Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of version 2 of the GNU General Public License as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it would be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. | |
11 | * | |
12 | * Further, this software is distributed without any warranty that it is | |
13 | * free of the rightful claim of any third person regarding infringement | |
14 | * or the like. Any license provided herein, whether implied or | |
15 | * otherwise, applies only to this software file. Patent licenses, if | |
16 | * any, provided herein do not apply to combinations of this program with | |
17 | * other software, or any other product whatsoever. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License along | |
20 | * with this program; if not, write the Free Software Foundation, Inc., 59 | |
21 | * Temple Place - Suite 330, Boston MA 02111-1307, USA. | |
22 | * | |
23 | * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy, | |
24 | * Mountain View, CA 94043, or: | |
25 | * | |
26 | * http://www.sgi.com | |
27 | * | |
28 | * For further information regarding this notice, see: | |
29 | * | |
30 | * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/ | |
31 | */ | |
1da177e4 | 32 | #include "xfs.h" |
a844f451 | 33 | #include "xfs_fs.h" |
1da177e4 | 34 | #include "xfs_types.h" |
a844f451 | 35 | #include "xfs_bit.h" |
1da177e4 | 36 | #include "xfs_log.h" |
a844f451 NS |
37 | #include "xfs_inum.h" |
38 | #include "xfs_imap.h" | |
1da177e4 LT |
39 | #include "xfs_trans.h" |
40 | #include "xfs_trans_priv.h" | |
41 | #include "xfs_sb.h" | |
42 | #include "xfs_ag.h" | |
43 | #include "xfs_dir.h" | |
44 | #include "xfs_dir2.h" | |
45 | #include "xfs_dmapi.h" | |
46 | #include "xfs_mount.h" | |
1da177e4 | 47 | #include "xfs_bmap_btree.h" |
a844f451 | 48 | #include "xfs_alloc_btree.h" |
1da177e4 | 49 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
50 | #include "xfs_dir_sf.h" |
51 | #include "xfs_dir2_sf.h" | |
a844f451 | 52 | #include "xfs_attr_sf.h" |
1da177e4 | 53 | #include "xfs_dinode.h" |
1da177e4 | 54 | #include "xfs_inode.h" |
1da177e4 | 55 | #include "xfs_buf_item.h" |
a844f451 NS |
56 | #include "xfs_inode_item.h" |
57 | #include "xfs_btree.h" | |
58 | #include "xfs_alloc.h" | |
59 | #include "xfs_ialloc.h" | |
60 | #include "xfs_bmap.h" | |
1da177e4 LT |
61 | #include "xfs_rw.h" |
62 | #include "xfs_error.h" | |
1da177e4 LT |
63 | #include "xfs_utils.h" |
64 | #include "xfs_dir2_trace.h" | |
65 | #include "xfs_quota.h" | |
66 | #include "xfs_mac.h" | |
67 | #include "xfs_acl.h" | |
68 | ||
69 | ||
70 | kmem_zone_t *xfs_ifork_zone; | |
71 | kmem_zone_t *xfs_inode_zone; | |
72 | kmem_zone_t *xfs_chashlist_zone; | |
73 | ||
74 | /* | |
75 | * Used in xfs_itruncate(). This is the maximum number of extents | |
76 | * freed from a file in a single transaction. | |
77 | */ | |
78 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
79 | ||
80 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
81 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
82 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
83 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
84 | ||
85 | ||
86 | #ifdef DEBUG | |
87 | /* | |
88 | * Make sure that the extents in the given memory buffer | |
89 | * are valid. | |
90 | */ | |
91 | STATIC void | |
92 | xfs_validate_extents( | |
93 | xfs_bmbt_rec_t *ep, | |
94 | int nrecs, | |
95 | int disk, | |
96 | xfs_exntfmt_t fmt) | |
97 | { | |
98 | xfs_bmbt_irec_t irec; | |
99 | xfs_bmbt_rec_t rec; | |
100 | int i; | |
101 | ||
102 | for (i = 0; i < nrecs; i++) { | |
103 | rec.l0 = get_unaligned((__uint64_t*)&ep->l0); | |
104 | rec.l1 = get_unaligned((__uint64_t*)&ep->l1); | |
105 | if (disk) | |
106 | xfs_bmbt_disk_get_all(&rec, &irec); | |
107 | else | |
108 | xfs_bmbt_get_all(&rec, &irec); | |
109 | if (fmt == XFS_EXTFMT_NOSTATE) | |
110 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
111 | ep++; | |
112 | } | |
113 | } | |
114 | #else /* DEBUG */ | |
115 | #define xfs_validate_extents(ep, nrecs, disk, fmt) | |
116 | #endif /* DEBUG */ | |
117 | ||
118 | /* | |
119 | * Check that none of the inode's in the buffer have a next | |
120 | * unlinked field of 0. | |
121 | */ | |
122 | #if defined(DEBUG) | |
123 | void | |
124 | xfs_inobp_check( | |
125 | xfs_mount_t *mp, | |
126 | xfs_buf_t *bp) | |
127 | { | |
128 | int i; | |
129 | int j; | |
130 | xfs_dinode_t *dip; | |
131 | ||
132 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
133 | ||
134 | for (i = 0; i < j; i++) { | |
135 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
136 | i * mp->m_sb.sb_inodesize); | |
137 | if (!dip->di_next_unlinked) { | |
138 | xfs_fs_cmn_err(CE_ALERT, mp, | |
139 | "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", | |
140 | bp); | |
141 | ASSERT(dip->di_next_unlinked); | |
142 | } | |
143 | } | |
144 | } | |
145 | #endif | |
146 | ||
1da177e4 LT |
147 | /* |
148 | * This routine is called to map an inode number within a file | |
149 | * system to the buffer containing the on-disk version of the | |
150 | * inode. It returns a pointer to the buffer containing the | |
151 | * on-disk inode in the bpp parameter, and in the dip parameter | |
152 | * it returns a pointer to the on-disk inode within that buffer. | |
153 | * | |
154 | * If a non-zero error is returned, then the contents of bpp and | |
155 | * dipp are undefined. | |
156 | * | |
157 | * Use xfs_imap() to determine the size and location of the | |
158 | * buffer to read from disk. | |
159 | */ | |
ba0f32d4 | 160 | STATIC int |
1da177e4 LT |
161 | xfs_inotobp( |
162 | xfs_mount_t *mp, | |
163 | xfs_trans_t *tp, | |
164 | xfs_ino_t ino, | |
165 | xfs_dinode_t **dipp, | |
166 | xfs_buf_t **bpp, | |
167 | int *offset) | |
168 | { | |
169 | int di_ok; | |
170 | xfs_imap_t imap; | |
171 | xfs_buf_t *bp; | |
172 | int error; | |
173 | xfs_dinode_t *dip; | |
174 | ||
175 | /* | |
176 | * Call the space managment code to find the location of the | |
177 | * inode on disk. | |
178 | */ | |
179 | imap.im_blkno = 0; | |
180 | error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP); | |
181 | if (error != 0) { | |
182 | cmn_err(CE_WARN, | |
183 | "xfs_inotobp: xfs_imap() returned an " | |
184 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
185 | return error; | |
186 | } | |
187 | ||
188 | /* | |
189 | * If the inode number maps to a block outside the bounds of the | |
190 | * file system then return NULL rather than calling read_buf | |
191 | * and panicing when we get an error from the driver. | |
192 | */ | |
193 | if ((imap.im_blkno + imap.im_len) > | |
194 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
195 | cmn_err(CE_WARN, | |
da1650a5 | 196 | "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds " |
1da177e4 | 197 | "of the file system %s. Returning EINVAL.", |
da1650a5 CH |
198 | (unsigned long long)imap.im_blkno, |
199 | imap.im_len, mp->m_fsname); | |
1da177e4 LT |
200 | return XFS_ERROR(EINVAL); |
201 | } | |
202 | ||
203 | /* | |
204 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
205 | * default to just a read_buf() call. | |
206 | */ | |
207 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
208 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
209 | ||
210 | if (error) { | |
211 | cmn_err(CE_WARN, | |
212 | "xfs_inotobp: xfs_trans_read_buf() returned an " | |
213 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
214 | return error; | |
215 | } | |
216 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0); | |
217 | di_ok = | |
218 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && | |
219 | XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); | |
220 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP, | |
221 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
222 | XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip); | |
223 | xfs_trans_brelse(tp, bp); | |
224 | cmn_err(CE_WARN, | |
225 | "xfs_inotobp: XFS_TEST_ERROR() returned an " | |
226 | "error on %s. Returning EFSCORRUPTED.", mp->m_fsname); | |
227 | return XFS_ERROR(EFSCORRUPTED); | |
228 | } | |
229 | ||
230 | xfs_inobp_check(mp, bp); | |
231 | ||
232 | /* | |
233 | * Set *dipp to point to the on-disk inode in the buffer. | |
234 | */ | |
235 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
236 | *bpp = bp; | |
237 | *offset = imap.im_boffset; | |
238 | return 0; | |
239 | } | |
240 | ||
241 | ||
242 | /* | |
243 | * This routine is called to map an inode to the buffer containing | |
244 | * the on-disk version of the inode. It returns a pointer to the | |
245 | * buffer containing the on-disk inode in the bpp parameter, and in | |
246 | * the dip parameter it returns a pointer to the on-disk inode within | |
247 | * that buffer. | |
248 | * | |
249 | * If a non-zero error is returned, then the contents of bpp and | |
250 | * dipp are undefined. | |
251 | * | |
252 | * If the inode is new and has not yet been initialized, use xfs_imap() | |
253 | * to determine the size and location of the buffer to read from disk. | |
254 | * If the inode has already been mapped to its buffer and read in once, | |
255 | * then use the mapping information stored in the inode rather than | |
256 | * calling xfs_imap(). This allows us to avoid the overhead of looking | |
257 | * at the inode btree for small block file systems (see xfs_dilocate()). | |
258 | * We can tell whether the inode has been mapped in before by comparing | |
259 | * its disk block address to 0. Only uninitialized inodes will have | |
260 | * 0 for the disk block address. | |
261 | */ | |
262 | int | |
263 | xfs_itobp( | |
264 | xfs_mount_t *mp, | |
265 | xfs_trans_t *tp, | |
266 | xfs_inode_t *ip, | |
267 | xfs_dinode_t **dipp, | |
268 | xfs_buf_t **bpp, | |
269 | xfs_daddr_t bno) | |
270 | { | |
271 | xfs_buf_t *bp; | |
272 | int error; | |
273 | xfs_imap_t imap; | |
274 | #ifdef __KERNEL__ | |
275 | int i; | |
276 | int ni; | |
277 | #endif | |
278 | ||
279 | if (ip->i_blkno == (xfs_daddr_t)0) { | |
280 | /* | |
281 | * Call the space management code to find the location of the | |
282 | * inode on disk. | |
283 | */ | |
284 | imap.im_blkno = bno; | |
285 | error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP); | |
286 | if (error != 0) { | |
287 | return error; | |
288 | } | |
289 | ||
290 | /* | |
291 | * If the inode number maps to a block outside the bounds | |
292 | * of the file system then return NULL rather than calling | |
293 | * read_buf and panicing when we get an error from the | |
294 | * driver. | |
295 | */ | |
296 | if ((imap.im_blkno + imap.im_len) > | |
297 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
298 | #ifdef DEBUG | |
299 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
300 | "(imap.im_blkno (0x%llx) " | |
301 | "+ imap.im_len (0x%llx)) > " | |
302 | " XFS_FSB_TO_BB(mp, " | |
303 | "mp->m_sb.sb_dblocks) (0x%llx)", | |
304 | (unsigned long long) imap.im_blkno, | |
305 | (unsigned long long) imap.im_len, | |
306 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
307 | #endif /* DEBUG */ | |
308 | return XFS_ERROR(EINVAL); | |
309 | } | |
310 | ||
311 | /* | |
312 | * Fill in the fields in the inode that will be used to | |
313 | * map the inode to its buffer from now on. | |
314 | */ | |
315 | ip->i_blkno = imap.im_blkno; | |
316 | ip->i_len = imap.im_len; | |
317 | ip->i_boffset = imap.im_boffset; | |
318 | } else { | |
319 | /* | |
320 | * We've already mapped the inode once, so just use the | |
321 | * mapping that we saved the first time. | |
322 | */ | |
323 | imap.im_blkno = ip->i_blkno; | |
324 | imap.im_len = ip->i_len; | |
325 | imap.im_boffset = ip->i_boffset; | |
326 | } | |
327 | ASSERT(bno == 0 || bno == imap.im_blkno); | |
328 | ||
329 | /* | |
330 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
331 | * default to just a read_buf() call. | |
332 | */ | |
333 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
334 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
335 | ||
336 | if (error) { | |
337 | #ifdef DEBUG | |
338 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
339 | "xfs_trans_read_buf() returned error %d, " | |
340 | "imap.im_blkno 0x%llx, imap.im_len 0x%llx", | |
341 | error, (unsigned long long) imap.im_blkno, | |
342 | (unsigned long long) imap.im_len); | |
343 | #endif /* DEBUG */ | |
344 | return error; | |
345 | } | |
346 | #ifdef __KERNEL__ | |
347 | /* | |
348 | * Validate the magic number and version of every inode in the buffer | |
349 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
350 | */ | |
351 | #ifdef DEBUG | |
352 | ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog; | |
353 | #else | |
354 | ni = 1; | |
355 | #endif | |
356 | for (i = 0; i < ni; i++) { | |
357 | int di_ok; | |
358 | xfs_dinode_t *dip; | |
359 | ||
360 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
361 | (i << mp->m_sb.sb_inodelog)); | |
362 | di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && | |
363 | XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); | |
364 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP, | |
365 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
366 | #ifdef DEBUG | |
367 | prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)", | |
368 | mp->m_ddev_targp, | |
369 | (unsigned long long)imap.im_blkno, i, | |
370 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT)); | |
371 | #endif | |
372 | XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH, | |
373 | mp, dip); | |
374 | xfs_trans_brelse(tp, bp); | |
375 | return XFS_ERROR(EFSCORRUPTED); | |
376 | } | |
377 | } | |
378 | #endif /* __KERNEL__ */ | |
379 | ||
380 | xfs_inobp_check(mp, bp); | |
381 | ||
382 | /* | |
383 | * Mark the buffer as an inode buffer now that it looks good | |
384 | */ | |
385 | XFS_BUF_SET_VTYPE(bp, B_FS_INO); | |
386 | ||
387 | /* | |
388 | * Set *dipp to point to the on-disk inode in the buffer. | |
389 | */ | |
390 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
391 | *bpp = bp; | |
392 | return 0; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Move inode type and inode format specific information from the | |
397 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
398 | * this means set if_rdev to the proper value. For files, directories, | |
399 | * and symlinks this means to bring in the in-line data or extent | |
400 | * pointers. For a file in B-tree format, only the root is immediately | |
401 | * brought in-core. The rest will be in-lined in if_extents when it | |
402 | * is first referenced (see xfs_iread_extents()). | |
403 | */ | |
404 | STATIC int | |
405 | xfs_iformat( | |
406 | xfs_inode_t *ip, | |
407 | xfs_dinode_t *dip) | |
408 | { | |
409 | xfs_attr_shortform_t *atp; | |
410 | int size; | |
411 | int error; | |
412 | xfs_fsize_t di_size; | |
413 | ip->i_df.if_ext_max = | |
414 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
415 | error = 0; | |
416 | ||
417 | if (unlikely( | |
418 | INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) + | |
419 | INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) > | |
420 | INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) { | |
421 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
422 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu." | |
423 | " Unmount and run xfs_repair.", | |
424 | (unsigned long long)ip->i_ino, | |
425 | (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) | |
426 | + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)), | |
427 | (unsigned long long) | |
428 | INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT)); | |
429 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, | |
430 | ip->i_mount, dip); | |
431 | return XFS_ERROR(EFSCORRUPTED); | |
432 | } | |
433 | ||
434 | if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) { | |
435 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
436 | "corrupt dinode %Lu, forkoff = 0x%x." | |
437 | " Unmount and run xfs_repair.", | |
438 | (unsigned long long)ip->i_ino, | |
439 | (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT))); | |
440 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, | |
441 | ip->i_mount, dip); | |
442 | return XFS_ERROR(EFSCORRUPTED); | |
443 | } | |
444 | ||
445 | switch (ip->i_d.di_mode & S_IFMT) { | |
446 | case S_IFIFO: | |
447 | case S_IFCHR: | |
448 | case S_IFBLK: | |
449 | case S_IFSOCK: | |
450 | if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) { | |
451 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, | |
452 | ip->i_mount, dip); | |
453 | return XFS_ERROR(EFSCORRUPTED); | |
454 | } | |
455 | ip->i_d.di_size = 0; | |
456 | ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT); | |
457 | break; | |
458 | ||
459 | case S_IFREG: | |
460 | case S_IFLNK: | |
461 | case S_IFDIR: | |
462 | switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) { | |
463 | case XFS_DINODE_FMT_LOCAL: | |
464 | /* | |
465 | * no local regular files yet | |
466 | */ | |
467 | if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) { | |
468 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
469 | "corrupt inode (local format for regular file) %Lu. Unmount and run xfs_repair.", | |
470 | (unsigned long long) ip->i_ino); | |
471 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
472 | XFS_ERRLEVEL_LOW, | |
473 | ip->i_mount, dip); | |
474 | return XFS_ERROR(EFSCORRUPTED); | |
475 | } | |
476 | ||
477 | di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT); | |
478 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { | |
479 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
480 | "corrupt inode %Lu (bad size %Ld for local inode). Unmount and run xfs_repair.", | |
481 | (unsigned long long) ip->i_ino, | |
482 | (long long) di_size); | |
483 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
484 | XFS_ERRLEVEL_LOW, | |
485 | ip->i_mount, dip); | |
486 | return XFS_ERROR(EFSCORRUPTED); | |
487 | } | |
488 | ||
489 | size = (int)di_size; | |
490 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
491 | break; | |
492 | case XFS_DINODE_FMT_EXTENTS: | |
493 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
494 | break; | |
495 | case XFS_DINODE_FMT_BTREE: | |
496 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
497 | break; | |
498 | default: | |
499 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
500 | ip->i_mount); | |
501 | return XFS_ERROR(EFSCORRUPTED); | |
502 | } | |
503 | break; | |
504 | ||
505 | default: | |
506 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
507 | return XFS_ERROR(EFSCORRUPTED); | |
508 | } | |
509 | if (error) { | |
510 | return error; | |
511 | } | |
512 | if (!XFS_DFORK_Q(dip)) | |
513 | return 0; | |
514 | ASSERT(ip->i_afp == NULL); | |
515 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
516 | ip->i_afp->if_ext_max = | |
517 | XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
518 | switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) { | |
519 | case XFS_DINODE_FMT_LOCAL: | |
520 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
521 | size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT); | |
522 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); | |
523 | break; | |
524 | case XFS_DINODE_FMT_EXTENTS: | |
525 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
526 | break; | |
527 | case XFS_DINODE_FMT_BTREE: | |
528 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
529 | break; | |
530 | default: | |
531 | error = XFS_ERROR(EFSCORRUPTED); | |
532 | break; | |
533 | } | |
534 | if (error) { | |
535 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
536 | ip->i_afp = NULL; | |
537 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
538 | } | |
539 | return error; | |
540 | } | |
541 | ||
542 | /* | |
543 | * The file is in-lined in the on-disk inode. | |
544 | * If it fits into if_inline_data, then copy | |
545 | * it there, otherwise allocate a buffer for it | |
546 | * and copy the data there. Either way, set | |
547 | * if_data to point at the data. | |
548 | * If we allocate a buffer for the data, make | |
549 | * sure that its size is a multiple of 4 and | |
550 | * record the real size in i_real_bytes. | |
551 | */ | |
552 | STATIC int | |
553 | xfs_iformat_local( | |
554 | xfs_inode_t *ip, | |
555 | xfs_dinode_t *dip, | |
556 | int whichfork, | |
557 | int size) | |
558 | { | |
559 | xfs_ifork_t *ifp; | |
560 | int real_size; | |
561 | ||
562 | /* | |
563 | * If the size is unreasonable, then something | |
564 | * is wrong and we just bail out rather than crash in | |
565 | * kmem_alloc() or memcpy() below. | |
566 | */ | |
567 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
568 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
569 | "corrupt inode %Lu (bad size %d for local fork, size = %d). Unmount and run xfs_repair.", | |
570 | (unsigned long long) ip->i_ino, size, | |
571 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
572 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
573 | ip->i_mount, dip); | |
574 | return XFS_ERROR(EFSCORRUPTED); | |
575 | } | |
576 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
577 | real_size = 0; | |
578 | if (size == 0) | |
579 | ifp->if_u1.if_data = NULL; | |
580 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
581 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
582 | else { | |
583 | real_size = roundup(size, 4); | |
584 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
585 | } | |
586 | ifp->if_bytes = size; | |
587 | ifp->if_real_bytes = real_size; | |
588 | if (size) | |
589 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
590 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
591 | ifp->if_flags |= XFS_IFINLINE; | |
592 | return 0; | |
593 | } | |
594 | ||
595 | /* | |
596 | * The file consists of a set of extents all | |
597 | * of which fit into the on-disk inode. | |
598 | * If there are few enough extents to fit into | |
599 | * the if_inline_ext, then copy them there. | |
600 | * Otherwise allocate a buffer for them and copy | |
601 | * them into it. Either way, set if_extents | |
602 | * to point at the extents. | |
603 | */ | |
604 | STATIC int | |
605 | xfs_iformat_extents( | |
606 | xfs_inode_t *ip, | |
607 | xfs_dinode_t *dip, | |
608 | int whichfork) | |
609 | { | |
610 | xfs_bmbt_rec_t *ep, *dp; | |
611 | xfs_ifork_t *ifp; | |
612 | int nex; | |
613 | int real_size; | |
614 | int size; | |
615 | int i; | |
616 | ||
617 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
618 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
619 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
620 | ||
621 | /* | |
622 | * If the number of extents is unreasonable, then something | |
623 | * is wrong and we just bail out rather than crash in | |
624 | * kmem_alloc() or memcpy() below. | |
625 | */ | |
626 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
627 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
628 | "corrupt inode %Lu ((a)extents = %d). Unmount and run xfs_repair.", | |
629 | (unsigned long long) ip->i_ino, nex); | |
630 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
631 | ip->i_mount, dip); | |
632 | return XFS_ERROR(EFSCORRUPTED); | |
633 | } | |
634 | ||
635 | real_size = 0; | |
636 | if (nex == 0) | |
637 | ifp->if_u1.if_extents = NULL; | |
638 | else if (nex <= XFS_INLINE_EXTS) | |
639 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
640 | else { | |
641 | ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP); | |
642 | ASSERT(ifp->if_u1.if_extents != NULL); | |
643 | real_size = size; | |
644 | } | |
645 | ifp->if_bytes = size; | |
646 | ifp->if_real_bytes = real_size; | |
647 | if (size) { | |
648 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
649 | xfs_validate_extents(dp, nex, 1, XFS_EXTFMT_INODE(ip)); | |
650 | ep = ifp->if_u1.if_extents; | |
651 | for (i = 0; i < nex; i++, ep++, dp++) { | |
652 | ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0), | |
653 | ARCH_CONVERT); | |
654 | ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1), | |
655 | ARCH_CONVERT); | |
656 | } | |
657 | xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex, | |
658 | whichfork); | |
659 | if (whichfork != XFS_DATA_FORK || | |
660 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
661 | if (unlikely(xfs_check_nostate_extents( | |
662 | ifp->if_u1.if_extents, nex))) { | |
663 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", | |
664 | XFS_ERRLEVEL_LOW, | |
665 | ip->i_mount); | |
666 | return XFS_ERROR(EFSCORRUPTED); | |
667 | } | |
668 | } | |
669 | ifp->if_flags |= XFS_IFEXTENTS; | |
670 | return 0; | |
671 | } | |
672 | ||
673 | /* | |
674 | * The file has too many extents to fit into | |
675 | * the inode, so they are in B-tree format. | |
676 | * Allocate a buffer for the root of the B-tree | |
677 | * and copy the root into it. The i_extents | |
678 | * field will remain NULL until all of the | |
679 | * extents are read in (when they are needed). | |
680 | */ | |
681 | STATIC int | |
682 | xfs_iformat_btree( | |
683 | xfs_inode_t *ip, | |
684 | xfs_dinode_t *dip, | |
685 | int whichfork) | |
686 | { | |
687 | xfs_bmdr_block_t *dfp; | |
688 | xfs_ifork_t *ifp; | |
689 | /* REFERENCED */ | |
690 | int nrecs; | |
691 | int size; | |
692 | ||
693 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
694 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
695 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
696 | nrecs = XFS_BMAP_BROOT_NUMRECS(dfp); | |
697 | ||
698 | /* | |
699 | * blow out if -- fork has less extents than can fit in | |
700 | * fork (fork shouldn't be a btree format), root btree | |
701 | * block has more records than can fit into the fork, | |
702 | * or the number of extents is greater than the number of | |
703 | * blocks. | |
704 | */ | |
705 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max | |
706 | || XFS_BMDR_SPACE_CALC(nrecs) > | |
707 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) | |
708 | || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
709 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, | |
710 | "corrupt inode %Lu (btree). Unmount and run xfs_repair.", | |
711 | (unsigned long long) ip->i_ino); | |
712 | XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, | |
713 | ip->i_mount); | |
714 | return XFS_ERROR(EFSCORRUPTED); | |
715 | } | |
716 | ||
717 | ifp->if_broot_bytes = size; | |
718 | ifp->if_broot = kmem_alloc(size, KM_SLEEP); | |
719 | ASSERT(ifp->if_broot != NULL); | |
720 | /* | |
721 | * Copy and convert from the on-disk structure | |
722 | * to the in-memory structure. | |
723 | */ | |
724 | xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
725 | ifp->if_broot, size); | |
726 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
727 | ifp->if_flags |= XFS_IFBROOT; | |
728 | ||
729 | return 0; | |
730 | } | |
731 | ||
732 | /* | |
733 | * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk | |
734 | * and native format | |
735 | * | |
736 | * buf = on-disk representation | |
737 | * dip = native representation | |
738 | * dir = direction - +ve -> disk to native | |
739 | * -ve -> native to disk | |
740 | */ | |
741 | void | |
742 | xfs_xlate_dinode_core( | |
743 | xfs_caddr_t buf, | |
744 | xfs_dinode_core_t *dip, | |
745 | int dir) | |
746 | { | |
747 | xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf; | |
748 | xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip; | |
749 | xfs_arch_t arch = ARCH_CONVERT; | |
750 | ||
751 | ASSERT(dir); | |
752 | ||
753 | INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch); | |
754 | INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch); | |
755 | INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch); | |
756 | INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch); | |
757 | INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch); | |
758 | INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch); | |
759 | INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch); | |
760 | INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch); | |
761 | INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch); | |
762 | ||
763 | if (dir > 0) { | |
764 | memcpy(mem_core->di_pad, buf_core->di_pad, | |
765 | sizeof(buf_core->di_pad)); | |
766 | } else { | |
767 | memcpy(buf_core->di_pad, mem_core->di_pad, | |
768 | sizeof(buf_core->di_pad)); | |
769 | } | |
770 | ||
771 | INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch); | |
772 | ||
773 | INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec, | |
774 | dir, arch); | |
775 | INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec, | |
776 | dir, arch); | |
777 | INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec, | |
778 | dir, arch); | |
779 | INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec, | |
780 | dir, arch); | |
781 | INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec, | |
782 | dir, arch); | |
783 | INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec, | |
784 | dir, arch); | |
785 | INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch); | |
786 | INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch); | |
787 | INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch); | |
788 | INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch); | |
789 | INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch); | |
790 | INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch); | |
791 | INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch); | |
792 | INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch); | |
793 | INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch); | |
794 | INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch); | |
795 | INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch); | |
796 | } | |
797 | ||
798 | STATIC uint | |
799 | _xfs_dic2xflags( | |
800 | xfs_dinode_core_t *dic, | |
801 | __uint16_t di_flags) | |
802 | { | |
803 | uint flags = 0; | |
804 | ||
805 | if (di_flags & XFS_DIFLAG_ANY) { | |
806 | if (di_flags & XFS_DIFLAG_REALTIME) | |
807 | flags |= XFS_XFLAG_REALTIME; | |
808 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
809 | flags |= XFS_XFLAG_PREALLOC; | |
810 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
811 | flags |= XFS_XFLAG_IMMUTABLE; | |
812 | if (di_flags & XFS_DIFLAG_APPEND) | |
813 | flags |= XFS_XFLAG_APPEND; | |
814 | if (di_flags & XFS_DIFLAG_SYNC) | |
815 | flags |= XFS_XFLAG_SYNC; | |
816 | if (di_flags & XFS_DIFLAG_NOATIME) | |
817 | flags |= XFS_XFLAG_NOATIME; | |
818 | if (di_flags & XFS_DIFLAG_NODUMP) | |
819 | flags |= XFS_XFLAG_NODUMP; | |
820 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
821 | flags |= XFS_XFLAG_RTINHERIT; | |
822 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
823 | flags |= XFS_XFLAG_PROJINHERIT; | |
824 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
825 | flags |= XFS_XFLAG_NOSYMLINKS; | |
826 | } | |
827 | ||
828 | return flags; | |
829 | } | |
830 | ||
831 | uint | |
832 | xfs_ip2xflags( | |
833 | xfs_inode_t *ip) | |
834 | { | |
835 | xfs_dinode_core_t *dic = &ip->i_d; | |
836 | ||
837 | return _xfs_dic2xflags(dic, dic->di_flags) | | |
838 | (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0); | |
839 | } | |
840 | ||
841 | uint | |
842 | xfs_dic2xflags( | |
843 | xfs_dinode_core_t *dic) | |
844 | { | |
845 | return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) | | |
846 | (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0); | |
847 | } | |
848 | ||
849 | /* | |
850 | * Given a mount structure and an inode number, return a pointer | |
851 | * to a newly allocated in-core inode coresponding to the given | |
852 | * inode number. | |
853 | * | |
854 | * Initialize the inode's attributes and extent pointers if it | |
855 | * already has them (it will not if the inode has no links). | |
856 | */ | |
857 | int | |
858 | xfs_iread( | |
859 | xfs_mount_t *mp, | |
860 | xfs_trans_t *tp, | |
861 | xfs_ino_t ino, | |
862 | xfs_inode_t **ipp, | |
863 | xfs_daddr_t bno) | |
864 | { | |
865 | xfs_buf_t *bp; | |
866 | xfs_dinode_t *dip; | |
867 | xfs_inode_t *ip; | |
868 | int error; | |
869 | ||
870 | ASSERT(xfs_inode_zone != NULL); | |
871 | ||
872 | ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP); | |
873 | ip->i_ino = ino; | |
874 | ip->i_mount = mp; | |
875 | ||
876 | /* | |
877 | * Get pointer's to the on-disk inode and the buffer containing it. | |
878 | * If the inode number refers to a block outside the file system | |
879 | * then xfs_itobp() will return NULL. In this case we should | |
880 | * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will | |
881 | * know that this is a new incore inode. | |
882 | */ | |
883 | error = xfs_itobp(mp, tp, ip, &dip, &bp, bno); | |
884 | ||
885 | if (error != 0) { | |
886 | kmem_zone_free(xfs_inode_zone, ip); | |
887 | return error; | |
888 | } | |
889 | ||
890 | /* | |
891 | * Initialize inode's trace buffers. | |
892 | * Do this before xfs_iformat in case it adds entries. | |
893 | */ | |
894 | #ifdef XFS_BMAP_TRACE | |
895 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP); | |
896 | #endif | |
897 | #ifdef XFS_BMBT_TRACE | |
898 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP); | |
899 | #endif | |
900 | #ifdef XFS_RW_TRACE | |
901 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP); | |
902 | #endif | |
903 | #ifdef XFS_ILOCK_TRACE | |
904 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP); | |
905 | #endif | |
906 | #ifdef XFS_DIR2_TRACE | |
907 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP); | |
908 | #endif | |
909 | ||
910 | /* | |
911 | * If we got something that isn't an inode it means someone | |
912 | * (nfs or dmi) has a stale handle. | |
913 | */ | |
914 | if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) { | |
915 | kmem_zone_free(xfs_inode_zone, ip); | |
916 | xfs_trans_brelse(tp, bp); | |
917 | #ifdef DEBUG | |
918 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
919 | "dip->di_core.di_magic (0x%x) != " | |
920 | "XFS_DINODE_MAGIC (0x%x)", | |
921 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT), | |
922 | XFS_DINODE_MAGIC); | |
923 | #endif /* DEBUG */ | |
924 | return XFS_ERROR(EINVAL); | |
925 | } | |
926 | ||
927 | /* | |
928 | * If the on-disk inode is already linked to a directory | |
929 | * entry, copy all of the inode into the in-core inode. | |
930 | * xfs_iformat() handles copying in the inode format | |
931 | * specific information. | |
932 | * Otherwise, just get the truly permanent information. | |
933 | */ | |
934 | if (dip->di_core.di_mode) { | |
935 | xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core, | |
936 | &(ip->i_d), 1); | |
937 | error = xfs_iformat(ip, dip); | |
938 | if (error) { | |
939 | kmem_zone_free(xfs_inode_zone, ip); | |
940 | xfs_trans_brelse(tp, bp); | |
941 | #ifdef DEBUG | |
942 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
943 | "xfs_iformat() returned error %d", | |
944 | error); | |
945 | #endif /* DEBUG */ | |
946 | return error; | |
947 | } | |
948 | } else { | |
949 | ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT); | |
950 | ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT); | |
951 | ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT); | |
952 | ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT); | |
953 | /* | |
954 | * Make sure to pull in the mode here as well in | |
955 | * case the inode is released without being used. | |
956 | * This ensures that xfs_inactive() will see that | |
957 | * the inode is already free and not try to mess | |
958 | * with the uninitialized part of it. | |
959 | */ | |
960 | ip->i_d.di_mode = 0; | |
961 | /* | |
962 | * Initialize the per-fork minima and maxima for a new | |
963 | * inode here. xfs_iformat will do it for old inodes. | |
964 | */ | |
965 | ip->i_df.if_ext_max = | |
966 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
967 | } | |
968 | ||
969 | INIT_LIST_HEAD(&ip->i_reclaim); | |
970 | ||
971 | /* | |
972 | * The inode format changed when we moved the link count and | |
973 | * made it 32 bits long. If this is an old format inode, | |
974 | * convert it in memory to look like a new one. If it gets | |
975 | * flushed to disk we will convert back before flushing or | |
976 | * logging it. We zero out the new projid field and the old link | |
977 | * count field. We'll handle clearing the pad field (the remains | |
978 | * of the old uuid field) when we actually convert the inode to | |
979 | * the new format. We don't change the version number so that we | |
980 | * can distinguish this from a real new format inode. | |
981 | */ | |
982 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
983 | ip->i_d.di_nlink = ip->i_d.di_onlink; | |
984 | ip->i_d.di_onlink = 0; | |
985 | ip->i_d.di_projid = 0; | |
986 | } | |
987 | ||
988 | ip->i_delayed_blks = 0; | |
989 | ||
990 | /* | |
991 | * Mark the buffer containing the inode as something to keep | |
992 | * around for a while. This helps to keep recently accessed | |
993 | * meta-data in-core longer. | |
994 | */ | |
995 | XFS_BUF_SET_REF(bp, XFS_INO_REF); | |
996 | ||
997 | /* | |
998 | * Use xfs_trans_brelse() to release the buffer containing the | |
999 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
1000 | * in xfs_itobp() above. If tp is NULL, this is just a normal | |
1001 | * brelse(). If we're within a transaction, then xfs_trans_brelse() | |
1002 | * will only release the buffer if it is not dirty within the | |
1003 | * transaction. It will be OK to release the buffer in this case, | |
1004 | * because inodes on disk are never destroyed and we will be | |
1005 | * locking the new in-core inode before putting it in the hash | |
1006 | * table where other processes can find it. Thus we don't have | |
1007 | * to worry about the inode being changed just because we released | |
1008 | * the buffer. | |
1009 | */ | |
1010 | xfs_trans_brelse(tp, bp); | |
1011 | *ipp = ip; | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Read in extents from a btree-format inode. | |
1017 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1018 | */ | |
1019 | int | |
1020 | xfs_iread_extents( | |
1021 | xfs_trans_t *tp, | |
1022 | xfs_inode_t *ip, | |
1023 | int whichfork) | |
1024 | { | |
1025 | int error; | |
1026 | xfs_ifork_t *ifp; | |
1027 | size_t size; | |
1028 | ||
1029 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1030 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1031 | ip->i_mount); | |
1032 | return XFS_ERROR(EFSCORRUPTED); | |
1033 | } | |
1034 | size = XFS_IFORK_NEXTENTS(ip, whichfork) * (uint)sizeof(xfs_bmbt_rec_t); | |
1035 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1036 | /* | |
1037 | * We know that the size is valid (it's checked in iformat_btree) | |
1038 | */ | |
1039 | ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP); | |
1040 | ASSERT(ifp->if_u1.if_extents != NULL); | |
1041 | ifp->if_lastex = NULLEXTNUM; | |
1042 | ifp->if_bytes = ifp->if_real_bytes = (int)size; | |
1043 | ifp->if_flags |= XFS_IFEXTENTS; | |
1044 | error = xfs_bmap_read_extents(tp, ip, whichfork); | |
1045 | if (error) { | |
1046 | kmem_free(ifp->if_u1.if_extents, size); | |
1047 | ifp->if_u1.if_extents = NULL; | |
1048 | ifp->if_bytes = ifp->if_real_bytes = 0; | |
1049 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
1050 | return error; | |
1051 | } | |
1052 | xfs_validate_extents((xfs_bmbt_rec_t *)ifp->if_u1.if_extents, | |
1053 | XFS_IFORK_NEXTENTS(ip, whichfork), 0, XFS_EXTFMT_INODE(ip)); | |
1054 | return 0; | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Allocate an inode on disk and return a copy of its in-core version. | |
1059 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1060 | * appropriately within the inode. The uid and gid for the inode are | |
1061 | * set according to the contents of the given cred structure. | |
1062 | * | |
1063 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
1064 | * has a free inode available, call xfs_iget() | |
1065 | * to obtain the in-core version of the allocated inode. Finally, | |
1066 | * fill in the inode and log its initial contents. In this case, | |
1067 | * ialloc_context would be set to NULL and call_again set to false. | |
1068 | * | |
1069 | * If xfs_dialloc() does not have an available inode, | |
1070 | * it will replenish its supply by doing an allocation. Since we can | |
1071 | * only do one allocation within a transaction without deadlocks, we | |
1072 | * must commit the current transaction before returning the inode itself. | |
1073 | * In this case, therefore, we will set call_again to true and return. | |
1074 | * The caller should then commit the current transaction, start a new | |
1075 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1076 | * | |
1077 | * To ensure that some other process does not grab the inode that | |
1078 | * was allocated during the first call to xfs_ialloc(), this routine | |
1079 | * also returns the [locked] bp pointing to the head of the freelist | |
1080 | * as ialloc_context. The caller should hold this buffer across | |
1081 | * the commit and pass it back into this routine on the second call. | |
1082 | */ | |
1083 | int | |
1084 | xfs_ialloc( | |
1085 | xfs_trans_t *tp, | |
1086 | xfs_inode_t *pip, | |
1087 | mode_t mode, | |
31b084ae | 1088 | xfs_nlink_t nlink, |
1da177e4 LT |
1089 | xfs_dev_t rdev, |
1090 | cred_t *cr, | |
1091 | xfs_prid_t prid, | |
1092 | int okalloc, | |
1093 | xfs_buf_t **ialloc_context, | |
1094 | boolean_t *call_again, | |
1095 | xfs_inode_t **ipp) | |
1096 | { | |
1097 | xfs_ino_t ino; | |
1098 | xfs_inode_t *ip; | |
1099 | vnode_t *vp; | |
1100 | uint flags; | |
1101 | int error; | |
1102 | ||
1103 | /* | |
1104 | * Call the space management code to pick | |
1105 | * the on-disk inode to be allocated. | |
1106 | */ | |
1107 | error = xfs_dialloc(tp, pip->i_ino, mode, okalloc, | |
1108 | ialloc_context, call_again, &ino); | |
1109 | if (error != 0) { | |
1110 | return error; | |
1111 | } | |
1112 | if (*call_again || ino == NULLFSINO) { | |
1113 | *ipp = NULL; | |
1114 | return 0; | |
1115 | } | |
1116 | ASSERT(*ialloc_context == NULL); | |
1117 | ||
1118 | /* | |
1119 | * Get the in-core inode with the lock held exclusively. | |
1120 | * This is because we're setting fields here we need | |
1121 | * to prevent others from looking at until we're done. | |
1122 | */ | |
1123 | error = xfs_trans_iget(tp->t_mountp, tp, ino, | |
1124 | IGET_CREATE, XFS_ILOCK_EXCL, &ip); | |
1125 | if (error != 0) { | |
1126 | return error; | |
1127 | } | |
1128 | ASSERT(ip != NULL); | |
1129 | ||
1130 | vp = XFS_ITOV(ip); | |
1da177e4 LT |
1131 | ip->i_d.di_mode = (__uint16_t)mode; |
1132 | ip->i_d.di_onlink = 0; | |
1133 | ip->i_d.di_nlink = nlink; | |
1134 | ASSERT(ip->i_d.di_nlink == nlink); | |
1135 | ip->i_d.di_uid = current_fsuid(cr); | |
1136 | ip->i_d.di_gid = current_fsgid(cr); | |
1137 | ip->i_d.di_projid = prid; | |
1138 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
1139 | ||
1140 | /* | |
1141 | * If the superblock version is up to where we support new format | |
1142 | * inodes and this is currently an old format inode, then change | |
1143 | * the inode version number now. This way we only do the conversion | |
1144 | * here rather than here and in the flush/logging code. | |
1145 | */ | |
1146 | if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) && | |
1147 | ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
1148 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
1149 | /* | |
1150 | * We've already zeroed the old link count, the projid field, | |
1151 | * and the pad field. | |
1152 | */ | |
1153 | } | |
1154 | ||
1155 | /* | |
1156 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1157 | */ | |
1158 | if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) | |
1159 | xfs_bump_ino_vers2(tp, ip); | |
1160 | ||
1161 | if (XFS_INHERIT_GID(pip, vp->v_vfsp)) { | |
1162 | ip->i_d.di_gid = pip->i_d.di_gid; | |
1163 | if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) { | |
1164 | ip->i_d.di_mode |= S_ISGID; | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * If the group ID of the new file does not match the effective group | |
1170 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1171 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1172 | */ | |
1173 | if ((irix_sgid_inherit) && | |
1174 | (ip->i_d.di_mode & S_ISGID) && | |
1175 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1176 | ip->i_d.di_mode &= ~S_ISGID; | |
1177 | } | |
1178 | ||
1179 | ip->i_d.di_size = 0; | |
1180 | ip->i_d.di_nextents = 0; | |
1181 | ASSERT(ip->i_d.di_nblocks == 0); | |
1182 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD); | |
1183 | /* | |
1184 | * di_gen will have been taken care of in xfs_iread. | |
1185 | */ | |
1186 | ip->i_d.di_extsize = 0; | |
1187 | ip->i_d.di_dmevmask = 0; | |
1188 | ip->i_d.di_dmstate = 0; | |
1189 | ip->i_d.di_flags = 0; | |
1190 | flags = XFS_ILOG_CORE; | |
1191 | switch (mode & S_IFMT) { | |
1192 | case S_IFIFO: | |
1193 | case S_IFCHR: | |
1194 | case S_IFBLK: | |
1195 | case S_IFSOCK: | |
1196 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1197 | ip->i_df.if_u2.if_rdev = rdev; | |
1198 | ip->i_df.if_flags = 0; | |
1199 | flags |= XFS_ILOG_DEV; | |
1200 | break; | |
1201 | case S_IFREG: | |
1202 | case S_IFDIR: | |
1203 | if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) { | |
365ca83d NS |
1204 | uint di_flags = 0; |
1205 | ||
1206 | if ((mode & S_IFMT) == S_IFDIR) { | |
1207 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) | |
1208 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
1209 | } else { | |
1210 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) { | |
1211 | di_flags |= XFS_DIFLAG_REALTIME; | |
1da177e4 LT |
1212 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; |
1213 | } | |
1214 | } | |
1215 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1216 | xfs_inherit_noatime) | |
365ca83d | 1217 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1218 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1219 | xfs_inherit_nodump) | |
365ca83d | 1220 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1221 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1222 | xfs_inherit_sync) | |
365ca83d | 1223 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1224 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1225 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1226 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1227 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1228 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
1229 | ip->i_d.di_flags |= di_flags; | |
1da177e4 LT |
1230 | } |
1231 | /* FALLTHROUGH */ | |
1232 | case S_IFLNK: | |
1233 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1234 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1235 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1236 | ip->i_df.if_u1.if_extents = NULL; | |
1237 | break; | |
1238 | default: | |
1239 | ASSERT(0); | |
1240 | } | |
1241 | /* | |
1242 | * Attribute fork settings for new inode. | |
1243 | */ | |
1244 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1245 | ip->i_d.di_anextents = 0; | |
1246 | ||
1247 | /* | |
1248 | * Log the new values stuffed into the inode. | |
1249 | */ | |
1250 | xfs_trans_log_inode(tp, ip, flags); | |
1251 | ||
0432dab2 | 1252 | /* now that we have an i_mode we can set Linux inode ops (& unlock) */ |
1da177e4 LT |
1253 | VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1); |
1254 | ||
1255 | *ipp = ip; | |
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Check to make sure that there are no blocks allocated to the | |
1261 | * file beyond the size of the file. We don't check this for | |
1262 | * files with fixed size extents or real time extents, but we | |
1263 | * at least do it for regular files. | |
1264 | */ | |
1265 | #ifdef DEBUG | |
1266 | void | |
1267 | xfs_isize_check( | |
1268 | xfs_mount_t *mp, | |
1269 | xfs_inode_t *ip, | |
1270 | xfs_fsize_t isize) | |
1271 | { | |
1272 | xfs_fileoff_t map_first; | |
1273 | int nimaps; | |
1274 | xfs_bmbt_irec_t imaps[2]; | |
1275 | ||
1276 | if ((ip->i_d.di_mode & S_IFMT) != S_IFREG) | |
1277 | return; | |
1278 | ||
1279 | if ( ip->i_d.di_flags & XFS_DIFLAG_REALTIME ) | |
1280 | return; | |
1281 | ||
1282 | nimaps = 2; | |
1283 | map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
1284 | /* | |
1285 | * The filesystem could be shutting down, so bmapi may return | |
1286 | * an error. | |
1287 | */ | |
1288 | if (xfs_bmapi(NULL, ip, map_first, | |
1289 | (XFS_B_TO_FSB(mp, | |
1290 | (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) - | |
1291 | map_first), | |
1292 | XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps, | |
1293 | NULL)) | |
1294 | return; | |
1295 | ASSERT(nimaps == 1); | |
1296 | ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK); | |
1297 | } | |
1298 | #endif /* DEBUG */ | |
1299 | ||
1300 | /* | |
1301 | * Calculate the last possible buffered byte in a file. This must | |
1302 | * include data that was buffered beyond the EOF by the write code. | |
1303 | * This also needs to deal with overflowing the xfs_fsize_t type | |
1304 | * which can happen for sizes near the limit. | |
1305 | * | |
1306 | * We also need to take into account any blocks beyond the EOF. It | |
1307 | * may be the case that they were buffered by a write which failed. | |
1308 | * In that case the pages will still be in memory, but the inode size | |
1309 | * will never have been updated. | |
1310 | */ | |
1311 | xfs_fsize_t | |
1312 | xfs_file_last_byte( | |
1313 | xfs_inode_t *ip) | |
1314 | { | |
1315 | xfs_mount_t *mp; | |
1316 | xfs_fsize_t last_byte; | |
1317 | xfs_fileoff_t last_block; | |
1318 | xfs_fileoff_t size_last_block; | |
1319 | int error; | |
1320 | ||
1321 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS)); | |
1322 | ||
1323 | mp = ip->i_mount; | |
1324 | /* | |
1325 | * Only check for blocks beyond the EOF if the extents have | |
1326 | * been read in. This eliminates the need for the inode lock, | |
1327 | * and it also saves us from looking when it really isn't | |
1328 | * necessary. | |
1329 | */ | |
1330 | if (ip->i_df.if_flags & XFS_IFEXTENTS) { | |
1331 | error = xfs_bmap_last_offset(NULL, ip, &last_block, | |
1332 | XFS_DATA_FORK); | |
1333 | if (error) { | |
1334 | last_block = 0; | |
1335 | } | |
1336 | } else { | |
1337 | last_block = 0; | |
1338 | } | |
1339 | size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size); | |
1340 | last_block = XFS_FILEOFF_MAX(last_block, size_last_block); | |
1341 | ||
1342 | last_byte = XFS_FSB_TO_B(mp, last_block); | |
1343 | if (last_byte < 0) { | |
1344 | return XFS_MAXIOFFSET(mp); | |
1345 | } | |
1346 | last_byte += (1 << mp->m_writeio_log); | |
1347 | if (last_byte < 0) { | |
1348 | return XFS_MAXIOFFSET(mp); | |
1349 | } | |
1350 | return last_byte; | |
1351 | } | |
1352 | ||
1353 | #if defined(XFS_RW_TRACE) | |
1354 | STATIC void | |
1355 | xfs_itrunc_trace( | |
1356 | int tag, | |
1357 | xfs_inode_t *ip, | |
1358 | int flag, | |
1359 | xfs_fsize_t new_size, | |
1360 | xfs_off_t toss_start, | |
1361 | xfs_off_t toss_finish) | |
1362 | { | |
1363 | if (ip->i_rwtrace == NULL) { | |
1364 | return; | |
1365 | } | |
1366 | ||
1367 | ktrace_enter(ip->i_rwtrace, | |
1368 | (void*)((long)tag), | |
1369 | (void*)ip, | |
1370 | (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff), | |
1371 | (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff), | |
1372 | (void*)((long)flag), | |
1373 | (void*)(unsigned long)((new_size >> 32) & 0xffffffff), | |
1374 | (void*)(unsigned long)(new_size & 0xffffffff), | |
1375 | (void*)(unsigned long)((toss_start >> 32) & 0xffffffff), | |
1376 | (void*)(unsigned long)(toss_start & 0xffffffff), | |
1377 | (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff), | |
1378 | (void*)(unsigned long)(toss_finish & 0xffffffff), | |
1379 | (void*)(unsigned long)current_cpu(), | |
1380 | (void*)0, | |
1381 | (void*)0, | |
1382 | (void*)0, | |
1383 | (void*)0); | |
1384 | } | |
1385 | #else | |
1386 | #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish) | |
1387 | #endif | |
1388 | ||
1389 | /* | |
1390 | * Start the truncation of the file to new_size. The new size | |
1391 | * must be smaller than the current size. This routine will | |
1392 | * clear the buffer and page caches of file data in the removed | |
1393 | * range, and xfs_itruncate_finish() will remove the underlying | |
1394 | * disk blocks. | |
1395 | * | |
1396 | * The inode must have its I/O lock locked EXCLUSIVELY, and it | |
1397 | * must NOT have the inode lock held at all. This is because we're | |
1398 | * calling into the buffer/page cache code and we can't hold the | |
1399 | * inode lock when we do so. | |
1400 | * | |
1401 | * The flags parameter can have either the value XFS_ITRUNC_DEFINITE | |
1402 | * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used | |
1403 | * in the case that the caller is locking things out of order and | |
1404 | * may not be able to call xfs_itruncate_finish() with the inode lock | |
1405 | * held without dropping the I/O lock. If the caller must drop the | |
1406 | * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start() | |
1407 | * must be called again with all the same restrictions as the initial | |
1408 | * call. | |
1409 | */ | |
1410 | void | |
1411 | xfs_itruncate_start( | |
1412 | xfs_inode_t *ip, | |
1413 | uint flags, | |
1414 | xfs_fsize_t new_size) | |
1415 | { | |
1416 | xfs_fsize_t last_byte; | |
1417 | xfs_off_t toss_start; | |
1418 | xfs_mount_t *mp; | |
1419 | vnode_t *vp; | |
1420 | ||
1421 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
1422 | ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size)); | |
1423 | ASSERT((flags == XFS_ITRUNC_DEFINITE) || | |
1424 | (flags == XFS_ITRUNC_MAYBE)); | |
1425 | ||
1426 | mp = ip->i_mount; | |
1427 | vp = XFS_ITOV(ip); | |
1428 | /* | |
1429 | * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers | |
1430 | * overlapping the region being removed. We have to use | |
1431 | * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the | |
1432 | * caller may not be able to finish the truncate without | |
1433 | * dropping the inode's I/O lock. Make sure | |
1434 | * to catch any pages brought in by buffers overlapping | |
1435 | * the EOF by searching out beyond the isize by our | |
1436 | * block size. We round new_size up to a block boundary | |
1437 | * so that we don't toss things on the same block as | |
1438 | * new_size but before it. | |
1439 | * | |
1440 | * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to | |
1441 | * call remapf() over the same region if the file is mapped. | |
1442 | * This frees up mapped file references to the pages in the | |
1443 | * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures | |
1444 | * that we get the latest mapped changes flushed out. | |
1445 | */ | |
1446 | toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1447 | toss_start = XFS_FSB_TO_B(mp, toss_start); | |
1448 | if (toss_start < 0) { | |
1449 | /* | |
1450 | * The place to start tossing is beyond our maximum | |
1451 | * file size, so there is no way that the data extended | |
1452 | * out there. | |
1453 | */ | |
1454 | return; | |
1455 | } | |
1456 | last_byte = xfs_file_last_byte(ip); | |
1457 | xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start, | |
1458 | last_byte); | |
1459 | if (last_byte > toss_start) { | |
1460 | if (flags & XFS_ITRUNC_DEFINITE) { | |
1461 | VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED); | |
1462 | } else { | |
1463 | VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED); | |
1464 | } | |
1465 | } | |
1466 | ||
1467 | #ifdef DEBUG | |
1468 | if (new_size == 0) { | |
1469 | ASSERT(VN_CACHED(vp) == 0); | |
1470 | } | |
1471 | #endif | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * Shrink the file to the given new_size. The new | |
1476 | * size must be smaller than the current size. | |
1477 | * This will free up the underlying blocks | |
1478 | * in the removed range after a call to xfs_itruncate_start() | |
1479 | * or xfs_atruncate_start(). | |
1480 | * | |
1481 | * The transaction passed to this routine must have made | |
1482 | * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES. | |
1483 | * This routine may commit the given transaction and | |
1484 | * start new ones, so make sure everything involved in | |
1485 | * the transaction is tidy before calling here. | |
1486 | * Some transaction will be returned to the caller to be | |
1487 | * committed. The incoming transaction must already include | |
1488 | * the inode, and both inode locks must be held exclusively. | |
1489 | * The inode must also be "held" within the transaction. On | |
1490 | * return the inode will be "held" within the returned transaction. | |
1491 | * This routine does NOT require any disk space to be reserved | |
1492 | * for it within the transaction. | |
1493 | * | |
1494 | * The fork parameter must be either xfs_attr_fork or xfs_data_fork, | |
1495 | * and it indicates the fork which is to be truncated. For the | |
1496 | * attribute fork we only support truncation to size 0. | |
1497 | * | |
1498 | * We use the sync parameter to indicate whether or not the first | |
1499 | * transaction we perform might have to be synchronous. For the attr fork, | |
1500 | * it needs to be so if the unlink of the inode is not yet known to be | |
1501 | * permanent in the log. This keeps us from freeing and reusing the | |
1502 | * blocks of the attribute fork before the unlink of the inode becomes | |
1503 | * permanent. | |
1504 | * | |
1505 | * For the data fork, we normally have to run synchronously if we're | |
1506 | * being called out of the inactive path or we're being called | |
1507 | * out of the create path where we're truncating an existing file. | |
1508 | * Either way, the truncate needs to be sync so blocks don't reappear | |
1509 | * in the file with altered data in case of a crash. wsync filesystems | |
1510 | * can run the first case async because anything that shrinks the inode | |
1511 | * has to run sync so by the time we're called here from inactive, the | |
1512 | * inode size is permanently set to 0. | |
1513 | * | |
1514 | * Calls from the truncate path always need to be sync unless we're | |
1515 | * in a wsync filesystem and the file has already been unlinked. | |
1516 | * | |
1517 | * The caller is responsible for correctly setting the sync parameter. | |
1518 | * It gets too hard for us to guess here which path we're being called | |
1519 | * out of just based on inode state. | |
1520 | */ | |
1521 | int | |
1522 | xfs_itruncate_finish( | |
1523 | xfs_trans_t **tp, | |
1524 | xfs_inode_t *ip, | |
1525 | xfs_fsize_t new_size, | |
1526 | int fork, | |
1527 | int sync) | |
1528 | { | |
1529 | xfs_fsblock_t first_block; | |
1530 | xfs_fileoff_t first_unmap_block; | |
1531 | xfs_fileoff_t last_block; | |
1532 | xfs_filblks_t unmap_len=0; | |
1533 | xfs_mount_t *mp; | |
1534 | xfs_trans_t *ntp; | |
1535 | int done; | |
1536 | int committed; | |
1537 | xfs_bmap_free_t free_list; | |
1538 | int error; | |
1539 | ||
1540 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
1541 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0); | |
1542 | ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size)); | |
1543 | ASSERT(*tp != NULL); | |
1544 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1545 | ASSERT(ip->i_transp == *tp); | |
1546 | ASSERT(ip->i_itemp != NULL); | |
1547 | ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD); | |
1548 | ||
1549 | ||
1550 | ntp = *tp; | |
1551 | mp = (ntp)->t_mountp; | |
1552 | ASSERT(! XFS_NOT_DQATTACHED(mp, ip)); | |
1553 | ||
1554 | /* | |
1555 | * We only support truncating the entire attribute fork. | |
1556 | */ | |
1557 | if (fork == XFS_ATTR_FORK) { | |
1558 | new_size = 0LL; | |
1559 | } | |
1560 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1561 | xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0); | |
1562 | /* | |
1563 | * The first thing we do is set the size to new_size permanently | |
1564 | * on disk. This way we don't have to worry about anyone ever | |
1565 | * being able to look at the data being freed even in the face | |
1566 | * of a crash. What we're getting around here is the case where | |
1567 | * we free a block, it is allocated to another file, it is written | |
1568 | * to, and then we crash. If the new data gets written to the | |
1569 | * file but the log buffers containing the free and reallocation | |
1570 | * don't, then we'd end up with garbage in the blocks being freed. | |
1571 | * As long as we make the new_size permanent before actually | |
1572 | * freeing any blocks it doesn't matter if they get writtten to. | |
1573 | * | |
1574 | * The callers must signal into us whether or not the size | |
1575 | * setting here must be synchronous. There are a few cases | |
1576 | * where it doesn't have to be synchronous. Those cases | |
1577 | * occur if the file is unlinked and we know the unlink is | |
1578 | * permanent or if the blocks being truncated are guaranteed | |
1579 | * to be beyond the inode eof (regardless of the link count) | |
1580 | * and the eof value is permanent. Both of these cases occur | |
1581 | * only on wsync-mounted filesystems. In those cases, we're | |
1582 | * guaranteed that no user will ever see the data in the blocks | |
1583 | * that are being truncated so the truncate can run async. | |
1584 | * In the free beyond eof case, the file may wind up with | |
1585 | * more blocks allocated to it than it needs if we crash | |
1586 | * and that won't get fixed until the next time the file | |
1587 | * is re-opened and closed but that's ok as that shouldn't | |
1588 | * be too many blocks. | |
1589 | * | |
1590 | * However, we can't just make all wsync xactions run async | |
1591 | * because there's one call out of the create path that needs | |
1592 | * to run sync where it's truncating an existing file to size | |
1593 | * 0 whose size is > 0. | |
1594 | * | |
1595 | * It's probably possible to come up with a test in this | |
1596 | * routine that would correctly distinguish all the above | |
1597 | * cases from the values of the function parameters and the | |
1598 | * inode state but for sanity's sake, I've decided to let the | |
1599 | * layers above just tell us. It's simpler to correctly figure | |
1600 | * out in the layer above exactly under what conditions we | |
1601 | * can run async and I think it's easier for others read and | |
1602 | * follow the logic in case something has to be changed. | |
1603 | * cscope is your friend -- rcc. | |
1604 | * | |
1605 | * The attribute fork is much simpler. | |
1606 | * | |
1607 | * For the attribute fork we allow the caller to tell us whether | |
1608 | * the unlink of the inode that led to this call is yet permanent | |
1609 | * in the on disk log. If it is not and we will be freeing extents | |
1610 | * in this inode then we make the first transaction synchronous | |
1611 | * to make sure that the unlink is permanent by the time we free | |
1612 | * the blocks. | |
1613 | */ | |
1614 | if (fork == XFS_DATA_FORK) { | |
1615 | if (ip->i_d.di_nextents > 0) { | |
1616 | ip->i_d.di_size = new_size; | |
1617 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1618 | } | |
1619 | } else if (sync) { | |
1620 | ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC)); | |
1621 | if (ip->i_d.di_anextents > 0) | |
1622 | xfs_trans_set_sync(ntp); | |
1623 | } | |
1624 | ASSERT(fork == XFS_DATA_FORK || | |
1625 | (fork == XFS_ATTR_FORK && | |
1626 | ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) || | |
1627 | (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC))))); | |
1628 | ||
1629 | /* | |
1630 | * Since it is possible for space to become allocated beyond | |
1631 | * the end of the file (in a crash where the space is allocated | |
1632 | * but the inode size is not yet updated), simply remove any | |
1633 | * blocks which show up between the new EOF and the maximum | |
1634 | * possible file size. If the first block to be removed is | |
1635 | * beyond the maximum file size (ie it is the same as last_block), | |
1636 | * then there is nothing to do. | |
1637 | */ | |
1638 | last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1639 | ASSERT(first_unmap_block <= last_block); | |
1640 | done = 0; | |
1641 | if (last_block == first_unmap_block) { | |
1642 | done = 1; | |
1643 | } else { | |
1644 | unmap_len = last_block - first_unmap_block + 1; | |
1645 | } | |
1646 | while (!done) { | |
1647 | /* | |
1648 | * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi() | |
1649 | * will tell us whether it freed the entire range or | |
1650 | * not. If this is a synchronous mount (wsync), | |
1651 | * then we can tell bunmapi to keep all the | |
1652 | * transactions asynchronous since the unlink | |
1653 | * transaction that made this inode inactive has | |
1654 | * already hit the disk. There's no danger of | |
1655 | * the freed blocks being reused, there being a | |
1656 | * crash, and the reused blocks suddenly reappearing | |
1657 | * in this file with garbage in them once recovery | |
1658 | * runs. | |
1659 | */ | |
1660 | XFS_BMAP_INIT(&free_list, &first_block); | |
1661 | error = xfs_bunmapi(ntp, ip, first_unmap_block, | |
1662 | unmap_len, | |
1663 | XFS_BMAPI_AFLAG(fork) | | |
1664 | (sync ? 0 : XFS_BMAPI_ASYNC), | |
1665 | XFS_ITRUNC_MAX_EXTENTS, | |
1666 | &first_block, &free_list, &done); | |
1667 | if (error) { | |
1668 | /* | |
1669 | * If the bunmapi call encounters an error, | |
1670 | * return to the caller where the transaction | |
1671 | * can be properly aborted. We just need to | |
1672 | * make sure we're not holding any resources | |
1673 | * that we were not when we came in. | |
1674 | */ | |
1675 | xfs_bmap_cancel(&free_list); | |
1676 | return error; | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Duplicate the transaction that has the permanent | |
1681 | * reservation and commit the old transaction. | |
1682 | */ | |
1683 | error = xfs_bmap_finish(tp, &free_list, first_block, | |
1684 | &committed); | |
1685 | ntp = *tp; | |
1686 | if (error) { | |
1687 | /* | |
1688 | * If the bmap finish call encounters an error, | |
1689 | * return to the caller where the transaction | |
1690 | * can be properly aborted. We just need to | |
1691 | * make sure we're not holding any resources | |
1692 | * that we were not when we came in. | |
1693 | * | |
1694 | * Aborting from this point might lose some | |
1695 | * blocks in the file system, but oh well. | |
1696 | */ | |
1697 | xfs_bmap_cancel(&free_list); | |
1698 | if (committed) { | |
1699 | /* | |
1700 | * If the passed in transaction committed | |
1701 | * in xfs_bmap_finish(), then we want to | |
1702 | * add the inode to this one before returning. | |
1703 | * This keeps things simple for the higher | |
1704 | * level code, because it always knows that | |
1705 | * the inode is locked and held in the | |
1706 | * transaction that returns to it whether | |
1707 | * errors occur or not. We don't mark the | |
1708 | * inode dirty so that this transaction can | |
1709 | * be easily aborted if possible. | |
1710 | */ | |
1711 | xfs_trans_ijoin(ntp, ip, | |
1712 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1713 | xfs_trans_ihold(ntp, ip); | |
1714 | } | |
1715 | return error; | |
1716 | } | |
1717 | ||
1718 | if (committed) { | |
1719 | /* | |
1720 | * The first xact was committed, | |
1721 | * so add the inode to the new one. | |
1722 | * Mark it dirty so it will be logged | |
1723 | * and moved forward in the log as | |
1724 | * part of every commit. | |
1725 | */ | |
1726 | xfs_trans_ijoin(ntp, ip, | |
1727 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1728 | xfs_trans_ihold(ntp, ip); | |
1729 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1730 | } | |
1731 | ntp = xfs_trans_dup(ntp); | |
1732 | (void) xfs_trans_commit(*tp, 0, NULL); | |
1733 | *tp = ntp; | |
1734 | error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1735 | XFS_TRANS_PERM_LOG_RES, | |
1736 | XFS_ITRUNCATE_LOG_COUNT); | |
1737 | /* | |
1738 | * Add the inode being truncated to the next chained | |
1739 | * transaction. | |
1740 | */ | |
1741 | xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1742 | xfs_trans_ihold(ntp, ip); | |
1743 | if (error) | |
1744 | return (error); | |
1745 | } | |
1746 | /* | |
1747 | * Only update the size in the case of the data fork, but | |
1748 | * always re-log the inode so that our permanent transaction | |
1749 | * can keep on rolling it forward in the log. | |
1750 | */ | |
1751 | if (fork == XFS_DATA_FORK) { | |
1752 | xfs_isize_check(mp, ip, new_size); | |
1753 | ip->i_d.di_size = new_size; | |
1754 | } | |
1755 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1756 | ASSERT((new_size != 0) || | |
1757 | (fork == XFS_ATTR_FORK) || | |
1758 | (ip->i_delayed_blks == 0)); | |
1759 | ASSERT((new_size != 0) || | |
1760 | (fork == XFS_ATTR_FORK) || | |
1761 | (ip->i_d.di_nextents == 0)); | |
1762 | xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0); | |
1763 | return 0; | |
1764 | } | |
1765 | ||
1766 | ||
1767 | /* | |
1768 | * xfs_igrow_start | |
1769 | * | |
1770 | * Do the first part of growing a file: zero any data in the last | |
1771 | * block that is beyond the old EOF. We need to do this before | |
1772 | * the inode is joined to the transaction to modify the i_size. | |
1773 | * That way we can drop the inode lock and call into the buffer | |
1774 | * cache to get the buffer mapping the EOF. | |
1775 | */ | |
1776 | int | |
1777 | xfs_igrow_start( | |
1778 | xfs_inode_t *ip, | |
1779 | xfs_fsize_t new_size, | |
1780 | cred_t *credp) | |
1781 | { | |
1782 | xfs_fsize_t isize; | |
1783 | int error; | |
1784 | ||
1785 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); | |
1786 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
1787 | ASSERT(new_size > ip->i_d.di_size); | |
1788 | ||
1789 | error = 0; | |
1790 | isize = ip->i_d.di_size; | |
1791 | /* | |
1792 | * Zero any pages that may have been created by | |
1793 | * xfs_write_file() beyond the end of the file | |
1794 | * and any blocks between the old and new file sizes. | |
1795 | */ | |
1796 | error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, isize, | |
1797 | new_size); | |
1798 | return error; | |
1799 | } | |
1800 | ||
1801 | /* | |
1802 | * xfs_igrow_finish | |
1803 | * | |
1804 | * This routine is called to extend the size of a file. | |
1805 | * The inode must have both the iolock and the ilock locked | |
1806 | * for update and it must be a part of the current transaction. | |
1807 | * The xfs_igrow_start() function must have been called previously. | |
1808 | * If the change_flag is not zero, the inode change timestamp will | |
1809 | * be updated. | |
1810 | */ | |
1811 | void | |
1812 | xfs_igrow_finish( | |
1813 | xfs_trans_t *tp, | |
1814 | xfs_inode_t *ip, | |
1815 | xfs_fsize_t new_size, | |
1816 | int change_flag) | |
1817 | { | |
1818 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); | |
1819 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
1820 | ASSERT(ip->i_transp == tp); | |
1821 | ASSERT(new_size > ip->i_d.di_size); | |
1822 | ||
1823 | /* | |
1824 | * Update the file size. Update the inode change timestamp | |
1825 | * if change_flag set. | |
1826 | */ | |
1827 | ip->i_d.di_size = new_size; | |
1828 | if (change_flag) | |
1829 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); | |
1830 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1831 | ||
1832 | } | |
1833 | ||
1834 | ||
1835 | /* | |
1836 | * This is called when the inode's link count goes to 0. | |
1837 | * We place the on-disk inode on a list in the AGI. It | |
1838 | * will be pulled from this list when the inode is freed. | |
1839 | */ | |
1840 | int | |
1841 | xfs_iunlink( | |
1842 | xfs_trans_t *tp, | |
1843 | xfs_inode_t *ip) | |
1844 | { | |
1845 | xfs_mount_t *mp; | |
1846 | xfs_agi_t *agi; | |
1847 | xfs_dinode_t *dip; | |
1848 | xfs_buf_t *agibp; | |
1849 | xfs_buf_t *ibp; | |
1850 | xfs_agnumber_t agno; | |
1851 | xfs_daddr_t agdaddr; | |
1852 | xfs_agino_t agino; | |
1853 | short bucket_index; | |
1854 | int offset; | |
1855 | int error; | |
1856 | int agi_ok; | |
1857 | ||
1858 | ASSERT(ip->i_d.di_nlink == 0); | |
1859 | ASSERT(ip->i_d.di_mode != 0); | |
1860 | ASSERT(ip->i_transp == tp); | |
1861 | ||
1862 | mp = tp->t_mountp; | |
1863 | ||
1864 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1865 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1866 | ||
1867 | /* | |
1868 | * Get the agi buffer first. It ensures lock ordering | |
1869 | * on the list. | |
1870 | */ | |
1871 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1872 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
1873 | if (error) { | |
1874 | return error; | |
1875 | } | |
1876 | /* | |
1877 | * Validate the magic number of the agi block. | |
1878 | */ | |
1879 | agi = XFS_BUF_TO_AGI(agibp); | |
1880 | agi_ok = | |
1881 | INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC && | |
1882 | XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT)); | |
1883 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK, | |
1884 | XFS_RANDOM_IUNLINK))) { | |
1885 | XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi); | |
1886 | xfs_trans_brelse(tp, agibp); | |
1887 | return XFS_ERROR(EFSCORRUPTED); | |
1888 | } | |
1889 | /* | |
1890 | * Get the index into the agi hash table for the | |
1891 | * list this inode will go on. | |
1892 | */ | |
1893 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1894 | ASSERT(agino != 0); | |
1895 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1896 | ASSERT(agi->agi_unlinked[bucket_index]); | |
1897 | ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != agino); | |
1898 | ||
1899 | if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO) { | |
1900 | /* | |
1901 | * There is already another inode in the bucket we need | |
1902 | * to add ourselves to. Add us at the front of the list. | |
1903 | * Here we put the head pointer into our next pointer, | |
1904 | * and then we fall through to point the head at us. | |
1905 | */ | |
1906 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0); | |
1907 | if (error) { | |
1908 | return error; | |
1909 | } | |
1910 | ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO); | |
1911 | ASSERT(dip->di_next_unlinked); | |
1912 | /* both on-disk, don't endian flip twice */ | |
1913 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; | |
1914 | offset = ip->i_boffset + | |
1915 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1916 | xfs_trans_inode_buf(tp, ibp); | |
1917 | xfs_trans_log_buf(tp, ibp, offset, | |
1918 | (offset + sizeof(xfs_agino_t) - 1)); | |
1919 | xfs_inobp_check(mp, ibp); | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * Point the bucket head pointer at the inode being inserted. | |
1924 | */ | |
1925 | ASSERT(agino != 0); | |
1926 | INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, agino); | |
1927 | offset = offsetof(xfs_agi_t, agi_unlinked) + | |
1928 | (sizeof(xfs_agino_t) * bucket_index); | |
1929 | xfs_trans_log_buf(tp, agibp, offset, | |
1930 | (offset + sizeof(xfs_agino_t) - 1)); | |
1931 | return 0; | |
1932 | } | |
1933 | ||
1934 | /* | |
1935 | * Pull the on-disk inode from the AGI unlinked list. | |
1936 | */ | |
1937 | STATIC int | |
1938 | xfs_iunlink_remove( | |
1939 | xfs_trans_t *tp, | |
1940 | xfs_inode_t *ip) | |
1941 | { | |
1942 | xfs_ino_t next_ino; | |
1943 | xfs_mount_t *mp; | |
1944 | xfs_agi_t *agi; | |
1945 | xfs_dinode_t *dip; | |
1946 | xfs_buf_t *agibp; | |
1947 | xfs_buf_t *ibp; | |
1948 | xfs_agnumber_t agno; | |
1949 | xfs_daddr_t agdaddr; | |
1950 | xfs_agino_t agino; | |
1951 | xfs_agino_t next_agino; | |
1952 | xfs_buf_t *last_ibp; | |
1953 | xfs_dinode_t *last_dip; | |
1954 | short bucket_index; | |
1955 | int offset, last_offset; | |
1956 | int error; | |
1957 | int agi_ok; | |
1958 | ||
1959 | /* | |
1960 | * First pull the on-disk inode from the AGI unlinked list. | |
1961 | */ | |
1962 | mp = tp->t_mountp; | |
1963 | ||
1964 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1965 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1966 | ||
1967 | /* | |
1968 | * Get the agi buffer first. It ensures lock ordering | |
1969 | * on the list. | |
1970 | */ | |
1971 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1972 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
1973 | if (error) { | |
1974 | cmn_err(CE_WARN, | |
1975 | "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.", | |
1976 | error, mp->m_fsname); | |
1977 | return error; | |
1978 | } | |
1979 | /* | |
1980 | * Validate the magic number of the agi block. | |
1981 | */ | |
1982 | agi = XFS_BUF_TO_AGI(agibp); | |
1983 | agi_ok = | |
1984 | INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC && | |
1985 | XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT)); | |
1986 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE, | |
1987 | XFS_RANDOM_IUNLINK_REMOVE))) { | |
1988 | XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW, | |
1989 | mp, agi); | |
1990 | xfs_trans_brelse(tp, agibp); | |
1991 | cmn_err(CE_WARN, | |
1992 | "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.", | |
1993 | mp->m_fsname); | |
1994 | return XFS_ERROR(EFSCORRUPTED); | |
1995 | } | |
1996 | /* | |
1997 | * Get the index into the agi hash table for the | |
1998 | * list this inode will go on. | |
1999 | */ | |
2000 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2001 | ASSERT(agino != 0); | |
2002 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
2003 | ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO); | |
2004 | ASSERT(agi->agi_unlinked[bucket_index]); | |
2005 | ||
2006 | if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) == agino) { | |
2007 | /* | |
2008 | * We're at the head of the list. Get the inode's | |
2009 | * on-disk buffer to see if there is anyone after us | |
2010 | * on the list. Only modify our next pointer if it | |
2011 | * is not already NULLAGINO. This saves us the overhead | |
2012 | * of dealing with the buffer when there is no need to | |
2013 | * change it. | |
2014 | */ | |
2015 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0); | |
2016 | if (error) { | |
2017 | cmn_err(CE_WARN, | |
2018 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2019 | error, mp->m_fsname); | |
2020 | return error; | |
2021 | } | |
2022 | next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); | |
2023 | ASSERT(next_agino != 0); | |
2024 | if (next_agino != NULLAGINO) { | |
2025 | INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); | |
2026 | offset = ip->i_boffset + | |
2027 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2028 | xfs_trans_inode_buf(tp, ibp); | |
2029 | xfs_trans_log_buf(tp, ibp, offset, | |
2030 | (offset + sizeof(xfs_agino_t) - 1)); | |
2031 | xfs_inobp_check(mp, ibp); | |
2032 | } else { | |
2033 | xfs_trans_brelse(tp, ibp); | |
2034 | } | |
2035 | /* | |
2036 | * Point the bucket head pointer at the next inode. | |
2037 | */ | |
2038 | ASSERT(next_agino != 0); | |
2039 | ASSERT(next_agino != agino); | |
2040 | INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, next_agino); | |
2041 | offset = offsetof(xfs_agi_t, agi_unlinked) + | |
2042 | (sizeof(xfs_agino_t) * bucket_index); | |
2043 | xfs_trans_log_buf(tp, agibp, offset, | |
2044 | (offset + sizeof(xfs_agino_t) - 1)); | |
2045 | } else { | |
2046 | /* | |
2047 | * We need to search the list for the inode being freed. | |
2048 | */ | |
2049 | next_agino = INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT); | |
2050 | last_ibp = NULL; | |
2051 | while (next_agino != agino) { | |
2052 | /* | |
2053 | * If the last inode wasn't the one pointing to | |
2054 | * us, then release its buffer since we're not | |
2055 | * going to do anything with it. | |
2056 | */ | |
2057 | if (last_ibp != NULL) { | |
2058 | xfs_trans_brelse(tp, last_ibp); | |
2059 | } | |
2060 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); | |
2061 | error = xfs_inotobp(mp, tp, next_ino, &last_dip, | |
2062 | &last_ibp, &last_offset); | |
2063 | if (error) { | |
2064 | cmn_err(CE_WARN, | |
2065 | "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.", | |
2066 | error, mp->m_fsname); | |
2067 | return error; | |
2068 | } | |
2069 | next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT); | |
2070 | ASSERT(next_agino != NULLAGINO); | |
2071 | ASSERT(next_agino != 0); | |
2072 | } | |
2073 | /* | |
2074 | * Now last_ibp points to the buffer previous to us on | |
2075 | * the unlinked list. Pull us from the list. | |
2076 | */ | |
2077 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0); | |
2078 | if (error) { | |
2079 | cmn_err(CE_WARN, | |
2080 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2081 | error, mp->m_fsname); | |
2082 | return error; | |
2083 | } | |
2084 | next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); | |
2085 | ASSERT(next_agino != 0); | |
2086 | ASSERT(next_agino != agino); | |
2087 | if (next_agino != NULLAGINO) { | |
2088 | INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); | |
2089 | offset = ip->i_boffset + | |
2090 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2091 | xfs_trans_inode_buf(tp, ibp); | |
2092 | xfs_trans_log_buf(tp, ibp, offset, | |
2093 | (offset + sizeof(xfs_agino_t) - 1)); | |
2094 | xfs_inobp_check(mp, ibp); | |
2095 | } else { | |
2096 | xfs_trans_brelse(tp, ibp); | |
2097 | } | |
2098 | /* | |
2099 | * Point the previous inode on the list to the next inode. | |
2100 | */ | |
2101 | INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino); | |
2102 | ASSERT(next_agino != 0); | |
2103 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
2104 | xfs_trans_inode_buf(tp, last_ibp); | |
2105 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2106 | (offset + sizeof(xfs_agino_t) - 1)); | |
2107 | xfs_inobp_check(mp, last_ibp); | |
2108 | } | |
2109 | return 0; | |
2110 | } | |
2111 | ||
2112 | static __inline__ int xfs_inode_clean(xfs_inode_t *ip) | |
2113 | { | |
2114 | return (((ip->i_itemp == NULL) || | |
2115 | !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
2116 | (ip->i_update_core == 0)); | |
2117 | } | |
2118 | ||
ba0f32d4 | 2119 | STATIC void |
1da177e4 LT |
2120 | xfs_ifree_cluster( |
2121 | xfs_inode_t *free_ip, | |
2122 | xfs_trans_t *tp, | |
2123 | xfs_ino_t inum) | |
2124 | { | |
2125 | xfs_mount_t *mp = free_ip->i_mount; | |
2126 | int blks_per_cluster; | |
2127 | int nbufs; | |
2128 | int ninodes; | |
2129 | int i, j, found, pre_flushed; | |
2130 | xfs_daddr_t blkno; | |
2131 | xfs_buf_t *bp; | |
2132 | xfs_ihash_t *ih; | |
2133 | xfs_inode_t *ip, **ip_found; | |
2134 | xfs_inode_log_item_t *iip; | |
2135 | xfs_log_item_t *lip; | |
2136 | SPLDECL(s); | |
2137 | ||
2138 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
2139 | blks_per_cluster = 1; | |
2140 | ninodes = mp->m_sb.sb_inopblock; | |
2141 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
2142 | } else { | |
2143 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
2144 | mp->m_sb.sb_blocksize; | |
2145 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
2146 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
2147 | } | |
2148 | ||
2149 | ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS); | |
2150 | ||
2151 | for (j = 0; j < nbufs; j++, inum += ninodes) { | |
2152 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
2153 | XFS_INO_TO_AGBNO(mp, inum)); | |
2154 | ||
2155 | ||
2156 | /* | |
2157 | * Look for each inode in memory and attempt to lock it, | |
2158 | * we can be racing with flush and tail pushing here. | |
2159 | * any inode we get the locks on, add to an array of | |
2160 | * inode items to process later. | |
2161 | * | |
2162 | * The get the buffer lock, we could beat a flush | |
2163 | * or tail pushing thread to the lock here, in which | |
2164 | * case they will go looking for the inode buffer | |
2165 | * and fail, we need some other form of interlock | |
2166 | * here. | |
2167 | */ | |
2168 | found = 0; | |
2169 | for (i = 0; i < ninodes; i++) { | |
2170 | ih = XFS_IHASH(mp, inum + i); | |
2171 | read_lock(&ih->ih_lock); | |
2172 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { | |
2173 | if (ip->i_ino == inum + i) | |
2174 | break; | |
2175 | } | |
2176 | ||
2177 | /* Inode not in memory or we found it already, | |
2178 | * nothing to do | |
2179 | */ | |
2180 | if (!ip || (ip->i_flags & XFS_ISTALE)) { | |
2181 | read_unlock(&ih->ih_lock); | |
2182 | continue; | |
2183 | } | |
2184 | ||
2185 | if (xfs_inode_clean(ip)) { | |
2186 | read_unlock(&ih->ih_lock); | |
2187 | continue; | |
2188 | } | |
2189 | ||
2190 | /* If we can get the locks then add it to the | |
2191 | * list, otherwise by the time we get the bp lock | |
2192 | * below it will already be attached to the | |
2193 | * inode buffer. | |
2194 | */ | |
2195 | ||
2196 | /* This inode will already be locked - by us, lets | |
2197 | * keep it that way. | |
2198 | */ | |
2199 | ||
2200 | if (ip == free_ip) { | |
2201 | if (xfs_iflock_nowait(ip)) { | |
2202 | ip->i_flags |= XFS_ISTALE; | |
2203 | ||
2204 | if (xfs_inode_clean(ip)) { | |
2205 | xfs_ifunlock(ip); | |
2206 | } else { | |
2207 | ip_found[found++] = ip; | |
2208 | } | |
2209 | } | |
2210 | read_unlock(&ih->ih_lock); | |
2211 | continue; | |
2212 | } | |
2213 | ||
2214 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2215 | if (xfs_iflock_nowait(ip)) { | |
2216 | ip->i_flags |= XFS_ISTALE; | |
2217 | ||
2218 | if (xfs_inode_clean(ip)) { | |
2219 | xfs_ifunlock(ip); | |
2220 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2221 | } else { | |
2222 | ip_found[found++] = ip; | |
2223 | } | |
2224 | } else { | |
2225 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2226 | } | |
2227 | } | |
2228 | ||
2229 | read_unlock(&ih->ih_lock); | |
2230 | } | |
2231 | ||
2232 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
2233 | mp->m_bsize * blks_per_cluster, | |
2234 | XFS_BUF_LOCK); | |
2235 | ||
2236 | pre_flushed = 0; | |
2237 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
2238 | while (lip) { | |
2239 | if (lip->li_type == XFS_LI_INODE) { | |
2240 | iip = (xfs_inode_log_item_t *)lip; | |
2241 | ASSERT(iip->ili_logged == 1); | |
2242 | lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done; | |
2243 | AIL_LOCK(mp,s); | |
2244 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
2245 | AIL_UNLOCK(mp, s); | |
2246 | iip->ili_inode->i_flags |= XFS_ISTALE; | |
2247 | pre_flushed++; | |
2248 | } | |
2249 | lip = lip->li_bio_list; | |
2250 | } | |
2251 | ||
2252 | for (i = 0; i < found; i++) { | |
2253 | ip = ip_found[i]; | |
2254 | iip = ip->i_itemp; | |
2255 | ||
2256 | if (!iip) { | |
2257 | ip->i_update_core = 0; | |
2258 | xfs_ifunlock(ip); | |
2259 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2260 | continue; | |
2261 | } | |
2262 | ||
2263 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
2264 | iip->ili_format.ilf_fields = 0; | |
2265 | iip->ili_logged = 1; | |
2266 | AIL_LOCK(mp,s); | |
2267 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
2268 | AIL_UNLOCK(mp, s); | |
2269 | ||
2270 | xfs_buf_attach_iodone(bp, | |
2271 | (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
2272 | xfs_istale_done, (xfs_log_item_t *)iip); | |
2273 | if (ip != free_ip) { | |
2274 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2275 | } | |
2276 | } | |
2277 | ||
2278 | if (found || pre_flushed) | |
2279 | xfs_trans_stale_inode_buf(tp, bp); | |
2280 | xfs_trans_binval(tp, bp); | |
2281 | } | |
2282 | ||
2283 | kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *)); | |
2284 | } | |
2285 | ||
2286 | /* | |
2287 | * This is called to return an inode to the inode free list. | |
2288 | * The inode should already be truncated to 0 length and have | |
2289 | * no pages associated with it. This routine also assumes that | |
2290 | * the inode is already a part of the transaction. | |
2291 | * | |
2292 | * The on-disk copy of the inode will have been added to the list | |
2293 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2294 | * that list atomically with respect to freeing it here. | |
2295 | */ | |
2296 | int | |
2297 | xfs_ifree( | |
2298 | xfs_trans_t *tp, | |
2299 | xfs_inode_t *ip, | |
2300 | xfs_bmap_free_t *flist) | |
2301 | { | |
2302 | int error; | |
2303 | int delete; | |
2304 | xfs_ino_t first_ino; | |
2305 | ||
2306 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2307 | ASSERT(ip->i_transp == tp); | |
2308 | ASSERT(ip->i_d.di_nlink == 0); | |
2309 | ASSERT(ip->i_d.di_nextents == 0); | |
2310 | ASSERT(ip->i_d.di_anextents == 0); | |
2311 | ASSERT((ip->i_d.di_size == 0) || | |
2312 | ((ip->i_d.di_mode & S_IFMT) != S_IFREG)); | |
2313 | ASSERT(ip->i_d.di_nblocks == 0); | |
2314 | ||
2315 | /* | |
2316 | * Pull the on-disk inode from the AGI unlinked list. | |
2317 | */ | |
2318 | error = xfs_iunlink_remove(tp, ip); | |
2319 | if (error != 0) { | |
2320 | return error; | |
2321 | } | |
2322 | ||
2323 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
2324 | if (error != 0) { | |
2325 | return error; | |
2326 | } | |
2327 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
2328 | ip->i_d.di_flags = 0; | |
2329 | ip->i_d.di_dmevmask = 0; | |
2330 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
2331 | ip->i_df.if_ext_max = | |
2332 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
2333 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
2334 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2335 | /* | |
2336 | * Bump the generation count so no one will be confused | |
2337 | * by reincarnations of this inode. | |
2338 | */ | |
2339 | ip->i_d.di_gen++; | |
2340 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2341 | ||
2342 | if (delete) { | |
2343 | xfs_ifree_cluster(ip, tp, first_ino); | |
2344 | } | |
2345 | ||
2346 | return 0; | |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * Reallocate the space for if_broot based on the number of records | |
2351 | * being added or deleted as indicated in rec_diff. Move the records | |
2352 | * and pointers in if_broot to fit the new size. When shrinking this | |
2353 | * will eliminate holes between the records and pointers created by | |
2354 | * the caller. When growing this will create holes to be filled in | |
2355 | * by the caller. | |
2356 | * | |
2357 | * The caller must not request to add more records than would fit in | |
2358 | * the on-disk inode root. If the if_broot is currently NULL, then | |
2359 | * if we adding records one will be allocated. The caller must also | |
2360 | * not request that the number of records go below zero, although | |
2361 | * it can go to zero. | |
2362 | * | |
2363 | * ip -- the inode whose if_broot area is changing | |
2364 | * ext_diff -- the change in the number of records, positive or negative, | |
2365 | * requested for the if_broot array. | |
2366 | */ | |
2367 | void | |
2368 | xfs_iroot_realloc( | |
2369 | xfs_inode_t *ip, | |
2370 | int rec_diff, | |
2371 | int whichfork) | |
2372 | { | |
2373 | int cur_max; | |
2374 | xfs_ifork_t *ifp; | |
2375 | xfs_bmbt_block_t *new_broot; | |
2376 | int new_max; | |
2377 | size_t new_size; | |
2378 | char *np; | |
2379 | char *op; | |
2380 | ||
2381 | /* | |
2382 | * Handle the degenerate case quietly. | |
2383 | */ | |
2384 | if (rec_diff == 0) { | |
2385 | return; | |
2386 | } | |
2387 | ||
2388 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2389 | if (rec_diff > 0) { | |
2390 | /* | |
2391 | * If there wasn't any memory allocated before, just | |
2392 | * allocate it now and get out. | |
2393 | */ | |
2394 | if (ifp->if_broot_bytes == 0) { | |
2395 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
2396 | ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size, | |
2397 | KM_SLEEP); | |
2398 | ifp->if_broot_bytes = (int)new_size; | |
2399 | return; | |
2400 | } | |
2401 | ||
2402 | /* | |
2403 | * If there is already an existing if_broot, then we need | |
2404 | * to realloc() it and shift the pointers to their new | |
2405 | * location. The records don't change location because | |
2406 | * they are kept butted up against the btree block header. | |
2407 | */ | |
2408 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2409 | new_max = cur_max + rec_diff; | |
2410 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2411 | ifp->if_broot = (xfs_bmbt_block_t *) | |
2412 | kmem_realloc(ifp->if_broot, | |
2413 | new_size, | |
2414 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ | |
2415 | KM_SLEEP); | |
2416 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2417 | ifp->if_broot_bytes); | |
2418 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2419 | (int)new_size); | |
2420 | ifp->if_broot_bytes = (int)new_size; | |
2421 | ASSERT(ifp->if_broot_bytes <= | |
2422 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2423 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2424 | return; | |
2425 | } | |
2426 | ||
2427 | /* | |
2428 | * rec_diff is less than 0. In this case, we are shrinking the | |
2429 | * if_broot buffer. It must already exist. If we go to zero | |
2430 | * records, just get rid of the root and clear the status bit. | |
2431 | */ | |
2432 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
2433 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2434 | new_max = cur_max + rec_diff; | |
2435 | ASSERT(new_max >= 0); | |
2436 | if (new_max > 0) | |
2437 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2438 | else | |
2439 | new_size = 0; | |
2440 | if (new_size > 0) { | |
2441 | new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP); | |
2442 | /* | |
2443 | * First copy over the btree block header. | |
2444 | */ | |
2445 | memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t)); | |
2446 | } else { | |
2447 | new_broot = NULL; | |
2448 | ifp->if_flags &= ~XFS_IFBROOT; | |
2449 | } | |
2450 | ||
2451 | /* | |
2452 | * Only copy the records and pointers if there are any. | |
2453 | */ | |
2454 | if (new_max > 0) { | |
2455 | /* | |
2456 | * First copy the records. | |
2457 | */ | |
2458 | op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1, | |
2459 | ifp->if_broot_bytes); | |
2460 | np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1, | |
2461 | (int)new_size); | |
2462 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); | |
2463 | ||
2464 | /* | |
2465 | * Then copy the pointers. | |
2466 | */ | |
2467 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2468 | ifp->if_broot_bytes); | |
2469 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1, | |
2470 | (int)new_size); | |
2471 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2472 | } | |
2473 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2474 | ifp->if_broot = new_broot; | |
2475 | ifp->if_broot_bytes = (int)new_size; | |
2476 | ASSERT(ifp->if_broot_bytes <= | |
2477 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2478 | return; | |
2479 | } | |
2480 | ||
2481 | ||
2482 | /* | |
2483 | * This is called when the amount of space needed for if_extents | |
2484 | * is increased or decreased. The change in size is indicated by | |
2485 | * the number of extents that need to be added or deleted in the | |
2486 | * ext_diff parameter. | |
2487 | * | |
2488 | * If the amount of space needed has decreased below the size of the | |
2489 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2490 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2491 | * to what is needed. | |
2492 | * | |
2493 | * ip -- the inode whose if_extents area is changing | |
2494 | * ext_diff -- the change in the number of extents, positive or negative, | |
2495 | * requested for the if_extents array. | |
2496 | */ | |
2497 | void | |
2498 | xfs_iext_realloc( | |
2499 | xfs_inode_t *ip, | |
2500 | int ext_diff, | |
2501 | int whichfork) | |
2502 | { | |
2503 | int byte_diff; | |
2504 | xfs_ifork_t *ifp; | |
2505 | int new_size; | |
2506 | uint rnew_size; | |
2507 | ||
2508 | if (ext_diff == 0) { | |
2509 | return; | |
2510 | } | |
2511 | ||
2512 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2513 | byte_diff = ext_diff * (uint)sizeof(xfs_bmbt_rec_t); | |
2514 | new_size = (int)ifp->if_bytes + byte_diff; | |
2515 | ASSERT(new_size >= 0); | |
2516 | ||
2517 | if (new_size == 0) { | |
2518 | if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) { | |
2519 | ASSERT(ifp->if_real_bytes != 0); | |
2520 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); | |
2521 | } | |
2522 | ifp->if_u1.if_extents = NULL; | |
2523 | rnew_size = 0; | |
2524 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_ext)) { | |
2525 | /* | |
2526 | * If the valid extents can fit in if_inline_ext, | |
2527 | * copy them from the malloc'd vector and free it. | |
2528 | */ | |
2529 | if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) { | |
2530 | /* | |
2531 | * For now, empty files are format EXTENTS, | |
2532 | * so the if_extents pointer is null. | |
2533 | */ | |
2534 | if (ifp->if_u1.if_extents) { | |
2535 | memcpy(ifp->if_u2.if_inline_ext, | |
2536 | ifp->if_u1.if_extents, new_size); | |
2537 | kmem_free(ifp->if_u1.if_extents, | |
2538 | ifp->if_real_bytes); | |
2539 | } | |
2540 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
2541 | } | |
2542 | rnew_size = 0; | |
2543 | } else { | |
2544 | rnew_size = new_size; | |
2545 | if ((rnew_size & (rnew_size - 1)) != 0) | |
2546 | rnew_size = xfs_iroundup(rnew_size); | |
2547 | /* | |
2548 | * Stuck with malloc/realloc. | |
2549 | */ | |
2550 | if (ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext) { | |
2551 | ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) | |
2552 | kmem_alloc(rnew_size, KM_SLEEP); | |
2553 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
2554 | sizeof(ifp->if_u2.if_inline_ext)); | |
2555 | } else if (rnew_size != ifp->if_real_bytes) { | |
2556 | ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) | |
2557 | kmem_realloc(ifp->if_u1.if_extents, | |
2558 | rnew_size, | |
2559 | ifp->if_real_bytes, | |
2560 | KM_NOFS); | |
2561 | } | |
2562 | } | |
2563 | ifp->if_real_bytes = rnew_size; | |
2564 | ifp->if_bytes = new_size; | |
2565 | } | |
2566 | ||
2567 | ||
2568 | /* | |
2569 | * This is called when the amount of space needed for if_data | |
2570 | * is increased or decreased. The change in size is indicated by | |
2571 | * the number of bytes that need to be added or deleted in the | |
2572 | * byte_diff parameter. | |
2573 | * | |
2574 | * If the amount of space needed has decreased below the size of the | |
2575 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2576 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2577 | * to what is needed. | |
2578 | * | |
2579 | * ip -- the inode whose if_data area is changing | |
2580 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2581 | * requested for the if_data array. | |
2582 | */ | |
2583 | void | |
2584 | xfs_idata_realloc( | |
2585 | xfs_inode_t *ip, | |
2586 | int byte_diff, | |
2587 | int whichfork) | |
2588 | { | |
2589 | xfs_ifork_t *ifp; | |
2590 | int new_size; | |
2591 | int real_size; | |
2592 | ||
2593 | if (byte_diff == 0) { | |
2594 | return; | |
2595 | } | |
2596 | ||
2597 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2598 | new_size = (int)ifp->if_bytes + byte_diff; | |
2599 | ASSERT(new_size >= 0); | |
2600 | ||
2601 | if (new_size == 0) { | |
2602 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2603 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2604 | } | |
2605 | ifp->if_u1.if_data = NULL; | |
2606 | real_size = 0; | |
2607 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2608 | /* | |
2609 | * If the valid extents/data can fit in if_inline_ext/data, | |
2610 | * copy them from the malloc'd vector and free it. | |
2611 | */ | |
2612 | if (ifp->if_u1.if_data == NULL) { | |
2613 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2614 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2615 | ASSERT(ifp->if_real_bytes != 0); | |
2616 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2617 | new_size); | |
2618 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2619 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2620 | } | |
2621 | real_size = 0; | |
2622 | } else { | |
2623 | /* | |
2624 | * Stuck with malloc/realloc. | |
2625 | * For inline data, the underlying buffer must be | |
2626 | * a multiple of 4 bytes in size so that it can be | |
2627 | * logged and stay on word boundaries. We enforce | |
2628 | * that here. | |
2629 | */ | |
2630 | real_size = roundup(new_size, 4); | |
2631 | if (ifp->if_u1.if_data == NULL) { | |
2632 | ASSERT(ifp->if_real_bytes == 0); | |
2633 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2634 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2635 | /* | |
2636 | * Only do the realloc if the underlying size | |
2637 | * is really changing. | |
2638 | */ | |
2639 | if (ifp->if_real_bytes != real_size) { | |
2640 | ifp->if_u1.if_data = | |
2641 | kmem_realloc(ifp->if_u1.if_data, | |
2642 | real_size, | |
2643 | ifp->if_real_bytes, | |
2644 | KM_SLEEP); | |
2645 | } | |
2646 | } else { | |
2647 | ASSERT(ifp->if_real_bytes == 0); | |
2648 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2649 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, | |
2650 | ifp->if_bytes); | |
2651 | } | |
2652 | } | |
2653 | ifp->if_real_bytes = real_size; | |
2654 | ifp->if_bytes = new_size; | |
2655 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2656 | } | |
2657 | ||
2658 | ||
2659 | ||
2660 | ||
2661 | /* | |
2662 | * Map inode to disk block and offset. | |
2663 | * | |
2664 | * mp -- the mount point structure for the current file system | |
2665 | * tp -- the current transaction | |
2666 | * ino -- the inode number of the inode to be located | |
2667 | * imap -- this structure is filled in with the information necessary | |
2668 | * to retrieve the given inode from disk | |
2669 | * flags -- flags to pass to xfs_dilocate indicating whether or not | |
2670 | * lookups in the inode btree were OK or not | |
2671 | */ | |
2672 | int | |
2673 | xfs_imap( | |
2674 | xfs_mount_t *mp, | |
2675 | xfs_trans_t *tp, | |
2676 | xfs_ino_t ino, | |
2677 | xfs_imap_t *imap, | |
2678 | uint flags) | |
2679 | { | |
2680 | xfs_fsblock_t fsbno; | |
2681 | int len; | |
2682 | int off; | |
2683 | int error; | |
2684 | ||
2685 | fsbno = imap->im_blkno ? | |
2686 | XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; | |
2687 | error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); | |
2688 | if (error != 0) { | |
2689 | return error; | |
2690 | } | |
2691 | imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); | |
2692 | imap->im_len = XFS_FSB_TO_BB(mp, len); | |
2693 | imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); | |
2694 | imap->im_ioffset = (ushort)off; | |
2695 | imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); | |
2696 | return 0; | |
2697 | } | |
2698 | ||
2699 | void | |
2700 | xfs_idestroy_fork( | |
2701 | xfs_inode_t *ip, | |
2702 | int whichfork) | |
2703 | { | |
2704 | xfs_ifork_t *ifp; | |
2705 | ||
2706 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2707 | if (ifp->if_broot != NULL) { | |
2708 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2709 | ifp->if_broot = NULL; | |
2710 | } | |
2711 | ||
2712 | /* | |
2713 | * If the format is local, then we can't have an extents | |
2714 | * array so just look for an inline data array. If we're | |
2715 | * not local then we may or may not have an extents list, | |
2716 | * so check and free it up if we do. | |
2717 | */ | |
2718 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2719 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2720 | (ifp->if_u1.if_data != NULL)) { | |
2721 | ASSERT(ifp->if_real_bytes != 0); | |
2722 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2723 | ifp->if_u1.if_data = NULL; | |
2724 | ifp->if_real_bytes = 0; | |
2725 | } | |
2726 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
2727 | (ifp->if_u1.if_extents != NULL) && | |
2728 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)) { | |
2729 | ASSERT(ifp->if_real_bytes != 0); | |
2730 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); | |
2731 | ifp->if_u1.if_extents = NULL; | |
2732 | ifp->if_real_bytes = 0; | |
2733 | } | |
2734 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2735 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2736 | ASSERT(ifp->if_real_bytes == 0); | |
2737 | if (whichfork == XFS_ATTR_FORK) { | |
2738 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2739 | ip->i_afp = NULL; | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | /* | |
2744 | * This is called free all the memory associated with an inode. | |
2745 | * It must free the inode itself and any buffers allocated for | |
2746 | * if_extents/if_data and if_broot. It must also free the lock | |
2747 | * associated with the inode. | |
2748 | */ | |
2749 | void | |
2750 | xfs_idestroy( | |
2751 | xfs_inode_t *ip) | |
2752 | { | |
2753 | ||
2754 | switch (ip->i_d.di_mode & S_IFMT) { | |
2755 | case S_IFREG: | |
2756 | case S_IFDIR: | |
2757 | case S_IFLNK: | |
2758 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
2759 | break; | |
2760 | } | |
2761 | if (ip->i_afp) | |
2762 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
2763 | mrfree(&ip->i_lock); | |
2764 | mrfree(&ip->i_iolock); | |
2765 | freesema(&ip->i_flock); | |
2766 | #ifdef XFS_BMAP_TRACE | |
2767 | ktrace_free(ip->i_xtrace); | |
2768 | #endif | |
2769 | #ifdef XFS_BMBT_TRACE | |
2770 | ktrace_free(ip->i_btrace); | |
2771 | #endif | |
2772 | #ifdef XFS_RW_TRACE | |
2773 | ktrace_free(ip->i_rwtrace); | |
2774 | #endif | |
2775 | #ifdef XFS_ILOCK_TRACE | |
2776 | ktrace_free(ip->i_lock_trace); | |
2777 | #endif | |
2778 | #ifdef XFS_DIR2_TRACE | |
2779 | ktrace_free(ip->i_dir_trace); | |
2780 | #endif | |
2781 | if (ip->i_itemp) { | |
2782 | /* XXXdpd should be able to assert this but shutdown | |
2783 | * is leaving the AIL behind. */ | |
2784 | ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) || | |
2785 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
2786 | xfs_inode_item_destroy(ip); | |
2787 | } | |
2788 | kmem_zone_free(xfs_inode_zone, ip); | |
2789 | } | |
2790 | ||
2791 | ||
2792 | /* | |
2793 | * Increment the pin count of the given buffer. | |
2794 | * This value is protected by ipinlock spinlock in the mount structure. | |
2795 | */ | |
2796 | void | |
2797 | xfs_ipin( | |
2798 | xfs_inode_t *ip) | |
2799 | { | |
2800 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2801 | ||
2802 | atomic_inc(&ip->i_pincount); | |
2803 | } | |
2804 | ||
2805 | /* | |
2806 | * Decrement the pin count of the given inode, and wake up | |
2807 | * anyone in xfs_iwait_unpin() if the count goes to 0. The | |
2808 | * inode must have been previoulsy pinned with a call to xfs_ipin(). | |
2809 | */ | |
2810 | void | |
2811 | xfs_iunpin( | |
2812 | xfs_inode_t *ip) | |
2813 | { | |
2814 | ASSERT(atomic_read(&ip->i_pincount) > 0); | |
2815 | ||
2816 | if (atomic_dec_and_test(&ip->i_pincount)) { | |
2817 | vnode_t *vp = XFS_ITOV_NULL(ip); | |
2818 | ||
2819 | /* make sync come back and flush this inode */ | |
2820 | if (vp) { | |
2821 | struct inode *inode = LINVFS_GET_IP(vp); | |
2822 | ||
2823 | if (!(inode->i_state & I_NEW)) | |
2824 | mark_inode_dirty_sync(inode); | |
2825 | } | |
2826 | ||
2827 | wake_up(&ip->i_ipin_wait); | |
2828 | } | |
2829 | } | |
2830 | ||
2831 | /* | |
2832 | * This is called to wait for the given inode to be unpinned. | |
2833 | * It will sleep until this happens. The caller must have the | |
2834 | * inode locked in at least shared mode so that the buffer cannot | |
2835 | * be subsequently pinned once someone is waiting for it to be | |
2836 | * unpinned. | |
2837 | */ | |
ba0f32d4 | 2838 | STATIC void |
1da177e4 LT |
2839 | xfs_iunpin_wait( |
2840 | xfs_inode_t *ip) | |
2841 | { | |
2842 | xfs_inode_log_item_t *iip; | |
2843 | xfs_lsn_t lsn; | |
2844 | ||
2845 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS)); | |
2846 | ||
2847 | if (atomic_read(&ip->i_pincount) == 0) { | |
2848 | return; | |
2849 | } | |
2850 | ||
2851 | iip = ip->i_itemp; | |
2852 | if (iip && iip->ili_last_lsn) { | |
2853 | lsn = iip->ili_last_lsn; | |
2854 | } else { | |
2855 | lsn = (xfs_lsn_t)0; | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * Give the log a push so we don't wait here too long. | |
2860 | */ | |
2861 | xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE); | |
2862 | ||
2863 | wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0)); | |
2864 | } | |
2865 | ||
2866 | ||
2867 | /* | |
2868 | * xfs_iextents_copy() | |
2869 | * | |
2870 | * This is called to copy the REAL extents (as opposed to the delayed | |
2871 | * allocation extents) from the inode into the given buffer. It | |
2872 | * returns the number of bytes copied into the buffer. | |
2873 | * | |
2874 | * If there are no delayed allocation extents, then we can just | |
2875 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2876 | * examine each extent in turn and skip those which are delayed. | |
2877 | */ | |
2878 | int | |
2879 | xfs_iextents_copy( | |
2880 | xfs_inode_t *ip, | |
2881 | xfs_bmbt_rec_t *buffer, | |
2882 | int whichfork) | |
2883 | { | |
2884 | int copied; | |
2885 | xfs_bmbt_rec_t *dest_ep; | |
2886 | xfs_bmbt_rec_t *ep; | |
2887 | #ifdef XFS_BMAP_TRACE | |
2888 | static char fname[] = "xfs_iextents_copy"; | |
2889 | #endif | |
2890 | int i; | |
2891 | xfs_ifork_t *ifp; | |
2892 | int nrecs; | |
2893 | xfs_fsblock_t start_block; | |
2894 | ||
2895 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2896 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
2897 | ASSERT(ifp->if_bytes > 0); | |
2898 | ||
2899 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2900 | xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork); | |
2901 | ASSERT(nrecs > 0); | |
2902 | ||
2903 | /* | |
2904 | * There are some delayed allocation extents in the | |
2905 | * inode, so copy the extents one at a time and skip | |
2906 | * the delayed ones. There must be at least one | |
2907 | * non-delayed extent. | |
2908 | */ | |
2909 | ep = ifp->if_u1.if_extents; | |
2910 | dest_ep = buffer; | |
2911 | copied = 0; | |
2912 | for (i = 0; i < nrecs; i++) { | |
2913 | start_block = xfs_bmbt_get_startblock(ep); | |
2914 | if (ISNULLSTARTBLOCK(start_block)) { | |
2915 | /* | |
2916 | * It's a delayed allocation extent, so skip it. | |
2917 | */ | |
2918 | ep++; | |
2919 | continue; | |
2920 | } | |
2921 | ||
2922 | /* Translate to on disk format */ | |
2923 | put_unaligned(INT_GET(ep->l0, ARCH_CONVERT), | |
2924 | (__uint64_t*)&dest_ep->l0); | |
2925 | put_unaligned(INT_GET(ep->l1, ARCH_CONVERT), | |
2926 | (__uint64_t*)&dest_ep->l1); | |
2927 | dest_ep++; | |
2928 | ep++; | |
2929 | copied++; | |
2930 | } | |
2931 | ASSERT(copied != 0); | |
2932 | xfs_validate_extents(buffer, copied, 1, XFS_EXTFMT_INODE(ip)); | |
2933 | ||
2934 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2935 | } | |
2936 | ||
2937 | /* | |
2938 | * Each of the following cases stores data into the same region | |
2939 | * of the on-disk inode, so only one of them can be valid at | |
2940 | * any given time. While it is possible to have conflicting formats | |
2941 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2942 | * in EXTENTS format, this can only happen when the fork has | |
2943 | * changed formats after being modified but before being flushed. | |
2944 | * In these cases, the format always takes precedence, because the | |
2945 | * format indicates the current state of the fork. | |
2946 | */ | |
2947 | /*ARGSUSED*/ | |
2948 | STATIC int | |
2949 | xfs_iflush_fork( | |
2950 | xfs_inode_t *ip, | |
2951 | xfs_dinode_t *dip, | |
2952 | xfs_inode_log_item_t *iip, | |
2953 | int whichfork, | |
2954 | xfs_buf_t *bp) | |
2955 | { | |
2956 | char *cp; | |
2957 | xfs_ifork_t *ifp; | |
2958 | xfs_mount_t *mp; | |
2959 | #ifdef XFS_TRANS_DEBUG | |
2960 | int first; | |
2961 | #endif | |
2962 | static const short brootflag[2] = | |
2963 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2964 | static const short dataflag[2] = | |
2965 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2966 | static const short extflag[2] = | |
2967 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2968 | ||
2969 | if (iip == NULL) | |
2970 | return 0; | |
2971 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2972 | /* | |
2973 | * This can happen if we gave up in iformat in an error path, | |
2974 | * for the attribute fork. | |
2975 | */ | |
2976 | if (ifp == NULL) { | |
2977 | ASSERT(whichfork == XFS_ATTR_FORK); | |
2978 | return 0; | |
2979 | } | |
2980 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2981 | mp = ip->i_mount; | |
2982 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2983 | case XFS_DINODE_FMT_LOCAL: | |
2984 | if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && | |
2985 | (ifp->if_bytes > 0)) { | |
2986 | ASSERT(ifp->if_u1.if_data != NULL); | |
2987 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2988 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2989 | } | |
2990 | if (whichfork == XFS_DATA_FORK) { | |
2991 | if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) { | |
2992 | XFS_ERROR_REPORT("xfs_iflush_fork", | |
2993 | XFS_ERRLEVEL_LOW, mp); | |
2994 | return XFS_ERROR(EFSCORRUPTED); | |
2995 | } | |
2996 | } | |
2997 | break; | |
2998 | ||
2999 | case XFS_DINODE_FMT_EXTENTS: | |
3000 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
3001 | !(iip->ili_format.ilf_fields & extflag[whichfork])); | |
3002 | ASSERT((ifp->if_u1.if_extents != NULL) || (ifp->if_bytes == 0)); | |
3003 | ASSERT((ifp->if_u1.if_extents == NULL) || (ifp->if_bytes > 0)); | |
3004 | if ((iip->ili_format.ilf_fields & extflag[whichfork]) && | |
3005 | (ifp->if_bytes > 0)) { | |
3006 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); | |
3007 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
3008 | whichfork); | |
3009 | } | |
3010 | break; | |
3011 | ||
3012 | case XFS_DINODE_FMT_BTREE: | |
3013 | if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && | |
3014 | (ifp->if_broot_bytes > 0)) { | |
3015 | ASSERT(ifp->if_broot != NULL); | |
3016 | ASSERT(ifp->if_broot_bytes <= | |
3017 | (XFS_IFORK_SIZE(ip, whichfork) + | |
3018 | XFS_BROOT_SIZE_ADJ)); | |
3019 | xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes, | |
3020 | (xfs_bmdr_block_t *)cp, | |
3021 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
3022 | } | |
3023 | break; | |
3024 | ||
3025 | case XFS_DINODE_FMT_DEV: | |
3026 | if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { | |
3027 | ASSERT(whichfork == XFS_DATA_FORK); | |
3028 | INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev); | |
3029 | } | |
3030 | break; | |
3031 | ||
3032 | case XFS_DINODE_FMT_UUID: | |
3033 | if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { | |
3034 | ASSERT(whichfork == XFS_DATA_FORK); | |
3035 | memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, | |
3036 | sizeof(uuid_t)); | |
3037 | } | |
3038 | break; | |
3039 | ||
3040 | default: | |
3041 | ASSERT(0); | |
3042 | break; | |
3043 | } | |
3044 | ||
3045 | return 0; | |
3046 | } | |
3047 | ||
3048 | /* | |
3049 | * xfs_iflush() will write a modified inode's changes out to the | |
3050 | * inode's on disk home. The caller must have the inode lock held | |
3051 | * in at least shared mode and the inode flush semaphore must be | |
3052 | * held as well. The inode lock will still be held upon return from | |
3053 | * the call and the caller is free to unlock it. | |
3054 | * The inode flush lock will be unlocked when the inode reaches the disk. | |
3055 | * The flags indicate how the inode's buffer should be written out. | |
3056 | */ | |
3057 | int | |
3058 | xfs_iflush( | |
3059 | xfs_inode_t *ip, | |
3060 | uint flags) | |
3061 | { | |
3062 | xfs_inode_log_item_t *iip; | |
3063 | xfs_buf_t *bp; | |
3064 | xfs_dinode_t *dip; | |
3065 | xfs_mount_t *mp; | |
3066 | int error; | |
3067 | /* REFERENCED */ | |
3068 | xfs_chash_t *ch; | |
3069 | xfs_inode_t *iq; | |
3070 | int clcount; /* count of inodes clustered */ | |
3071 | int bufwasdelwri; | |
3072 | enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) }; | |
3073 | SPLDECL(s); | |
3074 | ||
3075 | XFS_STATS_INC(xs_iflush_count); | |
3076 | ||
3077 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
3078 | ASSERT(valusema(&ip->i_flock) <= 0); | |
3079 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || | |
3080 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3081 | ||
3082 | iip = ip->i_itemp; | |
3083 | mp = ip->i_mount; | |
3084 | ||
3085 | /* | |
3086 | * If the inode isn't dirty, then just release the inode | |
3087 | * flush lock and do nothing. | |
3088 | */ | |
3089 | if ((ip->i_update_core == 0) && | |
3090 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3091 | ASSERT((iip != NULL) ? | |
3092 | !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1); | |
3093 | xfs_ifunlock(ip); | |
3094 | return 0; | |
3095 | } | |
3096 | ||
3097 | /* | |
3098 | * We can't flush the inode until it is unpinned, so | |
3099 | * wait for it. We know noone new can pin it, because | |
3100 | * we are holding the inode lock shared and you need | |
3101 | * to hold it exclusively to pin the inode. | |
3102 | */ | |
3103 | xfs_iunpin_wait(ip); | |
3104 | ||
3105 | /* | |
3106 | * This may have been unpinned because the filesystem is shutting | |
3107 | * down forcibly. If that's the case we must not write this inode | |
3108 | * to disk, because the log record didn't make it to disk! | |
3109 | */ | |
3110 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
3111 | ip->i_update_core = 0; | |
3112 | if (iip) | |
3113 | iip->ili_format.ilf_fields = 0; | |
3114 | xfs_ifunlock(ip); | |
3115 | return XFS_ERROR(EIO); | |
3116 | } | |
3117 | ||
3118 | /* | |
3119 | * Get the buffer containing the on-disk inode. | |
3120 | */ | |
3121 | error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0); | |
3122 | if (error != 0) { | |
3123 | xfs_ifunlock(ip); | |
3124 | return error; | |
3125 | } | |
3126 | ||
3127 | /* | |
3128 | * Decide how buffer will be flushed out. This is done before | |
3129 | * the call to xfs_iflush_int because this field is zeroed by it. | |
3130 | */ | |
3131 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3132 | /* | |
3133 | * Flush out the inode buffer according to the directions | |
3134 | * of the caller. In the cases where the caller has given | |
3135 | * us a choice choose the non-delwri case. This is because | |
3136 | * the inode is in the AIL and we need to get it out soon. | |
3137 | */ | |
3138 | switch (flags) { | |
3139 | case XFS_IFLUSH_SYNC: | |
3140 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3141 | flags = 0; | |
3142 | break; | |
3143 | case XFS_IFLUSH_ASYNC: | |
3144 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3145 | flags = INT_ASYNC; | |
3146 | break; | |
3147 | case XFS_IFLUSH_DELWRI: | |
3148 | flags = INT_DELWRI; | |
3149 | break; | |
3150 | default: | |
3151 | ASSERT(0); | |
3152 | flags = 0; | |
3153 | break; | |
3154 | } | |
3155 | } else { | |
3156 | switch (flags) { | |
3157 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3158 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3159 | case XFS_IFLUSH_DELWRI: | |
3160 | flags = INT_DELWRI; | |
3161 | break; | |
3162 | case XFS_IFLUSH_ASYNC: | |
3163 | flags = INT_ASYNC; | |
3164 | break; | |
3165 | case XFS_IFLUSH_SYNC: | |
3166 | flags = 0; | |
3167 | break; | |
3168 | default: | |
3169 | ASSERT(0); | |
3170 | flags = 0; | |
3171 | break; | |
3172 | } | |
3173 | } | |
3174 | ||
3175 | /* | |
3176 | * First flush out the inode that xfs_iflush was called with. | |
3177 | */ | |
3178 | error = xfs_iflush_int(ip, bp); | |
3179 | if (error) { | |
3180 | goto corrupt_out; | |
3181 | } | |
3182 | ||
3183 | /* | |
3184 | * inode clustering: | |
3185 | * see if other inodes can be gathered into this write | |
3186 | */ | |
3187 | ||
3188 | ip->i_chash->chl_buf = bp; | |
3189 | ||
3190 | ch = XFS_CHASH(mp, ip->i_blkno); | |
3191 | s = mutex_spinlock(&ch->ch_lock); | |
3192 | ||
3193 | clcount = 0; | |
3194 | for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) { | |
3195 | /* | |
3196 | * Do an un-protected check to see if the inode is dirty and | |
3197 | * is a candidate for flushing. These checks will be repeated | |
3198 | * later after the appropriate locks are acquired. | |
3199 | */ | |
3200 | iip = iq->i_itemp; | |
3201 | if ((iq->i_update_core == 0) && | |
3202 | ((iip == NULL) || | |
3203 | !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
3204 | xfs_ipincount(iq) == 0) { | |
3205 | continue; | |
3206 | } | |
3207 | ||
3208 | /* | |
3209 | * Try to get locks. If any are unavailable, | |
3210 | * then this inode cannot be flushed and is skipped. | |
3211 | */ | |
3212 | ||
3213 | /* get inode locks (just i_lock) */ | |
3214 | if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) { | |
3215 | /* get inode flush lock */ | |
3216 | if (xfs_iflock_nowait(iq)) { | |
3217 | /* check if pinned */ | |
3218 | if (xfs_ipincount(iq) == 0) { | |
3219 | /* arriving here means that | |
3220 | * this inode can be flushed. | |
3221 | * first re-check that it's | |
3222 | * dirty | |
3223 | */ | |
3224 | iip = iq->i_itemp; | |
3225 | if ((iq->i_update_core != 0)|| | |
3226 | ((iip != NULL) && | |
3227 | (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3228 | clcount++; | |
3229 | error = xfs_iflush_int(iq, bp); | |
3230 | if (error) { | |
3231 | xfs_iunlock(iq, | |
3232 | XFS_ILOCK_SHARED); | |
3233 | goto cluster_corrupt_out; | |
3234 | } | |
3235 | } else { | |
3236 | xfs_ifunlock(iq); | |
3237 | } | |
3238 | } else { | |
3239 | xfs_ifunlock(iq); | |
3240 | } | |
3241 | } | |
3242 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3243 | } | |
3244 | } | |
3245 | mutex_spinunlock(&ch->ch_lock, s); | |
3246 | ||
3247 | if (clcount) { | |
3248 | XFS_STATS_INC(xs_icluster_flushcnt); | |
3249 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
3250 | } | |
3251 | ||
3252 | /* | |
3253 | * If the buffer is pinned then push on the log so we won't | |
3254 | * get stuck waiting in the write for too long. | |
3255 | */ | |
3256 | if (XFS_BUF_ISPINNED(bp)){ | |
3257 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
3258 | } | |
3259 | ||
3260 | if (flags & INT_DELWRI) { | |
3261 | xfs_bdwrite(mp, bp); | |
3262 | } else if (flags & INT_ASYNC) { | |
3263 | xfs_bawrite(mp, bp); | |
3264 | } else { | |
3265 | error = xfs_bwrite(mp, bp); | |
3266 | } | |
3267 | return error; | |
3268 | ||
3269 | corrupt_out: | |
3270 | xfs_buf_relse(bp); | |
3271 | xfs_force_shutdown(mp, XFS_CORRUPT_INCORE); | |
3272 | xfs_iflush_abort(ip); | |
3273 | /* | |
3274 | * Unlocks the flush lock | |
3275 | */ | |
3276 | return XFS_ERROR(EFSCORRUPTED); | |
3277 | ||
3278 | cluster_corrupt_out: | |
3279 | /* Corruption detected in the clustering loop. Invalidate the | |
3280 | * inode buffer and shut down the filesystem. | |
3281 | */ | |
3282 | mutex_spinunlock(&ch->ch_lock, s); | |
3283 | ||
3284 | /* | |
3285 | * Clean up the buffer. If it was B_DELWRI, just release it -- | |
3286 | * brelse can handle it with no problems. If not, shut down the | |
3287 | * filesystem before releasing the buffer. | |
3288 | */ | |
3289 | if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) { | |
3290 | xfs_buf_relse(bp); | |
3291 | } | |
3292 | ||
3293 | xfs_force_shutdown(mp, XFS_CORRUPT_INCORE); | |
3294 | ||
3295 | if(!bufwasdelwri) { | |
3296 | /* | |
3297 | * Just like incore_relse: if we have b_iodone functions, | |
3298 | * mark the buffer as an error and call them. Otherwise | |
3299 | * mark it as stale and brelse. | |
3300 | */ | |
3301 | if (XFS_BUF_IODONE_FUNC(bp)) { | |
3302 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
3303 | XFS_BUF_UNDONE(bp); | |
3304 | XFS_BUF_STALE(bp); | |
3305 | XFS_BUF_SHUT(bp); | |
3306 | XFS_BUF_ERROR(bp,EIO); | |
3307 | xfs_biodone(bp); | |
3308 | } else { | |
3309 | XFS_BUF_STALE(bp); | |
3310 | xfs_buf_relse(bp); | |
3311 | } | |
3312 | } | |
3313 | ||
3314 | xfs_iflush_abort(iq); | |
3315 | /* | |
3316 | * Unlocks the flush lock | |
3317 | */ | |
3318 | return XFS_ERROR(EFSCORRUPTED); | |
3319 | } | |
3320 | ||
3321 | ||
3322 | STATIC int | |
3323 | xfs_iflush_int( | |
3324 | xfs_inode_t *ip, | |
3325 | xfs_buf_t *bp) | |
3326 | { | |
3327 | xfs_inode_log_item_t *iip; | |
3328 | xfs_dinode_t *dip; | |
3329 | xfs_mount_t *mp; | |
3330 | #ifdef XFS_TRANS_DEBUG | |
3331 | int first; | |
3332 | #endif | |
3333 | SPLDECL(s); | |
3334 | ||
3335 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
3336 | ASSERT(valusema(&ip->i_flock) <= 0); | |
3337 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || | |
3338 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3339 | ||
3340 | iip = ip->i_itemp; | |
3341 | mp = ip->i_mount; | |
3342 | ||
3343 | ||
3344 | /* | |
3345 | * If the inode isn't dirty, then just release the inode | |
3346 | * flush lock and do nothing. | |
3347 | */ | |
3348 | if ((ip->i_update_core == 0) && | |
3349 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3350 | xfs_ifunlock(ip); | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | /* set *dip = inode's place in the buffer */ | |
3355 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset); | |
3356 | ||
3357 | /* | |
3358 | * Clear i_update_core before copying out the data. | |
3359 | * This is for coordination with our timestamp updates | |
3360 | * that don't hold the inode lock. They will always | |
3361 | * update the timestamps BEFORE setting i_update_core, | |
3362 | * so if we clear i_update_core after they set it we | |
3363 | * are guaranteed to see their updates to the timestamps. | |
3364 | * I believe that this depends on strongly ordered memory | |
3365 | * semantics, but we have that. We use the SYNCHRONIZE | |
3366 | * macro to make sure that the compiler does not reorder | |
3367 | * the i_update_core access below the data copy below. | |
3368 | */ | |
3369 | ip->i_update_core = 0; | |
3370 | SYNCHRONIZE(); | |
3371 | ||
3372 | if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC, | |
3373 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { | |
3374 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3375 | "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3376 | ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip); | |
3377 | goto corrupt_out; | |
3378 | } | |
3379 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
3380 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
3381 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3382 | "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
3383 | ip->i_ino, ip, ip->i_d.di_magic); | |
3384 | goto corrupt_out; | |
3385 | } | |
3386 | if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
3387 | if (XFS_TEST_ERROR( | |
3388 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3389 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3390 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
3391 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3392 | "xfs_iflush: Bad regular inode %Lu, ptr 0x%p", | |
3393 | ip->i_ino, ip); | |
3394 | goto corrupt_out; | |
3395 | } | |
3396 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { | |
3397 | if (XFS_TEST_ERROR( | |
3398 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3399 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3400 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3401 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
3402 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3403 | "xfs_iflush: Bad directory inode %Lu, ptr 0x%p", | |
3404 | ip->i_ino, ip); | |
3405 | goto corrupt_out; | |
3406 | } | |
3407 | } | |
3408 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3409 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3410 | XFS_RANDOM_IFLUSH_5)) { | |
3411 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3412 | "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3413 | ip->i_ino, | |
3414 | ip->i_d.di_nextents + ip->i_d.di_anextents, | |
3415 | ip->i_d.di_nblocks, | |
3416 | ip); | |
3417 | goto corrupt_out; | |
3418 | } | |
3419 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3420 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
3421 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3422 | "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3423 | ip->i_ino, ip->i_d.di_forkoff, ip); | |
3424 | goto corrupt_out; | |
3425 | } | |
3426 | /* | |
3427 | * bump the flush iteration count, used to detect flushes which | |
3428 | * postdate a log record during recovery. | |
3429 | */ | |
3430 | ||
3431 | ip->i_d.di_flushiter++; | |
3432 | ||
3433 | /* | |
3434 | * Copy the dirty parts of the inode into the on-disk | |
3435 | * inode. We always copy out the core of the inode, | |
3436 | * because if the inode is dirty at all the core must | |
3437 | * be. | |
3438 | */ | |
3439 | xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1); | |
3440 | ||
3441 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3442 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3443 | ip->i_d.di_flushiter = 0; | |
3444 | ||
3445 | /* | |
3446 | * If this is really an old format inode and the superblock version | |
3447 | * has not been updated to support only new format inodes, then | |
3448 | * convert back to the old inode format. If the superblock version | |
3449 | * has been updated, then make the conversion permanent. | |
3450 | */ | |
3451 | ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 || | |
3452 | XFS_SB_VERSION_HASNLINK(&mp->m_sb)); | |
3453 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
3454 | if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { | |
3455 | /* | |
3456 | * Convert it back. | |
3457 | */ | |
3458 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
3459 | INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink); | |
3460 | } else { | |
3461 | /* | |
3462 | * The superblock version has already been bumped, | |
3463 | * so just make the conversion to the new inode | |
3464 | * format permanent. | |
3465 | */ | |
3466 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
3467 | INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2); | |
3468 | ip->i_d.di_onlink = 0; | |
3469 | dip->di_core.di_onlink = 0; | |
3470 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
3471 | memset(&(dip->di_core.di_pad[0]), 0, | |
3472 | sizeof(dip->di_core.di_pad)); | |
3473 | ASSERT(ip->i_d.di_projid == 0); | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) { | |
3478 | goto corrupt_out; | |
3479 | } | |
3480 | ||
3481 | if (XFS_IFORK_Q(ip)) { | |
3482 | /* | |
3483 | * The only error from xfs_iflush_fork is on the data fork. | |
3484 | */ | |
3485 | (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
3486 | } | |
3487 | xfs_inobp_check(mp, bp); | |
3488 | ||
3489 | /* | |
3490 | * We've recorded everything logged in the inode, so we'd | |
3491 | * like to clear the ilf_fields bits so we don't log and | |
3492 | * flush things unnecessarily. However, we can't stop | |
3493 | * logging all this information until the data we've copied | |
3494 | * into the disk buffer is written to disk. If we did we might | |
3495 | * overwrite the copy of the inode in the log with all the | |
3496 | * data after re-logging only part of it, and in the face of | |
3497 | * a crash we wouldn't have all the data we need to recover. | |
3498 | * | |
3499 | * What we do is move the bits to the ili_last_fields field. | |
3500 | * When logging the inode, these bits are moved back to the | |
3501 | * ilf_fields field. In the xfs_iflush_done() routine we | |
3502 | * clear ili_last_fields, since we know that the information | |
3503 | * those bits represent is permanently on disk. As long as | |
3504 | * the flush completes before the inode is logged again, then | |
3505 | * both ilf_fields and ili_last_fields will be cleared. | |
3506 | * | |
3507 | * We can play with the ilf_fields bits here, because the inode | |
3508 | * lock must be held exclusively in order to set bits there | |
3509 | * and the flush lock protects the ili_last_fields bits. | |
3510 | * Set ili_logged so the flush done | |
3511 | * routine can tell whether or not to look in the AIL. | |
3512 | * Also, store the current LSN of the inode so that we can tell | |
3513 | * whether the item has moved in the AIL from xfs_iflush_done(). | |
3514 | * In order to read the lsn we need the AIL lock, because | |
3515 | * it is a 64 bit value that cannot be read atomically. | |
3516 | */ | |
3517 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3518 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
3519 | iip->ili_format.ilf_fields = 0; | |
3520 | iip->ili_logged = 1; | |
3521 | ||
3522 | ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */ | |
3523 | AIL_LOCK(mp,s); | |
3524 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
3525 | AIL_UNLOCK(mp, s); | |
3526 | ||
3527 | /* | |
3528 | * Attach the function xfs_iflush_done to the inode's | |
3529 | * buffer. This will remove the inode from the AIL | |
3530 | * and unlock the inode's flush lock when the inode is | |
3531 | * completely written to disk. | |
3532 | */ | |
3533 | xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
3534 | xfs_iflush_done, (xfs_log_item_t *)iip); | |
3535 | ||
3536 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
3537 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL); | |
3538 | } else { | |
3539 | /* | |
3540 | * We're flushing an inode which is not in the AIL and has | |
3541 | * not been logged but has i_update_core set. For this | |
3542 | * case we can use a B_DELWRI flush and immediately drop | |
3543 | * the inode flush lock because we can avoid the whole | |
3544 | * AIL state thing. It's OK to drop the flush lock now, | |
3545 | * because we've already locked the buffer and to do anything | |
3546 | * you really need both. | |
3547 | */ | |
3548 | if (iip != NULL) { | |
3549 | ASSERT(iip->ili_logged == 0); | |
3550 | ASSERT(iip->ili_last_fields == 0); | |
3551 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
3552 | } | |
3553 | xfs_ifunlock(ip); | |
3554 | } | |
3555 | ||
3556 | return 0; | |
3557 | ||
3558 | corrupt_out: | |
3559 | return XFS_ERROR(EFSCORRUPTED); | |
3560 | } | |
3561 | ||
3562 | ||
3563 | /* | |
efa80278 | 3564 | * Flush all inactive inodes in mp. |
1da177e4 | 3565 | */ |
efa80278 | 3566 | void |
1da177e4 | 3567 | xfs_iflush_all( |
efa80278 | 3568 | xfs_mount_t *mp) |
1da177e4 | 3569 | { |
1da177e4 | 3570 | xfs_inode_t *ip; |
1da177e4 LT |
3571 | vnode_t *vp; |
3572 | ||
efa80278 CH |
3573 | again: |
3574 | XFS_MOUNT_ILOCK(mp); | |
3575 | ip = mp->m_inodes; | |
3576 | if (ip == NULL) | |
3577 | goto out; | |
1da177e4 | 3578 | |
efa80278 CH |
3579 | do { |
3580 | /* Make sure we skip markers inserted by sync */ | |
3581 | if (ip->i_mount == NULL) { | |
3582 | ip = ip->i_mnext; | |
3583 | continue; | |
3584 | } | |
1da177e4 | 3585 | |
efa80278 CH |
3586 | vp = XFS_ITOV_NULL(ip); |
3587 | if (!vp) { | |
1da177e4 | 3588 | XFS_MOUNT_IUNLOCK(mp); |
efa80278 CH |
3589 | xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC); |
3590 | goto again; | |
3591 | } | |
1da177e4 | 3592 | |
efa80278 | 3593 | ASSERT(vn_count(vp) == 0); |
1da177e4 | 3594 | |
efa80278 CH |
3595 | ip = ip->i_mnext; |
3596 | } while (ip != mp->m_inodes); | |
3597 | out: | |
1da177e4 | 3598 | XFS_MOUNT_IUNLOCK(mp); |
1da177e4 LT |
3599 | } |
3600 | ||
1da177e4 LT |
3601 | /* |
3602 | * xfs_iaccess: check accessibility of inode for mode. | |
3603 | */ | |
3604 | int | |
3605 | xfs_iaccess( | |
3606 | xfs_inode_t *ip, | |
3607 | mode_t mode, | |
3608 | cred_t *cr) | |
3609 | { | |
3610 | int error; | |
3611 | mode_t orgmode = mode; | |
3612 | struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip)); | |
3613 | ||
3614 | if (mode & S_IWUSR) { | |
3615 | umode_t imode = inode->i_mode; | |
3616 | ||
3617 | if (IS_RDONLY(inode) && | |
3618 | (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode))) | |
3619 | return XFS_ERROR(EROFS); | |
3620 | ||
3621 | if (IS_IMMUTABLE(inode)) | |
3622 | return XFS_ERROR(EACCES); | |
3623 | } | |
3624 | ||
3625 | /* | |
3626 | * If there's an Access Control List it's used instead of | |
3627 | * the mode bits. | |
3628 | */ | |
3629 | if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1) | |
3630 | return error ? XFS_ERROR(error) : 0; | |
3631 | ||
3632 | if (current_fsuid(cr) != ip->i_d.di_uid) { | |
3633 | mode >>= 3; | |
3634 | if (!in_group_p((gid_t)ip->i_d.di_gid)) | |
3635 | mode >>= 3; | |
3636 | } | |
3637 | ||
3638 | /* | |
3639 | * If the DACs are ok we don't need any capability check. | |
3640 | */ | |
3641 | if ((ip->i_d.di_mode & mode) == mode) | |
3642 | return 0; | |
3643 | /* | |
3644 | * Read/write DACs are always overridable. | |
3645 | * Executable DACs are overridable if at least one exec bit is set. | |
3646 | */ | |
3647 | if (!(orgmode & S_IXUSR) || | |
3648 | (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) | |
3649 | if (capable_cred(cr, CAP_DAC_OVERRIDE)) | |
3650 | return 0; | |
3651 | ||
3652 | if ((orgmode == S_IRUSR) || | |
3653 | (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) { | |
3654 | if (capable_cred(cr, CAP_DAC_READ_SEARCH)) | |
3655 | return 0; | |
3656 | #ifdef NOISE | |
3657 | cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode); | |
3658 | #endif /* NOISE */ | |
3659 | return XFS_ERROR(EACCES); | |
3660 | } | |
3661 | return XFS_ERROR(EACCES); | |
3662 | } | |
3663 | ||
3664 | /* | |
3665 | * xfs_iroundup: round up argument to next power of two | |
3666 | */ | |
3667 | uint | |
3668 | xfs_iroundup( | |
3669 | uint v) | |
3670 | { | |
3671 | int i; | |
3672 | uint m; | |
3673 | ||
3674 | if ((v & (v - 1)) == 0) | |
3675 | return v; | |
3676 | ASSERT((v & 0x80000000) == 0); | |
3677 | if ((v & (v + 1)) == 0) | |
3678 | return v + 1; | |
3679 | for (i = 0, m = 1; i < 31; i++, m <<= 1) { | |
3680 | if (v & m) | |
3681 | continue; | |
3682 | v |= m; | |
3683 | if ((v & (v + 1)) == 0) | |
3684 | return v + 1; | |
3685 | } | |
3686 | ASSERT(0); | |
3687 | return( 0 ); | |
3688 | } | |
3689 | ||
1da177e4 LT |
3690 | #ifdef XFS_ILOCK_TRACE |
3691 | ktrace_t *xfs_ilock_trace_buf; | |
3692 | ||
3693 | void | |
3694 | xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra) | |
3695 | { | |
3696 | ktrace_enter(ip->i_lock_trace, | |
3697 | (void *)ip, | |
3698 | (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */ | |
3699 | (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */ | |
3700 | (void *)ra, /* caller of ilock */ | |
3701 | (void *)(unsigned long)current_cpu(), | |
3702 | (void *)(unsigned long)current_pid(), | |
3703 | NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL); | |
3704 | } | |
3705 | #endif |