]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
40ebd81d RD |
18 | #include <linux/log2.h> |
19 | ||
1da177e4 | 20 | #include "xfs.h" |
a844f451 | 21 | #include "xfs_fs.h" |
1da177e4 | 22 | #include "xfs_types.h" |
a844f451 | 23 | #include "xfs_bit.h" |
1da177e4 | 24 | #include "xfs_log.h" |
a844f451 NS |
25 | #include "xfs_inum.h" |
26 | #include "xfs_imap.h" | |
1da177e4 LT |
27 | #include "xfs_trans.h" |
28 | #include "xfs_trans_priv.h" | |
29 | #include "xfs_sb.h" | |
30 | #include "xfs_ag.h" | |
1da177e4 LT |
31 | #include "xfs_dir2.h" |
32 | #include "xfs_dmapi.h" | |
33 | #include "xfs_mount.h" | |
1da177e4 | 34 | #include "xfs_bmap_btree.h" |
a844f451 | 35 | #include "xfs_alloc_btree.h" |
1da177e4 | 36 | #include "xfs_ialloc_btree.h" |
1da177e4 | 37 | #include "xfs_dir2_sf.h" |
a844f451 | 38 | #include "xfs_attr_sf.h" |
1da177e4 | 39 | #include "xfs_dinode.h" |
1da177e4 | 40 | #include "xfs_inode.h" |
1da177e4 | 41 | #include "xfs_buf_item.h" |
a844f451 NS |
42 | #include "xfs_inode_item.h" |
43 | #include "xfs_btree.h" | |
8c4ed633 | 44 | #include "xfs_btree_trace.h" |
a844f451 NS |
45 | #include "xfs_alloc.h" |
46 | #include "xfs_ialloc.h" | |
47 | #include "xfs_bmap.h" | |
1da177e4 LT |
48 | #include "xfs_rw.h" |
49 | #include "xfs_error.h" | |
1da177e4 LT |
50 | #include "xfs_utils.h" |
51 | #include "xfs_dir2_trace.h" | |
52 | #include "xfs_quota.h" | |
1da177e4 | 53 | #include "xfs_acl.h" |
2a82b8be | 54 | #include "xfs_filestream.h" |
739bfb2a | 55 | #include "xfs_vnodeops.h" |
1da177e4 | 56 | |
1da177e4 LT |
57 | kmem_zone_t *xfs_ifork_zone; |
58 | kmem_zone_t *xfs_inode_zone; | |
1da177e4 LT |
59 | |
60 | /* | |
61 | * Used in xfs_itruncate(). This is the maximum number of extents | |
62 | * freed from a file in a single transaction. | |
63 | */ | |
64 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
65 | ||
66 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
67 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
68 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
69 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
70 | ||
1da177e4 LT |
71 | #ifdef DEBUG |
72 | /* | |
73 | * Make sure that the extents in the given memory buffer | |
74 | * are valid. | |
75 | */ | |
76 | STATIC void | |
77 | xfs_validate_extents( | |
4eea22f0 | 78 | xfs_ifork_t *ifp, |
1da177e4 | 79 | int nrecs, |
1da177e4 LT |
80 | xfs_exntfmt_t fmt) |
81 | { | |
82 | xfs_bmbt_irec_t irec; | |
a6f64d4a | 83 | xfs_bmbt_rec_host_t rec; |
1da177e4 LT |
84 | int i; |
85 | ||
86 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a CH |
87 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
88 | rec.l0 = get_unaligned(&ep->l0); | |
89 | rec.l1 = get_unaligned(&ep->l1); | |
90 | xfs_bmbt_get_all(&rec, &irec); | |
1da177e4 LT |
91 | if (fmt == XFS_EXTFMT_NOSTATE) |
92 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
1da177e4 LT |
93 | } |
94 | } | |
95 | #else /* DEBUG */ | |
a6f64d4a | 96 | #define xfs_validate_extents(ifp, nrecs, fmt) |
1da177e4 LT |
97 | #endif /* DEBUG */ |
98 | ||
99 | /* | |
100 | * Check that none of the inode's in the buffer have a next | |
101 | * unlinked field of 0. | |
102 | */ | |
103 | #if defined(DEBUG) | |
104 | void | |
105 | xfs_inobp_check( | |
106 | xfs_mount_t *mp, | |
107 | xfs_buf_t *bp) | |
108 | { | |
109 | int i; | |
110 | int j; | |
111 | xfs_dinode_t *dip; | |
112 | ||
113 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
114 | ||
115 | for (i = 0; i < j; i++) { | |
116 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
117 | i * mp->m_sb.sb_inodesize); | |
118 | if (!dip->di_next_unlinked) { | |
119 | xfs_fs_cmn_err(CE_ALERT, mp, | |
120 | "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", | |
121 | bp); | |
122 | ASSERT(dip->di_next_unlinked); | |
123 | } | |
124 | } | |
125 | } | |
126 | #endif | |
127 | ||
4ae29b43 DC |
128 | /* |
129 | * Find the buffer associated with the given inode map | |
130 | * We do basic validation checks on the buffer once it has been | |
131 | * retrieved from disk. | |
132 | */ | |
133 | STATIC int | |
134 | xfs_imap_to_bp( | |
135 | xfs_mount_t *mp, | |
136 | xfs_trans_t *tp, | |
137 | xfs_imap_t *imap, | |
138 | xfs_buf_t **bpp, | |
139 | uint buf_flags, | |
140 | uint imap_flags) | |
141 | { | |
142 | int error; | |
143 | int i; | |
144 | int ni; | |
145 | xfs_buf_t *bp; | |
146 | ||
147 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, | |
a3f74ffb | 148 | (int)imap->im_len, buf_flags, &bp); |
4ae29b43 | 149 | if (error) { |
a3f74ffb DC |
150 | if (error != EAGAIN) { |
151 | cmn_err(CE_WARN, | |
152 | "xfs_imap_to_bp: xfs_trans_read_buf()returned " | |
4ae29b43 DC |
153 | "an error %d on %s. Returning error.", |
154 | error, mp->m_fsname); | |
a3f74ffb DC |
155 | } else { |
156 | ASSERT(buf_flags & XFS_BUF_TRYLOCK); | |
157 | } | |
4ae29b43 DC |
158 | return error; |
159 | } | |
160 | ||
161 | /* | |
162 | * Validate the magic number and version of every inode in the buffer | |
163 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
164 | */ | |
165 | #ifdef DEBUG | |
166 | ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog; | |
167 | #else /* usual case */ | |
168 | ni = 1; | |
169 | #endif | |
170 | ||
171 | for (i = 0; i < ni; i++) { | |
172 | int di_ok; | |
173 | xfs_dinode_t *dip; | |
174 | ||
175 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
176 | (i << mp->m_sb.sb_inodelog)); | |
177 | di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC && | |
178 | XFS_DINODE_GOOD_VERSION(dip->di_core.di_version); | |
179 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, | |
180 | XFS_ERRTAG_ITOBP_INOTOBP, | |
181 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
182 | if (imap_flags & XFS_IMAP_BULKSTAT) { | |
183 | xfs_trans_brelse(tp, bp); | |
184 | return XFS_ERROR(EINVAL); | |
185 | } | |
186 | XFS_CORRUPTION_ERROR("xfs_imap_to_bp", | |
187 | XFS_ERRLEVEL_HIGH, mp, dip); | |
188 | #ifdef DEBUG | |
189 | cmn_err(CE_PANIC, | |
190 | "Device %s - bad inode magic/vsn " | |
191 | "daddr %lld #%d (magic=%x)", | |
192 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
193 | (unsigned long long)imap->im_blkno, i, | |
194 | be16_to_cpu(dip->di_core.di_magic)); | |
195 | #endif | |
196 | xfs_trans_brelse(tp, bp); | |
197 | return XFS_ERROR(EFSCORRUPTED); | |
198 | } | |
199 | } | |
200 | ||
201 | xfs_inobp_check(mp, bp); | |
202 | ||
203 | /* | |
204 | * Mark the buffer as an inode buffer now that it looks good | |
205 | */ | |
206 | XFS_BUF_SET_VTYPE(bp, B_FS_INO); | |
207 | ||
208 | *bpp = bp; | |
209 | return 0; | |
210 | } | |
211 | ||
1da177e4 LT |
212 | /* |
213 | * This routine is called to map an inode number within a file | |
214 | * system to the buffer containing the on-disk version of the | |
215 | * inode. It returns a pointer to the buffer containing the | |
216 | * on-disk inode in the bpp parameter, and in the dip parameter | |
217 | * it returns a pointer to the on-disk inode within that buffer. | |
218 | * | |
219 | * If a non-zero error is returned, then the contents of bpp and | |
220 | * dipp are undefined. | |
221 | * | |
222 | * Use xfs_imap() to determine the size and location of the | |
223 | * buffer to read from disk. | |
224 | */ | |
c679eef0 | 225 | int |
1da177e4 LT |
226 | xfs_inotobp( |
227 | xfs_mount_t *mp, | |
228 | xfs_trans_t *tp, | |
229 | xfs_ino_t ino, | |
230 | xfs_dinode_t **dipp, | |
231 | xfs_buf_t **bpp, | |
c679eef0 CH |
232 | int *offset, |
233 | uint imap_flags) | |
1da177e4 | 234 | { |
1da177e4 LT |
235 | xfs_imap_t imap; |
236 | xfs_buf_t *bp; | |
237 | int error; | |
1da177e4 | 238 | |
1da177e4 | 239 | imap.im_blkno = 0; |
c679eef0 | 240 | error = xfs_imap(mp, tp, ino, &imap, imap_flags | XFS_IMAP_LOOKUP); |
4ae29b43 | 241 | if (error) |
1da177e4 | 242 | return error; |
1da177e4 | 243 | |
c679eef0 | 244 | error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags); |
4ae29b43 | 245 | if (error) |
1da177e4 | 246 | return error; |
1da177e4 | 247 | |
1da177e4 LT |
248 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); |
249 | *bpp = bp; | |
250 | *offset = imap.im_boffset; | |
251 | return 0; | |
252 | } | |
253 | ||
254 | ||
255 | /* | |
256 | * This routine is called to map an inode to the buffer containing | |
257 | * the on-disk version of the inode. It returns a pointer to the | |
258 | * buffer containing the on-disk inode in the bpp parameter, and in | |
259 | * the dip parameter it returns a pointer to the on-disk inode within | |
260 | * that buffer. | |
261 | * | |
262 | * If a non-zero error is returned, then the contents of bpp and | |
263 | * dipp are undefined. | |
264 | * | |
265 | * If the inode is new and has not yet been initialized, use xfs_imap() | |
266 | * to determine the size and location of the buffer to read from disk. | |
267 | * If the inode has already been mapped to its buffer and read in once, | |
268 | * then use the mapping information stored in the inode rather than | |
269 | * calling xfs_imap(). This allows us to avoid the overhead of looking | |
270 | * at the inode btree for small block file systems (see xfs_dilocate()). | |
271 | * We can tell whether the inode has been mapped in before by comparing | |
272 | * its disk block address to 0. Only uninitialized inodes will have | |
273 | * 0 for the disk block address. | |
274 | */ | |
275 | int | |
276 | xfs_itobp( | |
277 | xfs_mount_t *mp, | |
278 | xfs_trans_t *tp, | |
279 | xfs_inode_t *ip, | |
280 | xfs_dinode_t **dipp, | |
281 | xfs_buf_t **bpp, | |
b12dd342 | 282 | xfs_daddr_t bno, |
a3f74ffb DC |
283 | uint imap_flags, |
284 | uint buf_flags) | |
1da177e4 | 285 | { |
4d1a2ed3 | 286 | xfs_imap_t imap; |
1da177e4 LT |
287 | xfs_buf_t *bp; |
288 | int error; | |
1da177e4 LT |
289 | |
290 | if (ip->i_blkno == (xfs_daddr_t)0) { | |
1da177e4 | 291 | imap.im_blkno = bno; |
4ae29b43 DC |
292 | error = xfs_imap(mp, tp, ip->i_ino, &imap, |
293 | XFS_IMAP_LOOKUP | imap_flags); | |
294 | if (error) | |
1da177e4 | 295 | return error; |
1da177e4 | 296 | |
1da177e4 LT |
297 | /* |
298 | * Fill in the fields in the inode that will be used to | |
299 | * map the inode to its buffer from now on. | |
300 | */ | |
301 | ip->i_blkno = imap.im_blkno; | |
302 | ip->i_len = imap.im_len; | |
303 | ip->i_boffset = imap.im_boffset; | |
304 | } else { | |
305 | /* | |
306 | * We've already mapped the inode once, so just use the | |
307 | * mapping that we saved the first time. | |
308 | */ | |
309 | imap.im_blkno = ip->i_blkno; | |
310 | imap.im_len = ip->i_len; | |
311 | imap.im_boffset = ip->i_boffset; | |
312 | } | |
313 | ASSERT(bno == 0 || bno == imap.im_blkno); | |
314 | ||
a3f74ffb | 315 | error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags); |
4ae29b43 | 316 | if (error) |
1da177e4 | 317 | return error; |
1da177e4 | 318 | |
a3f74ffb DC |
319 | if (!bp) { |
320 | ASSERT(buf_flags & XFS_BUF_TRYLOCK); | |
321 | ASSERT(tp == NULL); | |
322 | *bpp = NULL; | |
323 | return EAGAIN; | |
324 | } | |
325 | ||
1da177e4 LT |
326 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); |
327 | *bpp = bp; | |
328 | return 0; | |
329 | } | |
330 | ||
331 | /* | |
332 | * Move inode type and inode format specific information from the | |
333 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
334 | * this means set if_rdev to the proper value. For files, directories, | |
335 | * and symlinks this means to bring in the in-line data or extent | |
336 | * pointers. For a file in B-tree format, only the root is immediately | |
337 | * brought in-core. The rest will be in-lined in if_extents when it | |
338 | * is first referenced (see xfs_iread_extents()). | |
339 | */ | |
340 | STATIC int | |
341 | xfs_iformat( | |
342 | xfs_inode_t *ip, | |
343 | xfs_dinode_t *dip) | |
344 | { | |
345 | xfs_attr_shortform_t *atp; | |
346 | int size; | |
347 | int error; | |
348 | xfs_fsize_t di_size; | |
349 | ip->i_df.if_ext_max = | |
350 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
351 | error = 0; | |
352 | ||
347d1c01 CH |
353 | if (unlikely(be32_to_cpu(dip->di_core.di_nextents) + |
354 | be16_to_cpu(dip->di_core.di_anextents) > | |
355 | be64_to_cpu(dip->di_core.di_nblocks))) { | |
3762ec6b NS |
356 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
357 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", | |
1da177e4 | 358 | (unsigned long long)ip->i_ino, |
347d1c01 CH |
359 | (int)(be32_to_cpu(dip->di_core.di_nextents) + |
360 | be16_to_cpu(dip->di_core.di_anextents)), | |
1da177e4 | 361 | (unsigned long long) |
347d1c01 | 362 | be64_to_cpu(dip->di_core.di_nblocks)); |
1da177e4 LT |
363 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
364 | ip->i_mount, dip); | |
365 | return XFS_ERROR(EFSCORRUPTED); | |
366 | } | |
367 | ||
347d1c01 | 368 | if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
3762ec6b NS |
369 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
370 | "corrupt dinode %Lu, forkoff = 0x%x.", | |
1da177e4 | 371 | (unsigned long long)ip->i_ino, |
347d1c01 | 372 | dip->di_core.di_forkoff); |
1da177e4 LT |
373 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
374 | ip->i_mount, dip); | |
375 | return XFS_ERROR(EFSCORRUPTED); | |
376 | } | |
377 | ||
378 | switch (ip->i_d.di_mode & S_IFMT) { | |
379 | case S_IFIFO: | |
380 | case S_IFCHR: | |
381 | case S_IFBLK: | |
382 | case S_IFSOCK: | |
347d1c01 | 383 | if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) { |
1da177e4 LT |
384 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
385 | ip->i_mount, dip); | |
386 | return XFS_ERROR(EFSCORRUPTED); | |
387 | } | |
388 | ip->i_d.di_size = 0; | |
ba87ea69 | 389 | ip->i_size = 0; |
347d1c01 | 390 | ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev); |
1da177e4 LT |
391 | break; |
392 | ||
393 | case S_IFREG: | |
394 | case S_IFLNK: | |
395 | case S_IFDIR: | |
347d1c01 | 396 | switch (dip->di_core.di_format) { |
1da177e4 LT |
397 | case XFS_DINODE_FMT_LOCAL: |
398 | /* | |
399 | * no local regular files yet | |
400 | */ | |
347d1c01 | 401 | if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) { |
3762ec6b NS |
402 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
403 | "corrupt inode %Lu " | |
404 | "(local format for regular file).", | |
1da177e4 LT |
405 | (unsigned long long) ip->i_ino); |
406 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
407 | XFS_ERRLEVEL_LOW, | |
408 | ip->i_mount, dip); | |
409 | return XFS_ERROR(EFSCORRUPTED); | |
410 | } | |
411 | ||
347d1c01 | 412 | di_size = be64_to_cpu(dip->di_core.di_size); |
1da177e4 | 413 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
3762ec6b NS |
414 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
415 | "corrupt inode %Lu " | |
416 | "(bad size %Ld for local inode).", | |
1da177e4 LT |
417 | (unsigned long long) ip->i_ino, |
418 | (long long) di_size); | |
419 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
420 | XFS_ERRLEVEL_LOW, | |
421 | ip->i_mount, dip); | |
422 | return XFS_ERROR(EFSCORRUPTED); | |
423 | } | |
424 | ||
425 | size = (int)di_size; | |
426 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
427 | break; | |
428 | case XFS_DINODE_FMT_EXTENTS: | |
429 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
430 | break; | |
431 | case XFS_DINODE_FMT_BTREE: | |
432 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
433 | break; | |
434 | default: | |
435 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
436 | ip->i_mount); | |
437 | return XFS_ERROR(EFSCORRUPTED); | |
438 | } | |
439 | break; | |
440 | ||
441 | default: | |
442 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
443 | return XFS_ERROR(EFSCORRUPTED); | |
444 | } | |
445 | if (error) { | |
446 | return error; | |
447 | } | |
448 | if (!XFS_DFORK_Q(dip)) | |
449 | return 0; | |
450 | ASSERT(ip->i_afp == NULL); | |
451 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
452 | ip->i_afp->if_ext_max = | |
453 | XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
347d1c01 | 454 | switch (dip->di_core.di_aformat) { |
1da177e4 LT |
455 | case XFS_DINODE_FMT_LOCAL: |
456 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
3b244aa8 | 457 | size = be16_to_cpu(atp->hdr.totsize); |
1da177e4 LT |
458 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
459 | break; | |
460 | case XFS_DINODE_FMT_EXTENTS: | |
461 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
462 | break; | |
463 | case XFS_DINODE_FMT_BTREE: | |
464 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
465 | break; | |
466 | default: | |
467 | error = XFS_ERROR(EFSCORRUPTED); | |
468 | break; | |
469 | } | |
470 | if (error) { | |
471 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
472 | ip->i_afp = NULL; | |
473 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
474 | } | |
475 | return error; | |
476 | } | |
477 | ||
478 | /* | |
479 | * The file is in-lined in the on-disk inode. | |
480 | * If it fits into if_inline_data, then copy | |
481 | * it there, otherwise allocate a buffer for it | |
482 | * and copy the data there. Either way, set | |
483 | * if_data to point at the data. | |
484 | * If we allocate a buffer for the data, make | |
485 | * sure that its size is a multiple of 4 and | |
486 | * record the real size in i_real_bytes. | |
487 | */ | |
488 | STATIC int | |
489 | xfs_iformat_local( | |
490 | xfs_inode_t *ip, | |
491 | xfs_dinode_t *dip, | |
492 | int whichfork, | |
493 | int size) | |
494 | { | |
495 | xfs_ifork_t *ifp; | |
496 | int real_size; | |
497 | ||
498 | /* | |
499 | * If the size is unreasonable, then something | |
500 | * is wrong and we just bail out rather than crash in | |
501 | * kmem_alloc() or memcpy() below. | |
502 | */ | |
503 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
504 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
505 | "corrupt inode %Lu " | |
506 | "(bad size %d for local fork, size = %d).", | |
1da177e4 LT |
507 | (unsigned long long) ip->i_ino, size, |
508 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
509 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
510 | ip->i_mount, dip); | |
511 | return XFS_ERROR(EFSCORRUPTED); | |
512 | } | |
513 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
514 | real_size = 0; | |
515 | if (size == 0) | |
516 | ifp->if_u1.if_data = NULL; | |
517 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
518 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
519 | else { | |
520 | real_size = roundup(size, 4); | |
521 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
522 | } | |
523 | ifp->if_bytes = size; | |
524 | ifp->if_real_bytes = real_size; | |
525 | if (size) | |
526 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
527 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
528 | ifp->if_flags |= XFS_IFINLINE; | |
529 | return 0; | |
530 | } | |
531 | ||
532 | /* | |
533 | * The file consists of a set of extents all | |
534 | * of which fit into the on-disk inode. | |
535 | * If there are few enough extents to fit into | |
536 | * the if_inline_ext, then copy them there. | |
537 | * Otherwise allocate a buffer for them and copy | |
538 | * them into it. Either way, set if_extents | |
539 | * to point at the extents. | |
540 | */ | |
541 | STATIC int | |
542 | xfs_iformat_extents( | |
543 | xfs_inode_t *ip, | |
544 | xfs_dinode_t *dip, | |
545 | int whichfork) | |
546 | { | |
a6f64d4a | 547 | xfs_bmbt_rec_t *dp; |
1da177e4 LT |
548 | xfs_ifork_t *ifp; |
549 | int nex; | |
1da177e4 LT |
550 | int size; |
551 | int i; | |
552 | ||
553 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
554 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
555 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
556 | ||
557 | /* | |
558 | * If the number of extents is unreasonable, then something | |
559 | * is wrong and we just bail out rather than crash in | |
560 | * kmem_alloc() or memcpy() below. | |
561 | */ | |
562 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
563 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
564 | "corrupt inode %Lu ((a)extents = %d).", | |
1da177e4 LT |
565 | (unsigned long long) ip->i_ino, nex); |
566 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
567 | ip->i_mount, dip); | |
568 | return XFS_ERROR(EFSCORRUPTED); | |
569 | } | |
570 | ||
4eea22f0 | 571 | ifp->if_real_bytes = 0; |
1da177e4 LT |
572 | if (nex == 0) |
573 | ifp->if_u1.if_extents = NULL; | |
574 | else if (nex <= XFS_INLINE_EXTS) | |
575 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
4eea22f0 MK |
576 | else |
577 | xfs_iext_add(ifp, 0, nex); | |
578 | ||
1da177e4 | 579 | ifp->if_bytes = size; |
1da177e4 LT |
580 | if (size) { |
581 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
a6f64d4a | 582 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
4eea22f0 | 583 | for (i = 0; i < nex; i++, dp++) { |
a6f64d4a | 584 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
597bca63 HH |
585 | ep->l0 = get_unaligned_be64(&dp->l0); |
586 | ep->l1 = get_unaligned_be64(&dp->l1); | |
1da177e4 | 587 | } |
3a59c94c | 588 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
1da177e4 LT |
589 | if (whichfork != XFS_DATA_FORK || |
590 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
591 | if (unlikely(xfs_check_nostate_extents( | |
4eea22f0 | 592 | ifp, 0, nex))) { |
1da177e4 LT |
593 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
594 | XFS_ERRLEVEL_LOW, | |
595 | ip->i_mount); | |
596 | return XFS_ERROR(EFSCORRUPTED); | |
597 | } | |
598 | } | |
599 | ifp->if_flags |= XFS_IFEXTENTS; | |
600 | return 0; | |
601 | } | |
602 | ||
603 | /* | |
604 | * The file has too many extents to fit into | |
605 | * the inode, so they are in B-tree format. | |
606 | * Allocate a buffer for the root of the B-tree | |
607 | * and copy the root into it. The i_extents | |
608 | * field will remain NULL until all of the | |
609 | * extents are read in (when they are needed). | |
610 | */ | |
611 | STATIC int | |
612 | xfs_iformat_btree( | |
613 | xfs_inode_t *ip, | |
614 | xfs_dinode_t *dip, | |
615 | int whichfork) | |
616 | { | |
617 | xfs_bmdr_block_t *dfp; | |
618 | xfs_ifork_t *ifp; | |
619 | /* REFERENCED */ | |
620 | int nrecs; | |
621 | int size; | |
622 | ||
623 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
624 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
625 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
60197e8d | 626 | nrecs = be16_to_cpu(dfp->bb_numrecs); |
1da177e4 LT |
627 | |
628 | /* | |
629 | * blow out if -- fork has less extents than can fit in | |
630 | * fork (fork shouldn't be a btree format), root btree | |
631 | * block has more records than can fit into the fork, | |
632 | * or the number of extents is greater than the number of | |
633 | * blocks. | |
634 | */ | |
635 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max | |
636 | || XFS_BMDR_SPACE_CALC(nrecs) > | |
637 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) | |
638 | || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
3762ec6b NS |
639 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
640 | "corrupt inode %Lu (btree).", | |
1da177e4 LT |
641 | (unsigned long long) ip->i_ino); |
642 | XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, | |
643 | ip->i_mount); | |
644 | return XFS_ERROR(EFSCORRUPTED); | |
645 | } | |
646 | ||
647 | ifp->if_broot_bytes = size; | |
648 | ifp->if_broot = kmem_alloc(size, KM_SLEEP); | |
649 | ASSERT(ifp->if_broot != NULL); | |
650 | /* | |
651 | * Copy and convert from the on-disk structure | |
652 | * to the in-memory structure. | |
653 | */ | |
60197e8d CH |
654 | xfs_bmdr_to_bmbt(ip->i_mount, dfp, |
655 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
656 | ifp->if_broot, size); | |
1da177e4 LT |
657 | ifp->if_flags &= ~XFS_IFEXTENTS; |
658 | ifp->if_flags |= XFS_IFBROOT; | |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
1da177e4 | 663 | void |
347d1c01 CH |
664 | xfs_dinode_from_disk( |
665 | xfs_icdinode_t *to, | |
666 | xfs_dinode_core_t *from) | |
1da177e4 | 667 | { |
347d1c01 CH |
668 | to->di_magic = be16_to_cpu(from->di_magic); |
669 | to->di_mode = be16_to_cpu(from->di_mode); | |
670 | to->di_version = from ->di_version; | |
671 | to->di_format = from->di_format; | |
672 | to->di_onlink = be16_to_cpu(from->di_onlink); | |
673 | to->di_uid = be32_to_cpu(from->di_uid); | |
674 | to->di_gid = be32_to_cpu(from->di_gid); | |
675 | to->di_nlink = be32_to_cpu(from->di_nlink); | |
676 | to->di_projid = be16_to_cpu(from->di_projid); | |
677 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
678 | to->di_flushiter = be16_to_cpu(from->di_flushiter); | |
679 | to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); | |
680 | to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); | |
681 | to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); | |
682 | to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); | |
683 | to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); | |
684 | to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); | |
685 | to->di_size = be64_to_cpu(from->di_size); | |
686 | to->di_nblocks = be64_to_cpu(from->di_nblocks); | |
687 | to->di_extsize = be32_to_cpu(from->di_extsize); | |
688 | to->di_nextents = be32_to_cpu(from->di_nextents); | |
689 | to->di_anextents = be16_to_cpu(from->di_anextents); | |
690 | to->di_forkoff = from->di_forkoff; | |
691 | to->di_aformat = from->di_aformat; | |
692 | to->di_dmevmask = be32_to_cpu(from->di_dmevmask); | |
693 | to->di_dmstate = be16_to_cpu(from->di_dmstate); | |
694 | to->di_flags = be16_to_cpu(from->di_flags); | |
695 | to->di_gen = be32_to_cpu(from->di_gen); | |
696 | } | |
697 | ||
698 | void | |
699 | xfs_dinode_to_disk( | |
700 | xfs_dinode_core_t *to, | |
701 | xfs_icdinode_t *from) | |
702 | { | |
703 | to->di_magic = cpu_to_be16(from->di_magic); | |
704 | to->di_mode = cpu_to_be16(from->di_mode); | |
705 | to->di_version = from ->di_version; | |
706 | to->di_format = from->di_format; | |
707 | to->di_onlink = cpu_to_be16(from->di_onlink); | |
708 | to->di_uid = cpu_to_be32(from->di_uid); | |
709 | to->di_gid = cpu_to_be32(from->di_gid); | |
710 | to->di_nlink = cpu_to_be32(from->di_nlink); | |
711 | to->di_projid = cpu_to_be16(from->di_projid); | |
712 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
713 | to->di_flushiter = cpu_to_be16(from->di_flushiter); | |
714 | to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); | |
715 | to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); | |
716 | to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); | |
717 | to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); | |
718 | to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); | |
719 | to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); | |
720 | to->di_size = cpu_to_be64(from->di_size); | |
721 | to->di_nblocks = cpu_to_be64(from->di_nblocks); | |
722 | to->di_extsize = cpu_to_be32(from->di_extsize); | |
723 | to->di_nextents = cpu_to_be32(from->di_nextents); | |
724 | to->di_anextents = cpu_to_be16(from->di_anextents); | |
725 | to->di_forkoff = from->di_forkoff; | |
726 | to->di_aformat = from->di_aformat; | |
727 | to->di_dmevmask = cpu_to_be32(from->di_dmevmask); | |
728 | to->di_dmstate = cpu_to_be16(from->di_dmstate); | |
729 | to->di_flags = cpu_to_be16(from->di_flags); | |
730 | to->di_gen = cpu_to_be32(from->di_gen); | |
1da177e4 LT |
731 | } |
732 | ||
733 | STATIC uint | |
734 | _xfs_dic2xflags( | |
1da177e4 LT |
735 | __uint16_t di_flags) |
736 | { | |
737 | uint flags = 0; | |
738 | ||
739 | if (di_flags & XFS_DIFLAG_ANY) { | |
740 | if (di_flags & XFS_DIFLAG_REALTIME) | |
741 | flags |= XFS_XFLAG_REALTIME; | |
742 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
743 | flags |= XFS_XFLAG_PREALLOC; | |
744 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
745 | flags |= XFS_XFLAG_IMMUTABLE; | |
746 | if (di_flags & XFS_DIFLAG_APPEND) | |
747 | flags |= XFS_XFLAG_APPEND; | |
748 | if (di_flags & XFS_DIFLAG_SYNC) | |
749 | flags |= XFS_XFLAG_SYNC; | |
750 | if (di_flags & XFS_DIFLAG_NOATIME) | |
751 | flags |= XFS_XFLAG_NOATIME; | |
752 | if (di_flags & XFS_DIFLAG_NODUMP) | |
753 | flags |= XFS_XFLAG_NODUMP; | |
754 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
755 | flags |= XFS_XFLAG_RTINHERIT; | |
756 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
757 | flags |= XFS_XFLAG_PROJINHERIT; | |
758 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
759 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
760 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
761 | flags |= XFS_XFLAG_EXTSIZE; | |
762 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
763 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
764 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
765 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
766 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
767 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
768 | } |
769 | ||
770 | return flags; | |
771 | } | |
772 | ||
773 | uint | |
774 | xfs_ip2xflags( | |
775 | xfs_inode_t *ip) | |
776 | { | |
347d1c01 | 777 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 778 | |
a916e2bd | 779 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 780 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
781 | } |
782 | ||
783 | uint | |
784 | xfs_dic2xflags( | |
45ba598e | 785 | xfs_dinode_t *dip) |
1da177e4 | 786 | { |
45ba598e CH |
787 | xfs_dinode_core_t *dic = &dip->di_core; |
788 | ||
347d1c01 | 789 | return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) | |
45ba598e | 790 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
791 | } |
792 | ||
07c8f675 DC |
793 | /* |
794 | * Allocate and initialise an xfs_inode. | |
795 | */ | |
c679eef0 | 796 | STATIC struct xfs_inode * |
07c8f675 DC |
797 | xfs_inode_alloc( |
798 | struct xfs_mount *mp, | |
799 | xfs_ino_t ino) | |
800 | { | |
801 | struct xfs_inode *ip; | |
802 | ||
803 | /* | |
804 | * if this didn't occur in transactions, we could use | |
805 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
806 | * code up to do this anyway. | |
807 | */ | |
808 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
809 | if (!ip) | |
810 | return NULL; | |
811 | ||
812 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
813 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
814 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
11654513 | 815 | ASSERT(completion_done(&ip->i_flush)); |
07c8f675 | 816 | |
bf904248 DC |
817 | /* |
818 | * initialise the VFS inode here to get failures | |
819 | * out of the way early. | |
820 | */ | |
821 | if (!inode_init_always(mp->m_super, VFS_I(ip))) { | |
822 | kmem_zone_free(xfs_inode_zone, ip); | |
823 | return NULL; | |
824 | } | |
825 | ||
826 | /* initialise the xfs inode */ | |
07c8f675 DC |
827 | ip->i_ino = ino; |
828 | ip->i_mount = mp; | |
829 | ip->i_blkno = 0; | |
830 | ip->i_len = 0; | |
831 | ip->i_boffset =0; | |
832 | ip->i_afp = NULL; | |
833 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
834 | ip->i_flags = 0; | |
835 | ip->i_update_core = 0; | |
836 | ip->i_update_size = 0; | |
837 | ip->i_delayed_blks = 0; | |
838 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
839 | ip->i_size = 0; | |
840 | ip->i_new_size = 0; | |
841 | ||
842 | /* | |
843 | * Initialize inode's trace buffers. | |
844 | */ | |
845 | #ifdef XFS_INODE_TRACE | |
846 | ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS); | |
847 | #endif | |
848 | #ifdef XFS_BMAP_TRACE | |
849 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS); | |
850 | #endif | |
8c4ed633 | 851 | #ifdef XFS_BTREE_TRACE |
07c8f675 DC |
852 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS); |
853 | #endif | |
854 | #ifdef XFS_RW_TRACE | |
855 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS); | |
856 | #endif | |
857 | #ifdef XFS_ILOCK_TRACE | |
858 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS); | |
859 | #endif | |
860 | #ifdef XFS_DIR2_TRACE | |
861 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS); | |
862 | #endif | |
863 | ||
864 | return ip; | |
865 | } | |
866 | ||
1da177e4 LT |
867 | /* |
868 | * Given a mount structure and an inode number, return a pointer | |
c41564b5 | 869 | * to a newly allocated in-core inode corresponding to the given |
1da177e4 LT |
870 | * inode number. |
871 | * | |
872 | * Initialize the inode's attributes and extent pointers if it | |
873 | * already has them (it will not if the inode has no links). | |
874 | */ | |
875 | int | |
876 | xfs_iread( | |
877 | xfs_mount_t *mp, | |
878 | xfs_trans_t *tp, | |
879 | xfs_ino_t ino, | |
880 | xfs_inode_t **ipp, | |
745b1f47 NS |
881 | xfs_daddr_t bno, |
882 | uint imap_flags) | |
1da177e4 LT |
883 | { |
884 | xfs_buf_t *bp; | |
885 | xfs_dinode_t *dip; | |
886 | xfs_inode_t *ip; | |
887 | int error; | |
888 | ||
07c8f675 DC |
889 | ip = xfs_inode_alloc(mp, ino); |
890 | if (!ip) | |
891 | return ENOMEM; | |
1da177e4 LT |
892 | |
893 | /* | |
894 | * Get pointer's to the on-disk inode and the buffer containing it. | |
895 | * If the inode number refers to a block outside the file system | |
896 | * then xfs_itobp() will return NULL. In this case we should | |
897 | * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will | |
898 | * know that this is a new incore inode. | |
899 | */ | |
a3f74ffb | 900 | error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK); |
9ed0451e CH |
901 | if (error) |
902 | goto out_destroy_inode; | |
1da177e4 | 903 | |
1da177e4 LT |
904 | /* |
905 | * If we got something that isn't an inode it means someone | |
906 | * (nfs or dmi) has a stale handle. | |
907 | */ | |
347d1c01 | 908 | if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) { |
1da177e4 LT |
909 | #ifdef DEBUG |
910 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
911 | "dip->di_core.di_magic (0x%x) != " | |
912 | "XFS_DINODE_MAGIC (0x%x)", | |
347d1c01 | 913 | be16_to_cpu(dip->di_core.di_magic), |
1da177e4 LT |
914 | XFS_DINODE_MAGIC); |
915 | #endif /* DEBUG */ | |
9ed0451e CH |
916 | error = XFS_ERROR(EINVAL); |
917 | goto out_brelse; | |
1da177e4 LT |
918 | } |
919 | ||
920 | /* | |
921 | * If the on-disk inode is already linked to a directory | |
922 | * entry, copy all of the inode into the in-core inode. | |
923 | * xfs_iformat() handles copying in the inode format | |
924 | * specific information. | |
925 | * Otherwise, just get the truly permanent information. | |
926 | */ | |
927 | if (dip->di_core.di_mode) { | |
347d1c01 | 928 | xfs_dinode_from_disk(&ip->i_d, &dip->di_core); |
1da177e4 LT |
929 | error = xfs_iformat(ip, dip); |
930 | if (error) { | |
1da177e4 LT |
931 | #ifdef DEBUG |
932 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
933 | "xfs_iformat() returned error %d", | |
934 | error); | |
935 | #endif /* DEBUG */ | |
9ed0451e | 936 | goto out_brelse; |
1da177e4 LT |
937 | } |
938 | } else { | |
347d1c01 CH |
939 | ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic); |
940 | ip->i_d.di_version = dip->di_core.di_version; | |
941 | ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen); | |
942 | ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter); | |
1da177e4 LT |
943 | /* |
944 | * Make sure to pull in the mode here as well in | |
945 | * case the inode is released without being used. | |
946 | * This ensures that xfs_inactive() will see that | |
947 | * the inode is already free and not try to mess | |
948 | * with the uninitialized part of it. | |
949 | */ | |
950 | ip->i_d.di_mode = 0; | |
951 | /* | |
952 | * Initialize the per-fork minima and maxima for a new | |
953 | * inode here. xfs_iformat will do it for old inodes. | |
954 | */ | |
955 | ip->i_df.if_ext_max = | |
956 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
957 | } | |
958 | ||
1da177e4 LT |
959 | /* |
960 | * The inode format changed when we moved the link count and | |
961 | * made it 32 bits long. If this is an old format inode, | |
962 | * convert it in memory to look like a new one. If it gets | |
963 | * flushed to disk we will convert back before flushing or | |
964 | * logging it. We zero out the new projid field and the old link | |
965 | * count field. We'll handle clearing the pad field (the remains | |
966 | * of the old uuid field) when we actually convert the inode to | |
967 | * the new format. We don't change the version number so that we | |
968 | * can distinguish this from a real new format inode. | |
969 | */ | |
970 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
971 | ip->i_d.di_nlink = ip->i_d.di_onlink; | |
972 | ip->i_d.di_onlink = 0; | |
973 | ip->i_d.di_projid = 0; | |
974 | } | |
975 | ||
976 | ip->i_delayed_blks = 0; | |
ba87ea69 | 977 | ip->i_size = ip->i_d.di_size; |
1da177e4 LT |
978 | |
979 | /* | |
980 | * Mark the buffer containing the inode as something to keep | |
981 | * around for a while. This helps to keep recently accessed | |
982 | * meta-data in-core longer. | |
983 | */ | |
984 | XFS_BUF_SET_REF(bp, XFS_INO_REF); | |
985 | ||
986 | /* | |
987 | * Use xfs_trans_brelse() to release the buffer containing the | |
988 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
989 | * in xfs_itobp() above. If tp is NULL, this is just a normal | |
990 | * brelse(). If we're within a transaction, then xfs_trans_brelse() | |
991 | * will only release the buffer if it is not dirty within the | |
992 | * transaction. It will be OK to release the buffer in this case, | |
993 | * because inodes on disk are never destroyed and we will be | |
994 | * locking the new in-core inode before putting it in the hash | |
995 | * table where other processes can find it. Thus we don't have | |
996 | * to worry about the inode being changed just because we released | |
997 | * the buffer. | |
998 | */ | |
999 | xfs_trans_brelse(tp, bp); | |
1000 | *ipp = ip; | |
1001 | return 0; | |
9ed0451e CH |
1002 | |
1003 | out_brelse: | |
1004 | xfs_trans_brelse(tp, bp); | |
1005 | out_destroy_inode: | |
1006 | xfs_destroy_inode(ip); | |
1007 | return error; | |
1da177e4 LT |
1008 | } |
1009 | ||
1010 | /* | |
1011 | * Read in extents from a btree-format inode. | |
1012 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1013 | */ | |
1014 | int | |
1015 | xfs_iread_extents( | |
1016 | xfs_trans_t *tp, | |
1017 | xfs_inode_t *ip, | |
1018 | int whichfork) | |
1019 | { | |
1020 | int error; | |
1021 | xfs_ifork_t *ifp; | |
4eea22f0 | 1022 | xfs_extnum_t nextents; |
1da177e4 LT |
1023 | size_t size; |
1024 | ||
1025 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1026 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1027 | ip->i_mount); | |
1028 | return XFS_ERROR(EFSCORRUPTED); | |
1029 | } | |
4eea22f0 MK |
1030 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
1031 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
1da177e4 | 1032 | ifp = XFS_IFORK_PTR(ip, whichfork); |
4eea22f0 | 1033 | |
1da177e4 LT |
1034 | /* |
1035 | * We know that the size is valid (it's checked in iformat_btree) | |
1036 | */ | |
1da177e4 | 1037 | ifp->if_lastex = NULLEXTNUM; |
4eea22f0 | 1038 | ifp->if_bytes = ifp->if_real_bytes = 0; |
1da177e4 | 1039 | ifp->if_flags |= XFS_IFEXTENTS; |
4eea22f0 | 1040 | xfs_iext_add(ifp, 0, nextents); |
1da177e4 LT |
1041 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
1042 | if (error) { | |
4eea22f0 | 1043 | xfs_iext_destroy(ifp); |
1da177e4 LT |
1044 | ifp->if_flags &= ~XFS_IFEXTENTS; |
1045 | return error; | |
1046 | } | |
a6f64d4a | 1047 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
1048 | return 0; |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * Allocate an inode on disk and return a copy of its in-core version. | |
1053 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1054 | * appropriately within the inode. The uid and gid for the inode are | |
1055 | * set according to the contents of the given cred structure. | |
1056 | * | |
1057 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
1058 | * has a free inode available, call xfs_iget() | |
1059 | * to obtain the in-core version of the allocated inode. Finally, | |
1060 | * fill in the inode and log its initial contents. In this case, | |
1061 | * ialloc_context would be set to NULL and call_again set to false. | |
1062 | * | |
1063 | * If xfs_dialloc() does not have an available inode, | |
1064 | * it will replenish its supply by doing an allocation. Since we can | |
1065 | * only do one allocation within a transaction without deadlocks, we | |
1066 | * must commit the current transaction before returning the inode itself. | |
1067 | * In this case, therefore, we will set call_again to true and return. | |
1068 | * The caller should then commit the current transaction, start a new | |
1069 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1070 | * | |
1071 | * To ensure that some other process does not grab the inode that | |
1072 | * was allocated during the first call to xfs_ialloc(), this routine | |
1073 | * also returns the [locked] bp pointing to the head of the freelist | |
1074 | * as ialloc_context. The caller should hold this buffer across | |
1075 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
1076 | * |
1077 | * If we are allocating quota inodes, we do not have a parent inode | |
1078 | * to attach to or associate with (i.e. pip == NULL) because they | |
1079 | * are not linked into the directory structure - they are attached | |
1080 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
1081 | */ |
1082 | int | |
1083 | xfs_ialloc( | |
1084 | xfs_trans_t *tp, | |
1085 | xfs_inode_t *pip, | |
1086 | mode_t mode, | |
31b084ae | 1087 | xfs_nlink_t nlink, |
1da177e4 LT |
1088 | xfs_dev_t rdev, |
1089 | cred_t *cr, | |
1090 | xfs_prid_t prid, | |
1091 | int okalloc, | |
1092 | xfs_buf_t **ialloc_context, | |
1093 | boolean_t *call_again, | |
1094 | xfs_inode_t **ipp) | |
1095 | { | |
1096 | xfs_ino_t ino; | |
1097 | xfs_inode_t *ip; | |
1da177e4 LT |
1098 | uint flags; |
1099 | int error; | |
dff35fd4 | 1100 | timespec_t tv; |
bf904248 | 1101 | int filestreams = 0; |
1da177e4 LT |
1102 | |
1103 | /* | |
1104 | * Call the space management code to pick | |
1105 | * the on-disk inode to be allocated. | |
1106 | */ | |
b11f94d5 | 1107 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
1da177e4 | 1108 | ialloc_context, call_again, &ino); |
bf904248 | 1109 | if (error) |
1da177e4 | 1110 | return error; |
1da177e4 LT |
1111 | if (*call_again || ino == NULLFSINO) { |
1112 | *ipp = NULL; | |
1113 | return 0; | |
1114 | } | |
1115 | ASSERT(*ialloc_context == NULL); | |
1116 | ||
1117 | /* | |
1118 | * Get the in-core inode with the lock held exclusively. | |
1119 | * This is because we're setting fields here we need | |
1120 | * to prevent others from looking at until we're done. | |
1121 | */ | |
1122 | error = xfs_trans_iget(tp->t_mountp, tp, ino, | |
745b1f47 | 1123 | XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip); |
bf904248 | 1124 | if (error) |
1da177e4 | 1125 | return error; |
1da177e4 LT |
1126 | ASSERT(ip != NULL); |
1127 | ||
1da177e4 LT |
1128 | ip->i_d.di_mode = (__uint16_t)mode; |
1129 | ip->i_d.di_onlink = 0; | |
1130 | ip->i_d.di_nlink = nlink; | |
1131 | ASSERT(ip->i_d.di_nlink == nlink); | |
9e2b2dc4 DH |
1132 | ip->i_d.di_uid = current_fsuid(); |
1133 | ip->i_d.di_gid = current_fsgid(); | |
1da177e4 LT |
1134 | ip->i_d.di_projid = prid; |
1135 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
1136 | ||
1137 | /* | |
1138 | * If the superblock version is up to where we support new format | |
1139 | * inodes and this is currently an old format inode, then change | |
1140 | * the inode version number now. This way we only do the conversion | |
1141 | * here rather than here and in the flush/logging code. | |
1142 | */ | |
62118709 | 1143 | if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) && |
1da177e4 LT |
1144 | ip->i_d.di_version == XFS_DINODE_VERSION_1) { |
1145 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
1146 | /* | |
1147 | * We've already zeroed the old link count, the projid field, | |
1148 | * and the pad field. | |
1149 | */ | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1154 | */ | |
2a82b8be | 1155 | if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) |
1da177e4 LT |
1156 | xfs_bump_ino_vers2(tp, ip); |
1157 | ||
bd186aa9 | 1158 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 LT |
1159 | ip->i_d.di_gid = pip->i_d.di_gid; |
1160 | if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) { | |
1161 | ip->i_d.di_mode |= S_ISGID; | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * If the group ID of the new file does not match the effective group | |
1167 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1168 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1169 | */ | |
1170 | if ((irix_sgid_inherit) && | |
1171 | (ip->i_d.di_mode & S_ISGID) && | |
1172 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1173 | ip->i_d.di_mode &= ~S_ISGID; | |
1174 | } | |
1175 | ||
1176 | ip->i_d.di_size = 0; | |
ba87ea69 | 1177 | ip->i_size = 0; |
1da177e4 LT |
1178 | ip->i_d.di_nextents = 0; |
1179 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
1180 | |
1181 | nanotime(&tv); | |
1182 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
1183 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
1184 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
1185 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
1186 | ||
1da177e4 LT |
1187 | /* |
1188 | * di_gen will have been taken care of in xfs_iread. | |
1189 | */ | |
1190 | ip->i_d.di_extsize = 0; | |
1191 | ip->i_d.di_dmevmask = 0; | |
1192 | ip->i_d.di_dmstate = 0; | |
1193 | ip->i_d.di_flags = 0; | |
1194 | flags = XFS_ILOG_CORE; | |
1195 | switch (mode & S_IFMT) { | |
1196 | case S_IFIFO: | |
1197 | case S_IFCHR: | |
1198 | case S_IFBLK: | |
1199 | case S_IFSOCK: | |
1200 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1201 | ip->i_df.if_u2.if_rdev = rdev; | |
1202 | ip->i_df.if_flags = 0; | |
1203 | flags |= XFS_ILOG_DEV; | |
1204 | break; | |
1205 | case S_IFREG: | |
bf904248 DC |
1206 | /* |
1207 | * we can't set up filestreams until after the VFS inode | |
1208 | * is set up properly. | |
1209 | */ | |
1210 | if (pip && xfs_inode_is_filestream(pip)) | |
1211 | filestreams = 1; | |
2a82b8be | 1212 | /* fall through */ |
1da177e4 | 1213 | case S_IFDIR: |
b11f94d5 | 1214 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
1215 | uint di_flags = 0; |
1216 | ||
1217 | if ((mode & S_IFMT) == S_IFDIR) { | |
1218 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) | |
1219 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
1220 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1221 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
1222 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1223 | } | |
1224 | } else if ((mode & S_IFMT) == S_IFREG) { | |
613d7043 | 1225 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 1226 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
1227 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1228 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1229 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1230 | } | |
1da177e4 LT |
1231 | } |
1232 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1233 | xfs_inherit_noatime) | |
365ca83d | 1234 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1235 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1236 | xfs_inherit_nodump) | |
365ca83d | 1237 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1238 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1239 | xfs_inherit_sync) | |
365ca83d | 1240 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1241 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1242 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1243 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1244 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1245 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
1246 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
1247 | xfs_inherit_nodefrag) | |
1248 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
1249 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
1250 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 1251 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
1252 | } |
1253 | /* FALLTHROUGH */ | |
1254 | case S_IFLNK: | |
1255 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1256 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1257 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1258 | ip->i_df.if_u1.if_extents = NULL; | |
1259 | break; | |
1260 | default: | |
1261 | ASSERT(0); | |
1262 | } | |
1263 | /* | |
1264 | * Attribute fork settings for new inode. | |
1265 | */ | |
1266 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1267 | ip->i_d.di_anextents = 0; | |
1268 | ||
1269 | /* | |
1270 | * Log the new values stuffed into the inode. | |
1271 | */ | |
1272 | xfs_trans_log_inode(tp, ip, flags); | |
1273 | ||
b83bd138 | 1274 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 1275 | xfs_setup_inode(ip); |
1da177e4 | 1276 | |
bf904248 DC |
1277 | /* now we have set up the vfs inode we can associate the filestream */ |
1278 | if (filestreams) { | |
1279 | error = xfs_filestream_associate(pip, ip); | |
1280 | if (error < 0) | |
1281 | return -error; | |
1282 | if (!error) | |
1283 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
1284 | } | |
1285 | ||
1da177e4 LT |
1286 | *ipp = ip; |
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Check to make sure that there are no blocks allocated to the | |
1292 | * file beyond the size of the file. We don't check this for | |
1293 | * files with fixed size extents or real time extents, but we | |
1294 | * at least do it for regular files. | |
1295 | */ | |
1296 | #ifdef DEBUG | |
1297 | void | |
1298 | xfs_isize_check( | |
1299 | xfs_mount_t *mp, | |
1300 | xfs_inode_t *ip, | |
1301 | xfs_fsize_t isize) | |
1302 | { | |
1303 | xfs_fileoff_t map_first; | |
1304 | int nimaps; | |
1305 | xfs_bmbt_irec_t imaps[2]; | |
1306 | ||
1307 | if ((ip->i_d.di_mode & S_IFMT) != S_IFREG) | |
1308 | return; | |
1309 | ||
71ddabb9 ES |
1310 | if (XFS_IS_REALTIME_INODE(ip)) |
1311 | return; | |
1312 | ||
1313 | if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) | |
1da177e4 LT |
1314 | return; |
1315 | ||
1316 | nimaps = 2; | |
1317 | map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
1318 | /* | |
1319 | * The filesystem could be shutting down, so bmapi may return | |
1320 | * an error. | |
1321 | */ | |
1322 | if (xfs_bmapi(NULL, ip, map_first, | |
1323 | (XFS_B_TO_FSB(mp, | |
1324 | (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) - | |
1325 | map_first), | |
1326 | XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps, | |
3e57ecf6 | 1327 | NULL, NULL)) |
1da177e4 LT |
1328 | return; |
1329 | ASSERT(nimaps == 1); | |
1330 | ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK); | |
1331 | } | |
1332 | #endif /* DEBUG */ | |
1333 | ||
1334 | /* | |
1335 | * Calculate the last possible buffered byte in a file. This must | |
1336 | * include data that was buffered beyond the EOF by the write code. | |
1337 | * This also needs to deal with overflowing the xfs_fsize_t type | |
1338 | * which can happen for sizes near the limit. | |
1339 | * | |
1340 | * We also need to take into account any blocks beyond the EOF. It | |
1341 | * may be the case that they were buffered by a write which failed. | |
1342 | * In that case the pages will still be in memory, but the inode size | |
1343 | * will never have been updated. | |
1344 | */ | |
1345 | xfs_fsize_t | |
1346 | xfs_file_last_byte( | |
1347 | xfs_inode_t *ip) | |
1348 | { | |
1349 | xfs_mount_t *mp; | |
1350 | xfs_fsize_t last_byte; | |
1351 | xfs_fileoff_t last_block; | |
1352 | xfs_fileoff_t size_last_block; | |
1353 | int error; | |
1354 | ||
579aa9ca | 1355 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)); |
1da177e4 LT |
1356 | |
1357 | mp = ip->i_mount; | |
1358 | /* | |
1359 | * Only check for blocks beyond the EOF if the extents have | |
1360 | * been read in. This eliminates the need for the inode lock, | |
1361 | * and it also saves us from looking when it really isn't | |
1362 | * necessary. | |
1363 | */ | |
1364 | if (ip->i_df.if_flags & XFS_IFEXTENTS) { | |
1365 | error = xfs_bmap_last_offset(NULL, ip, &last_block, | |
1366 | XFS_DATA_FORK); | |
1367 | if (error) { | |
1368 | last_block = 0; | |
1369 | } | |
1370 | } else { | |
1371 | last_block = 0; | |
1372 | } | |
ba87ea69 | 1373 | size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size); |
1da177e4 LT |
1374 | last_block = XFS_FILEOFF_MAX(last_block, size_last_block); |
1375 | ||
1376 | last_byte = XFS_FSB_TO_B(mp, last_block); | |
1377 | if (last_byte < 0) { | |
1378 | return XFS_MAXIOFFSET(mp); | |
1379 | } | |
1380 | last_byte += (1 << mp->m_writeio_log); | |
1381 | if (last_byte < 0) { | |
1382 | return XFS_MAXIOFFSET(mp); | |
1383 | } | |
1384 | return last_byte; | |
1385 | } | |
1386 | ||
1387 | #if defined(XFS_RW_TRACE) | |
1388 | STATIC void | |
1389 | xfs_itrunc_trace( | |
1390 | int tag, | |
1391 | xfs_inode_t *ip, | |
1392 | int flag, | |
1393 | xfs_fsize_t new_size, | |
1394 | xfs_off_t toss_start, | |
1395 | xfs_off_t toss_finish) | |
1396 | { | |
1397 | if (ip->i_rwtrace == NULL) { | |
1398 | return; | |
1399 | } | |
1400 | ||
1401 | ktrace_enter(ip->i_rwtrace, | |
1402 | (void*)((long)tag), | |
1403 | (void*)ip, | |
1404 | (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff), | |
1405 | (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff), | |
1406 | (void*)((long)flag), | |
1407 | (void*)(unsigned long)((new_size >> 32) & 0xffffffff), | |
1408 | (void*)(unsigned long)(new_size & 0xffffffff), | |
1409 | (void*)(unsigned long)((toss_start >> 32) & 0xffffffff), | |
1410 | (void*)(unsigned long)(toss_start & 0xffffffff), | |
1411 | (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff), | |
1412 | (void*)(unsigned long)(toss_finish & 0xffffffff), | |
1413 | (void*)(unsigned long)current_cpu(), | |
f1fdc848 YL |
1414 | (void*)(unsigned long)current_pid(), |
1415 | (void*)NULL, | |
1416 | (void*)NULL, | |
1417 | (void*)NULL); | |
1da177e4 LT |
1418 | } |
1419 | #else | |
1420 | #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish) | |
1421 | #endif | |
1422 | ||
1423 | /* | |
1424 | * Start the truncation of the file to new_size. The new size | |
1425 | * must be smaller than the current size. This routine will | |
1426 | * clear the buffer and page caches of file data in the removed | |
1427 | * range, and xfs_itruncate_finish() will remove the underlying | |
1428 | * disk blocks. | |
1429 | * | |
1430 | * The inode must have its I/O lock locked EXCLUSIVELY, and it | |
1431 | * must NOT have the inode lock held at all. This is because we're | |
1432 | * calling into the buffer/page cache code and we can't hold the | |
1433 | * inode lock when we do so. | |
1434 | * | |
38e2299a DC |
1435 | * We need to wait for any direct I/Os in flight to complete before we |
1436 | * proceed with the truncate. This is needed to prevent the extents | |
1437 | * being read or written by the direct I/Os from being removed while the | |
1438 | * I/O is in flight as there is no other method of synchronising | |
1439 | * direct I/O with the truncate operation. Also, because we hold | |
1440 | * the IOLOCK in exclusive mode, we prevent new direct I/Os from being | |
1441 | * started until the truncate completes and drops the lock. Essentially, | |
1442 | * the vn_iowait() call forms an I/O barrier that provides strict ordering | |
1443 | * between direct I/Os and the truncate operation. | |
1444 | * | |
1da177e4 LT |
1445 | * The flags parameter can have either the value XFS_ITRUNC_DEFINITE |
1446 | * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used | |
1447 | * in the case that the caller is locking things out of order and | |
1448 | * may not be able to call xfs_itruncate_finish() with the inode lock | |
1449 | * held without dropping the I/O lock. If the caller must drop the | |
1450 | * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start() | |
1451 | * must be called again with all the same restrictions as the initial | |
1452 | * call. | |
1453 | */ | |
d3cf2094 | 1454 | int |
1da177e4 LT |
1455 | xfs_itruncate_start( |
1456 | xfs_inode_t *ip, | |
1457 | uint flags, | |
1458 | xfs_fsize_t new_size) | |
1459 | { | |
1460 | xfs_fsize_t last_byte; | |
1461 | xfs_off_t toss_start; | |
1462 | xfs_mount_t *mp; | |
d3cf2094 | 1463 | int error = 0; |
1da177e4 | 1464 | |
579aa9ca | 1465 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
ba87ea69 | 1466 | ASSERT((new_size == 0) || (new_size <= ip->i_size)); |
1da177e4 LT |
1467 | ASSERT((flags == XFS_ITRUNC_DEFINITE) || |
1468 | (flags == XFS_ITRUNC_MAYBE)); | |
1469 | ||
1470 | mp = ip->i_mount; | |
9fa8046f | 1471 | |
c734c79b | 1472 | /* wait for the completion of any pending DIOs */ |
d112f298 | 1473 | if (new_size == 0 || new_size < ip->i_size) |
c734c79b LM |
1474 | vn_iowait(ip); |
1475 | ||
1da177e4 | 1476 | /* |
67fcaa73 | 1477 | * Call toss_pages or flushinval_pages to get rid of pages |
1da177e4 | 1478 | * overlapping the region being removed. We have to use |
67fcaa73 | 1479 | * the less efficient flushinval_pages in the case that the |
1da177e4 LT |
1480 | * caller may not be able to finish the truncate without |
1481 | * dropping the inode's I/O lock. Make sure | |
1482 | * to catch any pages brought in by buffers overlapping | |
1483 | * the EOF by searching out beyond the isize by our | |
1484 | * block size. We round new_size up to a block boundary | |
1485 | * so that we don't toss things on the same block as | |
1486 | * new_size but before it. | |
1487 | * | |
67fcaa73 | 1488 | * Before calling toss_page or flushinval_pages, make sure to |
1da177e4 LT |
1489 | * call remapf() over the same region if the file is mapped. |
1490 | * This frees up mapped file references to the pages in the | |
67fcaa73 | 1491 | * given range and for the flushinval_pages case it ensures |
1da177e4 LT |
1492 | * that we get the latest mapped changes flushed out. |
1493 | */ | |
1494 | toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1495 | toss_start = XFS_FSB_TO_B(mp, toss_start); | |
1496 | if (toss_start < 0) { | |
1497 | /* | |
1498 | * The place to start tossing is beyond our maximum | |
1499 | * file size, so there is no way that the data extended | |
1500 | * out there. | |
1501 | */ | |
d3cf2094 | 1502 | return 0; |
1da177e4 LT |
1503 | } |
1504 | last_byte = xfs_file_last_byte(ip); | |
1505 | xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start, | |
1506 | last_byte); | |
1507 | if (last_byte > toss_start) { | |
1508 | if (flags & XFS_ITRUNC_DEFINITE) { | |
739bfb2a CH |
1509 | xfs_tosspages(ip, toss_start, |
1510 | -1, FI_REMAPF_LOCKED); | |
1da177e4 | 1511 | } else { |
739bfb2a CH |
1512 | error = xfs_flushinval_pages(ip, toss_start, |
1513 | -1, FI_REMAPF_LOCKED); | |
1da177e4 LT |
1514 | } |
1515 | } | |
1516 | ||
1517 | #ifdef DEBUG | |
1518 | if (new_size == 0) { | |
df80c933 | 1519 | ASSERT(VN_CACHED(VFS_I(ip)) == 0); |
1da177e4 LT |
1520 | } |
1521 | #endif | |
d3cf2094 | 1522 | return error; |
1da177e4 LT |
1523 | } |
1524 | ||
1525 | /* | |
f6485057 DC |
1526 | * Shrink the file to the given new_size. The new size must be smaller than |
1527 | * the current size. This will free up the underlying blocks in the removed | |
1528 | * range after a call to xfs_itruncate_start() or xfs_atruncate_start(). | |
1da177e4 | 1529 | * |
f6485057 DC |
1530 | * The transaction passed to this routine must have made a permanent log |
1531 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1532 | * given transaction and start new ones, so make sure everything involved in | |
1533 | * the transaction is tidy before calling here. Some transaction will be | |
1534 | * returned to the caller to be committed. The incoming transaction must | |
1535 | * already include the inode, and both inode locks must be held exclusively. | |
1536 | * The inode must also be "held" within the transaction. On return the inode | |
1537 | * will be "held" within the returned transaction. This routine does NOT | |
1538 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1539 | * |
f6485057 DC |
1540 | * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it |
1541 | * indicates the fork which is to be truncated. For the attribute fork we only | |
1542 | * support truncation to size 0. | |
1da177e4 | 1543 | * |
f6485057 DC |
1544 | * We use the sync parameter to indicate whether or not the first transaction |
1545 | * we perform might have to be synchronous. For the attr fork, it needs to be | |
1546 | * so if the unlink of the inode is not yet known to be permanent in the log. | |
1547 | * This keeps us from freeing and reusing the blocks of the attribute fork | |
1548 | * before the unlink of the inode becomes permanent. | |
1da177e4 | 1549 | * |
f6485057 DC |
1550 | * For the data fork, we normally have to run synchronously if we're being |
1551 | * called out of the inactive path or we're being called out of the create path | |
1552 | * where we're truncating an existing file. Either way, the truncate needs to | |
1553 | * be sync so blocks don't reappear in the file with altered data in case of a | |
1554 | * crash. wsync filesystems can run the first case async because anything that | |
1555 | * shrinks the inode has to run sync so by the time we're called here from | |
1556 | * inactive, the inode size is permanently set to 0. | |
1da177e4 | 1557 | * |
f6485057 DC |
1558 | * Calls from the truncate path always need to be sync unless we're in a wsync |
1559 | * filesystem and the file has already been unlinked. | |
1da177e4 | 1560 | * |
f6485057 DC |
1561 | * The caller is responsible for correctly setting the sync parameter. It gets |
1562 | * too hard for us to guess here which path we're being called out of just | |
1563 | * based on inode state. | |
1564 | * | |
1565 | * If we get an error, we must return with the inode locked and linked into the | |
1566 | * current transaction. This keeps things simple for the higher level code, | |
1567 | * because it always knows that the inode is locked and held in the transaction | |
1568 | * that returns to it whether errors occur or not. We don't mark the inode | |
1569 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1570 | */ |
1571 | int | |
1572 | xfs_itruncate_finish( | |
1573 | xfs_trans_t **tp, | |
1574 | xfs_inode_t *ip, | |
1575 | xfs_fsize_t new_size, | |
1576 | int fork, | |
1577 | int sync) | |
1578 | { | |
1579 | xfs_fsblock_t first_block; | |
1580 | xfs_fileoff_t first_unmap_block; | |
1581 | xfs_fileoff_t last_block; | |
1582 | xfs_filblks_t unmap_len=0; | |
1583 | xfs_mount_t *mp; | |
1584 | xfs_trans_t *ntp; | |
1585 | int done; | |
1586 | int committed; | |
1587 | xfs_bmap_free_t free_list; | |
1588 | int error; | |
1589 | ||
579aa9ca | 1590 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); |
ba87ea69 | 1591 | ASSERT((new_size == 0) || (new_size <= ip->i_size)); |
1da177e4 LT |
1592 | ASSERT(*tp != NULL); |
1593 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1594 | ASSERT(ip->i_transp == *tp); | |
1595 | ASSERT(ip->i_itemp != NULL); | |
1596 | ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD); | |
1597 | ||
1598 | ||
1599 | ntp = *tp; | |
1600 | mp = (ntp)->t_mountp; | |
1601 | ASSERT(! XFS_NOT_DQATTACHED(mp, ip)); | |
1602 | ||
1603 | /* | |
1604 | * We only support truncating the entire attribute fork. | |
1605 | */ | |
1606 | if (fork == XFS_ATTR_FORK) { | |
1607 | new_size = 0LL; | |
1608 | } | |
1609 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1610 | xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0); | |
1611 | /* | |
1612 | * The first thing we do is set the size to new_size permanently | |
1613 | * on disk. This way we don't have to worry about anyone ever | |
1614 | * being able to look at the data being freed even in the face | |
1615 | * of a crash. What we're getting around here is the case where | |
1616 | * we free a block, it is allocated to another file, it is written | |
1617 | * to, and then we crash. If the new data gets written to the | |
1618 | * file but the log buffers containing the free and reallocation | |
1619 | * don't, then we'd end up with garbage in the blocks being freed. | |
1620 | * As long as we make the new_size permanent before actually | |
1621 | * freeing any blocks it doesn't matter if they get writtten to. | |
1622 | * | |
1623 | * The callers must signal into us whether or not the size | |
1624 | * setting here must be synchronous. There are a few cases | |
1625 | * where it doesn't have to be synchronous. Those cases | |
1626 | * occur if the file is unlinked and we know the unlink is | |
1627 | * permanent or if the blocks being truncated are guaranteed | |
1628 | * to be beyond the inode eof (regardless of the link count) | |
1629 | * and the eof value is permanent. Both of these cases occur | |
1630 | * only on wsync-mounted filesystems. In those cases, we're | |
1631 | * guaranteed that no user will ever see the data in the blocks | |
1632 | * that are being truncated so the truncate can run async. | |
1633 | * In the free beyond eof case, the file may wind up with | |
1634 | * more blocks allocated to it than it needs if we crash | |
1635 | * and that won't get fixed until the next time the file | |
1636 | * is re-opened and closed but that's ok as that shouldn't | |
1637 | * be too many blocks. | |
1638 | * | |
1639 | * However, we can't just make all wsync xactions run async | |
1640 | * because there's one call out of the create path that needs | |
1641 | * to run sync where it's truncating an existing file to size | |
1642 | * 0 whose size is > 0. | |
1643 | * | |
1644 | * It's probably possible to come up with a test in this | |
1645 | * routine that would correctly distinguish all the above | |
1646 | * cases from the values of the function parameters and the | |
1647 | * inode state but for sanity's sake, I've decided to let the | |
1648 | * layers above just tell us. It's simpler to correctly figure | |
1649 | * out in the layer above exactly under what conditions we | |
1650 | * can run async and I think it's easier for others read and | |
1651 | * follow the logic in case something has to be changed. | |
1652 | * cscope is your friend -- rcc. | |
1653 | * | |
1654 | * The attribute fork is much simpler. | |
1655 | * | |
1656 | * For the attribute fork we allow the caller to tell us whether | |
1657 | * the unlink of the inode that led to this call is yet permanent | |
1658 | * in the on disk log. If it is not and we will be freeing extents | |
1659 | * in this inode then we make the first transaction synchronous | |
1660 | * to make sure that the unlink is permanent by the time we free | |
1661 | * the blocks. | |
1662 | */ | |
1663 | if (fork == XFS_DATA_FORK) { | |
1664 | if (ip->i_d.di_nextents > 0) { | |
ba87ea69 LM |
1665 | /* |
1666 | * If we are not changing the file size then do | |
1667 | * not update the on-disk file size - we may be | |
1668 | * called from xfs_inactive_free_eofblocks(). If we | |
1669 | * update the on-disk file size and then the system | |
1670 | * crashes before the contents of the file are | |
1671 | * flushed to disk then the files may be full of | |
1672 | * holes (ie NULL files bug). | |
1673 | */ | |
1674 | if (ip->i_size != new_size) { | |
1675 | ip->i_d.di_size = new_size; | |
1676 | ip->i_size = new_size; | |
1677 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1678 | } | |
1da177e4 LT |
1679 | } |
1680 | } else if (sync) { | |
1681 | ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC)); | |
1682 | if (ip->i_d.di_anextents > 0) | |
1683 | xfs_trans_set_sync(ntp); | |
1684 | } | |
1685 | ASSERT(fork == XFS_DATA_FORK || | |
1686 | (fork == XFS_ATTR_FORK && | |
1687 | ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) || | |
1688 | (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC))))); | |
1689 | ||
1690 | /* | |
1691 | * Since it is possible for space to become allocated beyond | |
1692 | * the end of the file (in a crash where the space is allocated | |
1693 | * but the inode size is not yet updated), simply remove any | |
1694 | * blocks which show up between the new EOF and the maximum | |
1695 | * possible file size. If the first block to be removed is | |
1696 | * beyond the maximum file size (ie it is the same as last_block), | |
1697 | * then there is nothing to do. | |
1698 | */ | |
1699 | last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1700 | ASSERT(first_unmap_block <= last_block); | |
1701 | done = 0; | |
1702 | if (last_block == first_unmap_block) { | |
1703 | done = 1; | |
1704 | } else { | |
1705 | unmap_len = last_block - first_unmap_block + 1; | |
1706 | } | |
1707 | while (!done) { | |
1708 | /* | |
1709 | * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi() | |
1710 | * will tell us whether it freed the entire range or | |
1711 | * not. If this is a synchronous mount (wsync), | |
1712 | * then we can tell bunmapi to keep all the | |
1713 | * transactions asynchronous since the unlink | |
1714 | * transaction that made this inode inactive has | |
1715 | * already hit the disk. There's no danger of | |
1716 | * the freed blocks being reused, there being a | |
1717 | * crash, and the reused blocks suddenly reappearing | |
1718 | * in this file with garbage in them once recovery | |
1719 | * runs. | |
1720 | */ | |
1721 | XFS_BMAP_INIT(&free_list, &first_block); | |
541d7d3c | 1722 | error = xfs_bunmapi(ntp, ip, |
3e57ecf6 | 1723 | first_unmap_block, unmap_len, |
1da177e4 LT |
1724 | XFS_BMAPI_AFLAG(fork) | |
1725 | (sync ? 0 : XFS_BMAPI_ASYNC), | |
1726 | XFS_ITRUNC_MAX_EXTENTS, | |
3e57ecf6 OW |
1727 | &first_block, &free_list, |
1728 | NULL, &done); | |
1da177e4 LT |
1729 | if (error) { |
1730 | /* | |
1731 | * If the bunmapi call encounters an error, | |
1732 | * return to the caller where the transaction | |
1733 | * can be properly aborted. We just need to | |
1734 | * make sure we're not holding any resources | |
1735 | * that we were not when we came in. | |
1736 | */ | |
1737 | xfs_bmap_cancel(&free_list); | |
1738 | return error; | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * Duplicate the transaction that has the permanent | |
1743 | * reservation and commit the old transaction. | |
1744 | */ | |
f7c99b6f | 1745 | error = xfs_bmap_finish(tp, &free_list, &committed); |
1da177e4 | 1746 | ntp = *tp; |
f6485057 DC |
1747 | if (committed) { |
1748 | /* link the inode into the next xact in the chain */ | |
1749 | xfs_trans_ijoin(ntp, ip, | |
1750 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1751 | xfs_trans_ihold(ntp, ip); | |
1752 | } | |
1753 | ||
1da177e4 LT |
1754 | if (error) { |
1755 | /* | |
f6485057 DC |
1756 | * If the bmap finish call encounters an error, return |
1757 | * to the caller where the transaction can be properly | |
1758 | * aborted. We just need to make sure we're not | |
1759 | * holding any resources that we were not when we came | |
1760 | * in. | |
1da177e4 | 1761 | * |
f6485057 DC |
1762 | * Aborting from this point might lose some blocks in |
1763 | * the file system, but oh well. | |
1da177e4 LT |
1764 | */ |
1765 | xfs_bmap_cancel(&free_list); | |
1da177e4 LT |
1766 | return error; |
1767 | } | |
1768 | ||
1769 | if (committed) { | |
1770 | /* | |
f6485057 | 1771 | * Mark the inode dirty so it will be logged and |
e5720eec | 1772 | * moved forward in the log as part of every commit. |
1da177e4 | 1773 | */ |
1da177e4 LT |
1774 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); |
1775 | } | |
f6485057 | 1776 | |
1da177e4 | 1777 | ntp = xfs_trans_dup(ntp); |
e5720eec | 1778 | error = xfs_trans_commit(*tp, 0); |
1da177e4 | 1779 | *tp = ntp; |
e5720eec | 1780 | |
f6485057 DC |
1781 | /* link the inode into the next transaction in the chain */ |
1782 | xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1783 | xfs_trans_ihold(ntp, ip); | |
1784 | ||
1785 | if (!error) | |
1786 | error = xfs_trans_reserve(ntp, 0, | |
1787 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1788 | XFS_TRANS_PERM_LOG_RES, | |
1789 | XFS_ITRUNCATE_LOG_COUNT); | |
1790 | if (error) | |
1791 | return error; | |
1da177e4 LT |
1792 | } |
1793 | /* | |
1794 | * Only update the size in the case of the data fork, but | |
1795 | * always re-log the inode so that our permanent transaction | |
1796 | * can keep on rolling it forward in the log. | |
1797 | */ | |
1798 | if (fork == XFS_DATA_FORK) { | |
1799 | xfs_isize_check(mp, ip, new_size); | |
ba87ea69 LM |
1800 | /* |
1801 | * If we are not changing the file size then do | |
1802 | * not update the on-disk file size - we may be | |
1803 | * called from xfs_inactive_free_eofblocks(). If we | |
1804 | * update the on-disk file size and then the system | |
1805 | * crashes before the contents of the file are | |
1806 | * flushed to disk then the files may be full of | |
1807 | * holes (ie NULL files bug). | |
1808 | */ | |
1809 | if (ip->i_size != new_size) { | |
1810 | ip->i_d.di_size = new_size; | |
1811 | ip->i_size = new_size; | |
1812 | } | |
1da177e4 LT |
1813 | } |
1814 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1815 | ASSERT((new_size != 0) || | |
1816 | (fork == XFS_ATTR_FORK) || | |
1817 | (ip->i_delayed_blks == 0)); | |
1818 | ASSERT((new_size != 0) || | |
1819 | (fork == XFS_ATTR_FORK) || | |
1820 | (ip->i_d.di_nextents == 0)); | |
1821 | xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0); | |
1822 | return 0; | |
1823 | } | |
1824 | ||
1da177e4 LT |
1825 | /* |
1826 | * This is called when the inode's link count goes to 0. | |
1827 | * We place the on-disk inode on a list in the AGI. It | |
1828 | * will be pulled from this list when the inode is freed. | |
1829 | */ | |
1830 | int | |
1831 | xfs_iunlink( | |
1832 | xfs_trans_t *tp, | |
1833 | xfs_inode_t *ip) | |
1834 | { | |
1835 | xfs_mount_t *mp; | |
1836 | xfs_agi_t *agi; | |
1837 | xfs_dinode_t *dip; | |
1838 | xfs_buf_t *agibp; | |
1839 | xfs_buf_t *ibp; | |
1840 | xfs_agnumber_t agno; | |
1841 | xfs_daddr_t agdaddr; | |
1842 | xfs_agino_t agino; | |
1843 | short bucket_index; | |
1844 | int offset; | |
1845 | int error; | |
1846 | int agi_ok; | |
1847 | ||
1848 | ASSERT(ip->i_d.di_nlink == 0); | |
1849 | ASSERT(ip->i_d.di_mode != 0); | |
1850 | ASSERT(ip->i_transp == tp); | |
1851 | ||
1852 | mp = tp->t_mountp; | |
1853 | ||
1854 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1855 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1856 | ||
1857 | /* | |
1858 | * Get the agi buffer first. It ensures lock ordering | |
1859 | * on the list. | |
1860 | */ | |
1861 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1862 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
859d7182 | 1863 | if (error) |
1da177e4 | 1864 | return error; |
859d7182 | 1865 | |
1da177e4 LT |
1866 | /* |
1867 | * Validate the magic number of the agi block. | |
1868 | */ | |
1869 | agi = XFS_BUF_TO_AGI(agibp); | |
1870 | agi_ok = | |
16259e7d CH |
1871 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
1872 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
1873 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK, |
1874 | XFS_RANDOM_IUNLINK))) { | |
1875 | XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi); | |
1876 | xfs_trans_brelse(tp, agibp); | |
1877 | return XFS_ERROR(EFSCORRUPTED); | |
1878 | } | |
1879 | /* | |
1880 | * Get the index into the agi hash table for the | |
1881 | * list this inode will go on. | |
1882 | */ | |
1883 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1884 | ASSERT(agino != 0); | |
1885 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1886 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1887 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1888 | |
16259e7d | 1889 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) { |
1da177e4 LT |
1890 | /* |
1891 | * There is already another inode in the bucket we need | |
1892 | * to add ourselves to. Add us at the front of the list. | |
1893 | * Here we put the head pointer into our next pointer, | |
1894 | * and then we fall through to point the head at us. | |
1895 | */ | |
a3f74ffb | 1896 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK); |
c319b58b VA |
1897 | if (error) |
1898 | return error; | |
1899 | ||
347d1c01 | 1900 | ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO); |
1da177e4 LT |
1901 | /* both on-disk, don't endian flip twice */ |
1902 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; | |
1903 | offset = ip->i_boffset + | |
1904 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1905 | xfs_trans_inode_buf(tp, ibp); | |
1906 | xfs_trans_log_buf(tp, ibp, offset, | |
1907 | (offset + sizeof(xfs_agino_t) - 1)); | |
1908 | xfs_inobp_check(mp, ibp); | |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * Point the bucket head pointer at the inode being inserted. | |
1913 | */ | |
1914 | ASSERT(agino != 0); | |
16259e7d | 1915 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1916 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1917 | (sizeof(xfs_agino_t) * bucket_index); | |
1918 | xfs_trans_log_buf(tp, agibp, offset, | |
1919 | (offset + sizeof(xfs_agino_t) - 1)); | |
1920 | return 0; | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * Pull the on-disk inode from the AGI unlinked list. | |
1925 | */ | |
1926 | STATIC int | |
1927 | xfs_iunlink_remove( | |
1928 | xfs_trans_t *tp, | |
1929 | xfs_inode_t *ip) | |
1930 | { | |
1931 | xfs_ino_t next_ino; | |
1932 | xfs_mount_t *mp; | |
1933 | xfs_agi_t *agi; | |
1934 | xfs_dinode_t *dip; | |
1935 | xfs_buf_t *agibp; | |
1936 | xfs_buf_t *ibp; | |
1937 | xfs_agnumber_t agno; | |
1938 | xfs_daddr_t agdaddr; | |
1939 | xfs_agino_t agino; | |
1940 | xfs_agino_t next_agino; | |
1941 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1942 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1943 | short bucket_index; |
6fdf8ccc | 1944 | int offset, last_offset = 0; |
1da177e4 LT |
1945 | int error; |
1946 | int agi_ok; | |
1947 | ||
1948 | /* | |
1949 | * First pull the on-disk inode from the AGI unlinked list. | |
1950 | */ | |
1951 | mp = tp->t_mountp; | |
1952 | ||
1953 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1954 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1955 | ||
1956 | /* | |
1957 | * Get the agi buffer first. It ensures lock ordering | |
1958 | * on the list. | |
1959 | */ | |
1960 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1961 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
1962 | if (error) { | |
1963 | cmn_err(CE_WARN, | |
1964 | "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.", | |
1965 | error, mp->m_fsname); | |
1966 | return error; | |
1967 | } | |
1968 | /* | |
1969 | * Validate the magic number of the agi block. | |
1970 | */ | |
1971 | agi = XFS_BUF_TO_AGI(agibp); | |
1972 | agi_ok = | |
16259e7d CH |
1973 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
1974 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
1975 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE, |
1976 | XFS_RANDOM_IUNLINK_REMOVE))) { | |
1977 | XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW, | |
1978 | mp, agi); | |
1979 | xfs_trans_brelse(tp, agibp); | |
1980 | cmn_err(CE_WARN, | |
1981 | "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.", | |
1982 | mp->m_fsname); | |
1983 | return XFS_ERROR(EFSCORRUPTED); | |
1984 | } | |
1985 | /* | |
1986 | * Get the index into the agi hash table for the | |
1987 | * list this inode will go on. | |
1988 | */ | |
1989 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1990 | ASSERT(agino != 0); | |
1991 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
16259e7d | 1992 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO); |
1da177e4 LT |
1993 | ASSERT(agi->agi_unlinked[bucket_index]); |
1994 | ||
16259e7d | 1995 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 LT |
1996 | /* |
1997 | * We're at the head of the list. Get the inode's | |
1998 | * on-disk buffer to see if there is anyone after us | |
1999 | * on the list. Only modify our next pointer if it | |
2000 | * is not already NULLAGINO. This saves us the overhead | |
2001 | * of dealing with the buffer when there is no need to | |
2002 | * change it. | |
2003 | */ | |
a3f74ffb | 2004 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK); |
1da177e4 LT |
2005 | if (error) { |
2006 | cmn_err(CE_WARN, | |
2007 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2008 | error, mp->m_fsname); | |
2009 | return error; | |
2010 | } | |
347d1c01 | 2011 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2012 | ASSERT(next_agino != 0); |
2013 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2014 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
2015 | offset = ip->i_boffset + |
2016 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2017 | xfs_trans_inode_buf(tp, ibp); | |
2018 | xfs_trans_log_buf(tp, ibp, offset, | |
2019 | (offset + sizeof(xfs_agino_t) - 1)); | |
2020 | xfs_inobp_check(mp, ibp); | |
2021 | } else { | |
2022 | xfs_trans_brelse(tp, ibp); | |
2023 | } | |
2024 | /* | |
2025 | * Point the bucket head pointer at the next inode. | |
2026 | */ | |
2027 | ASSERT(next_agino != 0); | |
2028 | ASSERT(next_agino != agino); | |
16259e7d | 2029 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2030 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2031 | (sizeof(xfs_agino_t) * bucket_index); | |
2032 | xfs_trans_log_buf(tp, agibp, offset, | |
2033 | (offset + sizeof(xfs_agino_t) - 1)); | |
2034 | } else { | |
2035 | /* | |
2036 | * We need to search the list for the inode being freed. | |
2037 | */ | |
16259e7d | 2038 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2039 | last_ibp = NULL; |
2040 | while (next_agino != agino) { | |
2041 | /* | |
2042 | * If the last inode wasn't the one pointing to | |
2043 | * us, then release its buffer since we're not | |
2044 | * going to do anything with it. | |
2045 | */ | |
2046 | if (last_ibp != NULL) { | |
2047 | xfs_trans_brelse(tp, last_ibp); | |
2048 | } | |
2049 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); | |
2050 | error = xfs_inotobp(mp, tp, next_ino, &last_dip, | |
c679eef0 | 2051 | &last_ibp, &last_offset, 0); |
1da177e4 LT |
2052 | if (error) { |
2053 | cmn_err(CE_WARN, | |
2054 | "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.", | |
2055 | error, mp->m_fsname); | |
2056 | return error; | |
2057 | } | |
347d1c01 | 2058 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2059 | ASSERT(next_agino != NULLAGINO); |
2060 | ASSERT(next_agino != 0); | |
2061 | } | |
2062 | /* | |
2063 | * Now last_ibp points to the buffer previous to us on | |
2064 | * the unlinked list. Pull us from the list. | |
2065 | */ | |
a3f74ffb | 2066 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK); |
1da177e4 LT |
2067 | if (error) { |
2068 | cmn_err(CE_WARN, | |
2069 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2070 | error, mp->m_fsname); | |
2071 | return error; | |
2072 | } | |
347d1c01 | 2073 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2074 | ASSERT(next_agino != 0); |
2075 | ASSERT(next_agino != agino); | |
2076 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2077 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
2078 | offset = ip->i_boffset + |
2079 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2080 | xfs_trans_inode_buf(tp, ibp); | |
2081 | xfs_trans_log_buf(tp, ibp, offset, | |
2082 | (offset + sizeof(xfs_agino_t) - 1)); | |
2083 | xfs_inobp_check(mp, ibp); | |
2084 | } else { | |
2085 | xfs_trans_brelse(tp, ibp); | |
2086 | } | |
2087 | /* | |
2088 | * Point the previous inode on the list to the next inode. | |
2089 | */ | |
347d1c01 | 2090 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2091 | ASSERT(next_agino != 0); |
2092 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
2093 | xfs_trans_inode_buf(tp, last_ibp); | |
2094 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2095 | (offset + sizeof(xfs_agino_t) - 1)); | |
2096 | xfs_inobp_check(mp, last_ibp); | |
2097 | } | |
2098 | return 0; | |
2099 | } | |
2100 | ||
ba0f32d4 | 2101 | STATIC void |
1da177e4 LT |
2102 | xfs_ifree_cluster( |
2103 | xfs_inode_t *free_ip, | |
2104 | xfs_trans_t *tp, | |
2105 | xfs_ino_t inum) | |
2106 | { | |
2107 | xfs_mount_t *mp = free_ip->i_mount; | |
2108 | int blks_per_cluster; | |
2109 | int nbufs; | |
2110 | int ninodes; | |
2111 | int i, j, found, pre_flushed; | |
2112 | xfs_daddr_t blkno; | |
2113 | xfs_buf_t *bp; | |
1da177e4 LT |
2114 | xfs_inode_t *ip, **ip_found; |
2115 | xfs_inode_log_item_t *iip; | |
2116 | xfs_log_item_t *lip; | |
da353b0d | 2117 | xfs_perag_t *pag = xfs_get_perag(mp, inum); |
1da177e4 LT |
2118 | |
2119 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
2120 | blks_per_cluster = 1; | |
2121 | ninodes = mp->m_sb.sb_inopblock; | |
2122 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
2123 | } else { | |
2124 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
2125 | mp->m_sb.sb_blocksize; | |
2126 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
2127 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
2128 | } | |
2129 | ||
2130 | ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS); | |
2131 | ||
2132 | for (j = 0; j < nbufs; j++, inum += ninodes) { | |
2133 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
2134 | XFS_INO_TO_AGBNO(mp, inum)); | |
2135 | ||
2136 | ||
2137 | /* | |
2138 | * Look for each inode in memory and attempt to lock it, | |
2139 | * we can be racing with flush and tail pushing here. | |
2140 | * any inode we get the locks on, add to an array of | |
2141 | * inode items to process later. | |
2142 | * | |
2143 | * The get the buffer lock, we could beat a flush | |
2144 | * or tail pushing thread to the lock here, in which | |
2145 | * case they will go looking for the inode buffer | |
2146 | * and fail, we need some other form of interlock | |
2147 | * here. | |
2148 | */ | |
2149 | found = 0; | |
2150 | for (i = 0; i < ninodes; i++) { | |
da353b0d DC |
2151 | read_lock(&pag->pag_ici_lock); |
2152 | ip = radix_tree_lookup(&pag->pag_ici_root, | |
2153 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 LT |
2154 | |
2155 | /* Inode not in memory or we found it already, | |
2156 | * nothing to do | |
2157 | */ | |
7a18c386 | 2158 | if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) { |
da353b0d | 2159 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2160 | continue; |
2161 | } | |
2162 | ||
2163 | if (xfs_inode_clean(ip)) { | |
da353b0d | 2164 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2165 | continue; |
2166 | } | |
2167 | ||
2168 | /* If we can get the locks then add it to the | |
2169 | * list, otherwise by the time we get the bp lock | |
2170 | * below it will already be attached to the | |
2171 | * inode buffer. | |
2172 | */ | |
2173 | ||
2174 | /* This inode will already be locked - by us, lets | |
2175 | * keep it that way. | |
2176 | */ | |
2177 | ||
2178 | if (ip == free_ip) { | |
2179 | if (xfs_iflock_nowait(ip)) { | |
7a18c386 | 2180 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 LT |
2181 | if (xfs_inode_clean(ip)) { |
2182 | xfs_ifunlock(ip); | |
2183 | } else { | |
2184 | ip_found[found++] = ip; | |
2185 | } | |
2186 | } | |
da353b0d | 2187 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2188 | continue; |
2189 | } | |
2190 | ||
2191 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2192 | if (xfs_iflock_nowait(ip)) { | |
7a18c386 | 2193 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 LT |
2194 | |
2195 | if (xfs_inode_clean(ip)) { | |
2196 | xfs_ifunlock(ip); | |
2197 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2198 | } else { | |
2199 | ip_found[found++] = ip; | |
2200 | } | |
2201 | } else { | |
2202 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2203 | } | |
2204 | } | |
da353b0d | 2205 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2206 | } |
2207 | ||
2208 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
2209 | mp->m_bsize * blks_per_cluster, | |
2210 | XFS_BUF_LOCK); | |
2211 | ||
2212 | pre_flushed = 0; | |
2213 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
2214 | while (lip) { | |
2215 | if (lip->li_type == XFS_LI_INODE) { | |
2216 | iip = (xfs_inode_log_item_t *)lip; | |
2217 | ASSERT(iip->ili_logged == 1); | |
2218 | lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done; | |
7b2e2a31 DC |
2219 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2220 | &iip->ili_flush_lsn, | |
2221 | &iip->ili_item.li_lsn); | |
e5ffd2bb | 2222 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); |
1da177e4 LT |
2223 | pre_flushed++; |
2224 | } | |
2225 | lip = lip->li_bio_list; | |
2226 | } | |
2227 | ||
2228 | for (i = 0; i < found; i++) { | |
2229 | ip = ip_found[i]; | |
2230 | iip = ip->i_itemp; | |
2231 | ||
2232 | if (!iip) { | |
2233 | ip->i_update_core = 0; | |
2234 | xfs_ifunlock(ip); | |
2235 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2236 | continue; | |
2237 | } | |
2238 | ||
2239 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
2240 | iip->ili_format.ilf_fields = 0; | |
2241 | iip->ili_logged = 1; | |
7b2e2a31 DC |
2242 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2243 | &iip->ili_item.li_lsn); | |
1da177e4 LT |
2244 | |
2245 | xfs_buf_attach_iodone(bp, | |
2246 | (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
2247 | xfs_istale_done, (xfs_log_item_t *)iip); | |
2248 | if (ip != free_ip) { | |
2249 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2250 | } | |
2251 | } | |
2252 | ||
2253 | if (found || pre_flushed) | |
2254 | xfs_trans_stale_inode_buf(tp, bp); | |
2255 | xfs_trans_binval(tp, bp); | |
2256 | } | |
2257 | ||
f0e2d93c | 2258 | kmem_free(ip_found); |
da353b0d | 2259 | xfs_put_perag(mp, pag); |
1da177e4 LT |
2260 | } |
2261 | ||
2262 | /* | |
2263 | * This is called to return an inode to the inode free list. | |
2264 | * The inode should already be truncated to 0 length and have | |
2265 | * no pages associated with it. This routine also assumes that | |
2266 | * the inode is already a part of the transaction. | |
2267 | * | |
2268 | * The on-disk copy of the inode will have been added to the list | |
2269 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2270 | * that list atomically with respect to freeing it here. | |
2271 | */ | |
2272 | int | |
2273 | xfs_ifree( | |
2274 | xfs_trans_t *tp, | |
2275 | xfs_inode_t *ip, | |
2276 | xfs_bmap_free_t *flist) | |
2277 | { | |
2278 | int error; | |
2279 | int delete; | |
2280 | xfs_ino_t first_ino; | |
c319b58b VA |
2281 | xfs_dinode_t *dip; |
2282 | xfs_buf_t *ibp; | |
1da177e4 | 2283 | |
579aa9ca | 2284 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
2285 | ASSERT(ip->i_transp == tp); |
2286 | ASSERT(ip->i_d.di_nlink == 0); | |
2287 | ASSERT(ip->i_d.di_nextents == 0); | |
2288 | ASSERT(ip->i_d.di_anextents == 0); | |
ba87ea69 | 2289 | ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) || |
1da177e4 LT |
2290 | ((ip->i_d.di_mode & S_IFMT) != S_IFREG)); |
2291 | ASSERT(ip->i_d.di_nblocks == 0); | |
2292 | ||
2293 | /* | |
2294 | * Pull the on-disk inode from the AGI unlinked list. | |
2295 | */ | |
2296 | error = xfs_iunlink_remove(tp, ip); | |
2297 | if (error != 0) { | |
2298 | return error; | |
2299 | } | |
2300 | ||
2301 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
2302 | if (error != 0) { | |
2303 | return error; | |
2304 | } | |
2305 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
2306 | ip->i_d.di_flags = 0; | |
2307 | ip->i_d.di_dmevmask = 0; | |
2308 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
2309 | ip->i_df.if_ext_max = | |
2310 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
2311 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
2312 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2313 | /* | |
2314 | * Bump the generation count so no one will be confused | |
2315 | * by reincarnations of this inode. | |
2316 | */ | |
2317 | ip->i_d.di_gen++; | |
c319b58b | 2318 | |
1da177e4 LT |
2319 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2320 | ||
a3f74ffb | 2321 | error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK); |
c319b58b VA |
2322 | if (error) |
2323 | return error; | |
2324 | ||
2325 | /* | |
2326 | * Clear the on-disk di_mode. This is to prevent xfs_bulkstat | |
2327 | * from picking up this inode when it is reclaimed (its incore state | |
2328 | * initialzed but not flushed to disk yet). The in-core di_mode is | |
2329 | * already cleared and a corresponding transaction logged. | |
2330 | * The hack here just synchronizes the in-core to on-disk | |
2331 | * di_mode value in advance before the actual inode sync to disk. | |
2332 | * This is OK because the inode is already unlinked and would never | |
2333 | * change its di_mode again for this inode generation. | |
2334 | * This is a temporary hack that would require a proper fix | |
2335 | * in the future. | |
2336 | */ | |
2337 | dip->di_core.di_mode = 0; | |
2338 | ||
1da177e4 LT |
2339 | if (delete) { |
2340 | xfs_ifree_cluster(ip, tp, first_ino); | |
2341 | } | |
2342 | ||
2343 | return 0; | |
2344 | } | |
2345 | ||
2346 | /* | |
2347 | * Reallocate the space for if_broot based on the number of records | |
2348 | * being added or deleted as indicated in rec_diff. Move the records | |
2349 | * and pointers in if_broot to fit the new size. When shrinking this | |
2350 | * will eliminate holes between the records and pointers created by | |
2351 | * the caller. When growing this will create holes to be filled in | |
2352 | * by the caller. | |
2353 | * | |
2354 | * The caller must not request to add more records than would fit in | |
2355 | * the on-disk inode root. If the if_broot is currently NULL, then | |
2356 | * if we adding records one will be allocated. The caller must also | |
2357 | * not request that the number of records go below zero, although | |
2358 | * it can go to zero. | |
2359 | * | |
2360 | * ip -- the inode whose if_broot area is changing | |
2361 | * ext_diff -- the change in the number of records, positive or negative, | |
2362 | * requested for the if_broot array. | |
2363 | */ | |
2364 | void | |
2365 | xfs_iroot_realloc( | |
2366 | xfs_inode_t *ip, | |
2367 | int rec_diff, | |
2368 | int whichfork) | |
2369 | { | |
60197e8d | 2370 | struct xfs_mount *mp = ip->i_mount; |
1da177e4 LT |
2371 | int cur_max; |
2372 | xfs_ifork_t *ifp; | |
7cc95a82 | 2373 | struct xfs_btree_block *new_broot; |
1da177e4 LT |
2374 | int new_max; |
2375 | size_t new_size; | |
2376 | char *np; | |
2377 | char *op; | |
2378 | ||
2379 | /* | |
2380 | * Handle the degenerate case quietly. | |
2381 | */ | |
2382 | if (rec_diff == 0) { | |
2383 | return; | |
2384 | } | |
2385 | ||
2386 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2387 | if (rec_diff > 0) { | |
2388 | /* | |
2389 | * If there wasn't any memory allocated before, just | |
2390 | * allocate it now and get out. | |
2391 | */ | |
2392 | if (ifp->if_broot_bytes == 0) { | |
2393 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
7cc95a82 | 2394 | ifp->if_broot = kmem_alloc(new_size, KM_SLEEP); |
1da177e4 LT |
2395 | ifp->if_broot_bytes = (int)new_size; |
2396 | return; | |
2397 | } | |
2398 | ||
2399 | /* | |
2400 | * If there is already an existing if_broot, then we need | |
2401 | * to realloc() it and shift the pointers to their new | |
2402 | * location. The records don't change location because | |
2403 | * they are kept butted up against the btree block header. | |
2404 | */ | |
60197e8d | 2405 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2406 | new_max = cur_max + rec_diff; |
2407 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
7cc95a82 | 2408 | ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, |
1da177e4 LT |
2409 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ |
2410 | KM_SLEEP); | |
60197e8d CH |
2411 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
2412 | ifp->if_broot_bytes); | |
2413 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, | |
2414 | (int)new_size); | |
1da177e4 LT |
2415 | ifp->if_broot_bytes = (int)new_size; |
2416 | ASSERT(ifp->if_broot_bytes <= | |
2417 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2418 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2419 | return; | |
2420 | } | |
2421 | ||
2422 | /* | |
2423 | * rec_diff is less than 0. In this case, we are shrinking the | |
2424 | * if_broot buffer. It must already exist. If we go to zero | |
2425 | * records, just get rid of the root and clear the status bit. | |
2426 | */ | |
2427 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
60197e8d | 2428 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2429 | new_max = cur_max + rec_diff; |
2430 | ASSERT(new_max >= 0); | |
2431 | if (new_max > 0) | |
2432 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2433 | else | |
2434 | new_size = 0; | |
2435 | if (new_size > 0) { | |
7cc95a82 | 2436 | new_broot = kmem_alloc(new_size, KM_SLEEP); |
1da177e4 LT |
2437 | /* |
2438 | * First copy over the btree block header. | |
2439 | */ | |
7cc95a82 | 2440 | memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN); |
1da177e4 LT |
2441 | } else { |
2442 | new_broot = NULL; | |
2443 | ifp->if_flags &= ~XFS_IFBROOT; | |
2444 | } | |
2445 | ||
2446 | /* | |
2447 | * Only copy the records and pointers if there are any. | |
2448 | */ | |
2449 | if (new_max > 0) { | |
2450 | /* | |
2451 | * First copy the records. | |
2452 | */ | |
136341b4 CH |
2453 | op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); |
2454 | np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); | |
1da177e4 LT |
2455 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); |
2456 | ||
2457 | /* | |
2458 | * Then copy the pointers. | |
2459 | */ | |
60197e8d | 2460 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
1da177e4 | 2461 | ifp->if_broot_bytes); |
60197e8d | 2462 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, |
1da177e4 LT |
2463 | (int)new_size); |
2464 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2465 | } | |
f0e2d93c | 2466 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2467 | ifp->if_broot = new_broot; |
2468 | ifp->if_broot_bytes = (int)new_size; | |
2469 | ASSERT(ifp->if_broot_bytes <= | |
2470 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2471 | return; | |
2472 | } | |
2473 | ||
2474 | ||
1da177e4 LT |
2475 | /* |
2476 | * This is called when the amount of space needed for if_data | |
2477 | * is increased or decreased. The change in size is indicated by | |
2478 | * the number of bytes that need to be added or deleted in the | |
2479 | * byte_diff parameter. | |
2480 | * | |
2481 | * If the amount of space needed has decreased below the size of the | |
2482 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2483 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2484 | * to what is needed. | |
2485 | * | |
2486 | * ip -- the inode whose if_data area is changing | |
2487 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2488 | * requested for the if_data array. | |
2489 | */ | |
2490 | void | |
2491 | xfs_idata_realloc( | |
2492 | xfs_inode_t *ip, | |
2493 | int byte_diff, | |
2494 | int whichfork) | |
2495 | { | |
2496 | xfs_ifork_t *ifp; | |
2497 | int new_size; | |
2498 | int real_size; | |
2499 | ||
2500 | if (byte_diff == 0) { | |
2501 | return; | |
2502 | } | |
2503 | ||
2504 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2505 | new_size = (int)ifp->if_bytes + byte_diff; | |
2506 | ASSERT(new_size >= 0); | |
2507 | ||
2508 | if (new_size == 0) { | |
2509 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
f0e2d93c | 2510 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2511 | } |
2512 | ifp->if_u1.if_data = NULL; | |
2513 | real_size = 0; | |
2514 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2515 | /* | |
2516 | * If the valid extents/data can fit in if_inline_ext/data, | |
2517 | * copy them from the malloc'd vector and free it. | |
2518 | */ | |
2519 | if (ifp->if_u1.if_data == NULL) { | |
2520 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2521 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2522 | ASSERT(ifp->if_real_bytes != 0); | |
2523 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2524 | new_size); | |
f0e2d93c | 2525 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2526 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
2527 | } | |
2528 | real_size = 0; | |
2529 | } else { | |
2530 | /* | |
2531 | * Stuck with malloc/realloc. | |
2532 | * For inline data, the underlying buffer must be | |
2533 | * a multiple of 4 bytes in size so that it can be | |
2534 | * logged and stay on word boundaries. We enforce | |
2535 | * that here. | |
2536 | */ | |
2537 | real_size = roundup(new_size, 4); | |
2538 | if (ifp->if_u1.if_data == NULL) { | |
2539 | ASSERT(ifp->if_real_bytes == 0); | |
2540 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2541 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2542 | /* | |
2543 | * Only do the realloc if the underlying size | |
2544 | * is really changing. | |
2545 | */ | |
2546 | if (ifp->if_real_bytes != real_size) { | |
2547 | ifp->if_u1.if_data = | |
2548 | kmem_realloc(ifp->if_u1.if_data, | |
2549 | real_size, | |
2550 | ifp->if_real_bytes, | |
2551 | KM_SLEEP); | |
2552 | } | |
2553 | } else { | |
2554 | ASSERT(ifp->if_real_bytes == 0); | |
2555 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2556 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, | |
2557 | ifp->if_bytes); | |
2558 | } | |
2559 | } | |
2560 | ifp->if_real_bytes = real_size; | |
2561 | ifp->if_bytes = new_size; | |
2562 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2563 | } | |
2564 | ||
2565 | ||
2566 | ||
2567 | ||
2568 | /* | |
2569 | * Map inode to disk block and offset. | |
2570 | * | |
2571 | * mp -- the mount point structure for the current file system | |
2572 | * tp -- the current transaction | |
2573 | * ino -- the inode number of the inode to be located | |
2574 | * imap -- this structure is filled in with the information necessary | |
2575 | * to retrieve the given inode from disk | |
2576 | * flags -- flags to pass to xfs_dilocate indicating whether or not | |
2577 | * lookups in the inode btree were OK or not | |
2578 | */ | |
2579 | int | |
2580 | xfs_imap( | |
2581 | xfs_mount_t *mp, | |
2582 | xfs_trans_t *tp, | |
2583 | xfs_ino_t ino, | |
2584 | xfs_imap_t *imap, | |
2585 | uint flags) | |
2586 | { | |
2587 | xfs_fsblock_t fsbno; | |
2588 | int len; | |
2589 | int off; | |
2590 | int error; | |
2591 | ||
2592 | fsbno = imap->im_blkno ? | |
2593 | XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; | |
2594 | error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); | |
4ae29b43 | 2595 | if (error) |
1da177e4 | 2596 | return error; |
4ae29b43 | 2597 | |
1da177e4 LT |
2598 | imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); |
2599 | imap->im_len = XFS_FSB_TO_BB(mp, len); | |
2600 | imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); | |
2601 | imap->im_ioffset = (ushort)off; | |
2602 | imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); | |
4ae29b43 DC |
2603 | |
2604 | /* | |
2605 | * If the inode number maps to a block outside the bounds | |
2606 | * of the file system then return NULL rather than calling | |
2607 | * read_buf and panicing when we get an error from the | |
2608 | * driver. | |
2609 | */ | |
2610 | if ((imap->im_blkno + imap->im_len) > | |
2611 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
2612 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: " | |
2613 | "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > " | |
2614 | " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)", | |
2615 | (unsigned long long) imap->im_blkno, | |
2616 | (unsigned long long) imap->im_len, | |
2617 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2618 | return EINVAL; | |
2619 | } | |
1da177e4 LT |
2620 | return 0; |
2621 | } | |
2622 | ||
2623 | void | |
2624 | xfs_idestroy_fork( | |
2625 | xfs_inode_t *ip, | |
2626 | int whichfork) | |
2627 | { | |
2628 | xfs_ifork_t *ifp; | |
2629 | ||
2630 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2631 | if (ifp->if_broot != NULL) { | |
f0e2d93c | 2632 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2633 | ifp->if_broot = NULL; |
2634 | } | |
2635 | ||
2636 | /* | |
2637 | * If the format is local, then we can't have an extents | |
2638 | * array so just look for an inline data array. If we're | |
2639 | * not local then we may or may not have an extents list, | |
2640 | * so check and free it up if we do. | |
2641 | */ | |
2642 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2643 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2644 | (ifp->if_u1.if_data != NULL)) { | |
2645 | ASSERT(ifp->if_real_bytes != 0); | |
f0e2d93c | 2646 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2647 | ifp->if_u1.if_data = NULL; |
2648 | ifp->if_real_bytes = 0; | |
2649 | } | |
2650 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
0293ce3a MK |
2651 | ((ifp->if_flags & XFS_IFEXTIREC) || |
2652 | ((ifp->if_u1.if_extents != NULL) && | |
2653 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
1da177e4 | 2654 | ASSERT(ifp->if_real_bytes != 0); |
4eea22f0 | 2655 | xfs_iext_destroy(ifp); |
1da177e4 LT |
2656 | } |
2657 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2658 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2659 | ASSERT(ifp->if_real_bytes == 0); | |
2660 | if (whichfork == XFS_ATTR_FORK) { | |
2661 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2662 | ip->i_afp = NULL; | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | /* | |
2667 | * This is called free all the memory associated with an inode. | |
2668 | * It must free the inode itself and any buffers allocated for | |
2669 | * if_extents/if_data and if_broot. It must also free the lock | |
2670 | * associated with the inode. | |
bf904248 DC |
2671 | * |
2672 | * Note: because we don't initialise everything on reallocation out | |
2673 | * of the zone, we must ensure we nullify everything correctly before | |
2674 | * freeing the structure. | |
1da177e4 LT |
2675 | */ |
2676 | void | |
2677 | xfs_idestroy( | |
2678 | xfs_inode_t *ip) | |
2679 | { | |
1da177e4 LT |
2680 | switch (ip->i_d.di_mode & S_IFMT) { |
2681 | case S_IFREG: | |
2682 | case S_IFDIR: | |
2683 | case S_IFLNK: | |
2684 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
2685 | break; | |
2686 | } | |
2687 | if (ip->i_afp) | |
2688 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1543d79c | 2689 | |
cf441eeb | 2690 | #ifdef XFS_INODE_TRACE |
1543d79c CH |
2691 | ktrace_free(ip->i_trace); |
2692 | #endif | |
1da177e4 LT |
2693 | #ifdef XFS_BMAP_TRACE |
2694 | ktrace_free(ip->i_xtrace); | |
2695 | #endif | |
8c4ed633 | 2696 | #ifdef XFS_BTREE_TRACE |
1da177e4 LT |
2697 | ktrace_free(ip->i_btrace); |
2698 | #endif | |
2699 | #ifdef XFS_RW_TRACE | |
2700 | ktrace_free(ip->i_rwtrace); | |
2701 | #endif | |
2702 | #ifdef XFS_ILOCK_TRACE | |
2703 | ktrace_free(ip->i_lock_trace); | |
2704 | #endif | |
2705 | #ifdef XFS_DIR2_TRACE | |
2706 | ktrace_free(ip->i_dir_trace); | |
2707 | #endif | |
2708 | if (ip->i_itemp) { | |
f74eaf59 DC |
2709 | /* |
2710 | * Only if we are shutting down the fs will we see an | |
2711 | * inode still in the AIL. If it is there, we should remove | |
2712 | * it to prevent a use-after-free from occurring. | |
2713 | */ | |
f74eaf59 | 2714 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; |
783a2f65 | 2715 | struct xfs_ail *ailp = lip->li_ailp; |
f74eaf59 DC |
2716 | |
2717 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
2718 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
2719 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
783a2f65 | 2720 | spin_lock(&ailp->xa_lock); |
f74eaf59 | 2721 | if (lip->li_flags & XFS_LI_IN_AIL) |
783a2f65 | 2722 | xfs_trans_ail_delete(ailp, lip); |
f74eaf59 | 2723 | else |
783a2f65 | 2724 | spin_unlock(&ailp->xa_lock); |
f74eaf59 | 2725 | } |
1da177e4 | 2726 | xfs_inode_item_destroy(ip); |
07c8f675 | 2727 | ip->i_itemp = NULL; |
1da177e4 | 2728 | } |
07c8f675 DC |
2729 | /* asserts to verify all state is correct here */ |
2730 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
2731 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
2732 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
11654513 | 2733 | ASSERT(completion_done(&ip->i_flush)); |
1da177e4 LT |
2734 | kmem_zone_free(xfs_inode_zone, ip); |
2735 | } | |
2736 | ||
2737 | ||
2738 | /* | |
2739 | * Increment the pin count of the given buffer. | |
2740 | * This value is protected by ipinlock spinlock in the mount structure. | |
2741 | */ | |
2742 | void | |
2743 | xfs_ipin( | |
2744 | xfs_inode_t *ip) | |
2745 | { | |
579aa9ca | 2746 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
2747 | |
2748 | atomic_inc(&ip->i_pincount); | |
2749 | } | |
2750 | ||
2751 | /* | |
2752 | * Decrement the pin count of the given inode, and wake up | |
2753 | * anyone in xfs_iwait_unpin() if the count goes to 0. The | |
c41564b5 | 2754 | * inode must have been previously pinned with a call to xfs_ipin(). |
1da177e4 LT |
2755 | */ |
2756 | void | |
2757 | xfs_iunpin( | |
2758 | xfs_inode_t *ip) | |
2759 | { | |
2760 | ASSERT(atomic_read(&ip->i_pincount) > 0); | |
2761 | ||
5d51eff4 | 2762 | if (atomic_dec_and_test(&ip->i_pincount)) |
1da177e4 | 2763 | wake_up(&ip->i_ipin_wait); |
1da177e4 LT |
2764 | } |
2765 | ||
2766 | /* | |
a3f74ffb DC |
2767 | * This is called to unpin an inode. It can be directed to wait or to return |
2768 | * immediately without waiting for the inode to be unpinned. The caller must | |
2769 | * have the inode locked in at least shared mode so that the buffer cannot be | |
2770 | * subsequently pinned once someone is waiting for it to be unpinned. | |
1da177e4 | 2771 | */ |
ba0f32d4 | 2772 | STATIC void |
a3f74ffb DC |
2773 | __xfs_iunpin_wait( |
2774 | xfs_inode_t *ip, | |
2775 | int wait) | |
1da177e4 | 2776 | { |
a3f74ffb | 2777 | xfs_inode_log_item_t *iip = ip->i_itemp; |
1da177e4 | 2778 | |
579aa9ca | 2779 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
a3f74ffb | 2780 | if (atomic_read(&ip->i_pincount) == 0) |
1da177e4 | 2781 | return; |
1da177e4 | 2782 | |
a3f74ffb DC |
2783 | /* Give the log a push to start the unpinning I/O */ |
2784 | xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ? | |
2785 | iip->ili_last_lsn : 0, XFS_LOG_FORCE); | |
2786 | if (wait) | |
2787 | wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0)); | |
2788 | } | |
1da177e4 | 2789 | |
a3f74ffb DC |
2790 | static inline void |
2791 | xfs_iunpin_wait( | |
2792 | xfs_inode_t *ip) | |
2793 | { | |
2794 | __xfs_iunpin_wait(ip, 1); | |
2795 | } | |
1da177e4 | 2796 | |
a3f74ffb DC |
2797 | static inline void |
2798 | xfs_iunpin_nowait( | |
2799 | xfs_inode_t *ip) | |
2800 | { | |
2801 | __xfs_iunpin_wait(ip, 0); | |
1da177e4 LT |
2802 | } |
2803 | ||
2804 | ||
2805 | /* | |
2806 | * xfs_iextents_copy() | |
2807 | * | |
2808 | * This is called to copy the REAL extents (as opposed to the delayed | |
2809 | * allocation extents) from the inode into the given buffer. It | |
2810 | * returns the number of bytes copied into the buffer. | |
2811 | * | |
2812 | * If there are no delayed allocation extents, then we can just | |
2813 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2814 | * examine each extent in turn and skip those which are delayed. | |
2815 | */ | |
2816 | int | |
2817 | xfs_iextents_copy( | |
2818 | xfs_inode_t *ip, | |
a6f64d4a | 2819 | xfs_bmbt_rec_t *dp, |
1da177e4 LT |
2820 | int whichfork) |
2821 | { | |
2822 | int copied; | |
1da177e4 LT |
2823 | int i; |
2824 | xfs_ifork_t *ifp; | |
2825 | int nrecs; | |
2826 | xfs_fsblock_t start_block; | |
2827 | ||
2828 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
579aa9ca | 2829 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 LT |
2830 | ASSERT(ifp->if_bytes > 0); |
2831 | ||
2832 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3a59c94c | 2833 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
1da177e4 LT |
2834 | ASSERT(nrecs > 0); |
2835 | ||
2836 | /* | |
2837 | * There are some delayed allocation extents in the | |
2838 | * inode, so copy the extents one at a time and skip | |
2839 | * the delayed ones. There must be at least one | |
2840 | * non-delayed extent. | |
2841 | */ | |
1da177e4 LT |
2842 | copied = 0; |
2843 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a | 2844 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
1da177e4 LT |
2845 | start_block = xfs_bmbt_get_startblock(ep); |
2846 | if (ISNULLSTARTBLOCK(start_block)) { | |
2847 | /* | |
2848 | * It's a delayed allocation extent, so skip it. | |
2849 | */ | |
1da177e4 LT |
2850 | continue; |
2851 | } | |
2852 | ||
2853 | /* Translate to on disk format */ | |
cd8b0a97 CH |
2854 | put_unaligned(cpu_to_be64(ep->l0), &dp->l0); |
2855 | put_unaligned(cpu_to_be64(ep->l1), &dp->l1); | |
a6f64d4a | 2856 | dp++; |
1da177e4 LT |
2857 | copied++; |
2858 | } | |
2859 | ASSERT(copied != 0); | |
a6f64d4a | 2860 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
2861 | |
2862 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2863 | } | |
2864 | ||
2865 | /* | |
2866 | * Each of the following cases stores data into the same region | |
2867 | * of the on-disk inode, so only one of them can be valid at | |
2868 | * any given time. While it is possible to have conflicting formats | |
2869 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2870 | * in EXTENTS format, this can only happen when the fork has | |
2871 | * changed formats after being modified but before being flushed. | |
2872 | * In these cases, the format always takes precedence, because the | |
2873 | * format indicates the current state of the fork. | |
2874 | */ | |
2875 | /*ARGSUSED*/ | |
e4ac967b | 2876 | STATIC void |
1da177e4 LT |
2877 | xfs_iflush_fork( |
2878 | xfs_inode_t *ip, | |
2879 | xfs_dinode_t *dip, | |
2880 | xfs_inode_log_item_t *iip, | |
2881 | int whichfork, | |
2882 | xfs_buf_t *bp) | |
2883 | { | |
2884 | char *cp; | |
2885 | xfs_ifork_t *ifp; | |
2886 | xfs_mount_t *mp; | |
2887 | #ifdef XFS_TRANS_DEBUG | |
2888 | int first; | |
2889 | #endif | |
2890 | static const short brootflag[2] = | |
2891 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2892 | static const short dataflag[2] = | |
2893 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2894 | static const short extflag[2] = | |
2895 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2896 | ||
e4ac967b DC |
2897 | if (!iip) |
2898 | return; | |
1da177e4 LT |
2899 | ifp = XFS_IFORK_PTR(ip, whichfork); |
2900 | /* | |
2901 | * This can happen if we gave up in iformat in an error path, | |
2902 | * for the attribute fork. | |
2903 | */ | |
e4ac967b | 2904 | if (!ifp) { |
1da177e4 | 2905 | ASSERT(whichfork == XFS_ATTR_FORK); |
e4ac967b | 2906 | return; |
1da177e4 LT |
2907 | } |
2908 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2909 | mp = ip->i_mount; | |
2910 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2911 | case XFS_DINODE_FMT_LOCAL: | |
2912 | if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && | |
2913 | (ifp->if_bytes > 0)) { | |
2914 | ASSERT(ifp->if_u1.if_data != NULL); | |
2915 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2916 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2917 | } | |
1da177e4 LT |
2918 | break; |
2919 | ||
2920 | case XFS_DINODE_FMT_EXTENTS: | |
2921 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
2922 | !(iip->ili_format.ilf_fields & extflag[whichfork])); | |
4eea22f0 MK |
2923 | ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) || |
2924 | (ifp->if_bytes == 0)); | |
2925 | ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) || | |
2926 | (ifp->if_bytes > 0)); | |
1da177e4 LT |
2927 | if ((iip->ili_format.ilf_fields & extflag[whichfork]) && |
2928 | (ifp->if_bytes > 0)) { | |
2929 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); | |
2930 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
2931 | whichfork); | |
2932 | } | |
2933 | break; | |
2934 | ||
2935 | case XFS_DINODE_FMT_BTREE: | |
2936 | if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && | |
2937 | (ifp->if_broot_bytes > 0)) { | |
2938 | ASSERT(ifp->if_broot != NULL); | |
2939 | ASSERT(ifp->if_broot_bytes <= | |
2940 | (XFS_IFORK_SIZE(ip, whichfork) + | |
2941 | XFS_BROOT_SIZE_ADJ)); | |
60197e8d | 2942 | xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, |
1da177e4 LT |
2943 | (xfs_bmdr_block_t *)cp, |
2944 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
2945 | } | |
2946 | break; | |
2947 | ||
2948 | case XFS_DINODE_FMT_DEV: | |
2949 | if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { | |
2950 | ASSERT(whichfork == XFS_DATA_FORK); | |
347d1c01 | 2951 | dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev); |
1da177e4 LT |
2952 | } |
2953 | break; | |
2954 | ||
2955 | case XFS_DINODE_FMT_UUID: | |
2956 | if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { | |
2957 | ASSERT(whichfork == XFS_DATA_FORK); | |
2958 | memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, | |
2959 | sizeof(uuid_t)); | |
2960 | } | |
2961 | break; | |
2962 | ||
2963 | default: | |
2964 | ASSERT(0); | |
2965 | break; | |
2966 | } | |
1da177e4 LT |
2967 | } |
2968 | ||
bad55843 DC |
2969 | STATIC int |
2970 | xfs_iflush_cluster( | |
2971 | xfs_inode_t *ip, | |
2972 | xfs_buf_t *bp) | |
2973 | { | |
2974 | xfs_mount_t *mp = ip->i_mount; | |
2975 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
2976 | unsigned long first_index, mask; | |
c8f5f12e | 2977 | unsigned long inodes_per_cluster; |
bad55843 DC |
2978 | int ilist_size; |
2979 | xfs_inode_t **ilist; | |
2980 | xfs_inode_t *iq; | |
bad55843 DC |
2981 | int nr_found; |
2982 | int clcount = 0; | |
2983 | int bufwasdelwri; | |
2984 | int i; | |
2985 | ||
2986 | ASSERT(pag->pagi_inodeok); | |
2987 | ASSERT(pag->pag_ici_init); | |
2988 | ||
c8f5f12e DC |
2989 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
2990 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
49383b0e | 2991 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); |
bad55843 DC |
2992 | if (!ilist) |
2993 | return 0; | |
2994 | ||
2995 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); | |
2996 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
2997 | read_lock(&pag->pag_ici_lock); | |
2998 | /* really need a gang lookup range call here */ | |
2999 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
c8f5f12e | 3000 | first_index, inodes_per_cluster); |
bad55843 DC |
3001 | if (nr_found == 0) |
3002 | goto out_free; | |
3003 | ||
3004 | for (i = 0; i < nr_found; i++) { | |
3005 | iq = ilist[i]; | |
3006 | if (iq == ip) | |
3007 | continue; | |
3008 | /* if the inode lies outside this cluster, we're done. */ | |
3009 | if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) | |
3010 | break; | |
3011 | /* | |
3012 | * Do an un-protected check to see if the inode is dirty and | |
3013 | * is a candidate for flushing. These checks will be repeated | |
3014 | * later after the appropriate locks are acquired. | |
3015 | */ | |
33540408 | 3016 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 3017 | continue; |
bad55843 DC |
3018 | |
3019 | /* | |
3020 | * Try to get locks. If any are unavailable or it is pinned, | |
3021 | * then this inode cannot be flushed and is skipped. | |
3022 | */ | |
3023 | ||
3024 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
3025 | continue; | |
3026 | if (!xfs_iflock_nowait(iq)) { | |
3027 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3028 | continue; | |
3029 | } | |
3030 | if (xfs_ipincount(iq)) { | |
3031 | xfs_ifunlock(iq); | |
3032 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3033 | continue; | |
3034 | } | |
3035 | ||
3036 | /* | |
3037 | * arriving here means that this inode can be flushed. First | |
3038 | * re-check that it's dirty before flushing. | |
3039 | */ | |
33540408 DC |
3040 | if (!xfs_inode_clean(iq)) { |
3041 | int error; | |
bad55843 DC |
3042 | error = xfs_iflush_int(iq, bp); |
3043 | if (error) { | |
3044 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3045 | goto cluster_corrupt_out; | |
3046 | } | |
3047 | clcount++; | |
3048 | } else { | |
3049 | xfs_ifunlock(iq); | |
3050 | } | |
3051 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3052 | } | |
3053 | ||
3054 | if (clcount) { | |
3055 | XFS_STATS_INC(xs_icluster_flushcnt); | |
3056 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
3057 | } | |
3058 | ||
3059 | out_free: | |
3060 | read_unlock(&pag->pag_ici_lock); | |
f0e2d93c | 3061 | kmem_free(ilist); |
bad55843 DC |
3062 | return 0; |
3063 | ||
3064 | ||
3065 | cluster_corrupt_out: | |
3066 | /* | |
3067 | * Corruption detected in the clustering loop. Invalidate the | |
3068 | * inode buffer and shut down the filesystem. | |
3069 | */ | |
3070 | read_unlock(&pag->pag_ici_lock); | |
3071 | /* | |
3072 | * Clean up the buffer. If it was B_DELWRI, just release it -- | |
3073 | * brelse can handle it with no problems. If not, shut down the | |
3074 | * filesystem before releasing the buffer. | |
3075 | */ | |
3076 | bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp); | |
3077 | if (bufwasdelwri) | |
3078 | xfs_buf_relse(bp); | |
3079 | ||
3080 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3081 | ||
3082 | if (!bufwasdelwri) { | |
3083 | /* | |
3084 | * Just like incore_relse: if we have b_iodone functions, | |
3085 | * mark the buffer as an error and call them. Otherwise | |
3086 | * mark it as stale and brelse. | |
3087 | */ | |
3088 | if (XFS_BUF_IODONE_FUNC(bp)) { | |
3089 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
3090 | XFS_BUF_UNDONE(bp); | |
3091 | XFS_BUF_STALE(bp); | |
3092 | XFS_BUF_SHUT(bp); | |
3093 | XFS_BUF_ERROR(bp,EIO); | |
3094 | xfs_biodone(bp); | |
3095 | } else { | |
3096 | XFS_BUF_STALE(bp); | |
3097 | xfs_buf_relse(bp); | |
3098 | } | |
3099 | } | |
3100 | ||
3101 | /* | |
3102 | * Unlocks the flush lock | |
3103 | */ | |
3104 | xfs_iflush_abort(iq); | |
f0e2d93c | 3105 | kmem_free(ilist); |
bad55843 DC |
3106 | return XFS_ERROR(EFSCORRUPTED); |
3107 | } | |
3108 | ||
1da177e4 LT |
3109 | /* |
3110 | * xfs_iflush() will write a modified inode's changes out to the | |
3111 | * inode's on disk home. The caller must have the inode lock held | |
c63942d3 DC |
3112 | * in at least shared mode and the inode flush completion must be |
3113 | * active as well. The inode lock will still be held upon return from | |
1da177e4 | 3114 | * the call and the caller is free to unlock it. |
c63942d3 | 3115 | * The inode flush will be completed when the inode reaches the disk. |
1da177e4 LT |
3116 | * The flags indicate how the inode's buffer should be written out. |
3117 | */ | |
3118 | int | |
3119 | xfs_iflush( | |
3120 | xfs_inode_t *ip, | |
3121 | uint flags) | |
3122 | { | |
3123 | xfs_inode_log_item_t *iip; | |
3124 | xfs_buf_t *bp; | |
3125 | xfs_dinode_t *dip; | |
3126 | xfs_mount_t *mp; | |
3127 | int error; | |
a3f74ffb | 3128 | int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK); |
bad55843 | 3129 | enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) }; |
1da177e4 LT |
3130 | |
3131 | XFS_STATS_INC(xs_iflush_count); | |
3132 | ||
579aa9ca | 3133 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
c63942d3 | 3134 | ASSERT(!completion_done(&ip->i_flush)); |
1da177e4 LT |
3135 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3136 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3137 | ||
3138 | iip = ip->i_itemp; | |
3139 | mp = ip->i_mount; | |
3140 | ||
3141 | /* | |
3142 | * If the inode isn't dirty, then just release the inode | |
3143 | * flush lock and do nothing. | |
3144 | */ | |
33540408 | 3145 | if (xfs_inode_clean(ip)) { |
1da177e4 LT |
3146 | xfs_ifunlock(ip); |
3147 | return 0; | |
3148 | } | |
3149 | ||
3150 | /* | |
a3f74ffb DC |
3151 | * We can't flush the inode until it is unpinned, so wait for it if we |
3152 | * are allowed to block. We know noone new can pin it, because we are | |
3153 | * holding the inode lock shared and you need to hold it exclusively to | |
3154 | * pin the inode. | |
3155 | * | |
3156 | * If we are not allowed to block, force the log out asynchronously so | |
3157 | * that when we come back the inode will be unpinned. If other inodes | |
3158 | * in the same cluster are dirty, they will probably write the inode | |
3159 | * out for us if they occur after the log force completes. | |
1da177e4 | 3160 | */ |
a3f74ffb DC |
3161 | if (noblock && xfs_ipincount(ip)) { |
3162 | xfs_iunpin_nowait(ip); | |
3163 | xfs_ifunlock(ip); | |
3164 | return EAGAIN; | |
3165 | } | |
1da177e4 LT |
3166 | xfs_iunpin_wait(ip); |
3167 | ||
3168 | /* | |
3169 | * This may have been unpinned because the filesystem is shutting | |
3170 | * down forcibly. If that's the case we must not write this inode | |
3171 | * to disk, because the log record didn't make it to disk! | |
3172 | */ | |
3173 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
3174 | ip->i_update_core = 0; | |
3175 | if (iip) | |
3176 | iip->ili_format.ilf_fields = 0; | |
3177 | xfs_ifunlock(ip); | |
3178 | return XFS_ERROR(EIO); | |
3179 | } | |
3180 | ||
1da177e4 LT |
3181 | /* |
3182 | * Decide how buffer will be flushed out. This is done before | |
3183 | * the call to xfs_iflush_int because this field is zeroed by it. | |
3184 | */ | |
3185 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3186 | /* | |
3187 | * Flush out the inode buffer according to the directions | |
3188 | * of the caller. In the cases where the caller has given | |
3189 | * us a choice choose the non-delwri case. This is because | |
3190 | * the inode is in the AIL and we need to get it out soon. | |
3191 | */ | |
3192 | switch (flags) { | |
3193 | case XFS_IFLUSH_SYNC: | |
3194 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3195 | flags = 0; | |
3196 | break; | |
a3f74ffb | 3197 | case XFS_IFLUSH_ASYNC_NOBLOCK: |
1da177e4 LT |
3198 | case XFS_IFLUSH_ASYNC: |
3199 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3200 | flags = INT_ASYNC; | |
3201 | break; | |
3202 | case XFS_IFLUSH_DELWRI: | |
3203 | flags = INT_DELWRI; | |
3204 | break; | |
3205 | default: | |
3206 | ASSERT(0); | |
3207 | flags = 0; | |
3208 | break; | |
3209 | } | |
3210 | } else { | |
3211 | switch (flags) { | |
3212 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3213 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3214 | case XFS_IFLUSH_DELWRI: | |
3215 | flags = INT_DELWRI; | |
3216 | break; | |
a3f74ffb | 3217 | case XFS_IFLUSH_ASYNC_NOBLOCK: |
1da177e4 LT |
3218 | case XFS_IFLUSH_ASYNC: |
3219 | flags = INT_ASYNC; | |
3220 | break; | |
3221 | case XFS_IFLUSH_SYNC: | |
3222 | flags = 0; | |
3223 | break; | |
3224 | default: | |
3225 | ASSERT(0); | |
3226 | flags = 0; | |
3227 | break; | |
3228 | } | |
3229 | } | |
3230 | ||
a3f74ffb DC |
3231 | /* |
3232 | * Get the buffer containing the on-disk inode. | |
3233 | */ | |
3234 | error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0, | |
3235 | noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK); | |
3236 | if (error || !bp) { | |
3237 | xfs_ifunlock(ip); | |
3238 | return error; | |
3239 | } | |
3240 | ||
1da177e4 LT |
3241 | /* |
3242 | * First flush out the inode that xfs_iflush was called with. | |
3243 | */ | |
3244 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3245 | if (error) |
1da177e4 | 3246 | goto corrupt_out; |
1da177e4 | 3247 | |
a3f74ffb DC |
3248 | /* |
3249 | * If the buffer is pinned then push on the log now so we won't | |
3250 | * get stuck waiting in the write for too long. | |
3251 | */ | |
3252 | if (XFS_BUF_ISPINNED(bp)) | |
3253 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
3254 | ||
1da177e4 LT |
3255 | /* |
3256 | * inode clustering: | |
3257 | * see if other inodes can be gathered into this write | |
3258 | */ | |
bad55843 DC |
3259 | error = xfs_iflush_cluster(ip, bp); |
3260 | if (error) | |
3261 | goto cluster_corrupt_out; | |
1da177e4 | 3262 | |
1da177e4 LT |
3263 | if (flags & INT_DELWRI) { |
3264 | xfs_bdwrite(mp, bp); | |
3265 | } else if (flags & INT_ASYNC) { | |
db7a19f2 | 3266 | error = xfs_bawrite(mp, bp); |
1da177e4 LT |
3267 | } else { |
3268 | error = xfs_bwrite(mp, bp); | |
3269 | } | |
3270 | return error; | |
3271 | ||
3272 | corrupt_out: | |
3273 | xfs_buf_relse(bp); | |
7d04a335 | 3274 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3275 | cluster_corrupt_out: |
1da177e4 LT |
3276 | /* |
3277 | * Unlocks the flush lock | |
3278 | */ | |
bad55843 | 3279 | xfs_iflush_abort(ip); |
1da177e4 LT |
3280 | return XFS_ERROR(EFSCORRUPTED); |
3281 | } | |
3282 | ||
3283 | ||
3284 | STATIC int | |
3285 | xfs_iflush_int( | |
3286 | xfs_inode_t *ip, | |
3287 | xfs_buf_t *bp) | |
3288 | { | |
3289 | xfs_inode_log_item_t *iip; | |
3290 | xfs_dinode_t *dip; | |
3291 | xfs_mount_t *mp; | |
3292 | #ifdef XFS_TRANS_DEBUG | |
3293 | int first; | |
3294 | #endif | |
1da177e4 | 3295 | |
579aa9ca | 3296 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
c63942d3 | 3297 | ASSERT(!completion_done(&ip->i_flush)); |
1da177e4 LT |
3298 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3299 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3300 | ||
3301 | iip = ip->i_itemp; | |
3302 | mp = ip->i_mount; | |
3303 | ||
3304 | ||
3305 | /* | |
3306 | * If the inode isn't dirty, then just release the inode | |
3307 | * flush lock and do nothing. | |
3308 | */ | |
33540408 | 3309 | if (xfs_inode_clean(ip)) { |
1da177e4 LT |
3310 | xfs_ifunlock(ip); |
3311 | return 0; | |
3312 | } | |
3313 | ||
3314 | /* set *dip = inode's place in the buffer */ | |
3315 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset); | |
3316 | ||
3317 | /* | |
3318 | * Clear i_update_core before copying out the data. | |
3319 | * This is for coordination with our timestamp updates | |
3320 | * that don't hold the inode lock. They will always | |
3321 | * update the timestamps BEFORE setting i_update_core, | |
3322 | * so if we clear i_update_core after they set it we | |
3323 | * are guaranteed to see their updates to the timestamps. | |
3324 | * I believe that this depends on strongly ordered memory | |
3325 | * semantics, but we have that. We use the SYNCHRONIZE | |
3326 | * macro to make sure that the compiler does not reorder | |
3327 | * the i_update_core access below the data copy below. | |
3328 | */ | |
3329 | ip->i_update_core = 0; | |
3330 | SYNCHRONIZE(); | |
3331 | ||
42fe2b1f CH |
3332 | /* |
3333 | * Make sure to get the latest atime from the Linux inode. | |
3334 | */ | |
3335 | xfs_synchronize_atime(ip); | |
3336 | ||
347d1c01 | 3337 | if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC, |
1da177e4 LT |
3338 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
3339 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3340 | "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
347d1c01 | 3341 | ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip); |
1da177e4 LT |
3342 | goto corrupt_out; |
3343 | } | |
3344 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
3345 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
3346 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3347 | "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
3348 | ip->i_ino, ip, ip->i_d.di_magic); | |
3349 | goto corrupt_out; | |
3350 | } | |
3351 | if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
3352 | if (XFS_TEST_ERROR( | |
3353 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3354 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3355 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
3356 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3357 | "xfs_iflush: Bad regular inode %Lu, ptr 0x%p", | |
3358 | ip->i_ino, ip); | |
3359 | goto corrupt_out; | |
3360 | } | |
3361 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { | |
3362 | if (XFS_TEST_ERROR( | |
3363 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3364 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3365 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3366 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
3367 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3368 | "xfs_iflush: Bad directory inode %Lu, ptr 0x%p", | |
3369 | ip->i_ino, ip); | |
3370 | goto corrupt_out; | |
3371 | } | |
3372 | } | |
3373 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3374 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3375 | XFS_RANDOM_IFLUSH_5)) { | |
3376 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3377 | "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3378 | ip->i_ino, | |
3379 | ip->i_d.di_nextents + ip->i_d.di_anextents, | |
3380 | ip->i_d.di_nblocks, | |
3381 | ip); | |
3382 | goto corrupt_out; | |
3383 | } | |
3384 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3385 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
3386 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3387 | "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3388 | ip->i_ino, ip->i_d.di_forkoff, ip); | |
3389 | goto corrupt_out; | |
3390 | } | |
3391 | /* | |
3392 | * bump the flush iteration count, used to detect flushes which | |
3393 | * postdate a log record during recovery. | |
3394 | */ | |
3395 | ||
3396 | ip->i_d.di_flushiter++; | |
3397 | ||
3398 | /* | |
3399 | * Copy the dirty parts of the inode into the on-disk | |
3400 | * inode. We always copy out the core of the inode, | |
3401 | * because if the inode is dirty at all the core must | |
3402 | * be. | |
3403 | */ | |
347d1c01 | 3404 | xfs_dinode_to_disk(&dip->di_core, &ip->i_d); |
1da177e4 LT |
3405 | |
3406 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3407 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3408 | ip->i_d.di_flushiter = 0; | |
3409 | ||
3410 | /* | |
3411 | * If this is really an old format inode and the superblock version | |
3412 | * has not been updated to support only new format inodes, then | |
3413 | * convert back to the old inode format. If the superblock version | |
3414 | * has been updated, then make the conversion permanent. | |
3415 | */ | |
3416 | ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 || | |
62118709 | 3417 | xfs_sb_version_hasnlink(&mp->m_sb)); |
1da177e4 | 3418 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { |
62118709 | 3419 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
3420 | /* |
3421 | * Convert it back. | |
3422 | */ | |
3423 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
347d1c01 | 3424 | dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
3425 | } else { |
3426 | /* | |
3427 | * The superblock version has already been bumped, | |
3428 | * so just make the conversion to the new inode | |
3429 | * format permanent. | |
3430 | */ | |
3431 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
347d1c01 | 3432 | dip->di_core.di_version = XFS_DINODE_VERSION_2; |
1da177e4 LT |
3433 | ip->i_d.di_onlink = 0; |
3434 | dip->di_core.di_onlink = 0; | |
3435 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
3436 | memset(&(dip->di_core.di_pad[0]), 0, | |
3437 | sizeof(dip->di_core.di_pad)); | |
3438 | ASSERT(ip->i_d.di_projid == 0); | |
3439 | } | |
3440 | } | |
3441 | ||
e4ac967b DC |
3442 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
3443 | if (XFS_IFORK_Q(ip)) | |
3444 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
3445 | xfs_inobp_check(mp, bp); |
3446 | ||
3447 | /* | |
3448 | * We've recorded everything logged in the inode, so we'd | |
3449 | * like to clear the ilf_fields bits so we don't log and | |
3450 | * flush things unnecessarily. However, we can't stop | |
3451 | * logging all this information until the data we've copied | |
3452 | * into the disk buffer is written to disk. If we did we might | |
3453 | * overwrite the copy of the inode in the log with all the | |
3454 | * data after re-logging only part of it, and in the face of | |
3455 | * a crash we wouldn't have all the data we need to recover. | |
3456 | * | |
3457 | * What we do is move the bits to the ili_last_fields field. | |
3458 | * When logging the inode, these bits are moved back to the | |
3459 | * ilf_fields field. In the xfs_iflush_done() routine we | |
3460 | * clear ili_last_fields, since we know that the information | |
3461 | * those bits represent is permanently on disk. As long as | |
3462 | * the flush completes before the inode is logged again, then | |
3463 | * both ilf_fields and ili_last_fields will be cleared. | |
3464 | * | |
3465 | * We can play with the ilf_fields bits here, because the inode | |
3466 | * lock must be held exclusively in order to set bits there | |
3467 | * and the flush lock protects the ili_last_fields bits. | |
3468 | * Set ili_logged so the flush done | |
3469 | * routine can tell whether or not to look in the AIL. | |
3470 | * Also, store the current LSN of the inode so that we can tell | |
3471 | * whether the item has moved in the AIL from xfs_iflush_done(). | |
3472 | * In order to read the lsn we need the AIL lock, because | |
3473 | * it is a 64 bit value that cannot be read atomically. | |
3474 | */ | |
3475 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3476 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
3477 | iip->ili_format.ilf_fields = 0; | |
3478 | iip->ili_logged = 1; | |
3479 | ||
7b2e2a31 DC |
3480 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3481 | &iip->ili_item.li_lsn); | |
1da177e4 LT |
3482 | |
3483 | /* | |
3484 | * Attach the function xfs_iflush_done to the inode's | |
3485 | * buffer. This will remove the inode from the AIL | |
3486 | * and unlock the inode's flush lock when the inode is | |
3487 | * completely written to disk. | |
3488 | */ | |
3489 | xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
3490 | xfs_iflush_done, (xfs_log_item_t *)iip); | |
3491 | ||
3492 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
3493 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL); | |
3494 | } else { | |
3495 | /* | |
3496 | * We're flushing an inode which is not in the AIL and has | |
3497 | * not been logged but has i_update_core set. For this | |
3498 | * case we can use a B_DELWRI flush and immediately drop | |
3499 | * the inode flush lock because we can avoid the whole | |
3500 | * AIL state thing. It's OK to drop the flush lock now, | |
3501 | * because we've already locked the buffer and to do anything | |
3502 | * you really need both. | |
3503 | */ | |
3504 | if (iip != NULL) { | |
3505 | ASSERT(iip->ili_logged == 0); | |
3506 | ASSERT(iip->ili_last_fields == 0); | |
3507 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
3508 | } | |
3509 | xfs_ifunlock(ip); | |
3510 | } | |
3511 | ||
3512 | return 0; | |
3513 | ||
3514 | corrupt_out: | |
3515 | return XFS_ERROR(EFSCORRUPTED); | |
3516 | } | |
3517 | ||
3518 | ||
1da177e4 | 3519 | |
1da177e4 LT |
3520 | #ifdef XFS_ILOCK_TRACE |
3521 | ktrace_t *xfs_ilock_trace_buf; | |
3522 | ||
3523 | void | |
3524 | xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra) | |
3525 | { | |
3526 | ktrace_enter(ip->i_lock_trace, | |
3527 | (void *)ip, | |
3528 | (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */ | |
3529 | (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */ | |
3530 | (void *)ra, /* caller of ilock */ | |
3531 | (void *)(unsigned long)current_cpu(), | |
3532 | (void *)(unsigned long)current_pid(), | |
3533 | NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL); | |
3534 | } | |
3535 | #endif | |
4eea22f0 MK |
3536 | |
3537 | /* | |
3538 | * Return a pointer to the extent record at file index idx. | |
3539 | */ | |
a6f64d4a | 3540 | xfs_bmbt_rec_host_t * |
4eea22f0 MK |
3541 | xfs_iext_get_ext( |
3542 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3543 | xfs_extnum_t idx) /* index of target extent */ | |
3544 | { | |
3545 | ASSERT(idx >= 0); | |
0293ce3a MK |
3546 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
3547 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
3548 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
3549 | xfs_ext_irec_t *erp; /* irec pointer */ | |
3550 | int erp_idx = 0; /* irec index */ | |
3551 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
3552 | ||
3553 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3554 | return &erp->er_extbuf[page_idx]; | |
3555 | } else if (ifp->if_bytes) { | |
4eea22f0 MK |
3556 | return &ifp->if_u1.if_extents[idx]; |
3557 | } else { | |
3558 | return NULL; | |
3559 | } | |
3560 | } | |
3561 | ||
3562 | /* | |
3563 | * Insert new item(s) into the extent records for incore inode | |
3564 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
3565 | */ | |
3566 | void | |
3567 | xfs_iext_insert( | |
3568 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3569 | xfs_extnum_t idx, /* starting index of new items */ | |
3570 | xfs_extnum_t count, /* number of inserted items */ | |
3571 | xfs_bmbt_irec_t *new) /* items to insert */ | |
3572 | { | |
4eea22f0 MK |
3573 | xfs_extnum_t i; /* extent record index */ |
3574 | ||
3575 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
3576 | xfs_iext_add(ifp, idx, count); | |
a6f64d4a CH |
3577 | for (i = idx; i < idx + count; i++, new++) |
3578 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); | |
4eea22f0 MK |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * This is called when the amount of space required for incore file | |
3583 | * extents needs to be increased. The ext_diff parameter stores the | |
3584 | * number of new extents being added and the idx parameter contains | |
3585 | * the extent index where the new extents will be added. If the new | |
3586 | * extents are being appended, then we just need to (re)allocate and | |
3587 | * initialize the space. Otherwise, if the new extents are being | |
3588 | * inserted into the middle of the existing entries, a bit more work | |
3589 | * is required to make room for the new extents to be inserted. The | |
3590 | * caller is responsible for filling in the new extent entries upon | |
3591 | * return. | |
3592 | */ | |
3593 | void | |
3594 | xfs_iext_add( | |
3595 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3596 | xfs_extnum_t idx, /* index to begin adding exts */ | |
c41564b5 | 3597 | int ext_diff) /* number of extents to add */ |
4eea22f0 MK |
3598 | { |
3599 | int byte_diff; /* new bytes being added */ | |
3600 | int new_size; /* size of extents after adding */ | |
3601 | xfs_extnum_t nextents; /* number of extents in file */ | |
3602 | ||
3603 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3604 | ASSERT((idx >= 0) && (idx <= nextents)); | |
3605 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
3606 | new_size = ifp->if_bytes + byte_diff; | |
3607 | /* | |
3608 | * If the new number of extents (nextents + ext_diff) | |
3609 | * fits inside the inode, then continue to use the inline | |
3610 | * extent buffer. | |
3611 | */ | |
3612 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
3613 | if (idx < nextents) { | |
3614 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3615 | &ifp->if_u2.if_inline_ext[idx], | |
3616 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3617 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
3618 | } | |
3619 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
3620 | ifp->if_real_bytes = 0; | |
0293ce3a | 3621 | ifp->if_lastex = nextents + ext_diff; |
4eea22f0 MK |
3622 | } |
3623 | /* | |
3624 | * Otherwise use a linear (direct) extent list. | |
3625 | * If the extents are currently inside the inode, | |
3626 | * xfs_iext_realloc_direct will switch us from | |
3627 | * inline to direct extent allocation mode. | |
3628 | */ | |
0293ce3a | 3629 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
4eea22f0 MK |
3630 | xfs_iext_realloc_direct(ifp, new_size); |
3631 | if (idx < nextents) { | |
3632 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
3633 | &ifp->if_u1.if_extents[idx], | |
3634 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3635 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
3636 | } | |
3637 | } | |
0293ce3a MK |
3638 | /* Indirection array */ |
3639 | else { | |
3640 | xfs_ext_irec_t *erp; | |
3641 | int erp_idx = 0; | |
3642 | int page_idx = idx; | |
3643 | ||
3644 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
3645 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3646 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
3647 | } else { | |
3648 | xfs_iext_irec_init(ifp); | |
3649 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3650 | erp = ifp->if_u1.if_ext_irec; | |
3651 | } | |
3652 | /* Extents fit in target extent page */ | |
3653 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
3654 | if (page_idx < erp->er_extcount) { | |
3655 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
3656 | &erp->er_extbuf[page_idx], | |
3657 | (erp->er_extcount - page_idx) * | |
3658 | sizeof(xfs_bmbt_rec_t)); | |
3659 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
3660 | } | |
3661 | erp->er_extcount += ext_diff; | |
3662 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3663 | } | |
3664 | /* Insert a new extent page */ | |
3665 | else if (erp) { | |
3666 | xfs_iext_add_indirect_multi(ifp, | |
3667 | erp_idx, page_idx, ext_diff); | |
3668 | } | |
3669 | /* | |
3670 | * If extent(s) are being appended to the last page in | |
3671 | * the indirection array and the new extent(s) don't fit | |
3672 | * in the page, then erp is NULL and erp_idx is set to | |
3673 | * the next index needed in the indirection array. | |
3674 | */ | |
3675 | else { | |
3676 | int count = ext_diff; | |
3677 | ||
3678 | while (count) { | |
3679 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3680 | erp->er_extcount = count; | |
3681 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
3682 | if (count) { | |
3683 | erp_idx++; | |
3684 | } | |
3685 | } | |
3686 | } | |
3687 | } | |
4eea22f0 MK |
3688 | ifp->if_bytes = new_size; |
3689 | } | |
3690 | ||
0293ce3a MK |
3691 | /* |
3692 | * This is called when incore extents are being added to the indirection | |
3693 | * array and the new extents do not fit in the target extent list. The | |
3694 | * erp_idx parameter contains the irec index for the target extent list | |
3695 | * in the indirection array, and the idx parameter contains the extent | |
3696 | * index within the list. The number of extents being added is stored | |
3697 | * in the count parameter. | |
3698 | * | |
3699 | * |-------| |-------| | |
3700 | * | | | | idx - number of extents before idx | |
3701 | * | idx | | count | | |
3702 | * | | | | count - number of extents being inserted at idx | |
3703 | * |-------| |-------| | |
3704 | * | count | | nex2 | nex2 - number of extents after idx + count | |
3705 | * |-------| |-------| | |
3706 | */ | |
3707 | void | |
3708 | xfs_iext_add_indirect_multi( | |
3709 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3710 | int erp_idx, /* target extent irec index */ | |
3711 | xfs_extnum_t idx, /* index within target list */ | |
3712 | int count) /* new extents being added */ | |
3713 | { | |
3714 | int byte_diff; /* new bytes being added */ | |
3715 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
3716 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
3717 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
3718 | xfs_extnum_t nex2; /* extents after idx + count */ | |
3719 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
3720 | int nlists; /* number of irec's (lists) */ | |
3721 | ||
3722 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3723 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3724 | nex2 = erp->er_extcount - idx; | |
3725 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3726 | ||
3727 | /* | |
3728 | * Save second part of target extent list | |
3729 | * (all extents past */ | |
3730 | if (nex2) { | |
3731 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
6785073b | 3732 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); |
0293ce3a MK |
3733 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); |
3734 | erp->er_extcount -= nex2; | |
3735 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
3736 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
3737 | } | |
3738 | ||
3739 | /* | |
3740 | * Add the new extents to the end of the target | |
3741 | * list, then allocate new irec record(s) and | |
3742 | * extent buffer(s) as needed to store the rest | |
3743 | * of the new extents. | |
3744 | */ | |
3745 | ext_cnt = count; | |
3746 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
3747 | if (ext_diff) { | |
3748 | erp->er_extcount += ext_diff; | |
3749 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3750 | ext_cnt -= ext_diff; | |
3751 | } | |
3752 | while (ext_cnt) { | |
3753 | erp_idx++; | |
3754 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3755 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
3756 | erp->er_extcount = ext_diff; | |
3757 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3758 | ext_cnt -= ext_diff; | |
3759 | } | |
3760 | ||
3761 | /* Add nex2 extents back to indirection array */ | |
3762 | if (nex2) { | |
3763 | xfs_extnum_t ext_avail; | |
3764 | int i; | |
3765 | ||
3766 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3767 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
3768 | i = 0; | |
3769 | /* | |
3770 | * If nex2 extents fit in the current page, append | |
3771 | * nex2_ep after the new extents. | |
3772 | */ | |
3773 | if (nex2 <= ext_avail) { | |
3774 | i = erp->er_extcount; | |
3775 | } | |
3776 | /* | |
3777 | * Otherwise, check if space is available in the | |
3778 | * next page. | |
3779 | */ | |
3780 | else if ((erp_idx < nlists - 1) && | |
3781 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
3782 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
3783 | erp_idx++; | |
3784 | erp++; | |
3785 | /* Create a hole for nex2 extents */ | |
3786 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
3787 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
3788 | } | |
3789 | /* | |
3790 | * Final choice, create a new extent page for | |
3791 | * nex2 extents. | |
3792 | */ | |
3793 | else { | |
3794 | erp_idx++; | |
3795 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3796 | } | |
3797 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
f0e2d93c | 3798 | kmem_free(nex2_ep); |
0293ce3a MK |
3799 | erp->er_extcount += nex2; |
3800 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
3801 | } | |
3802 | } | |
3803 | ||
4eea22f0 MK |
3804 | /* |
3805 | * This is called when the amount of space required for incore file | |
3806 | * extents needs to be decreased. The ext_diff parameter stores the | |
3807 | * number of extents to be removed and the idx parameter contains | |
3808 | * the extent index where the extents will be removed from. | |
0293ce3a MK |
3809 | * |
3810 | * If the amount of space needed has decreased below the linear | |
3811 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
3812 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
3813 | * size to what is needed. | |
4eea22f0 MK |
3814 | */ |
3815 | void | |
3816 | xfs_iext_remove( | |
3817 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3818 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3819 | int ext_diff) /* number of extents to remove */ | |
3820 | { | |
3821 | xfs_extnum_t nextents; /* number of extents in file */ | |
3822 | int new_size; /* size of extents after removal */ | |
3823 | ||
3824 | ASSERT(ext_diff > 0); | |
3825 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3826 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
3827 | ||
3828 | if (new_size == 0) { | |
3829 | xfs_iext_destroy(ifp); | |
0293ce3a MK |
3830 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
3831 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
4eea22f0 MK |
3832 | } else if (ifp->if_real_bytes) { |
3833 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
3834 | } else { | |
3835 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
3836 | } | |
3837 | ifp->if_bytes = new_size; | |
3838 | } | |
3839 | ||
3840 | /* | |
3841 | * This removes ext_diff extents from the inline buffer, beginning | |
3842 | * at extent index idx. | |
3843 | */ | |
3844 | void | |
3845 | xfs_iext_remove_inline( | |
3846 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3847 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3848 | int ext_diff) /* number of extents to remove */ | |
3849 | { | |
3850 | int nextents; /* number of extents in file */ | |
3851 | ||
0293ce3a | 3852 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3853 | ASSERT(idx < XFS_INLINE_EXTS); |
3854 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3855 | ASSERT(((nextents - ext_diff) > 0) && | |
3856 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
3857 | ||
3858 | if (idx + ext_diff < nextents) { | |
3859 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
3860 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3861 | (nextents - (idx + ext_diff)) * | |
3862 | sizeof(xfs_bmbt_rec_t)); | |
3863 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
3864 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3865 | } else { | |
3866 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
3867 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3868 | } | |
3869 | } | |
3870 | ||
3871 | /* | |
3872 | * This removes ext_diff extents from a linear (direct) extent list, | |
3873 | * beginning at extent index idx. If the extents are being removed | |
3874 | * from the end of the list (ie. truncate) then we just need to re- | |
3875 | * allocate the list to remove the extra space. Otherwise, if the | |
3876 | * extents are being removed from the middle of the existing extent | |
3877 | * entries, then we first need to move the extent records beginning | |
3878 | * at idx + ext_diff up in the list to overwrite the records being | |
3879 | * removed, then remove the extra space via kmem_realloc. | |
3880 | */ | |
3881 | void | |
3882 | xfs_iext_remove_direct( | |
3883 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3884 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3885 | int ext_diff) /* number of extents to remove */ | |
3886 | { | |
3887 | xfs_extnum_t nextents; /* number of extents in file */ | |
3888 | int new_size; /* size of extents after removal */ | |
3889 | ||
0293ce3a | 3890 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3891 | new_size = ifp->if_bytes - |
3892 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3893 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3894 | ||
3895 | if (new_size == 0) { | |
3896 | xfs_iext_destroy(ifp); | |
3897 | return; | |
3898 | } | |
3899 | /* Move extents up in the list (if needed) */ | |
3900 | if (idx + ext_diff < nextents) { | |
3901 | memmove(&ifp->if_u1.if_extents[idx], | |
3902 | &ifp->if_u1.if_extents[idx + ext_diff], | |
3903 | (nextents - (idx + ext_diff)) * | |
3904 | sizeof(xfs_bmbt_rec_t)); | |
3905 | } | |
3906 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
3907 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3908 | /* | |
3909 | * Reallocate the direct extent list. If the extents | |
3910 | * will fit inside the inode then xfs_iext_realloc_direct | |
3911 | * will switch from direct to inline extent allocation | |
3912 | * mode for us. | |
3913 | */ | |
3914 | xfs_iext_realloc_direct(ifp, new_size); | |
3915 | ifp->if_bytes = new_size; | |
3916 | } | |
3917 | ||
0293ce3a MK |
3918 | /* |
3919 | * This is called when incore extents are being removed from the | |
3920 | * indirection array and the extents being removed span multiple extent | |
3921 | * buffers. The idx parameter contains the file extent index where we | |
3922 | * want to begin removing extents, and the count parameter contains | |
3923 | * how many extents need to be removed. | |
3924 | * | |
3925 | * |-------| |-------| | |
3926 | * | nex1 | | | nex1 - number of extents before idx | |
3927 | * |-------| | count | | |
3928 | * | | | | count - number of extents being removed at idx | |
3929 | * | count | |-------| | |
3930 | * | | | nex2 | nex2 - number of extents after idx + count | |
3931 | * |-------| |-------| | |
3932 | */ | |
3933 | void | |
3934 | xfs_iext_remove_indirect( | |
3935 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3936 | xfs_extnum_t idx, /* index to begin removing extents */ | |
3937 | int count) /* number of extents to remove */ | |
3938 | { | |
3939 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3940 | int erp_idx = 0; /* indirection array index */ | |
3941 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
3942 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
3943 | xfs_extnum_t nex1; /* number of extents before idx */ | |
3944 | xfs_extnum_t nex2; /* extents after idx + count */ | |
c41564b5 | 3945 | int nlists; /* entries in indirection array */ |
0293ce3a MK |
3946 | int page_idx = idx; /* index in target extent list */ |
3947 | ||
3948 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3949 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3950 | ASSERT(erp != NULL); | |
3951 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3952 | nex1 = page_idx; | |
3953 | ext_cnt = count; | |
3954 | while (ext_cnt) { | |
3955 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
3956 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
3957 | /* | |
3958 | * Check for deletion of entire list; | |
3959 | * xfs_iext_irec_remove() updates extent offsets. | |
3960 | */ | |
3961 | if (ext_diff == erp->er_extcount) { | |
3962 | xfs_iext_irec_remove(ifp, erp_idx); | |
3963 | ext_cnt -= ext_diff; | |
3964 | nex1 = 0; | |
3965 | if (ext_cnt) { | |
3966 | ASSERT(erp_idx < ifp->if_real_bytes / | |
3967 | XFS_IEXT_BUFSZ); | |
3968 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3969 | nex1 = 0; | |
3970 | continue; | |
3971 | } else { | |
3972 | break; | |
3973 | } | |
3974 | } | |
3975 | /* Move extents up (if needed) */ | |
3976 | if (nex2) { | |
3977 | memmove(&erp->er_extbuf[nex1], | |
3978 | &erp->er_extbuf[nex1 + ext_diff], | |
3979 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
3980 | } | |
3981 | /* Zero out rest of page */ | |
3982 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
3983 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
3984 | /* Update remaining counters */ | |
3985 | erp->er_extcount -= ext_diff; | |
3986 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
3987 | ext_cnt -= ext_diff; | |
3988 | nex1 = 0; | |
3989 | erp_idx++; | |
3990 | erp++; | |
3991 | } | |
3992 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
3993 | xfs_iext_irec_compact(ifp); | |
3994 | } | |
3995 | ||
4eea22f0 MK |
3996 | /* |
3997 | * Create, destroy, or resize a linear (direct) block of extents. | |
3998 | */ | |
3999 | void | |
4000 | xfs_iext_realloc_direct( | |
4001 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4002 | int new_size) /* new size of extents */ | |
4003 | { | |
4004 | int rnew_size; /* real new size of extents */ | |
4005 | ||
4006 | rnew_size = new_size; | |
4007 | ||
0293ce3a MK |
4008 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
4009 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
4010 | (new_size != ifp->if_real_bytes))); | |
4011 | ||
4eea22f0 MK |
4012 | /* Free extent records */ |
4013 | if (new_size == 0) { | |
4014 | xfs_iext_destroy(ifp); | |
4015 | } | |
4016 | /* Resize direct extent list and zero any new bytes */ | |
4017 | else if (ifp->if_real_bytes) { | |
4018 | /* Check if extents will fit inside the inode */ | |
4019 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
4020 | xfs_iext_direct_to_inline(ifp, new_size / | |
4021 | (uint)sizeof(xfs_bmbt_rec_t)); | |
4022 | ifp->if_bytes = new_size; | |
4023 | return; | |
4024 | } | |
16a087d8 | 4025 | if (!is_power_of_2(new_size)){ |
40ebd81d | 4026 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
4027 | } |
4028 | if (rnew_size != ifp->if_real_bytes) { | |
a6f64d4a | 4029 | ifp->if_u1.if_extents = |
4eea22f0 MK |
4030 | kmem_realloc(ifp->if_u1.if_extents, |
4031 | rnew_size, | |
6785073b | 4032 | ifp->if_real_bytes, KM_NOFS); |
4eea22f0 MK |
4033 | } |
4034 | if (rnew_size > ifp->if_real_bytes) { | |
4035 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
4036 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
4037 | rnew_size - ifp->if_real_bytes); | |
4038 | } | |
4039 | } | |
4040 | /* | |
4041 | * Switch from the inline extent buffer to a direct | |
4042 | * extent list. Be sure to include the inline extent | |
4043 | * bytes in new_size. | |
4044 | */ | |
4045 | else { | |
4046 | new_size += ifp->if_bytes; | |
16a087d8 | 4047 | if (!is_power_of_2(new_size)) { |
40ebd81d | 4048 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
4049 | } |
4050 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
4051 | } | |
4052 | ifp->if_real_bytes = rnew_size; | |
4053 | ifp->if_bytes = new_size; | |
4054 | } | |
4055 | ||
4056 | /* | |
4057 | * Switch from linear (direct) extent records to inline buffer. | |
4058 | */ | |
4059 | void | |
4060 | xfs_iext_direct_to_inline( | |
4061 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4062 | xfs_extnum_t nextents) /* number of extents in file */ | |
4063 | { | |
4064 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
4065 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
4066 | /* | |
4067 | * The inline buffer was zeroed when we switched | |
4068 | * from inline to direct extent allocation mode, | |
4069 | * so we don't need to clear it here. | |
4070 | */ | |
4071 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
4072 | nextents * sizeof(xfs_bmbt_rec_t)); | |
f0e2d93c | 4073 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
4074 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
4075 | ifp->if_real_bytes = 0; | |
4076 | } | |
4077 | ||
4078 | /* | |
4079 | * Switch from inline buffer to linear (direct) extent records. | |
4080 | * new_size should already be rounded up to the next power of 2 | |
4081 | * by the caller (when appropriate), so use new_size as it is. | |
4082 | * However, since new_size may be rounded up, we can't update | |
4083 | * if_bytes here. It is the caller's responsibility to update | |
4084 | * if_bytes upon return. | |
4085 | */ | |
4086 | void | |
4087 | xfs_iext_inline_to_direct( | |
4088 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4089 | int new_size) /* number of extents in file */ | |
4090 | { | |
6785073b | 4091 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); |
4eea22f0 MK |
4092 | memset(ifp->if_u1.if_extents, 0, new_size); |
4093 | if (ifp->if_bytes) { | |
4094 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
4095 | ifp->if_bytes); | |
4096 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4097 | sizeof(xfs_bmbt_rec_t)); | |
4098 | } | |
4099 | ifp->if_real_bytes = new_size; | |
4100 | } | |
4101 | ||
0293ce3a MK |
4102 | /* |
4103 | * Resize an extent indirection array to new_size bytes. | |
4104 | */ | |
4105 | void | |
4106 | xfs_iext_realloc_indirect( | |
4107 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4108 | int new_size) /* new indirection array size */ | |
4109 | { | |
4110 | int nlists; /* number of irec's (ex lists) */ | |
4111 | int size; /* current indirection array size */ | |
4112 | ||
4113 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4114 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4115 | size = nlists * sizeof(xfs_ext_irec_t); | |
4116 | ASSERT(ifp->if_real_bytes); | |
4117 | ASSERT((new_size >= 0) && (new_size != size)); | |
4118 | if (new_size == 0) { | |
4119 | xfs_iext_destroy(ifp); | |
4120 | } else { | |
4121 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
4122 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
6785073b | 4123 | new_size, size, KM_NOFS); |
0293ce3a MK |
4124 | } |
4125 | } | |
4126 | ||
4127 | /* | |
4128 | * Switch from indirection array to linear (direct) extent allocations. | |
4129 | */ | |
4130 | void | |
4131 | xfs_iext_indirect_to_direct( | |
4132 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4133 | { | |
a6f64d4a | 4134 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ |
0293ce3a MK |
4135 | xfs_extnum_t nextents; /* number of extents in file */ |
4136 | int size; /* size of file extents */ | |
4137 | ||
4138 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4139 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4140 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4141 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
4142 | ||
71a8c87f | 4143 | xfs_iext_irec_compact_pages(ifp); |
0293ce3a MK |
4144 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); |
4145 | ||
4146 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
f0e2d93c | 4147 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
4148 | ifp->if_flags &= ~XFS_IFEXTIREC; |
4149 | ifp->if_u1.if_extents = ep; | |
4150 | ifp->if_bytes = size; | |
4151 | if (nextents < XFS_LINEAR_EXTS) { | |
4152 | xfs_iext_realloc_direct(ifp, size); | |
4153 | } | |
4154 | } | |
4155 | ||
4eea22f0 MK |
4156 | /* |
4157 | * Free incore file extents. | |
4158 | */ | |
4159 | void | |
4160 | xfs_iext_destroy( | |
4161 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4162 | { | |
0293ce3a MK |
4163 | if (ifp->if_flags & XFS_IFEXTIREC) { |
4164 | int erp_idx; | |
4165 | int nlists; | |
4166 | ||
4167 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4168 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
4169 | xfs_iext_irec_remove(ifp, erp_idx); | |
4170 | } | |
4171 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
4172 | } else if (ifp->if_real_bytes) { | |
f0e2d93c | 4173 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
4174 | } else if (ifp->if_bytes) { |
4175 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4176 | sizeof(xfs_bmbt_rec_t)); | |
4177 | } | |
4178 | ifp->if_u1.if_extents = NULL; | |
4179 | ifp->if_real_bytes = 0; | |
4180 | ifp->if_bytes = 0; | |
4181 | } | |
0293ce3a | 4182 | |
8867bc9b MK |
4183 | /* |
4184 | * Return a pointer to the extent record for file system block bno. | |
4185 | */ | |
a6f64d4a | 4186 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ |
8867bc9b MK |
4187 | xfs_iext_bno_to_ext( |
4188 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4189 | xfs_fileoff_t bno, /* block number to search for */ | |
4190 | xfs_extnum_t *idxp) /* index of target extent */ | |
4191 | { | |
a6f64d4a | 4192 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ |
8867bc9b | 4193 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ |
a6f64d4a | 4194 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ |
8867bc9b | 4195 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
c41564b5 | 4196 | int high; /* upper boundary in search */ |
8867bc9b | 4197 | xfs_extnum_t idx = 0; /* index of target extent */ |
c41564b5 | 4198 | int low; /* lower boundary in search */ |
8867bc9b MK |
4199 | xfs_extnum_t nextents; /* number of file extents */ |
4200 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
4201 | ||
4202 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4203 | if (nextents == 0) { | |
4204 | *idxp = 0; | |
4205 | return NULL; | |
4206 | } | |
4207 | low = 0; | |
4208 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4209 | /* Find target extent list */ | |
4210 | int erp_idx = 0; | |
4211 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
4212 | base = erp->er_extbuf; | |
4213 | high = erp->er_extcount - 1; | |
4214 | } else { | |
4215 | base = ifp->if_u1.if_extents; | |
4216 | high = nextents - 1; | |
4217 | } | |
4218 | /* Binary search extent records */ | |
4219 | while (low <= high) { | |
4220 | idx = (low + high) >> 1; | |
4221 | ep = base + idx; | |
4222 | startoff = xfs_bmbt_get_startoff(ep); | |
4223 | blockcount = xfs_bmbt_get_blockcount(ep); | |
4224 | if (bno < startoff) { | |
4225 | high = idx - 1; | |
4226 | } else if (bno >= startoff + blockcount) { | |
4227 | low = idx + 1; | |
4228 | } else { | |
4229 | /* Convert back to file-based extent index */ | |
4230 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4231 | idx += erp->er_extoff; | |
4232 | } | |
4233 | *idxp = idx; | |
4234 | return ep; | |
4235 | } | |
4236 | } | |
4237 | /* Convert back to file-based extent index */ | |
4238 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4239 | idx += erp->er_extoff; | |
4240 | } | |
4241 | if (bno >= startoff + blockcount) { | |
4242 | if (++idx == nextents) { | |
4243 | ep = NULL; | |
4244 | } else { | |
4245 | ep = xfs_iext_get_ext(ifp, idx); | |
4246 | } | |
4247 | } | |
4248 | *idxp = idx; | |
4249 | return ep; | |
4250 | } | |
4251 | ||
0293ce3a MK |
4252 | /* |
4253 | * Return a pointer to the indirection array entry containing the | |
4254 | * extent record for filesystem block bno. Store the index of the | |
4255 | * target irec in *erp_idxp. | |
4256 | */ | |
8867bc9b | 4257 | xfs_ext_irec_t * /* pointer to found extent record */ |
0293ce3a MK |
4258 | xfs_iext_bno_to_irec( |
4259 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4260 | xfs_fileoff_t bno, /* block number to search for */ | |
4261 | int *erp_idxp) /* irec index of target ext list */ | |
4262 | { | |
4263 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
4264 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
8867bc9b | 4265 | int erp_idx; /* indirection array index */ |
0293ce3a MK |
4266 | int nlists; /* number of extent irec's (lists) */ |
4267 | int high; /* binary search upper limit */ | |
4268 | int low; /* binary search lower limit */ | |
4269 | ||
4270 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4271 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4272 | erp_idx = 0; | |
4273 | low = 0; | |
4274 | high = nlists - 1; | |
4275 | while (low <= high) { | |
4276 | erp_idx = (low + high) >> 1; | |
4277 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4278 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
4279 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
4280 | high = erp_idx - 1; | |
4281 | } else if (erp_next && bno >= | |
4282 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
4283 | low = erp_idx + 1; | |
4284 | } else { | |
4285 | break; | |
4286 | } | |
4287 | } | |
4288 | *erp_idxp = erp_idx; | |
4289 | return erp; | |
4290 | } | |
4291 | ||
4292 | /* | |
4293 | * Return a pointer to the indirection array entry containing the | |
4294 | * extent record at file extent index *idxp. Store the index of the | |
4295 | * target irec in *erp_idxp and store the page index of the target | |
4296 | * extent record in *idxp. | |
4297 | */ | |
4298 | xfs_ext_irec_t * | |
4299 | xfs_iext_idx_to_irec( | |
4300 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4301 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
4302 | int *erp_idxp, /* pointer to target irec */ | |
4303 | int realloc) /* new bytes were just added */ | |
4304 | { | |
4305 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
4306 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
4307 | int erp_idx; /* indirection array index */ | |
4308 | int nlists; /* number of irec's (ex lists) */ | |
4309 | int high; /* binary search upper limit */ | |
4310 | int low; /* binary search lower limit */ | |
4311 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
4312 | ||
4313 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4314 | ASSERT(page_idx >= 0 && page_idx <= | |
4315 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)); | |
4316 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4317 | erp_idx = 0; | |
4318 | low = 0; | |
4319 | high = nlists - 1; | |
4320 | ||
4321 | /* Binary search extent irec's */ | |
4322 | while (low <= high) { | |
4323 | erp_idx = (low + high) >> 1; | |
4324 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4325 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
4326 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
4327 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
4328 | high = erp_idx - 1; | |
4329 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
4330 | (page_idx == erp->er_extoff + erp->er_extcount && | |
4331 | !realloc)) { | |
4332 | low = erp_idx + 1; | |
4333 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
4334 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
4335 | ASSERT(realloc); | |
4336 | page_idx = 0; | |
4337 | erp_idx++; | |
4338 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
4339 | break; | |
4340 | } else { | |
4341 | page_idx -= erp->er_extoff; | |
4342 | break; | |
4343 | } | |
4344 | } | |
4345 | *idxp = page_idx; | |
4346 | *erp_idxp = erp_idx; | |
4347 | return(erp); | |
4348 | } | |
4349 | ||
4350 | /* | |
4351 | * Allocate and initialize an indirection array once the space needed | |
4352 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
4353 | */ | |
4354 | void | |
4355 | xfs_iext_irec_init( | |
4356 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4357 | { | |
4358 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4359 | xfs_extnum_t nextents; /* number of extents in file */ | |
4360 | ||
4361 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
4362 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4363 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4364 | ||
6785073b | 4365 | erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); |
0293ce3a MK |
4366 | |
4367 | if (nextents == 0) { | |
6785073b | 4368 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
4369 | } else if (!ifp->if_real_bytes) { |
4370 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
4371 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
4372 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
4373 | } | |
4374 | erp->er_extbuf = ifp->if_u1.if_extents; | |
4375 | erp->er_extcount = nextents; | |
4376 | erp->er_extoff = 0; | |
4377 | ||
4378 | ifp->if_flags |= XFS_IFEXTIREC; | |
4379 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
4380 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
4381 | ifp->if_u1.if_ext_irec = erp; | |
4382 | ||
4383 | return; | |
4384 | } | |
4385 | ||
4386 | /* | |
4387 | * Allocate and initialize a new entry in the indirection array. | |
4388 | */ | |
4389 | xfs_ext_irec_t * | |
4390 | xfs_iext_irec_new( | |
4391 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4392 | int erp_idx) /* index for new irec */ | |
4393 | { | |
4394 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4395 | int i; /* loop counter */ | |
4396 | int nlists; /* number of irec's (ex lists) */ | |
4397 | ||
4398 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4399 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4400 | ||
4401 | /* Resize indirection array */ | |
4402 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
4403 | sizeof(xfs_ext_irec_t)); | |
4404 | /* | |
4405 | * Move records down in the array so the | |
4406 | * new page can use erp_idx. | |
4407 | */ | |
4408 | erp = ifp->if_u1.if_ext_irec; | |
4409 | for (i = nlists - 1; i > erp_idx; i--) { | |
4410 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
4411 | } | |
4412 | ASSERT(i == erp_idx); | |
4413 | ||
4414 | /* Initialize new extent record */ | |
4415 | erp = ifp->if_u1.if_ext_irec; | |
6785073b | 4416 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
4417 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
4418 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
4419 | erp[erp_idx].er_extcount = 0; | |
4420 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
4421 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
4422 | return (&erp[erp_idx]); | |
4423 | } | |
4424 | ||
4425 | /* | |
4426 | * Remove a record from the indirection array. | |
4427 | */ | |
4428 | void | |
4429 | xfs_iext_irec_remove( | |
4430 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4431 | int erp_idx) /* irec index to remove */ | |
4432 | { | |
4433 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4434 | int i; /* loop counter */ | |
4435 | int nlists; /* number of irec's (ex lists) */ | |
4436 | ||
4437 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4438 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4439 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4440 | if (erp->er_extbuf) { | |
4441 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
4442 | -erp->er_extcount); | |
f0e2d93c | 4443 | kmem_free(erp->er_extbuf); |
0293ce3a MK |
4444 | } |
4445 | /* Compact extent records */ | |
4446 | erp = ifp->if_u1.if_ext_irec; | |
4447 | for (i = erp_idx; i < nlists - 1; i++) { | |
4448 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
4449 | } | |
4450 | /* | |
4451 | * Manually free the last extent record from the indirection | |
4452 | * array. A call to xfs_iext_realloc_indirect() with a size | |
4453 | * of zero would result in a call to xfs_iext_destroy() which | |
4454 | * would in turn call this function again, creating a nasty | |
4455 | * infinite loop. | |
4456 | */ | |
4457 | if (--nlists) { | |
4458 | xfs_iext_realloc_indirect(ifp, | |
4459 | nlists * sizeof(xfs_ext_irec_t)); | |
4460 | } else { | |
f0e2d93c | 4461 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
4462 | } |
4463 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
4464 | } | |
4465 | ||
4466 | /* | |
4467 | * This is called to clean up large amounts of unused memory allocated | |
4468 | * by the indirection array. Before compacting anything though, verify | |
4469 | * that the indirection array is still needed and switch back to the | |
4470 | * linear extent list (or even the inline buffer) if possible. The | |
4471 | * compaction policy is as follows: | |
4472 | * | |
4473 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
71a8c87f | 4474 | * Partial Compaction: Extents occupy less than 50% of allocated space |
0293ce3a MK |
4475 | * No Compaction: Extents occupy at least 50% of allocated space |
4476 | */ | |
4477 | void | |
4478 | xfs_iext_irec_compact( | |
4479 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4480 | { | |
4481 | xfs_extnum_t nextents; /* number of extents in file */ | |
4482 | int nlists; /* number of irec's (ex lists) */ | |
4483 | ||
4484 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4485 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4486 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4487 | ||
4488 | if (nextents == 0) { | |
4489 | xfs_iext_destroy(ifp); | |
4490 | } else if (nextents <= XFS_INLINE_EXTS) { | |
4491 | xfs_iext_indirect_to_direct(ifp); | |
4492 | xfs_iext_direct_to_inline(ifp, nextents); | |
4493 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
4494 | xfs_iext_indirect_to_direct(ifp); | |
0293ce3a MK |
4495 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { |
4496 | xfs_iext_irec_compact_pages(ifp); | |
4497 | } | |
4498 | } | |
4499 | ||
4500 | /* | |
4501 | * Combine extents from neighboring extent pages. | |
4502 | */ | |
4503 | void | |
4504 | xfs_iext_irec_compact_pages( | |
4505 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4506 | { | |
4507 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
4508 | int erp_idx = 0; /* indirection array index */ | |
4509 | int nlists; /* number of irec's (ex lists) */ | |
4510 | ||
4511 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4512 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4513 | while (erp_idx < nlists - 1) { | |
4514 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4515 | erp_next = erp + 1; | |
4516 | if (erp_next->er_extcount <= | |
4517 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
71a8c87f | 4518 | memcpy(&erp->er_extbuf[erp->er_extcount], |
0293ce3a MK |
4519 | erp_next->er_extbuf, erp_next->er_extcount * |
4520 | sizeof(xfs_bmbt_rec_t)); | |
4521 | erp->er_extcount += erp_next->er_extcount; | |
4522 | /* | |
4523 | * Free page before removing extent record | |
4524 | * so er_extoffs don't get modified in | |
4525 | * xfs_iext_irec_remove. | |
4526 | */ | |
f0e2d93c | 4527 | kmem_free(erp_next->er_extbuf); |
0293ce3a MK |
4528 | erp_next->er_extbuf = NULL; |
4529 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
4530 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4531 | } else { | |
4532 | erp_idx++; | |
4533 | } | |
4534 | } | |
4535 | } | |
4536 | ||
0293ce3a MK |
4537 | /* |
4538 | * This is called to update the er_extoff field in the indirection | |
4539 | * array when extents have been added or removed from one of the | |
4540 | * extent lists. erp_idx contains the irec index to begin updating | |
4541 | * at and ext_diff contains the number of extents that were added | |
4542 | * or removed. | |
4543 | */ | |
4544 | void | |
4545 | xfs_iext_irec_update_extoffs( | |
4546 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4547 | int erp_idx, /* irec index to update */ | |
4548 | int ext_diff) /* number of new extents */ | |
4549 | { | |
4550 | int i; /* loop counter */ | |
4551 | int nlists; /* number of irec's (ex lists */ | |
4552 | ||
4553 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4554 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4555 | for (i = erp_idx; i < nlists; i++) { | |
4556 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
4557 | } | |
4558 | } |