1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_ag.h"
14 #include "xfs_inode.h"
15 #include "xfs_errortag.h"
16 #include "xfs_error.h"
17 #include "xfs_icache.h"
18 #include "xfs_trans.h"
19 #include "xfs_ialloc.h"
20 #include "xfs_dir2.h"
21 #include "xfs_health.h"
22
23 #include <linux/iversion.h>
24
25 /*
26 * If we are doing readahead on an inode buffer, we might be in log recovery
27 * reading an inode allocation buffer that hasn't yet been replayed, and hence
28 * has not had the inode cores stamped into it. Hence for readahead, the buffer
29 * may be potentially invalid.
30 *
31 * If the readahead buffer is invalid, we need to mark it with an error and
32 * clear the DONE status of the buffer so that a followup read will re-read it
33 * from disk. We don't report the error otherwise to avoid warnings during log
34 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
35 * because all we want to do is say readahead failed; there is no-one to report
36 * the error to, so this will distinguish it from a non-ra verifier failure.
37 * Changes to this readahead error behaviour also need to be reflected in
38 * xfs_dquot_buf_readahead_verify().
39 */
40 static void
xfs_inode_buf_verify(struct xfs_buf * bp,bool readahead)41 xfs_inode_buf_verify(
42 struct xfs_buf *bp,
43 bool readahead)
44 {
45 struct xfs_mount *mp = bp->b_mount;
46 int i;
47 int ni;
48
49 /*
50 * Validate the magic number and version of every inode in the buffer
51 */
52 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 for (i = 0; i < ni; i++) {
54 struct xfs_dinode *dip;
55 xfs_agino_t unlinked_ino;
56 int di_ok;
57
58 dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
59 unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
60 di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
61 xfs_dinode_good_version(mp, dip->di_version) &&
62 xfs_verify_agino_or_null(bp->b_pag, unlinked_ino);
63 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
64 XFS_ERRTAG_ITOBP_INOTOBP))) {
65 if (readahead) {
66 bp->b_flags &= ~XBF_DONE;
67 xfs_buf_ioerror(bp, -EIO);
68 return;
69 }
70
71 #ifdef DEBUG
72 xfs_alert(mp,
73 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
74 (unsigned long long)xfs_buf_daddr(bp), i,
75 be16_to_cpu(dip->di_magic));
76 #endif
77 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
78 __func__, dip, sizeof(*dip),
79 NULL);
80 return;
81 }
82 }
83 }
84
85
86 static void
xfs_inode_buf_read_verify(struct xfs_buf * bp)87 xfs_inode_buf_read_verify(
88 struct xfs_buf *bp)
89 {
90 xfs_inode_buf_verify(bp, false);
91 }
92
93 static void
xfs_inode_buf_readahead_verify(struct xfs_buf * bp)94 xfs_inode_buf_readahead_verify(
95 struct xfs_buf *bp)
96 {
97 xfs_inode_buf_verify(bp, true);
98 }
99
100 static void
xfs_inode_buf_write_verify(struct xfs_buf * bp)101 xfs_inode_buf_write_verify(
102 struct xfs_buf *bp)
103 {
104 xfs_inode_buf_verify(bp, false);
105 }
106
107 const struct xfs_buf_ops xfs_inode_buf_ops = {
108 .name = "xfs_inode",
109 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
110 cpu_to_be16(XFS_DINODE_MAGIC) },
111 .verify_read = xfs_inode_buf_read_verify,
112 .verify_write = xfs_inode_buf_write_verify,
113 };
114
115 const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
116 .name = "xfs_inode_ra",
117 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
118 cpu_to_be16(XFS_DINODE_MAGIC) },
119 .verify_read = xfs_inode_buf_readahead_verify,
120 .verify_write = xfs_inode_buf_write_verify,
121 };
122
123
124 /*
125 * This routine is called to map an inode to the buffer containing the on-disk
126 * version of the inode. It returns a pointer to the buffer containing the
127 * on-disk inode in the bpp parameter.
128 */
129 int
xfs_imap_to_bp(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_imap * imap,struct xfs_buf ** bpp)130 xfs_imap_to_bp(
131 struct xfs_mount *mp,
132 struct xfs_trans *tp,
133 struct xfs_imap *imap,
134 struct xfs_buf **bpp)
135 {
136 int error;
137
138 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
139 imap->im_len, XBF_UNMAPPED, bpp, &xfs_inode_buf_ops);
140 if (xfs_metadata_is_sick(error))
141 xfs_agno_mark_sick(mp, xfs_daddr_to_agno(mp, imap->im_blkno),
142 XFS_SICK_AG_INODES);
143 return error;
144 }
145
xfs_inode_decode_bigtime(uint64_t ts)146 static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
147 {
148 struct timespec64 tv;
149 uint32_t n;
150
151 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
152 tv.tv_nsec = n;
153
154 return tv;
155 }
156
157 /* Convert an ondisk timestamp to an incore timestamp. */
158 struct timespec64
xfs_inode_from_disk_ts(struct xfs_dinode * dip,const xfs_timestamp_t ts)159 xfs_inode_from_disk_ts(
160 struct xfs_dinode *dip,
161 const xfs_timestamp_t ts)
162 {
163 struct timespec64 tv;
164 struct xfs_legacy_timestamp *lts;
165
166 if (xfs_dinode_has_bigtime(dip))
167 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
168
169 lts = (struct xfs_legacy_timestamp *)&ts;
170 tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
171 tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
172
173 return tv;
174 }
175
176 int
xfs_inode_from_disk(struct xfs_inode * ip,struct xfs_dinode * from)177 xfs_inode_from_disk(
178 struct xfs_inode *ip,
179 struct xfs_dinode *from)
180 {
181 struct inode *inode = VFS_I(ip);
182 int error;
183 xfs_failaddr_t fa;
184
185 ASSERT(ip->i_cowfp == NULL);
186
187 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
188 if (fa) {
189 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
190 sizeof(*from), fa);
191 return -EFSCORRUPTED;
192 }
193
194 /*
195 * First get the permanent information that is needed to allocate an
196 * inode. If the inode is unused, mode is zero and we shouldn't mess
197 * with the uninitialized part of it.
198 */
199 if (!xfs_has_v3inodes(ip->i_mount))
200 ip->i_flushiter = be16_to_cpu(from->di_flushiter);
201 inode->i_generation = be32_to_cpu(from->di_gen);
202 inode->i_mode = be16_to_cpu(from->di_mode);
203 if (!inode->i_mode)
204 return 0;
205
206 /*
207 * Convert v1 inodes immediately to v2 inode format as this is the
208 * minimum inode version format we support in the rest of the code.
209 * They will also be unconditionally written back to disk as v2 inodes.
210 */
211 if (unlikely(from->di_version == 1)) {
212 set_nlink(inode, be16_to_cpu(from->di_onlink));
213 ip->i_projid = 0;
214 } else {
215 set_nlink(inode, be32_to_cpu(from->di_nlink));
216 ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
217 be16_to_cpu(from->di_projid_lo);
218 }
219
220 i_uid_write(inode, be32_to_cpu(from->di_uid));
221 i_gid_write(inode, be32_to_cpu(from->di_gid));
222
223 /*
224 * Time is signed, so need to convert to signed 32 bit before
225 * storing in inode timestamp which may be 64 bit. Otherwise
226 * a time before epoch is converted to a time long after epoch
227 * on 64 bit systems.
228 */
229 inode_set_atime_to_ts(inode,
230 xfs_inode_from_disk_ts(from, from->di_atime));
231 inode_set_mtime_to_ts(inode,
232 xfs_inode_from_disk_ts(from, from->di_mtime));
233 inode_set_ctime_to_ts(inode,
234 xfs_inode_from_disk_ts(from, from->di_ctime));
235
236 ip->i_disk_size = be64_to_cpu(from->di_size);
237 ip->i_nblocks = be64_to_cpu(from->di_nblocks);
238 ip->i_extsize = be32_to_cpu(from->di_extsize);
239 ip->i_forkoff = from->di_forkoff;
240 ip->i_diflags = be16_to_cpu(from->di_flags);
241 ip->i_next_unlinked = be32_to_cpu(from->di_next_unlinked);
242
243 if (from->di_dmevmask || from->di_dmstate)
244 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
245
246 if (xfs_has_v3inodes(ip->i_mount)) {
247 inode_set_iversion_queried(inode,
248 be64_to_cpu(from->di_changecount));
249 ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
250 ip->i_diflags2 = be64_to_cpu(from->di_flags2);
251 ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
252 }
253
254 error = xfs_iformat_data_fork(ip, from);
255 if (error)
256 return error;
257 if (from->di_forkoff) {
258 error = xfs_iformat_attr_fork(ip, from);
259 if (error)
260 goto out_destroy_data_fork;
261 }
262 if (xfs_is_reflink_inode(ip))
263 xfs_ifork_init_cow(ip);
264 return 0;
265
266 out_destroy_data_fork:
267 xfs_idestroy_fork(&ip->i_df);
268 return error;
269 }
270
271 /* Convert an incore timestamp to an ondisk timestamp. */
272 static inline xfs_timestamp_t
xfs_inode_to_disk_ts(struct xfs_inode * ip,const struct timespec64 tv)273 xfs_inode_to_disk_ts(
274 struct xfs_inode *ip,
275 const struct timespec64 tv)
276 {
277 struct xfs_legacy_timestamp *lts;
278 xfs_timestamp_t ts;
279
280 if (xfs_inode_has_bigtime(ip))
281 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
282
283 lts = (struct xfs_legacy_timestamp *)&ts;
284 lts->t_sec = cpu_to_be32(tv.tv_sec);
285 lts->t_nsec = cpu_to_be32(tv.tv_nsec);
286
287 return ts;
288 }
289
290 static inline void
xfs_inode_to_disk_iext_counters(struct xfs_inode * ip,struct xfs_dinode * to)291 xfs_inode_to_disk_iext_counters(
292 struct xfs_inode *ip,
293 struct xfs_dinode *to)
294 {
295 if (xfs_inode_has_large_extent_counts(ip)) {
296 to->di_big_nextents = cpu_to_be64(xfs_ifork_nextents(&ip->i_df));
297 to->di_big_anextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_af));
298 /*
299 * We might be upgrading the inode to use larger extent counters
300 * than was previously used. Hence zero the unused field.
301 */
302 to->di_nrext64_pad = cpu_to_be16(0);
303 } else {
304 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
305 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(&ip->i_af));
306 }
307 }
308
309 void
xfs_inode_to_disk(struct xfs_inode * ip,struct xfs_dinode * to,xfs_lsn_t lsn)310 xfs_inode_to_disk(
311 struct xfs_inode *ip,
312 struct xfs_dinode *to,
313 xfs_lsn_t lsn)
314 {
315 struct inode *inode = VFS_I(ip);
316
317 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
318 to->di_onlink = 0;
319
320 to->di_format = xfs_ifork_format(&ip->i_df);
321 to->di_uid = cpu_to_be32(i_uid_read(inode));
322 to->di_gid = cpu_to_be32(i_gid_read(inode));
323 to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
324 to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
325
326 to->di_atime = xfs_inode_to_disk_ts(ip, inode_get_atime(inode));
327 to->di_mtime = xfs_inode_to_disk_ts(ip, inode_get_mtime(inode));
328 to->di_ctime = xfs_inode_to_disk_ts(ip, inode_get_ctime(inode));
329 to->di_nlink = cpu_to_be32(inode->i_nlink);
330 to->di_gen = cpu_to_be32(inode->i_generation);
331 to->di_mode = cpu_to_be16(inode->i_mode);
332
333 to->di_size = cpu_to_be64(ip->i_disk_size);
334 to->di_nblocks = cpu_to_be64(ip->i_nblocks);
335 to->di_extsize = cpu_to_be32(ip->i_extsize);
336 to->di_forkoff = ip->i_forkoff;
337 to->di_aformat = xfs_ifork_format(&ip->i_af);
338 to->di_flags = cpu_to_be16(ip->i_diflags);
339
340 if (xfs_has_v3inodes(ip->i_mount)) {
341 to->di_version = 3;
342 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
343 to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
344 to->di_flags2 = cpu_to_be64(ip->i_diflags2);
345 to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
346 to->di_ino = cpu_to_be64(ip->i_ino);
347 to->di_lsn = cpu_to_be64(lsn);
348 memset(to->di_pad2, 0, sizeof(to->di_pad2));
349 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
350 to->di_v3_pad = 0;
351 } else {
352 to->di_version = 2;
353 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
354 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
355 }
356
357 xfs_inode_to_disk_iext_counters(ip, to);
358 }
359
360 static xfs_failaddr_t
xfs_dinode_verify_fork(struct xfs_dinode * dip,struct xfs_mount * mp,int whichfork)361 xfs_dinode_verify_fork(
362 struct xfs_dinode *dip,
363 struct xfs_mount *mp,
364 int whichfork)
365 {
366 xfs_extnum_t di_nextents;
367 xfs_extnum_t max_extents;
368 mode_t mode = be16_to_cpu(dip->di_mode);
369 uint32_t fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
370 uint32_t fork_format = XFS_DFORK_FORMAT(dip, whichfork);
371
372 di_nextents = xfs_dfork_nextents(dip, whichfork);
373
374 /*
375 * For fork types that can contain local data, check that the fork
376 * format matches the size of local data contained within the fork.
377 */
378 if (whichfork == XFS_DATA_FORK) {
379 /*
380 * A directory small enough to fit in the inode must be stored
381 * in local format. The directory sf <-> extents conversion
382 * code updates the directory size accordingly. Directories
383 * being truncated have zero size and are not subject to this
384 * check.
385 */
386 if (S_ISDIR(mode)) {
387 if (dip->di_size &&
388 be64_to_cpu(dip->di_size) <= fork_size &&
389 fork_format != XFS_DINODE_FMT_LOCAL)
390 return __this_address;
391 }
392
393 /*
394 * A symlink with a target small enough to fit in the inode can
395 * be stored in extents format if xattrs were added (thus
396 * converting the data fork from shortform to remote format)
397 * and then removed.
398 */
399 if (S_ISLNK(mode)) {
400 if (be64_to_cpu(dip->di_size) <= fork_size &&
401 fork_format != XFS_DINODE_FMT_EXTENTS &&
402 fork_format != XFS_DINODE_FMT_LOCAL)
403 return __this_address;
404 }
405
406 /*
407 * For all types, check that when the size says the fork should
408 * be in extent or btree format, the inode isn't claiming to be
409 * in local format.
410 */
411 if (be64_to_cpu(dip->di_size) > fork_size &&
412 fork_format == XFS_DINODE_FMT_LOCAL)
413 return __this_address;
414 }
415
416 switch (fork_format) {
417 case XFS_DINODE_FMT_LOCAL:
418 /*
419 * No local regular files yet.
420 */
421 if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
422 return __this_address;
423 if (di_nextents)
424 return __this_address;
425 break;
426 case XFS_DINODE_FMT_EXTENTS:
427 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
428 return __this_address;
429 break;
430 case XFS_DINODE_FMT_BTREE:
431 max_extents = xfs_iext_max_nextents(
432 xfs_dinode_has_large_extent_counts(dip),
433 whichfork);
434 if (di_nextents > max_extents)
435 return __this_address;
436 break;
437 default:
438 return __this_address;
439 }
440 return NULL;
441 }
442
443 static xfs_failaddr_t
xfs_dinode_verify_forkoff(struct xfs_dinode * dip,struct xfs_mount * mp)444 xfs_dinode_verify_forkoff(
445 struct xfs_dinode *dip,
446 struct xfs_mount *mp)
447 {
448 if (!dip->di_forkoff)
449 return NULL;
450
451 switch (dip->di_format) {
452 case XFS_DINODE_FMT_DEV:
453 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
454 return __this_address;
455 break;
456 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
457 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
458 case XFS_DINODE_FMT_BTREE:
459 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
460 return __this_address;
461 break;
462 default:
463 return __this_address;
464 }
465 return NULL;
466 }
467
468 static xfs_failaddr_t
xfs_dinode_verify_nrext64(struct xfs_mount * mp,struct xfs_dinode * dip)469 xfs_dinode_verify_nrext64(
470 struct xfs_mount *mp,
471 struct xfs_dinode *dip)
472 {
473 if (xfs_dinode_has_large_extent_counts(dip)) {
474 if (!xfs_has_large_extent_counts(mp))
475 return __this_address;
476 if (dip->di_nrext64_pad != 0)
477 return __this_address;
478 } else if (dip->di_version >= 3) {
479 if (dip->di_v3_pad != 0)
480 return __this_address;
481 }
482
483 return NULL;
484 }
485
486 xfs_failaddr_t
xfs_dinode_verify(struct xfs_mount * mp,xfs_ino_t ino,struct xfs_dinode * dip)487 xfs_dinode_verify(
488 struct xfs_mount *mp,
489 xfs_ino_t ino,
490 struct xfs_dinode *dip)
491 {
492 xfs_failaddr_t fa;
493 uint16_t mode;
494 uint16_t flags;
495 uint64_t flags2;
496 uint64_t di_size;
497 xfs_extnum_t nextents;
498 xfs_extnum_t naextents;
499 xfs_filblks_t nblocks;
500
501 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
502 return __this_address;
503
504 /* Verify v3 integrity information first */
505 if (dip->di_version >= 3) {
506 if (!xfs_has_v3inodes(mp))
507 return __this_address;
508 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
509 XFS_DINODE_CRC_OFF))
510 return __this_address;
511 if (be64_to_cpu(dip->di_ino) != ino)
512 return __this_address;
513 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
514 return __this_address;
515 }
516
517 /*
518 * Historical note: xfsprogs in the 3.2 era set up its incore inodes to
519 * have di_nlink track the link count, even if the actual filesystem
520 * only supported V1 inodes (i.e. di_onlink). When writing out the
521 * ondisk inode, it would set both the ondisk di_nlink and di_onlink to
522 * the the incore di_nlink value, which is why we cannot check for
523 * di_nlink==0 on a V1 inode. V2/3 inodes would get written out with
524 * di_onlink==0, so we can check that.
525 */
526 if (dip->di_version >= 2) {
527 if (dip->di_onlink)
528 return __this_address;
529 }
530
531 /* don't allow invalid i_size */
532 di_size = be64_to_cpu(dip->di_size);
533 if (di_size & (1ULL << 63))
534 return __this_address;
535
536 mode = be16_to_cpu(dip->di_mode);
537 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
538 return __this_address;
539
540 /*
541 * No zero-length symlinks/dirs unless they're unlinked and hence being
542 * inactivated.
543 */
544 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0) {
545 if (dip->di_version > 1) {
546 if (dip->di_nlink)
547 return __this_address;
548 } else {
549 if (dip->di_onlink)
550 return __this_address;
551 }
552 }
553
554 fa = xfs_dinode_verify_nrext64(mp, dip);
555 if (fa)
556 return fa;
557
558 nextents = xfs_dfork_data_extents(dip);
559 naextents = xfs_dfork_attr_extents(dip);
560 nblocks = be64_to_cpu(dip->di_nblocks);
561
562 /* Fork checks carried over from xfs_iformat_fork */
563 if (mode && nextents + naextents > nblocks)
564 return __this_address;
565
566 if (nextents + naextents == 0 && nblocks != 0)
567 return __this_address;
568
569 if (S_ISDIR(mode) && nextents > mp->m_dir_geo->max_extents)
570 return __this_address;
571
572 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
573 return __this_address;
574
575 flags = be16_to_cpu(dip->di_flags);
576
577 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
578 return __this_address;
579
580 /* check for illegal values of forkoff */
581 fa = xfs_dinode_verify_forkoff(dip, mp);
582 if (fa)
583 return fa;
584
585 /* Do we have appropriate data fork formats for the mode? */
586 switch (mode & S_IFMT) {
587 case S_IFIFO:
588 case S_IFCHR:
589 case S_IFBLK:
590 case S_IFSOCK:
591 if (dip->di_format != XFS_DINODE_FMT_DEV)
592 return __this_address;
593 break;
594 case S_IFREG:
595 case S_IFLNK:
596 case S_IFDIR:
597 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
598 if (fa)
599 return fa;
600 break;
601 case 0:
602 /* Uninitialized inode ok. */
603 break;
604 default:
605 return __this_address;
606 }
607
608 if (dip->di_forkoff) {
609 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
610 if (fa)
611 return fa;
612 } else {
613 /*
614 * If there is no fork offset, this may be a freshly-made inode
615 * in a new disk cluster, in which case di_aformat is zeroed.
616 * Otherwise, such an inode must be in EXTENTS format; this goes
617 * for freed inodes as well.
618 */
619 switch (dip->di_aformat) {
620 case 0:
621 case XFS_DINODE_FMT_EXTENTS:
622 break;
623 default:
624 return __this_address;
625 }
626 if (naextents)
627 return __this_address;
628 }
629
630 /* extent size hint validation */
631 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
632 mode, flags);
633 if (fa)
634 return fa;
635
636 /* only version 3 or greater inodes are extensively verified here */
637 if (dip->di_version < 3)
638 return NULL;
639
640 flags2 = be64_to_cpu(dip->di_flags2);
641
642 /* don't allow reflink/cowextsize if we don't have reflink */
643 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
644 !xfs_has_reflink(mp))
645 return __this_address;
646
647 /* only regular files get reflink */
648 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
649 return __this_address;
650
651 /* don't let reflink and realtime mix */
652 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
653 return __this_address;
654
655 /* COW extent size hint validation */
656 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
657 mode, flags, flags2);
658 if (fa)
659 return fa;
660
661 /* bigtime iflag can only happen on bigtime filesystems */
662 if (xfs_dinode_has_bigtime(dip) &&
663 !xfs_has_bigtime(mp))
664 return __this_address;
665
666 return NULL;
667 }
668
669 void
xfs_dinode_calc_crc(struct xfs_mount * mp,struct xfs_dinode * dip)670 xfs_dinode_calc_crc(
671 struct xfs_mount *mp,
672 struct xfs_dinode *dip)
673 {
674 uint32_t crc;
675
676 if (dip->di_version < 3)
677 return;
678
679 ASSERT(xfs_has_crc(mp));
680 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
681 XFS_DINODE_CRC_OFF);
682 dip->di_crc = xfs_end_cksum(crc);
683 }
684
685 /*
686 * Validate di_extsize hint.
687 *
688 * 1. Extent size hint is only valid for directories and regular files.
689 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
690 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
691 * 4. Hint cannot be larger than MAXTEXTLEN.
692 * 5. Can be changed on directories at any time.
693 * 6. Hint value of 0 turns off hints, clears inode flags.
694 * 7. Extent size must be a multiple of the appropriate block size.
695 * For realtime files, this is the rt extent size.
696 * 8. For non-realtime files, the extent size hint must be limited
697 * to half the AG size to avoid alignment extending the extent beyond the
698 * limits of the AG.
699 */
700 xfs_failaddr_t
xfs_inode_validate_extsize(struct xfs_mount * mp,uint32_t extsize,uint16_t mode,uint16_t flags)701 xfs_inode_validate_extsize(
702 struct xfs_mount *mp,
703 uint32_t extsize,
704 uint16_t mode,
705 uint16_t flags)
706 {
707 bool rt_flag;
708 bool hint_flag;
709 bool inherit_flag;
710 uint32_t extsize_bytes;
711 uint32_t blocksize_bytes;
712
713 rt_flag = (flags & XFS_DIFLAG_REALTIME);
714 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
715 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
716 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
717
718 /*
719 * This comment describes a historic gap in this verifier function.
720 *
721 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
722 * function has never checked that the extent size hint is an integer
723 * multiple of the realtime extent size. Since we allow users to set
724 * this combination on non-rt filesystems /and/ to change the rt
725 * extent size when adding a rt device to a filesystem, the net effect
726 * is that users can configure a filesystem anticipating one rt
727 * geometry and change their minds later. Directories do not use the
728 * extent size hint, so this is harmless for them.
729 *
730 * If a directory with a misaligned extent size hint is allowed to
731 * propagate that hint into a new regular realtime file, the result
732 * is that the inode cluster buffer verifier will trigger a corruption
733 * shutdown the next time it is run, because the verifier has always
734 * enforced the alignment rule for regular files.
735 *
736 * Because we allow administrators to set a new rt extent size when
737 * adding a rt section, we cannot add a check to this verifier because
738 * that will result a new source of directory corruption errors when
739 * reading an existing filesystem. Instead, we rely on callers to
740 * decide when alignment checks are appropriate, and fix things up as
741 * needed.
742 */
743
744 if (rt_flag)
745 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
746 else
747 blocksize_bytes = mp->m_sb.sb_blocksize;
748
749 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
750 return __this_address;
751
752 if (hint_flag && !S_ISREG(mode))
753 return __this_address;
754
755 if (inherit_flag && !S_ISDIR(mode))
756 return __this_address;
757
758 if ((hint_flag || inherit_flag) && extsize == 0)
759 return __this_address;
760
761 /* free inodes get flags set to zero but extsize remains */
762 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
763 return __this_address;
764
765 if (extsize_bytes % blocksize_bytes)
766 return __this_address;
767
768 if (extsize > XFS_MAX_BMBT_EXTLEN)
769 return __this_address;
770
771 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
772 return __this_address;
773
774 return NULL;
775 }
776
777 /*
778 * Validate di_cowextsize hint.
779 *
780 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
781 * The inode does not have to have any shared blocks, but it must be a v3.
782 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
783 * for a directory, the hint is propagated to new files.
784 * 3. Can be changed on files & directories at any time.
785 * 4. Hint value of 0 turns off hints, clears inode flags.
786 * 5. Extent size must be a multiple of the appropriate block size.
787 * 6. The extent size hint must be limited to half the AG size to avoid
788 * alignment extending the extent beyond the limits of the AG.
789 */
790 xfs_failaddr_t
xfs_inode_validate_cowextsize(struct xfs_mount * mp,uint32_t cowextsize,uint16_t mode,uint16_t flags,uint64_t flags2)791 xfs_inode_validate_cowextsize(
792 struct xfs_mount *mp,
793 uint32_t cowextsize,
794 uint16_t mode,
795 uint16_t flags,
796 uint64_t flags2)
797 {
798 bool rt_flag;
799 bool hint_flag;
800 uint32_t cowextsize_bytes;
801
802 rt_flag = (flags & XFS_DIFLAG_REALTIME);
803 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
804 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
805
806 if (hint_flag && !xfs_has_reflink(mp))
807 return __this_address;
808
809 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
810 return __this_address;
811
812 if (hint_flag && cowextsize == 0)
813 return __this_address;
814
815 /* free inodes get flags set to zero but cowextsize remains */
816 if (mode && !hint_flag && cowextsize != 0)
817 return __this_address;
818
819 if (hint_flag && rt_flag)
820 return __this_address;
821
822 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
823 return __this_address;
824
825 if (cowextsize > XFS_MAX_BMBT_EXTLEN)
826 return __this_address;
827
828 if (cowextsize > mp->m_sb.sb_agblocks / 2)
829 return __this_address;
830
831 return NULL;
832 }
833