]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
5e656dbb | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
da23017d | 3 | * All Rights Reserved. |
5000d01d | 4 | * |
da23017d 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 | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d 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. | |
5000d01d | 13 | * |
da23017d 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 | |
2bd0ea18 NS |
17 | */ |
18 | ||
19 | #include <xfs.h> | |
20 | ||
5e656dbb BN |
21 | kmem_zone_t *xfs_ifork_zone; |
22 | kmem_zone_t *xfs_inode_zone; | |
23 | ||
24 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
25 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
26 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
2bd0ea18 NS |
27 | |
28 | #ifdef DEBUG | |
5e656dbb BN |
29 | /* |
30 | * Make sure that the extents in the given memory buffer | |
31 | * are valid. | |
32 | */ | |
33 | STATIC void | |
34 | xfs_validate_extents( | |
35 | xfs_ifork_t *ifp, | |
36 | int nrecs, | |
37 | xfs_exntfmt_t fmt) | |
38 | { | |
39 | xfs_bmbt_irec_t irec; | |
40 | xfs_bmbt_rec_host_t rec; | |
41 | int i; | |
42 | ||
43 | for (i = 0; i < nrecs; i++) { | |
44 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); | |
45 | rec.l0 = get_unaligned(&ep->l0); | |
46 | rec.l1 = get_unaligned(&ep->l1); | |
47 | xfs_bmbt_get_all(&rec, &irec); | |
48 | if (fmt == XFS_EXTFMT_NOSTATE) | |
49 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
50 | } | |
51 | } | |
52 | #else /* DEBUG */ | |
53 | #define xfs_validate_extents(ifp, nrecs, fmt) | |
54 | #endif /* DEBUG */ | |
55 | ||
56 | /* | |
57 | * Check that none of the inode's in the buffer have a next | |
58 | * unlinked field of 0. | |
59 | */ | |
60 | #if defined(DEBUG) | |
2bd0ea18 NS |
61 | void |
62 | xfs_inobp_check( | |
63 | xfs_mount_t *mp, | |
64 | xfs_buf_t *bp) | |
65 | { | |
66 | int i; | |
67 | int j; | |
68 | xfs_dinode_t *dip; | |
69 | ||
70 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
71 | ||
72 | for (i = 0; i < j; i++) { | |
73 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
74 | i * mp->m_sb.sb_inodesize); | |
46eca962 | 75 | if (!dip->di_next_unlinked) { |
2bd0ea18 | 76 | xfs_fs_cmn_err(CE_ALERT, mp, |
dfc130f3 | 77 | "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", |
2bd0ea18 | 78 | bp); |
46eca962 | 79 | ASSERT(dip->di_next_unlinked); |
2bd0ea18 NS |
80 | } |
81 | } | |
82 | } | |
83 | #endif | |
84 | ||
5e656dbb BN |
85 | /* |
86 | * Find the buffer associated with the given inode map | |
87 | * We do basic validation checks on the buffer once it has been | |
88 | * retrieved from disk. | |
89 | */ | |
90 | STATIC int | |
91 | xfs_imap_to_bp( | |
92 | xfs_mount_t *mp, | |
93 | xfs_trans_t *tp, | |
94 | xfs_imap_t *imap, | |
95 | xfs_buf_t **bpp, | |
96 | uint buf_flags, | |
97 | uint imap_flags) | |
98 | { | |
99 | int error; | |
100 | int i; | |
101 | int ni; | |
102 | xfs_buf_t *bp; | |
103 | ||
104 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, | |
105 | (int)imap->im_len, buf_flags, &bp); | |
106 | if (error) { | |
107 | if (error != EAGAIN) { | |
108 | cmn_err(CE_WARN, | |
109 | "xfs_imap_to_bp: xfs_trans_read_buf()returned " | |
110 | "an error %d on %s. Returning error.", | |
111 | error, mp->m_fsname); | |
112 | } else { | |
113 | ASSERT(buf_flags & XFS_BUF_TRYLOCK); | |
114 | } | |
115 | return error; | |
116 | } | |
117 | ||
118 | /* | |
119 | * Validate the magic number and version of every inode in the buffer | |
120 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
121 | */ | |
122 | #ifdef DEBUG | |
123 | ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog; | |
124 | #else /* usual case */ | |
125 | ni = 1; | |
126 | #endif | |
127 | ||
128 | for (i = 0; i < ni; i++) { | |
129 | int di_ok; | |
130 | xfs_dinode_t *dip; | |
131 | ||
132 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
133 | (i << mp->m_sb.sb_inodelog)); | |
134 | di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC && | |
135 | XFS_DINODE_GOOD_VERSION(dip->di_core.di_version); | |
136 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, | |
137 | XFS_ERRTAG_ITOBP_INOTOBP, | |
138 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
139 | if (imap_flags & XFS_IMAP_BULKSTAT) { | |
140 | xfs_trans_brelse(tp, bp); | |
141 | return XFS_ERROR(EINVAL); | |
142 | } | |
143 | XFS_CORRUPTION_ERROR("xfs_imap_to_bp", | |
144 | XFS_ERRLEVEL_HIGH, mp, dip); | |
145 | #ifdef DEBUG | |
146 | cmn_err(CE_PANIC, | |
147 | "Device %s - bad inode magic/vsn " | |
148 | "daddr %lld #%d (magic=%x)", | |
149 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
150 | (unsigned long long)imap->im_blkno, i, | |
151 | be16_to_cpu(dip->di_core.di_magic)); | |
152 | #endif | |
153 | xfs_trans_brelse(tp, bp); | |
154 | return XFS_ERROR(EFSCORRUPTED); | |
155 | } | |
156 | } | |
157 | ||
158 | xfs_inobp_check(mp, bp); | |
159 | ||
160 | /* | |
161 | * Mark the buffer as an inode buffer now that it looks good | |
162 | */ | |
163 | XFS_BUF_SET_VTYPE(bp, B_FS_INO); | |
164 | ||
165 | *bpp = bp; | |
166 | return 0; | |
167 | } | |
2bd0ea18 NS |
168 | |
169 | /* | |
170 | * This routine is called to map an inode to the buffer containing | |
171 | * the on-disk version of the inode. It returns a pointer to the | |
172 | * buffer containing the on-disk inode in the bpp parameter, and in | |
173 | * the dip parameter it returns a pointer to the on-disk inode within | |
174 | * that buffer. | |
175 | * | |
176 | * If a non-zero error is returned, then the contents of bpp and | |
177 | * dipp are undefined. | |
178 | * | |
179 | * If the inode is new and has not yet been initialized, use xfs_imap() | |
180 | * to determine the size and location of the buffer to read from disk. | |
181 | * If the inode has already been mapped to its buffer and read in once, | |
182 | * then use the mapping information stored in the inode rather than | |
dfc130f3 | 183 | * calling xfs_imap(). This allows us to avoid the overhead of looking |
2bd0ea18 NS |
184 | * at the inode btree for small block file systems (see xfs_dilocate()). |
185 | * We can tell whether the inode has been mapped in before by comparing | |
dfc130f3 | 186 | * its disk block address to 0. Only uninitialized inodes will have |
2bd0ea18 NS |
187 | * 0 for the disk block address. |
188 | */ | |
189 | int | |
190 | xfs_itobp( | |
191 | xfs_mount_t *mp, | |
192 | xfs_trans_t *tp, | |
5000d01d | 193 | xfs_inode_t *ip, |
2bd0ea18 NS |
194 | xfs_dinode_t **dipp, |
195 | xfs_buf_t **bpp, | |
5e656dbb BN |
196 | xfs_daddr_t bno, |
197 | uint imap_flags, | |
198 | uint buf_flags) | |
2bd0ea18 | 199 | { |
5e656dbb | 200 | xfs_imap_t imap; |
2bd0ea18 NS |
201 | xfs_buf_t *bp; |
202 | int error; | |
2bd0ea18 NS |
203 | |
204 | if (ip->i_blkno == (xfs_daddr_t)0) { | |
2bd0ea18 | 205 | imap.im_blkno = bno; |
5e656dbb BN |
206 | error = xfs_imap(mp, tp, ip->i_ino, &imap, |
207 | XFS_IMAP_LOOKUP | imap_flags); | |
208 | if (error) | |
2bd0ea18 | 209 | return error; |
2bd0ea18 NS |
210 | |
211 | /* | |
212 | * Fill in the fields in the inode that will be used to | |
213 | * map the inode to its buffer from now on. | |
214 | */ | |
215 | ip->i_blkno = imap.im_blkno; | |
216 | ip->i_len = imap.im_len; | |
217 | ip->i_boffset = imap.im_boffset; | |
218 | } else { | |
219 | /* | |
220 | * We've already mapped the inode once, so just use the | |
221 | * mapping that we saved the first time. | |
222 | */ | |
223 | imap.im_blkno = ip->i_blkno; | |
224 | imap.im_len = ip->i_len; | |
225 | imap.im_boffset = ip->i_boffset; | |
226 | } | |
227 | ASSERT(bno == 0 || bno == imap.im_blkno); | |
228 | ||
5e656dbb BN |
229 | error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags); |
230 | if (error) | |
2bd0ea18 | 231 | return error; |
2bd0ea18 | 232 | |
5e656dbb BN |
233 | if (!bp) { |
234 | ASSERT(buf_flags & XFS_BUF_TRYLOCK); | |
235 | ASSERT(tp == NULL); | |
236 | *bpp = NULL; | |
237 | return EAGAIN; | |
2bd0ea18 | 238 | } |
2bd0ea18 | 239 | |
2bd0ea18 NS |
240 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); |
241 | *bpp = bp; | |
242 | return 0; | |
243 | } | |
244 | ||
245 | /* | |
246 | * Move inode type and inode format specific information from the | |
dfc130f3 RC |
247 | * on-disk inode to the in-core inode. For fifos, devs, and sockets |
248 | * this means set if_rdev to the proper value. For files, directories, | |
2bd0ea18 NS |
249 | * and symlinks this means to bring in the in-line data or extent |
250 | * pointers. For a file in B-tree format, only the root is immediately | |
251 | * brought in-core. The rest will be in-lined in if_extents when it | |
252 | * is first referenced (see xfs_iread_extents()). | |
5e656dbb BN |
253 | * |
254 | * Note: this requires user-space public scope for libxfs_iread | |
2bd0ea18 | 255 | */ |
5e656dbb | 256 | int |
2bd0ea18 NS |
257 | xfs_iformat( |
258 | xfs_inode_t *ip, | |
2f8883de | 259 | xfs_dinode_t *dip) |
2bd0ea18 NS |
260 | { |
261 | xfs_attr_shortform_t *atp; | |
262 | int size; | |
263 | int error; | |
dfc130f3 | 264 | xfs_fsize_t di_size; |
2bd0ea18 NS |
265 | ip->i_df.if_ext_max = |
266 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
267 | error = 0; | |
268 | ||
5e656dbb BN |
269 | if (unlikely(be32_to_cpu(dip->di_core.di_nextents) + |
270 | be16_to_cpu(dip->di_core.di_anextents) > | |
271 | be64_to_cpu(dip->di_core.di_nblocks))) { | |
23297264 NS |
272 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
273 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", | |
31c5308f | 274 | (unsigned long long)ip->i_ino, |
5e656dbb BN |
275 | (int)(be32_to_cpu(dip->di_core.di_nextents) + |
276 | be16_to_cpu(dip->di_core.di_anextents)), | |
31c5308f | 277 | (unsigned long long) |
5e656dbb | 278 | be64_to_cpu(dip->di_core.di_nblocks)); |
4ca431fc NS |
279 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
280 | ip->i_mount, dip); | |
2bd0ea18 NS |
281 | return XFS_ERROR(EFSCORRUPTED); |
282 | } | |
283 | ||
5e656dbb | 284 | if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
23297264 NS |
285 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
286 | "corrupt dinode %Lu, forkoff = 0x%x.", | |
31c5308f | 287 | (unsigned long long)ip->i_ino, |
5e656dbb | 288 | dip->di_core.di_forkoff); |
4ca431fc NS |
289 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
290 | ip->i_mount, dip); | |
2bd0ea18 NS |
291 | return XFS_ERROR(EFSCORRUPTED); |
292 | } | |
293 | ||
322f2a29 SL |
294 | switch (ip->i_d.di_mode & S_IFMT) { |
295 | case S_IFIFO: | |
296 | case S_IFCHR: | |
297 | case S_IFBLK: | |
298 | case S_IFSOCK: | |
5e656dbb | 299 | if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) { |
4ca431fc NS |
300 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
301 | ip->i_mount, dip); | |
2bd0ea18 | 302 | return XFS_ERROR(EFSCORRUPTED); |
4ca431fc | 303 | } |
2bd0ea18 | 304 | ip->i_d.di_size = 0; |
5e656dbb BN |
305 | ip->i_size = 0; |
306 | ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev); | |
2bd0ea18 NS |
307 | break; |
308 | ||
322f2a29 SL |
309 | case S_IFREG: |
310 | case S_IFLNK: | |
311 | case S_IFDIR: | |
5e656dbb | 312 | switch (dip->di_core.di_format) { |
2bd0ea18 NS |
313 | case XFS_DINODE_FMT_LOCAL: |
314 | /* | |
315 | * no local regular files yet | |
316 | */ | |
5e656dbb | 317 | if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) { |
23297264 NS |
318 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
319 | "corrupt inode %Lu " | |
320 | "(local format for regular file).", | |
31c5308f | 321 | (unsigned long long) ip->i_ino); |
4ca431fc NS |
322 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", |
323 | XFS_ERRLEVEL_LOW, | |
324 | ip->i_mount, dip); | |
2bd0ea18 NS |
325 | return XFS_ERROR(EFSCORRUPTED); |
326 | } | |
31c5308f | 327 | |
5e656dbb | 328 | di_size = be64_to_cpu(dip->di_core.di_size); |
46eca962 | 329 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
23297264 NS |
330 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
331 | "corrupt inode %Lu " | |
332 | "(bad size %Ld for local inode).", | |
31c5308f NS |
333 | (unsigned long long) ip->i_ino, |
334 | (long long) di_size); | |
4ca431fc NS |
335 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", |
336 | XFS_ERRLEVEL_LOW, | |
337 | ip->i_mount, dip); | |
2bd0ea18 NS |
338 | return XFS_ERROR(EFSCORRUPTED); |
339 | } | |
340 | ||
341 | size = (int)di_size; | |
2f8883de | 342 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); |
2bd0ea18 NS |
343 | break; |
344 | case XFS_DINODE_FMT_EXTENTS: | |
2f8883de | 345 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); |
2bd0ea18 NS |
346 | break; |
347 | case XFS_DINODE_FMT_BTREE: | |
2f8883de | 348 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); |
2bd0ea18 NS |
349 | break; |
350 | default: | |
4ca431fc NS |
351 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, |
352 | ip->i_mount); | |
2bd0ea18 NS |
353 | return XFS_ERROR(EFSCORRUPTED); |
354 | } | |
355 | break; | |
356 | ||
357 | default: | |
4ca431fc | 358 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); |
2bd0ea18 NS |
359 | return XFS_ERROR(EFSCORRUPTED); |
360 | } | |
361 | if (error) { | |
362 | return error; | |
5000d01d | 363 | } |
46eca962 | 364 | if (!XFS_DFORK_Q(dip)) |
2bd0ea18 NS |
365 | return 0; |
366 | ASSERT(ip->i_afp == NULL); | |
2f8883de | 367 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); |
2bd0ea18 NS |
368 | ip->i_afp->if_ext_max = |
369 | XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
5e656dbb | 370 | switch (dip->di_core.di_aformat) { |
2bd0ea18 | 371 | case XFS_DINODE_FMT_LOCAL: |
46eca962 | 372 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); |
5e656dbb | 373 | size = be16_to_cpu(atp->hdr.totsize); |
2f8883de | 374 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
2bd0ea18 NS |
375 | break; |
376 | case XFS_DINODE_FMT_EXTENTS: | |
2f8883de | 377 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); |
2bd0ea18 NS |
378 | break; |
379 | case XFS_DINODE_FMT_BTREE: | |
2f8883de | 380 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); |
2bd0ea18 NS |
381 | break; |
382 | default: | |
383 | error = XFS_ERROR(EFSCORRUPTED); | |
384 | break; | |
385 | } | |
386 | if (error) { | |
387 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
388 | ip->i_afp = NULL; | |
389 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
390 | } | |
391 | return error; | |
392 | } | |
393 | ||
394 | /* | |
395 | * The file is in-lined in the on-disk inode. | |
396 | * If it fits into if_inline_data, then copy | |
397 | * it there, otherwise allocate a buffer for it | |
398 | * and copy the data there. Either way, set | |
399 | * if_data to point at the data. | |
400 | * If we allocate a buffer for the data, make | |
401 | * sure that its size is a multiple of 4 and | |
402 | * record the real size in i_real_bytes. | |
403 | */ | |
404 | STATIC int | |
405 | xfs_iformat_local( | |
406 | xfs_inode_t *ip, | |
407 | xfs_dinode_t *dip, | |
408 | int whichfork, | |
2f8883de | 409 | int size) |
2bd0ea18 NS |
410 | { |
411 | xfs_ifork_t *ifp; | |
412 | int real_size; | |
413 | ||
414 | /* | |
415 | * If the size is unreasonable, then something | |
416 | * is wrong and we just bail out rather than crash in | |
32181a02 | 417 | * kmem_alloc() or memcpy() below. |
2bd0ea18 | 418 | */ |
46eca962 | 419 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { |
23297264 NS |
420 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
421 | "corrupt inode %Lu " | |
422 | "(bad size %d for local fork, size = %d).", | |
31c5308f | 423 | (unsigned long long) ip->i_ino, size, |
46eca962 | 424 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); |
4ca431fc NS |
425 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, |
426 | ip->i_mount, dip); | |
2bd0ea18 NS |
427 | return XFS_ERROR(EFSCORRUPTED); |
428 | } | |
429 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
430 | real_size = 0; | |
431 | if (size == 0) | |
432 | ifp->if_u1.if_data = NULL; | |
433 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
434 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
435 | else { | |
436 | real_size = roundup(size, 4); | |
2f8883de | 437 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); |
2bd0ea18 NS |
438 | } |
439 | ifp->if_bytes = size; | |
440 | ifp->if_real_bytes = real_size; | |
441 | if (size) | |
46eca962 | 442 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); |
2bd0ea18 NS |
443 | ifp->if_flags &= ~XFS_IFEXTENTS; |
444 | ifp->if_flags |= XFS_IFINLINE; | |
445 | return 0; | |
446 | } | |
447 | ||
448 | /* | |
449 | * The file consists of a set of extents all | |
450 | * of which fit into the on-disk inode. | |
451 | * If there are few enough extents to fit into | |
452 | * the if_inline_ext, then copy them there. | |
453 | * Otherwise allocate a buffer for them and copy | |
454 | * them into it. Either way, set if_extents | |
455 | * to point at the extents. | |
456 | */ | |
457 | STATIC int | |
458 | xfs_iformat_extents( | |
459 | xfs_inode_t *ip, | |
460 | xfs_dinode_t *dip, | |
2f8883de | 461 | int whichfork) |
2bd0ea18 | 462 | { |
5e656dbb | 463 | xfs_bmbt_rec_t *dp; |
2bd0ea18 NS |
464 | xfs_ifork_t *ifp; |
465 | int nex; | |
2bd0ea18 | 466 | int size; |
f9e56f43 | 467 | int i; |
2bd0ea18 NS |
468 | |
469 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
46eca962 | 470 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); |
2bd0ea18 NS |
471 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); |
472 | ||
473 | /* | |
474 | * If the number of extents is unreasonable, then something | |
475 | * is wrong and we just bail out rather than crash in | |
32181a02 | 476 | * kmem_alloc() or memcpy() below. |
2bd0ea18 | 477 | */ |
46eca962 | 478 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { |
23297264 NS |
479 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
480 | "corrupt inode %Lu ((a)extents = %d).", | |
31c5308f | 481 | (unsigned long long) ip->i_ino, nex); |
4ca431fc NS |
482 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, |
483 | ip->i_mount, dip); | |
2bd0ea18 NS |
484 | return XFS_ERROR(EFSCORRUPTED); |
485 | } | |
486 | ||
5e656dbb | 487 | ifp->if_real_bytes = 0; |
2bd0ea18 NS |
488 | if (nex == 0) |
489 | ifp->if_u1.if_extents = NULL; | |
490 | else if (nex <= XFS_INLINE_EXTS) | |
491 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
5e656dbb BN |
492 | else |
493 | xfs_iext_add(ifp, 0, nex); | |
494 | ||
2bd0ea18 | 495 | ifp->if_bytes = size; |
2bd0ea18 | 496 | if (size) { |
46eca962 | 497 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); |
5e656dbb BN |
498 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
499 | for (i = 0; i < nex; i++, dp++) { | |
500 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); | |
501 | ep->l0 = get_unaligned_be64(&dp->l0); | |
502 | ep->l1 = get_unaligned_be64(&dp->l1); | |
f9e56f43 | 503 | } |
5e656dbb | 504 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
2bd0ea18 NS |
505 | if (whichfork != XFS_DATA_FORK || |
506 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
4ca431fc | 507 | if (unlikely(xfs_check_nostate_extents( |
5e656dbb | 508 | ifp, 0, nex))) { |
4ca431fc NS |
509 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
510 | XFS_ERRLEVEL_LOW, | |
511 | ip->i_mount); | |
2bd0ea18 | 512 | return XFS_ERROR(EFSCORRUPTED); |
4ca431fc | 513 | } |
2bd0ea18 NS |
514 | } |
515 | ifp->if_flags |= XFS_IFEXTENTS; | |
516 | return 0; | |
517 | } | |
518 | ||
519 | /* | |
520 | * The file has too many extents to fit into | |
521 | * the inode, so they are in B-tree format. | |
522 | * Allocate a buffer for the root of the B-tree | |
523 | * and copy the root into it. The i_extents | |
524 | * field will remain NULL until all of the | |
525 | * extents are read in (when they are needed). | |
526 | */ | |
527 | STATIC int | |
528 | xfs_iformat_btree( | |
529 | xfs_inode_t *ip, | |
530 | xfs_dinode_t *dip, | |
2f8883de | 531 | int whichfork) |
2bd0ea18 NS |
532 | { |
533 | xfs_bmdr_block_t *dfp; | |
534 | xfs_ifork_t *ifp; | |
535 | /* REFERENCED */ | |
536 | int nrecs; | |
537 | int size; | |
538 | ||
539 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
46eca962 | 540 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); |
2bd0ea18 | 541 | size = XFS_BMAP_BROOT_SPACE(dfp); |
b3563c19 | 542 | nrecs = be16_to_cpu(dfp->bb_numrecs); |
2bd0ea18 NS |
543 | |
544 | /* | |
545 | * blow out if -- fork has less extents than can fit in | |
546 | * fork (fork shouldn't be a btree format), root btree | |
547 | * block has more records than can fit into the fork, | |
548 | * or the number of extents is greater than the number of | |
549 | * blocks. | |
550 | */ | |
4ca431fc | 551 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max |
2bd0ea18 | 552 | || XFS_BMDR_SPACE_CALC(nrecs) > |
46eca962 | 553 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) |
4ca431fc | 554 | || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { |
23297264 NS |
555 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
556 | "corrupt inode %Lu (btree).", | |
31c5308f | 557 | (unsigned long long) ip->i_ino); |
4ca431fc NS |
558 | XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, |
559 | ip->i_mount); | |
2bd0ea18 NS |
560 | return XFS_ERROR(EFSCORRUPTED); |
561 | } | |
562 | ||
563 | ifp->if_broot_bytes = size; | |
2f8883de | 564 | ifp->if_broot = kmem_alloc(size, KM_SLEEP); |
2bd0ea18 NS |
565 | ASSERT(ifp->if_broot != NULL); |
566 | /* | |
567 | * Copy and convert from the on-disk structure | |
568 | * to the in-memory structure. | |
569 | */ | |
b3563c19 BN |
570 | xfs_bmdr_to_bmbt(ip->i_mount, dfp, |
571 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
572 | ifp->if_broot, size); | |
2bd0ea18 NS |
573 | ifp->if_flags &= ~XFS_IFEXTENTS; |
574 | ifp->if_flags |= XFS_IFBROOT; | |
575 | ||
576 | return 0; | |
577 | } | |
578 | ||
5000d01d | 579 | void |
5e656dbb BN |
580 | xfs_dinode_from_disk( |
581 | xfs_icdinode_t *to, | |
582 | xfs_dinode_core_t *from) | |
2bd0ea18 | 583 | { |
5e656dbb BN |
584 | to->di_magic = be16_to_cpu(from->di_magic); |
585 | to->di_mode = be16_to_cpu(from->di_mode); | |
586 | to->di_version = from ->di_version; | |
587 | to->di_format = from->di_format; | |
588 | to->di_onlink = be16_to_cpu(from->di_onlink); | |
589 | to->di_uid = be32_to_cpu(from->di_uid); | |
590 | to->di_gid = be32_to_cpu(from->di_gid); | |
591 | to->di_nlink = be32_to_cpu(from->di_nlink); | |
592 | to->di_projid = be16_to_cpu(from->di_projid); | |
593 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
594 | to->di_flushiter = be16_to_cpu(from->di_flushiter); | |
595 | to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); | |
596 | to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); | |
597 | to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); | |
598 | to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); | |
599 | to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); | |
600 | to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); | |
601 | to->di_size = be64_to_cpu(from->di_size); | |
602 | to->di_nblocks = be64_to_cpu(from->di_nblocks); | |
603 | to->di_extsize = be32_to_cpu(from->di_extsize); | |
604 | to->di_nextents = be32_to_cpu(from->di_nextents); | |
605 | to->di_anextents = be16_to_cpu(from->di_anextents); | |
606 | to->di_forkoff = from->di_forkoff; | |
607 | to->di_aformat = from->di_aformat; | |
608 | to->di_dmevmask = be32_to_cpu(from->di_dmevmask); | |
609 | to->di_dmstate = be16_to_cpu(from->di_dmstate); | |
610 | to->di_flags = be16_to_cpu(from->di_flags); | |
611 | to->di_gen = be32_to_cpu(from->di_gen); | |
612 | } | |
613 | ||
614 | void | |
615 | xfs_dinode_to_disk( | |
616 | xfs_dinode_core_t *to, | |
617 | xfs_icdinode_t *from) | |
618 | { | |
619 | to->di_magic = cpu_to_be16(from->di_magic); | |
620 | to->di_mode = cpu_to_be16(from->di_mode); | |
621 | to->di_version = from ->di_version; | |
622 | to->di_format = from->di_format; | |
623 | to->di_onlink = cpu_to_be16(from->di_onlink); | |
624 | to->di_uid = cpu_to_be32(from->di_uid); | |
625 | to->di_gid = cpu_to_be32(from->di_gid); | |
626 | to->di_nlink = cpu_to_be32(from->di_nlink); | |
627 | to->di_projid = cpu_to_be16(from->di_projid); | |
628 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
629 | to->di_flushiter = cpu_to_be16(from->di_flushiter); | |
630 | to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); | |
631 | to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); | |
632 | to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); | |
633 | to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); | |
634 | to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); | |
635 | to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); | |
636 | to->di_size = cpu_to_be64(from->di_size); | |
637 | to->di_nblocks = cpu_to_be64(from->di_nblocks); | |
638 | to->di_extsize = cpu_to_be32(from->di_extsize); | |
639 | to->di_nextents = cpu_to_be32(from->di_nextents); | |
640 | to->di_anextents = cpu_to_be16(from->di_anextents); | |
641 | to->di_forkoff = from->di_forkoff; | |
642 | to->di_aformat = from->di_aformat; | |
643 | to->di_dmevmask = cpu_to_be32(from->di_dmevmask); | |
644 | to->di_dmstate = cpu_to_be16(from->di_dmstate); | |
645 | to->di_flags = cpu_to_be16(from->di_flags); | |
646 | to->di_gen = cpu_to_be32(from->di_gen); | |
2bd0ea18 NS |
647 | } |
648 | ||
2bd0ea18 NS |
649 | /* |
650 | * Read in extents from a btree-format inode. | |
651 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
652 | */ | |
653 | int | |
654 | xfs_iread_extents( | |
655 | xfs_trans_t *tp, | |
656 | xfs_inode_t *ip, | |
657 | int whichfork) | |
658 | { | |
659 | int error; | |
660 | xfs_ifork_t *ifp; | |
5e656dbb | 661 | xfs_extnum_t nextents; |
2bd0ea18 NS |
662 | size_t size; |
663 | ||
4ca431fc NS |
664 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { |
665 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
666 | ip->i_mount); | |
2bd0ea18 | 667 | return XFS_ERROR(EFSCORRUPTED); |
4ca431fc | 668 | } |
5e656dbb BN |
669 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
670 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
2bd0ea18 | 671 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5e656dbb | 672 | |
2bd0ea18 | 673 | /* |
05bba5b7 | 674 | * We know that the size is valid (it's checked in iformat_btree) |
2bd0ea18 | 675 | */ |
2bd0ea18 | 676 | ifp->if_lastex = NULLEXTNUM; |
5e656dbb | 677 | ifp->if_bytes = ifp->if_real_bytes = 0; |
2bd0ea18 | 678 | ifp->if_flags |= XFS_IFEXTENTS; |
5e656dbb | 679 | xfs_iext_add(ifp, 0, nextents); |
2bd0ea18 NS |
680 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
681 | if (error) { | |
5e656dbb | 682 | xfs_iext_destroy(ifp); |
2bd0ea18 NS |
683 | ifp->if_flags &= ~XFS_IFEXTENTS; |
684 | return error; | |
685 | } | |
5e656dbb | 686 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
2bd0ea18 NS |
687 | return 0; |
688 | } | |
689 | ||
690 | /* | |
691 | * Reallocate the space for if_broot based on the number of records | |
692 | * being added or deleted as indicated in rec_diff. Move the records | |
693 | * and pointers in if_broot to fit the new size. When shrinking this | |
694 | * will eliminate holes between the records and pointers created by | |
5e656dbb | 695 | * the caller. When growing this will create holes to be filled in |
2bd0ea18 NS |
696 | * by the caller. |
697 | * | |
698 | * The caller must not request to add more records than would fit in | |
699 | * the on-disk inode root. If the if_broot is currently NULL, then | |
5e656dbb | 700 | * if we adding records one will be allocated. The caller must also |
2bd0ea18 NS |
701 | * not request that the number of records go below zero, although |
702 | * it can go to zero. | |
703 | * | |
704 | * ip -- the inode whose if_broot area is changing | |
705 | * ext_diff -- the change in the number of records, positive or negative, | |
706 | * requested for the if_broot array. | |
707 | */ | |
708 | void | |
709 | xfs_iroot_realloc( | |
5000d01d SL |
710 | xfs_inode_t *ip, |
711 | int rec_diff, | |
2bd0ea18 NS |
712 | int whichfork) |
713 | { | |
b3563c19 | 714 | struct xfs_mount *mp = ip->i_mount; |
2bd0ea18 NS |
715 | int cur_max; |
716 | xfs_ifork_t *ifp; | |
b3563c19 | 717 | struct xfs_btree_block *new_broot; |
2bd0ea18 NS |
718 | int new_max; |
719 | size_t new_size; | |
720 | char *np; | |
721 | char *op; | |
722 | ||
723 | /* | |
724 | * Handle the degenerate case quietly. | |
725 | */ | |
726 | if (rec_diff == 0) { | |
727 | return; | |
728 | } | |
729 | ||
730 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
731 | if (rec_diff > 0) { | |
732 | /* | |
733 | * If there wasn't any memory allocated before, just | |
734 | * allocate it now and get out. | |
735 | */ | |
736 | if (ifp->if_broot_bytes == 0) { | |
737 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
b3563c19 | 738 | ifp->if_broot = kmem_alloc(new_size, KM_SLEEP); |
2bd0ea18 NS |
739 | ifp->if_broot_bytes = (int)new_size; |
740 | return; | |
741 | } | |
742 | ||
743 | /* | |
744 | * If there is already an existing if_broot, then we need | |
745 | * to realloc() it and shift the pointers to their new | |
746 | * location. The records don't change location because | |
747 | * they are kept butted up against the btree block header. | |
748 | */ | |
b3563c19 | 749 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
2bd0ea18 NS |
750 | new_max = cur_max + rec_diff; |
751 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
b3563c19 | 752 | ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, |
2bd0ea18 NS |
753 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ |
754 | KM_SLEEP); | |
b3563c19 BN |
755 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
756 | ifp->if_broot_bytes); | |
757 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, | |
758 | (int)new_size); | |
2bd0ea18 NS |
759 | ifp->if_broot_bytes = (int)new_size; |
760 | ASSERT(ifp->if_broot_bytes <= | |
761 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
32181a02 | 762 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); |
2bd0ea18 NS |
763 | return; |
764 | } | |
765 | ||
766 | /* | |
767 | * rec_diff is less than 0. In this case, we are shrinking the | |
768 | * if_broot buffer. It must already exist. If we go to zero | |
769 | * records, just get rid of the root and clear the status bit. | |
770 | */ | |
771 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
b3563c19 | 772 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
2bd0ea18 NS |
773 | new_max = cur_max + rec_diff; |
774 | ASSERT(new_max >= 0); | |
775 | if (new_max > 0) | |
776 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
777 | else | |
778 | new_size = 0; | |
779 | if (new_size > 0) { | |
b3563c19 | 780 | new_broot = kmem_alloc(new_size, KM_SLEEP); |
2bd0ea18 NS |
781 | /* |
782 | * First copy over the btree block header. | |
783 | */ | |
b3563c19 | 784 | memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN); |
2bd0ea18 NS |
785 | } else { |
786 | new_broot = NULL; | |
787 | ifp->if_flags &= ~XFS_IFBROOT; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Only copy the records and pointers if there are any. | |
792 | */ | |
793 | if (new_max > 0) { | |
794 | /* | |
795 | * First copy the records. | |
796 | */ | |
b3563c19 BN |
797 | op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); |
798 | np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); | |
32181a02 | 799 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); |
2bd0ea18 NS |
800 | |
801 | /* | |
802 | * Then copy the pointers. | |
803 | */ | |
b3563c19 | 804 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
2bd0ea18 | 805 | ifp->if_broot_bytes); |
b3563c19 | 806 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, |
2bd0ea18 | 807 | (int)new_size); |
32181a02 | 808 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); |
2bd0ea18 | 809 | } |
5e656dbb | 810 | kmem_free(ifp->if_broot); |
2bd0ea18 NS |
811 | ifp->if_broot = new_broot; |
812 | ifp->if_broot_bytes = (int)new_size; | |
813 | ASSERT(ifp->if_broot_bytes <= | |
814 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
815 | return; | |
816 | } | |
817 | ||
2bd0ea18 NS |
818 | |
819 | /* | |
820 | * This is called when the amount of space needed for if_data | |
821 | * is increased or decreased. The change in size is indicated by | |
822 | * the number of bytes that need to be added or deleted in the | |
823 | * byte_diff parameter. | |
824 | * | |
825 | * If the amount of space needed has decreased below the size of the | |
826 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
827 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
828 | * to what is needed. | |
829 | * | |
830 | * ip -- the inode whose if_data area is changing | |
831 | * byte_diff -- the change in the number of bytes, positive or negative, | |
832 | * requested for the if_data array. | |
833 | */ | |
834 | void | |
835 | xfs_idata_realloc( | |
836 | xfs_inode_t *ip, | |
837 | int byte_diff, | |
838 | int whichfork) | |
839 | { | |
840 | xfs_ifork_t *ifp; | |
841 | int new_size; | |
842 | int real_size; | |
843 | ||
844 | if (byte_diff == 0) { | |
845 | return; | |
846 | } | |
847 | ||
848 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
849 | new_size = (int)ifp->if_bytes + byte_diff; | |
850 | ASSERT(new_size >= 0); | |
851 | ||
852 | if (new_size == 0) { | |
853 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
5e656dbb | 854 | kmem_free(ifp->if_u1.if_data); |
2bd0ea18 NS |
855 | } |
856 | ifp->if_u1.if_data = NULL; | |
857 | real_size = 0; | |
858 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
859 | /* | |
860 | * If the valid extents/data can fit in if_inline_ext/data, | |
861 | * copy them from the malloc'd vector and free it. | |
862 | */ | |
863 | if (ifp->if_u1.if_data == NULL) { | |
864 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
865 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
866 | ASSERT(ifp->if_real_bytes != 0); | |
32181a02 | 867 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, |
2bd0ea18 | 868 | new_size); |
5e656dbb | 869 | kmem_free(ifp->if_u1.if_data); |
2bd0ea18 NS |
870 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
871 | } | |
872 | real_size = 0; | |
873 | } else { | |
874 | /* | |
875 | * Stuck with malloc/realloc. | |
876 | * For inline data, the underlying buffer must be | |
877 | * a multiple of 4 bytes in size so that it can be | |
4ed50f8a | 878 | * logged and stay on word boundaries. We enforce |
2bd0ea18 NS |
879 | * that here. |
880 | */ | |
881 | real_size = roundup(new_size, 4); | |
882 | if (ifp->if_u1.if_data == NULL) { | |
883 | ASSERT(ifp->if_real_bytes == 0); | |
884 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
885 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
886 | /* | |
887 | * Only do the realloc if the underlying size | |
888 | * is really changing. | |
889 | */ | |
890 | if (ifp->if_real_bytes != real_size) { | |
891 | ifp->if_u1.if_data = | |
892 | kmem_realloc(ifp->if_u1.if_data, | |
893 | real_size, | |
894 | ifp->if_real_bytes, | |
895 | KM_SLEEP); | |
896 | } | |
897 | } else { | |
898 | ASSERT(ifp->if_real_bytes == 0); | |
899 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
32181a02 NS |
900 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, |
901 | ifp->if_bytes); | |
2bd0ea18 NS |
902 | } |
903 | } | |
904 | ifp->if_real_bytes = real_size; | |
905 | ifp->if_bytes = new_size; | |
906 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
907 | } | |
908 | ||
909 | ||
5e656dbb BN |
910 | |
911 | ||
2bd0ea18 NS |
912 | /* |
913 | * Map inode to disk block and offset. | |
914 | * | |
915 | * mp -- the mount point structure for the current file system | |
916 | * tp -- the current transaction | |
917 | * ino -- the inode number of the inode to be located | |
918 | * imap -- this structure is filled in with the information necessary | |
919 | * to retrieve the given inode from disk | |
920 | * flags -- flags to pass to xfs_dilocate indicating whether or not | |
921 | * lookups in the inode btree were OK or not | |
922 | */ | |
923 | int | |
924 | xfs_imap( | |
925 | xfs_mount_t *mp, | |
926 | xfs_trans_t *tp, | |
927 | xfs_ino_t ino, | |
928 | xfs_imap_t *imap, | |
929 | uint flags) | |
930 | { | |
931 | xfs_fsblock_t fsbno; | |
932 | int len; | |
933 | int off; | |
934 | int error; | |
935 | ||
936 | fsbno = imap->im_blkno ? | |
937 | XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; | |
938 | error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); | |
5e656dbb | 939 | if (error) |
2bd0ea18 | 940 | return error; |
5e656dbb | 941 | |
2bd0ea18 NS |
942 | imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); |
943 | imap->im_len = XFS_FSB_TO_BB(mp, len); | |
944 | imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); | |
945 | imap->im_ioffset = (ushort)off; | |
946 | imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); | |
5e656dbb BN |
947 | |
948 | /* | |
949 | * If the inode number maps to a block outside the bounds | |
950 | * of the file system then return NULL rather than calling | |
951 | * read_buf and panicing when we get an error from the | |
952 | * driver. | |
953 | */ | |
954 | if ((imap->im_blkno + imap->im_len) > | |
955 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
956 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: " | |
957 | "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > " | |
958 | " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)", | |
959 | (unsigned long long) imap->im_blkno, | |
960 | (unsigned long long) imap->im_len, | |
961 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
962 | return EINVAL; | |
963 | } | |
2bd0ea18 NS |
964 | return 0; |
965 | } | |
966 | ||
967 | void | |
968 | xfs_idestroy_fork( | |
969 | xfs_inode_t *ip, | |
970 | int whichfork) | |
971 | { | |
972 | xfs_ifork_t *ifp; | |
973 | ||
974 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
975 | if (ifp->if_broot != NULL) { | |
5e656dbb | 976 | kmem_free(ifp->if_broot); |
2bd0ea18 NS |
977 | ifp->if_broot = NULL; |
978 | } | |
979 | ||
980 | /* | |
981 | * If the format is local, then we can't have an extents | |
4ed50f8a | 982 | * array so just look for an inline data array. If we're |
2bd0ea18 NS |
983 | * not local then we may or may not have an extents list, |
984 | * so check and free it up if we do. | |
985 | */ | |
986 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
5000d01d | 987 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && |
2bd0ea18 NS |
988 | (ifp->if_u1.if_data != NULL)) { |
989 | ASSERT(ifp->if_real_bytes != 0); | |
5e656dbb | 990 | kmem_free(ifp->if_u1.if_data); |
2bd0ea18 NS |
991 | ifp->if_u1.if_data = NULL; |
992 | ifp->if_real_bytes = 0; | |
993 | } | |
994 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
5e656dbb BN |
995 | ((ifp->if_flags & XFS_IFEXTIREC) || |
996 | ((ifp->if_u1.if_extents != NULL) && | |
997 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
2bd0ea18 | 998 | ASSERT(ifp->if_real_bytes != 0); |
5e656dbb | 999 | xfs_iext_destroy(ifp); |
2bd0ea18 NS |
1000 | } |
1001 | ASSERT(ifp->if_u1.if_extents == NULL || | |
1002 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
1003 | ASSERT(ifp->if_real_bytes == 0); | |
1004 | if (whichfork == XFS_ATTR_FORK) { | |
1005 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
1006 | ip->i_afp = NULL; | |
1007 | } | |
1008 | } | |
1009 | ||
2bd0ea18 NS |
1010 | /* |
1011 | * xfs_iextents_copy() | |
1012 | * | |
1013 | * This is called to copy the REAL extents (as opposed to the delayed | |
1014 | * allocation extents) from the inode into the given buffer. It | |
1015 | * returns the number of bytes copied into the buffer. | |
1016 | * | |
1017 | * If there are no delayed allocation extents, then we can just | |
32181a02 | 1018 | * memcpy() the extents into the buffer. Otherwise, we need to |
2bd0ea18 NS |
1019 | * examine each extent in turn and skip those which are delayed. |
1020 | */ | |
1021 | int | |
1022 | xfs_iextents_copy( | |
1023 | xfs_inode_t *ip, | |
5e656dbb | 1024 | xfs_bmbt_rec_t *dp, |
2bd0ea18 NS |
1025 | int whichfork) |
1026 | { | |
1027 | int copied; | |
2bd0ea18 NS |
1028 | int i; |
1029 | xfs_ifork_t *ifp; | |
1030 | int nrecs; | |
1031 | xfs_fsblock_t start_block; | |
1032 | ||
1033 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
5e656dbb | 1034 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
2bd0ea18 NS |
1035 | ASSERT(ifp->if_bytes > 0); |
1036 | ||
1037 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
5e656dbb | 1038 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
2bd0ea18 | 1039 | ASSERT(nrecs > 0); |
2bd0ea18 | 1040 | |
2bd0ea18 NS |
1041 | /* |
1042 | * There are some delayed allocation extents in the | |
1043 | * inode, so copy the extents one at a time and skip | |
1044 | * the delayed ones. There must be at least one | |
1045 | * non-delayed extent. | |
1046 | */ | |
2bd0ea18 NS |
1047 | copied = 0; |
1048 | for (i = 0; i < nrecs; i++) { | |
5e656dbb | 1049 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
2bd0ea18 NS |
1050 | start_block = xfs_bmbt_get_startblock(ep); |
1051 | if (ISNULLSTARTBLOCK(start_block)) { | |
1052 | /* | |
1053 | * It's a delayed allocation extent, so skip it. | |
1054 | */ | |
2bd0ea18 NS |
1055 | continue; |
1056 | } | |
1057 | ||
f9e56f43 | 1058 | /* Translate to on disk format */ |
5e656dbb BN |
1059 | put_unaligned(cpu_to_be64(ep->l0), &dp->l0); |
1060 | put_unaligned(cpu_to_be64(ep->l1), &dp->l1); | |
1061 | dp++; | |
2bd0ea18 NS |
1062 | copied++; |
1063 | } | |
1064 | ASSERT(copied != 0); | |
5e656dbb | 1065 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
2bd0ea18 NS |
1066 | |
1067 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
3ded4129 | 1068 | } |
2bd0ea18 NS |
1069 | |
1070 | /* | |
1071 | * Each of the following cases stores data into the same region | |
1072 | * of the on-disk inode, so only one of them can be valid at | |
1073 | * any given time. While it is possible to have conflicting formats | |
1074 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
1075 | * in EXTENTS format, this can only happen when the fork has | |
1076 | * changed formats after being modified but before being flushed. | |
1077 | * In these cases, the format always takes precedence, because the | |
1078 | * format indicates the current state of the fork. | |
5e656dbb BN |
1079 | * |
1080 | * Note: this requires user-space public scope for libxfs_iread | |
2bd0ea18 | 1081 | */ |
5e656dbb BN |
1082 | /*ARGSUSED*/ |
1083 | void | |
2bd0ea18 NS |
1084 | xfs_iflush_fork( |
1085 | xfs_inode_t *ip, | |
1086 | xfs_dinode_t *dip, | |
1087 | xfs_inode_log_item_t *iip, | |
1088 | int whichfork, | |
1089 | xfs_buf_t *bp) | |
1090 | { | |
1091 | char *cp; | |
1092 | xfs_ifork_t *ifp; | |
1093 | xfs_mount_t *mp; | |
1094 | #ifdef XFS_TRANS_DEBUG | |
1095 | int first; | |
1096 | #endif | |
1097 | static const short brootflag[2] = | |
1098 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
1099 | static const short dataflag[2] = | |
1100 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
1101 | static const short extflag[2] = | |
1102 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
1103 | ||
5e656dbb BN |
1104 | if (!iip) |
1105 | return; | |
2bd0ea18 NS |
1106 | ifp = XFS_IFORK_PTR(ip, whichfork); |
1107 | /* | |
1108 | * This can happen if we gave up in iformat in an error path, | |
1109 | * for the attribute fork. | |
1110 | */ | |
5e656dbb | 1111 | if (!ifp) { |
2bd0ea18 | 1112 | ASSERT(whichfork == XFS_ATTR_FORK); |
5e656dbb | 1113 | return; |
2bd0ea18 | 1114 | } |
46eca962 | 1115 | cp = XFS_DFORK_PTR(dip, whichfork); |
2bd0ea18 NS |
1116 | mp = ip->i_mount; |
1117 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
1118 | case XFS_DINODE_FMT_LOCAL: | |
1119 | if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && | |
1120 | (ifp->if_bytes > 0)) { | |
1121 | ASSERT(ifp->if_u1.if_data != NULL); | |
1122 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
32181a02 | 1123 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); |
2bd0ea18 | 1124 | } |
2bd0ea18 NS |
1125 | break; |
1126 | ||
1127 | case XFS_DINODE_FMT_EXTENTS: | |
1128 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
1129 | !(iip->ili_format.ilf_fields & extflag[whichfork])); | |
5e656dbb BN |
1130 | ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) || |
1131 | (ifp->if_bytes == 0)); | |
1132 | ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) || | |
1133 | (ifp->if_bytes > 0)); | |
2bd0ea18 NS |
1134 | if ((iip->ili_format.ilf_fields & extflag[whichfork]) && |
1135 | (ifp->if_bytes > 0)) { | |
1136 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); | |
f9e56f43 | 1137 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, |
2bd0ea18 NS |
1138 | whichfork); |
1139 | } | |
1140 | break; | |
1141 | ||
1142 | case XFS_DINODE_FMT_BTREE: | |
1143 | if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && | |
1144 | (ifp->if_broot_bytes > 0)) { | |
1145 | ASSERT(ifp->if_broot != NULL); | |
1146 | ASSERT(ifp->if_broot_bytes <= | |
1147 | (XFS_IFORK_SIZE(ip, whichfork) + | |
1148 | XFS_BROOT_SIZE_ADJ)); | |
b3563c19 | 1149 | xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, |
2bd0ea18 | 1150 | (xfs_bmdr_block_t *)cp, |
46eca962 | 1151 | XFS_DFORK_SIZE(dip, mp, whichfork)); |
2bd0ea18 NS |
1152 | } |
1153 | break; | |
1154 | ||
1155 | case XFS_DINODE_FMT_DEV: | |
1156 | if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { | |
1157 | ASSERT(whichfork == XFS_DATA_FORK); | |
5e656dbb | 1158 | dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev); |
2bd0ea18 NS |
1159 | } |
1160 | break; | |
5000d01d | 1161 | |
2bd0ea18 NS |
1162 | case XFS_DINODE_FMT_UUID: |
1163 | if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { | |
1164 | ASSERT(whichfork == XFS_DATA_FORK); | |
32181a02 | 1165 | memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, |
2bd0ea18 NS |
1166 | sizeof(uuid_t)); |
1167 | } | |
1168 | break; | |
1169 | ||
1170 | default: | |
1171 | ASSERT(0); | |
1172 | break; | |
1173 | } | |
5e656dbb | 1174 | } |
2bd0ea18 | 1175 | |
5e656dbb BN |
1176 | /* |
1177 | * Return a pointer to the extent record at file index idx. | |
1178 | */ | |
1179 | xfs_bmbt_rec_host_t * | |
1180 | xfs_iext_get_ext( | |
1181 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1182 | xfs_extnum_t idx) /* index of target extent */ | |
1183 | { | |
1184 | ASSERT(idx >= 0); | |
1185 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { | |
1186 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
1187 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
1188 | xfs_ext_irec_t *erp; /* irec pointer */ | |
1189 | int erp_idx = 0; /* irec index */ | |
1190 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
1191 | ||
1192 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
1193 | return &erp->er_extbuf[page_idx]; | |
1194 | } else if (ifp->if_bytes) { | |
1195 | return &ifp->if_u1.if_extents[idx]; | |
1196 | } else { | |
1197 | return NULL; | |
1198 | } | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Insert new item(s) into the extent records for incore inode | |
1203 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
1204 | */ | |
1205 | void | |
1206 | xfs_iext_insert( | |
1207 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1208 | xfs_extnum_t idx, /* starting index of new items */ | |
1209 | xfs_extnum_t count, /* number of inserted items */ | |
1210 | xfs_bmbt_irec_t *new) /* items to insert */ | |
1211 | { | |
1212 | xfs_extnum_t i; /* extent record index */ | |
1213 | ||
1214 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
1215 | xfs_iext_add(ifp, idx, count); | |
1216 | for (i = idx; i < idx + count; i++, new++) | |
1217 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); | |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * This is called when the amount of space required for incore file | |
1222 | * extents needs to be increased. The ext_diff parameter stores the | |
1223 | * number of new extents being added and the idx parameter contains | |
1224 | * the extent index where the new extents will be added. If the new | |
1225 | * extents are being appended, then we just need to (re)allocate and | |
1226 | * initialize the space. Otherwise, if the new extents are being | |
1227 | * inserted into the middle of the existing entries, a bit more work | |
1228 | * is required to make room for the new extents to be inserted. The | |
1229 | * caller is responsible for filling in the new extent entries upon | |
1230 | * return. | |
1231 | */ | |
1232 | void | |
1233 | xfs_iext_add( | |
1234 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1235 | xfs_extnum_t idx, /* index to begin adding exts */ | |
1236 | int ext_diff) /* number of extents to add */ | |
1237 | { | |
1238 | int byte_diff; /* new bytes being added */ | |
1239 | int new_size; /* size of extents after adding */ | |
1240 | xfs_extnum_t nextents; /* number of extents in file */ | |
1241 | ||
1242 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1243 | ASSERT((idx >= 0) && (idx <= nextents)); | |
1244 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
1245 | new_size = ifp->if_bytes + byte_diff; | |
1246 | /* | |
1247 | * If the new number of extents (nextents + ext_diff) | |
1248 | * fits inside the inode, then continue to use the inline | |
1249 | * extent buffer. | |
1250 | */ | |
1251 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
1252 | if (idx < nextents) { | |
1253 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
1254 | &ifp->if_u2.if_inline_ext[idx], | |
1255 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
1256 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
1257 | } | |
1258 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
1259 | ifp->if_real_bytes = 0; | |
1260 | ifp->if_lastex = nextents + ext_diff; | |
1261 | } | |
1262 | /* | |
1263 | * Otherwise use a linear (direct) extent list. | |
1264 | * If the extents are currently inside the inode, | |
1265 | * xfs_iext_realloc_direct will switch us from | |
1266 | * inline to direct extent allocation mode. | |
1267 | */ | |
1268 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { | |
1269 | xfs_iext_realloc_direct(ifp, new_size); | |
1270 | if (idx < nextents) { | |
1271 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
1272 | &ifp->if_u1.if_extents[idx], | |
1273 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
1274 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
1275 | } | |
1276 | } | |
1277 | /* Indirection array */ | |
1278 | else { | |
1279 | xfs_ext_irec_t *erp; | |
1280 | int erp_idx = 0; | |
1281 | int page_idx = idx; | |
1282 | ||
1283 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
1284 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
1285 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
1286 | } else { | |
1287 | xfs_iext_irec_init(ifp); | |
1288 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1289 | erp = ifp->if_u1.if_ext_irec; | |
1290 | } | |
1291 | /* Extents fit in target extent page */ | |
1292 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
1293 | if (page_idx < erp->er_extcount) { | |
1294 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
1295 | &erp->er_extbuf[page_idx], | |
1296 | (erp->er_extcount - page_idx) * | |
1297 | sizeof(xfs_bmbt_rec_t)); | |
1298 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
1299 | } | |
1300 | erp->er_extcount += ext_diff; | |
1301 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
1302 | } | |
1303 | /* Insert a new extent page */ | |
1304 | else if (erp) { | |
1305 | xfs_iext_add_indirect_multi(ifp, | |
1306 | erp_idx, page_idx, ext_diff); | |
1307 | } | |
1308 | /* | |
1309 | * If extent(s) are being appended to the last page in | |
1310 | * the indirection array and the new extent(s) don't fit | |
1311 | * in the page, then erp is NULL and erp_idx is set to | |
1312 | * the next index needed in the indirection array. | |
1313 | */ | |
1314 | else { | |
1315 | int count = ext_diff; | |
1316 | ||
1317 | while (count) { | |
1318 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
1319 | erp->er_extcount = count; | |
1320 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
1321 | if (count) { | |
1322 | erp_idx++; | |
1323 | } | |
1324 | } | |
1325 | } | |
1326 | } | |
1327 | ifp->if_bytes = new_size; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * This is called when incore extents are being added to the indirection | |
1332 | * array and the new extents do not fit in the target extent list. The | |
1333 | * erp_idx parameter contains the irec index for the target extent list | |
1334 | * in the indirection array, and the idx parameter contains the extent | |
1335 | * index within the list. The number of extents being added is stored | |
1336 | * in the count parameter. | |
1337 | * | |
1338 | * |-------| |-------| | |
1339 | * | | | | idx - number of extents before idx | |
1340 | * | idx | | count | | |
1341 | * | | | | count - number of extents being inserted at idx | |
1342 | * |-------| |-------| | |
1343 | * | count | | nex2 | nex2 - number of extents after idx + count | |
1344 | * |-------| |-------| | |
1345 | */ | |
1346 | void | |
1347 | xfs_iext_add_indirect_multi( | |
1348 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1349 | int erp_idx, /* target extent irec index */ | |
1350 | xfs_extnum_t idx, /* index within target list */ | |
1351 | int count) /* new extents being added */ | |
1352 | { | |
1353 | int byte_diff; /* new bytes being added */ | |
1354 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
1355 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
1356 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
1357 | xfs_extnum_t nex2; /* extents after idx + count */ | |
1358 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
1359 | int nlists; /* number of irec's (lists) */ | |
1360 | ||
1361 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1362 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
1363 | nex2 = erp->er_extcount - idx; | |
1364 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1365 | ||
1366 | /* | |
1367 | * Save second part of target extent list | |
1368 | * (all extents past */ | |
1369 | if (nex2) { | |
1370 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
1371 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); | |
1372 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); | |
1373 | erp->er_extcount -= nex2; | |
1374 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
1375 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * Add the new extents to the end of the target | |
1380 | * list, then allocate new irec record(s) and | |
1381 | * extent buffer(s) as needed to store the rest | |
1382 | * of the new extents. | |
1383 | */ | |
1384 | ext_cnt = count; | |
1385 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
1386 | if (ext_diff) { | |
1387 | erp->er_extcount += ext_diff; | |
1388 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
1389 | ext_cnt -= ext_diff; | |
1390 | } | |
1391 | while (ext_cnt) { | |
1392 | erp_idx++; | |
1393 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
1394 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
1395 | erp->er_extcount = ext_diff; | |
1396 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
1397 | ext_cnt -= ext_diff; | |
1398 | } | |
1399 | ||
1400 | /* Add nex2 extents back to indirection array */ | |
1401 | if (nex2) { | |
1402 | xfs_extnum_t ext_avail; | |
1403 | int i; | |
1404 | ||
1405 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
1406 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
1407 | i = 0; | |
1408 | /* | |
1409 | * If nex2 extents fit in the current page, append | |
1410 | * nex2_ep after the new extents. | |
1411 | */ | |
1412 | if (nex2 <= ext_avail) { | |
1413 | i = erp->er_extcount; | |
1414 | } | |
1415 | /* | |
1416 | * Otherwise, check if space is available in the | |
1417 | * next page. | |
1418 | */ | |
1419 | else if ((erp_idx < nlists - 1) && | |
1420 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
1421 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
1422 | erp_idx++; | |
1423 | erp++; | |
1424 | /* Create a hole for nex2 extents */ | |
1425 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
1426 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
1427 | } | |
1428 | /* | |
1429 | * Final choice, create a new extent page for | |
1430 | * nex2 extents. | |
1431 | */ | |
1432 | else { | |
1433 | erp_idx++; | |
1434 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
1435 | } | |
1436 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
1437 | kmem_free(nex2_ep); | |
1438 | erp->er_extcount += nex2; | |
1439 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * This is called when the amount of space required for incore file | |
1445 | * extents needs to be decreased. The ext_diff parameter stores the | |
1446 | * number of extents to be removed and the idx parameter contains | |
1447 | * the extent index where the extents will be removed from. | |
1448 | * | |
1449 | * If the amount of space needed has decreased below the linear | |
1450 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
1451 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
1452 | * size to what is needed. | |
1453 | */ | |
1454 | void | |
1455 | xfs_iext_remove( | |
1456 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1457 | xfs_extnum_t idx, /* index to begin removing exts */ | |
1458 | int ext_diff) /* number of extents to remove */ | |
1459 | { | |
1460 | xfs_extnum_t nextents; /* number of extents in file */ | |
1461 | int new_size; /* size of extents after removal */ | |
1462 | ||
1463 | ASSERT(ext_diff > 0); | |
1464 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1465 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
1466 | ||
1467 | if (new_size == 0) { | |
1468 | xfs_iext_destroy(ifp); | |
1469 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
1470 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
1471 | } else if (ifp->if_real_bytes) { | |
1472 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
1473 | } else { | |
1474 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
1475 | } | |
1476 | ifp->if_bytes = new_size; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * This removes ext_diff extents from the inline buffer, beginning | |
1481 | * at extent index idx. | |
1482 | */ | |
1483 | void | |
1484 | xfs_iext_remove_inline( | |
1485 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1486 | xfs_extnum_t idx, /* index to begin removing exts */ | |
1487 | int ext_diff) /* number of extents to remove */ | |
1488 | { | |
1489 | int nextents; /* number of extents in file */ | |
1490 | ||
1491 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
1492 | ASSERT(idx < XFS_INLINE_EXTS); | |
1493 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1494 | ASSERT(((nextents - ext_diff) > 0) && | |
1495 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
1496 | ||
1497 | if (idx + ext_diff < nextents) { | |
1498 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
1499 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
1500 | (nextents - (idx + ext_diff)) * | |
1501 | sizeof(xfs_bmbt_rec_t)); | |
1502 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
1503 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
1504 | } else { | |
1505 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
1506 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | /* | |
1511 | * This removes ext_diff extents from a linear (direct) extent list, | |
1512 | * beginning at extent index idx. If the extents are being removed | |
1513 | * from the end of the list (ie. truncate) then we just need to re- | |
1514 | * allocate the list to remove the extra space. Otherwise, if the | |
1515 | * extents are being removed from the middle of the existing extent | |
1516 | * entries, then we first need to move the extent records beginning | |
1517 | * at idx + ext_diff up in the list to overwrite the records being | |
1518 | * removed, then remove the extra space via kmem_realloc. | |
1519 | */ | |
1520 | void | |
1521 | xfs_iext_remove_direct( | |
1522 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1523 | xfs_extnum_t idx, /* index to begin removing exts */ | |
1524 | int ext_diff) /* number of extents to remove */ | |
1525 | { | |
1526 | xfs_extnum_t nextents; /* number of extents in file */ | |
1527 | int new_size; /* size of extents after removal */ | |
1528 | ||
1529 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
1530 | new_size = ifp->if_bytes - | |
1531 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
1532 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1533 | ||
1534 | if (new_size == 0) { | |
1535 | xfs_iext_destroy(ifp); | |
1536 | return; | |
1537 | } | |
1538 | /* Move extents up in the list (if needed) */ | |
1539 | if (idx + ext_diff < nextents) { | |
1540 | memmove(&ifp->if_u1.if_extents[idx], | |
1541 | &ifp->if_u1.if_extents[idx + ext_diff], | |
1542 | (nextents - (idx + ext_diff)) * | |
1543 | sizeof(xfs_bmbt_rec_t)); | |
1544 | } | |
1545 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
1546 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
1547 | /* | |
1548 | * Reallocate the direct extent list. If the extents | |
1549 | * will fit inside the inode then xfs_iext_realloc_direct | |
1550 | * will switch from direct to inline extent allocation | |
1551 | * mode for us. | |
1552 | */ | |
1553 | xfs_iext_realloc_direct(ifp, new_size); | |
1554 | ifp->if_bytes = new_size; | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * This is called when incore extents are being removed from the | |
1559 | * indirection array and the extents being removed span multiple extent | |
1560 | * buffers. The idx parameter contains the file extent index where we | |
1561 | * want to begin removing extents, and the count parameter contains | |
1562 | * how many extents need to be removed. | |
1563 | * | |
1564 | * |-------| |-------| | |
1565 | * | nex1 | | | nex1 - number of extents before idx | |
1566 | * |-------| | count | | |
1567 | * | | | | count - number of extents being removed at idx | |
1568 | * | count | |-------| | |
1569 | * | | | nex2 | nex2 - number of extents after idx + count | |
1570 | * |-------| |-------| | |
1571 | */ | |
1572 | void | |
1573 | xfs_iext_remove_indirect( | |
1574 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1575 | xfs_extnum_t idx, /* index to begin removing extents */ | |
1576 | int count) /* number of extents to remove */ | |
1577 | { | |
1578 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
1579 | int erp_idx = 0; /* indirection array index */ | |
1580 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
1581 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
1582 | xfs_extnum_t nex1; /* number of extents before idx */ | |
1583 | xfs_extnum_t nex2; /* extents after idx + count */ | |
1584 | int nlists; /* entries in indirection array */ | |
1585 | int page_idx = idx; /* index in target extent list */ | |
1586 | ||
1587 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1588 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
1589 | ASSERT(erp != NULL); | |
1590 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1591 | nex1 = page_idx; | |
1592 | ext_cnt = count; | |
1593 | while (ext_cnt) { | |
1594 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
1595 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
1596 | /* | |
1597 | * Check for deletion of entire list; | |
1598 | * xfs_iext_irec_remove() updates extent offsets. | |
1599 | */ | |
1600 | if (ext_diff == erp->er_extcount) { | |
1601 | xfs_iext_irec_remove(ifp, erp_idx); | |
1602 | ext_cnt -= ext_diff; | |
1603 | nex1 = 0; | |
1604 | if (ext_cnt) { | |
1605 | ASSERT(erp_idx < ifp->if_real_bytes / | |
1606 | XFS_IEXT_BUFSZ); | |
1607 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
1608 | nex1 = 0; | |
1609 | continue; | |
1610 | } else { | |
1611 | break; | |
1612 | } | |
1613 | } | |
1614 | /* Move extents up (if needed) */ | |
1615 | if (nex2) { | |
1616 | memmove(&erp->er_extbuf[nex1], | |
1617 | &erp->er_extbuf[nex1 + ext_diff], | |
1618 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
1619 | } | |
1620 | /* Zero out rest of page */ | |
1621 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
1622 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
1623 | /* Update remaining counters */ | |
1624 | erp->er_extcount -= ext_diff; | |
1625 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
1626 | ext_cnt -= ext_diff; | |
1627 | nex1 = 0; | |
1628 | erp_idx++; | |
1629 | erp++; | |
1630 | } | |
1631 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
1632 | xfs_iext_irec_compact(ifp); | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Create, destroy, or resize a linear (direct) block of extents. | |
1637 | */ | |
1638 | void | |
1639 | xfs_iext_realloc_direct( | |
1640 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1641 | int new_size) /* new size of extents */ | |
1642 | { | |
1643 | int rnew_size; /* real new size of extents */ | |
1644 | ||
1645 | rnew_size = new_size; | |
1646 | ||
1647 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || | |
1648 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
1649 | (new_size != ifp->if_real_bytes))); | |
1650 | ||
1651 | /* Free extent records */ | |
1652 | if (new_size == 0) { | |
1653 | xfs_iext_destroy(ifp); | |
1654 | } | |
1655 | /* Resize direct extent list and zero any new bytes */ | |
1656 | else if (ifp->if_real_bytes) { | |
1657 | /* Check if extents will fit inside the inode */ | |
1658 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
1659 | xfs_iext_direct_to_inline(ifp, new_size / | |
1660 | (uint)sizeof(xfs_bmbt_rec_t)); | |
1661 | ifp->if_bytes = new_size; | |
1662 | return; | |
1663 | } | |
1664 | if (!is_power_of_2(new_size)){ | |
1665 | rnew_size = roundup_pow_of_two(new_size); | |
1666 | } | |
1667 | if (rnew_size != ifp->if_real_bytes) { | |
1668 | ifp->if_u1.if_extents = | |
1669 | kmem_realloc(ifp->if_u1.if_extents, | |
1670 | rnew_size, | |
1671 | ifp->if_real_bytes, KM_NOFS); | |
1672 | } | |
1673 | if (rnew_size > ifp->if_real_bytes) { | |
1674 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
1675 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
1676 | rnew_size - ifp->if_real_bytes); | |
1677 | } | |
1678 | } | |
1679 | /* | |
1680 | * Switch from the inline extent buffer to a direct | |
1681 | * extent list. Be sure to include the inline extent | |
1682 | * bytes in new_size. | |
1683 | */ | |
1684 | else { | |
1685 | new_size += ifp->if_bytes; | |
1686 | if (!is_power_of_2(new_size)) { | |
1687 | rnew_size = roundup_pow_of_two(new_size); | |
1688 | } | |
1689 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
1690 | } | |
1691 | ifp->if_real_bytes = rnew_size; | |
1692 | ifp->if_bytes = new_size; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * Switch from linear (direct) extent records to inline buffer. | |
1697 | */ | |
1698 | void | |
1699 | xfs_iext_direct_to_inline( | |
1700 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1701 | xfs_extnum_t nextents) /* number of extents in file */ | |
1702 | { | |
1703 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
1704 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
1705 | /* | |
1706 | * The inline buffer was zeroed when we switched | |
1707 | * from inline to direct extent allocation mode, | |
1708 | * so we don't need to clear it here. | |
1709 | */ | |
1710 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
1711 | nextents * sizeof(xfs_bmbt_rec_t)); | |
1712 | kmem_free(ifp->if_u1.if_extents); | |
1713 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
1714 | ifp->if_real_bytes = 0; | |
1715 | } | |
1716 | ||
1717 | /* | |
1718 | * Switch from inline buffer to linear (direct) extent records. | |
1719 | * new_size should already be rounded up to the next power of 2 | |
1720 | * by the caller (when appropriate), so use new_size as it is. | |
1721 | * However, since new_size may be rounded up, we can't update | |
1722 | * if_bytes here. It is the caller's responsibility to update | |
1723 | * if_bytes upon return. | |
1724 | */ | |
1725 | void | |
1726 | xfs_iext_inline_to_direct( | |
1727 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1728 | int new_size) /* number of extents in file */ | |
1729 | { | |
1730 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); | |
1731 | memset(ifp->if_u1.if_extents, 0, new_size); | |
1732 | if (ifp->if_bytes) { | |
1733 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
1734 | ifp->if_bytes); | |
1735 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
1736 | sizeof(xfs_bmbt_rec_t)); | |
1737 | } | |
1738 | ifp->if_real_bytes = new_size; | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * Resize an extent indirection array to new_size bytes. | |
1743 | */ | |
1744 | void | |
1745 | xfs_iext_realloc_indirect( | |
1746 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1747 | int new_size) /* new indirection array size */ | |
1748 | { | |
1749 | int nlists; /* number of irec's (ex lists) */ | |
1750 | int size; /* current indirection array size */ | |
1751 | ||
1752 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1753 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1754 | size = nlists * sizeof(xfs_ext_irec_t); | |
1755 | ASSERT(ifp->if_real_bytes); | |
1756 | ASSERT((new_size >= 0) && (new_size != size)); | |
1757 | if (new_size == 0) { | |
1758 | xfs_iext_destroy(ifp); | |
1759 | } else { | |
1760 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
1761 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
1762 | new_size, size, KM_NOFS); | |
1763 | } | |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * Switch from indirection array to linear (direct) extent allocations. | |
1768 | */ | |
1769 | void | |
1770 | xfs_iext_indirect_to_direct( | |
1771 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
1772 | { | |
1773 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
1774 | xfs_extnum_t nextents; /* number of extents in file */ | |
1775 | int size; /* size of file extents */ | |
1776 | ||
1777 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1778 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1779 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
1780 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
1781 | ||
b194c7d8 | 1782 | xfs_iext_irec_compact_pages(ifp); |
5e656dbb BN |
1783 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); |
1784 | ||
1785 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
1786 | kmem_free(ifp->if_u1.if_ext_irec); | |
1787 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
1788 | ifp->if_u1.if_extents = ep; | |
1789 | ifp->if_bytes = size; | |
1790 | if (nextents < XFS_LINEAR_EXTS) { | |
1791 | xfs_iext_realloc_direct(ifp, size); | |
1792 | } | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Free incore file extents. | |
1797 | */ | |
1798 | void | |
1799 | xfs_iext_destroy( | |
1800 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
1801 | { | |
1802 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
1803 | int erp_idx; | |
1804 | int nlists; | |
1805 | ||
1806 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1807 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
1808 | xfs_iext_irec_remove(ifp, erp_idx); | |
1809 | } | |
1810 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
1811 | } else if (ifp->if_real_bytes) { | |
1812 | kmem_free(ifp->if_u1.if_extents); | |
1813 | } else if (ifp->if_bytes) { | |
1814 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
1815 | sizeof(xfs_bmbt_rec_t)); | |
1816 | } | |
1817 | ifp->if_u1.if_extents = NULL; | |
1818 | ifp->if_real_bytes = 0; | |
1819 | ifp->if_bytes = 0; | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * Return a pointer to the extent record for file system block bno. | |
1824 | */ | |
1825 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ | |
1826 | xfs_iext_bno_to_ext( | |
1827 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1828 | xfs_fileoff_t bno, /* block number to search for */ | |
1829 | xfs_extnum_t *idxp) /* index of target extent */ | |
1830 | { | |
1831 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ | |
1832 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ | |
1833 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ | |
1834 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
1835 | int high; /* upper boundary in search */ | |
1836 | xfs_extnum_t idx = 0; /* index of target extent */ | |
1837 | int low; /* lower boundary in search */ | |
1838 | xfs_extnum_t nextents; /* number of file extents */ | |
1839 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
1840 | ||
1841 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
1842 | if (nextents == 0) { | |
1843 | *idxp = 0; | |
1844 | return NULL; | |
1845 | } | |
1846 | low = 0; | |
1847 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
1848 | /* Find target extent list */ | |
1849 | int erp_idx = 0; | |
1850 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
1851 | base = erp->er_extbuf; | |
1852 | high = erp->er_extcount - 1; | |
1853 | } else { | |
1854 | base = ifp->if_u1.if_extents; | |
1855 | high = nextents - 1; | |
1856 | } | |
1857 | /* Binary search extent records */ | |
1858 | while (low <= high) { | |
1859 | idx = (low + high) >> 1; | |
1860 | ep = base + idx; | |
1861 | startoff = xfs_bmbt_get_startoff(ep); | |
1862 | blockcount = xfs_bmbt_get_blockcount(ep); | |
1863 | if (bno < startoff) { | |
1864 | high = idx - 1; | |
1865 | } else if (bno >= startoff + blockcount) { | |
1866 | low = idx + 1; | |
1867 | } else { | |
1868 | /* Convert back to file-based extent index */ | |
1869 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
1870 | idx += erp->er_extoff; | |
1871 | } | |
1872 | *idxp = idx; | |
1873 | return ep; | |
1874 | } | |
1875 | } | |
1876 | /* Convert back to file-based extent index */ | |
1877 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
1878 | idx += erp->er_extoff; | |
1879 | } | |
1880 | if (bno >= startoff + blockcount) { | |
1881 | if (++idx == nextents) { | |
1882 | ep = NULL; | |
1883 | } else { | |
1884 | ep = xfs_iext_get_ext(ifp, idx); | |
1885 | } | |
1886 | } | |
1887 | *idxp = idx; | |
1888 | return ep; | |
1889 | } | |
1890 | ||
1891 | /* | |
1892 | * Return a pointer to the indirection array entry containing the | |
1893 | * extent record for filesystem block bno. Store the index of the | |
1894 | * target irec in *erp_idxp. | |
1895 | */ | |
1896 | xfs_ext_irec_t * /* pointer to found extent record */ | |
1897 | xfs_iext_bno_to_irec( | |
1898 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1899 | xfs_fileoff_t bno, /* block number to search for */ | |
1900 | int *erp_idxp) /* irec index of target ext list */ | |
1901 | { | |
1902 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
1903 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
1904 | int erp_idx; /* indirection array index */ | |
1905 | int nlists; /* number of extent irec's (lists) */ | |
1906 | int high; /* binary search upper limit */ | |
1907 | int low; /* binary search lower limit */ | |
1908 | ||
1909 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1910 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1911 | erp_idx = 0; | |
1912 | low = 0; | |
1913 | high = nlists - 1; | |
1914 | while (low <= high) { | |
1915 | erp_idx = (low + high) >> 1; | |
1916 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
1917 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
1918 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
1919 | high = erp_idx - 1; | |
1920 | } else if (erp_next && bno >= | |
1921 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
1922 | low = erp_idx + 1; | |
1923 | } else { | |
1924 | break; | |
1925 | } | |
1926 | } | |
1927 | *erp_idxp = erp_idx; | |
1928 | return erp; | |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * Return a pointer to the indirection array entry containing the | |
1933 | * extent record at file extent index *idxp. Store the index of the | |
1934 | * target irec in *erp_idxp and store the page index of the target | |
1935 | * extent record in *idxp. | |
1936 | */ | |
1937 | xfs_ext_irec_t * | |
1938 | xfs_iext_idx_to_irec( | |
1939 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
1940 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
1941 | int *erp_idxp, /* pointer to target irec */ | |
1942 | int realloc) /* new bytes were just added */ | |
1943 | { | |
1944 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
1945 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
1946 | int erp_idx; /* indirection array index */ | |
1947 | int nlists; /* number of irec's (ex lists) */ | |
1948 | int high; /* binary search upper limit */ | |
1949 | int low; /* binary search lower limit */ | |
1950 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
1951 | ||
1952 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
1953 | ASSERT(page_idx >= 0 && page_idx <= | |
1954 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)); | |
1955 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
1956 | erp_idx = 0; | |
1957 | low = 0; | |
1958 | high = nlists - 1; | |
1959 | ||
1960 | /* Binary search extent irec's */ | |
1961 | while (low <= high) { | |
1962 | erp_idx = (low + high) >> 1; | |
1963 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
1964 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
1965 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
1966 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
1967 | high = erp_idx - 1; | |
1968 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
1969 | (page_idx == erp->er_extoff + erp->er_extcount && | |
1970 | !realloc)) { | |
1971 | low = erp_idx + 1; | |
1972 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
1973 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
1974 | ASSERT(realloc); | |
1975 | page_idx = 0; | |
1976 | erp_idx++; | |
1977 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
1978 | break; | |
1979 | } else { | |
1980 | page_idx -= erp->er_extoff; | |
1981 | break; | |
1982 | } | |
1983 | } | |
1984 | *idxp = page_idx; | |
1985 | *erp_idxp = erp_idx; | |
1986 | return(erp); | |
1987 | } | |
1988 | ||
1989 | /* | |
1990 | * Allocate and initialize an indirection array once the space needed | |
1991 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
1992 | */ | |
1993 | void | |
1994 | xfs_iext_irec_init( | |
1995 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
1996 | { | |
1997 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
1998 | xfs_extnum_t nextents; /* number of extents in file */ | |
1999 | ||
2000 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
2001 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2002 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
2003 | ||
2004 | erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); | |
2005 | ||
2006 | if (nextents == 0) { | |
2007 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); | |
2008 | } else if (!ifp->if_real_bytes) { | |
2009 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
2010 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
2011 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
2012 | } | |
2013 | erp->er_extbuf = ifp->if_u1.if_extents; | |
2014 | erp->er_extcount = nextents; | |
2015 | erp->er_extoff = 0; | |
2016 | ||
2017 | ifp->if_flags |= XFS_IFEXTIREC; | |
2018 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
2019 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
2020 | ifp->if_u1.if_ext_irec = erp; | |
2021 | ||
2022 | return; | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Allocate and initialize a new entry in the indirection array. | |
2027 | */ | |
2028 | xfs_ext_irec_t * | |
2029 | xfs_iext_irec_new( | |
2030 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2031 | int erp_idx) /* index for new irec */ | |
2032 | { | |
2033 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
2034 | int i; /* loop counter */ | |
2035 | int nlists; /* number of irec's (ex lists) */ | |
2036 | ||
2037 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
2038 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2039 | ||
2040 | /* Resize indirection array */ | |
2041 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
2042 | sizeof(xfs_ext_irec_t)); | |
2043 | /* | |
2044 | * Move records down in the array so the | |
2045 | * new page can use erp_idx. | |
2046 | */ | |
2047 | erp = ifp->if_u1.if_ext_irec; | |
2048 | for (i = nlists - 1; i > erp_idx; i--) { | |
2049 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
2050 | } | |
2051 | ASSERT(i == erp_idx); | |
2052 | ||
2053 | /* Initialize new extent record */ | |
2054 | erp = ifp->if_u1.if_ext_irec; | |
2055 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); | |
2056 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
2057 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
2058 | erp[erp_idx].er_extcount = 0; | |
2059 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
2060 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
2061 | return (&erp[erp_idx]); | |
2062 | } | |
2063 | ||
2064 | /* | |
2065 | * Remove a record from the indirection array. | |
2066 | */ | |
2067 | void | |
2068 | xfs_iext_irec_remove( | |
2069 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2070 | int erp_idx) /* irec index to remove */ | |
2071 | { | |
2072 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
2073 | int i; /* loop counter */ | |
2074 | int nlists; /* number of irec's (ex lists) */ | |
2075 | ||
2076 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
2077 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2078 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
2079 | if (erp->er_extbuf) { | |
2080 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
2081 | -erp->er_extcount); | |
2082 | kmem_free(erp->er_extbuf); | |
2083 | } | |
2084 | /* Compact extent records */ | |
2085 | erp = ifp->if_u1.if_ext_irec; | |
2086 | for (i = erp_idx; i < nlists - 1; i++) { | |
2087 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
2088 | } | |
2089 | /* | |
2090 | * Manually free the last extent record from the indirection | |
2091 | * array. A call to xfs_iext_realloc_indirect() with a size | |
2092 | * of zero would result in a call to xfs_iext_destroy() which | |
2093 | * would in turn call this function again, creating a nasty | |
2094 | * infinite loop. | |
2095 | */ | |
2096 | if (--nlists) { | |
2097 | xfs_iext_realloc_indirect(ifp, | |
2098 | nlists * sizeof(xfs_ext_irec_t)); | |
2099 | } else { | |
2100 | kmem_free(ifp->if_u1.if_ext_irec); | |
2101 | } | |
2102 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
2103 | } | |
2104 | ||
2105 | /* | |
2106 | * This is called to clean up large amounts of unused memory allocated | |
2107 | * by the indirection array. Before compacting anything though, verify | |
2108 | * that the indirection array is still needed and switch back to the | |
2109 | * linear extent list (or even the inline buffer) if possible. The | |
2110 | * compaction policy is as follows: | |
2111 | * | |
2112 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
b194c7d8 | 2113 | * Partial Compaction: Extents occupy less than 50% of allocated space |
5e656dbb BN |
2114 | * No Compaction: Extents occupy at least 50% of allocated space |
2115 | */ | |
2116 | void | |
2117 | xfs_iext_irec_compact( | |
2118 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
2119 | { | |
2120 | xfs_extnum_t nextents; /* number of extents in file */ | |
2121 | int nlists; /* number of irec's (ex lists) */ | |
2122 | ||
2123 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
2124 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2125 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2126 | ||
2127 | if (nextents == 0) { | |
2128 | xfs_iext_destroy(ifp); | |
2129 | } else if (nextents <= XFS_INLINE_EXTS) { | |
2130 | xfs_iext_indirect_to_direct(ifp); | |
2131 | xfs_iext_direct_to_inline(ifp, nextents); | |
2132 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
2133 | xfs_iext_indirect_to_direct(ifp); | |
5e656dbb BN |
2134 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { |
2135 | xfs_iext_irec_compact_pages(ifp); | |
2136 | } | |
2137 | } | |
2138 | ||
2139 | /* | |
2140 | * Combine extents from neighboring extent pages. | |
2141 | */ | |
2142 | void | |
2143 | xfs_iext_irec_compact_pages( | |
2144 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
2145 | { | |
2146 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
2147 | int erp_idx = 0; /* indirection array index */ | |
2148 | int nlists; /* number of irec's (ex lists) */ | |
2149 | ||
2150 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
2151 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2152 | while (erp_idx < nlists - 1) { | |
2153 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
2154 | erp_next = erp + 1; | |
2155 | if (erp_next->er_extcount <= | |
2156 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
b194c7d8 | 2157 | memcpy(&erp->er_extbuf[erp->er_extcount], |
5e656dbb BN |
2158 | erp_next->er_extbuf, erp_next->er_extcount * |
2159 | sizeof(xfs_bmbt_rec_t)); | |
2160 | erp->er_extcount += erp_next->er_extcount; | |
2161 | /* | |
2162 | * Free page before removing extent record | |
2163 | * so er_extoffs don't get modified in | |
2164 | * xfs_iext_irec_remove. | |
2165 | */ | |
2166 | kmem_free(erp_next->er_extbuf); | |
2167 | erp_next->er_extbuf = NULL; | |
2168 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
2169 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2170 | } else { | |
2171 | erp_idx++; | |
2172 | } | |
2173 | } | |
2174 | } | |
2175 | ||
5e656dbb BN |
2176 | /* |
2177 | * This is called to update the er_extoff field in the indirection | |
2178 | * array when extents have been added or removed from one of the | |
2179 | * extent lists. erp_idx contains the irec index to begin updating | |
2180 | * at and ext_diff contains the number of extents that were added | |
2181 | * or removed. | |
2182 | */ | |
2183 | void | |
2184 | xfs_iext_irec_update_extoffs( | |
2185 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2186 | int erp_idx, /* irec index to update */ | |
2187 | int ext_diff) /* number of new extents */ | |
2188 | { | |
2189 | int i; /* loop counter */ | |
2190 | int nlists; /* number of irec's (ex lists */ | |
2191 | ||
2192 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
2193 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
2194 | for (i = erp_idx; i < nlists; i++) { | |
2195 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
2196 | } | |
2bd0ea18 | 2197 | } |