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" |
a4fbe6ab | 28 | #include "xfs_inode.h" |
57062787 | 29 | #include "xfs_da_format.h" |
c24b5dfa | 30 | #include "xfs_da_btree.h" |
c24b5dfa | 31 | #include "xfs_dir2.h" |
a844f451 | 32 | #include "xfs_attr_sf.h" |
c24b5dfa | 33 | #include "xfs_attr.h" |
239880ef DC |
34 | #include "xfs_trans_space.h" |
35 | #include "xfs_trans.h" | |
1da177e4 | 36 | #include "xfs_buf_item.h" |
a844f451 | 37 | #include "xfs_inode_item.h" |
a844f451 NS |
38 | #include "xfs_ialloc.h" |
39 | #include "xfs_bmap.h" | |
68988114 | 40 | #include "xfs_bmap_util.h" |
1da177e4 | 41 | #include "xfs_error.h" |
1da177e4 | 42 | #include "xfs_quota.h" |
2a82b8be | 43 | #include "xfs_filestream.h" |
93848a99 | 44 | #include "xfs_cksum.h" |
0b1b213f | 45 | #include "xfs_trace.h" |
33479e05 | 46 | #include "xfs_icache.h" |
c24b5dfa | 47 | #include "xfs_symlink.h" |
239880ef DC |
48 | #include "xfs_trans_priv.h" |
49 | #include "xfs_log.h" | |
a4fbe6ab | 50 | #include "xfs_bmap_btree.h" |
1da177e4 | 51 | |
1da177e4 | 52 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
53 | |
54 | /* | |
8f04c47a | 55 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
56 | * freed from a file in a single transaction. |
57 | */ | |
58 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
59 | ||
54d7b5c1 DC |
60 | STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); |
61 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); | |
62 | STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); | |
ab297431 | 63 | |
2a0ec1d9 DC |
64 | /* |
65 | * helper function to extract extent size hint from inode | |
66 | */ | |
67 | xfs_extlen_t | |
68 | xfs_get_extsz_hint( | |
69 | struct xfs_inode *ip) | |
70 | { | |
71 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
72 | return ip->i_d.di_extsize; | |
73 | if (XFS_IS_REALTIME_INODE(ip)) | |
74 | return ip->i_mount->m_sb.sb_rextsize; | |
75 | return 0; | |
76 | } | |
77 | ||
fa96acad | 78 | /* |
efa70be1 CH |
79 | * These two are wrapper routines around the xfs_ilock() routine used to |
80 | * centralize some grungy code. They are used in places that wish to lock the | |
81 | * inode solely for reading the extents. The reason these places can't just | |
82 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
83 | * bringing in of the extents from disk for a file in b-tree format. If the | |
84 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
85 | * the extents are read in. Locking it exclusively all the time would limit | |
86 | * our parallelism unnecessarily, though. What we do instead is check to see | |
87 | * if the extents have been read in yet, and only lock the inode exclusively | |
88 | * if they have not. | |
fa96acad | 89 | * |
efa70be1 | 90 | * The functions return a value which should be given to the corresponding |
01f4f327 | 91 | * xfs_iunlock() call. |
fa96acad DC |
92 | */ |
93 | uint | |
309ecac8 CH |
94 | xfs_ilock_data_map_shared( |
95 | struct xfs_inode *ip) | |
fa96acad | 96 | { |
309ecac8 | 97 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 98 | |
309ecac8 CH |
99 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && |
100 | (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) | |
fa96acad | 101 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 102 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
103 | return lock_mode; |
104 | } | |
105 | ||
efa70be1 CH |
106 | uint |
107 | xfs_ilock_attr_map_shared( | |
108 | struct xfs_inode *ip) | |
fa96acad | 109 | { |
efa70be1 CH |
110 | uint lock_mode = XFS_ILOCK_SHARED; |
111 | ||
112 | if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && | |
113 | (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) | |
114 | lock_mode = XFS_ILOCK_EXCL; | |
115 | xfs_ilock(ip, lock_mode); | |
116 | return lock_mode; | |
fa96acad DC |
117 | } |
118 | ||
119 | /* | |
653c60b6 DC |
120 | * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and |
121 | * the i_lock. This routine allows various combinations of the locks to be | |
122 | * obtained. | |
fa96acad | 123 | * |
653c60b6 DC |
124 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
125 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 126 | * |
653c60b6 DC |
127 | * Basic locking order: |
128 | * | |
129 | * i_iolock -> i_mmap_lock -> page_lock -> i_ilock | |
130 | * | |
131 | * mmap_sem locking order: | |
132 | * | |
133 | * i_iolock -> page lock -> mmap_sem | |
134 | * mmap_sem -> i_mmap_lock -> page_lock | |
135 | * | |
136 | * The difference in mmap_sem locking order mean that we cannot hold the | |
137 | * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can | |
138 | * fault in pages during copy in/out (for buffered IO) or require the mmap_sem | |
139 | * in get_user_pages() to map the user pages into the kernel address space for | |
140 | * direct IO. Similarly the i_iolock cannot be taken inside a page fault because | |
141 | * page faults already hold the mmap_sem. | |
142 | * | |
143 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
144 | * take both the i_iolock and the i_mmap_lock. These locks should *only* be both | |
145 | * taken in places where we need to invalidate the page cache in a race | |
146 | * free manner (e.g. truncate, hole punch and other extent manipulation | |
147 | * functions). | |
fa96acad DC |
148 | */ |
149 | void | |
150 | xfs_ilock( | |
151 | xfs_inode_t *ip, | |
152 | uint lock_flags) | |
153 | { | |
154 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
155 | ||
156 | /* | |
157 | * You can't set both SHARED and EXCL for the same lock, | |
158 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
159 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
160 | */ | |
161 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
162 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
163 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
164 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
165 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
166 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 167 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
168 | |
169 | if (lock_flags & XFS_IOLOCK_EXCL) | |
170 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
171 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
172 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
173 | ||
653c60b6 DC |
174 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
175 | mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
176 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
177 | mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
178 | ||
fa96acad DC |
179 | if (lock_flags & XFS_ILOCK_EXCL) |
180 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
181 | else if (lock_flags & XFS_ILOCK_SHARED) | |
182 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
183 | } | |
184 | ||
185 | /* | |
186 | * This is just like xfs_ilock(), except that the caller | |
187 | * is guaranteed not to sleep. It returns 1 if it gets | |
188 | * the requested locks and 0 otherwise. If the IO lock is | |
189 | * obtained but the inode lock cannot be, then the IO lock | |
190 | * is dropped before returning. | |
191 | * | |
192 | * ip -- the inode being locked | |
193 | * lock_flags -- this parameter indicates the inode's locks to be | |
194 | * to be locked. See the comment for xfs_ilock() for a list | |
195 | * of valid values. | |
196 | */ | |
197 | int | |
198 | xfs_ilock_nowait( | |
199 | xfs_inode_t *ip, | |
200 | uint lock_flags) | |
201 | { | |
202 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
203 | ||
204 | /* | |
205 | * You can't set both SHARED and EXCL for the same lock, | |
206 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
207 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
208 | */ | |
209 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
210 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
211 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
212 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
213 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
214 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 215 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
216 | |
217 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
218 | if (!mrtryupdate(&ip->i_iolock)) | |
219 | goto out; | |
220 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
221 | if (!mrtryaccess(&ip->i_iolock)) | |
222 | goto out; | |
223 | } | |
653c60b6 DC |
224 | |
225 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
226 | if (!mrtryupdate(&ip->i_mmaplock)) | |
227 | goto out_undo_iolock; | |
228 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
229 | if (!mrtryaccess(&ip->i_mmaplock)) | |
230 | goto out_undo_iolock; | |
231 | } | |
232 | ||
fa96acad DC |
233 | if (lock_flags & XFS_ILOCK_EXCL) { |
234 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 235 | goto out_undo_mmaplock; |
fa96acad DC |
236 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
237 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 238 | goto out_undo_mmaplock; |
fa96acad DC |
239 | } |
240 | return 1; | |
241 | ||
653c60b6 DC |
242 | out_undo_mmaplock: |
243 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
244 | mrunlock_excl(&ip->i_mmaplock); | |
245 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
246 | mrunlock_shared(&ip->i_mmaplock); | |
247 | out_undo_iolock: | |
fa96acad DC |
248 | if (lock_flags & XFS_IOLOCK_EXCL) |
249 | mrunlock_excl(&ip->i_iolock); | |
250 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
251 | mrunlock_shared(&ip->i_iolock); | |
653c60b6 | 252 | out: |
fa96acad DC |
253 | return 0; |
254 | } | |
255 | ||
256 | /* | |
257 | * xfs_iunlock() is used to drop the inode locks acquired with | |
258 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
259 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
260 | * that we know which locks to drop. | |
261 | * | |
262 | * ip -- the inode being unlocked | |
263 | * lock_flags -- this parameter indicates the inode's locks to be | |
264 | * to be unlocked. See the comment for xfs_ilock() for a list | |
265 | * of valid values for this parameter. | |
266 | * | |
267 | */ | |
268 | void | |
269 | xfs_iunlock( | |
270 | xfs_inode_t *ip, | |
271 | uint lock_flags) | |
272 | { | |
273 | /* | |
274 | * You can't set both SHARED and EXCL for the same lock, | |
275 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
276 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
277 | */ | |
278 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
279 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
280 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
281 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
282 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
283 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 284 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
285 | ASSERT(lock_flags != 0); |
286 | ||
287 | if (lock_flags & XFS_IOLOCK_EXCL) | |
288 | mrunlock_excl(&ip->i_iolock); | |
289 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
290 | mrunlock_shared(&ip->i_iolock); | |
291 | ||
653c60b6 DC |
292 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
293 | mrunlock_excl(&ip->i_mmaplock); | |
294 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
295 | mrunlock_shared(&ip->i_mmaplock); | |
296 | ||
fa96acad DC |
297 | if (lock_flags & XFS_ILOCK_EXCL) |
298 | mrunlock_excl(&ip->i_lock); | |
299 | else if (lock_flags & XFS_ILOCK_SHARED) | |
300 | mrunlock_shared(&ip->i_lock); | |
301 | ||
302 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
303 | } | |
304 | ||
305 | /* | |
306 | * give up write locks. the i/o lock cannot be held nested | |
307 | * if it is being demoted. | |
308 | */ | |
309 | void | |
310 | xfs_ilock_demote( | |
311 | xfs_inode_t *ip, | |
312 | uint lock_flags) | |
313 | { | |
653c60b6 DC |
314 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
315 | ASSERT((lock_flags & | |
316 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
317 | |
318 | if (lock_flags & XFS_ILOCK_EXCL) | |
319 | mrdemote(&ip->i_lock); | |
653c60b6 DC |
320 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
321 | mrdemote(&ip->i_mmaplock); | |
fa96acad DC |
322 | if (lock_flags & XFS_IOLOCK_EXCL) |
323 | mrdemote(&ip->i_iolock); | |
324 | ||
325 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
326 | } | |
327 | ||
742ae1e3 | 328 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
329 | int |
330 | xfs_isilocked( | |
331 | xfs_inode_t *ip, | |
332 | uint lock_flags) | |
333 | { | |
334 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
335 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
336 | return !!ip->i_lock.mr_writer; | |
337 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
338 | } | |
339 | ||
653c60b6 DC |
340 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
341 | if (!(lock_flags & XFS_MMAPLOCK_SHARED)) | |
342 | return !!ip->i_mmaplock.mr_writer; | |
343 | return rwsem_is_locked(&ip->i_mmaplock.mr_lock); | |
344 | } | |
345 | ||
fa96acad DC |
346 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
347 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
348 | return !!ip->i_iolock.mr_writer; | |
349 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
350 | } | |
351 | ||
352 | ASSERT(0); | |
353 | return 0; | |
354 | } | |
355 | #endif | |
356 | ||
c24b5dfa DC |
357 | #ifdef DEBUG |
358 | int xfs_locked_n; | |
359 | int xfs_small_retries; | |
360 | int xfs_middle_retries; | |
361 | int xfs_lots_retries; | |
362 | int xfs_lock_delays; | |
363 | #endif | |
364 | ||
b6a9947e DC |
365 | /* |
366 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
367 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
368 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
369 | * errors and warnings. | |
370 | */ | |
371 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
372 | static bool |
373 | xfs_lockdep_subclass_ok( | |
374 | int subclass) | |
375 | { | |
376 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
377 | } | |
378 | #else | |
379 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
380 | #endif | |
381 | ||
c24b5dfa | 382 | /* |
653c60b6 | 383 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
384 | * value. This can be called for any type of inode lock combination, including |
385 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
386 | * storage fields in the class mask we build. | |
c24b5dfa DC |
387 | */ |
388 | static inline int | |
389 | xfs_lock_inumorder(int lock_mode, int subclass) | |
390 | { | |
0952c818 DC |
391 | int class = 0; |
392 | ||
393 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
394 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 395 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 396 | |
653c60b6 | 397 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 398 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
3403ccc0 DC |
399 | ASSERT(xfs_lockdep_subclass_ok(subclass + |
400 | XFS_IOLOCK_PARENT_VAL)); | |
0952c818 DC |
401 | class += subclass << XFS_IOLOCK_SHIFT; |
402 | if (lock_mode & XFS_IOLOCK_PARENT) | |
403 | class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT; | |
653c60b6 DC |
404 | } |
405 | ||
406 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
407 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
408 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
409 | } |
410 | ||
0952c818 DC |
411 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
412 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
413 | class += subclass << XFS_ILOCK_SHIFT; | |
414 | } | |
c24b5dfa | 415 | |
0952c818 | 416 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
417 | } |
418 | ||
419 | /* | |
95afcf5c DC |
420 | * The following routine will lock n inodes in exclusive mode. We assume the |
421 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 422 | * |
95afcf5c DC |
423 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
424 | * start waiting for another inode that is locked by a thread in a long running | |
425 | * transaction (such as truncate). This can result in deadlock since the long | |
426 | * running trans might need to wait for the inode we just locked in order to | |
427 | * push the tail and free space in the log. | |
0952c818 DC |
428 | * |
429 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
430 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
431 | * lock more than one at a time, lockdep will report false positives saying we | |
432 | * have violated locking orders. | |
c24b5dfa DC |
433 | */ |
434 | void | |
435 | xfs_lock_inodes( | |
436 | xfs_inode_t **ips, | |
437 | int inodes, | |
438 | uint lock_mode) | |
439 | { | |
440 | int attempts = 0, i, j, try_lock; | |
441 | xfs_log_item_t *lp; | |
442 | ||
0952c818 DC |
443 | /* |
444 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
445 | * support an arbitrary depth of locking here, but absolute limits on | |
446 | * inodes depend on the the type of locking and the limits placed by | |
447 | * lockdep annotations in xfs_lock_inumorder. These are all checked by | |
448 | * the asserts. | |
449 | */ | |
95afcf5c | 450 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
451 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
452 | XFS_ILOCK_EXCL)); | |
453 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
454 | XFS_ILOCK_SHARED))); | |
455 | ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) || | |
456 | inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1); | |
457 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || | |
458 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
459 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
460 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
461 | ||
462 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
463 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
464 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
465 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa DC |
466 | |
467 | try_lock = 0; | |
468 | i = 0; | |
c24b5dfa DC |
469 | again: |
470 | for (; i < inodes; i++) { | |
471 | ASSERT(ips[i]); | |
472 | ||
95afcf5c | 473 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
474 | continue; |
475 | ||
476 | /* | |
95afcf5c DC |
477 | * If try_lock is not set yet, make sure all locked inodes are |
478 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 479 | */ |
c24b5dfa DC |
480 | if (!try_lock) { |
481 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
482 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
95afcf5c | 483 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) |
c24b5dfa | 484 | try_lock++; |
c24b5dfa DC |
485 | } |
486 | } | |
487 | ||
488 | /* | |
489 | * If any of the previous locks we have locked is in the AIL, | |
490 | * we must TRY to get the second and subsequent locks. If | |
491 | * we can't get any, we must release all we have | |
492 | * and try again. | |
493 | */ | |
95afcf5c DC |
494 | if (!try_lock) { |
495 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
496 | continue; | |
497 | } | |
498 | ||
499 | /* try_lock means we have an inode locked that is in the AIL. */ | |
500 | ASSERT(i != 0); | |
501 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
502 | continue; | |
c24b5dfa | 503 | |
95afcf5c DC |
504 | /* |
505 | * Unlock all previous guys and try again. xfs_iunlock will try | |
506 | * to push the tail if the inode is in the AIL. | |
507 | */ | |
508 | attempts++; | |
509 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 510 | /* |
95afcf5c DC |
511 | * Check to see if we've already unlocked this one. Not |
512 | * the first one going back, and the inode ptr is the | |
513 | * same. | |
c24b5dfa | 514 | */ |
95afcf5c DC |
515 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
516 | continue; | |
c24b5dfa | 517 | |
95afcf5c DC |
518 | xfs_iunlock(ips[j], lock_mode); |
519 | } | |
c24b5dfa | 520 | |
95afcf5c DC |
521 | if ((attempts % 5) == 0) { |
522 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 523 | #ifdef DEBUG |
95afcf5c | 524 | xfs_lock_delays++; |
c24b5dfa | 525 | #endif |
c24b5dfa | 526 | } |
95afcf5c DC |
527 | i = 0; |
528 | try_lock = 0; | |
529 | goto again; | |
c24b5dfa DC |
530 | } |
531 | ||
532 | #ifdef DEBUG | |
533 | if (attempts) { | |
534 | if (attempts < 5) xfs_small_retries++; | |
535 | else if (attempts < 100) xfs_middle_retries++; | |
536 | else xfs_lots_retries++; | |
537 | } else { | |
538 | xfs_locked_n++; | |
539 | } | |
540 | #endif | |
541 | } | |
542 | ||
543 | /* | |
653c60b6 DC |
544 | * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - |
545 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
546 | * lock more than one at a time, lockdep will report false positives saying we | |
547 | * have violated locking orders. | |
c24b5dfa DC |
548 | */ |
549 | void | |
550 | xfs_lock_two_inodes( | |
551 | xfs_inode_t *ip0, | |
552 | xfs_inode_t *ip1, | |
553 | uint lock_mode) | |
554 | { | |
555 | xfs_inode_t *temp; | |
556 | int attempts = 0; | |
557 | xfs_log_item_t *lp; | |
558 | ||
653c60b6 DC |
559 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
560 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); | |
561 | ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
562 | } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) | |
563 | ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
564 | ||
c24b5dfa DC |
565 | ASSERT(ip0->i_ino != ip1->i_ino); |
566 | ||
567 | if (ip0->i_ino > ip1->i_ino) { | |
568 | temp = ip0; | |
569 | ip0 = ip1; | |
570 | ip1 = temp; | |
571 | } | |
572 | ||
573 | again: | |
574 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
575 | ||
576 | /* | |
577 | * If the first lock we have locked is in the AIL, we must TRY to get | |
578 | * the second lock. If we can't get it, we must release the first one | |
579 | * and try again. | |
580 | */ | |
581 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
582 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
583 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
584 | xfs_iunlock(ip0, lock_mode); | |
585 | if ((++attempts % 5) == 0) | |
586 | delay(1); /* Don't just spin the CPU */ | |
587 | goto again; | |
588 | } | |
589 | } else { | |
590 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
591 | } | |
592 | } | |
593 | ||
594 | ||
fa96acad DC |
595 | void |
596 | __xfs_iflock( | |
597 | struct xfs_inode *ip) | |
598 | { | |
599 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
600 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
601 | ||
602 | do { | |
603 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
604 | if (xfs_isiflocked(ip)) | |
605 | io_schedule(); | |
606 | } while (!xfs_iflock_nowait(ip)); | |
607 | ||
608 | finish_wait(wq, &wait.wait); | |
609 | } | |
610 | ||
1da177e4 LT |
611 | STATIC uint |
612 | _xfs_dic2xflags( | |
58f88ca2 DC |
613 | __uint16_t di_flags, |
614 | uint64_t di_flags2, | |
615 | bool has_attr) | |
1da177e4 LT |
616 | { |
617 | uint flags = 0; | |
618 | ||
619 | if (di_flags & XFS_DIFLAG_ANY) { | |
620 | if (di_flags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 621 | flags |= FS_XFLAG_REALTIME; |
1da177e4 | 622 | if (di_flags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 623 | flags |= FS_XFLAG_PREALLOC; |
1da177e4 | 624 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 625 | flags |= FS_XFLAG_IMMUTABLE; |
1da177e4 | 626 | if (di_flags & XFS_DIFLAG_APPEND) |
e7b89481 | 627 | flags |= FS_XFLAG_APPEND; |
1da177e4 | 628 | if (di_flags & XFS_DIFLAG_SYNC) |
e7b89481 | 629 | flags |= FS_XFLAG_SYNC; |
1da177e4 | 630 | if (di_flags & XFS_DIFLAG_NOATIME) |
e7b89481 | 631 | flags |= FS_XFLAG_NOATIME; |
1da177e4 | 632 | if (di_flags & XFS_DIFLAG_NODUMP) |
e7b89481 | 633 | flags |= FS_XFLAG_NODUMP; |
1da177e4 | 634 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 635 | flags |= FS_XFLAG_RTINHERIT; |
1da177e4 | 636 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 637 | flags |= FS_XFLAG_PROJINHERIT; |
1da177e4 | 638 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 639 | flags |= FS_XFLAG_NOSYMLINKS; |
dd9f438e | 640 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 641 | flags |= FS_XFLAG_EXTSIZE; |
dd9f438e | 642 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 643 | flags |= FS_XFLAG_EXTSZINHERIT; |
d3446eac | 644 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 645 | flags |= FS_XFLAG_NODEFRAG; |
2a82b8be | 646 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 647 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
648 | } |
649 | ||
58f88ca2 DC |
650 | if (di_flags2 & XFS_DIFLAG2_ANY) { |
651 | if (di_flags2 & XFS_DIFLAG2_DAX) | |
652 | flags |= FS_XFLAG_DAX; | |
653 | } | |
654 | ||
655 | if (has_attr) | |
656 | flags |= FS_XFLAG_HASATTR; | |
657 | ||
1da177e4 LT |
658 | return flags; |
659 | } | |
660 | ||
661 | uint | |
662 | xfs_ip2xflags( | |
58f88ca2 | 663 | struct xfs_inode *ip) |
1da177e4 | 664 | { |
58f88ca2 | 665 | struct xfs_icdinode *dic = &ip->i_d; |
1da177e4 | 666 | |
58f88ca2 | 667 | return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); |
1da177e4 LT |
668 | } |
669 | ||
670 | uint | |
671 | xfs_dic2xflags( | |
58f88ca2 | 672 | struct xfs_dinode *dip) |
1da177e4 | 673 | { |
58f88ca2 DC |
674 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags), |
675 | be64_to_cpu(dip->di_flags2), XFS_DFORK_Q(dip)); | |
1da177e4 LT |
676 | } |
677 | ||
c24b5dfa DC |
678 | /* |
679 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
680 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
681 | * ci_name->name will point to a the actual name (caller must free) or | |
682 | * will be set to NULL if an exact match is found. | |
683 | */ | |
684 | int | |
685 | xfs_lookup( | |
686 | xfs_inode_t *dp, | |
687 | struct xfs_name *name, | |
688 | xfs_inode_t **ipp, | |
689 | struct xfs_name *ci_name) | |
690 | { | |
691 | xfs_ino_t inum; | |
692 | int error; | |
c24b5dfa DC |
693 | |
694 | trace_xfs_lookup(dp, name); | |
695 | ||
696 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
2451337d | 697 | return -EIO; |
c24b5dfa | 698 | |
dbad7c99 | 699 | xfs_ilock(dp, XFS_IOLOCK_SHARED); |
c24b5dfa | 700 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 701 | if (error) |
dbad7c99 | 702 | goto out_unlock; |
c24b5dfa DC |
703 | |
704 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
705 | if (error) | |
706 | goto out_free_name; | |
707 | ||
dbad7c99 | 708 | xfs_iunlock(dp, XFS_IOLOCK_SHARED); |
c24b5dfa DC |
709 | return 0; |
710 | ||
711 | out_free_name: | |
712 | if (ci_name) | |
713 | kmem_free(ci_name->name); | |
dbad7c99 DC |
714 | out_unlock: |
715 | xfs_iunlock(dp, XFS_IOLOCK_SHARED); | |
c24b5dfa DC |
716 | *ipp = NULL; |
717 | return error; | |
718 | } | |
719 | ||
1da177e4 LT |
720 | /* |
721 | * Allocate an inode on disk and return a copy of its in-core version. | |
722 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
723 | * appropriately within the inode. The uid and gid for the inode are | |
724 | * set according to the contents of the given cred structure. | |
725 | * | |
726 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
727 | * has a free inode available, call xfs_iget() to obtain the in-core |
728 | * version of the allocated inode. Finally, fill in the inode and | |
729 | * log its initial contents. In this case, ialloc_context would be | |
730 | * set to NULL. | |
1da177e4 | 731 | * |
cd856db6 CM |
732 | * If xfs_dialloc() does not have an available inode, it will replenish |
733 | * its supply by doing an allocation. Since we can only do one | |
734 | * allocation within a transaction without deadlocks, we must commit | |
735 | * the current transaction before returning the inode itself. | |
736 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
737 | * The caller should then commit the current transaction, start a new |
738 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
739 | * | |
740 | * To ensure that some other process does not grab the inode that | |
741 | * was allocated during the first call to xfs_ialloc(), this routine | |
742 | * also returns the [locked] bp pointing to the head of the freelist | |
743 | * as ialloc_context. The caller should hold this buffer across | |
744 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
745 | * |
746 | * If we are allocating quota inodes, we do not have a parent inode | |
747 | * to attach to or associate with (i.e. pip == NULL) because they | |
748 | * are not linked into the directory structure - they are attached | |
749 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
750 | */ |
751 | int | |
752 | xfs_ialloc( | |
753 | xfs_trans_t *tp, | |
754 | xfs_inode_t *pip, | |
576b1d67 | 755 | umode_t mode, |
31b084ae | 756 | xfs_nlink_t nlink, |
1da177e4 | 757 | xfs_dev_t rdev, |
6743099c | 758 | prid_t prid, |
1da177e4 LT |
759 | int okalloc, |
760 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
761 | xfs_inode_t **ipp) |
762 | { | |
93848a99 | 763 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
764 | xfs_ino_t ino; |
765 | xfs_inode_t *ip; | |
1da177e4 LT |
766 | uint flags; |
767 | int error; | |
e076b0f3 | 768 | struct timespec tv; |
3987848c | 769 | struct inode *inode; |
1da177e4 LT |
770 | |
771 | /* | |
772 | * Call the space management code to pick | |
773 | * the on-disk inode to be allocated. | |
774 | */ | |
b11f94d5 | 775 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 776 | ialloc_context, &ino); |
bf904248 | 777 | if (error) |
1da177e4 | 778 | return error; |
08358906 | 779 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
780 | *ipp = NULL; |
781 | return 0; | |
782 | } | |
783 | ASSERT(*ialloc_context == NULL); | |
784 | ||
785 | /* | |
786 | * Get the in-core inode with the lock held exclusively. | |
787 | * This is because we're setting fields here we need | |
788 | * to prevent others from looking at until we're done. | |
789 | */ | |
93848a99 | 790 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 791 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 792 | if (error) |
1da177e4 | 793 | return error; |
1da177e4 | 794 | ASSERT(ip != NULL); |
3987848c | 795 | inode = VFS_I(ip); |
1da177e4 | 796 | |
263997a6 DC |
797 | /* |
798 | * We always convert v1 inodes to v2 now - we only support filesystems | |
799 | * with >= v2 inode capability, so there is no reason for ever leaving | |
800 | * an inode in v1 format. | |
801 | */ | |
802 | if (ip->i_d.di_version == 1) | |
803 | ip->i_d.di_version = 2; | |
804 | ||
576b1d67 | 805 | ip->i_d.di_mode = mode; |
54d7b5c1 | 806 | set_nlink(inode, nlink); |
7aab1b28 DE |
807 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
808 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
6743099c | 809 | xfs_set_projid(ip, prid); |
1da177e4 | 810 | |
bd186aa9 | 811 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 812 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 813 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
814 | ip->i_d.di_mode |= S_ISGID; |
815 | } | |
816 | } | |
817 | ||
818 | /* | |
819 | * If the group ID of the new file does not match the effective group | |
820 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
821 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
822 | */ | |
823 | if ((irix_sgid_inherit) && | |
824 | (ip->i_d.di_mode & S_ISGID) && | |
7aab1b28 | 825 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) { |
1da177e4 LT |
826 | ip->i_d.di_mode &= ~S_ISGID; |
827 | } | |
828 | ||
829 | ip->i_d.di_size = 0; | |
830 | ip->i_d.di_nextents = 0; | |
831 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 | 832 | |
e076b0f3 | 833 | tv = current_fs_time(mp->m_super); |
3987848c DC |
834 | inode->i_mtime = tv; |
835 | inode->i_atime = tv; | |
836 | inode->i_ctime = tv; | |
dff35fd4 | 837 | |
1da177e4 LT |
838 | /* |
839 | * di_gen will have been taken care of in xfs_iread. | |
840 | */ | |
841 | ip->i_d.di_extsize = 0; | |
842 | ip->i_d.di_dmevmask = 0; | |
843 | ip->i_d.di_dmstate = 0; | |
844 | ip->i_d.di_flags = 0; | |
93848a99 CH |
845 | |
846 | if (ip->i_d.di_version == 3) { | |
93848a99 | 847 | ip->i_d.di_changecount = 1; |
93848a99 | 848 | ip->i_d.di_flags2 = 0; |
3987848c DC |
849 | ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec; |
850 | ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec; | |
93848a99 CH |
851 | } |
852 | ||
853 | ||
1da177e4 LT |
854 | flags = XFS_ILOG_CORE; |
855 | switch (mode & S_IFMT) { | |
856 | case S_IFIFO: | |
857 | case S_IFCHR: | |
858 | case S_IFBLK: | |
859 | case S_IFSOCK: | |
860 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
861 | ip->i_df.if_u2.if_rdev = rdev; | |
862 | ip->i_df.if_flags = 0; | |
863 | flags |= XFS_ILOG_DEV; | |
864 | break; | |
865 | case S_IFREG: | |
866 | case S_IFDIR: | |
b11f94d5 | 867 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
58f88ca2 DC |
868 | uint64_t di_flags2 = 0; |
869 | uint di_flags = 0; | |
365ca83d | 870 | |
abbede1b | 871 | if (S_ISDIR(mode)) { |
365ca83d NS |
872 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
873 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
874 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
875 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
876 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
877 | } | |
9336e3a7 DC |
878 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
879 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
abbede1b | 880 | } else if (S_ISREG(mode)) { |
613d7043 | 881 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 882 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
883 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
884 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
885 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
886 | } | |
1da177e4 LT |
887 | } |
888 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
889 | xfs_inherit_noatime) | |
365ca83d | 890 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
891 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
892 | xfs_inherit_nodump) | |
365ca83d | 893 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
894 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
895 | xfs_inherit_sync) | |
365ca83d | 896 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
897 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
898 | xfs_inherit_nosymlinks) | |
365ca83d | 899 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
d3446eac BN |
900 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
901 | xfs_inherit_nodefrag) | |
902 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
903 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
904 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
58f88ca2 DC |
905 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) |
906 | di_flags2 |= XFS_DIFLAG2_DAX; | |
907 | ||
365ca83d | 908 | ip->i_d.di_flags |= di_flags; |
58f88ca2 | 909 | ip->i_d.di_flags2 |= di_flags2; |
1da177e4 LT |
910 | } |
911 | /* FALLTHROUGH */ | |
912 | case S_IFLNK: | |
913 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
914 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
915 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
916 | ip->i_df.if_u1.if_extents = NULL; | |
917 | break; | |
918 | default: | |
919 | ASSERT(0); | |
920 | } | |
921 | /* | |
922 | * Attribute fork settings for new inode. | |
923 | */ | |
924 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
925 | ip->i_d.di_anextents = 0; | |
926 | ||
927 | /* | |
928 | * Log the new values stuffed into the inode. | |
929 | */ | |
ddc3415a | 930 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
931 | xfs_trans_log_inode(tp, ip, flags); |
932 | ||
58c90473 | 933 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 934 | xfs_setup_inode(ip); |
1da177e4 LT |
935 | |
936 | *ipp = ip; | |
937 | return 0; | |
938 | } | |
939 | ||
e546cb79 DC |
940 | /* |
941 | * Allocates a new inode from disk and return a pointer to the | |
942 | * incore copy. This routine will internally commit the current | |
943 | * transaction and allocate a new one if the Space Manager needed | |
944 | * to do an allocation to replenish the inode free-list. | |
945 | * | |
946 | * This routine is designed to be called from xfs_create and | |
947 | * xfs_create_dir. | |
948 | * | |
949 | */ | |
950 | int | |
951 | xfs_dir_ialloc( | |
952 | xfs_trans_t **tpp, /* input: current transaction; | |
953 | output: may be a new transaction. */ | |
954 | xfs_inode_t *dp, /* directory within whose allocate | |
955 | the inode. */ | |
956 | umode_t mode, | |
957 | xfs_nlink_t nlink, | |
958 | xfs_dev_t rdev, | |
959 | prid_t prid, /* project id */ | |
960 | int okalloc, /* ok to allocate new space */ | |
961 | xfs_inode_t **ipp, /* pointer to inode; it will be | |
962 | locked. */ | |
963 | int *committed) | |
964 | ||
965 | { | |
966 | xfs_trans_t *tp; | |
e546cb79 DC |
967 | xfs_inode_t *ip; |
968 | xfs_buf_t *ialloc_context = NULL; | |
969 | int code; | |
e546cb79 DC |
970 | void *dqinfo; |
971 | uint tflags; | |
972 | ||
973 | tp = *tpp; | |
974 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
975 | ||
976 | /* | |
977 | * xfs_ialloc will return a pointer to an incore inode if | |
978 | * the Space Manager has an available inode on the free | |
979 | * list. Otherwise, it will do an allocation and replenish | |
980 | * the freelist. Since we can only do one allocation per | |
981 | * transaction without deadlocks, we will need to commit the | |
982 | * current transaction and start a new one. We will then | |
983 | * need to call xfs_ialloc again to get the inode. | |
984 | * | |
985 | * If xfs_ialloc did an allocation to replenish the freelist, | |
986 | * it returns the bp containing the head of the freelist as | |
987 | * ialloc_context. We will hold a lock on it across the | |
988 | * transaction commit so that no other process can steal | |
989 | * the inode(s) that we've just allocated. | |
990 | */ | |
991 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, | |
992 | &ialloc_context, &ip); | |
993 | ||
994 | /* | |
995 | * Return an error if we were unable to allocate a new inode. | |
996 | * This should only happen if we run out of space on disk or | |
997 | * encounter a disk error. | |
998 | */ | |
999 | if (code) { | |
1000 | *ipp = NULL; | |
1001 | return code; | |
1002 | } | |
1003 | if (!ialloc_context && !ip) { | |
1004 | *ipp = NULL; | |
2451337d | 1005 | return -ENOSPC; |
e546cb79 DC |
1006 | } |
1007 | ||
1008 | /* | |
1009 | * If the AGI buffer is non-NULL, then we were unable to get an | |
1010 | * inode in one operation. We need to commit the current | |
1011 | * transaction and call xfs_ialloc() again. It is guaranteed | |
1012 | * to succeed the second time. | |
1013 | */ | |
1014 | if (ialloc_context) { | |
1015 | /* | |
1016 | * Normally, xfs_trans_commit releases all the locks. | |
1017 | * We call bhold to hang on to the ialloc_context across | |
1018 | * the commit. Holding this buffer prevents any other | |
1019 | * processes from doing any allocations in this | |
1020 | * allocation group. | |
1021 | */ | |
1022 | xfs_trans_bhold(tp, ialloc_context); | |
e546cb79 DC |
1023 | |
1024 | /* | |
1025 | * We want the quota changes to be associated with the next | |
1026 | * transaction, NOT this one. So, detach the dqinfo from this | |
1027 | * and attach it to the next transaction. | |
1028 | */ | |
1029 | dqinfo = NULL; | |
1030 | tflags = 0; | |
1031 | if (tp->t_dqinfo) { | |
1032 | dqinfo = (void *)tp->t_dqinfo; | |
1033 | tp->t_dqinfo = NULL; | |
1034 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; | |
1035 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); | |
1036 | } | |
1037 | ||
2e6db6c4 CH |
1038 | code = xfs_trans_roll(&tp, 0); |
1039 | if (committed != NULL) | |
e546cb79 | 1040 | *committed = 1; |
3d3c8b52 | 1041 | |
e546cb79 DC |
1042 | /* |
1043 | * Re-attach the quota info that we detached from prev trx. | |
1044 | */ | |
1045 | if (dqinfo) { | |
1046 | tp->t_dqinfo = dqinfo; | |
1047 | tp->t_flags |= tflags; | |
1048 | } | |
1049 | ||
1050 | if (code) { | |
1051 | xfs_buf_relse(ialloc_context); | |
2e6db6c4 | 1052 | *tpp = tp; |
e546cb79 DC |
1053 | *ipp = NULL; |
1054 | return code; | |
1055 | } | |
1056 | xfs_trans_bjoin(tp, ialloc_context); | |
1057 | ||
1058 | /* | |
1059 | * Call ialloc again. Since we've locked out all | |
1060 | * other allocations in this allocation group, | |
1061 | * this call should always succeed. | |
1062 | */ | |
1063 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, | |
1064 | okalloc, &ialloc_context, &ip); | |
1065 | ||
1066 | /* | |
1067 | * If we get an error at this point, return to the caller | |
1068 | * so that the current transaction can be aborted. | |
1069 | */ | |
1070 | if (code) { | |
1071 | *tpp = tp; | |
1072 | *ipp = NULL; | |
1073 | return code; | |
1074 | } | |
1075 | ASSERT(!ialloc_context && ip); | |
1076 | ||
1077 | } else { | |
1078 | if (committed != NULL) | |
1079 | *committed = 0; | |
1080 | } | |
1081 | ||
1082 | *ipp = ip; | |
1083 | *tpp = tp; | |
1084 | ||
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | /* | |
54d7b5c1 DC |
1089 | * Decrement the link count on an inode & log the change. If this causes the |
1090 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
1091 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 DC |
1092 | */ |
1093 | int /* error */ | |
1094 | xfs_droplink( | |
1095 | xfs_trans_t *tp, | |
1096 | xfs_inode_t *ip) | |
1097 | { | |
e546cb79 DC |
1098 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
1099 | ||
e546cb79 DC |
1100 | drop_nlink(VFS_I(ip)); |
1101 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1102 | ||
54d7b5c1 DC |
1103 | if (VFS_I(ip)->i_nlink) |
1104 | return 0; | |
1105 | ||
1106 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
1107 | } |
1108 | ||
e546cb79 DC |
1109 | /* |
1110 | * Increment the link count on an inode & log the change. | |
1111 | */ | |
1112 | int | |
1113 | xfs_bumplink( | |
1114 | xfs_trans_t *tp, | |
1115 | xfs_inode_t *ip) | |
1116 | { | |
1117 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1118 | ||
263997a6 | 1119 | ASSERT(ip->i_d.di_version > 1); |
e546cb79 | 1120 | inc_nlink(VFS_I(ip)); |
e546cb79 DC |
1121 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1122 | return 0; | |
1123 | } | |
1124 | ||
c24b5dfa DC |
1125 | int |
1126 | xfs_create( | |
1127 | xfs_inode_t *dp, | |
1128 | struct xfs_name *name, | |
1129 | umode_t mode, | |
1130 | xfs_dev_t rdev, | |
1131 | xfs_inode_t **ipp) | |
1132 | { | |
1133 | int is_dir = S_ISDIR(mode); | |
1134 | struct xfs_mount *mp = dp->i_mount; | |
1135 | struct xfs_inode *ip = NULL; | |
1136 | struct xfs_trans *tp = NULL; | |
1137 | int error; | |
1138 | xfs_bmap_free_t free_list; | |
1139 | xfs_fsblock_t first_block; | |
1140 | bool unlock_dp_on_error = false; | |
c24b5dfa DC |
1141 | prid_t prid; |
1142 | struct xfs_dquot *udqp = NULL; | |
1143 | struct xfs_dquot *gdqp = NULL; | |
1144 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 1145 | struct xfs_trans_res *tres; |
c24b5dfa | 1146 | uint resblks; |
c24b5dfa DC |
1147 | |
1148 | trace_xfs_create(dp, name); | |
1149 | ||
1150 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1151 | return -EIO; |
c24b5dfa | 1152 | |
163467d3 | 1153 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
1154 | |
1155 | /* | |
1156 | * Make sure that we have allocated dquot(s) on disk. | |
1157 | */ | |
7aab1b28 DE |
1158 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1159 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1160 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1161 | &udqp, &gdqp, &pdqp); | |
1162 | if (error) | |
1163 | return error; | |
1164 | ||
1165 | if (is_dir) { | |
1166 | rdev = 0; | |
1167 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
062647a8 | 1168 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
1169 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); |
1170 | } else { | |
1171 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 1172 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
1173 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); |
1174 | } | |
1175 | ||
c24b5dfa DC |
1176 | /* |
1177 | * Initially assume that the file does not exist and | |
1178 | * reserve the resources for that case. If that is not | |
1179 | * the case we'll drop the one we have and get a more | |
1180 | * appropriate transaction later. | |
1181 | */ | |
062647a8 | 1182 | error = xfs_trans_reserve(tp, tres, resblks, 0); |
2451337d | 1183 | if (error == -ENOSPC) { |
c24b5dfa DC |
1184 | /* flush outstanding delalloc blocks and retry */ |
1185 | xfs_flush_inodes(mp); | |
062647a8 | 1186 | error = xfs_trans_reserve(tp, tres, resblks, 0); |
c24b5dfa | 1187 | } |
2451337d | 1188 | if (error == -ENOSPC) { |
c24b5dfa DC |
1189 | /* No space at all so try a "no-allocation" reservation */ |
1190 | resblks = 0; | |
062647a8 | 1191 | error = xfs_trans_reserve(tp, tres, 0, 0); |
c24b5dfa | 1192 | } |
4906e215 | 1193 | if (error) |
c24b5dfa | 1194 | goto out_trans_cancel; |
4906e215 | 1195 | |
c24b5dfa | 1196 | |
dbad7c99 DC |
1197 | xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
1198 | XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT); | |
c24b5dfa DC |
1199 | unlock_dp_on_error = true; |
1200 | ||
1201 | xfs_bmap_init(&free_list, &first_block); | |
1202 | ||
1203 | /* | |
1204 | * Reserve disk quota and the inode. | |
1205 | */ | |
1206 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1207 | pdqp, resblks, 1, 0); | |
1208 | if (error) | |
1209 | goto out_trans_cancel; | |
1210 | ||
94f3cad5 ES |
1211 | if (!resblks) { |
1212 | error = xfs_dir_canenter(tp, dp, name); | |
1213 | if (error) | |
1214 | goto out_trans_cancel; | |
1215 | } | |
c24b5dfa DC |
1216 | |
1217 | /* | |
1218 | * A newly created regular or special file just has one directory | |
1219 | * entry pointing to them, but a directory also the "." entry | |
1220 | * pointing to itself. | |
1221 | */ | |
1222 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, | |
f6106efa | 1223 | prid, resblks > 0, &ip, NULL); |
d6077aa3 | 1224 | if (error) |
4906e215 | 1225 | goto out_trans_cancel; |
c24b5dfa DC |
1226 | |
1227 | /* | |
1228 | * Now we join the directory inode to the transaction. We do not do it | |
1229 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1230 | * (and release all the locks). An error from here on will result in | |
1231 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1232 | * error path. | |
1233 | */ | |
dbad7c99 | 1234 | xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
c24b5dfa DC |
1235 | unlock_dp_on_error = false; |
1236 | ||
1237 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
1238 | &first_block, &free_list, resblks ? | |
1239 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1240 | if (error) { | |
2451337d | 1241 | ASSERT(error != -ENOSPC); |
4906e215 | 1242 | goto out_trans_cancel; |
c24b5dfa DC |
1243 | } |
1244 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1245 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1246 | ||
1247 | if (is_dir) { | |
1248 | error = xfs_dir_init(tp, ip, dp); | |
1249 | if (error) | |
1250 | goto out_bmap_cancel; | |
1251 | ||
1252 | error = xfs_bumplink(tp, dp); | |
1253 | if (error) | |
1254 | goto out_bmap_cancel; | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * If this is a synchronous mount, make sure that the | |
1259 | * create transaction goes to disk before returning to | |
1260 | * the user. | |
1261 | */ | |
1262 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1263 | xfs_trans_set_sync(tp); | |
1264 | ||
1265 | /* | |
1266 | * Attach the dquot(s) to the inodes and modify them incore. | |
1267 | * These ids of the inode couldn't have changed since the new | |
1268 | * inode has been locked ever since it was created. | |
1269 | */ | |
1270 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1271 | ||
f6106efa | 1272 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
c24b5dfa DC |
1273 | if (error) |
1274 | goto out_bmap_cancel; | |
1275 | ||
70393313 | 1276 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1277 | if (error) |
1278 | goto out_release_inode; | |
1279 | ||
1280 | xfs_qm_dqrele(udqp); | |
1281 | xfs_qm_dqrele(gdqp); | |
1282 | xfs_qm_dqrele(pdqp); | |
1283 | ||
1284 | *ipp = ip; | |
1285 | return 0; | |
1286 | ||
1287 | out_bmap_cancel: | |
1288 | xfs_bmap_cancel(&free_list); | |
c24b5dfa | 1289 | out_trans_cancel: |
4906e215 | 1290 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1291 | out_release_inode: |
1292 | /* | |
58c90473 DC |
1293 | * Wait until after the current transaction is aborted to finish the |
1294 | * setup of the inode and release the inode. This prevents recursive | |
1295 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1296 | */ |
58c90473 DC |
1297 | if (ip) { |
1298 | xfs_finish_inode_setup(ip); | |
c24b5dfa | 1299 | IRELE(ip); |
58c90473 | 1300 | } |
c24b5dfa DC |
1301 | |
1302 | xfs_qm_dqrele(udqp); | |
1303 | xfs_qm_dqrele(gdqp); | |
1304 | xfs_qm_dqrele(pdqp); | |
1305 | ||
1306 | if (unlock_dp_on_error) | |
dbad7c99 | 1307 | xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
c24b5dfa DC |
1308 | return error; |
1309 | } | |
1310 | ||
99b6436b ZYW |
1311 | int |
1312 | xfs_create_tmpfile( | |
1313 | struct xfs_inode *dp, | |
1314 | struct dentry *dentry, | |
330033d6 BF |
1315 | umode_t mode, |
1316 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1317 | { |
1318 | struct xfs_mount *mp = dp->i_mount; | |
1319 | struct xfs_inode *ip = NULL; | |
1320 | struct xfs_trans *tp = NULL; | |
1321 | int error; | |
99b6436b ZYW |
1322 | prid_t prid; |
1323 | struct xfs_dquot *udqp = NULL; | |
1324 | struct xfs_dquot *gdqp = NULL; | |
1325 | struct xfs_dquot *pdqp = NULL; | |
1326 | struct xfs_trans_res *tres; | |
1327 | uint resblks; | |
1328 | ||
1329 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1330 | return -EIO; |
99b6436b ZYW |
1331 | |
1332 | prid = xfs_get_initial_prid(dp); | |
1333 | ||
1334 | /* | |
1335 | * Make sure that we have allocated dquot(s) on disk. | |
1336 | */ | |
1337 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), | |
1338 | xfs_kgid_to_gid(current_fsgid()), prid, | |
1339 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
1340 | &udqp, &gdqp, &pdqp); | |
1341 | if (error) | |
1342 | return error; | |
1343 | ||
1344 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
1345 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE); | |
1346 | ||
1347 | tres = &M_RES(mp)->tr_create_tmpfile; | |
1348 | error = xfs_trans_reserve(tp, tres, resblks, 0); | |
2451337d | 1349 | if (error == -ENOSPC) { |
99b6436b ZYW |
1350 | /* No space at all so try a "no-allocation" reservation */ |
1351 | resblks = 0; | |
1352 | error = xfs_trans_reserve(tp, tres, 0, 0); | |
1353 | } | |
4906e215 | 1354 | if (error) |
99b6436b | 1355 | goto out_trans_cancel; |
99b6436b ZYW |
1356 | |
1357 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1358 | pdqp, resblks, 1, 0); | |
1359 | if (error) | |
1360 | goto out_trans_cancel; | |
1361 | ||
1362 | error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, | |
1363 | prid, resblks > 0, &ip, NULL); | |
d6077aa3 | 1364 | if (error) |
4906e215 | 1365 | goto out_trans_cancel; |
99b6436b ZYW |
1366 | |
1367 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1368 | xfs_trans_set_sync(tp); | |
1369 | ||
1370 | /* | |
1371 | * Attach the dquot(s) to the inodes and modify them incore. | |
1372 | * These ids of the inode couldn't have changed since the new | |
1373 | * inode has been locked ever since it was created. | |
1374 | */ | |
1375 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1376 | ||
99b6436b ZYW |
1377 | error = xfs_iunlink(tp, ip); |
1378 | if (error) | |
4906e215 | 1379 | goto out_trans_cancel; |
99b6436b | 1380 | |
70393313 | 1381 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1382 | if (error) |
1383 | goto out_release_inode; | |
1384 | ||
1385 | xfs_qm_dqrele(udqp); | |
1386 | xfs_qm_dqrele(gdqp); | |
1387 | xfs_qm_dqrele(pdqp); | |
1388 | ||
330033d6 | 1389 | *ipp = ip; |
99b6436b ZYW |
1390 | return 0; |
1391 | ||
99b6436b | 1392 | out_trans_cancel: |
4906e215 | 1393 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1394 | out_release_inode: |
1395 | /* | |
58c90473 DC |
1396 | * Wait until after the current transaction is aborted to finish the |
1397 | * setup of the inode and release the inode. This prevents recursive | |
1398 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1399 | */ |
58c90473 DC |
1400 | if (ip) { |
1401 | xfs_finish_inode_setup(ip); | |
99b6436b | 1402 | IRELE(ip); |
58c90473 | 1403 | } |
99b6436b ZYW |
1404 | |
1405 | xfs_qm_dqrele(udqp); | |
1406 | xfs_qm_dqrele(gdqp); | |
1407 | xfs_qm_dqrele(pdqp); | |
1408 | ||
1409 | return error; | |
1410 | } | |
1411 | ||
c24b5dfa DC |
1412 | int |
1413 | xfs_link( | |
1414 | xfs_inode_t *tdp, | |
1415 | xfs_inode_t *sip, | |
1416 | struct xfs_name *target_name) | |
1417 | { | |
1418 | xfs_mount_t *mp = tdp->i_mount; | |
1419 | xfs_trans_t *tp; | |
1420 | int error; | |
1421 | xfs_bmap_free_t free_list; | |
1422 | xfs_fsblock_t first_block; | |
c24b5dfa DC |
1423 | int resblks; |
1424 | ||
1425 | trace_xfs_link(tdp, target_name); | |
1426 | ||
1427 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); | |
1428 | ||
1429 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1430 | return -EIO; |
c24b5dfa DC |
1431 | |
1432 | error = xfs_qm_dqattach(sip, 0); | |
1433 | if (error) | |
1434 | goto std_return; | |
1435 | ||
1436 | error = xfs_qm_dqattach(tdp, 0); | |
1437 | if (error) | |
1438 | goto std_return; | |
1439 | ||
1440 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
c24b5dfa | 1441 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
3d3c8b52 | 1442 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0); |
2451337d | 1443 | if (error == -ENOSPC) { |
c24b5dfa | 1444 | resblks = 0; |
3d3c8b52 | 1445 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0); |
c24b5dfa | 1446 | } |
4906e215 | 1447 | if (error) |
c24b5dfa | 1448 | goto error_return; |
c24b5dfa | 1449 | |
dbad7c99 | 1450 | xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); |
c24b5dfa DC |
1451 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1452 | ||
1453 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
dbad7c99 | 1454 | xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
c24b5dfa DC |
1455 | |
1456 | /* | |
1457 | * If we are using project inheritance, we only allow hard link | |
1458 | * creation in our tree when the project IDs are the same; else | |
1459 | * the tree quota mechanism could be circumvented. | |
1460 | */ | |
1461 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1462 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
2451337d | 1463 | error = -EXDEV; |
c24b5dfa DC |
1464 | goto error_return; |
1465 | } | |
1466 | ||
94f3cad5 ES |
1467 | if (!resblks) { |
1468 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1469 | if (error) | |
1470 | goto error_return; | |
1471 | } | |
c24b5dfa DC |
1472 | |
1473 | xfs_bmap_init(&free_list, &first_block); | |
1474 | ||
54d7b5c1 DC |
1475 | /* |
1476 | * Handle initial link state of O_TMPFILE inode | |
1477 | */ | |
1478 | if (VFS_I(sip)->i_nlink == 0) { | |
ab297431 ZYW |
1479 | error = xfs_iunlink_remove(tp, sip); |
1480 | if (error) | |
4906e215 | 1481 | goto error_return; |
ab297431 ZYW |
1482 | } |
1483 | ||
c24b5dfa DC |
1484 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
1485 | &first_block, &free_list, resblks); | |
1486 | if (error) | |
4906e215 | 1487 | goto error_return; |
c24b5dfa DC |
1488 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1489 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1490 | ||
1491 | error = xfs_bumplink(tp, sip); | |
1492 | if (error) | |
4906e215 | 1493 | goto error_return; |
c24b5dfa DC |
1494 | |
1495 | /* | |
1496 | * If this is a synchronous mount, make sure that the | |
1497 | * link transaction goes to disk before returning to | |
1498 | * the user. | |
1499 | */ | |
f6106efa | 1500 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
c24b5dfa | 1501 | xfs_trans_set_sync(tp); |
c24b5dfa | 1502 | |
f6106efa | 1503 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
c24b5dfa DC |
1504 | if (error) { |
1505 | xfs_bmap_cancel(&free_list); | |
4906e215 | 1506 | goto error_return; |
c24b5dfa DC |
1507 | } |
1508 | ||
70393313 | 1509 | return xfs_trans_commit(tp); |
c24b5dfa | 1510 | |
c24b5dfa | 1511 | error_return: |
4906e215 | 1512 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1513 | std_return: |
1514 | return error; | |
1515 | } | |
1516 | ||
1da177e4 | 1517 | /* |
8f04c47a CH |
1518 | * Free up the underlying blocks past new_size. The new size must be smaller |
1519 | * than the current size. This routine can be used both for the attribute and | |
1520 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1521 | * |
f6485057 DC |
1522 | * The transaction passed to this routine must have made a permanent log |
1523 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1524 | * given transaction and start new ones, so make sure everything involved in | |
1525 | * the transaction is tidy before calling here. Some transaction will be | |
1526 | * returned to the caller to be committed. The incoming transaction must | |
1527 | * already include the inode, and both inode locks must be held exclusively. | |
1528 | * The inode must also be "held" within the transaction. On return the inode | |
1529 | * will be "held" within the returned transaction. This routine does NOT | |
1530 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1531 | * |
f6485057 DC |
1532 | * If we get an error, we must return with the inode locked and linked into the |
1533 | * current transaction. This keeps things simple for the higher level code, | |
1534 | * because it always knows that the inode is locked and held in the transaction | |
1535 | * that returns to it whether errors occur or not. We don't mark the inode | |
1536 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1537 | */ |
1538 | int | |
8f04c47a CH |
1539 | xfs_itruncate_extents( |
1540 | struct xfs_trans **tpp, | |
1541 | struct xfs_inode *ip, | |
1542 | int whichfork, | |
1543 | xfs_fsize_t new_size) | |
1da177e4 | 1544 | { |
8f04c47a CH |
1545 | struct xfs_mount *mp = ip->i_mount; |
1546 | struct xfs_trans *tp = *tpp; | |
8f04c47a CH |
1547 | xfs_bmap_free_t free_list; |
1548 | xfs_fsblock_t first_block; | |
1549 | xfs_fileoff_t first_unmap_block; | |
1550 | xfs_fileoff_t last_block; | |
1551 | xfs_filblks_t unmap_len; | |
8f04c47a CH |
1552 | int error = 0; |
1553 | int done = 0; | |
1da177e4 | 1554 | |
0b56185b CH |
1555 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1556 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1557 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1558 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1559 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1560 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1561 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1562 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1563 | |
673e8e59 CH |
1564 | trace_xfs_itruncate_extents_start(ip, new_size); |
1565 | ||
1da177e4 LT |
1566 | /* |
1567 | * Since it is possible for space to become allocated beyond | |
1568 | * the end of the file (in a crash where the space is allocated | |
1569 | * but the inode size is not yet updated), simply remove any | |
1570 | * blocks which show up between the new EOF and the maximum | |
1571 | * possible file size. If the first block to be removed is | |
1572 | * beyond the maximum file size (ie it is the same as last_block), | |
1573 | * then there is nothing to do. | |
1574 | */ | |
8f04c47a | 1575 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1576 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1577 | if (first_unmap_block == last_block) |
1578 | return 0; | |
1579 | ||
1580 | ASSERT(first_unmap_block < last_block); | |
1581 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1582 | while (!done) { |
9d87c319 | 1583 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 1584 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1585 | first_unmap_block, unmap_len, |
8f04c47a | 1586 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1587 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 1588 | &first_block, &free_list, |
b4e9181e | 1589 | &done); |
8f04c47a CH |
1590 | if (error) |
1591 | goto out_bmap_cancel; | |
1da177e4 LT |
1592 | |
1593 | /* | |
1594 | * Duplicate the transaction that has the permanent | |
1595 | * reservation and commit the old transaction. | |
1596 | */ | |
f6106efa | 1597 | error = xfs_bmap_finish(&tp, &free_list, ip); |
8f04c47a CH |
1598 | if (error) |
1599 | goto out_bmap_cancel; | |
1da177e4 | 1600 | |
2e6db6c4 | 1601 | error = xfs_trans_roll(&tp, ip); |
f6485057 | 1602 | if (error) |
8f04c47a | 1603 | goto out; |
1da177e4 | 1604 | } |
8f04c47a | 1605 | |
673e8e59 CH |
1606 | /* |
1607 | * Always re-log the inode so that our permanent transaction can keep | |
1608 | * on rolling it forward in the log. | |
1609 | */ | |
1610 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1611 | ||
1612 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1613 | ||
8f04c47a CH |
1614 | out: |
1615 | *tpp = tp; | |
1616 | return error; | |
1617 | out_bmap_cancel: | |
1da177e4 | 1618 | /* |
8f04c47a CH |
1619 | * If the bunmapi call encounters an error, return to the caller where |
1620 | * the transaction can be properly aborted. We just need to make sure | |
1621 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1622 | */ |
8f04c47a CH |
1623 | xfs_bmap_cancel(&free_list); |
1624 | goto out; | |
1625 | } | |
1626 | ||
c24b5dfa DC |
1627 | int |
1628 | xfs_release( | |
1629 | xfs_inode_t *ip) | |
1630 | { | |
1631 | xfs_mount_t *mp = ip->i_mount; | |
1632 | int error; | |
1633 | ||
1634 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) | |
1635 | return 0; | |
1636 | ||
1637 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1638 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1639 | return 0; | |
1640 | ||
1641 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1642 | int truncated; | |
1643 | ||
c24b5dfa DC |
1644 | /* |
1645 | * If we previously truncated this file and removed old data | |
1646 | * in the process, we want to initiate "early" writeout on | |
1647 | * the last close. This is an attempt to combat the notorious | |
1648 | * NULL files problem which is particularly noticeable from a | |
1649 | * truncate down, buffered (re-)write (delalloc), followed by | |
1650 | * a crash. What we are effectively doing here is | |
1651 | * significantly reducing the time window where we'd otherwise | |
1652 | * be exposed to that problem. | |
1653 | */ | |
1654 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1655 | if (truncated) { | |
1656 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1657 | if (ip->i_delayed_blks > 0) { |
2451337d | 1658 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1659 | if (error) |
1660 | return error; | |
1661 | } | |
1662 | } | |
1663 | } | |
1664 | ||
54d7b5c1 | 1665 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1666 | return 0; |
1667 | ||
1668 | if (xfs_can_free_eofblocks(ip, false)) { | |
1669 | ||
1670 | /* | |
1671 | * If we can't get the iolock just skip truncating the blocks | |
1672 | * past EOF because we could deadlock with the mmap_sem | |
1673 | * otherwise. We'll get another chance to drop them once the | |
1674 | * last reference to the inode is dropped, so we'll never leak | |
1675 | * blocks permanently. | |
1676 | * | |
1677 | * Further, check if the inode is being opened, written and | |
1678 | * closed frequently and we have delayed allocation blocks | |
1679 | * outstanding (e.g. streaming writes from the NFS server), | |
1680 | * truncating the blocks past EOF will cause fragmentation to | |
1681 | * occur. | |
1682 | * | |
1683 | * In this case don't do the truncation, either, but we have to | |
1684 | * be careful how we detect this case. Blocks beyond EOF show | |
1685 | * up as i_delayed_blks even when the inode is clean, so we | |
1686 | * need to truncate them away first before checking for a dirty | |
1687 | * release. Hence on the first dirty close we will still remove | |
1688 | * the speculative allocation, but after that we will leave it | |
1689 | * in place. | |
1690 | */ | |
1691 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1692 | return 0; | |
1693 | ||
1694 | error = xfs_free_eofblocks(mp, ip, true); | |
2451337d | 1695 | if (error && error != -EAGAIN) |
c24b5dfa DC |
1696 | return error; |
1697 | ||
1698 | /* delalloc blocks after truncation means it really is dirty */ | |
1699 | if (ip->i_delayed_blks) | |
1700 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1701 | } | |
1702 | return 0; | |
1703 | } | |
1704 | ||
f7be2d7f BF |
1705 | /* |
1706 | * xfs_inactive_truncate | |
1707 | * | |
1708 | * Called to perform a truncate when an inode becomes unlinked. | |
1709 | */ | |
1710 | STATIC int | |
1711 | xfs_inactive_truncate( | |
1712 | struct xfs_inode *ip) | |
1713 | { | |
1714 | struct xfs_mount *mp = ip->i_mount; | |
1715 | struct xfs_trans *tp; | |
1716 | int error; | |
1717 | ||
1718 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1719 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); | |
1720 | if (error) { | |
1721 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
4906e215 | 1722 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1723 | return error; |
1724 | } | |
1725 | ||
1726 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1727 | xfs_trans_ijoin(tp, ip, 0); | |
1728 | ||
1729 | /* | |
1730 | * Log the inode size first to prevent stale data exposure in the event | |
1731 | * of a system crash before the truncate completes. See the related | |
1732 | * comment in xfs_setattr_size() for details. | |
1733 | */ | |
1734 | ip->i_d.di_size = 0; | |
1735 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1736 | ||
1737 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1738 | if (error) | |
1739 | goto error_trans_cancel; | |
1740 | ||
1741 | ASSERT(ip->i_d.di_nextents == 0); | |
1742 | ||
70393313 | 1743 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1744 | if (error) |
1745 | goto error_unlock; | |
1746 | ||
1747 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1748 | return 0; | |
1749 | ||
1750 | error_trans_cancel: | |
4906e215 | 1751 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1752 | error_unlock: |
1753 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1754 | return error; | |
1755 | } | |
1756 | ||
88877d2b BF |
1757 | /* |
1758 | * xfs_inactive_ifree() | |
1759 | * | |
1760 | * Perform the inode free when an inode is unlinked. | |
1761 | */ | |
1762 | STATIC int | |
1763 | xfs_inactive_ifree( | |
1764 | struct xfs_inode *ip) | |
1765 | { | |
1766 | xfs_bmap_free_t free_list; | |
1767 | xfs_fsblock_t first_block; | |
88877d2b BF |
1768 | struct xfs_mount *mp = ip->i_mount; |
1769 | struct xfs_trans *tp; | |
1770 | int error; | |
1771 | ||
1772 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
9d43b180 BF |
1773 | |
1774 | /* | |
1775 | * The ifree transaction might need to allocate blocks for record | |
1776 | * insertion to the finobt. We don't want to fail here at ENOSPC, so | |
1777 | * allow ifree to dip into the reserved block pool if necessary. | |
1778 | * | |
1779 | * Freeing large sets of inodes generally means freeing inode chunks, | |
1780 | * directory and file data blocks, so this should be relatively safe. | |
1781 | * Only under severe circumstances should it be possible to free enough | |
1782 | * inodes to exhaust the reserve block pool via finobt expansion while | |
1783 | * at the same time not creating free space in the filesystem. | |
1784 | * | |
1785 | * Send a warning if the reservation does happen to fail, as the inode | |
1786 | * now remains allocated and sits on the unlinked list until the fs is | |
1787 | * repaired. | |
1788 | */ | |
1789 | tp->t_flags |= XFS_TRANS_RESERVE; | |
1790 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, | |
1791 | XFS_IFREE_SPACE_RES(mp), 0); | |
88877d2b | 1792 | if (error) { |
2451337d | 1793 | if (error == -ENOSPC) { |
9d43b180 BF |
1794 | xfs_warn_ratelimited(mp, |
1795 | "Failed to remove inode(s) from unlinked list. " | |
1796 | "Please free space, unmount and run xfs_repair."); | |
1797 | } else { | |
1798 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1799 | } | |
4906e215 | 1800 | xfs_trans_cancel(tp); |
88877d2b BF |
1801 | return error; |
1802 | } | |
1803 | ||
1804 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1805 | xfs_trans_ijoin(tp, ip, 0); | |
1806 | ||
1807 | xfs_bmap_init(&free_list, &first_block); | |
1808 | error = xfs_ifree(tp, ip, &free_list); | |
1809 | if (error) { | |
1810 | /* | |
1811 | * If we fail to free the inode, shut down. The cancel | |
1812 | * might do that, we need to make sure. Otherwise the | |
1813 | * inode might be lost for a long time or forever. | |
1814 | */ | |
1815 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1816 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1817 | __func__, error); | |
1818 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1819 | } | |
4906e215 | 1820 | xfs_trans_cancel(tp); |
88877d2b BF |
1821 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1822 | return error; | |
1823 | } | |
1824 | ||
1825 | /* | |
1826 | * Credit the quota account(s). The inode is gone. | |
1827 | */ | |
1828 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1829 | ||
1830 | /* | |
d4a97a04 BF |
1831 | * Just ignore errors at this point. There is nothing we can do except |
1832 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1833 | */ |
f6106efa | 1834 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
d4a97a04 | 1835 | if (error) { |
88877d2b BF |
1836 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", |
1837 | __func__, error); | |
d4a97a04 BF |
1838 | xfs_bmap_cancel(&free_list); |
1839 | } | |
70393313 | 1840 | error = xfs_trans_commit(tp); |
88877d2b BF |
1841 | if (error) |
1842 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1843 | __func__, error); | |
1844 | ||
1845 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1846 | return 0; | |
1847 | } | |
1848 | ||
c24b5dfa DC |
1849 | /* |
1850 | * xfs_inactive | |
1851 | * | |
1852 | * This is called when the vnode reference count for the vnode | |
1853 | * goes to zero. If the file has been unlinked, then it must | |
1854 | * now be truncated. Also, we clear all of the read-ahead state | |
1855 | * kept for the inode here since the file is now closed. | |
1856 | */ | |
74564fb4 | 1857 | void |
c24b5dfa DC |
1858 | xfs_inactive( |
1859 | xfs_inode_t *ip) | |
1860 | { | |
3d3c8b52 | 1861 | struct xfs_mount *mp; |
3d3c8b52 JL |
1862 | int error; |
1863 | int truncate = 0; | |
c24b5dfa DC |
1864 | |
1865 | /* | |
1866 | * If the inode is already free, then there can be nothing | |
1867 | * to clean up here. | |
1868 | */ | |
d948709b | 1869 | if (ip->i_d.di_mode == 0) { |
c24b5dfa DC |
1870 | ASSERT(ip->i_df.if_real_bytes == 0); |
1871 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1872 | return; |
c24b5dfa DC |
1873 | } |
1874 | ||
1875 | mp = ip->i_mount; | |
1876 | ||
c24b5dfa DC |
1877 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1878 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1879 | return; |
c24b5dfa | 1880 | |
54d7b5c1 | 1881 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1882 | /* |
1883 | * force is true because we are evicting an inode from the | |
1884 | * cache. Post-eof blocks must be freed, lest we end up with | |
1885 | * broken free space accounting. | |
1886 | */ | |
74564fb4 BF |
1887 | if (xfs_can_free_eofblocks(ip, true)) |
1888 | xfs_free_eofblocks(mp, ip, false); | |
1889 | ||
1890 | return; | |
c24b5dfa DC |
1891 | } |
1892 | ||
1893 | if (S_ISREG(ip->i_d.di_mode) && | |
1894 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || | |
1895 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1896 | truncate = 1; | |
1897 | ||
1898 | error = xfs_qm_dqattach(ip, 0); | |
1899 | if (error) | |
74564fb4 | 1900 | return; |
c24b5dfa | 1901 | |
f7be2d7f | 1902 | if (S_ISLNK(ip->i_d.di_mode)) |
36b21dde | 1903 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1904 | else if (truncate) |
1905 | error = xfs_inactive_truncate(ip); | |
1906 | if (error) | |
74564fb4 | 1907 | return; |
c24b5dfa DC |
1908 | |
1909 | /* | |
1910 | * If there are attributes associated with the file then blow them away | |
1911 | * now. The code calls a routine that recursively deconstructs the | |
6dfe5a04 | 1912 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1913 | */ |
6dfe5a04 | 1914 | if (XFS_IFORK_Q(ip)) { |
c24b5dfa DC |
1915 | error = xfs_attr_inactive(ip); |
1916 | if (error) | |
74564fb4 | 1917 | return; |
c24b5dfa DC |
1918 | } |
1919 | ||
6dfe5a04 | 1920 | ASSERT(!ip->i_afp); |
c24b5dfa | 1921 | ASSERT(ip->i_d.di_anextents == 0); |
6dfe5a04 | 1922 | ASSERT(ip->i_d.di_forkoff == 0); |
c24b5dfa DC |
1923 | |
1924 | /* | |
1925 | * Free the inode. | |
1926 | */ | |
88877d2b BF |
1927 | error = xfs_inactive_ifree(ip); |
1928 | if (error) | |
74564fb4 | 1929 | return; |
c24b5dfa DC |
1930 | |
1931 | /* | |
1932 | * Release the dquots held by inode, if any. | |
1933 | */ | |
1934 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1935 | } |
1936 | ||
1da177e4 | 1937 | /* |
54d7b5c1 DC |
1938 | * This is called when the inode's link count goes to 0 or we are creating a |
1939 | * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be | |
1940 | * set to true as the link count is dropped to zero by the VFS after we've | |
1941 | * created the file successfully, so we have to add it to the unlinked list | |
1942 | * while the link count is non-zero. | |
1943 | * | |
1944 | * We place the on-disk inode on a list in the AGI. It will be pulled from this | |
1945 | * list when the inode is freed. | |
1da177e4 | 1946 | */ |
54d7b5c1 | 1947 | STATIC int |
1da177e4 | 1948 | xfs_iunlink( |
54d7b5c1 DC |
1949 | struct xfs_trans *tp, |
1950 | struct xfs_inode *ip) | |
1da177e4 | 1951 | { |
54d7b5c1 | 1952 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1953 | xfs_agi_t *agi; |
1954 | xfs_dinode_t *dip; | |
1955 | xfs_buf_t *agibp; | |
1956 | xfs_buf_t *ibp; | |
1da177e4 LT |
1957 | xfs_agino_t agino; |
1958 | short bucket_index; | |
1959 | int offset; | |
1960 | int error; | |
1da177e4 | 1961 | |
1da177e4 | 1962 | ASSERT(ip->i_d.di_mode != 0); |
1da177e4 | 1963 | |
1da177e4 LT |
1964 | /* |
1965 | * Get the agi buffer first. It ensures lock ordering | |
1966 | * on the list. | |
1967 | */ | |
5e1be0fb | 1968 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1969 | if (error) |
1da177e4 | 1970 | return error; |
1da177e4 | 1971 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1972 | |
1da177e4 LT |
1973 | /* |
1974 | * Get the index into the agi hash table for the | |
1975 | * list this inode will go on. | |
1976 | */ | |
1977 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1978 | ASSERT(agino != 0); | |
1979 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1980 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1981 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1982 | |
69ef921b | 1983 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1984 | /* |
1985 | * There is already another inode in the bucket we need | |
1986 | * to add ourselves to. Add us at the front of the list. | |
1987 | * Here we put the head pointer into our next pointer, | |
1988 | * and then we fall through to point the head at us. | |
1989 | */ | |
475ee413 CH |
1990 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1991 | 0, 0); | |
c319b58b VA |
1992 | if (error) |
1993 | return error; | |
1994 | ||
69ef921b | 1995 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1996 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1997 | offset = ip->i_imap.im_boffset + |
1da177e4 | 1998 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1999 | |
2000 | /* need to recalc the inode CRC if appropriate */ | |
2001 | xfs_dinode_calc_crc(mp, dip); | |
2002 | ||
1da177e4 LT |
2003 | xfs_trans_inode_buf(tp, ibp); |
2004 | xfs_trans_log_buf(tp, ibp, offset, | |
2005 | (offset + sizeof(xfs_agino_t) - 1)); | |
2006 | xfs_inobp_check(mp, ibp); | |
2007 | } | |
2008 | ||
2009 | /* | |
2010 | * Point the bucket head pointer at the inode being inserted. | |
2011 | */ | |
2012 | ASSERT(agino != 0); | |
16259e7d | 2013 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
2014 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2015 | (sizeof(xfs_agino_t) * bucket_index); | |
f19b872b | 2016 | xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); |
1da177e4 LT |
2017 | xfs_trans_log_buf(tp, agibp, offset, |
2018 | (offset + sizeof(xfs_agino_t) - 1)); | |
2019 | return 0; | |
2020 | } | |
2021 | ||
2022 | /* | |
2023 | * Pull the on-disk inode from the AGI unlinked list. | |
2024 | */ | |
2025 | STATIC int | |
2026 | xfs_iunlink_remove( | |
2027 | xfs_trans_t *tp, | |
2028 | xfs_inode_t *ip) | |
2029 | { | |
2030 | xfs_ino_t next_ino; | |
2031 | xfs_mount_t *mp; | |
2032 | xfs_agi_t *agi; | |
2033 | xfs_dinode_t *dip; | |
2034 | xfs_buf_t *agibp; | |
2035 | xfs_buf_t *ibp; | |
2036 | xfs_agnumber_t agno; | |
1da177e4 LT |
2037 | xfs_agino_t agino; |
2038 | xfs_agino_t next_agino; | |
2039 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2040 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2041 | short bucket_index; |
6fdf8ccc | 2042 | int offset, last_offset = 0; |
1da177e4 | 2043 | int error; |
1da177e4 | 2044 | |
1da177e4 | 2045 | mp = tp->t_mountp; |
1da177e4 | 2046 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
2047 | |
2048 | /* | |
2049 | * Get the agi buffer first. It ensures lock ordering | |
2050 | * on the list. | |
2051 | */ | |
5e1be0fb CH |
2052 | error = xfs_read_agi(mp, tp, agno, &agibp); |
2053 | if (error) | |
1da177e4 | 2054 | return error; |
5e1be0fb | 2055 | |
1da177e4 | 2056 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2057 | |
1da177e4 LT |
2058 | /* |
2059 | * Get the index into the agi hash table for the | |
2060 | * list this inode will go on. | |
2061 | */ | |
2062 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2063 | ASSERT(agino != 0); | |
2064 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 2065 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
2066 | ASSERT(agi->agi_unlinked[bucket_index]); |
2067 | ||
16259e7d | 2068 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2069 | /* |
475ee413 CH |
2070 | * We're at the head of the list. Get the inode's on-disk |
2071 | * buffer to see if there is anyone after us on the list. | |
2072 | * Only modify our next pointer if it is not already NULLAGINO. | |
2073 | * This saves us the overhead of dealing with the buffer when | |
2074 | * there is no need to change it. | |
1da177e4 | 2075 | */ |
475ee413 CH |
2076 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2077 | 0, 0); | |
1da177e4 | 2078 | if (error) { |
475ee413 | 2079 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2080 | __func__, error); |
1da177e4 LT |
2081 | return error; |
2082 | } | |
347d1c01 | 2083 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2084 | ASSERT(next_agino != 0); |
2085 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2086 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2087 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2088 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2089 | |
2090 | /* need to recalc the inode CRC if appropriate */ | |
2091 | xfs_dinode_calc_crc(mp, dip); | |
2092 | ||
1da177e4 LT |
2093 | xfs_trans_inode_buf(tp, ibp); |
2094 | xfs_trans_log_buf(tp, ibp, offset, | |
2095 | (offset + sizeof(xfs_agino_t) - 1)); | |
2096 | xfs_inobp_check(mp, ibp); | |
2097 | } else { | |
2098 | xfs_trans_brelse(tp, ibp); | |
2099 | } | |
2100 | /* | |
2101 | * Point the bucket head pointer at the next inode. | |
2102 | */ | |
2103 | ASSERT(next_agino != 0); | |
2104 | ASSERT(next_agino != agino); | |
16259e7d | 2105 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2106 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2107 | (sizeof(xfs_agino_t) * bucket_index); | |
f19b872b | 2108 | xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); |
1da177e4 LT |
2109 | xfs_trans_log_buf(tp, agibp, offset, |
2110 | (offset + sizeof(xfs_agino_t) - 1)); | |
2111 | } else { | |
2112 | /* | |
2113 | * We need to search the list for the inode being freed. | |
2114 | */ | |
16259e7d | 2115 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2116 | last_ibp = NULL; |
2117 | while (next_agino != agino) { | |
129dbc9a CH |
2118 | struct xfs_imap imap; |
2119 | ||
2120 | if (last_ibp) | |
1da177e4 | 2121 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2122 | |
2123 | imap.im_blkno = 0; | |
1da177e4 | 2124 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2125 | |
2126 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2127 | if (error) { | |
2128 | xfs_warn(mp, | |
2129 | "%s: xfs_imap returned error %d.", | |
2130 | __func__, error); | |
2131 | return error; | |
2132 | } | |
2133 | ||
2134 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
2135 | &last_ibp, 0, 0); | |
1da177e4 | 2136 | if (error) { |
0b932ccc | 2137 | xfs_warn(mp, |
129dbc9a | 2138 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2139 | __func__, error); |
1da177e4 LT |
2140 | return error; |
2141 | } | |
129dbc9a CH |
2142 | |
2143 | last_offset = imap.im_boffset; | |
347d1c01 | 2144 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2145 | ASSERT(next_agino != NULLAGINO); |
2146 | ASSERT(next_agino != 0); | |
2147 | } | |
475ee413 | 2148 | |
1da177e4 | 2149 | /* |
475ee413 CH |
2150 | * Now last_ibp points to the buffer previous to us on the |
2151 | * unlinked list. Pull us from the list. | |
1da177e4 | 2152 | */ |
475ee413 CH |
2153 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2154 | 0, 0); | |
1da177e4 | 2155 | if (error) { |
475ee413 | 2156 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2157 | __func__, error); |
1da177e4 LT |
2158 | return error; |
2159 | } | |
347d1c01 | 2160 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2161 | ASSERT(next_agino != 0); |
2162 | ASSERT(next_agino != agino); | |
2163 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2164 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2165 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2166 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2167 | |
2168 | /* need to recalc the inode CRC if appropriate */ | |
2169 | xfs_dinode_calc_crc(mp, dip); | |
2170 | ||
1da177e4 LT |
2171 | xfs_trans_inode_buf(tp, ibp); |
2172 | xfs_trans_log_buf(tp, ibp, offset, | |
2173 | (offset + sizeof(xfs_agino_t) - 1)); | |
2174 | xfs_inobp_check(mp, ibp); | |
2175 | } else { | |
2176 | xfs_trans_brelse(tp, ibp); | |
2177 | } | |
2178 | /* | |
2179 | * Point the previous inode on the list to the next inode. | |
2180 | */ | |
347d1c01 | 2181 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2182 | ASSERT(next_agino != 0); |
2183 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2184 | |
2185 | /* need to recalc the inode CRC if appropriate */ | |
2186 | xfs_dinode_calc_crc(mp, last_dip); | |
2187 | ||
1da177e4 LT |
2188 | xfs_trans_inode_buf(tp, last_ibp); |
2189 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2190 | (offset + sizeof(xfs_agino_t) - 1)); | |
2191 | xfs_inobp_check(mp, last_ibp); | |
2192 | } | |
2193 | return 0; | |
2194 | } | |
2195 | ||
5b3eed75 | 2196 | /* |
0b8182db | 2197 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2198 | * inodes that are in memory - they all must be marked stale and attached to |
2199 | * the cluster buffer. | |
2200 | */ | |
2a30f36d | 2201 | STATIC int |
1da177e4 | 2202 | xfs_ifree_cluster( |
09b56604 BF |
2203 | xfs_inode_t *free_ip, |
2204 | xfs_trans_t *tp, | |
2205 | struct xfs_icluster *xic) | |
1da177e4 LT |
2206 | { |
2207 | xfs_mount_t *mp = free_ip->i_mount; | |
2208 | int blks_per_cluster; | |
982e939e | 2209 | int inodes_per_cluster; |
1da177e4 | 2210 | int nbufs; |
5b257b4a | 2211 | int i, j; |
3cdaa189 | 2212 | int ioffset; |
1da177e4 LT |
2213 | xfs_daddr_t blkno; |
2214 | xfs_buf_t *bp; | |
5b257b4a | 2215 | xfs_inode_t *ip; |
1da177e4 LT |
2216 | xfs_inode_log_item_t *iip; |
2217 | xfs_log_item_t *lip; | |
5017e97d | 2218 | struct xfs_perag *pag; |
09b56604 | 2219 | xfs_ino_t inum; |
1da177e4 | 2220 | |
09b56604 | 2221 | inum = xic->first_ino; |
5017e97d | 2222 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
982e939e JL |
2223 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
2224 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
2225 | nbufs = mp->m_ialloc_blks / blks_per_cluster; | |
1da177e4 | 2226 | |
982e939e | 2227 | for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { |
09b56604 BF |
2228 | /* |
2229 | * The allocation bitmap tells us which inodes of the chunk were | |
2230 | * physically allocated. Skip the cluster if an inode falls into | |
2231 | * a sparse region. | |
2232 | */ | |
3cdaa189 BF |
2233 | ioffset = inum - xic->first_ino; |
2234 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
2235 | ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); | |
09b56604 BF |
2236 | continue; |
2237 | } | |
2238 | ||
1da177e4 LT |
2239 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2240 | XFS_INO_TO_AGBNO(mp, inum)); | |
2241 | ||
5b257b4a DC |
2242 | /* |
2243 | * We obtain and lock the backing buffer first in the process | |
2244 | * here, as we have to ensure that any dirty inode that we | |
2245 | * can't get the flush lock on is attached to the buffer. | |
2246 | * If we scan the in-memory inodes first, then buffer IO can | |
2247 | * complete before we get a lock on it, and hence we may fail | |
2248 | * to mark all the active inodes on the buffer stale. | |
2249 | */ | |
2250 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2251 | mp->m_bsize * blks_per_cluster, |
2252 | XBF_UNMAPPED); | |
5b257b4a | 2253 | |
2a30f36d | 2254 | if (!bp) |
2451337d | 2255 | return -ENOMEM; |
b0f539de DC |
2256 | |
2257 | /* | |
2258 | * This buffer may not have been correctly initialised as we | |
2259 | * didn't read it from disk. That's not important because we are | |
2260 | * only using to mark the buffer as stale in the log, and to | |
2261 | * attach stale cached inodes on it. That means it will never be | |
2262 | * dispatched for IO. If it is, we want to know about it, and we | |
2263 | * want it to fail. We can acheive this by adding a write | |
2264 | * verifier to the buffer. | |
2265 | */ | |
1813dd64 | 2266 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2267 | |
5b257b4a DC |
2268 | /* |
2269 | * Walk the inodes already attached to the buffer and mark them | |
2270 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2271 | * in-memory inode walk can't lock them. By marking them all |
2272 | * stale first, we will not attempt to lock them in the loop | |
2273 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 2274 | */ |
adadbeef | 2275 | lip = bp->b_fspriv; |
5b257b4a DC |
2276 | while (lip) { |
2277 | if (lip->li_type == XFS_LI_INODE) { | |
2278 | iip = (xfs_inode_log_item_t *)lip; | |
2279 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2280 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2281 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2282 | &iip->ili_flush_lsn, | |
2283 | &iip->ili_item.li_lsn); | |
2284 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
2285 | } |
2286 | lip = lip->li_bio_list; | |
2287 | } | |
1da177e4 | 2288 | |
5b3eed75 | 2289 | |
1da177e4 | 2290 | /* |
5b257b4a DC |
2291 | * For each inode in memory attempt to add it to the inode |
2292 | * buffer and set it up for being staled on buffer IO | |
2293 | * completion. This is safe as we've locked out tail pushing | |
2294 | * and flushing by locking the buffer. | |
1da177e4 | 2295 | * |
5b257b4a DC |
2296 | * We have already marked every inode that was part of a |
2297 | * transaction stale above, which means there is no point in | |
2298 | * even trying to lock them. | |
1da177e4 | 2299 | */ |
982e939e | 2300 | for (i = 0; i < inodes_per_cluster; i++) { |
5b3eed75 | 2301 | retry: |
1a3e8f3d | 2302 | rcu_read_lock(); |
da353b0d DC |
2303 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2304 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2305 | |
1a3e8f3d DC |
2306 | /* Inode not in memory, nothing to do */ |
2307 | if (!ip) { | |
2308 | rcu_read_unlock(); | |
1da177e4 LT |
2309 | continue; |
2310 | } | |
2311 | ||
1a3e8f3d DC |
2312 | /* |
2313 | * because this is an RCU protected lookup, we could | |
2314 | * find a recently freed or even reallocated inode | |
2315 | * during the lookup. We need to check under the | |
2316 | * i_flags_lock for a valid inode here. Skip it if it | |
2317 | * is not valid, the wrong inode or stale. | |
2318 | */ | |
2319 | spin_lock(&ip->i_flags_lock); | |
2320 | if (ip->i_ino != inum + i || | |
2321 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
2322 | spin_unlock(&ip->i_flags_lock); | |
2323 | rcu_read_unlock(); | |
2324 | continue; | |
2325 | } | |
2326 | spin_unlock(&ip->i_flags_lock); | |
2327 | ||
5b3eed75 DC |
2328 | /* |
2329 | * Don't try to lock/unlock the current inode, but we | |
2330 | * _cannot_ skip the other inodes that we did not find | |
2331 | * in the list attached to the buffer and are not | |
2332 | * already marked stale. If we can't lock it, back off | |
2333 | * and retry. | |
2334 | */ | |
5b257b4a DC |
2335 | if (ip != free_ip && |
2336 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 2337 | rcu_read_unlock(); |
5b3eed75 DC |
2338 | delay(1); |
2339 | goto retry; | |
1da177e4 | 2340 | } |
1a3e8f3d | 2341 | rcu_read_unlock(); |
1da177e4 | 2342 | |
5b3eed75 | 2343 | xfs_iflock(ip); |
5b257b4a | 2344 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2345 | |
5b3eed75 DC |
2346 | /* |
2347 | * we don't need to attach clean inodes or those only | |
2348 | * with unlogged changes (which we throw away, anyway). | |
2349 | */ | |
1da177e4 | 2350 | iip = ip->i_itemp; |
5b3eed75 | 2351 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2352 | ASSERT(ip != free_ip); |
1da177e4 LT |
2353 | xfs_ifunlock(ip); |
2354 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2355 | continue; | |
2356 | } | |
2357 | ||
f5d8d5c4 CH |
2358 | iip->ili_last_fields = iip->ili_fields; |
2359 | iip->ili_fields = 0; | |
fc0561ce | 2360 | iip->ili_fsync_fields = 0; |
1da177e4 | 2361 | iip->ili_logged = 1; |
7b2e2a31 DC |
2362 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2363 | &iip->ili_item.li_lsn); | |
1da177e4 | 2364 | |
ca30b2a7 CH |
2365 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2366 | &iip->ili_item); | |
5b257b4a DC |
2367 | |
2368 | if (ip != free_ip) | |
1da177e4 | 2369 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2370 | } |
2371 | ||
5b3eed75 | 2372 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2373 | xfs_trans_binval(tp, bp); |
2374 | } | |
2375 | ||
5017e97d | 2376 | xfs_perag_put(pag); |
2a30f36d | 2377 | return 0; |
1da177e4 LT |
2378 | } |
2379 | ||
2380 | /* | |
2381 | * This is called to return an inode to the inode free list. | |
2382 | * The inode should already be truncated to 0 length and have | |
2383 | * no pages associated with it. This routine also assumes that | |
2384 | * the inode is already a part of the transaction. | |
2385 | * | |
2386 | * The on-disk copy of the inode will have been added to the list | |
2387 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2388 | * that list atomically with respect to freeing it here. | |
2389 | */ | |
2390 | int | |
2391 | xfs_ifree( | |
2392 | xfs_trans_t *tp, | |
2393 | xfs_inode_t *ip, | |
2394 | xfs_bmap_free_t *flist) | |
2395 | { | |
2396 | int error; | |
09b56604 | 2397 | struct xfs_icluster xic = { 0 }; |
1da177e4 | 2398 | |
579aa9ca | 2399 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2400 | ASSERT(VFS_I(ip)->i_nlink == 0); |
1da177e4 LT |
2401 | ASSERT(ip->i_d.di_nextents == 0); |
2402 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 2403 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
2404 | ASSERT(ip->i_d.di_nblocks == 0); |
2405 | ||
2406 | /* | |
2407 | * Pull the on-disk inode from the AGI unlinked list. | |
2408 | */ | |
2409 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2410 | if (error) |
1da177e4 | 2411 | return error; |
1da177e4 | 2412 | |
09b56604 | 2413 | error = xfs_difree(tp, ip->i_ino, flist, &xic); |
1baaed8f | 2414 | if (error) |
1da177e4 | 2415 | return error; |
1baaed8f | 2416 | |
1da177e4 LT |
2417 | ip->i_d.di_mode = 0; /* mark incore inode as free */ |
2418 | ip->i_d.di_flags = 0; | |
2419 | ip->i_d.di_dmevmask = 0; | |
2420 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2421 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2422 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2423 | /* | |
2424 | * Bump the generation count so no one will be confused | |
2425 | * by reincarnations of this inode. | |
2426 | */ | |
2427 | ip->i_d.di_gen++; | |
2428 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2429 | ||
09b56604 BF |
2430 | if (xic.deleted) |
2431 | error = xfs_ifree_cluster(ip, tp, &xic); | |
1da177e4 | 2432 | |
2a30f36d | 2433 | return error; |
1da177e4 LT |
2434 | } |
2435 | ||
1da177e4 | 2436 | /* |
60ec6783 CH |
2437 | * This is called to unpin an inode. The caller must have the inode locked |
2438 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2439 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2440 | */ |
60ec6783 | 2441 | static void |
f392e631 | 2442 | xfs_iunpin( |
60ec6783 | 2443 | struct xfs_inode *ip) |
1da177e4 | 2444 | { |
579aa9ca | 2445 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2446 | |
4aaf15d1 DC |
2447 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2448 | ||
a3f74ffb | 2449 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2450 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2451 | |
a3f74ffb | 2452 | } |
1da177e4 | 2453 | |
f392e631 CH |
2454 | static void |
2455 | __xfs_iunpin_wait( | |
2456 | struct xfs_inode *ip) | |
2457 | { | |
2458 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2459 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2460 | ||
2461 | xfs_iunpin(ip); | |
2462 | ||
2463 | do { | |
2464 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2465 | if (xfs_ipincount(ip)) | |
2466 | io_schedule(); | |
2467 | } while (xfs_ipincount(ip)); | |
2468 | finish_wait(wq, &wait.wait); | |
2469 | } | |
2470 | ||
777df5af | 2471 | void |
a3f74ffb | 2472 | xfs_iunpin_wait( |
60ec6783 | 2473 | struct xfs_inode *ip) |
a3f74ffb | 2474 | { |
f392e631 CH |
2475 | if (xfs_ipincount(ip)) |
2476 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2477 | } |
2478 | ||
27320369 DC |
2479 | /* |
2480 | * Removing an inode from the namespace involves removing the directory entry | |
2481 | * and dropping the link count on the inode. Removing the directory entry can | |
2482 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2483 | * count can result in placing the inode on an unlinked list which results in | |
2484 | * locking an AGI. | |
2485 | * | |
2486 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2487 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2488 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2489 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2490 | * first, and then freeing the inode can result in an inode chunk being freed | |
2491 | * and hence freeing disk space requiring that we lock an AGF. | |
2492 | * | |
2493 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2494 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2495 | * reference count and put it on the unlinked list as this results in a lock | |
2496 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2497 | * freeing. Therefore we must drop the link counts before we remove the | |
2498 | * directory entry. | |
2499 | * | |
2500 | * This is still safe from a transactional point of view - it is not until we | |
2501 | * get to xfs_bmap_finish() that we have the possibility of multiple | |
2502 | * transactions in this operation. Hence as long as we remove the directory | |
2503 | * entry and drop the link count in the first transaction of the remove | |
2504 | * operation, there are no transactional constraints on the ordering here. | |
2505 | */ | |
c24b5dfa DC |
2506 | int |
2507 | xfs_remove( | |
2508 | xfs_inode_t *dp, | |
2509 | struct xfs_name *name, | |
2510 | xfs_inode_t *ip) | |
2511 | { | |
2512 | xfs_mount_t *mp = dp->i_mount; | |
2513 | xfs_trans_t *tp = NULL; | |
2514 | int is_dir = S_ISDIR(ip->i_d.di_mode); | |
2515 | int error = 0; | |
2516 | xfs_bmap_free_t free_list; | |
2517 | xfs_fsblock_t first_block; | |
c24b5dfa | 2518 | uint resblks; |
c24b5dfa DC |
2519 | |
2520 | trace_xfs_remove(dp, name); | |
2521 | ||
2522 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 2523 | return -EIO; |
c24b5dfa DC |
2524 | |
2525 | error = xfs_qm_dqattach(dp, 0); | |
2526 | if (error) | |
2527 | goto std_return; | |
2528 | ||
2529 | error = xfs_qm_dqattach(ip, 0); | |
2530 | if (error) | |
2531 | goto std_return; | |
2532 | ||
32296f86 | 2533 | if (is_dir) |
c24b5dfa | 2534 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); |
32296f86 | 2535 | else |
c24b5dfa | 2536 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); |
c24b5dfa DC |
2537 | |
2538 | /* | |
2539 | * We try to get the real space reservation first, | |
2540 | * allowing for directory btree deletion(s) implying | |
2541 | * possible bmap insert(s). If we can't get the space | |
2542 | * reservation then we use 0 instead, and avoid the bmap | |
2543 | * btree insert(s) in the directory code by, if the bmap | |
2544 | * insert tries to happen, instead trimming the LAST | |
2545 | * block from the directory. | |
2546 | */ | |
2547 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3d3c8b52 | 2548 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0); |
2451337d | 2549 | if (error == -ENOSPC) { |
c24b5dfa | 2550 | resblks = 0; |
3d3c8b52 | 2551 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0); |
c24b5dfa DC |
2552 | } |
2553 | if (error) { | |
2451337d | 2554 | ASSERT(error != -ENOSPC); |
c24b5dfa DC |
2555 | goto out_trans_cancel; |
2556 | } | |
2557 | ||
dbad7c99 | 2558 | xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); |
c24b5dfa DC |
2559 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
2560 | ||
dbad7c99 | 2561 | xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
c24b5dfa DC |
2562 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
2563 | ||
2564 | /* | |
2565 | * If we're removing a directory perform some additional validation. | |
2566 | */ | |
2567 | if (is_dir) { | |
54d7b5c1 DC |
2568 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2569 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2570 | error = -ENOTEMPTY; |
c24b5dfa DC |
2571 | goto out_trans_cancel; |
2572 | } | |
2573 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2574 | error = -ENOTEMPTY; |
c24b5dfa DC |
2575 | goto out_trans_cancel; |
2576 | } | |
c24b5dfa | 2577 | |
27320369 | 2578 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2579 | error = xfs_droplink(tp, dp); |
2580 | if (error) | |
27320369 | 2581 | goto out_trans_cancel; |
c24b5dfa | 2582 | |
27320369 | 2583 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2584 | error = xfs_droplink(tp, ip); |
2585 | if (error) | |
27320369 | 2586 | goto out_trans_cancel; |
c24b5dfa DC |
2587 | } else { |
2588 | /* | |
2589 | * When removing a non-directory we need to log the parent | |
2590 | * inode here. For a directory this is done implicitly | |
2591 | * by the xfs_droplink call for the ".." entry. | |
2592 | */ | |
2593 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2594 | } | |
27320369 | 2595 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2596 | |
27320369 | 2597 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2598 | error = xfs_droplink(tp, ip); |
2599 | if (error) | |
27320369 | 2600 | goto out_trans_cancel; |
c24b5dfa | 2601 | |
27320369 DC |
2602 | xfs_bmap_init(&free_list, &first_block); |
2603 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, | |
2604 | &first_block, &free_list, resblks); | |
2605 | if (error) { | |
2451337d | 2606 | ASSERT(error != -ENOENT); |
27320369 DC |
2607 | goto out_bmap_cancel; |
2608 | } | |
2609 | ||
c24b5dfa DC |
2610 | /* |
2611 | * If this is a synchronous mount, make sure that the | |
2612 | * remove transaction goes to disk before returning to | |
2613 | * the user. | |
2614 | */ | |
2615 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2616 | xfs_trans_set_sync(tp); | |
2617 | ||
f6106efa | 2618 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
c24b5dfa DC |
2619 | if (error) |
2620 | goto out_bmap_cancel; | |
2621 | ||
70393313 | 2622 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2623 | if (error) |
2624 | goto std_return; | |
2625 | ||
2cd2ef6a | 2626 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2627 | xfs_filestream_deassociate(ip); |
2628 | ||
2629 | return 0; | |
2630 | ||
2631 | out_bmap_cancel: | |
2632 | xfs_bmap_cancel(&free_list); | |
c24b5dfa | 2633 | out_trans_cancel: |
4906e215 | 2634 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2635 | std_return: |
2636 | return error; | |
2637 | } | |
2638 | ||
f6bba201 DC |
2639 | /* |
2640 | * Enter all inodes for a rename transaction into a sorted array. | |
2641 | */ | |
95afcf5c | 2642 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2643 | STATIC void |
2644 | xfs_sort_for_rename( | |
95afcf5c DC |
2645 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2646 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2647 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2648 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2649 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2650 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2651 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2652 | { |
f6bba201 DC |
2653 | int i, j; |
2654 | ||
95afcf5c DC |
2655 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2656 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2657 | ||
f6bba201 DC |
2658 | /* |
2659 | * i_tab contains a list of pointers to inodes. We initialize | |
2660 | * the table here & we'll sort it. We will then use it to | |
2661 | * order the acquisition of the inode locks. | |
2662 | * | |
2663 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2664 | */ | |
95afcf5c DC |
2665 | i = 0; |
2666 | i_tab[i++] = dp1; | |
2667 | i_tab[i++] = dp2; | |
2668 | i_tab[i++] = ip1; | |
2669 | if (ip2) | |
2670 | i_tab[i++] = ip2; | |
2671 | if (wip) | |
2672 | i_tab[i++] = wip; | |
2673 | *num_inodes = i; | |
f6bba201 DC |
2674 | |
2675 | /* | |
2676 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2677 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2678 | */ |
2679 | for (i = 0; i < *num_inodes; i++) { | |
2680 | for (j = 1; j < *num_inodes; j++) { | |
2681 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2682 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2683 | i_tab[j] = i_tab[j-1]; |
2684 | i_tab[j-1] = temp; | |
2685 | } | |
2686 | } | |
2687 | } | |
2688 | } | |
2689 | ||
310606b0 DC |
2690 | static int |
2691 | xfs_finish_rename( | |
2692 | struct xfs_trans *tp, | |
2693 | struct xfs_bmap_free *free_list) | |
2694 | { | |
310606b0 DC |
2695 | int error; |
2696 | ||
2697 | /* | |
2698 | * If this is a synchronous mount, make sure that the rename transaction | |
2699 | * goes to disk before returning to the user. | |
2700 | */ | |
2701 | if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2702 | xfs_trans_set_sync(tp); | |
2703 | ||
f6106efa | 2704 | error = xfs_bmap_finish(&tp, free_list, NULL); |
310606b0 DC |
2705 | if (error) { |
2706 | xfs_bmap_cancel(free_list); | |
4906e215 | 2707 | xfs_trans_cancel(tp); |
310606b0 DC |
2708 | return error; |
2709 | } | |
2710 | ||
70393313 | 2711 | return xfs_trans_commit(tp); |
310606b0 DC |
2712 | } |
2713 | ||
d31a1825 CM |
2714 | /* |
2715 | * xfs_cross_rename() | |
2716 | * | |
2717 | * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall | |
2718 | */ | |
2719 | STATIC int | |
2720 | xfs_cross_rename( | |
2721 | struct xfs_trans *tp, | |
2722 | struct xfs_inode *dp1, | |
2723 | struct xfs_name *name1, | |
2724 | struct xfs_inode *ip1, | |
2725 | struct xfs_inode *dp2, | |
2726 | struct xfs_name *name2, | |
2727 | struct xfs_inode *ip2, | |
2728 | struct xfs_bmap_free *free_list, | |
2729 | xfs_fsblock_t *first_block, | |
2730 | int spaceres) | |
2731 | { | |
2732 | int error = 0; | |
2733 | int ip1_flags = 0; | |
2734 | int ip2_flags = 0; | |
2735 | int dp2_flags = 0; | |
2736 | ||
2737 | /* Swap inode number for dirent in first parent */ | |
2738 | error = xfs_dir_replace(tp, dp1, name1, | |
2739 | ip2->i_ino, | |
2740 | first_block, free_list, spaceres); | |
2741 | if (error) | |
eeacd321 | 2742 | goto out_trans_abort; |
d31a1825 CM |
2743 | |
2744 | /* Swap inode number for dirent in second parent */ | |
2745 | error = xfs_dir_replace(tp, dp2, name2, | |
2746 | ip1->i_ino, | |
2747 | first_block, free_list, spaceres); | |
2748 | if (error) | |
eeacd321 | 2749 | goto out_trans_abort; |
d31a1825 CM |
2750 | |
2751 | /* | |
2752 | * If we're renaming one or more directories across different parents, | |
2753 | * update the respective ".." entries (and link counts) to match the new | |
2754 | * parents. | |
2755 | */ | |
2756 | if (dp1 != dp2) { | |
2757 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2758 | ||
2759 | if (S_ISDIR(ip2->i_d.di_mode)) { | |
2760 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, | |
2761 | dp1->i_ino, first_block, | |
2762 | free_list, spaceres); | |
2763 | if (error) | |
eeacd321 | 2764 | goto out_trans_abort; |
d31a1825 CM |
2765 | |
2766 | /* transfer ip2 ".." reference to dp1 */ | |
2767 | if (!S_ISDIR(ip1->i_d.di_mode)) { | |
2768 | error = xfs_droplink(tp, dp2); | |
2769 | if (error) | |
eeacd321 | 2770 | goto out_trans_abort; |
d31a1825 CM |
2771 | error = xfs_bumplink(tp, dp1); |
2772 | if (error) | |
eeacd321 | 2773 | goto out_trans_abort; |
d31a1825 CM |
2774 | } |
2775 | ||
2776 | /* | |
2777 | * Although ip1 isn't changed here, userspace needs | |
2778 | * to be warned about the change, so that applications | |
2779 | * relying on it (like backup ones), will properly | |
2780 | * notify the change | |
2781 | */ | |
2782 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2783 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2784 | } | |
2785 | ||
2786 | if (S_ISDIR(ip1->i_d.di_mode)) { | |
2787 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, | |
2788 | dp2->i_ino, first_block, | |
2789 | free_list, spaceres); | |
2790 | if (error) | |
eeacd321 | 2791 | goto out_trans_abort; |
d31a1825 CM |
2792 | |
2793 | /* transfer ip1 ".." reference to dp2 */ | |
2794 | if (!S_ISDIR(ip2->i_d.di_mode)) { | |
2795 | error = xfs_droplink(tp, dp1); | |
2796 | if (error) | |
eeacd321 | 2797 | goto out_trans_abort; |
d31a1825 CM |
2798 | error = xfs_bumplink(tp, dp2); |
2799 | if (error) | |
eeacd321 | 2800 | goto out_trans_abort; |
d31a1825 CM |
2801 | } |
2802 | ||
2803 | /* | |
2804 | * Although ip2 isn't changed here, userspace needs | |
2805 | * to be warned about the change, so that applications | |
2806 | * relying on it (like backup ones), will properly | |
2807 | * notify the change | |
2808 | */ | |
2809 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2810 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2811 | } | |
2812 | } | |
2813 | ||
2814 | if (ip1_flags) { | |
2815 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2816 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2817 | } | |
2818 | if (ip2_flags) { | |
2819 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2820 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2821 | } | |
2822 | if (dp2_flags) { | |
2823 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2824 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2825 | } | |
2826 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2827 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
eeacd321 DC |
2828 | return xfs_finish_rename(tp, free_list); |
2829 | ||
2830 | out_trans_abort: | |
2831 | xfs_bmap_cancel(free_list); | |
4906e215 | 2832 | xfs_trans_cancel(tp); |
d31a1825 CM |
2833 | return error; |
2834 | } | |
2835 | ||
7dcf5c3e DC |
2836 | /* |
2837 | * xfs_rename_alloc_whiteout() | |
2838 | * | |
2839 | * Return a referenced, unlinked, unlocked inode that that can be used as a | |
2840 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we | |
2841 | * crash between allocating the inode and linking it into the rename transaction | |
2842 | * recovery will free the inode and we won't leak it. | |
2843 | */ | |
2844 | static int | |
2845 | xfs_rename_alloc_whiteout( | |
2846 | struct xfs_inode *dp, | |
2847 | struct xfs_inode **wip) | |
2848 | { | |
2849 | struct xfs_inode *tmpfile; | |
2850 | int error; | |
2851 | ||
2852 | error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); | |
2853 | if (error) | |
2854 | return error; | |
2855 | ||
22419ac9 BF |
2856 | /* |
2857 | * Prepare the tmpfile inode as if it were created through the VFS. | |
2858 | * Otherwise, the link increment paths will complain about nlink 0->1. | |
2859 | * Drop the link count as done by d_tmpfile(), complete the inode setup | |
2860 | * and flag it as linkable. | |
2861 | */ | |
2862 | drop_nlink(VFS_I(tmpfile)); | |
7dcf5c3e DC |
2863 | xfs_finish_inode_setup(tmpfile); |
2864 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2865 | ||
2866 | *wip = tmpfile; | |
2867 | return 0; | |
2868 | } | |
2869 | ||
f6bba201 DC |
2870 | /* |
2871 | * xfs_rename | |
2872 | */ | |
2873 | int | |
2874 | xfs_rename( | |
7dcf5c3e DC |
2875 | struct xfs_inode *src_dp, |
2876 | struct xfs_name *src_name, | |
2877 | struct xfs_inode *src_ip, | |
2878 | struct xfs_inode *target_dp, | |
2879 | struct xfs_name *target_name, | |
2880 | struct xfs_inode *target_ip, | |
2881 | unsigned int flags) | |
f6bba201 | 2882 | { |
7dcf5c3e DC |
2883 | struct xfs_mount *mp = src_dp->i_mount; |
2884 | struct xfs_trans *tp; | |
2885 | struct xfs_bmap_free free_list; | |
2886 | xfs_fsblock_t first_block; | |
2887 | struct xfs_inode *wip = NULL; /* whiteout inode */ | |
2888 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
2889 | int num_inodes = __XFS_SORT_INODES; | |
2b93681f DC |
2890 | bool new_parent = (src_dp != target_dp); |
2891 | bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode); | |
7dcf5c3e DC |
2892 | int spaceres; |
2893 | int error; | |
f6bba201 DC |
2894 | |
2895 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2896 | ||
eeacd321 DC |
2897 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2898 | return -EINVAL; | |
2899 | ||
7dcf5c3e DC |
2900 | /* |
2901 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2902 | * we will be placing at the target and ensure the type is set | |
2903 | * appropriately. | |
2904 | */ | |
2905 | if (flags & RENAME_WHITEOUT) { | |
2906 | ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); | |
2907 | error = xfs_rename_alloc_whiteout(target_dp, &wip); | |
2908 | if (error) | |
2909 | return error; | |
2910 | ||
2911 | /* setup target dirent info as whiteout */ | |
2912 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2913 | } | |
f6bba201 | 2914 | |
7dcf5c3e | 2915 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2916 | inodes, &num_inodes); |
2917 | ||
f6bba201 | 2918 | tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME); |
f6bba201 | 2919 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
3d3c8b52 | 2920 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0); |
2451337d | 2921 | if (error == -ENOSPC) { |
f6bba201 | 2922 | spaceres = 0; |
3d3c8b52 | 2923 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0); |
f6bba201 | 2924 | } |
445883e8 DC |
2925 | if (error) |
2926 | goto out_trans_cancel; | |
f6bba201 DC |
2927 | |
2928 | /* | |
2929 | * Attach the dquots to the inodes | |
2930 | */ | |
2931 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2932 | if (error) |
2933 | goto out_trans_cancel; | |
f6bba201 DC |
2934 | |
2935 | /* | |
2936 | * Lock all the participating inodes. Depending upon whether | |
2937 | * the target_name exists in the target directory, and | |
2938 | * whether the target directory is the same as the source | |
2939 | * directory, we can lock from 2 to 4 inodes. | |
2940 | */ | |
dbad7c99 DC |
2941 | if (!new_parent) |
2942 | xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); | |
2943 | else | |
2944 | xfs_lock_two_inodes(src_dp, target_dp, | |
2945 | XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); | |
2946 | ||
f6bba201 DC |
2947 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); |
2948 | ||
2949 | /* | |
2950 | * Join all the inodes to the transaction. From this point on, | |
2951 | * we can rely on either trans_commit or trans_cancel to unlock | |
2952 | * them. | |
2953 | */ | |
dbad7c99 | 2954 | xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
f6bba201 | 2955 | if (new_parent) |
dbad7c99 | 2956 | xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
f6bba201 DC |
2957 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2958 | if (target_ip) | |
2959 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2960 | if (wip) |
2961 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
2962 | |
2963 | /* | |
2964 | * If we are using project inheritance, we only allow renames | |
2965 | * into our tree when the project IDs are the same; else the | |
2966 | * tree quota mechanism would be circumvented. | |
2967 | */ | |
2968 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2969 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2451337d | 2970 | error = -EXDEV; |
445883e8 | 2971 | goto out_trans_cancel; |
f6bba201 DC |
2972 | } |
2973 | ||
445883e8 DC |
2974 | xfs_bmap_init(&free_list, &first_block); |
2975 | ||
eeacd321 DC |
2976 | /* RENAME_EXCHANGE is unique from here on. */ |
2977 | if (flags & RENAME_EXCHANGE) | |
2978 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
2979 | target_dp, target_name, target_ip, | |
2980 | &free_list, &first_block, spaceres); | |
d31a1825 | 2981 | |
f6bba201 DC |
2982 | /* |
2983 | * Set up the target. | |
2984 | */ | |
2985 | if (target_ip == NULL) { | |
2986 | /* | |
2987 | * If there's no space reservation, check the entry will | |
2988 | * fit before actually inserting it. | |
2989 | */ | |
94f3cad5 ES |
2990 | if (!spaceres) { |
2991 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
2992 | if (error) | |
445883e8 | 2993 | goto out_trans_cancel; |
94f3cad5 | 2994 | } |
f6bba201 DC |
2995 | /* |
2996 | * If target does not exist and the rename crosses | |
2997 | * directories, adjust the target directory link count | |
2998 | * to account for the ".." reference from the new entry. | |
2999 | */ | |
3000 | error = xfs_dir_createname(tp, target_dp, target_name, | |
3001 | src_ip->i_ino, &first_block, | |
3002 | &free_list, spaceres); | |
f6bba201 | 3003 | if (error) |
4906e215 | 3004 | goto out_bmap_cancel; |
f6bba201 DC |
3005 | |
3006 | xfs_trans_ichgtime(tp, target_dp, | |
3007 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3008 | ||
3009 | if (new_parent && src_is_directory) { | |
3010 | error = xfs_bumplink(tp, target_dp); | |
3011 | if (error) | |
4906e215 | 3012 | goto out_bmap_cancel; |
f6bba201 DC |
3013 | } |
3014 | } else { /* target_ip != NULL */ | |
3015 | /* | |
3016 | * If target exists and it's a directory, check that both | |
3017 | * target and source are directories and that target can be | |
3018 | * destroyed, or that neither is a directory. | |
3019 | */ | |
3020 | if (S_ISDIR(target_ip->i_d.di_mode)) { | |
3021 | /* | |
3022 | * Make sure target dir is empty. | |
3023 | */ | |
3024 | if (!(xfs_dir_isempty(target_ip)) || | |
54d7b5c1 | 3025 | (VFS_I(target_ip)->i_nlink > 2)) { |
2451337d | 3026 | error = -EEXIST; |
445883e8 | 3027 | goto out_trans_cancel; |
f6bba201 DC |
3028 | } |
3029 | } | |
3030 | ||
3031 | /* | |
3032 | * Link the source inode under the target name. | |
3033 | * If the source inode is a directory and we are moving | |
3034 | * it across directories, its ".." entry will be | |
3035 | * inconsistent until we replace that down below. | |
3036 | * | |
3037 | * In case there is already an entry with the same | |
3038 | * name at the destination directory, remove it first. | |
3039 | */ | |
3040 | error = xfs_dir_replace(tp, target_dp, target_name, | |
3041 | src_ip->i_ino, | |
3042 | &first_block, &free_list, spaceres); | |
3043 | if (error) | |
4906e215 | 3044 | goto out_bmap_cancel; |
f6bba201 DC |
3045 | |
3046 | xfs_trans_ichgtime(tp, target_dp, | |
3047 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3048 | ||
3049 | /* | |
3050 | * Decrement the link count on the target since the target | |
3051 | * dir no longer points to it. | |
3052 | */ | |
3053 | error = xfs_droplink(tp, target_ip); | |
3054 | if (error) | |
4906e215 | 3055 | goto out_bmap_cancel; |
f6bba201 DC |
3056 | |
3057 | if (src_is_directory) { | |
3058 | /* | |
3059 | * Drop the link from the old "." entry. | |
3060 | */ | |
3061 | error = xfs_droplink(tp, target_ip); | |
3062 | if (error) | |
4906e215 | 3063 | goto out_bmap_cancel; |
f6bba201 DC |
3064 | } |
3065 | } /* target_ip != NULL */ | |
3066 | ||
3067 | /* | |
3068 | * Remove the source. | |
3069 | */ | |
3070 | if (new_parent && src_is_directory) { | |
3071 | /* | |
3072 | * Rewrite the ".." entry to point to the new | |
3073 | * directory. | |
3074 | */ | |
3075 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
3076 | target_dp->i_ino, | |
3077 | &first_block, &free_list, spaceres); | |
2451337d | 3078 | ASSERT(error != -EEXIST); |
f6bba201 | 3079 | if (error) |
4906e215 | 3080 | goto out_bmap_cancel; |
f6bba201 DC |
3081 | } |
3082 | ||
3083 | /* | |
3084 | * We always want to hit the ctime on the source inode. | |
3085 | * | |
3086 | * This isn't strictly required by the standards since the source | |
3087 | * inode isn't really being changed, but old unix file systems did | |
3088 | * it and some incremental backup programs won't work without it. | |
3089 | */ | |
3090 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3091 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3092 | ||
3093 | /* | |
3094 | * Adjust the link count on src_dp. This is necessary when | |
3095 | * renaming a directory, either within one parent when | |
3096 | * the target existed, or across two parent directories. | |
3097 | */ | |
3098 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3099 | ||
3100 | /* | |
3101 | * Decrement link count on src_directory since the | |
3102 | * entry that's moved no longer points to it. | |
3103 | */ | |
3104 | error = xfs_droplink(tp, src_dp); | |
3105 | if (error) | |
4906e215 | 3106 | goto out_bmap_cancel; |
f6bba201 DC |
3107 | } |
3108 | ||
7dcf5c3e DC |
3109 | /* |
3110 | * For whiteouts, we only need to update the source dirent with the | |
3111 | * inode number of the whiteout inode rather than removing it | |
3112 | * altogether. | |
3113 | */ | |
3114 | if (wip) { | |
3115 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, | |
f6bba201 | 3116 | &first_block, &free_list, spaceres); |
7dcf5c3e DC |
3117 | } else |
3118 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
3119 | &first_block, &free_list, spaceres); | |
f6bba201 | 3120 | if (error) |
4906e215 | 3121 | goto out_bmap_cancel; |
f6bba201 DC |
3122 | |
3123 | /* | |
7dcf5c3e DC |
3124 | * For whiteouts, we need to bump the link count on the whiteout inode. |
3125 | * This means that failures all the way up to this point leave the inode | |
3126 | * on the unlinked list and so cleanup is a simple matter of dropping | |
3127 | * the remaining reference to it. If we fail here after bumping the link | |
3128 | * count, we're shutting down the filesystem so we'll never see the | |
3129 | * intermediate state on disk. | |
f6bba201 | 3130 | */ |
7dcf5c3e | 3131 | if (wip) { |
54d7b5c1 | 3132 | ASSERT(VFS_I(wip)->i_nlink == 0); |
7dcf5c3e DC |
3133 | error = xfs_bumplink(tp, wip); |
3134 | if (error) | |
4906e215 | 3135 | goto out_bmap_cancel; |
7dcf5c3e DC |
3136 | error = xfs_iunlink_remove(tp, wip); |
3137 | if (error) | |
4906e215 | 3138 | goto out_bmap_cancel; |
7dcf5c3e | 3139 | xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); |
f6bba201 | 3140 | |
7dcf5c3e DC |
3141 | /* |
3142 | * Now we have a real link, clear the "I'm a tmpfile" state | |
3143 | * flag from the inode so it doesn't accidentally get misused in | |
3144 | * future. | |
3145 | */ | |
3146 | VFS_I(wip)->i_state &= ~I_LINKABLE; | |
f6bba201 DC |
3147 | } |
3148 | ||
f6bba201 DC |
3149 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3150 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3151 | if (new_parent) | |
3152 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3153 | |
7dcf5c3e DC |
3154 | error = xfs_finish_rename(tp, &free_list); |
3155 | if (wip) | |
3156 | IRELE(wip); | |
3157 | return error; | |
f6bba201 | 3158 | |
445883e8 | 3159 | out_bmap_cancel: |
f6bba201 | 3160 | xfs_bmap_cancel(&free_list); |
445883e8 | 3161 | out_trans_cancel: |
4906e215 | 3162 | xfs_trans_cancel(tp); |
7dcf5c3e DC |
3163 | if (wip) |
3164 | IRELE(wip); | |
f6bba201 DC |
3165 | return error; |
3166 | } | |
3167 | ||
5c4d97d0 DC |
3168 | STATIC int |
3169 | xfs_iflush_cluster( | |
3170 | xfs_inode_t *ip, | |
3171 | xfs_buf_t *bp) | |
1da177e4 | 3172 | { |
5c4d97d0 DC |
3173 | xfs_mount_t *mp = ip->i_mount; |
3174 | struct xfs_perag *pag; | |
3175 | unsigned long first_index, mask; | |
3176 | unsigned long inodes_per_cluster; | |
3177 | int ilist_size; | |
3178 | xfs_inode_t **ilist; | |
3179 | xfs_inode_t *iq; | |
3180 | int nr_found; | |
3181 | int clcount = 0; | |
3182 | int bufwasdelwri; | |
1da177e4 | 3183 | int i; |
1da177e4 | 3184 | |
5c4d97d0 | 3185 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 3186 | |
0f49efd8 | 3187 | inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; |
5c4d97d0 DC |
3188 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
3189 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); | |
3190 | if (!ilist) | |
3191 | goto out_put; | |
1da177e4 | 3192 | |
0f49efd8 | 3193 | mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); |
5c4d97d0 DC |
3194 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
3195 | rcu_read_lock(); | |
3196 | /* really need a gang lookup range call here */ | |
3197 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
3198 | first_index, inodes_per_cluster); | |
3199 | if (nr_found == 0) | |
3200 | goto out_free; | |
3201 | ||
3202 | for (i = 0; i < nr_found; i++) { | |
3203 | iq = ilist[i]; | |
3204 | if (iq == ip) | |
bad55843 | 3205 | continue; |
1a3e8f3d DC |
3206 | |
3207 | /* | |
3208 | * because this is an RCU protected lookup, we could find a | |
3209 | * recently freed or even reallocated inode during the lookup. | |
3210 | * We need to check under the i_flags_lock for a valid inode | |
3211 | * here. Skip it if it is not valid or the wrong inode. | |
3212 | */ | |
3213 | spin_lock(&ip->i_flags_lock); | |
3214 | if (!ip->i_ino || | |
3215 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
3216 | spin_unlock(&ip->i_flags_lock); | |
3217 | continue; | |
3218 | } | |
3219 | spin_unlock(&ip->i_flags_lock); | |
3220 | ||
bad55843 DC |
3221 | /* |
3222 | * Do an un-protected check to see if the inode is dirty and | |
3223 | * is a candidate for flushing. These checks will be repeated | |
3224 | * later after the appropriate locks are acquired. | |
3225 | */ | |
33540408 | 3226 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 3227 | continue; |
bad55843 DC |
3228 | |
3229 | /* | |
3230 | * Try to get locks. If any are unavailable or it is pinned, | |
3231 | * then this inode cannot be flushed and is skipped. | |
3232 | */ | |
3233 | ||
3234 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
3235 | continue; | |
3236 | if (!xfs_iflock_nowait(iq)) { | |
3237 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3238 | continue; | |
3239 | } | |
3240 | if (xfs_ipincount(iq)) { | |
3241 | xfs_ifunlock(iq); | |
3242 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3243 | continue; | |
3244 | } | |
3245 | ||
3246 | /* | |
3247 | * arriving here means that this inode can be flushed. First | |
3248 | * re-check that it's dirty before flushing. | |
3249 | */ | |
33540408 DC |
3250 | if (!xfs_inode_clean(iq)) { |
3251 | int error; | |
bad55843 DC |
3252 | error = xfs_iflush_int(iq, bp); |
3253 | if (error) { | |
3254 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3255 | goto cluster_corrupt_out; | |
3256 | } | |
3257 | clcount++; | |
3258 | } else { | |
3259 | xfs_ifunlock(iq); | |
3260 | } | |
3261 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3262 | } | |
3263 | ||
3264 | if (clcount) { | |
ff6d6af2 BD |
3265 | XFS_STATS_INC(mp, xs_icluster_flushcnt); |
3266 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
bad55843 DC |
3267 | } |
3268 | ||
3269 | out_free: | |
1a3e8f3d | 3270 | rcu_read_unlock(); |
f0e2d93c | 3271 | kmem_free(ilist); |
44b56e0a DC |
3272 | out_put: |
3273 | xfs_perag_put(pag); | |
bad55843 DC |
3274 | return 0; |
3275 | ||
3276 | ||
3277 | cluster_corrupt_out: | |
3278 | /* | |
3279 | * Corruption detected in the clustering loop. Invalidate the | |
3280 | * inode buffer and shut down the filesystem. | |
3281 | */ | |
1a3e8f3d | 3282 | rcu_read_unlock(); |
bad55843 | 3283 | /* |
43ff2122 | 3284 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
3285 | * brelse can handle it with no problems. If not, shut down the |
3286 | * filesystem before releasing the buffer. | |
3287 | */ | |
43ff2122 | 3288 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
3289 | if (bufwasdelwri) |
3290 | xfs_buf_relse(bp); | |
3291 | ||
3292 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3293 | ||
3294 | if (!bufwasdelwri) { | |
3295 | /* | |
3296 | * Just like incore_relse: if we have b_iodone functions, | |
3297 | * mark the buffer as an error and call them. Otherwise | |
3298 | * mark it as stale and brelse. | |
3299 | */ | |
cb669ca5 | 3300 | if (bp->b_iodone) { |
bad55843 | 3301 | XFS_BUF_UNDONE(bp); |
c867cb61 | 3302 | xfs_buf_stale(bp); |
2451337d | 3303 | xfs_buf_ioerror(bp, -EIO); |
e8aaba9a | 3304 | xfs_buf_ioend(bp); |
bad55843 | 3305 | } else { |
c867cb61 | 3306 | xfs_buf_stale(bp); |
bad55843 DC |
3307 | xfs_buf_relse(bp); |
3308 | } | |
3309 | } | |
3310 | ||
3311 | /* | |
3312 | * Unlocks the flush lock | |
3313 | */ | |
04913fdd | 3314 | xfs_iflush_abort(iq, false); |
f0e2d93c | 3315 | kmem_free(ilist); |
44b56e0a | 3316 | xfs_perag_put(pag); |
2451337d | 3317 | return -EFSCORRUPTED; |
bad55843 DC |
3318 | } |
3319 | ||
1da177e4 | 3320 | /* |
4c46819a CH |
3321 | * Flush dirty inode metadata into the backing buffer. |
3322 | * | |
3323 | * The caller must have the inode lock and the inode flush lock held. The | |
3324 | * inode lock will still be held upon return to the caller, and the inode | |
3325 | * flush lock will be released after the inode has reached the disk. | |
3326 | * | |
3327 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3328 | */ |
3329 | int | |
3330 | xfs_iflush( | |
4c46819a CH |
3331 | struct xfs_inode *ip, |
3332 | struct xfs_buf **bpp) | |
1da177e4 | 3333 | { |
4c46819a CH |
3334 | struct xfs_mount *mp = ip->i_mount; |
3335 | struct xfs_buf *bp; | |
3336 | struct xfs_dinode *dip; | |
1da177e4 | 3337 | int error; |
1da177e4 | 3338 | |
ff6d6af2 | 3339 | XFS_STATS_INC(mp, xs_iflush_count); |
1da177e4 | 3340 | |
579aa9ca | 3341 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3342 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3343 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3344 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3345 | |
4c46819a | 3346 | *bpp = NULL; |
1da177e4 | 3347 | |
1da177e4 LT |
3348 | xfs_iunpin_wait(ip); |
3349 | ||
4b6a4688 DC |
3350 | /* |
3351 | * For stale inodes we cannot rely on the backing buffer remaining | |
3352 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3353 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3354 | * inodes below. We have to check this after ensuring the inode is |
3355 | * unpinned so that it is safe to reclaim the stale inode after the | |
3356 | * flush call. | |
3357 | */ | |
3358 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3359 | xfs_ifunlock(ip); | |
3360 | return 0; | |
3361 | } | |
3362 | ||
1da177e4 LT |
3363 | /* |
3364 | * This may have been unpinned because the filesystem is shutting | |
3365 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3366 | * to disk, because the log record didn't make it to disk. |
3367 | * | |
3368 | * We also have to remove the log item from the AIL in this case, | |
3369 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3370 | */ |
3371 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
2451337d | 3372 | error = -EIO; |
32ce90a4 | 3373 | goto abort_out; |
1da177e4 LT |
3374 | } |
3375 | ||
a3f74ffb DC |
3376 | /* |
3377 | * Get the buffer containing the on-disk inode. | |
3378 | */ | |
475ee413 CH |
3379 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3380 | 0); | |
a3f74ffb DC |
3381 | if (error || !bp) { |
3382 | xfs_ifunlock(ip); | |
3383 | return error; | |
3384 | } | |
3385 | ||
1da177e4 LT |
3386 | /* |
3387 | * First flush out the inode that xfs_iflush was called with. | |
3388 | */ | |
3389 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3390 | if (error) |
1da177e4 | 3391 | goto corrupt_out; |
1da177e4 | 3392 | |
a3f74ffb DC |
3393 | /* |
3394 | * If the buffer is pinned then push on the log now so we won't | |
3395 | * get stuck waiting in the write for too long. | |
3396 | */ | |
811e64c7 | 3397 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3398 | xfs_log_force(mp, 0); |
a3f74ffb | 3399 | |
1da177e4 LT |
3400 | /* |
3401 | * inode clustering: | |
3402 | * see if other inodes can be gathered into this write | |
3403 | */ | |
bad55843 DC |
3404 | error = xfs_iflush_cluster(ip, bp); |
3405 | if (error) | |
3406 | goto cluster_corrupt_out; | |
1da177e4 | 3407 | |
4c46819a CH |
3408 | *bpp = bp; |
3409 | return 0; | |
1da177e4 LT |
3410 | |
3411 | corrupt_out: | |
3412 | xfs_buf_relse(bp); | |
7d04a335 | 3413 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3414 | cluster_corrupt_out: |
2451337d | 3415 | error = -EFSCORRUPTED; |
32ce90a4 | 3416 | abort_out: |
1da177e4 LT |
3417 | /* |
3418 | * Unlocks the flush lock | |
3419 | */ | |
04913fdd | 3420 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3421 | return error; |
1da177e4 LT |
3422 | } |
3423 | ||
1da177e4 LT |
3424 | STATIC int |
3425 | xfs_iflush_int( | |
93848a99 CH |
3426 | struct xfs_inode *ip, |
3427 | struct xfs_buf *bp) | |
1da177e4 | 3428 | { |
93848a99 CH |
3429 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3430 | struct xfs_dinode *dip; | |
3431 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3432 | |
579aa9ca | 3433 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3434 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3435 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3436 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3437 | ASSERT(iip != NULL && iip->ili_fields != 0); |
263997a6 | 3438 | ASSERT(ip->i_d.di_version > 1); |
1da177e4 | 3439 | |
1da177e4 | 3440 | /* set *dip = inode's place in the buffer */ |
88ee2df7 | 3441 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3442 | |
69ef921b | 3443 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 3444 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
3445 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3446 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3447 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
3448 | goto corrupt_out; |
3449 | } | |
abbede1b | 3450 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
3451 | if (XFS_TEST_ERROR( |
3452 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3453 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3454 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
3455 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3456 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
3457 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3458 | goto corrupt_out; |
3459 | } | |
abbede1b | 3460 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
3461 | if (XFS_TEST_ERROR( |
3462 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3463 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3464 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3465 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
3466 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3467 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
3468 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3469 | goto corrupt_out; |
3470 | } | |
3471 | } | |
3472 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3473 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3474 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
3475 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3476 | "%s: detected corrupt incore inode %Lu, " | |
3477 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3478 | __func__, ip->i_ino, | |
1da177e4 | 3479 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3480 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3481 | goto corrupt_out; |
3482 | } | |
3483 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3484 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
3485 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3486 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3487 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
3488 | goto corrupt_out; |
3489 | } | |
e60896d8 | 3490 | |
1da177e4 | 3491 | /* |
263997a6 | 3492 | * Inode item log recovery for v2 inodes are dependent on the |
e60896d8 DC |
3493 | * di_flushiter count for correct sequencing. We bump the flush |
3494 | * iteration count so we can detect flushes which postdate a log record | |
3495 | * during recovery. This is redundant as we now log every change and | |
3496 | * hence this can't happen but we need to still do it to ensure | |
3497 | * backwards compatibility with old kernels that predate logging all | |
3498 | * inode changes. | |
1da177e4 | 3499 | */ |
e60896d8 DC |
3500 | if (ip->i_d.di_version < 3) |
3501 | ip->i_d.di_flushiter++; | |
1da177e4 LT |
3502 | |
3503 | /* | |
3987848c DC |
3504 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3505 | * copy out the core of the inode, because if the inode is dirty at all | |
3506 | * the core must be. | |
1da177e4 | 3507 | */ |
93f958f9 | 3508 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3509 | |
3510 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3511 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3512 | ip->i_d.di_flushiter = 0; | |
3513 | ||
fd9fdba6 | 3514 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
e4ac967b | 3515 | if (XFS_IFORK_Q(ip)) |
fd9fdba6 | 3516 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); |
1da177e4 LT |
3517 | xfs_inobp_check(mp, bp); |
3518 | ||
3519 | /* | |
f5d8d5c4 CH |
3520 | * We've recorded everything logged in the inode, so we'd like to clear |
3521 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3522 | * However, we can't stop logging all this information until the data | |
3523 | * we've copied into the disk buffer is written to disk. If we did we | |
3524 | * might overwrite the copy of the inode in the log with all the data | |
3525 | * after re-logging only part of it, and in the face of a crash we | |
3526 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3527 | * |
f5d8d5c4 CH |
3528 | * What we do is move the bits to the ili_last_fields field. When |
3529 | * logging the inode, these bits are moved back to the ili_fields field. | |
3530 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3531 | * know that the information those bits represent is permanently on | |
3532 | * disk. As long as the flush completes before the inode is logged | |
3533 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3534 | * |
f5d8d5c4 CH |
3535 | * We can play with the ili_fields bits here, because the inode lock |
3536 | * must be held exclusively in order to set bits there and the flush | |
3537 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3538 | * done routine can tell whether or not to look in the AIL. Also, store | |
3539 | * the current LSN of the inode so that we can tell whether the item has | |
3540 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3541 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3542 | * atomically. | |
1da177e4 | 3543 | */ |
93848a99 CH |
3544 | iip->ili_last_fields = iip->ili_fields; |
3545 | iip->ili_fields = 0; | |
fc0561ce | 3546 | iip->ili_fsync_fields = 0; |
93848a99 | 3547 | iip->ili_logged = 1; |
1da177e4 | 3548 | |
93848a99 CH |
3549 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3550 | &iip->ili_item.li_lsn); | |
1da177e4 | 3551 | |
93848a99 CH |
3552 | /* |
3553 | * Attach the function xfs_iflush_done to the inode's | |
3554 | * buffer. This will remove the inode from the AIL | |
3555 | * and unlock the inode's flush lock when the inode is | |
3556 | * completely written to disk. | |
3557 | */ | |
3558 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3559 | |
93848a99 CH |
3560 | /* generate the checksum. */ |
3561 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3562 | |
93848a99 CH |
3563 | ASSERT(bp->b_fspriv != NULL); |
3564 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
3565 | return 0; |
3566 | ||
3567 | corrupt_out: | |
2451337d | 3568 | return -EFSCORRUPTED; |
1da177e4 | 3569 | } |