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" |
6ca1c906 | 22 | #include "xfs_format.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_trans_priv.h" | |
27 | #include "xfs_sb.h" | |
28 | #include "xfs_ag.h" | |
1da177e4 | 29 | #include "xfs_mount.h" |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
a844f451 | 33 | #include "xfs_attr_sf.h" |
1da177e4 | 34 | #include "xfs_dinode.h" |
1da177e4 | 35 | #include "xfs_inode.h" |
1da177e4 | 36 | #include "xfs_buf_item.h" |
a844f451 NS |
37 | #include "xfs_inode_item.h" |
38 | #include "xfs_btree.h" | |
39 | #include "xfs_alloc.h" | |
40 | #include "xfs_ialloc.h" | |
41 | #include "xfs_bmap.h" | |
68988114 | 42 | #include "xfs_bmap_util.h" |
1da177e4 | 43 | #include "xfs_error.h" |
1da177e4 | 44 | #include "xfs_utils.h" |
1da177e4 | 45 | #include "xfs_quota.h" |
2a82b8be | 46 | #include "xfs_filestream.h" |
739bfb2a | 47 | #include "xfs_vnodeops.h" |
93848a99 | 48 | #include "xfs_cksum.h" |
0b1b213f | 49 | #include "xfs_trace.h" |
33479e05 | 50 | #include "xfs_icache.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 | ||
60 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
1da177e4 | 61 | |
2a0ec1d9 DC |
62 | /* |
63 | * helper function to extract extent size hint from inode | |
64 | */ | |
65 | xfs_extlen_t | |
66 | xfs_get_extsz_hint( | |
67 | struct xfs_inode *ip) | |
68 | { | |
69 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
70 | return ip->i_d.di_extsize; | |
71 | if (XFS_IS_REALTIME_INODE(ip)) | |
72 | return ip->i_mount->m_sb.sb_rextsize; | |
73 | return 0; | |
74 | } | |
75 | ||
fa96acad DC |
76 | /* |
77 | * This is a wrapper routine around the xfs_ilock() routine used to centralize | |
78 | * some grungy code. It is used in places that wish to lock the inode solely | |
79 | * for reading the extents. The reason these places can't just call | |
80 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the | |
81 | * extents from disk for a file in b-tree format. If the inode is in b-tree | |
82 | * format, then we need to lock the inode exclusively until the extents are read | |
83 | * in. Locking it exclusively all the time would limit our parallelism | |
84 | * unnecessarily, though. What we do instead is check to see if the extents | |
85 | * have been read in yet, and only lock the inode exclusively if they have not. | |
86 | * | |
87 | * The function returns a value which should be given to the corresponding | |
88 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was | |
89 | * actually taken. | |
90 | */ | |
91 | uint | |
92 | xfs_ilock_map_shared( | |
93 | xfs_inode_t *ip) | |
94 | { | |
95 | uint lock_mode; | |
96 | ||
97 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
98 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
99 | lock_mode = XFS_ILOCK_EXCL; | |
100 | } else { | |
101 | lock_mode = XFS_ILOCK_SHARED; | |
102 | } | |
103 | ||
104 | xfs_ilock(ip, lock_mode); | |
105 | ||
106 | return lock_mode; | |
107 | } | |
108 | ||
109 | /* | |
110 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
111 | * All it does is call xfs_iunlock() with the given lock_mode. | |
112 | */ | |
113 | void | |
114 | xfs_iunlock_map_shared( | |
115 | xfs_inode_t *ip, | |
116 | unsigned int lock_mode) | |
117 | { | |
118 | xfs_iunlock(ip, lock_mode); | |
119 | } | |
120 | ||
121 | /* | |
122 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
123 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
124 | * allows either or both of the locks to be obtained. | |
125 | * | |
126 | * The 2 locks should always be ordered so that the IO lock is | |
127 | * obtained first in order to prevent deadlock. | |
128 | * | |
129 | * ip -- the inode being locked | |
130 | * lock_flags -- this parameter indicates the inode's locks | |
131 | * to be locked. It can be: | |
132 | * XFS_IOLOCK_SHARED, | |
133 | * XFS_IOLOCK_EXCL, | |
134 | * XFS_ILOCK_SHARED, | |
135 | * XFS_ILOCK_EXCL, | |
136 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
137 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
138 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
139 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
140 | */ | |
141 | void | |
142 | xfs_ilock( | |
143 | xfs_inode_t *ip, | |
144 | uint lock_flags) | |
145 | { | |
146 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
147 | ||
148 | /* | |
149 | * You can't set both SHARED and EXCL for the same lock, | |
150 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
151 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
152 | */ | |
153 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
154 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
155 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
156 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
157 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
158 | ||
159 | if (lock_flags & XFS_IOLOCK_EXCL) | |
160 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
161 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
162 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
163 | ||
164 | if (lock_flags & XFS_ILOCK_EXCL) | |
165 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
166 | else if (lock_flags & XFS_ILOCK_SHARED) | |
167 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
168 | } | |
169 | ||
170 | /* | |
171 | * This is just like xfs_ilock(), except that the caller | |
172 | * is guaranteed not to sleep. It returns 1 if it gets | |
173 | * the requested locks and 0 otherwise. If the IO lock is | |
174 | * obtained but the inode lock cannot be, then the IO lock | |
175 | * is dropped before returning. | |
176 | * | |
177 | * ip -- the inode being locked | |
178 | * lock_flags -- this parameter indicates the inode's locks to be | |
179 | * to be locked. See the comment for xfs_ilock() for a list | |
180 | * of valid values. | |
181 | */ | |
182 | int | |
183 | xfs_ilock_nowait( | |
184 | xfs_inode_t *ip, | |
185 | uint lock_flags) | |
186 | { | |
187 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
188 | ||
189 | /* | |
190 | * You can't set both SHARED and EXCL for the same lock, | |
191 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
192 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
193 | */ | |
194 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
195 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
196 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
197 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
198 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
199 | ||
200 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
201 | if (!mrtryupdate(&ip->i_iolock)) | |
202 | goto out; | |
203 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
204 | if (!mrtryaccess(&ip->i_iolock)) | |
205 | goto out; | |
206 | } | |
207 | if (lock_flags & XFS_ILOCK_EXCL) { | |
208 | if (!mrtryupdate(&ip->i_lock)) | |
209 | goto out_undo_iolock; | |
210 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
211 | if (!mrtryaccess(&ip->i_lock)) | |
212 | goto out_undo_iolock; | |
213 | } | |
214 | return 1; | |
215 | ||
216 | out_undo_iolock: | |
217 | if (lock_flags & XFS_IOLOCK_EXCL) | |
218 | mrunlock_excl(&ip->i_iolock); | |
219 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
220 | mrunlock_shared(&ip->i_iolock); | |
221 | out: | |
222 | return 0; | |
223 | } | |
224 | ||
225 | /* | |
226 | * xfs_iunlock() is used to drop the inode locks acquired with | |
227 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
228 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
229 | * that we know which locks to drop. | |
230 | * | |
231 | * ip -- the inode being unlocked | |
232 | * lock_flags -- this parameter indicates the inode's locks to be | |
233 | * to be unlocked. See the comment for xfs_ilock() for a list | |
234 | * of valid values for this parameter. | |
235 | * | |
236 | */ | |
237 | void | |
238 | xfs_iunlock( | |
239 | xfs_inode_t *ip, | |
240 | uint lock_flags) | |
241 | { | |
242 | /* | |
243 | * You can't set both SHARED and EXCL for the same lock, | |
244 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
245 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
246 | */ | |
247 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
248 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
249 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
250 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
251 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
252 | ASSERT(lock_flags != 0); | |
253 | ||
254 | if (lock_flags & XFS_IOLOCK_EXCL) | |
255 | mrunlock_excl(&ip->i_iolock); | |
256 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
257 | mrunlock_shared(&ip->i_iolock); | |
258 | ||
259 | if (lock_flags & XFS_ILOCK_EXCL) | |
260 | mrunlock_excl(&ip->i_lock); | |
261 | else if (lock_flags & XFS_ILOCK_SHARED) | |
262 | mrunlock_shared(&ip->i_lock); | |
263 | ||
264 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
265 | } | |
266 | ||
267 | /* | |
268 | * give up write locks. the i/o lock cannot be held nested | |
269 | * if it is being demoted. | |
270 | */ | |
271 | void | |
272 | xfs_ilock_demote( | |
273 | xfs_inode_t *ip, | |
274 | uint lock_flags) | |
275 | { | |
276 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
277 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
278 | ||
279 | if (lock_flags & XFS_ILOCK_EXCL) | |
280 | mrdemote(&ip->i_lock); | |
281 | if (lock_flags & XFS_IOLOCK_EXCL) | |
282 | mrdemote(&ip->i_iolock); | |
283 | ||
284 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
285 | } | |
286 | ||
742ae1e3 | 287 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
288 | int |
289 | xfs_isilocked( | |
290 | xfs_inode_t *ip, | |
291 | uint lock_flags) | |
292 | { | |
293 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
294 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
295 | return !!ip->i_lock.mr_writer; | |
296 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
297 | } | |
298 | ||
299 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { | |
300 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
301 | return !!ip->i_iolock.mr_writer; | |
302 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
303 | } | |
304 | ||
305 | ASSERT(0); | |
306 | return 0; | |
307 | } | |
308 | #endif | |
309 | ||
310 | void | |
311 | __xfs_iflock( | |
312 | struct xfs_inode *ip) | |
313 | { | |
314 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
315 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
316 | ||
317 | do { | |
318 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
319 | if (xfs_isiflocked(ip)) | |
320 | io_schedule(); | |
321 | } while (!xfs_iflock_nowait(ip)); | |
322 | ||
323 | finish_wait(wq, &wait.wait); | |
324 | } | |
325 | ||
1da177e4 LT |
326 | STATIC uint |
327 | _xfs_dic2xflags( | |
1da177e4 LT |
328 | __uint16_t di_flags) |
329 | { | |
330 | uint flags = 0; | |
331 | ||
332 | if (di_flags & XFS_DIFLAG_ANY) { | |
333 | if (di_flags & XFS_DIFLAG_REALTIME) | |
334 | flags |= XFS_XFLAG_REALTIME; | |
335 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
336 | flags |= XFS_XFLAG_PREALLOC; | |
337 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
338 | flags |= XFS_XFLAG_IMMUTABLE; | |
339 | if (di_flags & XFS_DIFLAG_APPEND) | |
340 | flags |= XFS_XFLAG_APPEND; | |
341 | if (di_flags & XFS_DIFLAG_SYNC) | |
342 | flags |= XFS_XFLAG_SYNC; | |
343 | if (di_flags & XFS_DIFLAG_NOATIME) | |
344 | flags |= XFS_XFLAG_NOATIME; | |
345 | if (di_flags & XFS_DIFLAG_NODUMP) | |
346 | flags |= XFS_XFLAG_NODUMP; | |
347 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
348 | flags |= XFS_XFLAG_RTINHERIT; | |
349 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
350 | flags |= XFS_XFLAG_PROJINHERIT; | |
351 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
352 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
353 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
354 | flags |= XFS_XFLAG_EXTSIZE; | |
355 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
356 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
357 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
358 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
359 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
360 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
361 | } |
362 | ||
363 | return flags; | |
364 | } | |
365 | ||
366 | uint | |
367 | xfs_ip2xflags( | |
368 | xfs_inode_t *ip) | |
369 | { | |
347d1c01 | 370 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 371 | |
a916e2bd | 372 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 373 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
374 | } |
375 | ||
376 | uint | |
377 | xfs_dic2xflags( | |
45ba598e | 378 | xfs_dinode_t *dip) |
1da177e4 | 379 | { |
81591fe2 | 380 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
45ba598e | 381 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
382 | } |
383 | ||
1da177e4 LT |
384 | /* |
385 | * Allocate an inode on disk and return a copy of its in-core version. | |
386 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
387 | * appropriately within the inode. The uid and gid for the inode are | |
388 | * set according to the contents of the given cred structure. | |
389 | * | |
390 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
391 | * has a free inode available, call xfs_iget() to obtain the in-core |
392 | * version of the allocated inode. Finally, fill in the inode and | |
393 | * log its initial contents. In this case, ialloc_context would be | |
394 | * set to NULL. | |
1da177e4 | 395 | * |
cd856db6 CM |
396 | * If xfs_dialloc() does not have an available inode, it will replenish |
397 | * its supply by doing an allocation. Since we can only do one | |
398 | * allocation within a transaction without deadlocks, we must commit | |
399 | * the current transaction before returning the inode itself. | |
400 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
401 | * The caller should then commit the current transaction, start a new |
402 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
403 | * | |
404 | * To ensure that some other process does not grab the inode that | |
405 | * was allocated during the first call to xfs_ialloc(), this routine | |
406 | * also returns the [locked] bp pointing to the head of the freelist | |
407 | * as ialloc_context. The caller should hold this buffer across | |
408 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
409 | * |
410 | * If we are allocating quota inodes, we do not have a parent inode | |
411 | * to attach to or associate with (i.e. pip == NULL) because they | |
412 | * are not linked into the directory structure - they are attached | |
413 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
414 | */ |
415 | int | |
416 | xfs_ialloc( | |
417 | xfs_trans_t *tp, | |
418 | xfs_inode_t *pip, | |
576b1d67 | 419 | umode_t mode, |
31b084ae | 420 | xfs_nlink_t nlink, |
1da177e4 | 421 | xfs_dev_t rdev, |
6743099c | 422 | prid_t prid, |
1da177e4 LT |
423 | int okalloc, |
424 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
425 | xfs_inode_t **ipp) |
426 | { | |
93848a99 | 427 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
428 | xfs_ino_t ino; |
429 | xfs_inode_t *ip; | |
1da177e4 LT |
430 | uint flags; |
431 | int error; | |
dff35fd4 | 432 | timespec_t tv; |
bf904248 | 433 | int filestreams = 0; |
1da177e4 LT |
434 | |
435 | /* | |
436 | * Call the space management code to pick | |
437 | * the on-disk inode to be allocated. | |
438 | */ | |
b11f94d5 | 439 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 440 | ialloc_context, &ino); |
bf904248 | 441 | if (error) |
1da177e4 | 442 | return error; |
08358906 | 443 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
444 | *ipp = NULL; |
445 | return 0; | |
446 | } | |
447 | ASSERT(*ialloc_context == NULL); | |
448 | ||
449 | /* | |
450 | * Get the in-core inode with the lock held exclusively. | |
451 | * This is because we're setting fields here we need | |
452 | * to prevent others from looking at until we're done. | |
453 | */ | |
93848a99 | 454 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 455 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 456 | if (error) |
1da177e4 | 457 | return error; |
1da177e4 LT |
458 | ASSERT(ip != NULL); |
459 | ||
576b1d67 | 460 | ip->i_d.di_mode = mode; |
1da177e4 LT |
461 | ip->i_d.di_onlink = 0; |
462 | ip->i_d.di_nlink = nlink; | |
463 | ASSERT(ip->i_d.di_nlink == nlink); | |
9e2b2dc4 DH |
464 | ip->i_d.di_uid = current_fsuid(); |
465 | ip->i_d.di_gid = current_fsgid(); | |
6743099c | 466 | xfs_set_projid(ip, prid); |
1da177e4 LT |
467 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
468 | ||
469 | /* | |
470 | * If the superblock version is up to where we support new format | |
471 | * inodes and this is currently an old format inode, then change | |
472 | * the inode version number now. This way we only do the conversion | |
473 | * here rather than here and in the flush/logging code. | |
474 | */ | |
93848a99 | 475 | if (xfs_sb_version_hasnlink(&mp->m_sb) && |
51ce16d5 CH |
476 | ip->i_d.di_version == 1) { |
477 | ip->i_d.di_version = 2; | |
1da177e4 LT |
478 | /* |
479 | * We've already zeroed the old link count, the projid field, | |
480 | * and the pad field. | |
481 | */ | |
482 | } | |
483 | ||
484 | /* | |
485 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
486 | */ | |
51ce16d5 | 487 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
1da177e4 LT |
488 | xfs_bump_ino_vers2(tp, ip); |
489 | ||
bd186aa9 | 490 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 491 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 492 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
493 | ip->i_d.di_mode |= S_ISGID; |
494 | } | |
495 | } | |
496 | ||
497 | /* | |
498 | * If the group ID of the new file does not match the effective group | |
499 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
500 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
501 | */ | |
502 | if ((irix_sgid_inherit) && | |
503 | (ip->i_d.di_mode & S_ISGID) && | |
504 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
505 | ip->i_d.di_mode &= ~S_ISGID; | |
506 | } | |
507 | ||
508 | ip->i_d.di_size = 0; | |
509 | ip->i_d.di_nextents = 0; | |
510 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
511 | |
512 | nanotime(&tv); | |
513 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
514 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
515 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
516 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
517 | ||
1da177e4 LT |
518 | /* |
519 | * di_gen will have been taken care of in xfs_iread. | |
520 | */ | |
521 | ip->i_d.di_extsize = 0; | |
522 | ip->i_d.di_dmevmask = 0; | |
523 | ip->i_d.di_dmstate = 0; | |
524 | ip->i_d.di_flags = 0; | |
93848a99 CH |
525 | |
526 | if (ip->i_d.di_version == 3) { | |
527 | ASSERT(ip->i_d.di_ino == ino); | |
528 | ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); | |
529 | ip->i_d.di_crc = 0; | |
530 | ip->i_d.di_changecount = 1; | |
531 | ip->i_d.di_lsn = 0; | |
532 | ip->i_d.di_flags2 = 0; | |
533 | memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); | |
534 | ip->i_d.di_crtime = ip->i_d.di_mtime; | |
535 | } | |
536 | ||
537 | ||
1da177e4 LT |
538 | flags = XFS_ILOG_CORE; |
539 | switch (mode & S_IFMT) { | |
540 | case S_IFIFO: | |
541 | case S_IFCHR: | |
542 | case S_IFBLK: | |
543 | case S_IFSOCK: | |
544 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
545 | ip->i_df.if_u2.if_rdev = rdev; | |
546 | ip->i_df.if_flags = 0; | |
547 | flags |= XFS_ILOG_DEV; | |
548 | break; | |
549 | case S_IFREG: | |
bf904248 DC |
550 | /* |
551 | * we can't set up filestreams until after the VFS inode | |
552 | * is set up properly. | |
553 | */ | |
554 | if (pip && xfs_inode_is_filestream(pip)) | |
555 | filestreams = 1; | |
2a82b8be | 556 | /* fall through */ |
1da177e4 | 557 | case S_IFDIR: |
b11f94d5 | 558 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
559 | uint di_flags = 0; |
560 | ||
abbede1b | 561 | if (S_ISDIR(mode)) { |
365ca83d NS |
562 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
563 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
564 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
565 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
566 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
567 | } | |
abbede1b | 568 | } else if (S_ISREG(mode)) { |
613d7043 | 569 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 570 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
571 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
572 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
573 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
574 | } | |
1da177e4 LT |
575 | } |
576 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
577 | xfs_inherit_noatime) | |
365ca83d | 578 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
579 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
580 | xfs_inherit_nodump) | |
365ca83d | 581 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
582 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
583 | xfs_inherit_sync) | |
365ca83d | 584 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
585 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
586 | xfs_inherit_nosymlinks) | |
365ca83d NS |
587 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
588 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
589 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
590 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
591 | xfs_inherit_nodefrag) | |
592 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
593 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
594 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 595 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
596 | } |
597 | /* FALLTHROUGH */ | |
598 | case S_IFLNK: | |
599 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
600 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
601 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
602 | ip->i_df.if_u1.if_extents = NULL; | |
603 | break; | |
604 | default: | |
605 | ASSERT(0); | |
606 | } | |
607 | /* | |
608 | * Attribute fork settings for new inode. | |
609 | */ | |
610 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
611 | ip->i_d.di_anextents = 0; | |
612 | ||
613 | /* | |
614 | * Log the new values stuffed into the inode. | |
615 | */ | |
ddc3415a | 616 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
617 | xfs_trans_log_inode(tp, ip, flags); |
618 | ||
b83bd138 | 619 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 620 | xfs_setup_inode(ip); |
1da177e4 | 621 | |
bf904248 DC |
622 | /* now we have set up the vfs inode we can associate the filestream */ |
623 | if (filestreams) { | |
624 | error = xfs_filestream_associate(pip, ip); | |
625 | if (error < 0) | |
626 | return -error; | |
627 | if (!error) | |
628 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
629 | } | |
630 | ||
1da177e4 LT |
631 | *ipp = ip; |
632 | return 0; | |
633 | } | |
634 | ||
1da177e4 | 635 | /* |
8f04c47a CH |
636 | * Free up the underlying blocks past new_size. The new size must be smaller |
637 | * than the current size. This routine can be used both for the attribute and | |
638 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 639 | * |
f6485057 DC |
640 | * The transaction passed to this routine must have made a permanent log |
641 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
642 | * given transaction and start new ones, so make sure everything involved in | |
643 | * the transaction is tidy before calling here. Some transaction will be | |
644 | * returned to the caller to be committed. The incoming transaction must | |
645 | * already include the inode, and both inode locks must be held exclusively. | |
646 | * The inode must also be "held" within the transaction. On return the inode | |
647 | * will be "held" within the returned transaction. This routine does NOT | |
648 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 649 | * |
f6485057 DC |
650 | * If we get an error, we must return with the inode locked and linked into the |
651 | * current transaction. This keeps things simple for the higher level code, | |
652 | * because it always knows that the inode is locked and held in the transaction | |
653 | * that returns to it whether errors occur or not. We don't mark the inode | |
654 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
655 | */ |
656 | int | |
8f04c47a CH |
657 | xfs_itruncate_extents( |
658 | struct xfs_trans **tpp, | |
659 | struct xfs_inode *ip, | |
660 | int whichfork, | |
661 | xfs_fsize_t new_size) | |
1da177e4 | 662 | { |
8f04c47a CH |
663 | struct xfs_mount *mp = ip->i_mount; |
664 | struct xfs_trans *tp = *tpp; | |
665 | struct xfs_trans *ntp; | |
666 | xfs_bmap_free_t free_list; | |
667 | xfs_fsblock_t first_block; | |
668 | xfs_fileoff_t first_unmap_block; | |
669 | xfs_fileoff_t last_block; | |
670 | xfs_filblks_t unmap_len; | |
671 | int committed; | |
672 | int error = 0; | |
673 | int done = 0; | |
1da177e4 | 674 | |
0b56185b CH |
675 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
676 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
677 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 678 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 679 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 680 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 681 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 682 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 683 | |
673e8e59 CH |
684 | trace_xfs_itruncate_extents_start(ip, new_size); |
685 | ||
1da177e4 LT |
686 | /* |
687 | * Since it is possible for space to become allocated beyond | |
688 | * the end of the file (in a crash where the space is allocated | |
689 | * but the inode size is not yet updated), simply remove any | |
690 | * blocks which show up between the new EOF and the maximum | |
691 | * possible file size. If the first block to be removed is | |
692 | * beyond the maximum file size (ie it is the same as last_block), | |
693 | * then there is nothing to do. | |
694 | */ | |
8f04c47a | 695 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 696 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
697 | if (first_unmap_block == last_block) |
698 | return 0; | |
699 | ||
700 | ASSERT(first_unmap_block < last_block); | |
701 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 702 | while (!done) { |
9d87c319 | 703 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 704 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 705 | first_unmap_block, unmap_len, |
8f04c47a | 706 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 707 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 708 | &first_block, &free_list, |
b4e9181e | 709 | &done); |
8f04c47a CH |
710 | if (error) |
711 | goto out_bmap_cancel; | |
1da177e4 LT |
712 | |
713 | /* | |
714 | * Duplicate the transaction that has the permanent | |
715 | * reservation and commit the old transaction. | |
716 | */ | |
8f04c47a | 717 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
898621d5 | 718 | if (committed) |
ddc3415a | 719 | xfs_trans_ijoin(tp, ip, 0); |
8f04c47a CH |
720 | if (error) |
721 | goto out_bmap_cancel; | |
1da177e4 LT |
722 | |
723 | if (committed) { | |
724 | /* | |
f6485057 | 725 | * Mark the inode dirty so it will be logged and |
e5720eec | 726 | * moved forward in the log as part of every commit. |
1da177e4 | 727 | */ |
8f04c47a | 728 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 729 | } |
f6485057 | 730 | |
8f04c47a CH |
731 | ntp = xfs_trans_dup(tp); |
732 | error = xfs_trans_commit(tp, 0); | |
733 | tp = ntp; | |
e5720eec | 734 | |
ddc3415a | 735 | xfs_trans_ijoin(tp, ip, 0); |
f6485057 | 736 | |
cc09c0dc | 737 | if (error) |
8f04c47a CH |
738 | goto out; |
739 | ||
cc09c0dc | 740 | /* |
8f04c47a | 741 | * Transaction commit worked ok so we can drop the extra ticket |
cc09c0dc DC |
742 | * reference that we gained in xfs_trans_dup() |
743 | */ | |
8f04c47a CH |
744 | xfs_log_ticket_put(tp->t_ticket); |
745 | error = xfs_trans_reserve(tp, 0, | |
f6485057 DC |
746 | XFS_ITRUNCATE_LOG_RES(mp), 0, |
747 | XFS_TRANS_PERM_LOG_RES, | |
748 | XFS_ITRUNCATE_LOG_COUNT); | |
749 | if (error) | |
8f04c47a | 750 | goto out; |
1da177e4 | 751 | } |
8f04c47a | 752 | |
673e8e59 CH |
753 | /* |
754 | * Always re-log the inode so that our permanent transaction can keep | |
755 | * on rolling it forward in the log. | |
756 | */ | |
757 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
758 | ||
759 | trace_xfs_itruncate_extents_end(ip, new_size); | |
760 | ||
8f04c47a CH |
761 | out: |
762 | *tpp = tp; | |
763 | return error; | |
764 | out_bmap_cancel: | |
1da177e4 | 765 | /* |
8f04c47a CH |
766 | * If the bunmapi call encounters an error, return to the caller where |
767 | * the transaction can be properly aborted. We just need to make sure | |
768 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 769 | */ |
8f04c47a CH |
770 | xfs_bmap_cancel(&free_list); |
771 | goto out; | |
772 | } | |
773 | ||
1da177e4 LT |
774 | /* |
775 | * This is called when the inode's link count goes to 0. | |
776 | * We place the on-disk inode on a list in the AGI. It | |
777 | * will be pulled from this list when the inode is freed. | |
778 | */ | |
779 | int | |
780 | xfs_iunlink( | |
781 | xfs_trans_t *tp, | |
782 | xfs_inode_t *ip) | |
783 | { | |
784 | xfs_mount_t *mp; | |
785 | xfs_agi_t *agi; | |
786 | xfs_dinode_t *dip; | |
787 | xfs_buf_t *agibp; | |
788 | xfs_buf_t *ibp; | |
1da177e4 LT |
789 | xfs_agino_t agino; |
790 | short bucket_index; | |
791 | int offset; | |
792 | int error; | |
1da177e4 LT |
793 | |
794 | ASSERT(ip->i_d.di_nlink == 0); | |
795 | ASSERT(ip->i_d.di_mode != 0); | |
1da177e4 LT |
796 | |
797 | mp = tp->t_mountp; | |
798 | ||
1da177e4 LT |
799 | /* |
800 | * Get the agi buffer first. It ensures lock ordering | |
801 | * on the list. | |
802 | */ | |
5e1be0fb | 803 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 804 | if (error) |
1da177e4 | 805 | return error; |
1da177e4 | 806 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 807 | |
1da177e4 LT |
808 | /* |
809 | * Get the index into the agi hash table for the | |
810 | * list this inode will go on. | |
811 | */ | |
812 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
813 | ASSERT(agino != 0); | |
814 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
815 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 816 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 817 | |
69ef921b | 818 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
819 | /* |
820 | * There is already another inode in the bucket we need | |
821 | * to add ourselves to. Add us at the front of the list. | |
822 | * Here we put the head pointer into our next pointer, | |
823 | * and then we fall through to point the head at us. | |
824 | */ | |
475ee413 CH |
825 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
826 | 0, 0); | |
c319b58b VA |
827 | if (error) |
828 | return error; | |
829 | ||
69ef921b | 830 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 831 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 832 | offset = ip->i_imap.im_boffset + |
1da177e4 | 833 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
834 | |
835 | /* need to recalc the inode CRC if appropriate */ | |
836 | xfs_dinode_calc_crc(mp, dip); | |
837 | ||
1da177e4 LT |
838 | xfs_trans_inode_buf(tp, ibp); |
839 | xfs_trans_log_buf(tp, ibp, offset, | |
840 | (offset + sizeof(xfs_agino_t) - 1)); | |
841 | xfs_inobp_check(mp, ibp); | |
842 | } | |
843 | ||
844 | /* | |
845 | * Point the bucket head pointer at the inode being inserted. | |
846 | */ | |
847 | ASSERT(agino != 0); | |
16259e7d | 848 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
849 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
850 | (sizeof(xfs_agino_t) * bucket_index); | |
851 | xfs_trans_log_buf(tp, agibp, offset, | |
852 | (offset + sizeof(xfs_agino_t) - 1)); | |
853 | return 0; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Pull the on-disk inode from the AGI unlinked list. | |
858 | */ | |
859 | STATIC int | |
860 | xfs_iunlink_remove( | |
861 | xfs_trans_t *tp, | |
862 | xfs_inode_t *ip) | |
863 | { | |
864 | xfs_ino_t next_ino; | |
865 | xfs_mount_t *mp; | |
866 | xfs_agi_t *agi; | |
867 | xfs_dinode_t *dip; | |
868 | xfs_buf_t *agibp; | |
869 | xfs_buf_t *ibp; | |
870 | xfs_agnumber_t agno; | |
1da177e4 LT |
871 | xfs_agino_t agino; |
872 | xfs_agino_t next_agino; | |
873 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 874 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 875 | short bucket_index; |
6fdf8ccc | 876 | int offset, last_offset = 0; |
1da177e4 | 877 | int error; |
1da177e4 | 878 | |
1da177e4 | 879 | mp = tp->t_mountp; |
1da177e4 | 880 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
881 | |
882 | /* | |
883 | * Get the agi buffer first. It ensures lock ordering | |
884 | * on the list. | |
885 | */ | |
5e1be0fb CH |
886 | error = xfs_read_agi(mp, tp, agno, &agibp); |
887 | if (error) | |
1da177e4 | 888 | return error; |
5e1be0fb | 889 | |
1da177e4 | 890 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 891 | |
1da177e4 LT |
892 | /* |
893 | * Get the index into the agi hash table for the | |
894 | * list this inode will go on. | |
895 | */ | |
896 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
897 | ASSERT(agino != 0); | |
898 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 899 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
900 | ASSERT(agi->agi_unlinked[bucket_index]); |
901 | ||
16259e7d | 902 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 903 | /* |
475ee413 CH |
904 | * We're at the head of the list. Get the inode's on-disk |
905 | * buffer to see if there is anyone after us on the list. | |
906 | * Only modify our next pointer if it is not already NULLAGINO. | |
907 | * This saves us the overhead of dealing with the buffer when | |
908 | * there is no need to change it. | |
1da177e4 | 909 | */ |
475ee413 CH |
910 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
911 | 0, 0); | |
1da177e4 | 912 | if (error) { |
475ee413 | 913 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 914 | __func__, error); |
1da177e4 LT |
915 | return error; |
916 | } | |
347d1c01 | 917 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
918 | ASSERT(next_agino != 0); |
919 | if (next_agino != NULLAGINO) { | |
347d1c01 | 920 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 921 | offset = ip->i_imap.im_boffset + |
1da177e4 | 922 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
923 | |
924 | /* need to recalc the inode CRC if appropriate */ | |
925 | xfs_dinode_calc_crc(mp, dip); | |
926 | ||
1da177e4 LT |
927 | xfs_trans_inode_buf(tp, ibp); |
928 | xfs_trans_log_buf(tp, ibp, offset, | |
929 | (offset + sizeof(xfs_agino_t) - 1)); | |
930 | xfs_inobp_check(mp, ibp); | |
931 | } else { | |
932 | xfs_trans_brelse(tp, ibp); | |
933 | } | |
934 | /* | |
935 | * Point the bucket head pointer at the next inode. | |
936 | */ | |
937 | ASSERT(next_agino != 0); | |
938 | ASSERT(next_agino != agino); | |
16259e7d | 939 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
940 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
941 | (sizeof(xfs_agino_t) * bucket_index); | |
942 | xfs_trans_log_buf(tp, agibp, offset, | |
943 | (offset + sizeof(xfs_agino_t) - 1)); | |
944 | } else { | |
945 | /* | |
946 | * We need to search the list for the inode being freed. | |
947 | */ | |
16259e7d | 948 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
949 | last_ibp = NULL; |
950 | while (next_agino != agino) { | |
129dbc9a CH |
951 | struct xfs_imap imap; |
952 | ||
953 | if (last_ibp) | |
1da177e4 | 954 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
955 | |
956 | imap.im_blkno = 0; | |
1da177e4 | 957 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
958 | |
959 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
960 | if (error) { | |
961 | xfs_warn(mp, | |
962 | "%s: xfs_imap returned error %d.", | |
963 | __func__, error); | |
964 | return error; | |
965 | } | |
966 | ||
967 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
968 | &last_ibp, 0, 0); | |
1da177e4 | 969 | if (error) { |
0b932ccc | 970 | xfs_warn(mp, |
129dbc9a | 971 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 972 | __func__, error); |
1da177e4 LT |
973 | return error; |
974 | } | |
129dbc9a CH |
975 | |
976 | last_offset = imap.im_boffset; | |
347d1c01 | 977 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
978 | ASSERT(next_agino != NULLAGINO); |
979 | ASSERT(next_agino != 0); | |
980 | } | |
475ee413 | 981 | |
1da177e4 | 982 | /* |
475ee413 CH |
983 | * Now last_ibp points to the buffer previous to us on the |
984 | * unlinked list. Pull us from the list. | |
1da177e4 | 985 | */ |
475ee413 CH |
986 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
987 | 0, 0); | |
1da177e4 | 988 | if (error) { |
475ee413 | 989 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 990 | __func__, error); |
1da177e4 LT |
991 | return error; |
992 | } | |
347d1c01 | 993 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
994 | ASSERT(next_agino != 0); |
995 | ASSERT(next_agino != agino); | |
996 | if (next_agino != NULLAGINO) { | |
347d1c01 | 997 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 998 | offset = ip->i_imap.im_boffset + |
1da177e4 | 999 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1000 | |
1001 | /* need to recalc the inode CRC if appropriate */ | |
1002 | xfs_dinode_calc_crc(mp, dip); | |
1003 | ||
1da177e4 LT |
1004 | xfs_trans_inode_buf(tp, ibp); |
1005 | xfs_trans_log_buf(tp, ibp, offset, | |
1006 | (offset + sizeof(xfs_agino_t) - 1)); | |
1007 | xfs_inobp_check(mp, ibp); | |
1008 | } else { | |
1009 | xfs_trans_brelse(tp, ibp); | |
1010 | } | |
1011 | /* | |
1012 | * Point the previous inode on the list to the next inode. | |
1013 | */ | |
347d1c01 | 1014 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
1015 | ASSERT(next_agino != 0); |
1016 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
1017 | |
1018 | /* need to recalc the inode CRC if appropriate */ | |
1019 | xfs_dinode_calc_crc(mp, last_dip); | |
1020 | ||
1da177e4 LT |
1021 | xfs_trans_inode_buf(tp, last_ibp); |
1022 | xfs_trans_log_buf(tp, last_ibp, offset, | |
1023 | (offset + sizeof(xfs_agino_t) - 1)); | |
1024 | xfs_inobp_check(mp, last_ibp); | |
1025 | } | |
1026 | return 0; | |
1027 | } | |
1028 | ||
5b3eed75 DC |
1029 | /* |
1030 | * A big issue when freeing the inode cluster is is that we _cannot_ skip any | |
1031 | * inodes that are in memory - they all must be marked stale and attached to | |
1032 | * the cluster buffer. | |
1033 | */ | |
2a30f36d | 1034 | STATIC int |
1da177e4 LT |
1035 | xfs_ifree_cluster( |
1036 | xfs_inode_t *free_ip, | |
1037 | xfs_trans_t *tp, | |
1038 | xfs_ino_t inum) | |
1039 | { | |
1040 | xfs_mount_t *mp = free_ip->i_mount; | |
1041 | int blks_per_cluster; | |
1042 | int nbufs; | |
1043 | int ninodes; | |
5b257b4a | 1044 | int i, j; |
1da177e4 LT |
1045 | xfs_daddr_t blkno; |
1046 | xfs_buf_t *bp; | |
5b257b4a | 1047 | xfs_inode_t *ip; |
1da177e4 LT |
1048 | xfs_inode_log_item_t *iip; |
1049 | xfs_log_item_t *lip; | |
5017e97d | 1050 | struct xfs_perag *pag; |
1da177e4 | 1051 | |
5017e97d | 1052 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
1da177e4 LT |
1053 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
1054 | blks_per_cluster = 1; | |
1055 | ninodes = mp->m_sb.sb_inopblock; | |
1056 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
1057 | } else { | |
1058 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
1059 | mp->m_sb.sb_blocksize; | |
1060 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
1061 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
1062 | } | |
1063 | ||
1da177e4 LT |
1064 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
1065 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
1066 | XFS_INO_TO_AGBNO(mp, inum)); | |
1067 | ||
5b257b4a DC |
1068 | /* |
1069 | * We obtain and lock the backing buffer first in the process | |
1070 | * here, as we have to ensure that any dirty inode that we | |
1071 | * can't get the flush lock on is attached to the buffer. | |
1072 | * If we scan the in-memory inodes first, then buffer IO can | |
1073 | * complete before we get a lock on it, and hence we may fail | |
1074 | * to mark all the active inodes on the buffer stale. | |
1075 | */ | |
1076 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
1077 | mp->m_bsize * blks_per_cluster, |
1078 | XBF_UNMAPPED); | |
5b257b4a | 1079 | |
2a30f36d CS |
1080 | if (!bp) |
1081 | return ENOMEM; | |
b0f539de DC |
1082 | |
1083 | /* | |
1084 | * This buffer may not have been correctly initialised as we | |
1085 | * didn't read it from disk. That's not important because we are | |
1086 | * only using to mark the buffer as stale in the log, and to | |
1087 | * attach stale cached inodes on it. That means it will never be | |
1088 | * dispatched for IO. If it is, we want to know about it, and we | |
1089 | * want it to fail. We can acheive this by adding a write | |
1090 | * verifier to the buffer. | |
1091 | */ | |
1813dd64 | 1092 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 1093 | |
5b257b4a DC |
1094 | /* |
1095 | * Walk the inodes already attached to the buffer and mark them | |
1096 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
1097 | * in-memory inode walk can't lock them. By marking them all |
1098 | * stale first, we will not attempt to lock them in the loop | |
1099 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 1100 | */ |
adadbeef | 1101 | lip = bp->b_fspriv; |
5b257b4a DC |
1102 | while (lip) { |
1103 | if (lip->li_type == XFS_LI_INODE) { | |
1104 | iip = (xfs_inode_log_item_t *)lip; | |
1105 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 1106 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
1107 | xfs_trans_ail_copy_lsn(mp->m_ail, |
1108 | &iip->ili_flush_lsn, | |
1109 | &iip->ili_item.li_lsn); | |
1110 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
1111 | } |
1112 | lip = lip->li_bio_list; | |
1113 | } | |
1da177e4 | 1114 | |
5b3eed75 | 1115 | |
1da177e4 | 1116 | /* |
5b257b4a DC |
1117 | * For each inode in memory attempt to add it to the inode |
1118 | * buffer and set it up for being staled on buffer IO | |
1119 | * completion. This is safe as we've locked out tail pushing | |
1120 | * and flushing by locking the buffer. | |
1da177e4 | 1121 | * |
5b257b4a DC |
1122 | * We have already marked every inode that was part of a |
1123 | * transaction stale above, which means there is no point in | |
1124 | * even trying to lock them. | |
1da177e4 | 1125 | */ |
1da177e4 | 1126 | for (i = 0; i < ninodes; i++) { |
5b3eed75 | 1127 | retry: |
1a3e8f3d | 1128 | rcu_read_lock(); |
da353b0d DC |
1129 | ip = radix_tree_lookup(&pag->pag_ici_root, |
1130 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 1131 | |
1a3e8f3d DC |
1132 | /* Inode not in memory, nothing to do */ |
1133 | if (!ip) { | |
1134 | rcu_read_unlock(); | |
1da177e4 LT |
1135 | continue; |
1136 | } | |
1137 | ||
1a3e8f3d DC |
1138 | /* |
1139 | * because this is an RCU protected lookup, we could | |
1140 | * find a recently freed or even reallocated inode | |
1141 | * during the lookup. We need to check under the | |
1142 | * i_flags_lock for a valid inode here. Skip it if it | |
1143 | * is not valid, the wrong inode or stale. | |
1144 | */ | |
1145 | spin_lock(&ip->i_flags_lock); | |
1146 | if (ip->i_ino != inum + i || | |
1147 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
1148 | spin_unlock(&ip->i_flags_lock); | |
1149 | rcu_read_unlock(); | |
1150 | continue; | |
1151 | } | |
1152 | spin_unlock(&ip->i_flags_lock); | |
1153 | ||
5b3eed75 DC |
1154 | /* |
1155 | * Don't try to lock/unlock the current inode, but we | |
1156 | * _cannot_ skip the other inodes that we did not find | |
1157 | * in the list attached to the buffer and are not | |
1158 | * already marked stale. If we can't lock it, back off | |
1159 | * and retry. | |
1160 | */ | |
5b257b4a DC |
1161 | if (ip != free_ip && |
1162 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 1163 | rcu_read_unlock(); |
5b3eed75 DC |
1164 | delay(1); |
1165 | goto retry; | |
1da177e4 | 1166 | } |
1a3e8f3d | 1167 | rcu_read_unlock(); |
1da177e4 | 1168 | |
5b3eed75 | 1169 | xfs_iflock(ip); |
5b257b4a | 1170 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 1171 | |
5b3eed75 DC |
1172 | /* |
1173 | * we don't need to attach clean inodes or those only | |
1174 | * with unlogged changes (which we throw away, anyway). | |
1175 | */ | |
1da177e4 | 1176 | iip = ip->i_itemp; |
5b3eed75 | 1177 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 1178 | ASSERT(ip != free_ip); |
1da177e4 LT |
1179 | xfs_ifunlock(ip); |
1180 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1181 | continue; | |
1182 | } | |
1183 | ||
f5d8d5c4 CH |
1184 | iip->ili_last_fields = iip->ili_fields; |
1185 | iip->ili_fields = 0; | |
1da177e4 | 1186 | iip->ili_logged = 1; |
7b2e2a31 DC |
1187 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
1188 | &iip->ili_item.li_lsn); | |
1da177e4 | 1189 | |
ca30b2a7 CH |
1190 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
1191 | &iip->ili_item); | |
5b257b4a DC |
1192 | |
1193 | if (ip != free_ip) | |
1da177e4 | 1194 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1195 | } |
1196 | ||
5b3eed75 | 1197 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
1198 | xfs_trans_binval(tp, bp); |
1199 | } | |
1200 | ||
5017e97d | 1201 | xfs_perag_put(pag); |
2a30f36d | 1202 | return 0; |
1da177e4 LT |
1203 | } |
1204 | ||
1205 | /* | |
1206 | * This is called to return an inode to the inode free list. | |
1207 | * The inode should already be truncated to 0 length and have | |
1208 | * no pages associated with it. This routine also assumes that | |
1209 | * the inode is already a part of the transaction. | |
1210 | * | |
1211 | * The on-disk copy of the inode will have been added to the list | |
1212 | * of unlinked inodes in the AGI. We need to remove the inode from | |
1213 | * that list atomically with respect to freeing it here. | |
1214 | */ | |
1215 | int | |
1216 | xfs_ifree( | |
1217 | xfs_trans_t *tp, | |
1218 | xfs_inode_t *ip, | |
1219 | xfs_bmap_free_t *flist) | |
1220 | { | |
1221 | int error; | |
1222 | int delete; | |
1223 | xfs_ino_t first_ino; | |
1224 | ||
579aa9ca | 1225 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1226 | ASSERT(ip->i_d.di_nlink == 0); |
1227 | ASSERT(ip->i_d.di_nextents == 0); | |
1228 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 1229 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
1230 | ASSERT(ip->i_d.di_nblocks == 0); |
1231 | ||
1232 | /* | |
1233 | * Pull the on-disk inode from the AGI unlinked list. | |
1234 | */ | |
1235 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 1236 | if (error) |
1da177e4 | 1237 | return error; |
1da177e4 LT |
1238 | |
1239 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
1baaed8f | 1240 | if (error) |
1da177e4 | 1241 | return error; |
1baaed8f | 1242 | |
1da177e4 LT |
1243 | ip->i_d.di_mode = 0; /* mark incore inode as free */ |
1244 | ip->i_d.di_flags = 0; | |
1245 | ip->i_d.di_dmevmask = 0; | |
1246 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
1247 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
1248 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1249 | /* | |
1250 | * Bump the generation count so no one will be confused | |
1251 | * by reincarnations of this inode. | |
1252 | */ | |
1253 | ip->i_d.di_gen++; | |
1254 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1255 | ||
1baaed8f | 1256 | if (delete) |
2a30f36d | 1257 | error = xfs_ifree_cluster(ip, tp, first_ino); |
1da177e4 | 1258 | |
2a30f36d | 1259 | return error; |
1da177e4 LT |
1260 | } |
1261 | ||
1da177e4 | 1262 | /* |
60ec6783 CH |
1263 | * This is called to unpin an inode. The caller must have the inode locked |
1264 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
1265 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 1266 | */ |
60ec6783 | 1267 | static void |
f392e631 | 1268 | xfs_iunpin( |
60ec6783 | 1269 | struct xfs_inode *ip) |
1da177e4 | 1270 | { |
579aa9ca | 1271 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 1272 | |
4aaf15d1 DC |
1273 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
1274 | ||
a3f74ffb | 1275 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 1276 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 1277 | |
a3f74ffb | 1278 | } |
1da177e4 | 1279 | |
f392e631 CH |
1280 | static void |
1281 | __xfs_iunpin_wait( | |
1282 | struct xfs_inode *ip) | |
1283 | { | |
1284 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
1285 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
1286 | ||
1287 | xfs_iunpin(ip); | |
1288 | ||
1289 | do { | |
1290 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1291 | if (xfs_ipincount(ip)) | |
1292 | io_schedule(); | |
1293 | } while (xfs_ipincount(ip)); | |
1294 | finish_wait(wq, &wait.wait); | |
1295 | } | |
1296 | ||
777df5af | 1297 | void |
a3f74ffb | 1298 | xfs_iunpin_wait( |
60ec6783 | 1299 | struct xfs_inode *ip) |
a3f74ffb | 1300 | { |
f392e631 CH |
1301 | if (xfs_ipincount(ip)) |
1302 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
1303 | } |
1304 | ||
5c4d97d0 DC |
1305 | STATIC int |
1306 | xfs_iflush_cluster( | |
1307 | xfs_inode_t *ip, | |
1308 | xfs_buf_t *bp) | |
1da177e4 | 1309 | { |
5c4d97d0 DC |
1310 | xfs_mount_t *mp = ip->i_mount; |
1311 | struct xfs_perag *pag; | |
1312 | unsigned long first_index, mask; | |
1313 | unsigned long inodes_per_cluster; | |
1314 | int ilist_size; | |
1315 | xfs_inode_t **ilist; | |
1316 | xfs_inode_t *iq; | |
1317 | int nr_found; | |
1318 | int clcount = 0; | |
1319 | int bufwasdelwri; | |
1da177e4 | 1320 | int i; |
1da177e4 | 1321 | |
5c4d97d0 | 1322 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 1323 | |
5c4d97d0 DC |
1324 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
1325 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
1326 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); | |
1327 | if (!ilist) | |
1328 | goto out_put; | |
1da177e4 | 1329 | |
5c4d97d0 DC |
1330 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
1331 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
1332 | rcu_read_lock(); | |
1333 | /* really need a gang lookup range call here */ | |
1334 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
1335 | first_index, inodes_per_cluster); | |
1336 | if (nr_found == 0) | |
1337 | goto out_free; | |
1338 | ||
1339 | for (i = 0; i < nr_found; i++) { | |
1340 | iq = ilist[i]; | |
1341 | if (iq == ip) | |
bad55843 | 1342 | continue; |
1a3e8f3d DC |
1343 | |
1344 | /* | |
1345 | * because this is an RCU protected lookup, we could find a | |
1346 | * recently freed or even reallocated inode during the lookup. | |
1347 | * We need to check under the i_flags_lock for a valid inode | |
1348 | * here. Skip it if it is not valid or the wrong inode. | |
1349 | */ | |
1350 | spin_lock(&ip->i_flags_lock); | |
1351 | if (!ip->i_ino || | |
1352 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
1353 | spin_unlock(&ip->i_flags_lock); | |
1354 | continue; | |
1355 | } | |
1356 | spin_unlock(&ip->i_flags_lock); | |
1357 | ||
bad55843 DC |
1358 | /* |
1359 | * Do an un-protected check to see if the inode is dirty and | |
1360 | * is a candidate for flushing. These checks will be repeated | |
1361 | * later after the appropriate locks are acquired. | |
1362 | */ | |
33540408 | 1363 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 1364 | continue; |
bad55843 DC |
1365 | |
1366 | /* | |
1367 | * Try to get locks. If any are unavailable or it is pinned, | |
1368 | * then this inode cannot be flushed and is skipped. | |
1369 | */ | |
1370 | ||
1371 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
1372 | continue; | |
1373 | if (!xfs_iflock_nowait(iq)) { | |
1374 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
1375 | continue; | |
1376 | } | |
1377 | if (xfs_ipincount(iq)) { | |
1378 | xfs_ifunlock(iq); | |
1379 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
1380 | continue; | |
1381 | } | |
1382 | ||
1383 | /* | |
1384 | * arriving here means that this inode can be flushed. First | |
1385 | * re-check that it's dirty before flushing. | |
1386 | */ | |
33540408 DC |
1387 | if (!xfs_inode_clean(iq)) { |
1388 | int error; | |
bad55843 DC |
1389 | error = xfs_iflush_int(iq, bp); |
1390 | if (error) { | |
1391 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
1392 | goto cluster_corrupt_out; | |
1393 | } | |
1394 | clcount++; | |
1395 | } else { | |
1396 | xfs_ifunlock(iq); | |
1397 | } | |
1398 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
1399 | } | |
1400 | ||
1401 | if (clcount) { | |
1402 | XFS_STATS_INC(xs_icluster_flushcnt); | |
1403 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
1404 | } | |
1405 | ||
1406 | out_free: | |
1a3e8f3d | 1407 | rcu_read_unlock(); |
f0e2d93c | 1408 | kmem_free(ilist); |
44b56e0a DC |
1409 | out_put: |
1410 | xfs_perag_put(pag); | |
bad55843 DC |
1411 | return 0; |
1412 | ||
1413 | ||
1414 | cluster_corrupt_out: | |
1415 | /* | |
1416 | * Corruption detected in the clustering loop. Invalidate the | |
1417 | * inode buffer and shut down the filesystem. | |
1418 | */ | |
1a3e8f3d | 1419 | rcu_read_unlock(); |
bad55843 | 1420 | /* |
43ff2122 | 1421 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
1422 | * brelse can handle it with no problems. If not, shut down the |
1423 | * filesystem before releasing the buffer. | |
1424 | */ | |
43ff2122 | 1425 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
1426 | if (bufwasdelwri) |
1427 | xfs_buf_relse(bp); | |
1428 | ||
1429 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
1430 | ||
1431 | if (!bufwasdelwri) { | |
1432 | /* | |
1433 | * Just like incore_relse: if we have b_iodone functions, | |
1434 | * mark the buffer as an error and call them. Otherwise | |
1435 | * mark it as stale and brelse. | |
1436 | */ | |
cb669ca5 | 1437 | if (bp->b_iodone) { |
bad55843 | 1438 | XFS_BUF_UNDONE(bp); |
c867cb61 | 1439 | xfs_buf_stale(bp); |
5a52c2a5 | 1440 | xfs_buf_ioerror(bp, EIO); |
1a1a3e97 | 1441 | xfs_buf_ioend(bp, 0); |
bad55843 | 1442 | } else { |
c867cb61 | 1443 | xfs_buf_stale(bp); |
bad55843 DC |
1444 | xfs_buf_relse(bp); |
1445 | } | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * Unlocks the flush lock | |
1450 | */ | |
04913fdd | 1451 | xfs_iflush_abort(iq, false); |
f0e2d93c | 1452 | kmem_free(ilist); |
44b56e0a | 1453 | xfs_perag_put(pag); |
bad55843 DC |
1454 | return XFS_ERROR(EFSCORRUPTED); |
1455 | } | |
1456 | ||
1da177e4 | 1457 | /* |
4c46819a CH |
1458 | * Flush dirty inode metadata into the backing buffer. |
1459 | * | |
1460 | * The caller must have the inode lock and the inode flush lock held. The | |
1461 | * inode lock will still be held upon return to the caller, and the inode | |
1462 | * flush lock will be released after the inode has reached the disk. | |
1463 | * | |
1464 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
1465 | */ |
1466 | int | |
1467 | xfs_iflush( | |
4c46819a CH |
1468 | struct xfs_inode *ip, |
1469 | struct xfs_buf **bpp) | |
1da177e4 | 1470 | { |
4c46819a CH |
1471 | struct xfs_mount *mp = ip->i_mount; |
1472 | struct xfs_buf *bp; | |
1473 | struct xfs_dinode *dip; | |
1da177e4 | 1474 | int error; |
1da177e4 LT |
1475 | |
1476 | XFS_STATS_INC(xs_iflush_count); | |
1477 | ||
579aa9ca | 1478 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 1479 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 1480 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 1481 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 1482 | |
4c46819a | 1483 | *bpp = NULL; |
1da177e4 | 1484 | |
1da177e4 LT |
1485 | xfs_iunpin_wait(ip); |
1486 | ||
4b6a4688 DC |
1487 | /* |
1488 | * For stale inodes we cannot rely on the backing buffer remaining | |
1489 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 1490 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
1491 | * inodes below. We have to check this after ensuring the inode is |
1492 | * unpinned so that it is safe to reclaim the stale inode after the | |
1493 | * flush call. | |
1494 | */ | |
1495 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
1496 | xfs_ifunlock(ip); | |
1497 | return 0; | |
1498 | } | |
1499 | ||
1da177e4 LT |
1500 | /* |
1501 | * This may have been unpinned because the filesystem is shutting | |
1502 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
1503 | * to disk, because the log record didn't make it to disk. |
1504 | * | |
1505 | * We also have to remove the log item from the AIL in this case, | |
1506 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
1507 | */ |
1508 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
32ce90a4 CH |
1509 | error = XFS_ERROR(EIO); |
1510 | goto abort_out; | |
1da177e4 LT |
1511 | } |
1512 | ||
a3f74ffb DC |
1513 | /* |
1514 | * Get the buffer containing the on-disk inode. | |
1515 | */ | |
475ee413 CH |
1516 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
1517 | 0); | |
a3f74ffb DC |
1518 | if (error || !bp) { |
1519 | xfs_ifunlock(ip); | |
1520 | return error; | |
1521 | } | |
1522 | ||
1da177e4 LT |
1523 | /* |
1524 | * First flush out the inode that xfs_iflush was called with. | |
1525 | */ | |
1526 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 1527 | if (error) |
1da177e4 | 1528 | goto corrupt_out; |
1da177e4 | 1529 | |
a3f74ffb DC |
1530 | /* |
1531 | * If the buffer is pinned then push on the log now so we won't | |
1532 | * get stuck waiting in the write for too long. | |
1533 | */ | |
811e64c7 | 1534 | if (xfs_buf_ispinned(bp)) |
a14a348b | 1535 | xfs_log_force(mp, 0); |
a3f74ffb | 1536 | |
1da177e4 LT |
1537 | /* |
1538 | * inode clustering: | |
1539 | * see if other inodes can be gathered into this write | |
1540 | */ | |
bad55843 DC |
1541 | error = xfs_iflush_cluster(ip, bp); |
1542 | if (error) | |
1543 | goto cluster_corrupt_out; | |
1da177e4 | 1544 | |
4c46819a CH |
1545 | *bpp = bp; |
1546 | return 0; | |
1da177e4 LT |
1547 | |
1548 | corrupt_out: | |
1549 | xfs_buf_relse(bp); | |
7d04a335 | 1550 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 1551 | cluster_corrupt_out: |
32ce90a4 CH |
1552 | error = XFS_ERROR(EFSCORRUPTED); |
1553 | abort_out: | |
1da177e4 LT |
1554 | /* |
1555 | * Unlocks the flush lock | |
1556 | */ | |
04913fdd | 1557 | xfs_iflush_abort(ip, false); |
32ce90a4 | 1558 | return error; |
1da177e4 LT |
1559 | } |
1560 | ||
1561 | ||
1562 | STATIC int | |
1563 | xfs_iflush_int( | |
93848a99 CH |
1564 | struct xfs_inode *ip, |
1565 | struct xfs_buf *bp) | |
1da177e4 | 1566 | { |
93848a99 CH |
1567 | struct xfs_inode_log_item *iip = ip->i_itemp; |
1568 | struct xfs_dinode *dip; | |
1569 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 1570 | |
579aa9ca | 1571 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 1572 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 1573 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 1574 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 1575 | ASSERT(iip != NULL && iip->ili_fields != 0); |
1da177e4 | 1576 | |
1da177e4 | 1577 | /* set *dip = inode's place in the buffer */ |
92bfc6e7 | 1578 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 1579 | |
69ef921b | 1580 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 1581 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
1582 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1583 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
1584 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
1585 | goto corrupt_out; |
1586 | } | |
1587 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
1588 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
6a19d939 DC |
1589 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1590 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
1591 | __func__, ip->i_ino, ip, ip->i_d.di_magic); | |
1da177e4 LT |
1592 | goto corrupt_out; |
1593 | } | |
abbede1b | 1594 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
1595 | if (XFS_TEST_ERROR( |
1596 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
1597 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
1598 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
1599 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1600 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
1601 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
1602 | goto corrupt_out; |
1603 | } | |
abbede1b | 1604 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
1605 | if (XFS_TEST_ERROR( |
1606 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
1607 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
1608 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
1609 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
1610 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1611 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
1612 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
1613 | goto corrupt_out; |
1614 | } | |
1615 | } | |
1616 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
1617 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
1618 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
1619 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1620 | "%s: detected corrupt incore inode %Lu, " | |
1621 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
1622 | __func__, ip->i_ino, | |
1da177e4 | 1623 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 1624 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
1625 | goto corrupt_out; |
1626 | } | |
1627 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
1628 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
1629 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
1630 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
1631 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
1632 | goto corrupt_out; |
1633 | } | |
e60896d8 | 1634 | |
1da177e4 | 1635 | /* |
e60896d8 DC |
1636 | * Inode item log recovery for v1/v2 inodes are dependent on the |
1637 | * di_flushiter count for correct sequencing. We bump the flush | |
1638 | * iteration count so we can detect flushes which postdate a log record | |
1639 | * during recovery. This is redundant as we now log every change and | |
1640 | * hence this can't happen but we need to still do it to ensure | |
1641 | * backwards compatibility with old kernels that predate logging all | |
1642 | * inode changes. | |
1da177e4 | 1643 | */ |
e60896d8 DC |
1644 | if (ip->i_d.di_version < 3) |
1645 | ip->i_d.di_flushiter++; | |
1da177e4 LT |
1646 | |
1647 | /* | |
1648 | * Copy the dirty parts of the inode into the on-disk | |
1649 | * inode. We always copy out the core of the inode, | |
1650 | * because if the inode is dirty at all the core must | |
1651 | * be. | |
1652 | */ | |
81591fe2 | 1653 | xfs_dinode_to_disk(dip, &ip->i_d); |
1da177e4 LT |
1654 | |
1655 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
1656 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
1657 | ip->i_d.di_flushiter = 0; | |
1658 | ||
1659 | /* | |
1660 | * If this is really an old format inode and the superblock version | |
1661 | * has not been updated to support only new format inodes, then | |
1662 | * convert back to the old inode format. If the superblock version | |
1663 | * has been updated, then make the conversion permanent. | |
1664 | */ | |
51ce16d5 CH |
1665 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
1666 | if (ip->i_d.di_version == 1) { | |
62118709 | 1667 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
1668 | /* |
1669 | * Convert it back. | |
1670 | */ | |
1671 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
81591fe2 | 1672 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
1673 | } else { |
1674 | /* | |
1675 | * The superblock version has already been bumped, | |
1676 | * so just make the conversion to the new inode | |
1677 | * format permanent. | |
1678 | */ | |
51ce16d5 CH |
1679 | ip->i_d.di_version = 2; |
1680 | dip->di_version = 2; | |
1da177e4 | 1681 | ip->i_d.di_onlink = 0; |
81591fe2 | 1682 | dip->di_onlink = 0; |
1da177e4 | 1683 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
81591fe2 CH |
1684 | memset(&(dip->di_pad[0]), 0, |
1685 | sizeof(dip->di_pad)); | |
6743099c | 1686 | ASSERT(xfs_get_projid(ip) == 0); |
1da177e4 LT |
1687 | } |
1688 | } | |
1689 | ||
e4ac967b DC |
1690 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
1691 | if (XFS_IFORK_Q(ip)) | |
1692 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
1693 | xfs_inobp_check(mp, bp); |
1694 | ||
1695 | /* | |
f5d8d5c4 CH |
1696 | * We've recorded everything logged in the inode, so we'd like to clear |
1697 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
1698 | * However, we can't stop logging all this information until the data | |
1699 | * we've copied into the disk buffer is written to disk. If we did we | |
1700 | * might overwrite the copy of the inode in the log with all the data | |
1701 | * after re-logging only part of it, and in the face of a crash we | |
1702 | * wouldn't have all the data we need to recover. | |
1da177e4 | 1703 | * |
f5d8d5c4 CH |
1704 | * What we do is move the bits to the ili_last_fields field. When |
1705 | * logging the inode, these bits are moved back to the ili_fields field. | |
1706 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
1707 | * know that the information those bits represent is permanently on | |
1708 | * disk. As long as the flush completes before the inode is logged | |
1709 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 1710 | * |
f5d8d5c4 CH |
1711 | * We can play with the ili_fields bits here, because the inode lock |
1712 | * must be held exclusively in order to set bits there and the flush | |
1713 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
1714 | * done routine can tell whether or not to look in the AIL. Also, store | |
1715 | * the current LSN of the inode so that we can tell whether the item has | |
1716 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
1717 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
1718 | * atomically. | |
1da177e4 | 1719 | */ |
93848a99 CH |
1720 | iip->ili_last_fields = iip->ili_fields; |
1721 | iip->ili_fields = 0; | |
1722 | iip->ili_logged = 1; | |
1da177e4 | 1723 | |
93848a99 CH |
1724 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
1725 | &iip->ili_item.li_lsn); | |
1da177e4 | 1726 | |
93848a99 CH |
1727 | /* |
1728 | * Attach the function xfs_iflush_done to the inode's | |
1729 | * buffer. This will remove the inode from the AIL | |
1730 | * and unlock the inode's flush lock when the inode is | |
1731 | * completely written to disk. | |
1732 | */ | |
1733 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 1734 | |
93848a99 CH |
1735 | /* update the lsn in the on disk inode if required */ |
1736 | if (ip->i_d.di_version == 3) | |
1737 | dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); | |
1738 | ||
1739 | /* generate the checksum. */ | |
1740 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 1741 | |
93848a99 CH |
1742 | ASSERT(bp->b_fspriv != NULL); |
1743 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
1744 | return 0; |
1745 | ||
1746 | corrupt_out: | |
1747 | return XFS_ERROR(EFSCORRUPTED); | |
1748 | } | |
1749 | ||
72b53efa BF |
1750 | /* |
1751 | * Test whether it is appropriate to check an inode for and free post EOF | |
1752 | * blocks. The 'force' parameter determines whether we should also consider | |
1753 | * regular files that are marked preallocated or append-only. | |
1754 | */ | |
1755 | bool | |
1756 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
1757 | { | |
1758 | /* prealloc/delalloc exists only on regular files */ | |
1759 | if (!S_ISREG(ip->i_d.di_mode)) | |
1760 | return false; | |
1761 | ||
1762 | /* | |
1763 | * Zero sized files with no cached pages and delalloc blocks will not | |
1764 | * have speculative prealloc/delalloc blocks to remove. | |
1765 | */ | |
1766 | if (VFS_I(ip)->i_size == 0 && | |
1767 | VN_CACHED(VFS_I(ip)) == 0 && | |
1768 | ip->i_delayed_blks == 0) | |
1769 | return false; | |
1770 | ||
1771 | /* If we haven't read in the extent list, then don't do it now. */ | |
1772 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
1773 | return false; | |
1774 | ||
1775 | /* | |
1776 | * Do not free real preallocated or append-only files unless the file | |
1777 | * has delalloc blocks and we are forced to remove them. | |
1778 | */ | |
1779 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
1780 | if (!force || ip->i_delayed_blks == 0) | |
1781 | return false; | |
1782 | ||
1783 | return true; | |
1784 | } |