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_inode.h"
14#include "xfs_errortag.h"
15#include "xfs_error.h"
16#include "xfs_icache.h"
17#include "xfs_trans.h"
18#include "xfs_ialloc.h"
19#include "xfs_dir2.h"
20
21#include <linux/iversion.h>
22
23/*
24 * If we are doing readahead on an inode buffer, we might be in log recovery
25 * reading an inode allocation buffer that hasn't yet been replayed, and hence
26 * has not had the inode cores stamped into it. Hence for readahead, the buffer
27 * may be potentially invalid.
28 *
29 * If the readahead buffer is invalid, we need to mark it with an error and
30 * clear the DONE status of the buffer so that a followup read will re-read it
31 * from disk. We don't report the error otherwise to avoid warnings during log
32 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
33 * because all we want to do is say readahead failed; there is no-one to report
34 * the error to, so this will distinguish it from a non-ra verifier failure.
35 * Changes to this readahead error behaviour also need to be reflected in
36 * xfs_dquot_buf_readahead_verify().
37 */
38static void
39xfs_inode_buf_verify(
40 struct xfs_buf *bp,
41 bool readahead)
42{
43 struct xfs_mount *mp = bp->b_mount;
44 xfs_agnumber_t agno;
45 int i;
46 int ni;
47
48 /*
49 * Validate the magic number and version of every inode in the buffer
50 */
51 agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
52 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 for (i = 0; i < ni; i++) {
54 int di_ok;
55 xfs_dinode_t *dip;
56 xfs_agino_t unlinked_ino;
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->m_sb, dip->di_version) &&
62 xfs_verify_agino_or_null(mp, agno, 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)bp->b_bn, 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 return xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
137 imap->im_len, XBF_UNMAPPED, bpp,
138 &xfs_inode_buf_ops);
139}
140
141static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
142{
143 struct timespec64 tv;
144 uint32_t n;
145
146 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
147 tv.tv_nsec = n;
148
149 return tv;
150}
151
152/* Convert an ondisk timestamp to an incore timestamp. */
153struct timespec64
154xfs_inode_from_disk_ts(
155 struct xfs_dinode *dip,
156 const xfs_timestamp_t ts)
157{
158 struct timespec64 tv;
159 struct xfs_legacy_timestamp *lts;
160
161 if (xfs_dinode_has_bigtime(dip))
162 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
163
164 lts = (struct xfs_legacy_timestamp *)&ts;
165 tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
166 tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
167
168 return tv;
169}
170
171int
172xfs_inode_from_disk(
173 struct xfs_inode *ip,
174 struct xfs_dinode *from)
175{
176 struct inode *inode = VFS_I(ip);
177 int error;
178 xfs_failaddr_t fa;
179
180 ASSERT(ip->i_cowfp == NULL);
181 ASSERT(ip->i_afp == NULL);
182
183 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
184 if (fa) {
185 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
186 sizeof(*from), fa);
187 return -EFSCORRUPTED;
188 }
189
190 /*
191 * First get the permanent information that is needed to allocate an
192 * inode. If the inode is unused, mode is zero and we shouldn't mess
193 * with the uninitialized part of it.
194 */
195 if (!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb))
196 ip->i_flushiter = be16_to_cpu(from->di_flushiter);
197 inode->i_generation = be32_to_cpu(from->di_gen);
198 inode->i_mode = be16_to_cpu(from->di_mode);
199 if (!inode->i_mode)
200 return 0;
201
202 /*
203 * Convert v1 inodes immediately to v2 inode format as this is the
204 * minimum inode version format we support in the rest of the code.
205 * They will also be unconditionally written back to disk as v2 inodes.
206 */
207 if (unlikely(from->di_version == 1)) {
208 set_nlink(inode, be16_to_cpu(from->di_onlink));
209 ip->i_projid = 0;
210 } else {
211 set_nlink(inode, be32_to_cpu(from->di_nlink));
212 ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
213 be16_to_cpu(from->di_projid_lo);
214 }
215
216 i_uid_write(inode, be32_to_cpu(from->di_uid));
217 i_gid_write(inode, be32_to_cpu(from->di_gid));
218
219 /*
220 * Time is signed, so need to convert to signed 32 bit before
221 * storing in inode timestamp which may be 64 bit. Otherwise
222 * a time before epoch is converted to a time long after epoch
223 * on 64 bit systems.
224 */
225 inode->i_atime = xfs_inode_from_disk_ts(from, from->di_atime);
226 inode->i_mtime = xfs_inode_from_disk_ts(from, from->di_mtime);
227 inode->i_ctime = xfs_inode_from_disk_ts(from, from->di_ctime);
228
229 ip->i_disk_size = be64_to_cpu(from->di_size);
230 ip->i_nblocks = be64_to_cpu(from->di_nblocks);
231 ip->i_extsize = be32_to_cpu(from->di_extsize);
232 ip->i_forkoff = from->di_forkoff;
233 ip->i_diflags = be16_to_cpu(from->di_flags);
234
235 if (from->di_dmevmask || from->di_dmstate)
236 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
237
238 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
239 inode_set_iversion_queried(inode,
240 be64_to_cpu(from->di_changecount));
241 ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
242 ip->i_diflags2 = be64_to_cpu(from->di_flags2);
243 ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
244 }
245
246 error = xfs_iformat_data_fork(ip, from);
247 if (error)
248 return error;
249 if (from->di_forkoff) {
250 error = xfs_iformat_attr_fork(ip, from);
251 if (error)
252 goto out_destroy_data_fork;
253 }
254 if (xfs_is_reflink_inode(ip))
255 xfs_ifork_init_cow(ip);
256 return 0;
257
258out_destroy_data_fork:
259 xfs_idestroy_fork(&ip->i_df);
260 return error;
261}
262
263/* Convert an incore timestamp to an ondisk timestamp. */
264static inline xfs_timestamp_t
265xfs_inode_to_disk_ts(
266 struct xfs_inode *ip,
267 const struct timespec64 tv)
268{
269 struct xfs_legacy_timestamp *lts;
270 xfs_timestamp_t ts;
271
272 if (xfs_inode_has_bigtime(ip))
273 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
274
275 lts = (struct xfs_legacy_timestamp *)&ts;
276 lts->t_sec = cpu_to_be32(tv.tv_sec);
277 lts->t_nsec = cpu_to_be32(tv.tv_nsec);
278
279 return ts;
280}
281
282void
283xfs_inode_to_disk(
284 struct xfs_inode *ip,
285 struct xfs_dinode *to,
286 xfs_lsn_t lsn)
287{
288 struct inode *inode = VFS_I(ip);
289
290 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
291 to->di_onlink = 0;
292
293 to->di_format = xfs_ifork_format(&ip->i_df);
294 to->di_uid = cpu_to_be32(i_uid_read(inode));
295 to->di_gid = cpu_to_be32(i_gid_read(inode));
296 to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
297 to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
298
299 memset(to->di_pad, 0, sizeof(to->di_pad));
300 to->di_atime = xfs_inode_to_disk_ts(ip, inode->i_atime);
301 to->di_mtime = xfs_inode_to_disk_ts(ip, inode->i_mtime);
302 to->di_ctime = xfs_inode_to_disk_ts(ip, inode->i_ctime);
303 to->di_nlink = cpu_to_be32(inode->i_nlink);
304 to->di_gen = cpu_to_be32(inode->i_generation);
305 to->di_mode = cpu_to_be16(inode->i_mode);
306
307 to->di_size = cpu_to_be64(ip->i_disk_size);
308 to->di_nblocks = cpu_to_be64(ip->i_nblocks);
309 to->di_extsize = cpu_to_be32(ip->i_extsize);
310 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
311 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(ip->i_afp));
312 to->di_forkoff = ip->i_forkoff;
313 to->di_aformat = xfs_ifork_format(ip->i_afp);
314 to->di_flags = cpu_to_be16(ip->i_diflags);
315
316 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
317 to->di_version = 3;
318 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
319 to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
320 to->di_flags2 = cpu_to_be64(ip->i_diflags2);
321 to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
322 to->di_ino = cpu_to_be64(ip->i_ino);
323 to->di_lsn = cpu_to_be64(lsn);
324 memset(to->di_pad2, 0, sizeof(to->di_pad2));
325 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
326 to->di_flushiter = 0;
327 } else {
328 to->di_version = 2;
329 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
330 }
331}
332
333static xfs_failaddr_t
334xfs_dinode_verify_fork(
335 struct xfs_dinode *dip,
336 struct xfs_mount *mp,
337 int whichfork)
338{
339 uint32_t di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
340
341 switch (XFS_DFORK_FORMAT(dip, whichfork)) {
342 case XFS_DINODE_FMT_LOCAL:
343 /*
344 * no local regular files yet
345 */
346 if (whichfork == XFS_DATA_FORK) {
347 if (S_ISREG(be16_to_cpu(dip->di_mode)))
348 return __this_address;
349 if (be64_to_cpu(dip->di_size) >
350 XFS_DFORK_SIZE(dip, mp, whichfork))
351 return __this_address;
352 }
353 if (di_nextents)
354 return __this_address;
355 break;
356 case XFS_DINODE_FMT_EXTENTS:
357 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
358 return __this_address;
359 break;
360 case XFS_DINODE_FMT_BTREE:
361 if (whichfork == XFS_ATTR_FORK) {
362 if (di_nextents > MAXAEXTNUM)
363 return __this_address;
364 } else if (di_nextents > MAXEXTNUM) {
365 return __this_address;
366 }
367 break;
368 default:
369 return __this_address;
370 }
371 return NULL;
372}
373
374static xfs_failaddr_t
375xfs_dinode_verify_forkoff(
376 struct xfs_dinode *dip,
377 struct xfs_mount *mp)
378{
379 if (!dip->di_forkoff)
380 return NULL;
381
382 switch (dip->di_format) {
383 case XFS_DINODE_FMT_DEV:
384 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
385 return __this_address;
386 break;
387 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
388 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
389 case XFS_DINODE_FMT_BTREE:
390 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
391 return __this_address;
392 break;
393 default:
394 return __this_address;
395 }
396 return NULL;
397}
398
399xfs_failaddr_t
400xfs_dinode_verify(
401 struct xfs_mount *mp,
402 xfs_ino_t ino,
403 struct xfs_dinode *dip)
404{
405 xfs_failaddr_t fa;
406 uint16_t mode;
407 uint16_t flags;
408 uint64_t flags2;
409 uint64_t di_size;
410
411 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
412 return __this_address;
413
414 /* Verify v3 integrity information first */
415 if (dip->di_version >= 3) {
416 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
417 return __this_address;
418 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
419 XFS_DINODE_CRC_OFF))
420 return __this_address;
421 if (be64_to_cpu(dip->di_ino) != ino)
422 return __this_address;
423 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
424 return __this_address;
425 }
426
427 /* don't allow invalid i_size */
428 di_size = be64_to_cpu(dip->di_size);
429 if (di_size & (1ULL << 63))
430 return __this_address;
431
432 mode = be16_to_cpu(dip->di_mode);
433 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
434 return __this_address;
435
436 /* No zero-length symlinks/dirs. */
437 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
438 return __this_address;
439
440 /* Fork checks carried over from xfs_iformat_fork */
441 if (mode &&
442 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
443 be64_to_cpu(dip->di_nblocks))
444 return __this_address;
445
446 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
447 return __this_address;
448
449 flags = be16_to_cpu(dip->di_flags);
450
451 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
452 return __this_address;
453
454 /* check for illegal values of forkoff */
455 fa = xfs_dinode_verify_forkoff(dip, mp);
456 if (fa)
457 return fa;
458
459 /* Do we have appropriate data fork formats for the mode? */
460 switch (mode & S_IFMT) {
461 case S_IFIFO:
462 case S_IFCHR:
463 case S_IFBLK:
464 case S_IFSOCK:
465 if (dip->di_format != XFS_DINODE_FMT_DEV)
466 return __this_address;
467 break;
468 case S_IFREG:
469 case S_IFLNK:
470 case S_IFDIR:
471 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
472 if (fa)
473 return fa;
474 break;
475 case 0:
476 /* Uninitialized inode ok. */
477 break;
478 default:
479 return __this_address;
480 }
481
482 if (dip->di_forkoff) {
483 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
484 if (fa)
485 return fa;
486 } else {
487 /*
488 * If there is no fork offset, this may be a freshly-made inode
489 * in a new disk cluster, in which case di_aformat is zeroed.
490 * Otherwise, such an inode must be in EXTENTS format; this goes
491 * for freed inodes as well.
492 */
493 switch (dip->di_aformat) {
494 case 0:
495 case XFS_DINODE_FMT_EXTENTS:
496 break;
497 default:
498 return __this_address;
499 }
500 if (dip->di_anextents)
501 return __this_address;
502 }
503
504 /* extent size hint validation */
505 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
506 mode, flags);
507 if (fa)
508 return fa;
509
510 /* only version 3 or greater inodes are extensively verified here */
511 if (dip->di_version < 3)
512 return NULL;
513
514 flags2 = be64_to_cpu(dip->di_flags2);
515
516 /* don't allow reflink/cowextsize if we don't have reflink */
517 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
518 !xfs_sb_version_hasreflink(&mp->m_sb))
519 return __this_address;
520
521 /* only regular files get reflink */
522 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
523 return __this_address;
524
525 /* don't let reflink and realtime mix */
526 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
527 return __this_address;
528
529 /* COW extent size hint validation */
530 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
531 mode, flags, flags2);
532 if (fa)
533 return fa;
534
535 /* bigtime iflag can only happen on bigtime filesystems */
536 if (xfs_dinode_has_bigtime(dip) &&
537 !xfs_sb_version_hasbigtime(&mp->m_sb))
538 return __this_address;
539
540 return NULL;
541}
542
543void
544xfs_dinode_calc_crc(
545 struct xfs_mount *mp,
546 struct xfs_dinode *dip)
547{
548 uint32_t crc;
549
550 if (dip->di_version < 3)
551 return;
552
553 ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
554 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
555 XFS_DINODE_CRC_OFF);
556 dip->di_crc = xfs_end_cksum(crc);
557}
558
559/*
560 * Validate di_extsize hint.
561 *
562 * 1. Extent size hint is only valid for directories and regular files.
563 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
564 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
565 * 4. Hint cannot be larger than MAXTEXTLEN.
566 * 5. Can be changed on directories at any time.
567 * 6. Hint value of 0 turns off hints, clears inode flags.
568 * 7. Extent size must be a multiple of the appropriate block size.
569 * For realtime files, this is the rt extent size.
570 * 8. For non-realtime files, the extent size hint must be limited
571 * to half the AG size to avoid alignment extending the extent beyond the
572 * limits of the AG.
573 */
574xfs_failaddr_t
575xfs_inode_validate_extsize(
576 struct xfs_mount *mp,
577 uint32_t extsize,
578 uint16_t mode,
579 uint16_t flags)
580{
581 bool rt_flag;
582 bool hint_flag;
583 bool inherit_flag;
584 uint32_t extsize_bytes;
585 uint32_t blocksize_bytes;
586
587 rt_flag = (flags & XFS_DIFLAG_REALTIME);
588 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
589 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
590 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
591
592 /*
593 * This comment describes a historic gap in this verifier function.
594 *
595 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
596 * function has never checked that the extent size hint is an integer
597 * multiple of the realtime extent size. Since we allow users to set
598 * this combination on non-rt filesystems /and/ to change the rt
599 * extent size when adding a rt device to a filesystem, the net effect
600 * is that users can configure a filesystem anticipating one rt
601 * geometry and change their minds later. Directories do not use the
602 * extent size hint, so this is harmless for them.
603 *
604 * If a directory with a misaligned extent size hint is allowed to
605 * propagate that hint into a new regular realtime file, the result
606 * is that the inode cluster buffer verifier will trigger a corruption
607 * shutdown the next time it is run, because the verifier has always
608 * enforced the alignment rule for regular files.
609 *
610 * Because we allow administrators to set a new rt extent size when
611 * adding a rt section, we cannot add a check to this verifier because
612 * that will result a new source of directory corruption errors when
613 * reading an existing filesystem. Instead, we rely on callers to
614 * decide when alignment checks are appropriate, and fix things up as
615 * needed.
616 */
617
618 if (rt_flag)
619 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
620 else
621 blocksize_bytes = mp->m_sb.sb_blocksize;
622
623 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
624 return __this_address;
625
626 if (hint_flag && !S_ISREG(mode))
627 return __this_address;
628
629 if (inherit_flag && !S_ISDIR(mode))
630 return __this_address;
631
632 if ((hint_flag || inherit_flag) && extsize == 0)
633 return __this_address;
634
635 /* free inodes get flags set to zero but extsize remains */
636 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
637 return __this_address;
638
639 if (extsize_bytes % blocksize_bytes)
640 return __this_address;
641
642 if (extsize > MAXEXTLEN)
643 return __this_address;
644
645 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
646 return __this_address;
647
648 return NULL;
649}
650
651/*
652 * Validate di_cowextsize hint.
653 *
654 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
655 * The inode does not have to have any shared blocks, but it must be a v3.
656 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
657 * for a directory, the hint is propagated to new files.
658 * 3. Can be changed on files & directories at any time.
659 * 4. Hint value of 0 turns off hints, clears inode flags.
660 * 5. Extent size must be a multiple of the appropriate block size.
661 * 6. The extent size hint must be limited to half the AG size to avoid
662 * alignment extending the extent beyond the limits of the AG.
663 */
664xfs_failaddr_t
665xfs_inode_validate_cowextsize(
666 struct xfs_mount *mp,
667 uint32_t cowextsize,
668 uint16_t mode,
669 uint16_t flags,
670 uint64_t flags2)
671{
672 bool rt_flag;
673 bool hint_flag;
674 uint32_t cowextsize_bytes;
675
676 rt_flag = (flags & XFS_DIFLAG_REALTIME);
677 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
678 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
679
680 if (hint_flag && !xfs_sb_version_hasreflink(&mp->m_sb))
681 return __this_address;
682
683 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
684 return __this_address;
685
686 if (hint_flag && cowextsize == 0)
687 return __this_address;
688
689 /* free inodes get flags set to zero but cowextsize remains */
690 if (mode && !hint_flag && cowextsize != 0)
691 return __this_address;
692
693 if (hint_flag && rt_flag)
694 return __this_address;
695
696 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
697 return __this_address;
698
699 if (cowextsize > MAXEXTLEN)
700 return __this_address;
701
702 if (cowextsize > mp->m_sb.sb_agblocks / 2)
703 return __this_address;
704
705 return NULL;
706}
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_inode.h"
14#include "xfs_errortag.h"
15#include "xfs_error.h"
16#include "xfs_icache.h"
17#include "xfs_trans.h"
18#include "xfs_ialloc.h"
19#include "xfs_dir2.h"
20
21#include <linux/iversion.h>
22
23/*
24 * If we are doing readahead on an inode buffer, we might be in log recovery
25 * reading an inode allocation buffer that hasn't yet been replayed, and hence
26 * has not had the inode cores stamped into it. Hence for readahead, the buffer
27 * may be potentially invalid.
28 *
29 * If the readahead buffer is invalid, we need to mark it with an error and
30 * clear the DONE status of the buffer so that a followup read will re-read it
31 * from disk. We don't report the error otherwise to avoid warnings during log
32 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
33 * because all we want to do is say readahead failed; there is no-one to report
34 * the error to, so this will distinguish it from a non-ra verifier failure.
35 * Changes to this readahead error behaviour also need to be reflected in
36 * xfs_dquot_buf_readahead_verify().
37 */
38static void
39xfs_inode_buf_verify(
40 struct xfs_buf *bp,
41 bool readahead)
42{
43 struct xfs_mount *mp = bp->b_mount;
44 xfs_agnumber_t agno;
45 int i;
46 int ni;
47
48 /*
49 * Validate the magic number and version of every inode in the buffer
50 */
51 agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
52 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 for (i = 0; i < ni; i++) {
54 int di_ok;
55 xfs_dinode_t *dip;
56 xfs_agino_t unlinked_ino;
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->m_sb, dip->di_version) &&
62 xfs_verify_agino_or_null(mp, agno, 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)bp->b_bn, 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, and in the dipp parameter it returns a
128 * pointer to the on-disk inode within that buffer.
129 *
130 * If a non-zero error is returned, then the contents of bpp and dipp are
131 * undefined.
132 */
133int
134xfs_imap_to_bp(
135 struct xfs_mount *mp,
136 struct xfs_trans *tp,
137 struct xfs_imap *imap,
138 struct xfs_dinode **dipp,
139 struct xfs_buf **bpp,
140 uint buf_flags)
141{
142 struct xfs_buf *bp;
143 int error;
144
145 buf_flags |= XBF_UNMAPPED;
146 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
147 (int)imap->im_len, buf_flags, &bp,
148 &xfs_inode_buf_ops);
149 if (error) {
150 ASSERT(error != -EAGAIN || (buf_flags & XBF_TRYLOCK));
151 return error;
152 }
153
154 *bpp = bp;
155 if (dipp)
156 *dipp = xfs_buf_offset(bp, imap->im_boffset);
157 return 0;
158}
159
160int
161xfs_inode_from_disk(
162 struct xfs_inode *ip,
163 struct xfs_dinode *from)
164{
165 struct xfs_icdinode *to = &ip->i_d;
166 struct inode *inode = VFS_I(ip);
167 int error;
168 xfs_failaddr_t fa;
169
170 ASSERT(ip->i_cowfp == NULL);
171 ASSERT(ip->i_afp == NULL);
172
173 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
174 if (fa) {
175 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
176 sizeof(*from), fa);
177 return -EFSCORRUPTED;
178 }
179
180 /*
181 * First get the permanent information that is needed to allocate an
182 * inode. If the inode is unused, mode is zero and we shouldn't mess
183 * with the uninitialized part of it.
184 */
185 to->di_flushiter = be16_to_cpu(from->di_flushiter);
186 inode->i_generation = be32_to_cpu(from->di_gen);
187 inode->i_mode = be16_to_cpu(from->di_mode);
188 if (!inode->i_mode)
189 return 0;
190
191 /*
192 * Convert v1 inodes immediately to v2 inode format as this is the
193 * minimum inode version format we support in the rest of the code.
194 * They will also be unconditionally written back to disk as v2 inodes.
195 */
196 if (unlikely(from->di_version == 1)) {
197 set_nlink(inode, be16_to_cpu(from->di_onlink));
198 to->di_projid = 0;
199 } else {
200 set_nlink(inode, be32_to_cpu(from->di_nlink));
201 to->di_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
202 be16_to_cpu(from->di_projid_lo);
203 }
204
205 i_uid_write(inode, be32_to_cpu(from->di_uid));
206 i_gid_write(inode, be32_to_cpu(from->di_gid));
207
208 /*
209 * Time is signed, so need to convert to signed 32 bit before
210 * storing in inode timestamp which may be 64 bit. Otherwise
211 * a time before epoch is converted to a time long after epoch
212 * on 64 bit systems.
213 */
214 inode->i_atime.tv_sec = (int)be32_to_cpu(from->di_atime.t_sec);
215 inode->i_atime.tv_nsec = (int)be32_to_cpu(from->di_atime.t_nsec);
216 inode->i_mtime.tv_sec = (int)be32_to_cpu(from->di_mtime.t_sec);
217 inode->i_mtime.tv_nsec = (int)be32_to_cpu(from->di_mtime.t_nsec);
218 inode->i_ctime.tv_sec = (int)be32_to_cpu(from->di_ctime.t_sec);
219 inode->i_ctime.tv_nsec = (int)be32_to_cpu(from->di_ctime.t_nsec);
220
221 to->di_size = be64_to_cpu(from->di_size);
222 to->di_nblocks = be64_to_cpu(from->di_nblocks);
223 to->di_extsize = be32_to_cpu(from->di_extsize);
224 to->di_forkoff = from->di_forkoff;
225 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
226 to->di_dmstate = be16_to_cpu(from->di_dmstate);
227 to->di_flags = be16_to_cpu(from->di_flags);
228
229 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
230 inode_set_iversion_queried(inode,
231 be64_to_cpu(from->di_changecount));
232 to->di_crtime.tv_sec = be32_to_cpu(from->di_crtime.t_sec);
233 to->di_crtime.tv_nsec = be32_to_cpu(from->di_crtime.t_nsec);
234 to->di_flags2 = be64_to_cpu(from->di_flags2);
235 to->di_cowextsize = be32_to_cpu(from->di_cowextsize);
236 }
237
238 error = xfs_iformat_data_fork(ip, from);
239 if (error)
240 return error;
241 if (from->di_forkoff) {
242 error = xfs_iformat_attr_fork(ip, from);
243 if (error)
244 goto out_destroy_data_fork;
245 }
246 if (xfs_is_reflink_inode(ip))
247 xfs_ifork_init_cow(ip);
248 return 0;
249
250out_destroy_data_fork:
251 xfs_idestroy_fork(&ip->i_df);
252 return error;
253}
254
255void
256xfs_inode_to_disk(
257 struct xfs_inode *ip,
258 struct xfs_dinode *to,
259 xfs_lsn_t lsn)
260{
261 struct xfs_icdinode *from = &ip->i_d;
262 struct inode *inode = VFS_I(ip);
263
264 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
265 to->di_onlink = 0;
266
267 to->di_format = xfs_ifork_format(&ip->i_df);
268 to->di_uid = cpu_to_be32(i_uid_read(inode));
269 to->di_gid = cpu_to_be32(i_gid_read(inode));
270 to->di_projid_lo = cpu_to_be16(from->di_projid & 0xffff);
271 to->di_projid_hi = cpu_to_be16(from->di_projid >> 16);
272
273 memset(to->di_pad, 0, sizeof(to->di_pad));
274 to->di_atime.t_sec = cpu_to_be32(inode->i_atime.tv_sec);
275 to->di_atime.t_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
276 to->di_mtime.t_sec = cpu_to_be32(inode->i_mtime.tv_sec);
277 to->di_mtime.t_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
278 to->di_ctime.t_sec = cpu_to_be32(inode->i_ctime.tv_sec);
279 to->di_ctime.t_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
280 to->di_nlink = cpu_to_be32(inode->i_nlink);
281 to->di_gen = cpu_to_be32(inode->i_generation);
282 to->di_mode = cpu_to_be16(inode->i_mode);
283
284 to->di_size = cpu_to_be64(from->di_size);
285 to->di_nblocks = cpu_to_be64(from->di_nblocks);
286 to->di_extsize = cpu_to_be32(from->di_extsize);
287 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
288 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(ip->i_afp));
289 to->di_forkoff = from->di_forkoff;
290 to->di_aformat = xfs_ifork_format(ip->i_afp);
291 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
292 to->di_dmstate = cpu_to_be16(from->di_dmstate);
293 to->di_flags = cpu_to_be16(from->di_flags);
294
295 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
296 to->di_version = 3;
297 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
298 to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.tv_sec);
299 to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.tv_nsec);
300 to->di_flags2 = cpu_to_be64(from->di_flags2);
301 to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
302 to->di_ino = cpu_to_be64(ip->i_ino);
303 to->di_lsn = cpu_to_be64(lsn);
304 memset(to->di_pad2, 0, sizeof(to->di_pad2));
305 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
306 to->di_flushiter = 0;
307 } else {
308 to->di_version = 2;
309 to->di_flushiter = cpu_to_be16(from->di_flushiter);
310 }
311}
312
313void
314xfs_log_dinode_to_disk(
315 struct xfs_log_dinode *from,
316 struct xfs_dinode *to)
317{
318 to->di_magic = cpu_to_be16(from->di_magic);
319 to->di_mode = cpu_to_be16(from->di_mode);
320 to->di_version = from->di_version;
321 to->di_format = from->di_format;
322 to->di_onlink = 0;
323 to->di_uid = cpu_to_be32(from->di_uid);
324 to->di_gid = cpu_to_be32(from->di_gid);
325 to->di_nlink = cpu_to_be32(from->di_nlink);
326 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
327 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
328 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
329
330 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
331 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
332 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
333 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
334 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
335 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
336
337 to->di_size = cpu_to_be64(from->di_size);
338 to->di_nblocks = cpu_to_be64(from->di_nblocks);
339 to->di_extsize = cpu_to_be32(from->di_extsize);
340 to->di_nextents = cpu_to_be32(from->di_nextents);
341 to->di_anextents = cpu_to_be16(from->di_anextents);
342 to->di_forkoff = from->di_forkoff;
343 to->di_aformat = from->di_aformat;
344 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
345 to->di_dmstate = cpu_to_be16(from->di_dmstate);
346 to->di_flags = cpu_to_be16(from->di_flags);
347 to->di_gen = cpu_to_be32(from->di_gen);
348
349 if (from->di_version == 3) {
350 to->di_changecount = cpu_to_be64(from->di_changecount);
351 to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
352 to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
353 to->di_flags2 = cpu_to_be64(from->di_flags2);
354 to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
355 to->di_ino = cpu_to_be64(from->di_ino);
356 to->di_lsn = cpu_to_be64(from->di_lsn);
357 memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
358 uuid_copy(&to->di_uuid, &from->di_uuid);
359 to->di_flushiter = 0;
360 } else {
361 to->di_flushiter = cpu_to_be16(from->di_flushiter);
362 }
363}
364
365static xfs_failaddr_t
366xfs_dinode_verify_fork(
367 struct xfs_dinode *dip,
368 struct xfs_mount *mp,
369 int whichfork)
370{
371 uint32_t di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
372
373 switch (XFS_DFORK_FORMAT(dip, whichfork)) {
374 case XFS_DINODE_FMT_LOCAL:
375 /*
376 * no local regular files yet
377 */
378 if (whichfork == XFS_DATA_FORK) {
379 if (S_ISREG(be16_to_cpu(dip->di_mode)))
380 return __this_address;
381 if (be64_to_cpu(dip->di_size) >
382 XFS_DFORK_SIZE(dip, mp, whichfork))
383 return __this_address;
384 }
385 if (di_nextents)
386 return __this_address;
387 break;
388 case XFS_DINODE_FMT_EXTENTS:
389 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
390 return __this_address;
391 break;
392 case XFS_DINODE_FMT_BTREE:
393 if (whichfork == XFS_ATTR_FORK) {
394 if (di_nextents > MAXAEXTNUM)
395 return __this_address;
396 } else if (di_nextents > MAXEXTNUM) {
397 return __this_address;
398 }
399 break;
400 default:
401 return __this_address;
402 }
403 return NULL;
404}
405
406static xfs_failaddr_t
407xfs_dinode_verify_forkoff(
408 struct xfs_dinode *dip,
409 struct xfs_mount *mp)
410{
411 if (!dip->di_forkoff)
412 return NULL;
413
414 switch (dip->di_format) {
415 case XFS_DINODE_FMT_DEV:
416 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
417 return __this_address;
418 break;
419 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
420 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
421 case XFS_DINODE_FMT_BTREE:
422 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
423 return __this_address;
424 break;
425 default:
426 return __this_address;
427 }
428 return NULL;
429}
430
431xfs_failaddr_t
432xfs_dinode_verify(
433 struct xfs_mount *mp,
434 xfs_ino_t ino,
435 struct xfs_dinode *dip)
436{
437 xfs_failaddr_t fa;
438 uint16_t mode;
439 uint16_t flags;
440 uint64_t flags2;
441 uint64_t di_size;
442
443 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
444 return __this_address;
445
446 /* Verify v3 integrity information first */
447 if (dip->di_version >= 3) {
448 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
449 return __this_address;
450 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
451 XFS_DINODE_CRC_OFF))
452 return __this_address;
453 if (be64_to_cpu(dip->di_ino) != ino)
454 return __this_address;
455 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
456 return __this_address;
457 }
458
459 /* don't allow invalid i_size */
460 di_size = be64_to_cpu(dip->di_size);
461 if (di_size & (1ULL << 63))
462 return __this_address;
463
464 mode = be16_to_cpu(dip->di_mode);
465 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
466 return __this_address;
467
468 /* No zero-length symlinks/dirs. */
469 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
470 return __this_address;
471
472 /* Fork checks carried over from xfs_iformat_fork */
473 if (mode &&
474 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
475 be64_to_cpu(dip->di_nblocks))
476 return __this_address;
477
478 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
479 return __this_address;
480
481 flags = be16_to_cpu(dip->di_flags);
482
483 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
484 return __this_address;
485
486 /* check for illegal values of forkoff */
487 fa = xfs_dinode_verify_forkoff(dip, mp);
488 if (fa)
489 return fa;
490
491 /* Do we have appropriate data fork formats for the mode? */
492 switch (mode & S_IFMT) {
493 case S_IFIFO:
494 case S_IFCHR:
495 case S_IFBLK:
496 case S_IFSOCK:
497 if (dip->di_format != XFS_DINODE_FMT_DEV)
498 return __this_address;
499 break;
500 case S_IFREG:
501 case S_IFLNK:
502 case S_IFDIR:
503 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
504 if (fa)
505 return fa;
506 break;
507 case 0:
508 /* Uninitialized inode ok. */
509 break;
510 default:
511 return __this_address;
512 }
513
514 if (dip->di_forkoff) {
515 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
516 if (fa)
517 return fa;
518 } else {
519 /*
520 * If there is no fork offset, this may be a freshly-made inode
521 * in a new disk cluster, in which case di_aformat is zeroed.
522 * Otherwise, such an inode must be in EXTENTS format; this goes
523 * for freed inodes as well.
524 */
525 switch (dip->di_aformat) {
526 case 0:
527 case XFS_DINODE_FMT_EXTENTS:
528 break;
529 default:
530 return __this_address;
531 }
532 if (dip->di_anextents)
533 return __this_address;
534 }
535
536 /* extent size hint validation */
537 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
538 mode, flags);
539 if (fa)
540 return fa;
541
542 /* only version 3 or greater inodes are extensively verified here */
543 if (dip->di_version < 3)
544 return NULL;
545
546 flags2 = be64_to_cpu(dip->di_flags2);
547
548 /* don't allow reflink/cowextsize if we don't have reflink */
549 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
550 !xfs_sb_version_hasreflink(&mp->m_sb))
551 return __this_address;
552
553 /* only regular files get reflink */
554 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
555 return __this_address;
556
557 /* don't let reflink and realtime mix */
558 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
559 return __this_address;
560
561 /* don't let reflink and dax mix */
562 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags2 & XFS_DIFLAG2_DAX))
563 return __this_address;
564
565 /* COW extent size hint validation */
566 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
567 mode, flags, flags2);
568 if (fa)
569 return fa;
570
571 return NULL;
572}
573
574void
575xfs_dinode_calc_crc(
576 struct xfs_mount *mp,
577 struct xfs_dinode *dip)
578{
579 uint32_t crc;
580
581 if (dip->di_version < 3)
582 return;
583
584 ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
585 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
586 XFS_DINODE_CRC_OFF);
587 dip->di_crc = xfs_end_cksum(crc);
588}
589
590/*
591 * Validate di_extsize hint.
592 *
593 * The rules are documented at xfs_ioctl_setattr_check_extsize().
594 * These functions must be kept in sync with each other.
595 */
596xfs_failaddr_t
597xfs_inode_validate_extsize(
598 struct xfs_mount *mp,
599 uint32_t extsize,
600 uint16_t mode,
601 uint16_t flags)
602{
603 bool rt_flag;
604 bool hint_flag;
605 bool inherit_flag;
606 uint32_t extsize_bytes;
607 uint32_t blocksize_bytes;
608
609 rt_flag = (flags & XFS_DIFLAG_REALTIME);
610 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
611 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
612 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
613
614 if (rt_flag)
615 blocksize_bytes = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
616 else
617 blocksize_bytes = mp->m_sb.sb_blocksize;
618
619 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
620 return __this_address;
621
622 if (hint_flag && !S_ISREG(mode))
623 return __this_address;
624
625 if (inherit_flag && !S_ISDIR(mode))
626 return __this_address;
627
628 if ((hint_flag || inherit_flag) && extsize == 0)
629 return __this_address;
630
631 /* free inodes get flags set to zero but extsize remains */
632 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
633 return __this_address;
634
635 if (extsize_bytes % blocksize_bytes)
636 return __this_address;
637
638 if (extsize > MAXEXTLEN)
639 return __this_address;
640
641 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
642 return __this_address;
643
644 return NULL;
645}
646
647/*
648 * Validate di_cowextsize hint.
649 *
650 * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
651 * These functions must be kept in sync with each other.
652 */
653xfs_failaddr_t
654xfs_inode_validate_cowextsize(
655 struct xfs_mount *mp,
656 uint32_t cowextsize,
657 uint16_t mode,
658 uint16_t flags,
659 uint64_t flags2)
660{
661 bool rt_flag;
662 bool hint_flag;
663 uint32_t cowextsize_bytes;
664
665 rt_flag = (flags & XFS_DIFLAG_REALTIME);
666 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
667 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
668
669 if (hint_flag && !xfs_sb_version_hasreflink(&mp->m_sb))
670 return __this_address;
671
672 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
673 return __this_address;
674
675 if (hint_flag && cowextsize == 0)
676 return __this_address;
677
678 /* free inodes get flags set to zero but cowextsize remains */
679 if (mode && !hint_flag && cowextsize != 0)
680 return __this_address;
681
682 if (hint_flag && rt_flag)
683 return __this_address;
684
685 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
686 return __this_address;
687
688 if (cowextsize > MAXEXTLEN)
689 return __this_address;
690
691 if (cowextsize > mp->m_sb.sb_agblocks / 2)
692 return __this_address;
693
694 return NULL;
695}