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