NOATIME) |
e7b89481 | 633 | flags |= FS_XFLAG_NOATIME; |
1da177e4 | 634 | if (di_flags & XFS_DIFLAG_NODUMP) |
e7b89481 | 635 | flags |= FS_XFLAG_NODUMP; |
1da177e4 | 636 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 637 | flags |= FS_XFLAG_RTINHERIT; |
1da177e4 | 638 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 639 | flags |= FS_XFLAG_PROJINHERIT; |
1da177e4 | 640 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 641 | flags |= FS_XFLAG_NOSYMLINKS; |
dd9f438e | 642 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 643 | flags |= FS_XFLAG_EXTSIZE; |
dd9f438e | 644 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 645 | flags |= FS_XFLAG_EXTSZINHERIT; |
d3446eac | 646 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 647 | flags |= FS_XFLAG_NODEFRAG; |
2a82b8be | 648 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 649 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
650 | } |
651 | ||
58f88ca2 DC |
652 | if (di_flags2 & XFS_DIFLAG2_ANY) { |
653 | if (di_flags2 & XFS_DIFLAG2_DAX) | |
654 | flags |= FS_XFLAG_DAX; | |
f7ca3522 DW |
655 | if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) |
656 | flags |= FS_XFLAG_COWEXTSIZE; | |
58f88ca2 DC |
657 | } |
658 | ||
659 | if (has_attr) | |
660 | flags |= FS_XFLAG_HASATTR; | |
661 | ||
1da177e4 LT |
662 | return flags; |
663 | } | |
664 | ||
665 | uint | |
666 | xfs_ip2xflags( | |
58f88ca2 | 667 | struct xfs_inode *ip) |
1da177e4 | 668 | { |
58f88ca2 | 669 | struct xfs_icdinode *dic = &ip->i_d; |
1da177e4 | 670 | |
58f88ca2 | 671 | return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); |
1da177e4 LT |
672 | } |
673 | ||
c24b5dfa DC |
674 | /* |
675 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
676 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
677 | * ci_name->name will point to a the actual name (caller must free) or | |
678 | * will be set to NULL if an exact match is found. | |
679 | */ | |
680 | int | |
681 | xfs_lookup( | |
682 | xfs_inode_t *dp, | |
683 | struct xfs_name *name, | |
684 | xfs_inode_t **ipp, | |
685 | struct xfs_name *ci_name) | |
686 | { | |
687 | xfs_ino_t inum; | |
688 | int error; | |
c24b5dfa DC |
689 | |
690 | trace_xfs_lookup(dp, name); | |
691 | ||
692 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
2451337d | 693 | return -EIO; |
c24b5dfa | 694 | |
c24b5dfa | 695 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 696 | if (error) |
dbad7c99 | 697 | goto out_unlock; |
c24b5dfa DC |
698 | |
699 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
700 | if (error) | |
701 | goto out_free_name; | |
702 | ||
703 | return 0; | |
704 | ||
705 | out_free_name: | |
706 | if (ci_name) | |
707 | kmem_free(ci_name->name); | |
dbad7c99 | 708 | out_unlock: |
c24b5dfa DC |
709 | *ipp = NULL; |
710 | return error; | |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * Allocate an inode on disk and return a copy of its in-core version. | |
715 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
716 | * appropriately within the inode. The uid and gid for the inode are | |
717 | * set according to the contents of the given cred structure. | |
718 | * | |
719 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
720 | * has a free inode available, call xfs_iget() to obtain the in-core |
721 | * version of the allocated inode. Finally, fill in the inode and | |
722 | * log its initial contents. In this case, ialloc_context would be | |
723 | * set to NULL. | |
1da177e4 | 724 | * |
cd856db6 CM |
725 | * If xfs_dialloc() does not have an available inode, it will replenish |
726 | * its supply by doing an allocation. Since we can only do one | |
727 | * allocation within a transaction without deadlocks, we must commit | |
728 | * the current transaction before returning the inode itself. | |
729 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
730 | * The caller should then commit the current transaction, start a new |
731 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
732 | * | |
733 | * To ensure that some other process does not grab the inode that | |
734 | * was allocated during the first call to xfs_ialloc(), this routine | |
735 | * also returns the [locked] bp pointing to the head of the freelist | |
736 | * as ialloc_context. The caller should hold this buffer across | |
737 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
738 | * |
739 | * If we are allocating quota inodes, we do not have a parent inode | |
740 | * to attach to or associate with (i.e. pip == NULL) because they | |
741 | * are not linked into the directory structure - they are attached | |
742 | * directly to the superblock - and so have no parent. | |
1da177e4 | 743 | */ |
0d5a75e9 | 744 | static int |
1da177e4 LT |
745 | xfs_ialloc( |
746 | xfs_trans_t *tp, | |
747 | xfs_inode_t *pip, | |
576b1d67 | 748 | umode_t mode, |
31b084ae | 749 | xfs_nlink_t nlink, |
66f36464 | 750 | dev_t rdev, |
6743099c | 751 | prid_t prid, |
1da177e4 LT |
752 | int okalloc, |
753 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
754 | xfs_inode_t **ipp) |
755 | { | |
93848a99 | 756 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
757 | xfs_ino_t ino; |
758 | xfs_inode_t *ip; | |
1da177e4 LT |
759 | uint flags; |
760 | int error; | |
e076b0f3 | 761 | struct timespec tv; |
3987848c | 762 | struct inode *inode; |
1da177e4 LT |
763 | |
764 | /* | |
765 | * Call the space management code to pick | |
766 | * the on-disk inode to be allocated. | |
767 | */ | |
b11f94d5 | 768 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 769 | ialloc_context, &ino); |
bf904248 | 770 | if (error) |
1da177e4 | 771 | return error; |
08358906 | 772 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
773 | *ipp = NULL; |
774 | return 0; | |
775 | } | |
776 | ASSERT(*ialloc_context == NULL); | |
777 | ||
778 | /* | |
779 | * Get the in-core inode with the lock held exclusively. | |
780 | * This is because we're setting fields here we need | |
781 | * to prevent others from looking at until we're done. | |
782 | */ | |
93848a99 | 783 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 784 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 785 | if (error) |
1da177e4 | 786 | return error; |
1da177e4 | 787 | ASSERT(ip != NULL); |
3987848c | 788 | inode = VFS_I(ip); |
1da177e4 | 789 | |
263997a6 DC |
790 | /* |
791 | * We always convert v1 inodes to v2 now - we only support filesystems | |
792 | * with >= v2 inode capability, so there is no reason for ever leaving | |
793 | * an inode in v1 format. | |
794 | */ | |
795 | if (ip->i_d.di_version == 1) | |
796 | ip->i_d.di_version = 2; | |
797 | ||
c19b3b05 | 798 | inode->i_mode = mode; |
54d7b5c1 | 799 | set_nlink(inode, nlink); |
7aab1b28 DE |
800 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
801 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
66f36464 | 802 | inode->i_rdev = rdev; |
6743099c | 803 | xfs_set_projid(ip, prid); |
1da177e4 | 804 | |
bd186aa9 | 805 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 806 | ip->i_d.di_gid = pip->i_d.di_gid; |
c19b3b05 DC |
807 | if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) |
808 | inode->i_mode |= S_ISGID; | |
1da177e4 LT |
809 | } |
810 | ||
811 | /* | |
812 | * If the group ID of the new file does not match the effective group | |
813 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
814 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
815 | */ | |
816 | if ((irix_sgid_inherit) && | |
c19b3b05 DC |
817 | (inode->i_mode & S_ISGID) && |
818 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) | |
819 | inode->i_mode &= ~S_ISGID; | |
1da177e4 LT |
820 | |
821 | ip->i_d.di_size = 0; | |
822 | ip->i_d.di_nextents = 0; | |
823 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 | 824 | |
c2050a45 | 825 | tv = current_time(inode); |
3987848c DC |
826 | inode->i_mtime = tv; |
827 | inode->i_atime = tv; | |
828 | inode->i_ctime = tv; | |
dff35fd4 | 829 | |
1da177e4 LT |
830 | ip->i_d.di_extsize = 0; |
831 | ip->i_d.di_dmevmask = 0; | |
832 | ip->i_d.di_dmstate = 0; | |
833 | ip->i_d.di_flags = 0; | |
93848a99 CH |
834 | |
835 | if (ip->i_d.di_version == 3) { | |
83e06f21 | 836 | inode->i_version = 1; |
93848a99 | 837 | ip->i_d.di_flags2 = 0; |
f7ca3522 | 838 | ip->i_d.di_cowextsize = 0; |
c8ce540d DW |
839 | ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; |
840 | ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; | |
93848a99 CH |
841 | } |
842 | ||
843 | ||
1da177e4 LT |
844 | flags = XFS_ILOG_CORE; |
845 | switch (mode & S_IFMT) { | |
846 | case S_IFIFO: | |
847 | case S_IFCHR: | |
848 | case S_IFBLK: | |
849 | case S_IFSOCK: | |
850 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1da177e4 LT |
851 | ip->i_df.if_flags = 0; |
852 | flags |= XFS_ILOG_DEV; | |
853 | break; | |
854 | case S_IFREG: | |
855 | case S_IFDIR: | |
b11f94d5 | 856 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
58f88ca2 | 857 | uint di_flags = 0; |
365ca83d | 858 | |
abbede1b | 859 | if (S_ISDIR(mode)) { |
365ca83d NS |
860 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
861 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
862 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
863 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
864 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
865 | } | |
9336e3a7 DC |
866 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
867 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
abbede1b | 868 | } else if (S_ISREG(mode)) { |
613d7043 | 869 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 870 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
871 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
872 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
873 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
874 | } | |
1da177e4 LT |
875 | } |
876 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
877 | xfs_inherit_noatime) | |
365ca83d | 878 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
879 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
880 | xfs_inherit_nodump) | |
365ca83d | 881 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
882 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
883 | xfs_inherit_sync) | |
365ca83d | 884 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
885 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
886 | xfs_inherit_nosymlinks) | |
365ca83d | 887 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
d3446eac BN |
888 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
889 | xfs_inherit_nodefrag) | |
890 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
891 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
892 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
58f88ca2 | 893 | |
365ca83d | 894 | ip->i_d.di_flags |= di_flags; |
1da177e4 | 895 | } |
f7ca3522 DW |
896 | if (pip && |
897 | (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && | |
898 | pip->i_d.di_version == 3 && | |
899 | ip->i_d.di_version == 3) { | |
56bdf855 LC |
900 | uint64_t di_flags2 = 0; |
901 | ||
f7ca3522 | 902 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { |
56bdf855 | 903 | di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; |
f7ca3522 DW |
904 | ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; |
905 | } | |
56bdf855 LC |
906 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) |
907 | di_flags2 |= XFS_DIFLAG2_DAX; | |
908 | ||
909 | ip->i_d.di_flags2 |= di_flags2; | |
f7ca3522 | 910 | } |
1da177e4 LT |
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; | |
6bdcf26a | 916 | ip->i_df.if_u1.if_root = NULL; |
1da177e4 LT |
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 | */ | |
ddc3415a | 930 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
931 | xfs_trans_log_inode(tp, ip, flags); |
932 | ||
58c90473 | 933 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 934 | xfs_setup_inode(ip); |
1da177e4 LT |
935 | |
936 | *ipp = ip; | |
937 | return 0; | |
938 | } | |
939 | ||
e546cb79 DC |
940 | /* |
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 | */ | |
950 | int | |
951 | xfs_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, | |
66f36464 | 958 | dev_t rdev, |
e546cb79 DC |
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; | |
e546cb79 DC |
967 | xfs_inode_t *ip; |
968 | xfs_buf_t *ialloc_context = NULL; | |
969 | int code; | |
e546cb79 DC |
970 | 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; | |
2451337d | 1005 | return -ENOSPC; |
e546cb79 DC |
1006 | } |
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); | |
e546cb79 DC |
1023 | |
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 | ||
411350df | 1038 | code = xfs_trans_roll(&tp); |
2e6db6c4 | 1039 | if (committed != NULL) |
e546cb79 | 1040 | *committed = 1; |
3d3c8b52 | 1041 | |
e546cb79 DC |
1042 | /* |
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); | |
2e6db6c4 | 1052 | *tpp = tp; |
e546cb79 DC |
1053 | *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 | /* | |
54d7b5c1 DC |
1089 | * Decrement the link count on an inode & log the change. If this causes the |
1090 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
1091 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 | 1092 | */ |
0d5a75e9 | 1093 | static int /* error */ |
e546cb79 DC |
1094 | xfs_droplink( |
1095 | xfs_trans_t *tp, | |
1096 | xfs_inode_t *ip) | |
1097 | { | |
e546cb79 DC |
1098 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
1099 | ||
e546cb79 DC |
1100 | drop_nlink(VFS_I(ip)); |
1101 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1102 | ||
54d7b5c1 DC |
1103 | if (VFS_I(ip)->i_nlink) |
1104 | return 0; | |
1105 | ||
1106 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
1107 | } |
1108 | ||
e546cb79 DC |
1109 | /* |
1110 | * Increment the link count on an inode & log the change. | |
1111 | */ | |
0d5a75e9 | 1112 | static int |
e546cb79 DC |
1113 | xfs_bumplink( |
1114 | xfs_trans_t *tp, | |
1115 | xfs_inode_t *ip) | |
1116 | { | |
1117 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1118 | ||
263997a6 | 1119 | ASSERT(ip->i_d.di_version > 1); |
e546cb79 | 1120 | inc_nlink(VFS_I(ip)); |
e546cb79 DC |
1121 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1122 | return 0; | |
1123 | } | |
1124 | ||
c24b5dfa DC |
1125 | int |
1126 | xfs_create( | |
1127 | xfs_inode_t *dp, | |
1128 | struct xfs_name *name, | |
1129 | umode_t mode, | |
66f36464 | 1130 | dev_t rdev, |
c24b5dfa DC |
1131 | xfs_inode_t **ipp) |
1132 | { | |
1133 | int is_dir = S_ISDIR(mode); | |
1134 | struct xfs_mount *mp = dp->i_mount; | |
1135 | struct xfs_inode *ip = NULL; | |
1136 | struct xfs_trans *tp = NULL; | |
1137 | int error; | |
2c3234d1 | 1138 | struct xfs_defer_ops dfops; |
c24b5dfa DC |
1139 | xfs_fsblock_t first_block; |
1140 | bool unlock_dp_on_error = false; | |
c24b5dfa DC |
1141 | prid_t prid; |
1142 | struct xfs_dquot *udqp = NULL; | |
1143 | struct xfs_dquot *gdqp = NULL; | |
1144 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 1145 | struct xfs_trans_res *tres; |
c24b5dfa | 1146 | uint resblks; |
c24b5dfa DC |
1147 | |
1148 | trace_xfs_create(dp, name); | |
1149 | ||
1150 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1151 | return -EIO; |
c24b5dfa | 1152 | |
163467d3 | 1153 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
1154 | |
1155 | /* | |
1156 | * Make sure that we have allocated dquot(s) on disk. | |
1157 | */ | |
7aab1b28 DE |
1158 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1159 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1160 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1161 | &udqp, &gdqp, &pdqp); | |
1162 | if (error) | |
1163 | return error; | |
1164 | ||
1165 | if (is_dir) { | |
c24b5dfa | 1166 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
062647a8 | 1167 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
1168 | } else { |
1169 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 1170 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
1171 | } |
1172 | ||
c24b5dfa DC |
1173 | /* |
1174 | * Initially assume that the file does not exist and | |
1175 | * reserve the resources for that case. If that is not | |
1176 | * the case we'll drop the one we have and get a more | |
1177 | * appropriate transaction later. | |
1178 | */ | |
253f4911 | 1179 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
2451337d | 1180 | if (error == -ENOSPC) { |
c24b5dfa DC |
1181 | /* flush outstanding delalloc blocks and retry */ |
1182 | xfs_flush_inodes(mp); | |
253f4911 | 1183 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
c24b5dfa | 1184 | } |
2451337d | 1185 | if (error == -ENOSPC) { |
c24b5dfa DC |
1186 | /* No space at all so try a "no-allocation" reservation */ |
1187 | resblks = 0; | |
253f4911 | 1188 | error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); |
c24b5dfa | 1189 | } |
4906e215 | 1190 | if (error) |
253f4911 | 1191 | goto out_release_inode; |
c24b5dfa | 1192 | |
65523218 | 1193 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
c24b5dfa DC |
1194 | unlock_dp_on_error = true; |
1195 | ||
2c3234d1 | 1196 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa DC |
1197 | |
1198 | /* | |
1199 | * Reserve disk quota and the inode. | |
1200 | */ | |
1201 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1202 | pdqp, resblks, 1, 0); | |
1203 | if (error) | |
1204 | goto out_trans_cancel; | |
1205 | ||
94f3cad5 ES |
1206 | if (!resblks) { |
1207 | error = xfs_dir_canenter(tp, dp, name); | |
1208 | if (error) | |
1209 | goto out_trans_cancel; | |
1210 | } | |
c24b5dfa DC |
1211 | |
1212 | /* | |
1213 | * A newly created regular or special file just has one directory | |
1214 | * entry pointing to them, but a directory also the "." entry | |
1215 | * pointing to itself. | |
1216 | */ | |
1217 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, | |
f6106efa | 1218 | prid, resblks > 0, &ip, NULL); |
d6077aa3 | 1219 | if (error) |
4906e215 | 1220 | goto out_trans_cancel; |
c24b5dfa DC |
1221 | |
1222 | /* | |
1223 | * Now we join the directory inode to the transaction. We do not do it | |
1224 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1225 | * (and release all the locks). An error from here on will result in | |
1226 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1227 | * error path. | |
1228 | */ | |
65523218 | 1229 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1230 | unlock_dp_on_error = false; |
1231 | ||
1232 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
2c3234d1 | 1233 | &first_block, &dfops, resblks ? |
c24b5dfa DC |
1234 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); |
1235 | if (error) { | |
2451337d | 1236 | ASSERT(error != -ENOSPC); |
4906e215 | 1237 | goto out_trans_cancel; |
c24b5dfa DC |
1238 | } |
1239 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1240 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1241 | ||
1242 | if (is_dir) { | |
1243 | error = xfs_dir_init(tp, ip, dp); | |
1244 | if (error) | |
1245 | goto out_bmap_cancel; | |
1246 | ||
1247 | error = xfs_bumplink(tp, dp); | |
1248 | if (error) | |
1249 | goto out_bmap_cancel; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * If this is a synchronous mount, make sure that the | |
1254 | * create transaction goes to disk before returning to | |
1255 | * the user. | |
1256 | */ | |
1257 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1258 | xfs_trans_set_sync(tp); | |
1259 | ||
1260 | /* | |
1261 | * Attach the dquot(s) to the inodes and modify them incore. | |
1262 | * These ids of the inode couldn't have changed since the new | |
1263 | * inode has been locked ever since it was created. | |
1264 | */ | |
1265 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1266 | ||
8ad7c629 | 1267 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
1268 | if (error) |
1269 | goto out_bmap_cancel; | |
1270 | ||
70393313 | 1271 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1272 | if (error) |
1273 | goto out_release_inode; | |
1274 | ||
1275 | xfs_qm_dqrele(udqp); | |
1276 | xfs_qm_dqrele(gdqp); | |
1277 | xfs_qm_dqrele(pdqp); | |
1278 | ||
1279 | *ipp = ip; | |
1280 | return 0; | |
1281 | ||
1282 | out_bmap_cancel: | |
2c3234d1 | 1283 | xfs_defer_cancel(&dfops); |
c24b5dfa | 1284 | out_trans_cancel: |
4906e215 | 1285 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1286 | out_release_inode: |
1287 | /* | |
58c90473 DC |
1288 | * Wait until after the current transaction is aborted to finish the |
1289 | * setup of the inode and release the inode. This prevents recursive | |
1290 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1291 | */ |
58c90473 DC |
1292 | if (ip) { |
1293 | xfs_finish_inode_setup(ip); | |
c24b5dfa | 1294 | IRELE(ip); |
58c90473 | 1295 | } |
c24b5dfa DC |
1296 | |
1297 | xfs_qm_dqrele(udqp); | |
1298 | xfs_qm_dqrele(gdqp); | |
1299 | xfs_qm_dqrele(pdqp); | |
1300 | ||
1301 | if (unlock_dp_on_error) | |
65523218 | 1302 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1303 | return error; |
1304 | } | |
1305 | ||
99b6436b ZYW |
1306 | int |
1307 | xfs_create_tmpfile( | |
1308 | struct xfs_inode *dp, | |
1309 | struct dentry *dentry, | |
330033d6 BF |
1310 | umode_t mode, |
1311 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1312 | { |
1313 | struct xfs_mount *mp = dp->i_mount; | |
1314 | struct xfs_inode *ip = NULL; | |
1315 | struct xfs_trans *tp = NULL; | |
1316 | int error; | |
99b6436b ZYW |
1317 | prid_t prid; |
1318 | struct xfs_dquot *udqp = NULL; | |
1319 | struct xfs_dquot *gdqp = NULL; | |
1320 | struct xfs_dquot *pdqp = NULL; | |
1321 | struct xfs_trans_res *tres; | |
1322 | uint resblks; | |
1323 | ||
1324 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1325 | return -EIO; |
99b6436b ZYW |
1326 | |
1327 | prid = xfs_get_initial_prid(dp); | |
1328 | ||
1329 | /* | |
1330 | * Make sure that we have allocated dquot(s) on disk. | |
1331 | */ | |
1332 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), | |
1333 | xfs_kgid_to_gid(current_fsgid()), prid, | |
1334 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
1335 | &udqp, &gdqp, &pdqp); | |
1336 | if (error) | |
1337 | return error; | |
1338 | ||
1339 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
99b6436b | 1340 | tres = &M_RES(mp)->tr_create_tmpfile; |
253f4911 CH |
1341 | |
1342 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); | |
2451337d | 1343 | if (error == -ENOSPC) { |
99b6436b ZYW |
1344 | /* No space at all so try a "no-allocation" reservation */ |
1345 | resblks = 0; | |
253f4911 | 1346 | error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); |
99b6436b | 1347 | } |
4906e215 | 1348 | if (error) |
253f4911 | 1349 | goto out_release_inode; |
99b6436b ZYW |
1350 | |
1351 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1352 | pdqp, resblks, 1, 0); | |
1353 | if (error) | |
1354 | goto out_trans_cancel; | |
1355 | ||
1356 | error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, | |
1357 | prid, resblks > 0, &ip, NULL); | |
d6077aa3 | 1358 | if (error) |
4906e215 | 1359 | goto out_trans_cancel; |
99b6436b ZYW |
1360 | |
1361 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1362 | xfs_trans_set_sync(tp); | |
1363 | ||
1364 | /* | |
1365 | * Attach the dquot(s) to the inodes and modify them incore. | |
1366 | * These ids of the inode couldn't have changed since the new | |
1367 | * inode has been locked ever since it was created. | |
1368 | */ | |
1369 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1370 | ||
99b6436b ZYW |
1371 | error = xfs_iunlink(tp, ip); |
1372 | if (error) | |
4906e215 | 1373 | goto out_trans_cancel; |
99b6436b | 1374 | |
70393313 | 1375 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1376 | if (error) |
1377 | goto out_release_inode; | |
1378 | ||
1379 | xfs_qm_dqrele(udqp); | |
1380 | xfs_qm_dqrele(gdqp); | |
1381 | xfs_qm_dqrele(pdqp); | |
1382 | ||
330033d6 | 1383 | *ipp = ip; |
99b6436b ZYW |
1384 | return 0; |
1385 | ||
99b6436b | 1386 | out_trans_cancel: |
4906e215 | 1387 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1388 | out_release_inode: |
1389 | /* | |
58c90473 DC |
1390 | * Wait until after the current transaction is aborted to finish the |
1391 | * setup of the inode and release the inode. This prevents recursive | |
1392 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1393 | */ |
58c90473 DC |
1394 | if (ip) { |
1395 | xfs_finish_inode_setup(ip); | |
99b6436b | 1396 | IRELE(ip); |
58c90473 | 1397 | } |
99b6436b ZYW |
1398 | |
1399 | xfs_qm_dqrele(udqp); | |
1400 | xfs_qm_dqrele(gdqp); | |
1401 | xfs_qm_dqrele(pdqp); | |
1402 | ||
1403 | return error; | |
1404 | } | |
1405 | ||
c24b5dfa DC |
1406 | int |
1407 | xfs_link( | |
1408 | xfs_inode_t *tdp, | |
1409 | xfs_inode_t *sip, | |
1410 | struct xfs_name *target_name) | |
1411 | { | |
1412 | xfs_mount_t *mp = tdp->i_mount; | |
1413 | xfs_trans_t *tp; | |
1414 | int error; | |
2c3234d1 | 1415 | struct xfs_defer_ops dfops; |
c24b5dfa | 1416 | xfs_fsblock_t first_block; |
c24b5dfa DC |
1417 | int resblks; |
1418 | ||
1419 | trace_xfs_link(tdp, target_name); | |
1420 | ||
c19b3b05 | 1421 | ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); |
c24b5dfa DC |
1422 | |
1423 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1424 | return -EIO; |
c24b5dfa DC |
1425 | |
1426 | error = xfs_qm_dqattach(sip, 0); | |
1427 | if (error) | |
1428 | goto std_return; | |
1429 | ||
1430 | error = xfs_qm_dqattach(tdp, 0); | |
1431 | if (error) | |
1432 | goto std_return; | |
1433 | ||
c24b5dfa | 1434 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
253f4911 | 1435 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); |
2451337d | 1436 | if (error == -ENOSPC) { |
c24b5dfa | 1437 | resblks = 0; |
253f4911 | 1438 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); |
c24b5dfa | 1439 | } |
4906e215 | 1440 | if (error) |
253f4911 | 1441 | goto std_return; |
c24b5dfa DC |
1442 | |
1443 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); | |
1444 | ||
1445 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
65523218 | 1446 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1447 | |
1448 | /* | |
1449 | * If we are using project inheritance, we only allow hard link | |
1450 | * creation in our tree when the project IDs are the same; else | |
1451 | * the tree quota mechanism could be circumvented. | |
1452 | */ | |
1453 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1454 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
2451337d | 1455 | error = -EXDEV; |
c24b5dfa DC |
1456 | goto error_return; |
1457 | } | |
1458 | ||
94f3cad5 ES |
1459 | if (!resblks) { |
1460 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1461 | if (error) | |
1462 | goto error_return; | |
1463 | } | |
c24b5dfa | 1464 | |
2c3234d1 | 1465 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa | 1466 | |
54d7b5c1 DC |
1467 | /* |
1468 | * Handle initial link state of O_TMPFILE inode | |
1469 | */ | |
1470 | if (VFS_I(sip)->i_nlink == 0) { | |
ab297431 ZYW |
1471 | error = xfs_iunlink_remove(tp, sip); |
1472 | if (error) | |
4906e215 | 1473 | goto error_return; |
ab297431 ZYW |
1474 | } |
1475 | ||
c24b5dfa | 1476 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2c3234d1 | 1477 | &first_block, &dfops, resblks); |
c24b5dfa | 1478 | if (error) |
4906e215 | 1479 | goto error_return; |
c24b5dfa DC |
1480 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1481 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1482 | ||
1483 | error = xfs_bumplink(tp, sip); | |
1484 | if (error) | |
4906e215 | 1485 | goto error_return; |
c24b5dfa DC |
1486 | |
1487 | /* | |
1488 | * If this is a synchronous mount, make sure that the | |
1489 | * link transaction goes to disk before returning to | |
1490 | * the user. | |
1491 | */ | |
f6106efa | 1492 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
c24b5dfa | 1493 | xfs_trans_set_sync(tp); |
c24b5dfa | 1494 | |
8ad7c629 | 1495 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa | 1496 | if (error) { |
2c3234d1 | 1497 | xfs_defer_cancel(&dfops); |
4906e215 | 1498 | goto error_return; |
c24b5dfa DC |
1499 | } |
1500 | ||
70393313 | 1501 | return xfs_trans_commit(tp); |
c24b5dfa | 1502 | |
c24b5dfa | 1503 | error_return: |
4906e215 | 1504 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1505 | std_return: |
1506 | return error; | |
1507 | } | |
1508 | ||
1da177e4 | 1509 | /* |
8f04c47a CH |
1510 | * Free up the underlying blocks past new_size. The new size must be smaller |
1511 | * than the current size. This routine can be used both for the attribute and | |
1512 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1513 | * |
f6485057 DC |
1514 | * The transaction passed to this routine must have made a permanent log |
1515 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1516 | * given transaction and start new ones, so make sure everything involved in | |
1517 | * the transaction is tidy before calling here. Some transaction will be | |
1518 | * returned to the caller to be committed. The incoming transaction must | |
1519 | * already include the inode, and both inode locks must be held exclusively. | |
1520 | * The inode must also be "held" within the transaction. On return the inode | |
1521 | * will be "held" within the returned transaction. This routine does NOT | |
1522 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1523 | * |
f6485057 DC |
1524 | * If we get an error, we must return with the inode locked and linked into the |
1525 | * current transaction. This keeps things simple for the higher level code, | |
1526 | * because it always knows that the inode is locked and held in the transaction | |
1527 | * that returns to it whether errors occur or not. We don't mark the inode | |
1528 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1529 | */ |
1530 | int | |
8f04c47a CH |
1531 | xfs_itruncate_extents( |
1532 | struct xfs_trans **tpp, | |
1533 | struct xfs_inode *ip, | |
1534 | int whichfork, | |
1535 | xfs_fsize_t new_size) | |
1da177e4 | 1536 | { |
8f04c47a CH |
1537 | struct xfs_mount *mp = ip->i_mount; |
1538 | struct xfs_trans *tp = *tpp; | |
2c3234d1 | 1539 | struct xfs_defer_ops dfops; |
8f04c47a CH |
1540 | xfs_fsblock_t first_block; |
1541 | xfs_fileoff_t first_unmap_block; | |
1542 | xfs_fileoff_t last_block; | |
1543 | xfs_filblks_t unmap_len; | |
8f04c47a CH |
1544 | int error = 0; |
1545 | int done = 0; | |
1da177e4 | 1546 | |
0b56185b CH |
1547 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1548 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1549 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1550 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1551 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1552 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1553 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1554 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1555 | |
673e8e59 CH |
1556 | trace_xfs_itruncate_extents_start(ip, new_size); |
1557 | ||
1da177e4 LT |
1558 | /* |
1559 | * Since it is possible for space to become allocated beyond | |
1560 | * the end of the file (in a crash where the space is allocated | |
1561 | * but the inode size is not yet updated), simply remove any | |
1562 | * blocks which show up between the new EOF and the maximum | |
1563 | * possible file size. If the first block to be removed is | |
1564 | * beyond the maximum file size (ie it is the same as last_block), | |
1565 | * then there is nothing to do. | |
1566 | */ | |
8f04c47a | 1567 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1568 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1569 | if (first_unmap_block == last_block) |
1570 | return 0; | |
1571 | ||
1572 | ASSERT(first_unmap_block < last_block); | |
1573 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1574 | while (!done) { |
2c3234d1 | 1575 | xfs_defer_init(&dfops, &first_block); |
8f04c47a | 1576 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1577 | first_unmap_block, unmap_len, |
8f04c47a | 1578 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1579 | XFS_ITRUNC_MAX_EXTENTS, |
2c3234d1 | 1580 | &first_block, &dfops, |
b4e9181e | 1581 | &done); |
8f04c47a CH |
1582 | if (error) |
1583 | goto out_bmap_cancel; | |
1da177e4 LT |
1584 | |
1585 | /* | |
1586 | * Duplicate the transaction that has the permanent | |
1587 | * reservation and commit the old transaction. | |
1588 | */ | |
8ad7c629 CH |
1589 | xfs_defer_ijoin(&dfops, ip); |
1590 | error = xfs_defer_finish(&tp, &dfops); | |
8f04c47a CH |
1591 | if (error) |
1592 | goto out_bmap_cancel; | |
1da177e4 | 1593 | |
411350df | 1594 | error = xfs_trans_roll_inode(&tp, ip); |
f6485057 | 1595 | if (error) |
8f04c47a | 1596 | goto out; |
1da177e4 | 1597 | } |
8f04c47a | 1598 | |
aa8968f2 DW |
1599 | /* Remove all pending CoW reservations. */ |
1600 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, | |
3802a345 | 1601 | last_block, true); |
aa8968f2 DW |
1602 | if (error) |
1603 | goto out; | |
1604 | ||
1605 | /* | |
cc6f7771 DW |
1606 | * Clear the reflink flag if there are no data fork blocks and |
1607 | * there are no extents staged in the cow fork. | |
aa8968f2 | 1608 | */ |
cc6f7771 DW |
1609 | if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) { |
1610 | if (ip->i_d.di_nblocks == 0) | |
1611 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
83104d44 DW |
1612 | xfs_inode_clear_cowblocks_tag(ip); |
1613 | } | |
aa8968f2 | 1614 | |
673e8e59 CH |
1615 | /* |
1616 | * Always re-log the inode so that our permanent transaction can keep | |
1617 | * on rolling it forward in the log. | |
1618 | */ | |
1619 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1620 | ||
1621 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1622 | ||
8f04c47a CH |
1623 | out: |
1624 | *tpp = tp; | |
1625 | return error; | |
1626 | out_bmap_cancel: | |
1da177e4 | 1627 | /* |
8f04c47a CH |
1628 | * If the bunmapi call encounters an error, return to the caller where |
1629 | * the transaction can be properly aborted. We just need to make sure | |
1630 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1631 | */ |
2c3234d1 | 1632 | xfs_defer_cancel(&dfops); |
8f04c47a CH |
1633 | goto out; |
1634 | } | |
1635 | ||
c24b5dfa DC |
1636 | int |
1637 | xfs_release( | |
1638 | xfs_inode_t *ip) | |
1639 | { | |
1640 | xfs_mount_t *mp = ip->i_mount; | |
1641 | int error; | |
1642 | ||
c19b3b05 | 1643 | if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) |
c24b5dfa DC |
1644 | return 0; |
1645 | ||
1646 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1647 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1648 | return 0; | |
1649 | ||
1650 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1651 | int truncated; | |
1652 | ||
c24b5dfa DC |
1653 | /* |
1654 | * If we previously truncated this file and removed old data | |
1655 | * in the process, we want to initiate "early" writeout on | |
1656 | * the last close. This is an attempt to combat the notorious | |
1657 | * NULL files problem which is particularly noticeable from a | |
1658 | * truncate down, buffered (re-)write (delalloc), followed by | |
1659 | * a crash. What we are effectively doing here is | |
1660 | * significantly reducing the time window where we'd otherwise | |
1661 | * be exposed to that problem. | |
1662 | */ | |
1663 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1664 | if (truncated) { | |
1665 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1666 | if (ip->i_delayed_blks > 0) { |
2451337d | 1667 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1668 | if (error) |
1669 | return error; | |
1670 | } | |
1671 | } | |
1672 | } | |
1673 | ||
54d7b5c1 | 1674 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1675 | return 0; |
1676 | ||
1677 | if (xfs_can_free_eofblocks(ip, false)) { | |
1678 | ||
a36b9261 BF |
1679 | /* |
1680 | * Check if the inode is being opened, written and closed | |
1681 | * frequently and we have delayed allocation blocks outstanding | |
1682 | * (e.g. streaming writes from the NFS server), truncating the | |
1683 | * blocks past EOF will cause fragmentation to occur. | |
1684 | * | |
1685 | * In this case don't do the truncation, but we have to be | |
1686 | * careful how we detect this case. Blocks beyond EOF show up as | |
1687 | * i_delayed_blks even when the inode is clean, so we need to | |
1688 | * truncate them away first before checking for a dirty release. | |
1689 | * Hence on the first dirty close we will still remove the | |
1690 | * speculative allocation, but after that we will leave it in | |
1691 | * place. | |
1692 | */ | |
1693 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1694 | return 0; | |
c24b5dfa DC |
1695 | /* |
1696 | * If we can't get the iolock just skip truncating the blocks | |
1697 | * past EOF because we could deadlock with the mmap_sem | |
a36b9261 | 1698 | * otherwise. We'll get another chance to drop them once the |
c24b5dfa DC |
1699 | * last reference to the inode is dropped, so we'll never leak |
1700 | * blocks permanently. | |
c24b5dfa | 1701 | */ |
a36b9261 BF |
1702 | if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
1703 | error = xfs_free_eofblocks(ip); | |
1704 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1705 | if (error) | |
1706 | return error; | |
1707 | } | |
c24b5dfa DC |
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 | ||
f7be2d7f BF |
1716 | /* |
1717 | * xfs_inactive_truncate | |
1718 | * | |
1719 | * Called to perform a truncate when an inode becomes unlinked. | |
1720 | */ | |
1721 | STATIC int | |
1722 | xfs_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 | ||
253f4911 | 1729 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
f7be2d7f BF |
1730 | if (error) { |
1731 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
f7be2d7f BF |
1732 | return error; |
1733 | } | |
1734 | ||
1735 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1736 | xfs_trans_ijoin(tp, ip, 0); | |
1737 | ||
1738 | /* | |
1739 | * Log the inode size first to prevent stale data exposure in the event | |
1740 | * of a system crash before the truncate completes. See the related | |
69bca807 | 1741 | * comment in xfs_vn_setattr_size() for details. |
f7be2d7f BF |
1742 | */ |
1743 | ip->i_d.di_size = 0; | |
1744 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1745 | ||
1746 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1747 | if (error) | |
1748 | goto error_trans_cancel; | |
1749 | ||
1750 | ASSERT(ip->i_d.di_nextents == 0); | |
1751 | ||
70393313 | 1752 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1753 | if (error) |
1754 | goto error_unlock; | |
1755 | ||
1756 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1757 | return 0; | |
1758 | ||
1759 | error_trans_cancel: | |
4906e215 | 1760 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1761 | error_unlock: |
1762 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1763 | return error; | |
1764 | } | |
1765 | ||
88877d2b BF |
1766 | /* |
1767 | * xfs_inactive_ifree() | |
1768 | * | |
1769 | * Perform the inode free when an inode is unlinked. | |
1770 | */ | |
1771 | STATIC int | |
1772 | xfs_inactive_ifree( | |
1773 | struct xfs_inode *ip) | |
1774 | { | |
2c3234d1 | 1775 | struct xfs_defer_ops dfops; |
88877d2b | 1776 | xfs_fsblock_t first_block; |
88877d2b BF |
1777 | struct xfs_mount *mp = ip->i_mount; |
1778 | struct xfs_trans *tp; | |
1779 | int error; | |
1780 | ||
9d43b180 | 1781 | /* |
76d771b4 CH |
1782 | * We try to use a per-AG reservation for any block needed by the finobt |
1783 | * tree, but as the finobt feature predates the per-AG reservation | |
1784 | * support a degraded file system might not have enough space for the | |
1785 | * reservation at mount time. In that case try to dip into the reserved | |
1786 | * pool and pray. | |
9d43b180 BF |
1787 | * |
1788 | * Send a warning if the reservation does happen to fail, as the inode | |
1789 | * now remains allocated and sits on the unlinked list until the fs is | |
1790 | * repaired. | |
1791 | */ | |
76d771b4 CH |
1792 | if (unlikely(mp->m_inotbt_nores)) { |
1793 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, | |
1794 | XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, | |
1795 | &tp); | |
1796 | } else { | |
1797 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); | |
1798 | } | |
88877d2b | 1799 | if (error) { |
2451337d | 1800 | if (error == -ENOSPC) { |
9d43b180 BF |
1801 | xfs_warn_ratelimited(mp, |
1802 | "Failed to remove inode(s) from unlinked list. " | |
1803 | "Please free space, unmount and run xfs_repair."); | |
1804 | } else { | |
1805 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1806 | } | |
88877d2b BF |
1807 | return error; |
1808 | } | |
1809 | ||
1810 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1811 | xfs_trans_ijoin(tp, ip, 0); | |
1812 | ||
2c3234d1 DW |
1813 | xfs_defer_init(&dfops, &first_block); |
1814 | error = xfs_ifree(tp, ip, &dfops); | |
88877d2b BF |
1815 | if (error) { |
1816 | /* | |
1817 | * If we fail to free the inode, shut down. The cancel | |
1818 | * might do that, we need to make sure. Otherwise the | |
1819 | * inode might be lost for a long time or forever. | |
1820 | */ | |
1821 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1822 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1823 | __func__, error); | |
1824 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1825 | } | |
4906e215 | 1826 | xfs_trans_cancel(tp); |
88877d2b BF |
1827 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1828 | return error; | |
1829 | } | |
1830 | ||
1831 | /* | |
1832 | * Credit the quota account(s). The inode is gone. | |
1833 | */ | |
1834 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1835 | ||
1836 | /* | |
d4a97a04 BF |
1837 | * Just ignore errors at this point. There is nothing we can do except |
1838 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1839 | */ |
8ad7c629 | 1840 | error = xfs_defer_finish(&tp, &dfops); |
d4a97a04 | 1841 | if (error) { |
310a75a3 | 1842 | xfs_notice(mp, "%s: xfs_defer_finish returned error %d", |
88877d2b | 1843 | __func__, error); |
2c3234d1 | 1844 | xfs_defer_cancel(&dfops); |
d4a97a04 | 1845 | } |
70393313 | 1846 | error = xfs_trans_commit(tp); |
88877d2b BF |
1847 | if (error) |
1848 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1849 | __func__, error); | |
1850 | ||
1851 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1852 | return 0; | |
1853 | } | |
1854 | ||
c24b5dfa DC |
1855 | /* |
1856 | * xfs_inactive | |
1857 | * | |
1858 | * This is called when the vnode reference count for the vnode | |
1859 | * goes to zero. If the file has been unlinked, then it must | |
1860 | * now be truncated. Also, we clear all of the read-ahead state | |
1861 | * kept for the inode here since the file is now closed. | |
1862 | */ | |
74564fb4 | 1863 | void |
c24b5dfa DC |
1864 | xfs_inactive( |
1865 | xfs_inode_t *ip) | |
1866 | { | |
3d3c8b52 | 1867 | struct xfs_mount *mp; |
3d3c8b52 JL |
1868 | int error; |
1869 | int truncate = 0; | |
c24b5dfa DC |
1870 | |
1871 | /* | |
1872 | * If the inode is already free, then there can be nothing | |
1873 | * to clean up here. | |
1874 | */ | |
c19b3b05 | 1875 | if (VFS_I(ip)->i_mode == 0) { |
c24b5dfa DC |
1876 | ASSERT(ip->i_df.if_real_bytes == 0); |
1877 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1878 | return; |
c24b5dfa DC |
1879 | } |
1880 | ||
1881 | mp = ip->i_mount; | |
17c12bcd | 1882 | ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); |
c24b5dfa | 1883 | |
c24b5dfa DC |
1884 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1885 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1886 | return; |
c24b5dfa | 1887 | |
54d7b5c1 | 1888 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1889 | /* |
1890 | * force is true because we are evicting an inode from the | |
1891 | * cache. Post-eof blocks must be freed, lest we end up with | |
1892 | * broken free space accounting. | |
3b4683c2 BF |
1893 | * |
1894 | * Note: don't bother with iolock here since lockdep complains | |
1895 | * about acquiring it in reclaim context. We have the only | |
1896 | * reference to the inode at this point anyways. | |
c24b5dfa | 1897 | */ |
3b4683c2 | 1898 | if (xfs_can_free_eofblocks(ip, true)) |
a36b9261 | 1899 | xfs_free_eofblocks(ip); |
74564fb4 BF |
1900 | |
1901 | return; | |
c24b5dfa DC |
1902 | } |
1903 | ||
c19b3b05 | 1904 | if (S_ISREG(VFS_I(ip)->i_mode) && |
c24b5dfa DC |
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) | |
74564fb4 | 1911 | return; |
c24b5dfa | 1912 | |
c19b3b05 | 1913 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
36b21dde | 1914 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1915 | else if (truncate) |
1916 | error = xfs_inactive_truncate(ip); | |
1917 | if (error) | |
74564fb4 | 1918 | return; |
c24b5dfa DC |
1919 | |
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 | |
6dfe5a04 | 1923 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1924 | */ |
6dfe5a04 | 1925 | if (XFS_IFORK_Q(ip)) { |
c24b5dfa DC |
1926 | error = xfs_attr_inactive(ip); |
1927 | if (error) | |
74564fb4 | 1928 | return; |
c24b5dfa DC |
1929 | } |
1930 | ||
6dfe5a04 | 1931 | ASSERT(!ip->i_afp); |
c24b5dfa | 1932 | ASSERT(ip->i_d.di_anextents == 0); |
6dfe5a04 | 1933 | ASSERT(ip->i_d.di_forkoff == 0); |
c24b5dfa DC |
1934 | |
1935 | /* | |
1936 | * Free the inode. | |
1937 | */ | |
88877d2b BF |
1938 | error = xfs_inactive_ifree(ip); |
1939 | if (error) | |
74564fb4 | 1940 | return; |
c24b5dfa DC |
1941 | |
1942 | /* | |
1943 | * Release the dquots held by inode, if any. | |
1944 | */ | |
1945 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1946 | } |
1947 | ||
1da177e4 | 1948 | /* |
54d7b5c1 DC |
1949 | * This is called when the inode's link count goes to 0 or we are creating a |
1950 | * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be | |
1951 | * set to true as the link count is dropped to zero by the VFS after we've | |
1952 | * created the file successfully, so we have to add it to the unlinked list | |
1953 | * while the link count is non-zero. | |
1954 | * | |
1955 | * We place the on-disk inode on a list in the AGI. It will be pulled from this | |
1956 | * list when the inode is freed. | |
1da177e4 | 1957 | */ |
54d7b5c1 | 1958 | STATIC int |
1da177e4 | 1959 | xfs_iunlink( |
54d7b5c1 DC |
1960 | struct xfs_trans *tp, |
1961 | struct xfs_inode *ip) | |
1da177e4 | 1962 | { |
54d7b5c1 | 1963 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1964 | xfs_agi_t *agi; |
1965 | xfs_dinode_t *dip; | |
1966 | xfs_buf_t *agibp; | |
1967 | xfs_buf_t *ibp; | |
1da177e4 LT |
1968 | xfs_agino_t agino; |
1969 | short bucket_index; | |
1970 | int offset; | |
1971 | int error; | |
1da177e4 | 1972 | |
c19b3b05 | 1973 | ASSERT(VFS_I(ip)->i_mode != 0); |
1da177e4 | 1974 | |
1da177e4 LT |
1975 | /* |
1976 | * Get the agi buffer first. It ensures lock ordering | |
1977 | * on the list. | |
1978 | */ | |
5e1be0fb | 1979 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1980 | if (error) |
1da177e4 | 1981 | return error; |
1da177e4 | 1982 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1983 | |
1da177e4 LT |
1984 | /* |
1985 | * Get the index into the agi hash table for the | |
1986 | * list this inode will go on. | |
1987 | */ | |
1988 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1989 | ASSERT(agino != 0); | |
1990 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1991 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1992 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1993 | |
69ef921b | 1994 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1995 | /* |
1996 | * There is already another inode in the bucket we need | |
1997 | * to add ourselves to. Add us at the front of the list. | |
1998 | * Here we put the head pointer into our next pointer, | |
1999 | * and then we fall through to point the head at us. | |
2000 | */ | |
475ee413 CH |
2001 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2002 | 0, 0); | |
c319b58b VA |
2003 | if (error) |
2004 | return error; | |
2005 | ||
69ef921b | 2006 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 2007 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 2008 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2009 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2010 | |
2011 | /* need to recalc the inode CRC if appropriate */ | |
2012 | xfs_dinode_calc_crc(mp, dip); | |
2013 | ||
1da177e4 LT |
2014 | xfs_trans_inode_buf(tp, ibp); |
2015 | xfs_trans_log_buf(tp, ibp, offset, | |
2016 | (offset + sizeof(xfs_agino_t) - 1)); | |
2017 | xfs_inobp_check(mp, ibp); | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * Point the bucket head pointer at the inode being inserted. | |
2022 | */ | |
2023 | ASSERT(agino != 0); | |
16259e7d | 2024 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
2025 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2026 | (sizeof(xfs_agino_t) * bucket_index); | |
2027 | xfs_trans_log_buf(tp, agibp, offset, | |
2028 | (offset + sizeof(xfs_agino_t) - 1)); | |
2029 | return 0; | |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * Pull the on-disk inode from the AGI unlinked list. | |
2034 | */ | |
2035 | STATIC int | |
2036 | xfs_iunlink_remove( | |
2037 | xfs_trans_t *tp, | |
2038 | xfs_inode_t *ip) | |
2039 | { | |
2040 | xfs_ino_t next_ino; | |
2041 | xfs_mount_t *mp; | |
2042 | xfs_agi_t *agi; | |
2043 | xfs_dinode_t *dip; | |
2044 | xfs_buf_t *agibp; | |
2045 | xfs_buf_t *ibp; | |
2046 | xfs_agnumber_t agno; | |
1da177e4 LT |
2047 | xfs_agino_t agino; |
2048 | xfs_agino_t next_agino; | |
2049 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2050 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2051 | short bucket_index; |
6fdf8ccc | 2052 | int offset, last_offset = 0; |
1da177e4 | 2053 | int error; |
1da177e4 | 2054 | |
1da177e4 | 2055 | mp = tp->t_mountp; |
1da177e4 | 2056 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * Get the agi buffer first. It ensures lock ordering | |
2060 | * on the list. | |
2061 | */ | |
5e1be0fb CH |
2062 | error = xfs_read_agi(mp, tp, agno, &agibp); |
2063 | if (error) | |
1da177e4 | 2064 | return error; |
5e1be0fb | 2065 | |
1da177e4 | 2066 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2067 | |
1da177e4 LT |
2068 | /* |
2069 | * Get the index into the agi hash table for the | |
2070 | * list this inode will go on. | |
2071 | */ | |
2072 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2073 | ASSERT(agino != 0); | |
2074 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 2075 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
2076 | ASSERT(agi->agi_unlinked[bucket_index]); |
2077 | ||
16259e7d | 2078 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2079 | /* |
475ee413 CH |
2080 | * We're at the head of the list. Get the inode's on-disk |
2081 | * buffer to see if there is anyone after us on the list. | |
2082 | * Only modify our next pointer if it is not already NULLAGINO. | |
2083 | * This saves us the overhead of dealing with the buffer when | |
2084 | * there is no need to change it. | |
1da177e4 | 2085 | */ |
475ee413 CH |
2086 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2087 | 0, 0); | |
1da177e4 | 2088 | if (error) { |
475ee413 | 2089 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2090 | __func__, error); |
1da177e4 LT |
2091 | return error; |
2092 | } | |
347d1c01 | 2093 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2094 | ASSERT(next_agino != 0); |
2095 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2096 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2097 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2098 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2099 | |
2100 | /* need to recalc the inode CRC if appropriate */ | |
2101 | xfs_dinode_calc_crc(mp, dip); | |
2102 | ||
1da177e4 LT |
2103 | xfs_trans_inode_buf(tp, ibp); |
2104 | xfs_trans_log_buf(tp, ibp, offset, | |
2105 | (offset + sizeof(xfs_agino_t) - 1)); | |
2106 | xfs_inobp_check(mp, ibp); | |
2107 | } else { | |
2108 | xfs_trans_brelse(tp, ibp); | |
2109 | } | |
2110 | /* | |
2111 | * Point the bucket head pointer at the next inode. | |
2112 | */ | |
2113 | ASSERT(next_agino != 0); | |
2114 | ASSERT(next_agino != agino); | |
16259e7d | 2115 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2116 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2117 | (sizeof(xfs_agino_t) * bucket_index); | |
2118 | 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 | */ | |
16259e7d | 2124 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2125 | last_ibp = NULL; |
2126 | while (next_agino != agino) { | |
129dbc9a CH |
2127 | struct xfs_imap imap; |
2128 | ||
2129 | if (last_ibp) | |
1da177e4 | 2130 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2131 | |
2132 | imap.im_blkno = 0; | |
1da177e4 | 2133 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2134 | |
2135 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2136 | if (error) { | |
2137 | xfs_warn(mp, | |
2138 | "%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); | |
1da177e4 | 2145 | if (error) { |
0b932ccc | 2146 | xfs_warn(mp, |
129dbc9a | 2147 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2148 | __func__, error); |
1da177e4 LT |
2149 | return error; |
2150 | } | |
129dbc9a CH |
2151 | |
2152 | last_offset = imap.im_boffset; | |
347d1c01 | 2153 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2154 | ASSERT(next_agino != NULLAGINO); |
2155 | ASSERT(next_agino != 0); | |
2156 | } | |
475ee413 | 2157 | |
1da177e4 | 2158 | /* |
475ee413 CH |
2159 | * Now last_ibp points to the buffer previous to us on the |
2160 | * unlinked list. Pull us from the list. | |
1da177e4 | 2161 | */ |
475ee413 CH |
2162 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2163 | 0, 0); | |
1da177e4 | 2164 | if (error) { |
475ee413 | 2165 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2166 | __func__, error); |
1da177e4 LT |
2167 | return error; |
2168 | } | |
347d1c01 | 2169 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2170 | ASSERT(next_agino != 0); |
2171 | ASSERT(next_agino != agino); | |
2172 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2173 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2174 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2175 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2176 | |
2177 | /* need to recalc the inode CRC if appropriate */ | |
2178 | xfs_dinode_calc_crc(mp, dip); | |
2179 | ||
1da177e4 LT |
2180 | 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 | */ | |
347d1c01 | 2190 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2191 | ASSERT(next_agino != 0); |
2192 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2193 | |
2194 | /* need to recalc the inode CRC if appropriate */ | |
2195 | xfs_dinode_calc_crc(mp, last_dip); | |
2196 | ||
1da177e4 LT |
2197 | 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 | ||
5b3eed75 | 2205 | /* |
0b8182db | 2206 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2207 | * inodes that are in memory - they all must be marked stale and attached to |
2208 | * the cluster buffer. | |
2209 | */ | |
2a30f36d | 2210 | STATIC int |
1da177e4 | 2211 | xfs_ifree_cluster( |
09b56604 BF |
2212 | xfs_inode_t *free_ip, |
2213 | xfs_trans_t *tp, | |
2214 | struct xfs_icluster *xic) | |
1da177e4 LT |
2215 | { |
2216 | xfs_mount_t *mp = free_ip->i_mount; | |
2217 | int blks_per_cluster; | |
982e939e | 2218 | int inodes_per_cluster; |
1da177e4 | 2219 | int nbufs; |
5b257b4a | 2220 | int i, j; |
3cdaa189 | 2221 | int ioffset; |
1da177e4 LT |
2222 | xfs_daddr_t blkno; |
2223 | xfs_buf_t *bp; | |
5b257b4a | 2224 | xfs_inode_t *ip; |
1da177e4 LT |
2225 | xfs_inode_log_item_t *iip; |
2226 | xfs_log_item_t *lip; | |
5017e97d | 2227 | struct xfs_perag *pag; |
09b56604 | 2228 | xfs_ino_t inum; |
1da177e4 | 2229 | |
09b56604 | 2230 | inum = xic->first_ino; |
5017e97d | 2231 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
982e939e JL |
2232 | 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; | |
1da177e4 | 2235 | |
982e939e | 2236 | for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { |
09b56604 BF |
2237 | /* |
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 | */ | |
3cdaa189 BF |
2242 | ioffset = inum - xic->first_ino; |
2243 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
2244 | ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); | |
09b56604 BF |
2245 | continue; |
2246 | } | |
2247 | ||
1da177e4 LT |
2248 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2249 | XFS_INO_TO_AGBNO(mp, inum)); | |
2250 | ||
5b257b4a DC |
2251 | /* |
2252 | * 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. | |
2258 | */ | |
2259 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2260 | mp->m_bsize * blks_per_cluster, |
2261 | XBF_UNMAPPED); | |
5b257b4a | 2262 | |
2a30f36d | 2263 | if (!bp) |
2451337d | 2264 | return -ENOMEM; |
b0f539de DC |
2265 | |
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 | */ | |
1813dd64 | 2275 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2276 | |
5b257b4a DC |
2277 | /* |
2278 | * Walk the inodes already attached to the buffer and mark them | |
2279 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2280 | * 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. | |
5b257b4a | 2283 | */ |
adadbeef | 2284 | lip = bp->b_fspriv; |
5b257b4a DC |
2285 | while (lip) { |
2286 | if (lip->li_type == XFS_LI_INODE) { | |
2287 | iip = (xfs_inode_log_item_t *)lip; | |
2288 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2289 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2290 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2291 | &iip->ili_flush_lsn, | |
2292 | &iip->ili_item.li_lsn); | |
2293 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
2294 | } |
2295 | lip = lip->li_bio_list; | |
2296 | } | |
1da177e4 | 2297 | |
5b3eed75 | 2298 | |
1da177e4 | 2299 | /* |
5b257b4a DC |
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. | |
1da177e4 | 2304 | * |
5b257b4a DC |
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. | |
1da177e4 | 2308 | */ |
982e939e | 2309 | for (i = 0; i < inodes_per_cluster; i++) { |
5b3eed75 | 2310 | retry: |
1a3e8f3d | 2311 | rcu_read_lock(); |
da353b0d DC |
2312 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2313 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2314 | |
1a3e8f3d DC |
2315 | /* Inode not in memory, nothing to do */ |
2316 | if (!ip) { | |
2317 | rcu_read_unlock(); | |
1da177e4 LT |
2318 | continue; |
2319 | } | |
2320 | ||
1a3e8f3d DC |
2321 | /* |
2322 | * 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 | ||
5b3eed75 DC |
2337 | /* |
2338 | * 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 | */ | |
f2e9ad21 OS |
2344 | if (ip != free_ip) { |
2345 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2346 | rcu_read_unlock(); | |
2347 | delay(1); | |
2348 | goto retry; | |
2349 | } | |
2350 | ||
2351 | /* | |
2352 | * Check the inode number again in case we're | |
2353 | * racing with freeing in xfs_reclaim_inode(). | |
2354 | * See the comments in that function for more | |
2355 | * information as to why the initial check is | |
2356 | * not sufficient. | |
2357 | */ | |
2358 | if (ip->i_ino != inum + i) { | |
2359 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
962cc1ad | 2360 | rcu_read_unlock(); |
f2e9ad21 OS |
2361 | continue; |
2362 | } | |
1da177e4 | 2363 | } |
1a3e8f3d | 2364 | rcu_read_unlock(); |
1da177e4 | 2365 | |
5b3eed75 | 2366 | xfs_iflock(ip); |
5b257b4a | 2367 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2368 | |
5b3eed75 DC |
2369 | /* |
2370 | * we don't need to attach clean inodes or those only | |
2371 | * with unlogged changes (which we throw away, anyway). | |
2372 | */ | |
1da177e4 | 2373 | iip = ip->i_itemp; |
5b3eed75 | 2374 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2375 | ASSERT(ip != free_ip); |
1da177e4 LT |
2376 | xfs_ifunlock(ip); |
2377 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2378 | continue; | |
2379 | } | |
2380 | ||
f5d8d5c4 CH |
2381 | iip->ili_last_fields = iip->ili_fields; |
2382 | iip->ili_fields = 0; | |
fc0561ce | 2383 | iip->ili_fsync_fields = 0; |
1da177e4 | 2384 | iip->ili_logged = 1; |
7b2e2a31 DC |
2385 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2386 | &iip->ili_item.li_lsn); | |
1da177e4 | 2387 | |
ca30b2a7 CH |
2388 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2389 | &iip->ili_item); | |
5b257b4a DC |
2390 | |
2391 | if (ip != free_ip) | |
1da177e4 | 2392 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2393 | } |
2394 | ||
5b3eed75 | 2395 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2396 | xfs_trans_binval(tp, bp); |
2397 | } | |
2398 | ||
5017e97d | 2399 | xfs_perag_put(pag); |
2a30f36d | 2400 | return 0; |
1da177e4 LT |
2401 | } |
2402 | ||
98c4f78d DW |
2403 | /* |
2404 | * Free any local-format buffers sitting around before we reset to | |
2405 | * extents format. | |
2406 | */ | |
2407 | static inline void | |
2408 | xfs_ifree_local_data( | |
2409 | struct xfs_inode *ip, | |
2410 | int whichfork) | |
2411 | { | |
2412 | struct xfs_ifork *ifp; | |
2413 | ||
2414 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL) | |
2415 | return; | |
2416 | ||
2417 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2418 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); | |
2419 | } | |
2420 | ||
1da177e4 LT |
2421 | /* |
2422 | * This is called to return an inode to the inode free list. | |
2423 | * The inode should already be truncated to 0 length and have | |
2424 | * no pages associated with it. This routine also assumes that | |
2425 | * the inode is already a part of the transaction. | |
2426 | * | |
2427 | * The on-disk copy of the inode will have been added to the list | |
2428 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2429 | * that list atomically with respect to freeing it here. | |
2430 | */ | |
2431 | int | |
2432 | xfs_ifree( | |
2433 | xfs_trans_t *tp, | |
2434 | xfs_inode_t *ip, | |
2c3234d1 | 2435 | struct xfs_defer_ops *dfops) |
1da177e4 LT |
2436 | { |
2437 | int error; | |
09b56604 | 2438 | struct xfs_icluster xic = { 0 }; |
1da177e4 | 2439 | |
579aa9ca | 2440 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2441 | ASSERT(VFS_I(ip)->i_nlink == 0); |
1da177e4 LT |
2442 | ASSERT(ip->i_d.di_nextents == 0); |
2443 | ASSERT(ip->i_d.di_anextents == 0); | |
c19b3b05 | 2444 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); |
1da177e4 LT |
2445 | ASSERT(ip->i_d.di_nblocks == 0); |
2446 | ||
2447 | /* | |
2448 | * Pull the on-disk inode from the AGI unlinked list. | |
2449 | */ | |
2450 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2451 | if (error) |
1da177e4 | 2452 | return error; |
1da177e4 | 2453 | |
2c3234d1 | 2454 | error = xfs_difree(tp, ip->i_ino, dfops, &xic); |
1baaed8f | 2455 | if (error) |
1da177e4 | 2456 | return error; |
1baaed8f | 2457 | |
98c4f78d DW |
2458 | xfs_ifree_local_data(ip, XFS_DATA_FORK); |
2459 | xfs_ifree_local_data(ip, XFS_ATTR_FORK); | |
2460 | ||
c19b3b05 | 2461 | VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ |
1da177e4 LT |
2462 | ip->i_d.di_flags = 0; |
2463 | ip->i_d.di_dmevmask = 0; | |
2464 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2465 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2466 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2467 | /* | |
2468 | * Bump the generation count so no one will be confused | |
2469 | * by reincarnations of this inode. | |
2470 | */ | |
9e9a2674 | 2471 | VFS_I(ip)->i_generation++; |
1da177e4 LT |
2472 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2473 | ||
09b56604 BF |
2474 | if (xic.deleted) |
2475 | error = xfs_ifree_cluster(ip, tp, &xic); | |
1da177e4 | 2476 | |
2a30f36d | 2477 | return error; |
1da177e4 LT |
2478 | } |
2479 | ||
1da177e4 | 2480 | /* |
60ec6783 CH |
2481 | * This is called to unpin an inode. The caller must have the inode locked |
2482 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2483 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2484 | */ |
60ec6783 | 2485 | static void |
f392e631 | 2486 | xfs_iunpin( |
60ec6783 | 2487 | struct xfs_inode *ip) |
1da177e4 | 2488 | { |
579aa9ca | 2489 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2490 | |
4aaf15d1 DC |
2491 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2492 | ||
a3f74ffb | 2493 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2494 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2495 | |
a3f74ffb | 2496 | } |
1da177e4 | 2497 | |
f392e631 CH |
2498 | static void |
2499 | __xfs_iunpin_wait( | |
2500 | struct xfs_inode *ip) | |
2501 | { | |
2502 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2503 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2504 | ||
2505 | xfs_iunpin(ip); | |
2506 | ||
2507 | do { | |
21417136 | 2508 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
f392e631 CH |
2509 | if (xfs_ipincount(ip)) |
2510 | io_schedule(); | |
2511 | } while (xfs_ipincount(ip)); | |
21417136 | 2512 | finish_wait(wq, &wait.wq_entry); |
f392e631 CH |
2513 | } |
2514 | ||
777df5af | 2515 | void |
a3f74ffb | 2516 | xfs_iunpin_wait( |
60ec6783 | 2517 | struct xfs_inode *ip) |
a3f74ffb | 2518 | { |
f392e631 CH |
2519 | if (xfs_ipincount(ip)) |
2520 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2521 | } |
2522 | ||
27320369 DC |
2523 | /* |
2524 | * Removing an inode from the namespace involves removing the directory entry | |
2525 | * and dropping the link count on the inode. Removing the directory entry can | |
2526 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2527 | * count can result in placing the inode on an unlinked list which results in | |
2528 | * locking an AGI. | |
2529 | * | |
2530 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2531 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2532 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2533 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2534 | * first, and then freeing the inode can result in an inode chunk being freed | |
2535 | * and hence freeing disk space requiring that we lock an AGF. | |
2536 | * | |
2537 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2538 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2539 | * reference count and put it on the unlinked list as this results in a lock | |
2540 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2541 | * freeing. Therefore we must drop the link counts before we remove the | |
2542 | * directory entry. | |
2543 | * | |
2544 | * This is still safe from a transactional point of view - it is not until we | |
310a75a3 | 2545 | * get to xfs_defer_finish() that we have the possibility of multiple |
27320369 DC |
2546 | * transactions in this operation. Hence as long as we remove the directory |
2547 | * entry and drop the link count in the first transaction of the remove | |
2548 | * operation, there are no transactional constraints on the ordering here. | |
2549 | */ | |
c24b5dfa DC |
2550 | int |
2551 | xfs_remove( | |
2552 | xfs_inode_t *dp, | |
2553 | struct xfs_name *name, | |
2554 | xfs_inode_t *ip) | |
2555 | { | |
2556 | xfs_mount_t *mp = dp->i_mount; | |
2557 | xfs_trans_t *tp = NULL; | |
c19b3b05 | 2558 | int is_dir = S_ISDIR(VFS_I(ip)->i_mode); |
c24b5dfa | 2559 | int error = 0; |
2c3234d1 | 2560 | struct xfs_defer_ops dfops; |
c24b5dfa | 2561 | xfs_fsblock_t first_block; |
c24b5dfa | 2562 | uint resblks; |
c24b5dfa DC |
2563 | |
2564 | trace_xfs_remove(dp, name); | |
2565 | ||
2566 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 2567 | return -EIO; |
c24b5dfa DC |
2568 | |
2569 | error = xfs_qm_dqattach(dp, 0); | |
2570 | if (error) | |
2571 | goto std_return; | |
2572 | ||
2573 | error = xfs_qm_dqattach(ip, 0); | |
2574 | if (error) | |
2575 | goto std_return; | |
2576 | ||
c24b5dfa DC |
2577 | /* |
2578 | * We try to get the real space reservation first, | |
2579 | * allowing for directory btree deletion(s) implying | |
2580 | * possible bmap insert(s). If we can't get the space | |
2581 | * reservation then we use 0 instead, and avoid the bmap | |
2582 | * btree insert(s) in the directory code by, if the bmap | |
2583 | * insert tries to happen, instead trimming the LAST | |
2584 | * block from the directory. | |
2585 | */ | |
2586 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
253f4911 | 2587 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); |
2451337d | 2588 | if (error == -ENOSPC) { |
c24b5dfa | 2589 | resblks = 0; |
253f4911 CH |
2590 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, |
2591 | &tp); | |
c24b5dfa DC |
2592 | } |
2593 | if (error) { | |
2451337d | 2594 | ASSERT(error != -ENOSPC); |
253f4911 | 2595 | goto std_return; |
c24b5dfa DC |
2596 | } |
2597 | ||
2598 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); | |
2599 | ||
65523218 | 2600 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
2601 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
2602 | ||
2603 | /* | |
2604 | * If we're removing a directory perform some additional validation. | |
2605 | */ | |
2606 | if (is_dir) { | |
54d7b5c1 DC |
2607 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2608 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2609 | error = -ENOTEMPTY; |
c24b5dfa DC |
2610 | goto out_trans_cancel; |
2611 | } | |
2612 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2613 | error = -ENOTEMPTY; |
c24b5dfa DC |
2614 | goto out_trans_cancel; |
2615 | } | |
c24b5dfa | 2616 | |
27320369 | 2617 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2618 | error = xfs_droplink(tp, dp); |
2619 | if (error) | |
27320369 | 2620 | goto out_trans_cancel; |
c24b5dfa | 2621 | |
27320369 | 2622 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2623 | error = xfs_droplink(tp, ip); |
2624 | if (error) | |
27320369 | 2625 | goto out_trans_cancel; |
c24b5dfa DC |
2626 | } else { |
2627 | /* | |
2628 | * When removing a non-directory we need to log the parent | |
2629 | * inode here. For a directory this is done implicitly | |
2630 | * by the xfs_droplink call for the ".." entry. | |
2631 | */ | |
2632 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2633 | } | |
27320369 | 2634 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2635 | |
27320369 | 2636 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2637 | error = xfs_droplink(tp, ip); |
2638 | if (error) | |
27320369 | 2639 | goto out_trans_cancel; |
c24b5dfa | 2640 | |
2c3234d1 | 2641 | xfs_defer_init(&dfops, &first_block); |
27320369 | 2642 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
2c3234d1 | 2643 | &first_block, &dfops, resblks); |
27320369 | 2644 | if (error) { |
2451337d | 2645 | ASSERT(error != -ENOENT); |
27320369 DC |
2646 | goto out_bmap_cancel; |
2647 | } | |
2648 | ||
c24b5dfa DC |
2649 | /* |
2650 | * If this is a synchronous mount, make sure that the | |
2651 | * remove transaction goes to disk before returning to | |
2652 | * the user. | |
2653 | */ | |
2654 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2655 | xfs_trans_set_sync(tp); | |
2656 | ||
8ad7c629 | 2657 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
2658 | if (error) |
2659 | goto out_bmap_cancel; | |
2660 | ||
70393313 | 2661 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2662 | if (error) |
2663 | goto std_return; | |
2664 | ||
2cd2ef6a | 2665 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2666 | xfs_filestream_deassociate(ip); |
2667 | ||
2668 | return 0; | |
2669 | ||
2670 | out_bmap_cancel: | |
2c3234d1 | 2671 | xfs_defer_cancel(&dfops); |
c24b5dfa | 2672 | out_trans_cancel: |
4906e215 | 2673 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2674 | std_return: |
2675 | return error; | |
2676 | } | |
2677 | ||
f6bba201 DC |
2678 | /* |
2679 | * Enter all inodes for a rename transaction into a sorted array. | |
2680 | */ | |
95afcf5c | 2681 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2682 | STATIC void |
2683 | xfs_sort_for_rename( | |
95afcf5c DC |
2684 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2685 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2686 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2687 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2688 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2689 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2690 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2691 | { |
f6bba201 DC |
2692 | int i, j; |
2693 | ||
95afcf5c DC |
2694 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2695 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2696 | ||
f6bba201 DC |
2697 | /* |
2698 | * i_tab contains a list of pointers to inodes. We initialize | |
2699 | * the table here & we'll sort it. We will then use it to | |
2700 | * order the acquisition of the inode locks. | |
2701 | * | |
2702 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2703 | */ | |
95afcf5c DC |
2704 | i = 0; |
2705 | i_tab[i++] = dp1; | |
2706 | i_tab[i++] = dp2; | |
2707 | i_tab[i++] = ip1; | |
2708 | if (ip2) | |
2709 | i_tab[i++] = ip2; | |
2710 | if (wip) | |
2711 | i_tab[i++] = wip; | |
2712 | *num_inodes = i; | |
f6bba201 DC |
2713 | |
2714 | /* | |
2715 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2716 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2717 | */ |
2718 | for (i = 0; i < *num_inodes; i++) { | |
2719 | for (j = 1; j < *num_inodes; j++) { | |
2720 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2721 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2722 | i_tab[j] = i_tab[j-1]; |
2723 | i_tab[j-1] = temp; | |
2724 | } | |
2725 | } | |
2726 | } | |
2727 | } | |
2728 | ||
310606b0 DC |
2729 | static int |
2730 | xfs_finish_rename( | |
2731 | struct xfs_trans *tp, | |
2c3234d1 | 2732 | struct xfs_defer_ops *dfops) |
310606b0 | 2733 | { |
310606b0 DC |
2734 | int error; |
2735 | ||
2736 | /* | |
2737 | * If this is a synchronous mount, make sure that the rename transaction | |
2738 | * goes to disk before returning to the user. | |
2739 | */ | |
2740 | if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2741 | xfs_trans_set_sync(tp); | |
2742 | ||
8ad7c629 | 2743 | error = xfs_defer_finish(&tp, dfops); |
310606b0 | 2744 | if (error) { |
2c3234d1 | 2745 | xfs_defer_cancel(dfops); |
4906e215 | 2746 | xfs_trans_cancel(tp); |
310606b0 DC |
2747 | return error; |
2748 | } | |
2749 | ||
70393313 | 2750 | return xfs_trans_commit(tp); |
310606b0 DC |
2751 | } |
2752 | ||
d31a1825 CM |
2753 | /* |
2754 | * xfs_cross_rename() | |
2755 | * | |
2756 | * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall | |
2757 | */ | |
2758 | STATIC int | |
2759 | xfs_cross_rename( | |
2760 | struct xfs_trans *tp, | |
2761 | struct xfs_inode *dp1, | |
2762 | struct xfs_name *name1, | |
2763 | struct xfs_inode *ip1, | |
2764 | struct xfs_inode *dp2, | |
2765 | struct xfs_name *name2, | |
2766 | struct xfs_inode *ip2, | |
2c3234d1 | 2767 | struct xfs_defer_ops *dfops, |
d31a1825 CM |
2768 | xfs_fsblock_t *first_block, |
2769 | int spaceres) | |
2770 | { | |
2771 | int error = 0; | |
2772 | int ip1_flags = 0; | |
2773 | int ip2_flags = 0; | |
2774 | int dp2_flags = 0; | |
2775 | ||
2776 | /* Swap inode number for dirent in first parent */ | |
2777 | error = xfs_dir_replace(tp, dp1, name1, | |
2778 | ip2->i_ino, | |
2c3234d1 | 2779 | first_block, dfops, spaceres); |
d31a1825 | 2780 | if (error) |
eeacd321 | 2781 | goto out_trans_abort; |
d31a1825 CM |
2782 | |
2783 | /* Swap inode number for dirent in second parent */ | |
2784 | error = xfs_dir_replace(tp, dp2, name2, | |
2785 | ip1->i_ino, | |
2c3234d1 | 2786 | first_block, dfops, spaceres); |
d31a1825 | 2787 | if (error) |
eeacd321 | 2788 | goto out_trans_abort; |
d31a1825 CM |
2789 | |
2790 | /* | |
2791 | * If we're renaming one or more directories across different parents, | |
2792 | * update the respective ".." entries (and link counts) to match the new | |
2793 | * parents. | |
2794 | */ | |
2795 | if (dp1 != dp2) { | |
2796 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2797 | ||
c19b3b05 | 2798 | if (S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2799 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, |
2800 | dp1->i_ino, first_block, | |
2c3234d1 | 2801 | dfops, spaceres); |
d31a1825 | 2802 | if (error) |
eeacd321 | 2803 | goto out_trans_abort; |
d31a1825 CM |
2804 | |
2805 | /* transfer ip2 ".." reference to dp1 */ | |
c19b3b05 | 2806 | if (!S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2807 | error = xfs_droplink(tp, dp2); |
2808 | if (error) | |
eeacd321 | 2809 | goto out_trans_abort; |
d31a1825 CM |
2810 | error = xfs_bumplink(tp, dp1); |
2811 | if (error) | |
eeacd321 | 2812 | goto out_trans_abort; |
d31a1825 CM |
2813 | } |
2814 | ||
2815 | /* | |
2816 | * Although ip1 isn't changed here, userspace needs | |
2817 | * to be warned about the change, so that applications | |
2818 | * relying on it (like backup ones), will properly | |
2819 | * notify the change | |
2820 | */ | |
2821 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2822 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2823 | } | |
2824 | ||
c19b3b05 | 2825 | if (S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2826 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, |
2827 | dp2->i_ino, first_block, | |
2c3234d1 | 2828 | dfops, spaceres); |
d31a1825 | 2829 | if (error) |
eeacd321 | 2830 | goto out_trans_abort; |
d31a1825 CM |
2831 | |
2832 | /* transfer ip1 ".." reference to dp2 */ | |
c19b3b05 | 2833 | if (!S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2834 | error = xfs_droplink(tp, dp1); |
2835 | if (error) | |
eeacd321 | 2836 | goto out_trans_abort; |
d31a1825 CM |
2837 | error = xfs_bumplink(tp, dp2); |
2838 | if (error) | |
eeacd321 | 2839 | goto out_trans_abort; |
d31a1825 CM |
2840 | } |
2841 | ||
2842 | /* | |
2843 | * Although ip2 isn't changed here, userspace needs | |
2844 | * to be warned about the change, so that applications | |
2845 | * relying on it (like backup ones), will properly | |
2846 | * notify the change | |
2847 | */ | |
2848 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2849 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2850 | } | |
2851 | } | |
2852 | ||
2853 | if (ip1_flags) { | |
2854 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2855 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2856 | } | |
2857 | if (ip2_flags) { | |
2858 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2859 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2860 | } | |
2861 | if (dp2_flags) { | |
2862 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2863 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2864 | } | |
2865 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2866 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
2c3234d1 | 2867 | return xfs_finish_rename(tp, dfops); |
eeacd321 DC |
2868 | |
2869 | out_trans_abort: | |
2c3234d1 | 2870 | xfs_defer_cancel(dfops); |
4906e215 | 2871 | xfs_trans_cancel(tp); |
d31a1825 CM |
2872 | return error; |
2873 | } | |
2874 | ||
7dcf5c3e DC |
2875 | /* |
2876 | * xfs_rename_alloc_whiteout() | |
2877 | * | |
2878 | * Return a referenced, unlinked, unlocked inode that that can be used as a | |
2879 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we | |
2880 | * crash between allocating the inode and linking it into the rename transaction | |
2881 | * recovery will free the inode and we won't leak it. | |
2882 | */ | |
2883 | static int | |
2884 | xfs_rename_alloc_whiteout( | |
2885 | struct xfs_inode *dp, | |
2886 | struct xfs_inode **wip) | |
2887 | { | |
2888 | struct xfs_inode *tmpfile; | |
2889 | int error; | |
2890 | ||
2891 | error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); | |
2892 | if (error) | |
2893 | return error; | |
2894 | ||
22419ac9 BF |
2895 | /* |
2896 | * Prepare the tmpfile inode as if it were created through the VFS. | |
2897 | * Otherwise, the link increment paths will complain about nlink 0->1. | |
2898 | * Drop the link count as done by d_tmpfile(), complete the inode setup | |
2899 | * and flag it as linkable. | |
2900 | */ | |
2901 | drop_nlink(VFS_I(tmpfile)); | |
2b3d1d41 | 2902 | xfs_setup_iops(tmpfile); |
7dcf5c3e DC |
2903 | xfs_finish_inode_setup(tmpfile); |
2904 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2905 | ||
2906 | *wip = tmpfile; | |
2907 | return 0; | |
2908 | } | |
2909 | ||
f6bba201 DC |
2910 | /* |
2911 | * xfs_rename | |
2912 | */ | |
2913 | int | |
2914 | xfs_rename( | |
7dcf5c3e DC |
2915 | struct xfs_inode *src_dp, |
2916 | struct xfs_name *src_name, | |
2917 | struct xfs_inode *src_ip, | |
2918 | struct xfs_inode *target_dp, | |
2919 | struct xfs_name *target_name, | |
2920 | struct xfs_inode *target_ip, | |
2921 | unsigned int flags) | |
f6bba201 | 2922 | { |
7dcf5c3e DC |
2923 | struct xfs_mount *mp = src_dp->i_mount; |
2924 | struct xfs_trans *tp; | |
2c3234d1 | 2925 | struct xfs_defer_ops dfops; |
7dcf5c3e DC |
2926 | xfs_fsblock_t first_block; |
2927 | struct xfs_inode *wip = NULL; /* whiteout inode */ | |
2928 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
2929 | int num_inodes = __XFS_SORT_INODES; | |
2b93681f | 2930 | bool new_parent = (src_dp != target_dp); |
c19b3b05 | 2931 | bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); |
7dcf5c3e DC |
2932 | int spaceres; |
2933 | int error; | |
f6bba201 DC |
2934 | |
2935 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2936 | ||
eeacd321 DC |
2937 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2938 | return -EINVAL; | |
2939 | ||
7dcf5c3e DC |
2940 | /* |
2941 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2942 | * we will be placing at the target and ensure the type is set | |
2943 | * appropriately. | |
2944 | */ | |
2945 | if (flags & RENAME_WHITEOUT) { | |
2946 | ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); | |
2947 | error = xfs_rename_alloc_whiteout(target_dp, &wip); | |
2948 | if (error) | |
2949 | return error; | |
2950 | ||
2951 | /* setup target dirent info as whiteout */ | |
2952 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2953 | } | |
f6bba201 | 2954 | |
7dcf5c3e | 2955 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2956 | inodes, &num_inodes); |
2957 | ||
f6bba201 | 2958 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
253f4911 | 2959 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); |
2451337d | 2960 | if (error == -ENOSPC) { |
f6bba201 | 2961 | spaceres = 0; |
253f4911 CH |
2962 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, |
2963 | &tp); | |
f6bba201 | 2964 | } |
445883e8 | 2965 | if (error) |
253f4911 | 2966 | goto out_release_wip; |
f6bba201 DC |
2967 | |
2968 | /* | |
2969 | * Attach the dquots to the inodes | |
2970 | */ | |
2971 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2972 | if (error) |
2973 | goto out_trans_cancel; | |
f6bba201 DC |
2974 | |
2975 | /* | |
2976 | * Lock all the participating inodes. Depending upon whether | |
2977 | * the target_name exists in the target directory, and | |
2978 | * whether the target directory is the same as the source | |
2979 | * directory, we can lock from 2 to 4 inodes. | |
2980 | */ | |
2981 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2982 | ||
2983 | /* | |
2984 | * Join all the inodes to the transaction. From this point on, | |
2985 | * we can rely on either trans_commit or trans_cancel to unlock | |
2986 | * them. | |
2987 | */ | |
65523218 | 2988 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
f6bba201 | 2989 | if (new_parent) |
65523218 | 2990 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
f6bba201 DC |
2991 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2992 | if (target_ip) | |
2993 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2994 | if (wip) |
2995 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
2996 | |
2997 | /* | |
2998 | * If we are using project inheritance, we only allow renames | |
2999 | * into our tree when the project IDs are the same; else the | |
3000 | * tree quota mechanism would be circumvented. | |
3001 | */ | |
3002 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
3003 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2451337d | 3004 | error = -EXDEV; |
445883e8 | 3005 | goto out_trans_cancel; |
f6bba201 DC |
3006 | } |
3007 | ||
2c3234d1 | 3008 | xfs_defer_init(&dfops, &first_block); |
445883e8 | 3009 | |
eeacd321 DC |
3010 | /* RENAME_EXCHANGE is unique from here on. */ |
3011 | if (flags & RENAME_EXCHANGE) | |
3012 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
3013 | target_dp, target_name, target_ip, | |
2c3234d1 | 3014 | &dfops, &first_block, spaceres); |
d31a1825 | 3015 | |
f6bba201 DC |
3016 | /* |
3017 | * Set up the target. | |
3018 | */ | |
3019 | if (target_ip == NULL) { | |
3020 | /* | |
3021 | * If there's no space reservation, check the entry will | |
3022 | * fit before actually inserting it. | |
3023 | */ | |
94f3cad5 ES |
3024 | if (!spaceres) { |
3025 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
3026 | if (error) | |
445883e8 | 3027 | goto out_trans_cancel; |
94f3cad5 | 3028 | } |
f6bba201 DC |
3029 | /* |
3030 | * If target does not exist and the rename crosses | |
3031 | * directories, adjust the target directory link count | |
3032 | * to account for the ".." reference from the new entry. | |
3033 | */ | |
3034 | error = xfs_dir_createname(tp, target_dp, target_name, | |
3035 | src_ip->i_ino, &first_block, | |
2c3234d1 | 3036 | &dfops, spaceres); |
f6bba201 | 3037 | if (error) |
4906e215 | 3038 | goto out_bmap_cancel; |
f6bba201 DC |
3039 | |
3040 | xfs_trans_ichgtime(tp, target_dp, | |
3041 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3042 | ||
3043 | if (new_parent && src_is_directory) { | |
3044 | error = xfs_bumplink(tp, target_dp); | |
3045 | if (error) | |
4906e215 | 3046 | goto out_bmap_cancel; |
f6bba201 DC |
3047 | } |
3048 | } else { /* target_ip != NULL */ | |
3049 | /* | |
3050 | * If target exists and it's a directory, check that both | |
3051 | * target and source are directories and that target can be | |
3052 | * destroyed, or that neither is a directory. | |
3053 | */ | |
c19b3b05 | 3054 | if (S_ISDIR(VFS_I(target_ip)->i_mode)) { |
f6bba201 DC |
3055 | /* |
3056 | * Make sure target dir is empty. | |
3057 | */ | |
3058 | if (!(xfs_dir_isempty(target_ip)) || | |
54d7b5c1 | 3059 | (VFS_I(target_ip)->i_nlink > 2)) { |
2451337d | 3060 | error = -EEXIST; |
445883e8 | 3061 | goto out_trans_cancel; |
f6bba201 DC |
3062 | } |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * Link the source inode under the target name. | |
3067 | * If the source inode is a directory and we are moving | |
3068 | * it across directories, its ".." entry will be | |
3069 | * inconsistent until we replace that down below. | |
3070 | * | |
3071 | * In case there is already an entry with the same | |
3072 | * name at the destination directory, remove it first. | |
3073 | */ | |
3074 | error = xfs_dir_replace(tp, target_dp, target_name, | |
3075 | src_ip->i_ino, | |
2c3234d1 | 3076 | &first_block, &dfops, spaceres); |
f6bba201 | 3077 | if (error) |
4906e215 | 3078 | goto out_bmap_cancel; |
f6bba201 DC |
3079 | |
3080 | xfs_trans_ichgtime(tp, target_dp, | |
3081 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3082 | ||
3083 | /* | |
3084 | * Decrement the link count on the target since the target | |
3085 | * dir no longer points to it. | |
3086 | */ | |
3087 | error = xfs_droplink(tp, target_ip); | |
3088 | if (error) | |
4906e215 | 3089 | goto out_bmap_cancel; |
f6bba201 DC |
3090 | |
3091 | if (src_is_directory) { | |
3092 | /* | |
3093 | * Drop the link from the old "." entry. | |
3094 | */ | |
3095 | error = xfs_droplink(tp, target_ip); | |
3096 | if (error) | |
4906e215 | 3097 | goto out_bmap_cancel; |
f6bba201 DC |
3098 | } |
3099 | } /* target_ip != NULL */ | |
3100 | ||
3101 | /* | |
3102 | * Remove the source. | |
3103 | */ | |
3104 | if (new_parent && src_is_directory) { | |
3105 | /* | |
3106 | * Rewrite the ".." entry to point to the new | |
3107 | * directory. | |
3108 | */ | |
3109 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
3110 | target_dp->i_ino, | |
2c3234d1 | 3111 | &first_block, &dfops, spaceres); |
2451337d | 3112 | ASSERT(error != -EEXIST); |
f6bba201 | 3113 | if (error) |
4906e215 | 3114 | goto out_bmap_cancel; |
f6bba201 DC |
3115 | } |
3116 | ||
3117 | /* | |
3118 | * We always want to hit the ctime on the source inode. | |
3119 | * | |
3120 | * This isn't strictly required by the standards since the source | |
3121 | * inode isn't really being changed, but old unix file systems did | |
3122 | * it and some incremental backup programs won't work without it. | |
3123 | */ | |
3124 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3125 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3126 | ||
3127 | /* | |
3128 | * Adjust the link count on src_dp. This is necessary when | |
3129 | * renaming a directory, either within one parent when | |
3130 | * the target existed, or across two parent directories. | |
3131 | */ | |
3132 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3133 | ||
3134 | /* | |
3135 | * Decrement link count on src_directory since the | |
3136 | * entry that's moved no longer points to it. | |
3137 | */ | |
3138 | error = xfs_droplink(tp, src_dp); | |
3139 | if (error) | |
4906e215 | 3140 | goto out_bmap_cancel; |
f6bba201 DC |
3141 | } |
3142 | ||
7dcf5c3e DC |
3143 | /* |
3144 | * For whiteouts, we only need to update the source dirent with the | |
3145 | * inode number of the whiteout inode rather than removing it | |
3146 | * altogether. | |
3147 | */ | |
3148 | if (wip) { | |
3149 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, | |
2c3234d1 | 3150 | &first_block, &dfops, spaceres); |
7dcf5c3e DC |
3151 | } else |
3152 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
2c3234d1 | 3153 | &first_block, &dfops, spaceres); |
f6bba201 | 3154 | if (error) |
4906e215 | 3155 | goto out_bmap_cancel; |
f6bba201 DC |
3156 | |
3157 | /* | |
7dcf5c3e DC |
3158 | * For whiteouts, we need to bump the link count on the whiteout inode. |
3159 | * This means that failures all the way up to this point leave the inode | |
3160 | * on the unlinked list and so cleanup is a simple matter of dropping | |
3161 | * the remaining reference to it. If we fail here after bumping the link | |
3162 | * count, we're shutting down the filesystem so we'll never see the | |
3163 | * intermediate state on disk. | |
f6bba201 | 3164 | */ |
7dcf5c3e | 3165 | if (wip) { |
54d7b5c1 | 3166 | ASSERT(VFS_I(wip)->i_nlink == 0); |
7dcf5c3e DC |
3167 | error = xfs_bumplink(tp, wip); |
3168 | if (error) | |
4906e215 | 3169 | goto out_bmap_cancel; |
7dcf5c3e DC |
3170 | error = xfs_iunlink_remove(tp, wip); |
3171 | if (error) | |
4906e215 | 3172 | goto out_bmap_cancel; |
7dcf5c3e | 3173 | xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); |
f6bba201 | 3174 | |
7dcf5c3e DC |
3175 | /* |
3176 | * Now we have a real link, clear the "I'm a tmpfile" state | |
3177 | * flag from the inode so it doesn't accidentally get misused in | |
3178 | * future. | |
3179 | */ | |
3180 | VFS_I(wip)->i_state &= ~I_LINKABLE; | |
f6bba201 DC |
3181 | } |
3182 | ||
f6bba201 DC |
3183 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3184 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3185 | if (new_parent) | |
3186 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3187 | |
2c3234d1 | 3188 | error = xfs_finish_rename(tp, &dfops); |
7dcf5c3e DC |
3189 | if (wip) |
3190 | IRELE(wip); | |
3191 | return error; | |
f6bba201 | 3192 | |
445883e8 | 3193 | out_bmap_cancel: |
2c3234d1 | 3194 | xfs_defer_cancel(&dfops); |
445883e8 | 3195 | out_trans_cancel: |
4906e215 | 3196 | xfs_trans_cancel(tp); |
253f4911 | 3197 | out_release_wip: |
7dcf5c3e DC |
3198 | if (wip) |
3199 | IRELE(wip); | |
f6bba201 DC |
3200 | return error; |
3201 | } | |
3202 | ||
5c4d97d0 DC |
3203 | STATIC int |
3204 | xfs_iflush_cluster( | |
19429363 DC |
3205 | struct xfs_inode *ip, |
3206 | struct xfs_buf *bp) | |
1da177e4 | 3207 | { |
19429363 | 3208 | struct xfs_mount *mp = ip->i_mount; |
5c4d97d0 DC |
3209 | struct xfs_perag *pag; |
3210 | unsigned long first_index, mask; | |
3211 | unsigned long inodes_per_cluster; | |
19429363 DC |
3212 | int cilist_size; |
3213 | struct xfs_inode **cilist; | |
3214 | struct xfs_inode *cip; | |
5c4d97d0 DC |
3215 | int nr_found; |
3216 | int clcount = 0; | |
3217 | int bufwasdelwri; | |
1da177e4 | 3218 | int i; |
1da177e4 | 3219 | |
5c4d97d0 | 3220 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 3221 | |
0f49efd8 | 3222 | inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; |
19429363 DC |
3223 | cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
3224 | cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); | |
3225 | if (!cilist) | |
5c4d97d0 | 3226 | goto out_put; |
1da177e4 | 3227 | |
0f49efd8 | 3228 | mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); |
5c4d97d0 DC |
3229 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
3230 | rcu_read_lock(); | |
3231 | /* really need a gang lookup range call here */ | |
19429363 | 3232 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, |
5c4d97d0 DC |
3233 | first_index, inodes_per_cluster); |
3234 | if (nr_found == 0) | |
3235 | goto out_free; | |
3236 | ||
3237 | for (i = 0; i < nr_found; i++) { | |
19429363 DC |
3238 | cip = cilist[i]; |
3239 | if (cip == ip) | |
bad55843 | 3240 | continue; |
1a3e8f3d DC |
3241 | |
3242 | /* | |
3243 | * because this is an RCU protected lookup, we could find a | |
3244 | * recently freed or even reallocated inode during the lookup. | |
3245 | * We need to check under the i_flags_lock for a valid inode | |
3246 | * here. Skip it if it is not valid or the wrong inode. | |
3247 | */ | |
19429363 DC |
3248 | spin_lock(&cip->i_flags_lock); |
3249 | if (!cip->i_ino || | |
3250 | __xfs_iflags_test(cip, XFS_ISTALE)) { | |
3251 | spin_unlock(&cip->i_flags_lock); | |
1a3e8f3d DC |
3252 | continue; |
3253 | } | |
5a90e53e DC |
3254 | |
3255 | /* | |
3256 | * Once we fall off the end of the cluster, no point checking | |
3257 | * any more inodes in the list because they will also all be | |
3258 | * outside the cluster. | |
3259 | */ | |
19429363 DC |
3260 | if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { |
3261 | spin_unlock(&cip->i_flags_lock); | |
5a90e53e DC |
3262 | break; |
3263 | } | |
19429363 | 3264 | spin_unlock(&cip->i_flags_lock); |
1a3e8f3d | 3265 | |
bad55843 DC |
3266 | /* |
3267 | * Do an un-protected check to see if the inode is dirty and | |
3268 | * is a candidate for flushing. These checks will be repeated | |
3269 | * later after the appropriate locks are acquired. | |
3270 | */ | |
19429363 | 3271 | if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) |
bad55843 | 3272 | continue; |
bad55843 DC |
3273 | |
3274 | /* | |
3275 | * Try to get locks. If any are unavailable or it is pinned, | |
3276 | * then this inode cannot be flushed and is skipped. | |
3277 | */ | |
3278 | ||
19429363 | 3279 | if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) |
bad55843 | 3280 | continue; |
19429363 DC |
3281 | if (!xfs_iflock_nowait(cip)) { |
3282 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3283 | continue; |
3284 | } | |
19429363 DC |
3285 | if (xfs_ipincount(cip)) { |
3286 | xfs_ifunlock(cip); | |
3287 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3288 | continue; |
3289 | } | |
3290 | ||
8a17d7dd DC |
3291 | |
3292 | /* | |
3293 | * Check the inode number again, just to be certain we are not | |
3294 | * racing with freeing in xfs_reclaim_inode(). See the comments | |
3295 | * in that function for more information as to why the initial | |
3296 | * check is not sufficient. | |
3297 | */ | |
19429363 DC |
3298 | if (!cip->i_ino) { |
3299 | xfs_ifunlock(cip); | |
3300 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3301 | continue; |
3302 | } | |
3303 | ||
3304 | /* | |
3305 | * arriving here means that this inode can be flushed. First | |
3306 | * re-check that it's dirty before flushing. | |
3307 | */ | |
19429363 | 3308 | if (!xfs_inode_clean(cip)) { |
33540408 | 3309 | int error; |
19429363 | 3310 | error = xfs_iflush_int(cip, bp); |
bad55843 | 3311 | if (error) { |
19429363 | 3312 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3313 | goto cluster_corrupt_out; |
3314 | } | |
3315 | clcount++; | |
3316 | } else { | |
19429363 | 3317 | xfs_ifunlock(cip); |
bad55843 | 3318 | } |
19429363 | 3319 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3320 | } |
3321 | ||
3322 | if (clcount) { | |
ff6d6af2 BD |
3323 | XFS_STATS_INC(mp, xs_icluster_flushcnt); |
3324 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
bad55843 DC |
3325 | } |
3326 | ||
3327 | out_free: | |
1a3e8f3d | 3328 | rcu_read_unlock(); |
19429363 | 3329 | kmem_free(cilist); |
44b56e0a DC |
3330 | out_put: |
3331 | xfs_perag_put(pag); | |
bad55843 DC |
3332 | return 0; |
3333 | ||
3334 | ||
3335 | cluster_corrupt_out: | |
3336 | /* | |
3337 | * Corruption detected in the clustering loop. Invalidate the | |
3338 | * inode buffer and shut down the filesystem. | |
3339 | */ | |
1a3e8f3d | 3340 | rcu_read_unlock(); |
bad55843 | 3341 | /* |
43ff2122 | 3342 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
3343 | * brelse can handle it with no problems. If not, shut down the |
3344 | * filesystem before releasing the buffer. | |
3345 | */ | |
43ff2122 | 3346 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
3347 | if (bufwasdelwri) |
3348 | xfs_buf_relse(bp); | |
3349 | ||
3350 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3351 | ||
3352 | if (!bufwasdelwri) { | |
3353 | /* | |
3354 | * Just like incore_relse: if we have b_iodone functions, | |
3355 | * mark the buffer as an error and call them. Otherwise | |
3356 | * mark it as stale and brelse. | |
3357 | */ | |
cb669ca5 | 3358 | if (bp->b_iodone) { |
b0388bf1 | 3359 | bp->b_flags &= ~XBF_DONE; |
c867cb61 | 3360 | xfs_buf_stale(bp); |
2451337d | 3361 | xfs_buf_ioerror(bp, -EIO); |
e8aaba9a | 3362 | xfs_buf_ioend(bp); |
bad55843 | 3363 | } else { |
c867cb61 | 3364 | xfs_buf_stale(bp); |
bad55843 DC |
3365 | xfs_buf_relse(bp); |
3366 | } | |
3367 | } | |
3368 | ||
3369 | /* | |
3370 | * Unlocks the flush lock | |
3371 | */ | |
19429363 DC |
3372 | xfs_iflush_abort(cip, false); |
3373 | kmem_free(cilist); | |
44b56e0a | 3374 | xfs_perag_put(pag); |
2451337d | 3375 | return -EFSCORRUPTED; |
bad55843 DC |
3376 | } |
3377 | ||
1da177e4 | 3378 | /* |
4c46819a CH |
3379 | * Flush dirty inode metadata into the backing buffer. |
3380 | * | |
3381 | * The caller must have the inode lock and the inode flush lock held. The | |
3382 | * inode lock will still be held upon return to the caller, and the inode | |
3383 | * flush lock will be released after the inode has reached the disk. | |
3384 | * | |
3385 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3386 | */ |
3387 | int | |
3388 | xfs_iflush( | |
4c46819a CH |
3389 | struct xfs_inode *ip, |
3390 | struct xfs_buf **bpp) | |
1da177e4 | 3391 | { |
4c46819a | 3392 | struct xfs_mount *mp = ip->i_mount; |
b1438f47 | 3393 | struct xfs_buf *bp = NULL; |
4c46819a | 3394 | struct xfs_dinode *dip; |
1da177e4 | 3395 | int error; |
1da177e4 | 3396 | |
ff6d6af2 | 3397 | XFS_STATS_INC(mp, xs_iflush_count); |
1da177e4 | 3398 | |
579aa9ca | 3399 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3400 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3401 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3402 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3403 | |
4c46819a | 3404 | *bpp = NULL; |
1da177e4 | 3405 | |
1da177e4 LT |
3406 | xfs_iunpin_wait(ip); |
3407 | ||
4b6a4688 DC |
3408 | /* |
3409 | * For stale inodes we cannot rely on the backing buffer remaining | |
3410 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3411 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3412 | * inodes below. We have to check this after ensuring the inode is |
3413 | * unpinned so that it is safe to reclaim the stale inode after the | |
3414 | * flush call. | |
3415 | */ | |
3416 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3417 | xfs_ifunlock(ip); | |
3418 | return 0; | |
3419 | } | |
3420 | ||
1da177e4 LT |
3421 | /* |
3422 | * This may have been unpinned because the filesystem is shutting | |
3423 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3424 | * to disk, because the log record didn't make it to disk. |
3425 | * | |
3426 | * We also have to remove the log item from the AIL in this case, | |
3427 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3428 | */ |
3429 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
2451337d | 3430 | error = -EIO; |
32ce90a4 | 3431 | goto abort_out; |
1da177e4 LT |
3432 | } |
3433 | ||
a3f74ffb | 3434 | /* |
b1438f47 DC |
3435 | * Get the buffer containing the on-disk inode. We are doing a try-lock |
3436 | * operation here, so we may get an EAGAIN error. In that case, we | |
3437 | * simply want to return with the inode still dirty. | |
3438 | * | |
3439 | * If we get any other error, we effectively have a corruption situation | |
3440 | * and we cannot flush the inode, so we treat it the same as failing | |
3441 | * xfs_iflush_int(). | |
a3f74ffb | 3442 | */ |
475ee413 CH |
3443 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3444 | 0); | |
b1438f47 | 3445 | if (error == -EAGAIN) { |
a3f74ffb DC |
3446 | xfs_ifunlock(ip); |
3447 | return error; | |
3448 | } | |
b1438f47 DC |
3449 | if (error) |
3450 | goto corrupt_out; | |
a3f74ffb | 3451 | |
1da177e4 LT |
3452 | /* |
3453 | * First flush out the inode that xfs_iflush was called with. | |
3454 | */ | |
3455 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3456 | if (error) |
1da177e4 | 3457 | goto corrupt_out; |
1da177e4 | 3458 | |
a3f74ffb DC |
3459 | /* |
3460 | * If the buffer is pinned then push on the log now so we won't | |
3461 | * get stuck waiting in the write for too long. | |
3462 | */ | |
811e64c7 | 3463 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3464 | xfs_log_force(mp, 0); |
a3f74ffb | 3465 | |
1da177e4 LT |
3466 | /* |
3467 | * inode clustering: | |
3468 | * see if other inodes can be gathered into this write | |
3469 | */ | |
bad55843 DC |
3470 | error = xfs_iflush_cluster(ip, bp); |
3471 | if (error) | |
3472 | goto cluster_corrupt_out; | |
1da177e4 | 3473 | |
4c46819a CH |
3474 | *bpp = bp; |
3475 | return 0; | |
1da177e4 LT |
3476 | |
3477 | corrupt_out: | |
b1438f47 DC |
3478 | if (bp) |
3479 | xfs_buf_relse(bp); | |
7d04a335 | 3480 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3481 | cluster_corrupt_out: |
2451337d | 3482 | error = -EFSCORRUPTED; |
32ce90a4 | 3483 | abort_out: |
1da177e4 LT |
3484 | /* |
3485 | * Unlocks the flush lock | |
3486 | */ | |
04913fdd | 3487 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3488 | return error; |
1da177e4 LT |
3489 | } |
3490 | ||
1da177e4 LT |
3491 | STATIC int |
3492 | xfs_iflush_int( | |
93848a99 CH |
3493 | struct xfs_inode *ip, |
3494 | struct xfs_buf *bp) | |
1da177e4 | 3495 | { |
93848a99 CH |
3496 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3497 | struct xfs_dinode *dip; | |
3498 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3499 | |
579aa9ca | 3500 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3501 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3502 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3503 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3504 | ASSERT(iip != NULL && iip->ili_fields != 0); |
263997a6 | 3505 | ASSERT(ip->i_d.di_version > 1); |
1da177e4 | 3506 | |
1da177e4 | 3507 | /* set *dip = inode's place in the buffer */ |
88ee2df7 | 3508 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3509 | |
69ef921b | 3510 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
9e24cfd0 | 3511 | mp, XFS_ERRTAG_IFLUSH_1)) { |
6a19d939 DC |
3512 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3513 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3514 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
3515 | goto corrupt_out; |
3516 | } | |
c19b3b05 | 3517 | if (S_ISREG(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3518 | if (XFS_TEST_ERROR( |
3519 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3520 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 3521 | mp, XFS_ERRTAG_IFLUSH_3)) { |
6a19d939 DC |
3522 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3523 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
3524 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3525 | goto corrupt_out; |
3526 | } | |
c19b3b05 | 3527 | } else if (S_ISDIR(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3528 | if (XFS_TEST_ERROR( |
3529 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3530 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3531 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
9e24cfd0 | 3532 | mp, XFS_ERRTAG_IFLUSH_4)) { |
6a19d939 DC |
3533 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3534 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
3535 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3536 | goto corrupt_out; |
3537 | } | |
3538 | } | |
3539 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
9e24cfd0 | 3540 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { |
6a19d939 DC |
3541 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3542 | "%s: detected corrupt incore inode %Lu, " | |
3543 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3544 | __func__, ip->i_ino, | |
1da177e4 | 3545 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3546 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3547 | goto corrupt_out; |
3548 | } | |
3549 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
9e24cfd0 | 3550 | mp, XFS_ERRTAG_IFLUSH_6)) { |
6a19d939 DC |
3551 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3552 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3553 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
3554 | goto corrupt_out; |
3555 | } | |
e60896d8 | 3556 | |
1da177e4 | 3557 | /* |
263997a6 | 3558 | * Inode item log recovery for v2 inodes are dependent on the |
e60896d8 DC |
3559 | * di_flushiter count for correct sequencing. We bump the flush |
3560 | * iteration count so we can detect flushes which postdate a log record | |
3561 | * during recovery. This is redundant as we now log every change and | |
3562 | * hence this can't happen but we need to still do it to ensure | |
3563 | * backwards compatibility with old kernels that predate logging all | |
3564 | * inode changes. | |
1da177e4 | 3565 | */ |
e60896d8 DC |
3566 | if (ip->i_d.di_version < 3) |
3567 | ip->i_d.di_flushiter++; | |
1da177e4 | 3568 | |
005c5db8 DW |
3569 | /* Check the inline directory data. */ |
3570 | if (S_ISDIR(VFS_I(ip)->i_mode) && | |
3571 | ip->i_d.di_format == XFS_DINODE_FMT_LOCAL && | |
3572 | xfs_dir2_sf_verify(ip)) | |
3573 | goto corrupt_out; | |
3574 | ||
1da177e4 | 3575 | /* |
3987848c DC |
3576 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3577 | * copy out the core of the inode, because if the inode is dirty at all | |
3578 | * the core must be. | |
1da177e4 | 3579 | */ |
93f958f9 | 3580 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3581 | |
3582 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3583 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3584 | ip->i_d.di_flushiter = 0; | |
3585 | ||
005c5db8 DW |
3586 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
3587 | if (XFS_IFORK_Q(ip)) | |
3588 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); | |
1da177e4 LT |
3589 | xfs_inobp_check(mp, bp); |
3590 | ||
3591 | /* | |
f5d8d5c4 CH |
3592 | * We've recorded everything logged in the inode, so we'd like to clear |
3593 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3594 | * However, we can't stop logging all this information until the data | |
3595 | * we've copied into the disk buffer is written to disk. If we did we | |
3596 | * might overwrite the copy of the inode in the log with all the data | |
3597 | * after re-logging only part of it, and in the face of a crash we | |
3598 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3599 | * |
f5d8d5c4 CH |
3600 | * What we do is move the bits to the ili_last_fields field. When |
3601 | * logging the inode, these bits are moved back to the ili_fields field. | |
3602 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3603 | * know that the information those bits represent is permanently on | |
3604 | * disk. As long as the flush completes before the inode is logged | |
3605 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3606 | * |
f5d8d5c4 CH |
3607 | * We can play with the ili_fields bits here, because the inode lock |
3608 | * must be held exclusively in order to set bits there and the flush | |
3609 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3610 | * done routine can tell whether or not to look in the AIL. Also, store | |
3611 | * the current LSN of the inode so that we can tell whether the item has | |
3612 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3613 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3614 | * atomically. | |
1da177e4 | 3615 | */ |
93848a99 CH |
3616 | iip->ili_last_fields = iip->ili_fields; |
3617 | iip->ili_fields = 0; | |
fc0561ce | 3618 | iip->ili_fsync_fields = 0; |
93848a99 | 3619 | iip->ili_logged = 1; |
1da177e4 | 3620 | |
93848a99 CH |
3621 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3622 | &iip->ili_item.li_lsn); | |
1da177e4 | 3623 | |
93848a99 CH |
3624 | /* |
3625 | * Attach the function xfs_iflush_done to the inode's | |
3626 | * buffer. This will remove the inode from the AIL | |
3627 | * and unlock the inode's flush lock when the inode is | |
3628 | * completely written to disk. | |
3629 | */ | |
3630 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3631 | |
93848a99 CH |
3632 | /* generate the checksum. */ |
3633 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3634 | |
93848a99 CH |
3635 | ASSERT(bp->b_fspriv != NULL); |
3636 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
3637 | return 0; |
3638 | ||
3639 | corrupt_out: | |
2451337d | 3640 | return -EFSCORRUPTED; |
1da177e4 | 3641 | } |