Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1e69dd0e | 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 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
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 NS |
33 | #include "xfs_dir2_sf.h" |
34 | #include "xfs_attr_sf.h" | |
35 | #include "xfs_dinode.h" | |
36 | #include "xfs_inode.h" | |
37 | #include "xfs_inode_item.h" | |
1da177e4 LT |
38 | #include "xfs_itable.h" |
39 | #include "xfs_btree.h" | |
40 | #include "xfs_alloc.h" | |
41 | #include "xfs_ialloc.h" | |
42 | #include "xfs_attr.h" | |
1da177e4 LT |
43 | #include "xfs_bmap.h" |
44 | #include "xfs_acl.h" | |
1da177e4 LT |
45 | #include "xfs_error.h" |
46 | #include "xfs_buf_item.h" | |
47 | #include "xfs_rw.h" | |
48 | ||
49 | /* | |
50 | * This is a subroutine for xfs_write() and other writers (xfs_ioctl) | |
51 | * which clears the setuid and setgid bits when a file is written. | |
52 | */ | |
53 | int | |
54 | xfs_write_clear_setuid( | |
55 | xfs_inode_t *ip) | |
56 | { | |
57 | xfs_mount_t *mp; | |
58 | xfs_trans_t *tp; | |
59 | int error; | |
60 | ||
61 | mp = ip->i_mount; | |
62 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
63 | if ((error = xfs_trans_reserve(tp, 0, | |
64 | XFS_WRITEID_LOG_RES(mp), | |
65 | 0, 0, 0))) { | |
66 | xfs_trans_cancel(tp, 0); | |
67 | return error; | |
68 | } | |
69 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
70 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
71 | xfs_trans_ihold(tp, ip); | |
72 | ip->i_d.di_mode &= ~S_ISUID; | |
73 | ||
74 | /* | |
75 | * Note that we don't have to worry about mandatory | |
76 | * file locking being disabled here because we only | |
77 | * clear the S_ISGID bit if the Group execute bit is | |
78 | * on, but if it was on then mandatory locking wouldn't | |
79 | * have been enabled. | |
80 | */ | |
81 | if (ip->i_d.di_mode & S_IXGRP) { | |
82 | ip->i_d.di_mode &= ~S_ISGID; | |
83 | } | |
84 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
85 | xfs_trans_set_sync(tp); | |
1c72bf90 | 86 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
87 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
88 | return 0; | |
89 | } | |
90 | ||
1e69dd0e NS |
91 | /* |
92 | * Handle logging requirements of various synchronous types of write. | |
93 | */ | |
94 | int | |
95 | xfs_write_sync_logforce( | |
96 | xfs_mount_t *mp, | |
97 | xfs_inode_t *ip) | |
98 | { | |
99 | int error = 0; | |
100 | ||
101 | /* | |
102 | * If we're treating this as O_DSYNC and we have not updated the | |
103 | * size, force the log. | |
104 | */ | |
105 | if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC) && | |
106 | !(ip->i_update_size)) { | |
107 | xfs_inode_log_item_t *iip = ip->i_itemp; | |
108 | ||
109 | /* | |
110 | * If an allocation transaction occurred | |
111 | * without extending the size, then we have to force | |
112 | * the log up the proper point to ensure that the | |
113 | * allocation is permanent. We can't count on | |
114 | * the fact that buffered writes lock out direct I/O | |
115 | * writes - the direct I/O write could have extended | |
116 | * the size nontransactionally, then finished before | |
117 | * we started. xfs_write_file will think that the file | |
118 | * didn't grow but the update isn't safe unless the | |
119 | * size change is logged. | |
120 | * | |
121 | * Force the log if we've committed a transaction | |
122 | * against the inode or if someone else has and | |
123 | * the commit record hasn't gone to disk (e.g. | |
124 | * the inode is pinned). This guarantees that | |
125 | * all changes affecting the inode are permanent | |
126 | * when we return. | |
127 | */ | |
128 | if (iip && iip->ili_last_lsn) { | |
129 | xfs_log_force(mp, iip->ili_last_lsn, | |
130 | XFS_LOG_FORCE | XFS_LOG_SYNC); | |
131 | } else if (xfs_ipincount(ip) > 0) { | |
132 | xfs_log_force(mp, (xfs_lsn_t)0, | |
133 | XFS_LOG_FORCE | XFS_LOG_SYNC); | |
134 | } | |
135 | ||
136 | } else { | |
137 | xfs_trans_t *tp; | |
138 | ||
139 | /* | |
140 | * O_SYNC or O_DSYNC _with_ a size update are handled | |
141 | * the same way. | |
142 | * | |
143 | * If the write was synchronous then we need to make | |
144 | * sure that the inode modification time is permanent. | |
145 | * We'll have updated the timestamp above, so here | |
146 | * we use a synchronous transaction to log the inode. | |
147 | * It's not fast, but it's necessary. | |
148 | * | |
149 | * If this a dsync write and the size got changed | |
150 | * non-transactionally, then we need to ensure that | |
151 | * the size change gets logged in a synchronous | |
152 | * transaction. | |
153 | */ | |
154 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC); | |
155 | if ((error = xfs_trans_reserve(tp, 0, | |
156 | XFS_SWRITE_LOG_RES(mp), | |
157 | 0, 0, 0))) { | |
158 | /* Transaction reserve failed */ | |
159 | xfs_trans_cancel(tp, 0); | |
160 | } else { | |
161 | /* Transaction reserve successful */ | |
162 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
163 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
164 | xfs_trans_ihold(tp, ip); | |
165 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
166 | xfs_trans_set_sync(tp); | |
1c72bf90 | 167 | error = xfs_trans_commit(tp, 0); |
1e69dd0e NS |
168 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
169 | } | |
170 | } | |
171 | ||
172 | return error; | |
173 | } | |
174 | ||
1da177e4 LT |
175 | /* |
176 | * Force a shutdown of the filesystem instantly while keeping | |
177 | * the filesystem consistent. We don't do an unmount here; just shutdown | |
178 | * the shop, make sure that absolutely nothing persistent happens to | |
179 | * this filesystem after this point. | |
180 | */ | |
1da177e4 LT |
181 | void |
182 | xfs_do_force_shutdown( | |
48c872a9 | 183 | xfs_mount_t *mp, |
1da177e4 LT |
184 | int flags, |
185 | char *fname, | |
186 | int lnnum) | |
187 | { | |
188 | int logerror; | |
1da177e4 | 189 | |
7d04a335 | 190 | logerror = flags & SHUTDOWN_LOG_IO_ERROR; |
1da177e4 | 191 | |
7d04a335 NS |
192 | if (!(flags & SHUTDOWN_FORCE_UMOUNT)) { |
193 | cmn_err(CE_NOTE, "xfs_force_shutdown(%s,0x%x) called from " | |
194 | "line %d of file %s. Return address = 0x%p", | |
195 | mp->m_fsname, flags, lnnum, fname, __return_address); | |
1da177e4 LT |
196 | } |
197 | /* | |
198 | * No need to duplicate efforts. | |
199 | */ | |
200 | if (XFS_FORCED_SHUTDOWN(mp) && !logerror) | |
201 | return; | |
202 | ||
203 | /* | |
204 | * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't | |
205 | * queue up anybody new on the log reservations, and wakes up | |
7d04a335 NS |
206 | * everybody who's sleeping on log reservations to tell them |
207 | * the bad news. | |
1da177e4 LT |
208 | */ |
209 | if (xfs_log_force_umount(mp, logerror)) | |
210 | return; | |
211 | ||
7d04a335 | 212 | if (flags & SHUTDOWN_CORRUPT_INCORE) { |
1da177e4 LT |
213 | xfs_cmn_err(XFS_PTAG_SHUTDOWN_CORRUPT, CE_ALERT, mp, |
214 | "Corruption of in-memory data detected. Shutting down filesystem: %s", | |
215 | mp->m_fsname); | |
216 | if (XFS_ERRLEVEL_HIGH <= xfs_error_level) { | |
217 | xfs_stack_trace(); | |
218 | } | |
7d04a335 | 219 | } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) { |
1da177e4 LT |
220 | if (logerror) { |
221 | xfs_cmn_err(XFS_PTAG_SHUTDOWN_LOGERROR, CE_ALERT, mp, | |
7d04a335 | 222 | "Log I/O Error Detected. Shutting down filesystem: %s", |
1da177e4 | 223 | mp->m_fsname); |
7d04a335 | 224 | } else if (flags & SHUTDOWN_DEVICE_REQ) { |
1da177e4 | 225 | xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp, |
7d04a335 NS |
226 | "All device paths lost. Shutting down filesystem: %s", |
227 | mp->m_fsname); | |
228 | } else if (!(flags & SHUTDOWN_REMOTE_REQ)) { | |
229 | xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp, | |
230 | "I/O Error Detected. Shutting down filesystem: %s", | |
1da177e4 LT |
231 | mp->m_fsname); |
232 | } | |
233 | } | |
7d04a335 NS |
234 | if (!(flags & SHUTDOWN_FORCE_UMOUNT)) { |
235 | cmn_err(CE_ALERT, "Please umount the filesystem, " | |
236 | "and rectify the problem(s)"); | |
1da177e4 LT |
237 | } |
238 | } | |
239 | ||
240 | ||
241 | /* | |
242 | * Called when we want to stop a buffer from getting written or read. | |
243 | * We attach the EIO error, muck with its flags, and call biodone | |
244 | * so that the proper iodone callbacks get called. | |
245 | */ | |
246 | int | |
247 | xfs_bioerror( | |
248 | xfs_buf_t *bp) | |
249 | { | |
250 | ||
251 | #ifdef XFSERRORDEBUG | |
252 | ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone); | |
253 | #endif | |
254 | ||
255 | /* | |
256 | * No need to wait until the buffer is unpinned. | |
257 | * We aren't flushing it. | |
258 | */ | |
259 | xfs_buftrace("XFS IOERROR", bp); | |
260 | XFS_BUF_ERROR(bp, EIO); | |
261 | /* | |
262 | * We're calling biodone, so delete B_DONE flag. Either way | |
263 | * we have to call the iodone callback, and calling biodone | |
264 | * probably is the best way since it takes care of | |
265 | * GRIO as well. | |
266 | */ | |
267 | XFS_BUF_UNREAD(bp); | |
268 | XFS_BUF_UNDELAYWRITE(bp); | |
269 | XFS_BUF_UNDONE(bp); | |
270 | XFS_BUF_STALE(bp); | |
271 | ||
272 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
273 | xfs_biodone(bp); | |
274 | ||
275 | return (EIO); | |
276 | } | |
277 | ||
278 | /* | |
279 | * Same as xfs_bioerror, except that we are releasing the buffer | |
280 | * here ourselves, and avoiding the biodone call. | |
281 | * This is meant for userdata errors; metadata bufs come with | |
282 | * iodone functions attached, so that we can track down errors. | |
283 | */ | |
284 | int | |
285 | xfs_bioerror_relse( | |
286 | xfs_buf_t *bp) | |
287 | { | |
288 | int64_t fl; | |
289 | ||
290 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != xfs_buf_iodone_callbacks); | |
291 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != xlog_iodone); | |
292 | ||
293 | xfs_buftrace("XFS IOERRELSE", bp); | |
294 | fl = XFS_BUF_BFLAGS(bp); | |
295 | /* | |
296 | * No need to wait until the buffer is unpinned. | |
297 | * We aren't flushing it. | |
298 | * | |
299 | * chunkhold expects B_DONE to be set, whether | |
300 | * we actually finish the I/O or not. We don't want to | |
301 | * change that interface. | |
302 | */ | |
303 | XFS_BUF_UNREAD(bp); | |
304 | XFS_BUF_UNDELAYWRITE(bp); | |
305 | XFS_BUF_DONE(bp); | |
306 | XFS_BUF_STALE(bp); | |
307 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
308 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
309 | if (!(fl & XFS_B_ASYNC)) { | |
310 | /* | |
311 | * Mark b_error and B_ERROR _both_. | |
312 | * Lot's of chunkcache code assumes that. | |
313 | * There's no reason to mark error for | |
314 | * ASYNC buffers. | |
315 | */ | |
316 | XFS_BUF_ERROR(bp, EIO); | |
317 | XFS_BUF_V_IODONESEMA(bp); | |
318 | } else { | |
319 | xfs_buf_relse(bp); | |
320 | } | |
321 | return (EIO); | |
322 | } | |
ce8e922c | 323 | |
1da177e4 LT |
324 | /* |
325 | * Prints out an ALERT message about I/O error. | |
326 | */ | |
327 | void | |
328 | xfs_ioerror_alert( | |
329 | char *func, | |
330 | struct xfs_mount *mp, | |
331 | xfs_buf_t *bp, | |
332 | xfs_daddr_t blkno) | |
333 | { | |
334 | cmn_err(CE_ALERT, | |
335 | "I/O error in filesystem (\"%s\") meta-data dev %s block 0x%llx" | |
da1650a5 | 336 | " (\"%s\") error %d buf count %zd", |
1da177e4 | 337 | (!mp || !mp->m_fsname) ? "(fs name not set)" : mp->m_fsname, |
ce8e922c NS |
338 | XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)), |
339 | (__uint64_t)blkno, func, | |
340 | XFS_BUF_GETERROR(bp), XFS_BUF_COUNT(bp)); | |
1da177e4 LT |
341 | } |
342 | ||
343 | /* | |
344 | * This isn't an absolute requirement, but it is | |
345 | * just a good idea to call xfs_read_buf instead of | |
346 | * directly doing a read_buf call. For one, we shouldn't | |
347 | * be doing this disk read if we are in SHUTDOWN state anyway, | |
348 | * so this stops that from happening. Secondly, this does all | |
349 | * the error checking stuff and the brelse if appropriate for | |
350 | * the caller, so the code can be a little leaner. | |
351 | */ | |
352 | ||
353 | int | |
354 | xfs_read_buf( | |
355 | struct xfs_mount *mp, | |
356 | xfs_buftarg_t *target, | |
357 | xfs_daddr_t blkno, | |
358 | int len, | |
359 | uint flags, | |
360 | xfs_buf_t **bpp) | |
361 | { | |
362 | xfs_buf_t *bp; | |
363 | int error; | |
364 | ||
365 | if (flags) | |
366 | bp = xfs_buf_read_flags(target, blkno, len, flags); | |
367 | else | |
368 | bp = xfs_buf_read(target, blkno, len, flags); | |
369 | if (!bp) | |
370 | return XFS_ERROR(EIO); | |
371 | error = XFS_BUF_GETERROR(bp); | |
372 | if (bp && !error && !XFS_FORCED_SHUTDOWN(mp)) { | |
373 | *bpp = bp; | |
374 | } else { | |
375 | *bpp = NULL; | |
376 | if (error) { | |
377 | xfs_ioerror_alert("xfs_read_buf", mp, bp, XFS_BUF_ADDR(bp)); | |
378 | } else { | |
379 | error = XFS_ERROR(EIO); | |
380 | } | |
381 | if (bp) { | |
382 | XFS_BUF_UNDONE(bp); | |
383 | XFS_BUF_UNDELAYWRITE(bp); | |
384 | XFS_BUF_STALE(bp); | |
385 | /* | |
386 | * brelse clears B_ERROR and b_error | |
387 | */ | |
388 | xfs_buf_relse(bp); | |
389 | } | |
390 | } | |
391 | return (error); | |
392 | } | |
393 | ||
394 | /* | |
395 | * Wrapper around bwrite() so that we can trap | |
396 | * write errors, and act accordingly. | |
397 | */ | |
398 | int | |
399 | xfs_bwrite( | |
400 | struct xfs_mount *mp, | |
401 | struct xfs_buf *bp) | |
402 | { | |
403 | int error; | |
404 | ||
405 | /* | |
406 | * XXXsup how does this work for quotas. | |
407 | */ | |
408 | XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb); | |
409 | XFS_BUF_SET_FSPRIVATE3(bp, mp); | |
410 | XFS_BUF_WRITE(bp); | |
411 | ||
412 | if ((error = XFS_bwrite(bp))) { | |
413 | ASSERT(mp); | |
414 | /* | |
415 | * Cannot put a buftrace here since if the buffer is not | |
416 | * B_HOLD then we will brelse() the buffer before returning | |
417 | * from bwrite and we could be tracing a buffer that has | |
418 | * been reused. | |
419 | */ | |
7d04a335 | 420 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
421 | } |
422 | return (error); | |
423 | } |