]>
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" |
1da177e4 | 42 | #include "xfs_error.h" |
1da177e4 | 43 | #include "xfs_quota.h" |
2a82b8be | 44 | #include "xfs_filestream.h" |
93848a99 | 45 | #include "xfs_cksum.h" |
0b1b213f | 46 | #include "xfs_trace.h" |
33479e05 | 47 | #include "xfs_icache.h" |
c24b5dfa | 48 | #include "xfs_symlink.h" |
239880ef DC |
49 | #include "xfs_trans_priv.h" |
50 | #include "xfs_log.h" | |
a4fbe6ab | 51 | #include "xfs_bmap_btree.h" |
aa8968f2 | 52 | #include "xfs_reflink.h" |
005c5db8 | 53 | #include "xfs_dir2_priv.h" |
1da177e4 | 54 | |
1da177e4 | 55 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
56 | |
57 | /* | |
8f04c47a | 58 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
59 | * freed from a file in a single transaction. |
60 | */ | |
61 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
62 | ||
54d7b5c1 DC |
63 | STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); |
64 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); | |
65 | STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); | |
ab297431 | 66 | |
2a0ec1d9 DC |
67 | /* |
68 | * helper function to extract extent size hint from inode | |
69 | */ | |
70 | xfs_extlen_t | |
71 | xfs_get_extsz_hint( | |
72 | struct xfs_inode *ip) | |
73 | { | |
74 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
75 | return ip->i_d.di_extsize; | |
76 | if (XFS_IS_REALTIME_INODE(ip)) | |
77 | return ip->i_mount->m_sb.sb_rextsize; | |
78 | return 0; | |
79 | } | |
80 | ||
f7ca3522 DW |
81 | /* |
82 | * Helper function to extract CoW extent size hint from inode. | |
83 | * Between the extent size hint and the CoW extent size hint, we | |
e153aa79 DW |
84 | * return the greater of the two. If the value is zero (automatic), |
85 | * use the default size. | |
f7ca3522 DW |
86 | */ |
87 | xfs_extlen_t | |
88 | xfs_get_cowextsz_hint( | |
89 | struct xfs_inode *ip) | |
90 | { | |
91 | xfs_extlen_t a, b; | |
92 | ||
93 | a = 0; | |
94 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) | |
95 | a = ip->i_d.di_cowextsize; | |
96 | b = xfs_get_extsz_hint(ip); | |
97 | ||
e153aa79 DW |
98 | a = max(a, b); |
99 | if (a == 0) | |
100 | return XFS_DEFAULT_COWEXTSZ_HINT; | |
101 | return a; | |
f7ca3522 DW |
102 | } |
103 | ||
fa96acad | 104 | /* |
efa70be1 CH |
105 | * These two are wrapper routines around the xfs_ilock() routine used to |
106 | * centralize some grungy code. They are used in places that wish to lock the | |
107 | * inode solely for reading the extents. The reason these places can't just | |
108 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
109 | * bringing in of the extents from disk for a file in b-tree format. If the | |
110 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
111 | * the extents are read in. Locking it exclusively all the time would limit | |
112 | * our parallelism unnecessarily, though. What we do instead is check to see | |
113 | * if the extents have been read in yet, and only lock the inode exclusively | |
114 | * if they have not. | |
fa96acad | 115 | * |
efa70be1 | 116 | * The functions return a value which should be given to the corresponding |
01f4f327 | 117 | * xfs_iunlock() call. |
fa96acad DC |
118 | */ |
119 | uint | |
309ecac8 CH |
120 | xfs_ilock_data_map_shared( |
121 | struct xfs_inode *ip) | |
fa96acad | 122 | { |
309ecac8 | 123 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 124 | |
309ecac8 CH |
125 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && |
126 | (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) | |
fa96acad | 127 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 128 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
129 | return lock_mode; |
130 | } | |
131 | ||
efa70be1 CH |
132 | uint |
133 | xfs_ilock_attr_map_shared( | |
134 | struct xfs_inode *ip) | |
fa96acad | 135 | { |
efa70be1 CH |
136 | uint lock_mode = XFS_ILOCK_SHARED; |
137 | ||
138 | if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && | |
139 | (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) | |
140 | lock_mode = XFS_ILOCK_EXCL; | |
141 | xfs_ilock(ip, lock_mode); | |
142 | return lock_mode; | |
fa96acad DC |
143 | } |
144 | ||
145 | /* | |
65523218 CH |
146 | * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 |
147 | * multi-reader locks: i_mmap_lock and the i_lock. This routine allows | |
148 | * various combinations of the locks to be obtained. | |
fa96acad | 149 | * |
653c60b6 DC |
150 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
151 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 152 | * |
653c60b6 DC |
153 | * Basic locking order: |
154 | * | |
65523218 | 155 | * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock |
653c60b6 DC |
156 | * |
157 | * mmap_sem locking order: | |
158 | * | |
65523218 | 159 | * i_rwsem -> page lock -> mmap_sem |
653c60b6 DC |
160 | * mmap_sem -> i_mmap_lock -> page_lock |
161 | * | |
162 | * The difference in mmap_sem locking order mean that we cannot hold the | |
163 | * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can | |
164 | * fault in pages during copy in/out (for buffered IO) or require the mmap_sem | |
165 | * in get_user_pages() to map the user pages into the kernel address space for | |
65523218 | 166 | * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because |
653c60b6 DC |
167 | * page faults already hold the mmap_sem. |
168 | * | |
169 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
65523218 | 170 | * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both |
653c60b6 DC |
171 | * taken in places where we need to invalidate the page cache in a race |
172 | * free manner (e.g. truncate, hole punch and other extent manipulation | |
173 | * functions). | |
fa96acad DC |
174 | */ |
175 | void | |
176 | xfs_ilock( | |
177 | xfs_inode_t *ip, | |
178 | uint lock_flags) | |
179 | { | |
180 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
181 | ||
182 | /* | |
183 | * You can't set both SHARED and EXCL for the same lock, | |
184 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
185 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
186 | */ | |
187 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
188 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
189 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
190 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
191 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
192 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 193 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad | 194 | |
65523218 CH |
195 | if (lock_flags & XFS_IOLOCK_EXCL) { |
196 | down_write_nested(&VFS_I(ip)->i_rwsem, | |
197 | XFS_IOLOCK_DEP(lock_flags)); | |
198 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
199 | down_read_nested(&VFS_I(ip)->i_rwsem, | |
200 | XFS_IOLOCK_DEP(lock_flags)); | |
201 | } | |
fa96acad | 202 | |
653c60b6 DC |
203 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
204 | mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
205 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
206 | mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
207 | ||
fa96acad DC |
208 | if (lock_flags & XFS_ILOCK_EXCL) |
209 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
210 | else if (lock_flags & XFS_ILOCK_SHARED) | |
211 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
212 | } | |
213 | ||
214 | /* | |
215 | * This is just like xfs_ilock(), except that the caller | |
216 | * is guaranteed not to sleep. It returns 1 if it gets | |
217 | * the requested locks and 0 otherwise. If the IO lock is | |
218 | * obtained but the inode lock cannot be, then the IO lock | |
219 | * is dropped before returning. | |
220 | * | |
221 | * ip -- the inode being locked | |
222 | * lock_flags -- this parameter indicates the inode's locks to be | |
223 | * to be locked. See the comment for xfs_ilock() for a list | |
224 | * of valid values. | |
225 | */ | |
226 | int | |
227 | xfs_ilock_nowait( | |
228 | xfs_inode_t *ip, | |
229 | uint lock_flags) | |
230 | { | |
231 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
232 | ||
233 | /* | |
234 | * You can't set both SHARED and EXCL for the same lock, | |
235 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
236 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
237 | */ | |
238 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
239 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
240 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
241 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
242 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
243 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 244 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
245 | |
246 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
65523218 | 247 | if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
248 | goto out; |
249 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
65523218 | 250 | if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
251 | goto out; |
252 | } | |
653c60b6 DC |
253 | |
254 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
255 | if (!mrtryupdate(&ip->i_mmaplock)) | |
256 | goto out_undo_iolock; | |
257 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
258 | if (!mrtryaccess(&ip->i_mmaplock)) | |
259 | goto out_undo_iolock; | |
260 | } | |
261 | ||
fa96acad DC |
262 | if (lock_flags & XFS_ILOCK_EXCL) { |
263 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 264 | goto out_undo_mmaplock; |
fa96acad DC |
265 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
266 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 267 | goto out_undo_mmaplock; |
fa96acad DC |
268 | } |
269 | return 1; | |
270 | ||
653c60b6 DC |
271 | out_undo_mmaplock: |
272 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
273 | mrunlock_excl(&ip->i_mmaplock); | |
274 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
275 | mrunlock_shared(&ip->i_mmaplock); | |
276 | out_undo_iolock: | |
fa96acad | 277 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 278 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 279 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 280 | up_read(&VFS_I(ip)->i_rwsem); |
653c60b6 | 281 | out: |
fa96acad DC |
282 | return 0; |
283 | } | |
284 | ||
285 | /* | |
286 | * xfs_iunlock() is used to drop the inode locks acquired with | |
287 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
288 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
289 | * that we know which locks to drop. | |
290 | * | |
291 | * ip -- the inode being unlocked | |
292 | * lock_flags -- this parameter indicates the inode's locks to be | |
293 | * to be unlocked. See the comment for xfs_ilock() for a list | |
294 | * of valid values for this parameter. | |
295 | * | |
296 | */ | |
297 | void | |
298 | xfs_iunlock( | |
299 | xfs_inode_t *ip, | |
300 | uint lock_flags) | |
301 | { | |
302 | /* | |
303 | * You can't set both SHARED and EXCL for the same lock, | |
304 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
305 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
306 | */ | |
307 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
308 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
309 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
310 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
311 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
312 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 313 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
314 | ASSERT(lock_flags != 0); |
315 | ||
316 | if (lock_flags & XFS_IOLOCK_EXCL) | |
65523218 | 317 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 318 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 319 | up_read(&VFS_I(ip)->i_rwsem); |
fa96acad | 320 | |
653c60b6 DC |
321 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
322 | mrunlock_excl(&ip->i_mmaplock); | |
323 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
324 | mrunlock_shared(&ip->i_mmaplock); | |
325 | ||
fa96acad DC |
326 | if (lock_flags & XFS_ILOCK_EXCL) |
327 | mrunlock_excl(&ip->i_lock); | |
328 | else if (lock_flags & XFS_ILOCK_SHARED) | |
329 | mrunlock_shared(&ip->i_lock); | |
330 | ||
331 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
332 | } | |
333 | ||
334 | /* | |
335 | * give up write locks. the i/o lock cannot be held nested | |
336 | * if it is being demoted. | |
337 | */ | |
338 | void | |
339 | xfs_ilock_demote( | |
340 | xfs_inode_t *ip, | |
341 | uint lock_flags) | |
342 | { | |
653c60b6 DC |
343 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
344 | ASSERT((lock_flags & | |
345 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
346 | |
347 | if (lock_flags & XFS_ILOCK_EXCL) | |
348 | mrdemote(&ip->i_lock); | |
653c60b6 DC |
349 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
350 | mrdemote(&ip->i_mmaplock); | |
fa96acad | 351 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 352 | downgrade_write(&VFS_I(ip)->i_rwsem); |
fa96acad DC |
353 | |
354 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
355 | } | |
356 | ||
742ae1e3 | 357 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
358 | int |
359 | xfs_isilocked( | |
360 | xfs_inode_t *ip, | |
361 | uint lock_flags) | |
362 | { | |
363 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
364 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
365 | return !!ip->i_lock.mr_writer; | |
366 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
367 | } | |
368 | ||
653c60b6 DC |
369 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
370 | if (!(lock_flags & XFS_MMAPLOCK_SHARED)) | |
371 | return !!ip->i_mmaplock.mr_writer; | |
372 | return rwsem_is_locked(&ip->i_mmaplock.mr_lock); | |
373 | } | |
374 | ||
fa96acad DC |
375 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
376 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
65523218 CH |
377 | return !debug_locks || |
378 | lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); | |
379 | return rwsem_is_locked(&VFS_I(ip)->i_rwsem); | |
fa96acad DC |
380 | } |
381 | ||
382 | ASSERT(0); | |
383 | return 0; | |
384 | } | |
385 | #endif | |
386 | ||
c24b5dfa DC |
387 | #ifdef DEBUG |
388 | int xfs_locked_n; | |
389 | int xfs_small_retries; | |
390 | int xfs_middle_retries; | |
391 | int xfs_lots_retries; | |
392 | int xfs_lock_delays; | |
393 | #endif | |
394 | ||
b6a9947e DC |
395 | /* |
396 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
397 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
398 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
399 | * errors and warnings. | |
400 | */ | |
401 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
402 | static bool |
403 | xfs_lockdep_subclass_ok( | |
404 | int subclass) | |
405 | { | |
406 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
407 | } | |
408 | #else | |
409 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
410 | #endif | |
411 | ||
c24b5dfa | 412 | /* |
653c60b6 | 413 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
414 | * value. This can be called for any type of inode lock combination, including |
415 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
416 | * storage fields in the class mask we build. | |
c24b5dfa DC |
417 | */ |
418 | static inline int | |
419 | xfs_lock_inumorder(int lock_mode, int subclass) | |
420 | { | |
0952c818 DC |
421 | int class = 0; |
422 | ||
423 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
424 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 425 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 426 | |
653c60b6 | 427 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 428 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
0952c818 | 429 | class += subclass << XFS_IOLOCK_SHIFT; |
653c60b6 DC |
430 | } |
431 | ||
432 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
433 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
434 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
435 | } |
436 | ||
0952c818 DC |
437 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
438 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
439 | class += subclass << XFS_ILOCK_SHIFT; | |
440 | } | |
c24b5dfa | 441 | |
0952c818 | 442 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
443 | } |
444 | ||
445 | /* | |
95afcf5c DC |
446 | * The following routine will lock n inodes in exclusive mode. We assume the |
447 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 448 | * |
95afcf5c DC |
449 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
450 | * start waiting for another inode that is locked by a thread in a long running | |
451 | * transaction (such as truncate). This can result in deadlock since the long | |
452 | * running trans might need to wait for the inode we just locked in order to | |
453 | * push the tail and free space in the log. | |
0952c818 DC |
454 | * |
455 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
456 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
457 | * lock more than one at a time, lockdep will report false positives saying we | |
458 | * have violated locking orders. | |
c24b5dfa | 459 | */ |
0d5a75e9 | 460 | static void |
c24b5dfa DC |
461 | xfs_lock_inodes( |
462 | xfs_inode_t **ips, | |
463 | int inodes, | |
464 | uint lock_mode) | |
465 | { | |
466 | int attempts = 0, i, j, try_lock; | |
467 | xfs_log_item_t *lp; | |
468 | ||
0952c818 DC |
469 | /* |
470 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
471 | * support an arbitrary depth of locking here, but absolute limits on | |
472 | * inodes depend on the the type of locking and the limits placed by | |
473 | * lockdep annotations in xfs_lock_inumorder. These are all checked by | |
474 | * the asserts. | |
475 | */ | |
95afcf5c | 476 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
477 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
478 | XFS_ILOCK_EXCL)); | |
479 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
480 | XFS_ILOCK_SHARED))); | |
0952c818 DC |
481 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || |
482 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
483 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
484 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
485 | ||
486 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
487 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
488 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
489 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa DC |
490 | |
491 | try_lock = 0; | |
492 | i = 0; | |
c24b5dfa DC |
493 | again: |
494 | for (; i < inodes; i++) { | |
495 | ASSERT(ips[i]); | |
496 | ||
95afcf5c | 497 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
498 | continue; |
499 | ||
500 | /* | |
95afcf5c DC |
501 | * If try_lock is not set yet, make sure all locked inodes are |
502 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 503 | */ |
c24b5dfa DC |
504 | if (!try_lock) { |
505 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
506 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
95afcf5c | 507 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) |
c24b5dfa | 508 | try_lock++; |
c24b5dfa DC |
509 | } |
510 | } | |
511 | ||
512 | /* | |
513 | * If any of the previous locks we have locked is in the AIL, | |
514 | * we must TRY to get the second and subsequent locks. If | |
515 | * we can't get any, we must release all we have | |
516 | * and try again. | |
517 | */ | |
95afcf5c DC |
518 | if (!try_lock) { |
519 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
520 | continue; | |
521 | } | |
522 | ||
523 | /* try_lock means we have an inode locked that is in the AIL. */ | |
524 | ASSERT(i != 0); | |
525 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
526 | continue; | |
c24b5dfa | 527 | |
95afcf5c DC |
528 | /* |
529 | * Unlock all previous guys and try again. xfs_iunlock will try | |
530 | * to push the tail if the inode is in the AIL. | |
531 | */ | |
532 | attempts++; | |
533 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 534 | /* |
95afcf5c DC |
535 | * Check to see if we've already unlocked this one. Not |
536 | * the first one going back, and the inode ptr is the | |
537 | * same. | |
c24b5dfa | 538 | */ |
95afcf5c DC |
539 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
540 | continue; | |
c24b5dfa | 541 | |
95afcf5c DC |
542 | xfs_iunlock(ips[j], lock_mode); |
543 | } | |
c24b5dfa | 544 | |
95afcf5c DC |
545 | if ((attempts % 5) == 0) { |
546 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 547 | #ifdef DEBUG |
95afcf5c | 548 | xfs_lock_delays++; |
c24b5dfa | 549 | #endif |
c24b5dfa | 550 | } |
95afcf5c DC |
551 | i = 0; |
552 | try_lock = 0; | |
553 | goto again; | |
c24b5dfa DC |
554 | } |
555 | ||
556 | #ifdef DEBUG | |
557 | if (attempts) { | |
558 | if (attempts < 5) xfs_small_retries++; | |
559 | else if (attempts < 100) xfs_middle_retries++; | |
560 | else xfs_lots_retries++; | |
561 | } else { | |
562 | xfs_locked_n++; | |
563 | } | |
564 | #endif | |
565 | } | |
566 | ||
567 | /* | |
653c60b6 DC |
568 | * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - |
569 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
570 | * lock more than one at a time, lockdep will report false positives saying we | |
571 | * have violated locking orders. | |
c24b5dfa DC |
572 | */ |
573 | void | |
574 | xfs_lock_two_inodes( | |
575 | xfs_inode_t *ip0, | |
576 | xfs_inode_t *ip1, | |
577 | uint lock_mode) | |
578 | { | |
579 | xfs_inode_t *temp; | |
580 | int attempts = 0; | |
581 | xfs_log_item_t *lp; | |
582 | ||
65523218 CH |
583 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); |
584 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) | |
653c60b6 DC |
585 | ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); |
586 | ||
c24b5dfa DC |
587 | ASSERT(ip0->i_ino != ip1->i_ino); |
588 | ||
589 | if (ip0->i_ino > ip1->i_ino) { | |
590 | temp = ip0; | |
591 | ip0 = ip1; | |
592 | ip1 = temp; | |
593 | } | |
594 | ||
595 | again: | |
596 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
597 | ||
598 | /* | |
599 | * If the first lock we have locked is in the AIL, we must TRY to get | |
600 | * the second lock. If we can't get it, we must release the first one | |
601 | * and try again. | |
602 | */ | |
603 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
604 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
605 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
606 | xfs_iunlock(ip0, lock_mode); | |
607 | if ((++attempts % 5) == 0) | |
608 | delay(1); /* Don't just spin the CPU */ | |
609 | goto again; | |
610 | } | |
611 | } else { | |
612 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
613 | } | |
614 | } | |
615 | ||
616 | ||
fa96acad DC |
617 | void |
618 | __xfs_iflock( | |
619 | struct xfs_inode *ip) | |
620 | { | |
621 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
622 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
623 | ||
624 | do { | |
21417136 | 625 | prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
fa96acad DC |
626 | if (xfs_isiflocked(ip)) |
627 | io_schedule(); | |
628 | } while (!xfs_iflock_nowait(ip)); | |
629 | ||
21417136 | 630 | finish_wait(wq, &wait.wq_entry); |
fa96acad DC |
631 | } |
632 | ||
1da177e4 LT |
633 | STATIC uint |
634 | _xfs_dic2xflags( | |
c8ce540d | 635 | uint16_t di_flags, |
58f88ca2 DC |
636 | uint64_t di_flags2, |
637 | bool has_attr) | |
1da177e4 LT |
638 | { |
639 | uint flags = 0; | |
640 | ||
641 | if (di_flags & XFS_DIFLAG_ANY) { | |
642 | if (di_flags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 643 | flags |= FS_XFLAG_REALTIME; |
1da177e4 | 644 | if (di_flags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 645 | flags |= FS_XFLAG_PREALLOC; |
1da177e4 | 646 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 647 | flags |= FS_XFLAG_IMMUTABLE; |
1da177e4 | 648 | if (di_flags & XFS_DIFLAG_APPEND) |
e7b89481 | 649 | flags |= FS_XFLAG_APPEND; |
1da177e4 | 650 | if (di_flags & XFS_DIFLAG_SYNC) |
e7b89481 | 651 | flags |= FS_XFLAG_SYNC; |
1da177e4 | 652 | if (di_flags & XFS_DIFLAG_NOATIME) |
e7b89481 | 653 | flags |= FS_XFLAG_NOATIME; |
1da177e4 | 654 | if (di_flags & XFS_DIFLAG_NODUMP) |
e7b89481 | 655 | flags |= FS_XFLAG_NODUMP; |
1da177e4 | 656 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 657 | flags |= FS_XFLAG_RTINHERIT; |
1da177e4 | 658 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 659 | flags |= FS_XFLAG_PROJINHERIT; |
1da177e4 | 660 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 661 | flags |= FS_XFLAG_NOSYMLINKS; |
dd9f438e | 662 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 663 | flags |= FS_XFLAG_EXTSIZE; |
dd9f438e | 664 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 665 | flags |= FS_XFLAG_EXTSZINHERIT; |
d3446eac | 666 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 667 | flags |= FS_XFLAG_NODEFRAG; |
2a82b8be | 668 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 669 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
670 | } |
671 | ||
58f88ca2 DC |
672 | if (di_flags2 & XFS_DIFLAG2_ANY) { |
673 | if (di_flags2 & XFS_DIFLAG2_DAX) | |
674 | flags |= FS_XFLAG_DAX; | |
f7ca3522 DW |
675 | if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) |
676 | flags |= FS_XFLAG_COWEXTSIZE; | |
58f88ca2 DC |
677 | } |
678 | ||
679 | if (has_attr) | |
680 | flags |= FS_XFLAG_HASATTR; | |
681 | ||
1da177e4 LT |
682 | return flags; |
683 | } | |
684 | ||
685 | uint | |
686 | xfs_ip2xflags( | |
58f88ca2 | 687 | struct xfs_inode *ip) |
1da177e4 | 688 | { |
58f88ca2 | 689 | struct xfs_icdinode *dic = &ip->i_d; |
1da177e4 | 690 | |
58f88ca2 | 691 | return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); |
1da177e4 LT |
692 | } |
693 | ||
c24b5dfa DC |
694 | /* |
695 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
696 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
697 | * ci_name->name will point to a the actual name (caller must free) or | |
698 | * will be set to NULL if an exact match is found. | |
699 | */ | |
700 | int | |
701 | xfs_lookup( | |
702 | xfs_inode_t *dp, | |
703 | struct xfs_name *name, | |
704 | xfs_inode_t **ipp, | |
705 | struct xfs_name *ci_name) | |
706 | { | |
707 | xfs_ino_t inum; | |
708 | int error; | |
c24b5dfa DC |
709 | |
710 | trace_xfs_lookup(dp, name); | |
711 | ||
712 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
2451337d | 713 | return -EIO; |
c24b5dfa | 714 | |
c24b5dfa | 715 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 716 | if (error) |
dbad7c99 | 717 | goto out_unlock; |
c24b5dfa DC |
718 | |
719 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
720 | if (error) | |
721 | goto out_free_name; | |
722 | ||
723 | return 0; | |
724 | ||
725 | out_free_name: | |
726 | if (ci_name) | |
727 | kmem_free(ci_name->name); | |
dbad7c99 | 728 | out_unlock: |
c24b5dfa DC |
729 | *ipp = NULL; |
730 | return error; | |
731 | } | |
732 | ||
1da177e4 LT |
733 | /* |
734 | * Allocate an inode on disk and return a copy of its in-core version. | |
735 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
736 | * appropriately within the inode. The uid and gid for the inode are | |
737 | * set according to the contents of the given cred structure. | |
738 | * | |
739 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
740 | * has a free inode available, call xfs_iget() to obtain the in-core |
741 | * version of the allocated inode. Finally, fill in the inode and | |
742 | * log its initial contents. In this case, ialloc_context would be | |
743 | * set to NULL. | |
1da177e4 | 744 | * |
cd856db6 CM |
745 | * If xfs_dialloc() does not have an available inode, it will replenish |
746 | * its supply by doing an allocation. Since we can only do one | |
747 | * allocation within a transaction without deadlocks, we must commit | |
748 | * the current transaction before returning the inode itself. | |
749 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
750 | * The caller should then commit the current transaction, start a new |
751 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
752 | * | |
753 | * To ensure that some other process does not grab the inode that | |
754 | * was allocated during the first call to xfs_ialloc(), this routine | |
755 | * also returns the [locked] bp pointing to the head of the freelist | |
756 | * as ialloc_context. The caller should hold this buffer across | |
757 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
758 | * |
759 | * If we are allocating quota inodes, we do not have a parent inode | |
760 | * to attach to or associate with (i.e. pip == NULL) because they | |
761 | * are not linked into the directory structure - they are attached | |
762 | * directly to the superblock - and so have no parent. | |
1da177e4 | 763 | */ |
0d5a75e9 | 764 | static int |
1da177e4 LT |
765 | xfs_ialloc( |
766 | xfs_trans_t *tp, | |
767 | xfs_inode_t *pip, | |
576b1d67 | 768 | umode_t mode, |
31b084ae | 769 | xfs_nlink_t nlink, |
66f36464 | 770 | dev_t rdev, |
6743099c | 771 | prid_t prid, |
1da177e4 LT |
772 | int okalloc, |
773 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
774 | xfs_inode_t **ipp) |
775 | { | |
93848a99 | 776 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
777 | xfs_ino_t ino; |
778 | xfs_inode_t *ip; | |
1da177e4 LT |
779 | uint flags; |
780 | int error; | |
e076b0f3 | 781 | struct timespec tv; |
3987848c | 782 | struct inode *inode; |
1da177e4 LT |
783 | |
784 | /* | |
785 | * Call the space management code to pick | |
786 | * the on-disk inode to be allocated. | |
787 | */ | |
b11f94d5 | 788 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 789 | ialloc_context, &ino); |
bf904248 | 790 | if (error) |
1da177e4 | 791 | return error; |
08358906 | 792 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
793 | *ipp = NULL; |
794 | return 0; | |
795 | } | |
796 | ASSERT(*ialloc_context == NULL); | |
797 | ||
798 | /* | |
799 | * Get the in-core inode with the lock held exclusively. | |
800 | * This is because we're setting fields here we need | |
801 | * to prevent others from looking at until we're done. | |
802 | */ | |
93848a99 | 803 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 804 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 805 | if (error) |
1da177e4 | 806 | return error; |
1da177e4 | 807 | ASSERT(ip != NULL); |
3987848c | 808 | inode = VFS_I(ip); |
1da177e4 | 809 | |
263997a6 DC |
810 | /* |
811 | * We always convert v1 inodes to v2 now - we only support filesystems | |
812 | * with >= v2 inode capability, so there is no reason for ever leaving | |
813 | * an inode in v1 format. | |
814 | */ | |
815 | if (ip->i_d.di_version == 1) | |
816 | ip->i_d.di_version = 2; | |
817 | ||
c19b3b05 | 818 | inode->i_mode = mode; |
54d7b5c1 | 819 | set_nlink(inode, nlink); |
7aab1b28 DE |
820 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
821 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
66f36464 | 822 | inode->i_rdev = rdev; |
6743099c | 823 | xfs_set_projid(ip, prid); |
1da177e4 | 824 | |
bd186aa9 | 825 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 826 | ip->i_d.di_gid = pip->i_d.di_gid; |
c19b3b05 DC |
827 | if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) |
828 | inode->i_mode |= S_ISGID; | |
1da177e4 LT |
829 | } |
830 | ||
831 | /* | |
832 | * If the group ID of the new file does not match the effective group | |
833 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
834 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
835 | */ | |
836 | if ((irix_sgid_inherit) && | |
c19b3b05 DC |
837 | (inode->i_mode & S_ISGID) && |
838 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) | |
839 | inode->i_mode &= ~S_ISGID; | |
1da177e4 LT |
840 | |
841 | ip->i_d.di_size = 0; | |
842 | ip->i_d.di_nextents = 0; | |
843 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 | 844 | |
c2050a45 | 845 | tv = current_time(inode); |
3987848c DC |
846 | inode->i_mtime = tv; |
847 | inode->i_atime = tv; | |
848 | inode->i_ctime = tv; | |
dff35fd4 | 849 | |
1da177e4 LT |
850 | ip->i_d.di_extsize = 0; |
851 | ip->i_d.di_dmevmask = 0; | |
852 | ip->i_d.di_dmstate = 0; | |
853 | ip->i_d.di_flags = 0; | |
93848a99 CH |
854 | |
855 | if (ip->i_d.di_version == 3) { | |
83e06f21 | 856 | inode->i_version = 1; |
93848a99 | 857 | ip->i_d.di_flags2 = 0; |
f7ca3522 | 858 | ip->i_d.di_cowextsize = 0; |
c8ce540d DW |
859 | ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; |
860 | ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; | |
93848a99 CH |
861 | } |
862 | ||
863 | ||
1da177e4 LT |
864 | flags = XFS_ILOG_CORE; |
865 | switch (mode & S_IFMT) { | |
866 | case S_IFIFO: | |
867 | case S_IFCHR: | |
868 | case S_IFBLK: | |
869 | case S_IFSOCK: | |
870 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1da177e4 LT |
871 | ip->i_df.if_flags = 0; |
872 | flags |= XFS_ILOG_DEV; | |
873 | break; | |
874 | case S_IFREG: | |
875 | case S_IFDIR: | |
b11f94d5 | 876 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
58f88ca2 | 877 | uint di_flags = 0; |
365ca83d | 878 | |
abbede1b | 879 | if (S_ISDIR(mode)) { |
365ca83d NS |
880 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
881 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
882 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
883 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
884 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
885 | } | |
9336e3a7 DC |
886 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
887 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
abbede1b | 888 | } else if (S_ISREG(mode)) { |
613d7043 | 889 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 890 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
891 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
892 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
893 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
894 | } | |
1da177e4 LT |
895 | } |
896 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
897 | xfs_inherit_noatime) | |
365ca83d | 898 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
899 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
900 | xfs_inherit_nodump) | |
365ca83d | 901 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
902 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
903 | xfs_inherit_sync) | |
365ca83d | 904 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
905 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
906 | xfs_inherit_nosymlinks) | |
365ca83d | 907 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
d3446eac BN |
908 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
909 | xfs_inherit_nodefrag) | |
910 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
911 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
912 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
58f88ca2 | 913 | |
365ca83d | 914 | ip->i_d.di_flags |= di_flags; |
1da177e4 | 915 | } |
f7ca3522 DW |
916 | if (pip && |
917 | (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && | |
918 | pip->i_d.di_version == 3 && | |
919 | ip->i_d.di_version == 3) { | |
56bdf855 LC |
920 | uint64_t di_flags2 = 0; |
921 | ||
f7ca3522 | 922 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { |
56bdf855 | 923 | di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; |
f7ca3522 DW |
924 | ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; |
925 | } | |
56bdf855 LC |
926 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) |
927 | di_flags2 |= XFS_DIFLAG2_DAX; | |
928 | ||
929 | ip->i_d.di_flags2 |= di_flags2; | |
f7ca3522 | 930 | } |
1da177e4 LT |
931 | /* FALLTHROUGH */ |
932 | case S_IFLNK: | |
933 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
934 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
935 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
936 | ip->i_df.if_u1.if_extents = NULL; | |
937 | break; | |
938 | default: | |
939 | ASSERT(0); | |
940 | } | |
941 | /* | |
942 | * Attribute fork settings for new inode. | |
943 | */ | |
944 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
945 | ip->i_d.di_anextents = 0; | |
946 | ||
947 | /* | |
948 | * Log the new values stuffed into the inode. | |
949 | */ | |
ddc3415a | 950 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
951 | xfs_trans_log_inode(tp, ip, flags); |
952 | ||
58c90473 | 953 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 954 | xfs_setup_inode(ip); |
1da177e4 LT |
955 | |
956 | *ipp = ip; | |
957 | return 0; | |
958 | } | |
959 | ||
e546cb79 DC |
960 | /* |
961 | * Allocates a new inode from disk and return a pointer to the | |
962 | * incore copy. This routine will internally commit the current | |
963 | * transaction and allocate a new one if the Space Manager needed | |
964 | * to do an allocation to replenish the inode free-list. | |
965 | * | |
966 | * This routine is designed to be called from xfs_create and | |
967 | * xfs_create_dir. | |
968 | * | |
969 | */ | |
970 | int | |
971 | xfs_dir_ialloc( | |
972 | xfs_trans_t **tpp, /* input: current transaction; | |
973 | output: may be a new transaction. */ | |
974 | xfs_inode_t *dp, /* directory within whose allocate | |
975 | the inode. */ | |
976 | umode_t mode, | |
977 | xfs_nlink_t nlink, | |
66f36464 | 978 | dev_t rdev, |
e546cb79 DC |
979 | prid_t prid, /* project id */ |
980 | int okalloc, /* ok to allocate new space */ | |
981 | xfs_inode_t **ipp, /* pointer to inode; it will be | |
982 | locked. */ | |
983 | int *committed) | |
984 | ||
985 | { | |
986 | xfs_trans_t *tp; | |
e546cb79 DC |
987 | xfs_inode_t *ip; |
988 | xfs_buf_t *ialloc_context = NULL; | |
989 | int code; | |
e546cb79 DC |
990 | void *dqinfo; |
991 | uint tflags; | |
992 | ||
993 | tp = *tpp; | |
994 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
995 | ||
996 | /* | |
997 | * xfs_ialloc will return a pointer to an incore inode if | |
998 | * the Space Manager has an available inode on the free | |
999 | * list. Otherwise, it will do an allocation and replenish | |
1000 | * the freelist. Since we can only do one allocation per | |
1001 | * transaction without deadlocks, we will need to commit the | |
1002 | * current transaction and start a new one. We will then | |
1003 | * need to call xfs_ialloc again to get the inode. | |
1004 | * | |
1005 | * If xfs_ialloc did an allocation to replenish the freelist, | |
1006 | * it returns the bp containing the head of the freelist as | |
1007 | * ialloc_context. We will hold a lock on it across the | |
1008 | * transaction commit so that no other process can steal | |
1009 | * the inode(s) that we've just allocated. | |
1010 | */ | |
1011 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, | |
1012 | &ialloc_context, &ip); | |
1013 | ||
1014 | /* | |
1015 | * Return an error if we were unable to allocate a new inode. | |
1016 | * This should only happen if we run out of space on disk or | |
1017 | * encounter a disk error. | |
1018 | */ | |
1019 | if (code) { | |
1020 | *ipp = NULL; | |
1021 | return code; | |
1022 | } | |
1023 | if (!ialloc_context && !ip) { | |
1024 | *ipp = NULL; | |
2451337d | 1025 | return -ENOSPC; |
e546cb79 DC |
1026 | } |
1027 | ||
1028 | /* | |
1029 | * If the AGI buffer is non-NULL, then we were unable to get an | |
1030 | * inode in one operation. We need to commit the current | |
1031 | * transaction and call xfs_ialloc() again. It is guaranteed | |
1032 | * to succeed the second time. | |
1033 | */ | |
1034 | if (ialloc_context) { | |
1035 | /* | |
1036 | * Normally, xfs_trans_commit releases all the locks. | |
1037 | * We call bhold to hang on to the ialloc_context across | |
1038 | * the commit. Holding this buffer prevents any other | |
1039 | * processes from doing any allocations in this | |
1040 | * allocation group. | |
1041 | */ | |
1042 | xfs_trans_bhold(tp, ialloc_context); | |
e546cb79 DC |
1043 | |
1044 | /* | |
1045 | * We want the quota changes to be associated with the next | |
1046 | * transaction, NOT this one. So, detach the dqinfo from this | |
1047 | * and attach it to the next transaction. | |
1048 | */ | |
1049 | dqinfo = NULL; | |
1050 | tflags = 0; | |
1051 | if (tp->t_dqinfo) { | |
1052 | dqinfo = (void *)tp->t_dqinfo; | |
1053 | tp->t_dqinfo = NULL; | |
1054 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; | |
1055 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); | |
1056 | } | |
1057 | ||
411350df | 1058 | code = xfs_trans_roll(&tp); |
2e6db6c4 | 1059 | if (committed != NULL) |
e546cb79 | 1060 | *committed = 1; |
3d3c8b52 | 1061 | |
e546cb79 DC |
1062 | /* |
1063 | * Re-attach the quota info that we detached from prev trx. | |
1064 | */ | |
1065 | if (dqinfo) { | |
1066 | tp->t_dqinfo = dqinfo; | |
1067 | tp->t_flags |= tflags; | |
1068 | } | |
1069 | ||
1070 | if (code) { | |
1071 | xfs_buf_relse(ialloc_context); | |
2e6db6c4 | 1072 | *tpp = tp; |
e546cb79 DC |
1073 | *ipp = NULL; |
1074 | return code; | |
1075 | } | |
1076 | xfs_trans_bjoin(tp, ialloc_context); | |
1077 | ||
1078 | /* | |
1079 | * Call ialloc again. Since we've locked out all | |
1080 | * other allocations in this allocation group, | |
1081 | * this call should always succeed. | |
1082 | */ | |
1083 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, | |
1084 | okalloc, &ialloc_context, &ip); | |
1085 | ||
1086 | /* | |
1087 | * If we get an error at this point, return to the caller | |
1088 | * so that the current transaction can be aborted. | |
1089 | */ | |
1090 | if (code) { | |
1091 | *tpp = tp; | |
1092 | *ipp = NULL; | |
1093 | return code; | |
1094 | } | |
1095 | ASSERT(!ialloc_context && ip); | |
1096 | ||
1097 | } else { | |
1098 | if (committed != NULL) | |
1099 | *committed = 0; | |
1100 | } | |
1101 | ||
1102 | *ipp = ip; | |
1103 | *tpp = tp; | |
1104 | ||
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /* | |
54d7b5c1 DC |
1109 | * Decrement the link count on an inode & log the change. If this causes the |
1110 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
1111 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 | 1112 | */ |
0d5a75e9 | 1113 | static int /* error */ |
e546cb79 DC |
1114 | xfs_droplink( |
1115 | xfs_trans_t *tp, | |
1116 | xfs_inode_t *ip) | |
1117 | { | |
e546cb79 DC |
1118 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
1119 | ||
e546cb79 DC |
1120 | drop_nlink(VFS_I(ip)); |
1121 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1122 | ||
54d7b5c1 DC |
1123 | if (VFS_I(ip)->i_nlink) |
1124 | return 0; | |
1125 | ||
1126 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
1127 | } |
1128 | ||
e546cb79 DC |
1129 | /* |
1130 | * Increment the link count on an inode & log the change. | |
1131 | */ | |
0d5a75e9 | 1132 | static int |
e546cb79 DC |
1133 | xfs_bumplink( |
1134 | xfs_trans_t *tp, | |
1135 | xfs_inode_t *ip) | |
1136 | { | |
1137 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1138 | ||
263997a6 | 1139 | ASSERT(ip->i_d.di_version > 1); |
e546cb79 | 1140 | inc_nlink(VFS_I(ip)); |
e546cb79 DC |
1141 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1142 | return 0; | |
1143 | } | |
1144 | ||
c24b5dfa DC |
1145 | int |
1146 | xfs_create( | |
1147 | xfs_inode_t *dp, | |
1148 | struct xfs_name *name, | |
1149 | umode_t mode, | |
66f36464 | 1150 | dev_t rdev, |
c24b5dfa DC |
1151 | xfs_inode_t **ipp) |
1152 | { | |
1153 | int is_dir = S_ISDIR(mode); | |
1154 | struct xfs_mount *mp = dp->i_mount; | |
1155 | struct xfs_inode *ip = NULL; | |
1156 | struct xfs_trans *tp = NULL; | |
1157 | int error; | |
2c3234d1 | 1158 | struct xfs_defer_ops dfops; |
c24b5dfa DC |
1159 | xfs_fsblock_t first_block; |
1160 | bool unlock_dp_on_error = false; | |
c24b5dfa DC |
1161 | prid_t prid; |
1162 | struct xfs_dquot *udqp = NULL; | |
1163 | struct xfs_dquot *gdqp = NULL; | |
1164 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 1165 | struct xfs_trans_res *tres; |
c24b5dfa | 1166 | uint resblks; |
c24b5dfa DC |
1167 | |
1168 | trace_xfs_create(dp, name); | |
1169 | ||
1170 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1171 | return -EIO; |
c24b5dfa | 1172 | |
163467d3 | 1173 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
1174 | |
1175 | /* | |
1176 | * Make sure that we have allocated dquot(s) on disk. | |
1177 | */ | |
7aab1b28 DE |
1178 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1179 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1180 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1181 | &udqp, &gdqp, &pdqp); | |
1182 | if (error) | |
1183 | return error; | |
1184 | ||
1185 | if (is_dir) { | |
c24b5dfa | 1186 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
062647a8 | 1187 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
1188 | } else { |
1189 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 1190 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
1191 | } |
1192 | ||
c24b5dfa DC |
1193 | /* |
1194 | * Initially assume that the file does not exist and | |
1195 | * reserve the resources for that case. If that is not | |
1196 | * the case we'll drop the one we have and get a more | |
1197 | * appropriate transaction later. | |
1198 | */ | |
253f4911 | 1199 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
2451337d | 1200 | if (error == -ENOSPC) { |
c24b5dfa DC |
1201 | /* flush outstanding delalloc blocks and retry */ |
1202 | xfs_flush_inodes(mp); | |
253f4911 | 1203 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
c24b5dfa | 1204 | } |
2451337d | 1205 | if (error == -ENOSPC) { |
c24b5dfa DC |
1206 | /* No space at all so try a "no-allocation" reservation */ |
1207 | resblks = 0; | |
253f4911 | 1208 | error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); |
c24b5dfa | 1209 | } |
4906e215 | 1210 | if (error) |
253f4911 | 1211 | goto out_release_inode; |
c24b5dfa | 1212 | |
65523218 | 1213 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
c24b5dfa DC |
1214 | unlock_dp_on_error = true; |
1215 | ||
2c3234d1 | 1216 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa DC |
1217 | |
1218 | /* | |
1219 | * Reserve disk quota and the inode. | |
1220 | */ | |
1221 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1222 | pdqp, resblks, 1, 0); | |
1223 | if (error) | |
1224 | goto out_trans_cancel; | |
1225 | ||
94f3cad5 ES |
1226 | if (!resblks) { |
1227 | error = xfs_dir_canenter(tp, dp, name); | |
1228 | if (error) | |
1229 | goto out_trans_cancel; | |
1230 | } | |
c24b5dfa DC |
1231 | |
1232 | /* | |
1233 | * A newly created regular or special file just has one directory | |
1234 | * entry pointing to them, but a directory also the "." entry | |
1235 | * pointing to itself. | |
1236 | */ | |
1237 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, | |
f6106efa | 1238 | prid, resblks > 0, &ip, NULL); |
d6077aa3 | 1239 | if (error) |
4906e215 | 1240 | goto out_trans_cancel; |
c24b5dfa DC |
1241 | |
1242 | /* | |
1243 | * Now we join the directory inode to the transaction. We do not do it | |
1244 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1245 | * (and release all the locks). An error from here on will result in | |
1246 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1247 | * error path. | |
1248 | */ | |
65523218 | 1249 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1250 | unlock_dp_on_error = false; |
1251 | ||
1252 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
2c3234d1 | 1253 | &first_block, &dfops, resblks ? |
c24b5dfa DC |
1254 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); |
1255 | if (error) { | |
2451337d | 1256 | ASSERT(error != -ENOSPC); |
4906e215 | 1257 | goto out_trans_cancel; |
c24b5dfa DC |
1258 | } |
1259 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1260 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1261 | ||
1262 | if (is_dir) { | |
1263 | error = xfs_dir_init(tp, ip, dp); | |
1264 | if (error) | |
1265 | goto out_bmap_cancel; | |
1266 | ||
1267 | error = xfs_bumplink(tp, dp); | |
1268 | if (error) | |
1269 | goto out_bmap_cancel; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * If this is a synchronous mount, make sure that the | |
1274 | * create transaction goes to disk before returning to | |
1275 | * the user. | |
1276 | */ | |
1277 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1278 | xfs_trans_set_sync(tp); | |
1279 | ||
1280 | /* | |
1281 | * Attach the dquot(s) to the inodes and modify them incore. | |
1282 | * These ids of the inode couldn't have changed since the new | |
1283 | * inode has been locked ever since it was created. | |
1284 | */ | |
1285 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1286 | ||
8ad7c629 | 1287 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
1288 | if (error) |
1289 | goto out_bmap_cancel; | |
1290 | ||
70393313 | 1291 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1292 | if (error) |
1293 | goto out_release_inode; | |
1294 | ||
1295 | xfs_qm_dqrele(udqp); | |
1296 | xfs_qm_dqrele(gdqp); | |
1297 | xfs_qm_dqrele(pdqp); | |
1298 | ||
1299 | *ipp = ip; | |
1300 | return 0; | |
1301 | ||
1302 | out_bmap_cancel: | |
2c3234d1 | 1303 | xfs_defer_cancel(&dfops); |
c24b5dfa | 1304 | out_trans_cancel: |
4906e215 | 1305 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1306 | out_release_inode: |
1307 | /* | |
58c90473 DC |
1308 | * Wait until after the current transaction is aborted to finish the |
1309 | * setup of the inode and release the inode. This prevents recursive | |
1310 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1311 | */ |
58c90473 DC |
1312 | if (ip) { |
1313 | xfs_finish_inode_setup(ip); | |
c24b5dfa | 1314 | IRELE(ip); |
58c90473 | 1315 | } |
c24b5dfa DC |
1316 | |
1317 | xfs_qm_dqrele(udqp); | |
1318 | xfs_qm_dqrele(gdqp); | |
1319 | xfs_qm_dqrele(pdqp); | |
1320 | ||
1321 | if (unlock_dp_on_error) | |
65523218 | 1322 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1323 | return error; |
1324 | } | |
1325 | ||
99b6436b ZYW |
1326 | int |
1327 | xfs_create_tmpfile( | |
1328 | struct xfs_inode *dp, | |
1329 | struct dentry *dentry, | |
330033d6 BF |
1330 | umode_t mode, |
1331 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1332 | { |
1333 | struct xfs_mount *mp = dp->i_mount; | |
1334 | struct xfs_inode *ip = NULL; | |
1335 | struct xfs_trans *tp = NULL; | |
1336 | int error; | |
99b6436b ZYW |
1337 | prid_t prid; |
1338 | struct xfs_dquot *udqp = NULL; | |
1339 | struct xfs_dquot *gdqp = NULL; | |
1340 | struct xfs_dquot *pdqp = NULL; | |
1341 | struct xfs_trans_res *tres; | |
1342 | uint resblks; | |
1343 | ||
1344 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1345 | return -EIO; |
99b6436b ZYW |
1346 | |
1347 | prid = xfs_get_initial_prid(dp); | |
1348 | ||
1349 | /* | |
1350 | * Make sure that we have allocated dquot(s) on disk. | |
1351 | */ | |
1352 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), | |
1353 | xfs_kgid_to_gid(current_fsgid()), prid, | |
1354 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
1355 | &udqp, &gdqp, &pdqp); | |
1356 | if (error) | |
1357 | return error; | |
1358 | ||
1359 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
99b6436b | 1360 | tres = &M_RES(mp)->tr_create_tmpfile; |
253f4911 CH |
1361 | |
1362 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); | |
2451337d | 1363 | if (error == -ENOSPC) { |
99b6436b ZYW |
1364 | /* No space at all so try a "no-allocation" reservation */ |
1365 | resblks = 0; | |
253f4911 | 1366 | error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); |
99b6436b | 1367 | } |
4906e215 | 1368 | if (error) |
253f4911 | 1369 | goto out_release_inode; |
99b6436b ZYW |
1370 | |
1371 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1372 | pdqp, resblks, 1, 0); | |
1373 | if (error) | |
1374 | goto out_trans_cancel; | |
1375 | ||
1376 | error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, | |
1377 | prid, resblks > 0, &ip, NULL); | |
d6077aa3 | 1378 | if (error) |
4906e215 | 1379 | goto out_trans_cancel; |
99b6436b ZYW |
1380 | |
1381 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1382 | xfs_trans_set_sync(tp); | |
1383 | ||
1384 | /* | |
1385 | * Attach the dquot(s) to the inodes and modify them incore. | |
1386 | * These ids of the inode couldn't have changed since the new | |
1387 | * inode has been locked ever since it was created. | |
1388 | */ | |
1389 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1390 | ||
99b6436b ZYW |
1391 | error = xfs_iunlink(tp, ip); |
1392 | if (error) | |
4906e215 | 1393 | goto out_trans_cancel; |
99b6436b | 1394 | |
70393313 | 1395 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1396 | if (error) |
1397 | goto out_release_inode; | |
1398 | ||
1399 | xfs_qm_dqrele(udqp); | |
1400 | xfs_qm_dqrele(gdqp); | |
1401 | xfs_qm_dqrele(pdqp); | |
1402 | ||
330033d6 | 1403 | *ipp = ip; |
99b6436b ZYW |
1404 | return 0; |
1405 | ||
99b6436b | 1406 | out_trans_cancel: |
4906e215 | 1407 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1408 | out_release_inode: |
1409 | /* | |
58c90473 DC |
1410 | * Wait until after the current transaction is aborted to finish the |
1411 | * setup of the inode and release the inode. This prevents recursive | |
1412 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1413 | */ |
58c90473 DC |
1414 | if (ip) { |
1415 | xfs_finish_inode_setup(ip); | |
99b6436b | 1416 | IRELE(ip); |
58c90473 | 1417 | } |
99b6436b ZYW |
1418 | |
1419 | xfs_qm_dqrele(udqp); | |
1420 | xfs_qm_dqrele(gdqp); | |
1421 | xfs_qm_dqrele(pdqp); | |
1422 | ||
1423 | return error; | |
1424 | } | |
1425 | ||
c24b5dfa DC |
1426 | int |
1427 | xfs_link( | |
1428 | xfs_inode_t *tdp, | |
1429 | xfs_inode_t *sip, | |
1430 | struct xfs_name *target_name) | |
1431 | { | |
1432 | xfs_mount_t *mp = tdp->i_mount; | |
1433 | xfs_trans_t *tp; | |
1434 | int error; | |
2c3234d1 | 1435 | struct xfs_defer_ops dfops; |
c24b5dfa | 1436 | xfs_fsblock_t first_block; |
c24b5dfa DC |
1437 | int resblks; |
1438 | ||
1439 | trace_xfs_link(tdp, target_name); | |
1440 | ||
c19b3b05 | 1441 | ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); |
c24b5dfa DC |
1442 | |
1443 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1444 | return -EIO; |
c24b5dfa DC |
1445 | |
1446 | error = xfs_qm_dqattach(sip, 0); | |
1447 | if (error) | |
1448 | goto std_return; | |
1449 | ||
1450 | error = xfs_qm_dqattach(tdp, 0); | |
1451 | if (error) | |
1452 | goto std_return; | |
1453 | ||
c24b5dfa | 1454 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
253f4911 | 1455 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); |
2451337d | 1456 | if (error == -ENOSPC) { |
c24b5dfa | 1457 | resblks = 0; |
253f4911 | 1458 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); |
c24b5dfa | 1459 | } |
4906e215 | 1460 | if (error) |
253f4911 | 1461 | goto std_return; |
c24b5dfa DC |
1462 | |
1463 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); | |
1464 | ||
1465 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
65523218 | 1466 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1467 | |
1468 | /* | |
1469 | * If we are using project inheritance, we only allow hard link | |
1470 | * creation in our tree when the project IDs are the same; else | |
1471 | * the tree quota mechanism could be circumvented. | |
1472 | */ | |
1473 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1474 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
2451337d | 1475 | error = -EXDEV; |
c24b5dfa DC |
1476 | goto error_return; |
1477 | } | |
1478 | ||
94f3cad5 ES |
1479 | if (!resblks) { |
1480 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1481 | if (error) | |
1482 | goto error_return; | |
1483 | } | |
c24b5dfa | 1484 | |
2c3234d1 | 1485 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa | 1486 | |
54d7b5c1 DC |
1487 | /* |
1488 | * Handle initial link state of O_TMPFILE inode | |
1489 | */ | |
1490 | if (VFS_I(sip)->i_nlink == 0) { | |
ab297431 ZYW |
1491 | error = xfs_iunlink_remove(tp, sip); |
1492 | if (error) | |
4906e215 | 1493 | goto error_return; |
ab297431 ZYW |
1494 | } |
1495 | ||
c24b5dfa | 1496 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2c3234d1 | 1497 | &first_block, &dfops, resblks); |
c24b5dfa | 1498 | if (error) |
4906e215 | 1499 | goto error_return; |
c24b5dfa DC |
1500 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1501 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1502 | ||
1503 | error = xfs_bumplink(tp, sip); | |
1504 | if (error) | |
4906e215 | 1505 | goto error_return; |
c24b5dfa DC |
1506 | |
1507 | /* | |
1508 | * If this is a synchronous mount, make sure that the | |
1509 | * link transaction goes to disk before returning to | |
1510 | * the user. | |
1511 | */ | |
f6106efa | 1512 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
c24b5dfa | 1513 | xfs_trans_set_sync(tp); |
c24b5dfa | 1514 | |
8ad7c629 | 1515 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa | 1516 | if (error) { |
2c3234d1 | 1517 | xfs_defer_cancel(&dfops); |
4906e215 | 1518 | goto error_return; |
c24b5dfa DC |
1519 | } |
1520 | ||
70393313 | 1521 | return xfs_trans_commit(tp); |
c24b5dfa | 1522 | |
c24b5dfa | 1523 | error_return: |
4906e215 | 1524 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1525 | std_return: |
1526 | return error; | |
1527 | } | |
1528 | ||
1da177e4 | 1529 | /* |
8f04c47a CH |
1530 | * Free up the underlying blocks past new_size. The new size must be smaller |
1531 | * than the current size. This routine can be used both for the attribute and | |
1532 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1533 | * |
f6485057 DC |
1534 | * The transaction passed to this routine must have made a permanent log |
1535 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1536 | * given transaction and start new ones, so make sure everything involved in | |
1537 | * the transaction is tidy before calling here. Some transaction will be | |
1538 | * returned to the caller to be committed. The incoming transaction must | |
1539 | * already include the inode, and both inode locks must be held exclusively. | |
1540 | * The inode must also be "held" within the transaction. On return the inode | |
1541 | * will be "held" within the returned transaction. This routine does NOT | |
1542 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1543 | * |
f6485057 DC |
1544 | * If we get an error, we must return with the inode locked and linked into the |
1545 | * current transaction. This keeps things simple for the higher level code, | |
1546 | * because it always knows that the inode is locked and held in the transaction | |
1547 | * that returns to it whether errors occur or not. We don't mark the inode | |
1548 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1549 | */ |
1550 | int | |
8f04c47a CH |
1551 | xfs_itruncate_extents( |
1552 | struct xfs_trans **tpp, | |
1553 | struct xfs_inode *ip, | |
1554 | int whichfork, | |
1555 | xfs_fsize_t new_size) | |
1da177e4 | 1556 | { |
8f04c47a CH |
1557 | struct xfs_mount *mp = ip->i_mount; |
1558 | struct xfs_trans *tp = *tpp; | |
2c3234d1 | 1559 | struct xfs_defer_ops dfops; |
8f04c47a CH |
1560 | xfs_fsblock_t first_block; |
1561 | xfs_fileoff_t first_unmap_block; | |
1562 | xfs_fileoff_t last_block; | |
1563 | xfs_filblks_t unmap_len; | |
8f04c47a CH |
1564 | int error = 0; |
1565 | int done = 0; | |
1da177e4 | 1566 | |
0b56185b CH |
1567 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1568 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1569 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1570 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1571 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1572 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1573 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1574 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1575 | |
673e8e59 CH |
1576 | trace_xfs_itruncate_extents_start(ip, new_size); |
1577 | ||
1da177e4 LT |
1578 | /* |
1579 | * Since it is possible for space to become allocated beyond | |
1580 | * the end of the file (in a crash where the space is allocated | |
1581 | * but the inode size is not yet updated), simply remove any | |
1582 | * blocks which show up between the new EOF and the maximum | |
1583 | * possible file size. If the first block to be removed is | |
1584 | * beyond the maximum file size (ie it is the same as last_block), | |
1585 | * then there is nothing to do. | |
1586 | */ | |
8f04c47a | 1587 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1588 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1589 | if (first_unmap_block == last_block) |
1590 | return 0; | |
1591 | ||
1592 | ASSERT(first_unmap_block < last_block); | |
1593 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1594 | while (!done) { |
2c3234d1 | 1595 | xfs_defer_init(&dfops, &first_block); |
8f04c47a | 1596 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1597 | first_unmap_block, unmap_len, |
8f04c47a | 1598 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1599 | XFS_ITRUNC_MAX_EXTENTS, |
2c3234d1 | 1600 | &first_block, &dfops, |
b4e9181e | 1601 | &done); |
8f04c47a CH |
1602 | if (error) |
1603 | goto out_bmap_cancel; | |
1da177e4 LT |
1604 | |
1605 | /* | |
1606 | * Duplicate the transaction that has the permanent | |
1607 | * reservation and commit the old transaction. | |
1608 | */ | |
8ad7c629 CH |
1609 | xfs_defer_ijoin(&dfops, ip); |
1610 | error = xfs_defer_finish(&tp, &dfops); | |
8f04c47a CH |
1611 | if (error) |
1612 | goto out_bmap_cancel; | |
1da177e4 | 1613 | |
411350df | 1614 | error = xfs_trans_roll_inode(&tp, ip); |
f6485057 | 1615 | if (error) |
8f04c47a | 1616 | goto out; |
1da177e4 | 1617 | } |
8f04c47a | 1618 | |
aa8968f2 DW |
1619 | /* Remove all pending CoW reservations. */ |
1620 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, | |
3802a345 | 1621 | last_block, true); |
aa8968f2 DW |
1622 | if (error) |
1623 | goto out; | |
1624 | ||
1625 | /* | |
cc6f7771 DW |
1626 | * Clear the reflink flag if there are no data fork blocks and |
1627 | * there are no extents staged in the cow fork. | |
aa8968f2 | 1628 | */ |
cc6f7771 DW |
1629 | if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) { |
1630 | if (ip->i_d.di_nblocks == 0) | |
1631 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
83104d44 DW |
1632 | xfs_inode_clear_cowblocks_tag(ip); |
1633 | } | |
aa8968f2 | 1634 | |
673e8e59 CH |
1635 | /* |
1636 | * Always re-log the inode so that our permanent transaction can keep | |
1637 | * on rolling it forward in the log. | |
1638 | */ | |
1639 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1640 | ||
1641 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1642 | ||
8f04c47a CH |
1643 | out: |
1644 | *tpp = tp; | |
1645 | return error; | |
1646 | out_bmap_cancel: | |
1da177e4 | 1647 | /* |
8f04c47a CH |
1648 | * If the bunmapi call encounters an error, return to the caller where |
1649 | * the transaction can be properly aborted. We just need to make sure | |
1650 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1651 | */ |
2c3234d1 | 1652 | xfs_defer_cancel(&dfops); |
8f04c47a CH |
1653 | goto out; |
1654 | } | |
1655 | ||
c24b5dfa DC |
1656 | int |
1657 | xfs_release( | |
1658 | xfs_inode_t *ip) | |
1659 | { | |
1660 | xfs_mount_t *mp = ip->i_mount; | |
1661 | int error; | |
1662 | ||
c19b3b05 | 1663 | if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) |
c24b5dfa DC |
1664 | return 0; |
1665 | ||
1666 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1667 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1668 | return 0; | |
1669 | ||
1670 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1671 | int truncated; | |
1672 | ||
c24b5dfa DC |
1673 | /* |
1674 | * If we previously truncated this file and removed old data | |
1675 | * in the process, we want to initiate "early" writeout on | |
1676 | * the last close. This is an attempt to combat the notorious | |
1677 | * NULL files problem which is particularly noticeable from a | |
1678 | * truncate down, buffered (re-)write (delalloc), followed by | |
1679 | * a crash. What we are effectively doing here is | |
1680 | * significantly reducing the time window where we'd otherwise | |
1681 | * be exposed to that problem. | |
1682 | */ | |
1683 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1684 | if (truncated) { | |
1685 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1686 | if (ip->i_delayed_blks > 0) { |
2451337d | 1687 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1688 | if (error) |
1689 | return error; | |
1690 | } | |
1691 | } | |
1692 | } | |
1693 | ||
54d7b5c1 | 1694 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1695 | return 0; |
1696 | ||
1697 | if (xfs_can_free_eofblocks(ip, false)) { | |
1698 | ||
a36b9261 BF |
1699 | /* |
1700 | * Check if the inode is being opened, written and closed | |
1701 | * frequently and we have delayed allocation blocks outstanding | |
1702 | * (e.g. streaming writes from the NFS server), truncating the | |
1703 | * blocks past EOF will cause fragmentation to occur. | |
1704 | * | |
1705 | * In this case don't do the truncation, but we have to be | |
1706 | * careful how we detect this case. Blocks beyond EOF show up as | |
1707 | * i_delayed_blks even when the inode is clean, so we need to | |
1708 | * truncate them away first before checking for a dirty release. | |
1709 | * Hence on the first dirty close we will still remove the | |
1710 | * speculative allocation, but after that we will leave it in | |
1711 | * place. | |
1712 | */ | |
1713 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1714 | return 0; | |
c24b5dfa DC |
1715 | /* |
1716 | * If we can't get the iolock just skip truncating the blocks | |
1717 | * past EOF because we could deadlock with the mmap_sem | |
a36b9261 | 1718 | * otherwise. We'll get another chance to drop them once the |
c24b5dfa DC |
1719 | * last reference to the inode is dropped, so we'll never leak |
1720 | * blocks permanently. | |
c24b5dfa | 1721 | */ |
a36b9261 BF |
1722 | if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
1723 | error = xfs_free_eofblocks(ip); | |
1724 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1725 | if (error) | |
1726 | return error; | |
1727 | } | |
c24b5dfa DC |
1728 | |
1729 | /* delalloc blocks after truncation means it really is dirty */ | |
1730 | if (ip->i_delayed_blks) | |
1731 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1732 | } | |
1733 | return 0; | |
1734 | } | |
1735 | ||
f7be2d7f BF |
1736 | /* |
1737 | * xfs_inactive_truncate | |
1738 | * | |
1739 | * Called to perform a truncate when an inode becomes unlinked. | |
1740 | */ | |
1741 | STATIC int | |
1742 | xfs_inactive_truncate( | |
1743 | struct xfs_inode *ip) | |
1744 | { | |
1745 | struct xfs_mount *mp = ip->i_mount; | |
1746 | struct xfs_trans *tp; | |
1747 | int error; | |
1748 | ||
253f4911 | 1749 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
f7be2d7f BF |
1750 | if (error) { |
1751 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
f7be2d7f BF |
1752 | return error; |
1753 | } | |
1754 | ||
1755 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1756 | xfs_trans_ijoin(tp, ip, 0); | |
1757 | ||
1758 | /* | |
1759 | * Log the inode size first to prevent stale data exposure in the event | |
1760 | * of a system crash before the truncate completes. See the related | |
69bca807 | 1761 | * comment in xfs_vn_setattr_size() for details. |
f7be2d7f BF |
1762 | */ |
1763 | ip->i_d.di_size = 0; | |
1764 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1765 | ||
1766 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1767 | if (error) | |
1768 | goto error_trans_cancel; | |
1769 | ||
1770 | ASSERT(ip->i_d.di_nextents == 0); | |
1771 | ||
70393313 | 1772 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1773 | if (error) |
1774 | goto error_unlock; | |
1775 | ||
1776 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1777 | return 0; | |
1778 | ||
1779 | error_trans_cancel: | |
4906e215 | 1780 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1781 | error_unlock: |
1782 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1783 | return error; | |
1784 | } | |
1785 | ||
88877d2b BF |
1786 | /* |
1787 | * xfs_inactive_ifree() | |
1788 | * | |
1789 | * Perform the inode free when an inode is unlinked. | |
1790 | */ | |
1791 | STATIC int | |
1792 | xfs_inactive_ifree( | |
1793 | struct xfs_inode *ip) | |
1794 | { | |
2c3234d1 | 1795 | struct xfs_defer_ops dfops; |
88877d2b | 1796 | xfs_fsblock_t first_block; |
88877d2b BF |
1797 | struct xfs_mount *mp = ip->i_mount; |
1798 | struct xfs_trans *tp; | |
1799 | int error; | |
1800 | ||
9d43b180 | 1801 | /* |
76d771b4 CH |
1802 | * We try to use a per-AG reservation for any block needed by the finobt |
1803 | * tree, but as the finobt feature predates the per-AG reservation | |
1804 | * support a degraded file system might not have enough space for the | |
1805 | * reservation at mount time. In that case try to dip into the reserved | |
1806 | * pool and pray. | |
9d43b180 BF |
1807 | * |
1808 | * Send a warning if the reservation does happen to fail, as the inode | |
1809 | * now remains allocated and sits on the unlinked list until the fs is | |
1810 | * repaired. | |
1811 | */ | |
76d771b4 CH |
1812 | if (unlikely(mp->m_inotbt_nores)) { |
1813 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, | |
1814 | XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, | |
1815 | &tp); | |
1816 | } else { | |
1817 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); | |
1818 | } | |
88877d2b | 1819 | if (error) { |
2451337d | 1820 | if (error == -ENOSPC) { |
9d43b180 BF |
1821 | xfs_warn_ratelimited(mp, |
1822 | "Failed to remove inode(s) from unlinked list. " | |
1823 | "Please free space, unmount and run xfs_repair."); | |
1824 | } else { | |
1825 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1826 | } | |
88877d2b BF |
1827 | return error; |
1828 | } | |
1829 | ||
1830 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1831 | xfs_trans_ijoin(tp, ip, 0); | |
1832 | ||
2c3234d1 DW |
1833 | xfs_defer_init(&dfops, &first_block); |
1834 | error = xfs_ifree(tp, ip, &dfops); | |
88877d2b BF |
1835 | if (error) { |
1836 | /* | |
1837 | * If we fail to free the inode, shut down. The cancel | |
1838 | * might do that, we need to make sure. Otherwise the | |
1839 | * inode might be lost for a long time or forever. | |
1840 | */ | |
1841 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1842 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1843 | __func__, error); | |
1844 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1845 | } | |
4906e215 | 1846 | xfs_trans_cancel(tp); |
88877d2b BF |
1847 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1848 | return error; | |
1849 | } | |
1850 | ||
1851 | /* | |
1852 | * Credit the quota account(s). The inode is gone. | |
1853 | */ | |
1854 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1855 | ||
1856 | /* | |
d4a97a04 BF |
1857 | * Just ignore errors at this point. There is nothing we can do except |
1858 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1859 | */ |
8ad7c629 | 1860 | error = xfs_defer_finish(&tp, &dfops); |
d4a97a04 | 1861 | if (error) { |
310a75a3 | 1862 | xfs_notice(mp, "%s: xfs_defer_finish returned error %d", |
88877d2b | 1863 | __func__, error); |
2c3234d1 | 1864 | xfs_defer_cancel(&dfops); |
d4a97a04 | 1865 | } |
70393313 | 1866 | error = xfs_trans_commit(tp); |
88877d2b BF |
1867 | if (error) |
1868 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1869 | __func__, error); | |
1870 | ||
1871 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1872 | return 0; | |
1873 | } | |
1874 | ||
c24b5dfa DC |
1875 | /* |
1876 | * xfs_inactive | |
1877 | * | |
1878 | * This is called when the vnode reference count for the vnode | |
1879 | * goes to zero. If the file has been unlinked, then it must | |
1880 | * now be truncated. Also, we clear all of the read-ahead state | |
1881 | * kept for the inode here since the file is now closed. | |
1882 | */ | |
74564fb4 | 1883 | void |
c24b5dfa DC |
1884 | xfs_inactive( |
1885 | xfs_inode_t *ip) | |
1886 | { | |
3d3c8b52 | 1887 | struct xfs_mount *mp; |
3d3c8b52 JL |
1888 | int error; |
1889 | int truncate = 0; | |
c24b5dfa DC |
1890 | |
1891 | /* | |
1892 | * If the inode is already free, then there can be nothing | |
1893 | * to clean up here. | |
1894 | */ | |
c19b3b05 | 1895 | if (VFS_I(ip)->i_mode == 0) { |
c24b5dfa DC |
1896 | ASSERT(ip->i_df.if_real_bytes == 0); |
1897 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1898 | return; |
c24b5dfa DC |
1899 | } |
1900 | ||
1901 | mp = ip->i_mount; | |
17c12bcd | 1902 | ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); |
c24b5dfa | 1903 | |
c24b5dfa DC |
1904 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1905 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1906 | return; |
c24b5dfa | 1907 | |
54d7b5c1 | 1908 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1909 | /* |
1910 | * force is true because we are evicting an inode from the | |
1911 | * cache. Post-eof blocks must be freed, lest we end up with | |
1912 | * broken free space accounting. | |
3b4683c2 BF |
1913 | * |
1914 | * Note: don't bother with iolock here since lockdep complains | |
1915 | * about acquiring it in reclaim context. We have the only | |
1916 | * reference to the inode at this point anyways. | |
c24b5dfa | 1917 | */ |
3b4683c2 | 1918 | if (xfs_can_free_eofblocks(ip, true)) |
a36b9261 | 1919 | xfs_free_eofblocks(ip); |
74564fb4 BF |
1920 | |
1921 | return; | |
c24b5dfa DC |
1922 | } |
1923 | ||
c19b3b05 | 1924 | if (S_ISREG(VFS_I(ip)->i_mode) && |
c24b5dfa DC |
1925 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || |
1926 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1927 | truncate = 1; | |
1928 | ||
1929 | error = xfs_qm_dqattach(ip, 0); | |
1930 | if (error) | |
74564fb4 | 1931 | return; |
c24b5dfa | 1932 | |
c19b3b05 | 1933 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
36b21dde | 1934 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1935 | else if (truncate) |
1936 | error = xfs_inactive_truncate(ip); | |
1937 | if (error) | |
74564fb4 | 1938 | return; |
c24b5dfa DC |
1939 | |
1940 | /* | |
1941 | * If there are attributes associated with the file then blow them away | |
1942 | * now. The code calls a routine that recursively deconstructs the | |
6dfe5a04 | 1943 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1944 | */ |
6dfe5a04 | 1945 | if (XFS_IFORK_Q(ip)) { |
c24b5dfa DC |
1946 | error = xfs_attr_inactive(ip); |
1947 | if (error) | |
74564fb4 | 1948 | return; |
c24b5dfa DC |
1949 | } |
1950 | ||
6dfe5a04 | 1951 | ASSERT(!ip->i_afp); |
c24b5dfa | 1952 | ASSERT(ip->i_d.di_anextents == 0); |
6dfe5a04 | 1953 | ASSERT(ip->i_d.di_forkoff == 0); |
c24b5dfa DC |
1954 | |
1955 | /* | |
1956 | * Free the inode. | |
1957 | */ | |
88877d2b BF |
1958 | error = xfs_inactive_ifree(ip); |
1959 | if (error) | |
74564fb4 | 1960 | return; |
c24b5dfa DC |
1961 | |
1962 | /* | |
1963 | * Release the dquots held by inode, if any. | |
1964 | */ | |
1965 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1966 | } |
1967 | ||
1da177e4 | 1968 | /* |
54d7b5c1 DC |
1969 | * This is called when the inode's link count goes to 0 or we are creating a |
1970 | * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be | |
1971 | * set to true as the link count is dropped to zero by the VFS after we've | |
1972 | * created the file successfully, so we have to add it to the unlinked list | |
1973 | * while the link count is non-zero. | |
1974 | * | |
1975 | * We place the on-disk inode on a list in the AGI. It will be pulled from this | |
1976 | * list when the inode is freed. | |
1da177e4 | 1977 | */ |
54d7b5c1 | 1978 | STATIC int |
1da177e4 | 1979 | xfs_iunlink( |
54d7b5c1 DC |
1980 | struct xfs_trans *tp, |
1981 | struct xfs_inode *ip) | |
1da177e4 | 1982 | { |
54d7b5c1 | 1983 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1984 | xfs_agi_t *agi; |
1985 | xfs_dinode_t *dip; | |
1986 | xfs_buf_t *agibp; | |
1987 | xfs_buf_t *ibp; | |
1da177e4 LT |
1988 | xfs_agino_t agino; |
1989 | short bucket_index; | |
1990 | int offset; | |
1991 | int error; | |
1da177e4 | 1992 | |
c19b3b05 | 1993 | ASSERT(VFS_I(ip)->i_mode != 0); |
1da177e4 | 1994 | |
1da177e4 LT |
1995 | /* |
1996 | * Get the agi buffer first. It ensures lock ordering | |
1997 | * on the list. | |
1998 | */ | |
5e1be0fb | 1999 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 2000 | if (error) |
1da177e4 | 2001 | return error; |
1da177e4 | 2002 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2003 | |
1da177e4 LT |
2004 | /* |
2005 | * Get the index into the agi hash table for the | |
2006 | * list this inode will go on. | |
2007 | */ | |
2008 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2009 | ASSERT(agino != 0); | |
2010 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
2011 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 2012 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 2013 | |
69ef921b | 2014 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
2015 | /* |
2016 | * There is already another inode in the bucket we need | |
2017 | * to add ourselves to. Add us at the front of the list. | |
2018 | * Here we put the head pointer into our next pointer, | |
2019 | * and then we fall through to point the head at us. | |
2020 | */ | |
475ee413 CH |
2021 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2022 | 0, 0); | |
c319b58b VA |
2023 | if (error) |
2024 | return error; | |
2025 | ||
69ef921b | 2026 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 2027 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 2028 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2029 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2030 | |
2031 | /* need to recalc the inode CRC if appropriate */ | |
2032 | xfs_dinode_calc_crc(mp, dip); | |
2033 | ||
1da177e4 LT |
2034 | xfs_trans_inode_buf(tp, ibp); |
2035 | xfs_trans_log_buf(tp, ibp, offset, | |
2036 | (offset + sizeof(xfs_agino_t) - 1)); | |
2037 | xfs_inobp_check(mp, ibp); | |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * Point the bucket head pointer at the inode being inserted. | |
2042 | */ | |
2043 | ASSERT(agino != 0); | |
16259e7d | 2044 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
2045 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2046 | (sizeof(xfs_agino_t) * bucket_index); | |
2047 | xfs_trans_log_buf(tp, agibp, offset, | |
2048 | (offset + sizeof(xfs_agino_t) - 1)); | |
2049 | return 0; | |
2050 | } | |
2051 | ||
2052 | /* | |
2053 | * Pull the on-disk inode from the AGI unlinked list. | |
2054 | */ | |
2055 | STATIC int | |
2056 | xfs_iunlink_remove( | |
2057 | xfs_trans_t *tp, | |
2058 | xfs_inode_t *ip) | |
2059 | { | |
2060 | xfs_ino_t next_ino; | |
2061 | xfs_mount_t *mp; | |
2062 | xfs_agi_t *agi; | |
2063 | xfs_dinode_t *dip; | |
2064 | xfs_buf_t *agibp; | |
2065 | xfs_buf_t *ibp; | |
2066 | xfs_agnumber_t agno; | |
1da177e4 LT |
2067 | xfs_agino_t agino; |
2068 | xfs_agino_t next_agino; | |
2069 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2070 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2071 | short bucket_index; |
6fdf8ccc | 2072 | int offset, last_offset = 0; |
1da177e4 | 2073 | int error; |
1da177e4 | 2074 | |
1da177e4 | 2075 | mp = tp->t_mountp; |
1da177e4 | 2076 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
2077 | |
2078 | /* | |
2079 | * Get the agi buffer first. It ensures lock ordering | |
2080 | * on the list. | |
2081 | */ | |
5e1be0fb CH |
2082 | error = xfs_read_agi(mp, tp, agno, &agibp); |
2083 | if (error) | |
1da177e4 | 2084 | return error; |
5e1be0fb | 2085 | |
1da177e4 | 2086 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2087 | |
1da177e4 LT |
2088 | /* |
2089 | * Get the index into the agi hash table for the | |
2090 | * list this inode will go on. | |
2091 | */ | |
2092 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2093 | ASSERT(agino != 0); | |
2094 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 2095 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
2096 | ASSERT(agi->agi_unlinked[bucket_index]); |
2097 | ||
16259e7d | 2098 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2099 | /* |
475ee413 CH |
2100 | * We're at the head of the list. Get the inode's on-disk |
2101 | * buffer to see if there is anyone after us on the list. | |
2102 | * Only modify our next pointer if it is not already NULLAGINO. | |
2103 | * This saves us the overhead of dealing with the buffer when | |
2104 | * there is no need to change it. | |
1da177e4 | 2105 | */ |
475ee413 CH |
2106 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2107 | 0, 0); | |
1da177e4 | 2108 | if (error) { |
475ee413 | 2109 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2110 | __func__, error); |
1da177e4 LT |
2111 | return error; |
2112 | } | |
347d1c01 | 2113 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2114 | ASSERT(next_agino != 0); |
2115 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2116 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2117 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2118 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2119 | |
2120 | /* need to recalc the inode CRC if appropriate */ | |
2121 | xfs_dinode_calc_crc(mp, dip); | |
2122 | ||
1da177e4 LT |
2123 | xfs_trans_inode_buf(tp, ibp); |
2124 | xfs_trans_log_buf(tp, ibp, offset, | |
2125 | (offset + sizeof(xfs_agino_t) - 1)); | |
2126 | xfs_inobp_check(mp, ibp); | |
2127 | } else { | |
2128 | xfs_trans_brelse(tp, ibp); | |
2129 | } | |
2130 | /* | |
2131 | * Point the bucket head pointer at the next inode. | |
2132 | */ | |
2133 | ASSERT(next_agino != 0); | |
2134 | ASSERT(next_agino != agino); | |
16259e7d | 2135 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2136 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2137 | (sizeof(xfs_agino_t) * bucket_index); | |
2138 | xfs_trans_log_buf(tp, agibp, offset, | |
2139 | (offset + sizeof(xfs_agino_t) - 1)); | |
2140 | } else { | |
2141 | /* | |
2142 | * We need to search the list for the inode being freed. | |
2143 | */ | |
16259e7d | 2144 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2145 | last_ibp = NULL; |
2146 | while (next_agino != agino) { | |
129dbc9a CH |
2147 | struct xfs_imap imap; |
2148 | ||
2149 | if (last_ibp) | |
1da177e4 | 2150 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2151 | |
2152 | imap.im_blkno = 0; | |
1da177e4 | 2153 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2154 | |
2155 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2156 | if (error) { | |
2157 | xfs_warn(mp, | |
2158 | "%s: xfs_imap returned error %d.", | |
2159 | __func__, error); | |
2160 | return error; | |
2161 | } | |
2162 | ||
2163 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
2164 | &last_ibp, 0, 0); | |
1da177e4 | 2165 | if (error) { |
0b932ccc | 2166 | xfs_warn(mp, |
129dbc9a | 2167 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2168 | __func__, error); |
1da177e4 LT |
2169 | return error; |
2170 | } | |
129dbc9a CH |
2171 | |
2172 | last_offset = imap.im_boffset; | |
347d1c01 | 2173 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2174 | ASSERT(next_agino != NULLAGINO); |
2175 | ASSERT(next_agino != 0); | |
2176 | } | |
475ee413 | 2177 | |
1da177e4 | 2178 | /* |
475ee413 CH |
2179 | * Now last_ibp points to the buffer previous to us on the |
2180 | * unlinked list. Pull us from the list. | |
1da177e4 | 2181 | */ |
475ee413 CH |
2182 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2183 | 0, 0); | |
1da177e4 | 2184 | if (error) { |
475ee413 | 2185 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2186 | __func__, error); |
1da177e4 LT |
2187 | return error; |
2188 | } | |
347d1c01 | 2189 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2190 | ASSERT(next_agino != 0); |
2191 | ASSERT(next_agino != agino); | |
2192 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2193 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2194 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2195 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2196 | |
2197 | /* need to recalc the inode CRC if appropriate */ | |
2198 | xfs_dinode_calc_crc(mp, dip); | |
2199 | ||
1da177e4 LT |
2200 | xfs_trans_inode_buf(tp, ibp); |
2201 | xfs_trans_log_buf(tp, ibp, offset, | |
2202 | (offset + sizeof(xfs_agino_t) - 1)); | |
2203 | xfs_inobp_check(mp, ibp); | |
2204 | } else { | |
2205 | xfs_trans_brelse(tp, ibp); | |
2206 | } | |
2207 | /* | |
2208 | * Point the previous inode on the list to the next inode. | |
2209 | */ | |
347d1c01 | 2210 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2211 | ASSERT(next_agino != 0); |
2212 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2213 | |
2214 | /* need to recalc the inode CRC if appropriate */ | |
2215 | xfs_dinode_calc_crc(mp, last_dip); | |
2216 | ||
1da177e4 LT |
2217 | xfs_trans_inode_buf(tp, last_ibp); |
2218 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2219 | (offset + sizeof(xfs_agino_t) - 1)); | |
2220 | xfs_inobp_check(mp, last_ibp); | |
2221 | } | |
2222 | return 0; | |
2223 | } | |
2224 | ||
5b3eed75 | 2225 | /* |
0b8182db | 2226 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2227 | * inodes that are in memory - they all must be marked stale and attached to |
2228 | * the cluster buffer. | |
2229 | */ | |
2a30f36d | 2230 | STATIC int |
1da177e4 | 2231 | xfs_ifree_cluster( |
09b56604 BF |
2232 | xfs_inode_t *free_ip, |
2233 | xfs_trans_t *tp, | |
2234 | struct xfs_icluster *xic) | |
1da177e4 LT |
2235 | { |
2236 | xfs_mount_t *mp = free_ip->i_mount; | |
2237 | int blks_per_cluster; | |
982e939e | 2238 | int inodes_per_cluster; |
1da177e4 | 2239 | int nbufs; |
5b257b4a | 2240 | int i, j; |
3cdaa189 | 2241 | int ioffset; |
1da177e4 LT |
2242 | xfs_daddr_t blkno; |
2243 | xfs_buf_t *bp; | |
5b257b4a | 2244 | xfs_inode_t *ip; |
1da177e4 LT |
2245 | xfs_inode_log_item_t *iip; |
2246 | xfs_log_item_t *lip; | |
5017e97d | 2247 | struct xfs_perag *pag; |
09b56604 | 2248 | xfs_ino_t inum; |
1da177e4 | 2249 | |
09b56604 | 2250 | inum = xic->first_ino; |
5017e97d | 2251 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
982e939e JL |
2252 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
2253 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
2254 | nbufs = mp->m_ialloc_blks / blks_per_cluster; | |
1da177e4 | 2255 | |
982e939e | 2256 | for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { |
09b56604 BF |
2257 | /* |
2258 | * The allocation bitmap tells us which inodes of the chunk were | |
2259 | * physically allocated. Skip the cluster if an inode falls into | |
2260 | * a sparse region. | |
2261 | */ | |
3cdaa189 BF |
2262 | ioffset = inum - xic->first_ino; |
2263 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
2264 | ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); | |
09b56604 BF |
2265 | continue; |
2266 | } | |
2267 | ||
1da177e4 LT |
2268 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2269 | XFS_INO_TO_AGBNO(mp, inum)); | |
2270 | ||
5b257b4a DC |
2271 | /* |
2272 | * We obtain and lock the backing buffer first in the process | |
2273 | * here, as we have to ensure that any dirty inode that we | |
2274 | * can't get the flush lock on is attached to the buffer. | |
2275 | * If we scan the in-memory inodes first, then buffer IO can | |
2276 | * complete before we get a lock on it, and hence we may fail | |
2277 | * to mark all the active inodes on the buffer stale. | |
2278 | */ | |
2279 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2280 | mp->m_bsize * blks_per_cluster, |
2281 | XBF_UNMAPPED); | |
5b257b4a | 2282 | |
2a30f36d | 2283 | if (!bp) |
2451337d | 2284 | return -ENOMEM; |
b0f539de DC |
2285 | |
2286 | /* | |
2287 | * This buffer may not have been correctly initialised as we | |
2288 | * didn't read it from disk. That's not important because we are | |
2289 | * only using to mark the buffer as stale in the log, and to | |
2290 | * attach stale cached inodes on it. That means it will never be | |
2291 | * dispatched for IO. If it is, we want to know about it, and we | |
2292 | * want it to fail. We can acheive this by adding a write | |
2293 | * verifier to the buffer. | |
2294 | */ | |
1813dd64 | 2295 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2296 | |
5b257b4a DC |
2297 | /* |
2298 | * Walk the inodes already attached to the buffer and mark them | |
2299 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2300 | * in-memory inode walk can't lock them. By marking them all |
2301 | * stale first, we will not attempt to lock them in the loop | |
2302 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 2303 | */ |
adadbeef | 2304 | lip = bp->b_fspriv; |
5b257b4a DC |
2305 | while (lip) { |
2306 | if (lip->li_type == XFS_LI_INODE) { | |
2307 | iip = (xfs_inode_log_item_t *)lip; | |
2308 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2309 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2310 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2311 | &iip->ili_flush_lsn, | |
2312 | &iip->ili_item.li_lsn); | |
2313 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
2314 | } |
2315 | lip = lip->li_bio_list; | |
2316 | } | |
1da177e4 | 2317 | |
5b3eed75 | 2318 | |
1da177e4 | 2319 | /* |
5b257b4a DC |
2320 | * For each inode in memory attempt to add it to the inode |
2321 | * buffer and set it up for being staled on buffer IO | |
2322 | * completion. This is safe as we've locked out tail pushing | |
2323 | * and flushing by locking the buffer. | |
1da177e4 | 2324 | * |
5b257b4a DC |
2325 | * We have already marked every inode that was part of a |
2326 | * transaction stale above, which means there is no point in | |
2327 | * even trying to lock them. | |
1da177e4 | 2328 | */ |
982e939e | 2329 | for (i = 0; i < inodes_per_cluster; i++) { |
5b3eed75 | 2330 | retry: |
1a3e8f3d | 2331 | rcu_read_lock(); |
da353b0d DC |
2332 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2333 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2334 | |
1a3e8f3d DC |
2335 | /* Inode not in memory, nothing to do */ |
2336 | if (!ip) { | |
2337 | rcu_read_unlock(); | |
1da177e4 LT |
2338 | continue; |
2339 | } | |
2340 | ||
1a3e8f3d DC |
2341 | /* |
2342 | * because this is an RCU protected lookup, we could | |
2343 | * find a recently freed or even reallocated inode | |
2344 | * during the lookup. We need to check under the | |
2345 | * i_flags_lock for a valid inode here. Skip it if it | |
2346 | * is not valid, the wrong inode or stale. | |
2347 | */ | |
2348 | spin_lock(&ip->i_flags_lock); | |
2349 | if (ip->i_ino != inum + i || | |
2350 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
2351 | spin_unlock(&ip->i_flags_lock); | |
2352 | rcu_read_unlock(); | |
2353 | continue; | |
2354 | } | |
2355 | spin_unlock(&ip->i_flags_lock); | |
2356 | ||
5b3eed75 DC |
2357 | /* |
2358 | * Don't try to lock/unlock the current inode, but we | |
2359 | * _cannot_ skip the other inodes that we did not find | |
2360 | * in the list attached to the buffer and are not | |
2361 | * already marked stale. If we can't lock it, back off | |
2362 | * and retry. | |
2363 | */ | |
f2e9ad21 OS |
2364 | if (ip != free_ip) { |
2365 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2366 | rcu_read_unlock(); | |
2367 | delay(1); | |
2368 | goto retry; | |
2369 | } | |
2370 | ||
2371 | /* | |
2372 | * Check the inode number again in case we're | |
2373 | * racing with freeing in xfs_reclaim_inode(). | |
2374 | * See the comments in that function for more | |
2375 | * information as to why the initial check is | |
2376 | * not sufficient. | |
2377 | */ | |
2378 | if (ip->i_ino != inum + i) { | |
2379 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2380 | continue; | |
2381 | } | |
1da177e4 | 2382 | } |
1a3e8f3d | 2383 | rcu_read_unlock(); |
1da177e4 | 2384 | |
5b3eed75 | 2385 | xfs_iflock(ip); |
5b257b4a | 2386 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2387 | |
5b3eed75 DC |
2388 | /* |
2389 | * we don't need to attach clean inodes or those only | |
2390 | * with unlogged changes (which we throw away, anyway). | |
2391 | */ | |
1da177e4 | 2392 | iip = ip->i_itemp; |
5b3eed75 | 2393 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2394 | ASSERT(ip != free_ip); |
1da177e4 LT |
2395 | xfs_ifunlock(ip); |
2396 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2397 | continue; | |
2398 | } | |
2399 | ||
f5d8d5c4 CH |
2400 | iip->ili_last_fields = iip->ili_fields; |
2401 | iip->ili_fields = 0; | |
fc0561ce | 2402 | iip->ili_fsync_fields = 0; |
1da177e4 | 2403 | iip->ili_logged = 1; |
7b2e2a31 DC |
2404 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2405 | &iip->ili_item.li_lsn); | |
1da177e4 | 2406 | |
ca30b2a7 CH |
2407 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2408 | &iip->ili_item); | |
5b257b4a DC |
2409 | |
2410 | if (ip != free_ip) | |
1da177e4 | 2411 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2412 | } |
2413 | ||
5b3eed75 | 2414 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2415 | xfs_trans_binval(tp, bp); |
2416 | } | |
2417 | ||
5017e97d | 2418 | xfs_perag_put(pag); |
2a30f36d | 2419 | return 0; |
1da177e4 LT |
2420 | } |
2421 | ||
2422 | /* | |
2423 | * This is called to return an inode to the inode free list. | |
2424 | * The inode should already be truncated to 0 length and have | |
2425 | * no pages associated with it. This routine also assumes that | |
2426 | * the inode is already a part of the transaction. | |
2427 | * | |
2428 | * The on-disk copy of the inode will have been added to the list | |
2429 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2430 | * that list atomically with respect to freeing it here. | |
2431 | */ | |
2432 | int | |
2433 | xfs_ifree( | |
2434 | xfs_trans_t *tp, | |
2435 | xfs_inode_t *ip, | |
2c3234d1 | 2436 | struct xfs_defer_ops *dfops) |
1da177e4 LT |
2437 | { |
2438 | int error; | |
09b56604 | 2439 | struct xfs_icluster xic = { 0 }; |
1da177e4 | 2440 | |
579aa9ca | 2441 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2442 | ASSERT(VFS_I(ip)->i_nlink == 0); |
1da177e4 LT |
2443 | ASSERT(ip->i_d.di_nextents == 0); |
2444 | ASSERT(ip->i_d.di_anextents == 0); | |
c19b3b05 | 2445 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); |
1da177e4 LT |
2446 | ASSERT(ip->i_d.di_nblocks == 0); |
2447 | ||
2448 | /* | |
2449 | * Pull the on-disk inode from the AGI unlinked list. | |
2450 | */ | |
2451 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2452 | if (error) |
1da177e4 | 2453 | return error; |
1da177e4 | 2454 | |
2c3234d1 | 2455 | error = xfs_difree(tp, ip->i_ino, dfops, &xic); |
1baaed8f | 2456 | if (error) |
1da177e4 | 2457 | return error; |
1baaed8f | 2458 | |
c19b3b05 | 2459 | VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ |
1da177e4 LT |
2460 | ip->i_d.di_flags = 0; |
2461 | ip->i_d.di_dmevmask = 0; | |
2462 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2463 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2464 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2465 | /* | |
2466 | * Bump the generation count so no one will be confused | |
2467 | * by reincarnations of this inode. | |
2468 | */ | |
9e9a2674 | 2469 | VFS_I(ip)->i_generation++; |
1da177e4 LT |
2470 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2471 | ||
09b56604 BF |
2472 | if (xic.deleted) |
2473 | error = xfs_ifree_cluster(ip, tp, &xic); | |
1da177e4 | 2474 | |
2a30f36d | 2475 | return error; |
1da177e4 LT |
2476 | } |
2477 | ||
1da177e4 | 2478 | /* |
60ec6783 CH |
2479 | * This is called to unpin an inode. The caller must have the inode locked |
2480 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2481 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2482 | */ |
60ec6783 | 2483 | static void |
f392e631 | 2484 | xfs_iunpin( |
60ec6783 | 2485 | struct xfs_inode *ip) |
1da177e4 | 2486 | { |
579aa9ca | 2487 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2488 | |
4aaf15d1 DC |
2489 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2490 | ||
a3f74ffb | 2491 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2492 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2493 | |
a3f74ffb | 2494 | } |
1da177e4 | 2495 | |
f392e631 CH |
2496 | static void |
2497 | __xfs_iunpin_wait( | |
2498 | struct xfs_inode *ip) | |
2499 | { | |
2500 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2501 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2502 | ||
2503 | xfs_iunpin(ip); | |
2504 | ||
2505 | do { | |
21417136 | 2506 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
f392e631 CH |
2507 | if (xfs_ipincount(ip)) |
2508 | io_schedule(); | |
2509 | } while (xfs_ipincount(ip)); | |
21417136 | 2510 | finish_wait(wq, &wait.wq_entry); |
f392e631 CH |
2511 | } |
2512 | ||
777df5af | 2513 | void |
a3f74ffb | 2514 | xfs_iunpin_wait( |
60ec6783 | 2515 | struct xfs_inode *ip) |
a3f74ffb | 2516 | { |
f392e631 CH |
2517 | if (xfs_ipincount(ip)) |
2518 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2519 | } |
2520 | ||
27320369 DC |
2521 | /* |
2522 | * Removing an inode from the namespace involves removing the directory entry | |
2523 | * and dropping the link count on the inode. Removing the directory entry can | |
2524 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2525 | * count can result in placing the inode on an unlinked list which results in | |
2526 | * locking an AGI. | |
2527 | * | |
2528 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2529 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2530 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2531 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2532 | * first, and then freeing the inode can result in an inode chunk being freed | |
2533 | * and hence freeing disk space requiring that we lock an AGF. | |
2534 | * | |
2535 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2536 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2537 | * reference count and put it on the unlinked list as this results in a lock | |
2538 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2539 | * freeing. Therefore we must drop the link counts before we remove the | |
2540 | * directory entry. | |
2541 | * | |
2542 | * This is still safe from a transactional point of view - it is not until we | |
310a75a3 | 2543 | * get to xfs_defer_finish() that we have the possibility of multiple |
27320369 DC |
2544 | * transactions in this operation. Hence as long as we remove the directory |
2545 | * entry and drop the link count in the first transaction of the remove | |
2546 | * operation, there are no transactional constraints on the ordering here. | |
2547 | */ | |
c24b5dfa DC |
2548 | int |
2549 | xfs_remove( | |
2550 | xfs_inode_t *dp, | |
2551 | struct xfs_name *name, | |
2552 | xfs_inode_t *ip) | |
2553 | { | |
2554 | xfs_mount_t *mp = dp->i_mount; | |
2555 | xfs_trans_t *tp = NULL; | |
c19b3b05 | 2556 | int is_dir = S_ISDIR(VFS_I(ip)->i_mode); |
c24b5dfa | 2557 | int error = 0; |
2c3234d1 | 2558 | struct xfs_defer_ops dfops; |
c24b5dfa | 2559 | xfs_fsblock_t first_block; |
c24b5dfa | 2560 | uint resblks; |
c24b5dfa DC |
2561 | |
2562 | trace_xfs_remove(dp, name); | |
2563 | ||
2564 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 2565 | return -EIO; |
c24b5dfa DC |
2566 | |
2567 | error = xfs_qm_dqattach(dp, 0); | |
2568 | if (error) | |
2569 | goto std_return; | |
2570 | ||
2571 | error = xfs_qm_dqattach(ip, 0); | |
2572 | if (error) | |
2573 | goto std_return; | |
2574 | ||
c24b5dfa DC |
2575 | /* |
2576 | * We try to get the real space reservation first, | |
2577 | * allowing for directory btree deletion(s) implying | |
2578 | * possible bmap insert(s). If we can't get the space | |
2579 | * reservation then we use 0 instead, and avoid the bmap | |
2580 | * btree insert(s) in the directory code by, if the bmap | |
2581 | * insert tries to happen, instead trimming the LAST | |
2582 | * block from the directory. | |
2583 | */ | |
2584 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
253f4911 | 2585 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); |
2451337d | 2586 | if (error == -ENOSPC) { |
c24b5dfa | 2587 | resblks = 0; |
253f4911 CH |
2588 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, |
2589 | &tp); | |
c24b5dfa DC |
2590 | } |
2591 | if (error) { | |
2451337d | 2592 | ASSERT(error != -ENOSPC); |
253f4911 | 2593 | goto std_return; |
c24b5dfa DC |
2594 | } |
2595 | ||
2596 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); | |
2597 | ||
65523218 | 2598 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
2599 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
2600 | ||
2601 | /* | |
2602 | * If we're removing a directory perform some additional validation. | |
2603 | */ | |
2604 | if (is_dir) { | |
54d7b5c1 DC |
2605 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2606 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2607 | error = -ENOTEMPTY; |
c24b5dfa DC |
2608 | goto out_trans_cancel; |
2609 | } | |
2610 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2611 | error = -ENOTEMPTY; |
c24b5dfa DC |
2612 | goto out_trans_cancel; |
2613 | } | |
c24b5dfa | 2614 | |
27320369 | 2615 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2616 | error = xfs_droplink(tp, dp); |
2617 | if (error) | |
27320369 | 2618 | goto out_trans_cancel; |
c24b5dfa | 2619 | |
27320369 | 2620 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2621 | error = xfs_droplink(tp, ip); |
2622 | if (error) | |
27320369 | 2623 | goto out_trans_cancel; |
c24b5dfa DC |
2624 | } else { |
2625 | /* | |
2626 | * When removing a non-directory we need to log the parent | |
2627 | * inode here. For a directory this is done implicitly | |
2628 | * by the xfs_droplink call for the ".." entry. | |
2629 | */ | |
2630 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2631 | } | |
27320369 | 2632 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2633 | |
27320369 | 2634 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2635 | error = xfs_droplink(tp, ip); |
2636 | if (error) | |
27320369 | 2637 | goto out_trans_cancel; |
c24b5dfa | 2638 | |
2c3234d1 | 2639 | xfs_defer_init(&dfops, &first_block); |
27320369 | 2640 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
2c3234d1 | 2641 | &first_block, &dfops, resblks); |
27320369 | 2642 | if (error) { |
2451337d | 2643 | ASSERT(error != -ENOENT); |
27320369 DC |
2644 | goto out_bmap_cancel; |
2645 | } | |
2646 | ||
c24b5dfa DC |
2647 | /* |
2648 | * If this is a synchronous mount, make sure that the | |
2649 | * remove transaction goes to disk before returning to | |
2650 | * the user. | |
2651 | */ | |
2652 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2653 | xfs_trans_set_sync(tp); | |
2654 | ||
8ad7c629 | 2655 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
2656 | if (error) |
2657 | goto out_bmap_cancel; | |
2658 | ||
70393313 | 2659 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2660 | if (error) |
2661 | goto std_return; | |
2662 | ||
2cd2ef6a | 2663 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2664 | xfs_filestream_deassociate(ip); |
2665 | ||
2666 | return 0; | |
2667 | ||
2668 | out_bmap_cancel: | |
2c3234d1 | 2669 | xfs_defer_cancel(&dfops); |
c24b5dfa | 2670 | out_trans_cancel: |
4906e215 | 2671 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2672 | std_return: |
2673 | return error; | |
2674 | } | |
2675 | ||
f6bba201 DC |
2676 | /* |
2677 | * Enter all inodes for a rename transaction into a sorted array. | |
2678 | */ | |
95afcf5c | 2679 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2680 | STATIC void |
2681 | xfs_sort_for_rename( | |
95afcf5c DC |
2682 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2683 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2684 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2685 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2686 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2687 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2688 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2689 | { |
f6bba201 DC |
2690 | int i, j; |
2691 | ||
95afcf5c DC |
2692 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2693 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2694 | ||
f6bba201 DC |
2695 | /* |
2696 | * i_tab contains a list of pointers to inodes. We initialize | |
2697 | * the table here & we'll sort it. We will then use it to | |
2698 | * order the acquisition of the inode locks. | |
2699 | * | |
2700 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2701 | */ | |
95afcf5c DC |
2702 | i = 0; |
2703 | i_tab[i++] = dp1; | |
2704 | i_tab[i++] = dp2; | |
2705 | i_tab[i++] = ip1; | |
2706 | if (ip2) | |
2707 | i_tab[i++] = ip2; | |
2708 | if (wip) | |
2709 | i_tab[i++] = wip; | |
2710 | *num_inodes = i; | |
f6bba201 DC |
2711 | |
2712 | /* | |
2713 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2714 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2715 | */ |
2716 | for (i = 0; i < *num_inodes; i++) { | |
2717 | for (j = 1; j < *num_inodes; j++) { | |
2718 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2719 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2720 | i_tab[j] = i_tab[j-1]; |
2721 | i_tab[j-1] = temp; | |
2722 | } | |
2723 | } | |
2724 | } | |
2725 | } | |
2726 | ||
310606b0 DC |
2727 | static int |
2728 | xfs_finish_rename( | |
2729 | struct xfs_trans *tp, | |
2c3234d1 | 2730 | struct xfs_defer_ops *dfops) |
310606b0 | 2731 | { |
310606b0 DC |
2732 | int error; |
2733 | ||
2734 | /* | |
2735 | * If this is a synchronous mount, make sure that the rename transaction | |
2736 | * goes to disk before returning to the user. | |
2737 | */ | |
2738 | if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2739 | xfs_trans_set_sync(tp); | |
2740 | ||
8ad7c629 | 2741 | error = xfs_defer_finish(&tp, dfops); |
310606b0 | 2742 | if (error) { |
2c3234d1 | 2743 | xfs_defer_cancel(dfops); |
4906e215 | 2744 | xfs_trans_cancel(tp); |
310606b0 DC |
2745 | return error; |
2746 | } | |
2747 | ||
70393313 | 2748 | return xfs_trans_commit(tp); |
310606b0 DC |
2749 | } |
2750 | ||
d31a1825 CM |
2751 | /* |
2752 | * xfs_cross_rename() | |
2753 | * | |
2754 | * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall | |
2755 | */ | |
2756 | STATIC int | |
2757 | xfs_cross_rename( | |
2758 | struct xfs_trans *tp, | |
2759 | struct xfs_inode *dp1, | |
2760 | struct xfs_name *name1, | |
2761 | struct xfs_inode *ip1, | |
2762 | struct xfs_inode *dp2, | |
2763 | struct xfs_name *name2, | |
2764 | struct xfs_inode *ip2, | |
2c3234d1 | 2765 | struct xfs_defer_ops *dfops, |
d31a1825 CM |
2766 | xfs_fsblock_t *first_block, |
2767 | int spaceres) | |
2768 | { | |
2769 | int error = 0; | |
2770 | int ip1_flags = 0; | |
2771 | int ip2_flags = 0; | |
2772 | int dp2_flags = 0; | |
2773 | ||
2774 | /* Swap inode number for dirent in first parent */ | |
2775 | error = xfs_dir_replace(tp, dp1, name1, | |
2776 | ip2->i_ino, | |
2c3234d1 | 2777 | first_block, dfops, spaceres); |
d31a1825 | 2778 | if (error) |
eeacd321 | 2779 | goto out_trans_abort; |
d31a1825 CM |
2780 | |
2781 | /* Swap inode number for dirent in second parent */ | |
2782 | error = xfs_dir_replace(tp, dp2, name2, | |
2783 | ip1->i_ino, | |
2c3234d1 | 2784 | first_block, dfops, spaceres); |
d31a1825 | 2785 | if (error) |
eeacd321 | 2786 | goto out_trans_abort; |
d31a1825 CM |
2787 | |
2788 | /* | |
2789 | * If we're renaming one or more directories across different parents, | |
2790 | * update the respective ".." entries (and link counts) to match the new | |
2791 | * parents. | |
2792 | */ | |
2793 | if (dp1 != dp2) { | |
2794 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2795 | ||
c19b3b05 | 2796 | if (S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2797 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, |
2798 | dp1->i_ino, first_block, | |
2c3234d1 | 2799 | dfops, spaceres); |
d31a1825 | 2800 | if (error) |
eeacd321 | 2801 | goto out_trans_abort; |
d31a1825 CM |
2802 | |
2803 | /* transfer ip2 ".." reference to dp1 */ | |
c19b3b05 | 2804 | if (!S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2805 | error = xfs_droplink(tp, dp2); |
2806 | if (error) | |
eeacd321 | 2807 | goto out_trans_abort; |
d31a1825 CM |
2808 | error = xfs_bumplink(tp, dp1); |
2809 | if (error) | |
eeacd321 | 2810 | goto out_trans_abort; |
d31a1825 CM |
2811 | } |
2812 | ||
2813 | /* | |
2814 | * Although ip1 isn't changed here, userspace needs | |
2815 | * to be warned about the change, so that applications | |
2816 | * relying on it (like backup ones), will properly | |
2817 | * notify the change | |
2818 | */ | |
2819 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2820 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2821 | } | |
2822 | ||
c19b3b05 | 2823 | if (S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2824 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, |
2825 | dp2->i_ino, first_block, | |
2c3234d1 | 2826 | dfops, spaceres); |
d31a1825 | 2827 | if (error) |
eeacd321 | 2828 | goto out_trans_abort; |
d31a1825 CM |
2829 | |
2830 | /* transfer ip1 ".." reference to dp2 */ | |
c19b3b05 | 2831 | if (!S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2832 | error = xfs_droplink(tp, dp1); |
2833 | if (error) | |
eeacd321 | 2834 | goto out_trans_abort; |
d31a1825 CM |
2835 | error = xfs_bumplink(tp, dp2); |
2836 | if (error) | |
eeacd321 | 2837 | goto out_trans_abort; |
d31a1825 CM |
2838 | } |
2839 | ||
2840 | /* | |
2841 | * Although ip2 isn't changed here, userspace needs | |
2842 | * to be warned about the change, so that applications | |
2843 | * relying on it (like backup ones), will properly | |
2844 | * notify the change | |
2845 | */ | |
2846 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2847 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2848 | } | |
2849 | } | |
2850 | ||
2851 | if (ip1_flags) { | |
2852 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2853 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2854 | } | |
2855 | if (ip2_flags) { | |
2856 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2857 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2858 | } | |
2859 | if (dp2_flags) { | |
2860 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2861 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2862 | } | |
2863 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2864 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
2c3234d1 | 2865 | return xfs_finish_rename(tp, dfops); |
eeacd321 DC |
2866 | |
2867 | out_trans_abort: | |
2c3234d1 | 2868 | xfs_defer_cancel(dfops); |
4906e215 | 2869 | xfs_trans_cancel(tp); |
d31a1825 CM |
2870 | return error; |
2871 | } | |
2872 | ||
7dcf5c3e DC |
2873 | /* |
2874 | * xfs_rename_alloc_whiteout() | |
2875 | * | |
2876 | * Return a referenced, unlinked, unlocked inode that that can be used as a | |
2877 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we | |
2878 | * crash between allocating the inode and linking it into the rename transaction | |
2879 | * recovery will free the inode and we won't leak it. | |
2880 | */ | |
2881 | static int | |
2882 | xfs_rename_alloc_whiteout( | |
2883 | struct xfs_inode *dp, | |
2884 | struct xfs_inode **wip) | |
2885 | { | |
2886 | struct xfs_inode *tmpfile; | |
2887 | int error; | |
2888 | ||
2889 | error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); | |
2890 | if (error) | |
2891 | return error; | |
2892 | ||
22419ac9 BF |
2893 | /* |
2894 | * Prepare the tmpfile inode as if it were created through the VFS. | |
2895 | * Otherwise, the link increment paths will complain about nlink 0->1. | |
2896 | * Drop the link count as done by d_tmpfile(), complete the inode setup | |
2897 | * and flag it as linkable. | |
2898 | */ | |
2899 | drop_nlink(VFS_I(tmpfile)); | |
2b3d1d41 | 2900 | xfs_setup_iops(tmpfile); |
7dcf5c3e DC |
2901 | xfs_finish_inode_setup(tmpfile); |
2902 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2903 | ||
2904 | *wip = tmpfile; | |
2905 | return 0; | |
2906 | } | |
2907 | ||
f6bba201 DC |
2908 | /* |
2909 | * xfs_rename | |
2910 | */ | |
2911 | int | |
2912 | xfs_rename( | |
7dcf5c3e DC |
2913 | struct xfs_inode *src_dp, |
2914 | struct xfs_name *src_name, | |
2915 | struct xfs_inode *src_ip, | |
2916 | struct xfs_inode *target_dp, | |
2917 | struct xfs_name *target_name, | |
2918 | struct xfs_inode *target_ip, | |
2919 | unsigned int flags) | |
f6bba201 | 2920 | { |
7dcf5c3e DC |
2921 | struct xfs_mount *mp = src_dp->i_mount; |
2922 | struct xfs_trans *tp; | |
2c3234d1 | 2923 | struct xfs_defer_ops dfops; |
7dcf5c3e DC |
2924 | xfs_fsblock_t first_block; |
2925 | struct xfs_inode *wip = NULL; /* whiteout inode */ | |
2926 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
2927 | int num_inodes = __XFS_SORT_INODES; | |
2b93681f | 2928 | bool new_parent = (src_dp != target_dp); |
c19b3b05 | 2929 | bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); |
7dcf5c3e DC |
2930 | int spaceres; |
2931 | int error; | |
f6bba201 DC |
2932 | |
2933 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2934 | ||
eeacd321 DC |
2935 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2936 | return -EINVAL; | |
2937 | ||
7dcf5c3e DC |
2938 | /* |
2939 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2940 | * we will be placing at the target and ensure the type is set | |
2941 | * appropriately. | |
2942 | */ | |
2943 | if (flags & RENAME_WHITEOUT) { | |
2944 | ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); | |
2945 | error = xfs_rename_alloc_whiteout(target_dp, &wip); | |
2946 | if (error) | |
2947 | return error; | |
2948 | ||
2949 | /* setup target dirent info as whiteout */ | |
2950 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2951 | } | |
f6bba201 | 2952 | |
7dcf5c3e | 2953 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2954 | inodes, &num_inodes); |
2955 | ||
f6bba201 | 2956 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
253f4911 | 2957 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); |
2451337d | 2958 | if (error == -ENOSPC) { |
f6bba201 | 2959 | spaceres = 0; |
253f4911 CH |
2960 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, |
2961 | &tp); | |
f6bba201 | 2962 | } |
445883e8 | 2963 | if (error) |
253f4911 | 2964 | goto out_release_wip; |
f6bba201 DC |
2965 | |
2966 | /* | |
2967 | * Attach the dquots to the inodes | |
2968 | */ | |
2969 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2970 | if (error) |
2971 | goto out_trans_cancel; | |
f6bba201 DC |
2972 | |
2973 | /* | |
2974 | * Lock all the participating inodes. Depending upon whether | |
2975 | * the target_name exists in the target directory, and | |
2976 | * whether the target directory is the same as the source | |
2977 | * directory, we can lock from 2 to 4 inodes. | |
2978 | */ | |
2979 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2980 | ||
2981 | /* | |
2982 | * Join all the inodes to the transaction. From this point on, | |
2983 | * we can rely on either trans_commit or trans_cancel to unlock | |
2984 | * them. | |
2985 | */ | |
65523218 | 2986 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
f6bba201 | 2987 | if (new_parent) |
65523218 | 2988 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
f6bba201 DC |
2989 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2990 | if (target_ip) | |
2991 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2992 | if (wip) |
2993 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
2994 | |
2995 | /* | |
2996 | * If we are using project inheritance, we only allow renames | |
2997 | * into our tree when the project IDs are the same; else the | |
2998 | * tree quota mechanism would be circumvented. | |
2999 | */ | |
3000 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
3001 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2451337d | 3002 | error = -EXDEV; |
445883e8 | 3003 | goto out_trans_cancel; |
f6bba201 DC |
3004 | } |
3005 | ||
2c3234d1 | 3006 | xfs_defer_init(&dfops, &first_block); |
445883e8 | 3007 | |
eeacd321 DC |
3008 | /* RENAME_EXCHANGE is unique from here on. */ |
3009 | if (flags & RENAME_EXCHANGE) | |
3010 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
3011 | target_dp, target_name, target_ip, | |
2c3234d1 | 3012 | &dfops, &first_block, spaceres); |
d31a1825 | 3013 | |
f6bba201 DC |
3014 | /* |
3015 | * Set up the target. | |
3016 | */ | |
3017 | if (target_ip == NULL) { | |
3018 | /* | |
3019 | * If there's no space reservation, check the entry will | |
3020 | * fit before actually inserting it. | |
3021 | */ | |
94f3cad5 ES |
3022 | if (!spaceres) { |
3023 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
3024 | if (error) | |
445883e8 | 3025 | goto out_trans_cancel; |
94f3cad5 | 3026 | } |
f6bba201 DC |
3027 | /* |
3028 | * If target does not exist and the rename crosses | |
3029 | * directories, adjust the target directory link count | |
3030 | * to account for the ".." reference from the new entry. | |
3031 | */ | |
3032 | error = xfs_dir_createname(tp, target_dp, target_name, | |
3033 | src_ip->i_ino, &first_block, | |
2c3234d1 | 3034 | &dfops, spaceres); |
f6bba201 | 3035 | if (error) |
4906e215 | 3036 | goto out_bmap_cancel; |
f6bba201 DC |
3037 | |
3038 | xfs_trans_ichgtime(tp, target_dp, | |
3039 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3040 | ||
3041 | if (new_parent && src_is_directory) { | |
3042 | error = xfs_bumplink(tp, target_dp); | |
3043 | if (error) | |
4906e215 | 3044 | goto out_bmap_cancel; |
f6bba201 DC |
3045 | } |
3046 | } else { /* target_ip != NULL */ | |
3047 | /* | |
3048 | * If target exists and it's a directory, check that both | |
3049 | * target and source are directories and that target can be | |
3050 | * destroyed, or that neither is a directory. | |
3051 | */ | |
c19b3b05 | 3052 | if (S_ISDIR(VFS_I(target_ip)->i_mode)) { |
f6bba201 DC |
3053 | /* |
3054 | * Make sure target dir is empty. | |
3055 | */ | |
3056 | if (!(xfs_dir_isempty(target_ip)) || | |
54d7b5c1 | 3057 | (VFS_I(target_ip)->i_nlink > 2)) { |
2451337d | 3058 | error = -EEXIST; |
445883e8 | 3059 | goto out_trans_cancel; |
f6bba201 DC |
3060 | } |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * Link the source inode under the target name. | |
3065 | * If the source inode is a directory and we are moving | |
3066 | * it across directories, its ".." entry will be | |
3067 | * inconsistent until we replace that down below. | |
3068 | * | |
3069 | * In case there is already an entry with the same | |
3070 | * name at the destination directory, remove it first. | |
3071 | */ | |
3072 | error = xfs_dir_replace(tp, target_dp, target_name, | |
3073 | src_ip->i_ino, | |
2c3234d1 | 3074 | &first_block, &dfops, spaceres); |
f6bba201 | 3075 | if (error) |
4906e215 | 3076 | goto out_bmap_cancel; |
f6bba201 DC |
3077 | |
3078 | xfs_trans_ichgtime(tp, target_dp, | |
3079 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3080 | ||
3081 | /* | |
3082 | * Decrement the link count on the target since the target | |
3083 | * dir no longer points to it. | |
3084 | */ | |
3085 | error = xfs_droplink(tp, target_ip); | |
3086 | if (error) | |
4906e215 | 3087 | goto out_bmap_cancel; |
f6bba201 DC |
3088 | |
3089 | if (src_is_directory) { | |
3090 | /* | |
3091 | * Drop the link from the old "." entry. | |
3092 | */ | |
3093 | error = xfs_droplink(tp, target_ip); | |
3094 | if (error) | |
4906e215 | 3095 | goto out_bmap_cancel; |
f6bba201 DC |
3096 | } |
3097 | } /* target_ip != NULL */ | |
3098 | ||
3099 | /* | |
3100 | * Remove the source. | |
3101 | */ | |
3102 | if (new_parent && src_is_directory) { | |
3103 | /* | |
3104 | * Rewrite the ".." entry to point to the new | |
3105 | * directory. | |
3106 | */ | |
3107 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
3108 | target_dp->i_ino, | |
2c3234d1 | 3109 | &first_block, &dfops, spaceres); |
2451337d | 3110 | ASSERT(error != -EEXIST); |
f6bba201 | 3111 | if (error) |
4906e215 | 3112 | goto out_bmap_cancel; |
f6bba201 DC |
3113 | } |
3114 | ||
3115 | /* | |
3116 | * We always want to hit the ctime on the source inode. | |
3117 | * | |
3118 | * This isn't strictly required by the standards since the source | |
3119 | * inode isn't really being changed, but old unix file systems did | |
3120 | * it and some incremental backup programs won't work without it. | |
3121 | */ | |
3122 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3123 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3124 | ||
3125 | /* | |
3126 | * Adjust the link count on src_dp. This is necessary when | |
3127 | * renaming a directory, either within one parent when | |
3128 | * the target existed, or across two parent directories. | |
3129 | */ | |
3130 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3131 | ||
3132 | /* | |
3133 | * Decrement link count on src_directory since the | |
3134 | * entry that's moved no longer points to it. | |
3135 | */ | |
3136 | error = xfs_droplink(tp, src_dp); | |
3137 | if (error) | |
4906e215 | 3138 | goto out_bmap_cancel; |
f6bba201 DC |
3139 | } |
3140 | ||
7dcf5c3e DC |
3141 | /* |
3142 | * For whiteouts, we only need to update the source dirent with the | |
3143 | * inode number of the whiteout inode rather than removing it | |
3144 | * altogether. | |
3145 | */ | |
3146 | if (wip) { | |
3147 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, | |
2c3234d1 | 3148 | &first_block, &dfops, spaceres); |
7dcf5c3e DC |
3149 | } else |
3150 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
2c3234d1 | 3151 | &first_block, &dfops, spaceres); |
f6bba201 | 3152 | if (error) |
4906e215 | 3153 | goto out_bmap_cancel; |
f6bba201 DC |
3154 | |
3155 | /* | |
7dcf5c3e DC |
3156 | * For whiteouts, we need to bump the link count on the whiteout inode. |
3157 | * This means that failures all the way up to this point leave the inode | |
3158 | * on the unlinked list and so cleanup is a simple matter of dropping | |
3159 | * the remaining reference to it. If we fail here after bumping the link | |
3160 | * count, we're shutting down the filesystem so we'll never see the | |
3161 | * intermediate state on disk. | |
f6bba201 | 3162 | */ |
7dcf5c3e | 3163 | if (wip) { |
54d7b5c1 | 3164 | ASSERT(VFS_I(wip)->i_nlink == 0); |
7dcf5c3e DC |
3165 | error = xfs_bumplink(tp, wip); |
3166 | if (error) | |
4906e215 | 3167 | goto out_bmap_cancel; |
7dcf5c3e DC |
3168 | error = xfs_iunlink_remove(tp, wip); |
3169 | if (error) | |
4906e215 | 3170 | goto out_bmap_cancel; |
7dcf5c3e | 3171 | xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); |
f6bba201 | 3172 | |
7dcf5c3e DC |
3173 | /* |
3174 | * Now we have a real link, clear the "I'm a tmpfile" state | |
3175 | * flag from the inode so it doesn't accidentally get misused in | |
3176 | * future. | |
3177 | */ | |
3178 | VFS_I(wip)->i_state &= ~I_LINKABLE; | |
f6bba201 DC |
3179 | } |
3180 | ||
f6bba201 DC |
3181 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3182 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3183 | if (new_parent) | |
3184 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3185 | |
2c3234d1 | 3186 | error = xfs_finish_rename(tp, &dfops); |
7dcf5c3e DC |
3187 | if (wip) |
3188 | IRELE(wip); | |
3189 | return error; | |
f6bba201 | 3190 | |
445883e8 | 3191 | out_bmap_cancel: |
2c3234d1 | 3192 | xfs_defer_cancel(&dfops); |
445883e8 | 3193 | out_trans_cancel: |
4906e215 | 3194 | xfs_trans_cancel(tp); |
253f4911 | 3195 | out_release_wip: |
7dcf5c3e DC |
3196 | if (wip) |
3197 | IRELE(wip); | |
f6bba201 DC |
3198 | return error; |
3199 | } | |
3200 | ||
5c4d97d0 DC |
3201 | STATIC int |
3202 | xfs_iflush_cluster( | |
19429363 DC |
3203 | struct xfs_inode *ip, |
3204 | struct xfs_buf *bp) | |
1da177e4 | 3205 | { |
19429363 | 3206 | struct xfs_mount *mp = ip->i_mount; |
5c4d97d0 DC |
3207 | struct xfs_perag *pag; |
3208 | unsigned long first_index, mask; | |
3209 | unsigned long inodes_per_cluster; | |
19429363 DC |
3210 | int cilist_size; |
3211 | struct xfs_inode **cilist; | |
3212 | struct xfs_inode *cip; | |
5c4d97d0 DC |
3213 | int nr_found; |
3214 | int clcount = 0; | |
3215 | int bufwasdelwri; | |
1da177e4 | 3216 | int i; |
1da177e4 | 3217 | |
5c4d97d0 | 3218 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 3219 | |
0f49efd8 | 3220 | inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; |
19429363 DC |
3221 | cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
3222 | cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); | |
3223 | if (!cilist) | |
5c4d97d0 | 3224 | goto out_put; |
1da177e4 | 3225 | |
0f49efd8 | 3226 | mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); |
5c4d97d0 DC |
3227 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
3228 | rcu_read_lock(); | |
3229 | /* really need a gang lookup range call here */ | |
19429363 | 3230 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, |
5c4d97d0 DC |
3231 | first_index, inodes_per_cluster); |
3232 | if (nr_found == 0) | |
3233 | goto out_free; | |
3234 | ||
3235 | for (i = 0; i < nr_found; i++) { | |
19429363 DC |
3236 | cip = cilist[i]; |
3237 | if (cip == ip) | |
bad55843 | 3238 | continue; |
1a3e8f3d DC |
3239 | |
3240 | /* | |
3241 | * because this is an RCU protected lookup, we could find a | |
3242 | * recently freed or even reallocated inode during the lookup. | |
3243 | * We need to check under the i_flags_lock for a valid inode | |
3244 | * here. Skip it if it is not valid or the wrong inode. | |
3245 | */ | |
19429363 DC |
3246 | spin_lock(&cip->i_flags_lock); |
3247 | if (!cip->i_ino || | |
3248 | __xfs_iflags_test(cip, XFS_ISTALE)) { | |
3249 | spin_unlock(&cip->i_flags_lock); | |
1a3e8f3d DC |
3250 | continue; |
3251 | } | |
5a90e53e DC |
3252 | |
3253 | /* | |
3254 | * Once we fall off the end of the cluster, no point checking | |
3255 | * any more inodes in the list because they will also all be | |
3256 | * outside the cluster. | |
3257 | */ | |
19429363 DC |
3258 | if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { |
3259 | spin_unlock(&cip->i_flags_lock); | |
5a90e53e DC |
3260 | break; |
3261 | } | |
19429363 | 3262 | spin_unlock(&cip->i_flags_lock); |
1a3e8f3d | 3263 | |
bad55843 DC |
3264 | /* |
3265 | * Do an un-protected check to see if the inode is dirty and | |
3266 | * is a candidate for flushing. These checks will be repeated | |
3267 | * later after the appropriate locks are acquired. | |
3268 | */ | |
19429363 | 3269 | if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) |
bad55843 | 3270 | continue; |
bad55843 DC |
3271 | |
3272 | /* | |
3273 | * Try to get locks. If any are unavailable or it is pinned, | |
3274 | * then this inode cannot be flushed and is skipped. | |
3275 | */ | |
3276 | ||
19429363 | 3277 | if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) |
bad55843 | 3278 | continue; |
19429363 DC |
3279 | if (!xfs_iflock_nowait(cip)) { |
3280 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3281 | continue; |
3282 | } | |
19429363 DC |
3283 | if (xfs_ipincount(cip)) { |
3284 | xfs_ifunlock(cip); | |
3285 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3286 | continue; |
3287 | } | |
3288 | ||
8a17d7dd DC |
3289 | |
3290 | /* | |
3291 | * Check the inode number again, just to be certain we are not | |
3292 | * racing with freeing in xfs_reclaim_inode(). See the comments | |
3293 | * in that function for more information as to why the initial | |
3294 | * check is not sufficient. | |
3295 | */ | |
19429363 DC |
3296 | if (!cip->i_ino) { |
3297 | xfs_ifunlock(cip); | |
3298 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3299 | continue; |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * arriving here means that this inode can be flushed. First | |
3304 | * re-check that it's dirty before flushing. | |
3305 | */ | |
19429363 | 3306 | if (!xfs_inode_clean(cip)) { |
33540408 | 3307 | int error; |
19429363 | 3308 | error = xfs_iflush_int(cip, bp); |
bad55843 | 3309 | if (error) { |
19429363 | 3310 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3311 | goto cluster_corrupt_out; |
3312 | } | |
3313 | clcount++; | |
3314 | } else { | |
19429363 | 3315 | xfs_ifunlock(cip); |
bad55843 | 3316 | } |
19429363 | 3317 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3318 | } |
3319 | ||
3320 | if (clcount) { | |
ff6d6af2 BD |
3321 | XFS_STATS_INC(mp, xs_icluster_flushcnt); |
3322 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
bad55843 DC |
3323 | } |
3324 | ||
3325 | out_free: | |
1a3e8f3d | 3326 | rcu_read_unlock(); |
19429363 | 3327 | kmem_free(cilist); |
44b56e0a DC |
3328 | out_put: |
3329 | xfs_perag_put(pag); | |
bad55843 DC |
3330 | return 0; |
3331 | ||
3332 | ||
3333 | cluster_corrupt_out: | |
3334 | /* | |
3335 | * Corruption detected in the clustering loop. Invalidate the | |
3336 | * inode buffer and shut down the filesystem. | |
3337 | */ | |
1a3e8f3d | 3338 | rcu_read_unlock(); |
bad55843 | 3339 | /* |
43ff2122 | 3340 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
3341 | * brelse can handle it with no problems. If not, shut down the |
3342 | * filesystem before releasing the buffer. | |
3343 | */ | |
43ff2122 | 3344 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
3345 | if (bufwasdelwri) |
3346 | xfs_buf_relse(bp); | |
3347 | ||
3348 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3349 | ||
3350 | if (!bufwasdelwri) { | |
3351 | /* | |
3352 | * Just like incore_relse: if we have b_iodone functions, | |
3353 | * mark the buffer as an error and call them. Otherwise | |
3354 | * mark it as stale and brelse. | |
3355 | */ | |
cb669ca5 | 3356 | if (bp->b_iodone) { |
b0388bf1 | 3357 | bp->b_flags &= ~XBF_DONE; |
c867cb61 | 3358 | xfs_buf_stale(bp); |
2451337d | 3359 | xfs_buf_ioerror(bp, -EIO); |
e8aaba9a | 3360 | xfs_buf_ioend(bp); |
bad55843 | 3361 | } else { |
c867cb61 | 3362 | xfs_buf_stale(bp); |
bad55843 DC |
3363 | xfs_buf_relse(bp); |
3364 | } | |
3365 | } | |
3366 | ||
3367 | /* | |
3368 | * Unlocks the flush lock | |
3369 | */ | |
19429363 DC |
3370 | xfs_iflush_abort(cip, false); |
3371 | kmem_free(cilist); | |
44b56e0a | 3372 | xfs_perag_put(pag); |
2451337d | 3373 | return -EFSCORRUPTED; |
bad55843 DC |
3374 | } |
3375 | ||
1da177e4 | 3376 | /* |
4c46819a CH |
3377 | * Flush dirty inode metadata into the backing buffer. |
3378 | * | |
3379 | * The caller must have the inode lock and the inode flush lock held. The | |
3380 | * inode lock will still be held upon return to the caller, and the inode | |
3381 | * flush lock will be released after the inode has reached the disk. | |
3382 | * | |
3383 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3384 | */ |
3385 | int | |
3386 | xfs_iflush( | |
4c46819a CH |
3387 | struct xfs_inode *ip, |
3388 | struct xfs_buf **bpp) | |
1da177e4 | 3389 | { |
4c46819a | 3390 | struct xfs_mount *mp = ip->i_mount; |
b1438f47 | 3391 | struct xfs_buf *bp = NULL; |
4c46819a | 3392 | struct xfs_dinode *dip; |
1da177e4 | 3393 | int error; |
1da177e4 | 3394 | |
ff6d6af2 | 3395 | XFS_STATS_INC(mp, xs_iflush_count); |
1da177e4 | 3396 | |
579aa9ca | 3397 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3398 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3399 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3400 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3401 | |
4c46819a | 3402 | *bpp = NULL; |
1da177e4 | 3403 | |
1da177e4 LT |
3404 | xfs_iunpin_wait(ip); |
3405 | ||
4b6a4688 DC |
3406 | /* |
3407 | * For stale inodes we cannot rely on the backing buffer remaining | |
3408 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3409 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3410 | * inodes below. We have to check this after ensuring the inode is |
3411 | * unpinned so that it is safe to reclaim the stale inode after the | |
3412 | * flush call. | |
3413 | */ | |
3414 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3415 | xfs_ifunlock(ip); | |
3416 | return 0; | |
3417 | } | |
3418 | ||
1da177e4 LT |
3419 | /* |
3420 | * This may have been unpinned because the filesystem is shutting | |
3421 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3422 | * to disk, because the log record didn't make it to disk. |
3423 | * | |
3424 | * We also have to remove the log item from the AIL in this case, | |
3425 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3426 | */ |
3427 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
2451337d | 3428 | error = -EIO; |
32ce90a4 | 3429 | goto abort_out; |
1da177e4 LT |
3430 | } |
3431 | ||
a3f74ffb | 3432 | /* |
b1438f47 DC |
3433 | * Get the buffer containing the on-disk inode. We are doing a try-lock |
3434 | * operation here, so we may get an EAGAIN error. In that case, we | |
3435 | * simply want to return with the inode still dirty. | |
3436 | * | |
3437 | * If we get any other error, we effectively have a corruption situation | |
3438 | * and we cannot flush the inode, so we treat it the same as failing | |
3439 | * xfs_iflush_int(). | |
a3f74ffb | 3440 | */ |
475ee413 CH |
3441 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3442 | 0); | |
b1438f47 | 3443 | if (error == -EAGAIN) { |
a3f74ffb DC |
3444 | xfs_ifunlock(ip); |
3445 | return error; | |
3446 | } | |
b1438f47 DC |
3447 | if (error) |
3448 | goto corrupt_out; | |
a3f74ffb | 3449 | |
1da177e4 LT |
3450 | /* |
3451 | * First flush out the inode that xfs_iflush was called with. | |
3452 | */ | |
3453 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3454 | if (error) |
1da177e4 | 3455 | goto corrupt_out; |
1da177e4 | 3456 | |
a3f74ffb DC |
3457 | /* |
3458 | * If the buffer is pinned then push on the log now so we won't | |
3459 | * get stuck waiting in the write for too long. | |
3460 | */ | |
811e64c7 | 3461 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3462 | xfs_log_force(mp, 0); |
a3f74ffb | 3463 | |
1da177e4 LT |
3464 | /* |
3465 | * inode clustering: | |
3466 | * see if other inodes can be gathered into this write | |
3467 | */ | |
bad55843 DC |
3468 | error = xfs_iflush_cluster(ip, bp); |
3469 | if (error) | |
3470 | goto cluster_corrupt_out; | |
1da177e4 | 3471 | |
4c46819a CH |
3472 | *bpp = bp; |
3473 | return 0; | |
1da177e4 LT |
3474 | |
3475 | corrupt_out: | |
b1438f47 DC |
3476 | if (bp) |
3477 | xfs_buf_relse(bp); | |
7d04a335 | 3478 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3479 | cluster_corrupt_out: |
2451337d | 3480 | error = -EFSCORRUPTED; |
32ce90a4 | 3481 | abort_out: |
1da177e4 LT |
3482 | /* |
3483 | * Unlocks the flush lock | |
3484 | */ | |
04913fdd | 3485 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3486 | return error; |
1da177e4 LT |
3487 | } |
3488 | ||
1da177e4 LT |
3489 | STATIC int |
3490 | xfs_iflush_int( | |
93848a99 CH |
3491 | struct xfs_inode *ip, |
3492 | struct xfs_buf *bp) | |
1da177e4 | 3493 | { |
93848a99 CH |
3494 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3495 | struct xfs_dinode *dip; | |
3496 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3497 | |
579aa9ca | 3498 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3499 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3500 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3501 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3502 | ASSERT(iip != NULL && iip->ili_fields != 0); |
263997a6 | 3503 | ASSERT(ip->i_d.di_version > 1); |
1da177e4 | 3504 | |
1da177e4 | 3505 | /* set *dip = inode's place in the buffer */ |
88ee2df7 | 3506 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3507 | |
69ef921b | 3508 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
9e24cfd0 | 3509 | mp, XFS_ERRTAG_IFLUSH_1)) { |
6a19d939 DC |
3510 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3511 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3512 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
3513 | goto corrupt_out; |
3514 | } | |
c19b3b05 | 3515 | if (S_ISREG(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3516 | if (XFS_TEST_ERROR( |
3517 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3518 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 3519 | mp, XFS_ERRTAG_IFLUSH_3)) { |
6a19d939 DC |
3520 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3521 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
3522 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3523 | goto corrupt_out; |
3524 | } | |
c19b3b05 | 3525 | } else if (S_ISDIR(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3526 | if (XFS_TEST_ERROR( |
3527 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3528 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3529 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
9e24cfd0 | 3530 | mp, XFS_ERRTAG_IFLUSH_4)) { |
6a19d939 DC |
3531 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3532 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
3533 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3534 | goto corrupt_out; |
3535 | } | |
3536 | } | |
3537 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
9e24cfd0 | 3538 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { |
6a19d939 DC |
3539 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3540 | "%s: detected corrupt incore inode %Lu, " | |
3541 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3542 | __func__, ip->i_ino, | |
1da177e4 | 3543 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3544 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3545 | goto corrupt_out; |
3546 | } | |
3547 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
9e24cfd0 | 3548 | mp, XFS_ERRTAG_IFLUSH_6)) { |
6a19d939 DC |
3549 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3550 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3551 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
3552 | goto corrupt_out; |
3553 | } | |
e60896d8 | 3554 | |
1da177e4 | 3555 | /* |
263997a6 | 3556 | * Inode item log recovery for v2 inodes are dependent on the |
e60896d8 DC |
3557 | * di_flushiter count for correct sequencing. We bump the flush |
3558 | * iteration count so we can detect flushes which postdate a log record | |
3559 | * during recovery. This is redundant as we now log every change and | |
3560 | * hence this can't happen but we need to still do it to ensure | |
3561 | * backwards compatibility with old kernels that predate logging all | |
3562 | * inode changes. | |
1da177e4 | 3563 | */ |
e60896d8 DC |
3564 | if (ip->i_d.di_version < 3) |
3565 | ip->i_d.di_flushiter++; | |
1da177e4 | 3566 | |
005c5db8 DW |
3567 | /* Check the inline directory data. */ |
3568 | if (S_ISDIR(VFS_I(ip)->i_mode) && | |
3569 | ip->i_d.di_format == XFS_DINODE_FMT_LOCAL && | |
3570 | xfs_dir2_sf_verify(ip)) | |
3571 | goto corrupt_out; | |
3572 | ||
1da177e4 | 3573 | /* |
3987848c DC |
3574 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3575 | * copy out the core of the inode, because if the inode is dirty at all | |
3576 | * the core must be. | |
1da177e4 | 3577 | */ |
93f958f9 | 3578 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3579 | |
3580 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3581 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3582 | ip->i_d.di_flushiter = 0; | |
3583 | ||
005c5db8 DW |
3584 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
3585 | if (XFS_IFORK_Q(ip)) | |
3586 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); | |
1da177e4 LT |
3587 | xfs_inobp_check(mp, bp); |
3588 | ||
3589 | /* | |
f5d8d5c4 CH |
3590 | * We've recorded everything logged in the inode, so we'd like to clear |
3591 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3592 | * However, we can't stop logging all this information until the data | |
3593 | * we've copied into the disk buffer is written to disk. If we did we | |
3594 | * might overwrite the copy of the inode in the log with all the data | |
3595 | * after re-logging only part of it, and in the face of a crash we | |
3596 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3597 | * |
f5d8d5c4 CH |
3598 | * What we do is move the bits to the ili_last_fields field. When |
3599 | * logging the inode, these bits are moved back to the ili_fields field. | |
3600 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3601 | * know that the information those bits represent is permanently on | |
3602 | * disk. As long as the flush completes before the inode is logged | |
3603 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3604 | * |
f5d8d5c4 CH |
3605 | * We can play with the ili_fields bits here, because the inode lock |
3606 | * must be held exclusively in order to set bits there and the flush | |
3607 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3608 | * done routine can tell whether or not to look in the AIL. Also, store | |
3609 | * the current LSN of the inode so that we can tell whether the item has | |
3610 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3611 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3612 | * atomically. | |
1da177e4 | 3613 | */ |
93848a99 CH |
3614 | iip->ili_last_fields = iip->ili_fields; |
3615 | iip->ili_fields = 0; | |
fc0561ce | 3616 | iip->ili_fsync_fields = 0; |
93848a99 | 3617 | iip->ili_logged = 1; |
1da177e4 | 3618 | |
93848a99 CH |
3619 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3620 | &iip->ili_item.li_lsn); | |
1da177e4 | 3621 | |
93848a99 CH |
3622 | /* |
3623 | * Attach the function xfs_iflush_done to the inode's | |
3624 | * buffer. This will remove the inode from the AIL | |
3625 | * and unlock the inode's flush lock when the inode is | |
3626 | * completely written to disk. | |
3627 | */ | |
3628 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3629 | |
93848a99 CH |
3630 | /* generate the checksum. */ |
3631 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3632 | |
93848a99 CH |
3633 | ASSERT(bp->b_fspriv != NULL); |
3634 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
3635 | return 0; |
3636 | ||
3637 | corrupt_out: | |
2451337d | 3638 | return -EFSCORRUPTED; |
1da177e4 | 3639 | } |