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 * Check that none of the inode's in the buffer have a next
25 * unlinked field of 0.
26 */
27#if defined(DEBUG)
28void
29xfs_inobp_check(
30 xfs_mount_t *mp,
31 xfs_buf_t *bp)
32{
33 int i;
34 xfs_dinode_t *dip;
35
36 for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
37 dip = xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize);
38 if (!dip->di_next_unlinked) {
39 xfs_alert(mp,
40 "Detected bogus zero next_unlinked field in inode %d buffer 0x%llx.",
41 i, (long long)bp->b_bn);
42 }
43 }
44}
45#endif
46
47bool
48xfs_dinode_good_version(
49 struct xfs_mount *mp,
50 __u8 version)
51{
52 if (xfs_sb_version_hascrc(&mp->m_sb))
53 return version == 3;
54
55 return version == 1 || version == 2;
56}
57
58/*
59 * If we are doing readahead on an inode buffer, we might be in log recovery
60 * reading an inode allocation buffer that hasn't yet been replayed, and hence
61 * has not had the inode cores stamped into it. Hence for readahead, the buffer
62 * may be potentially invalid.
63 *
64 * If the readahead buffer is invalid, we need to mark it with an error and
65 * clear the DONE status of the buffer so that a followup read will re-read it
66 * from disk. We don't report the error otherwise to avoid warnings during log
67 * recovery and we don't get unnecssary panics on debug kernels. We use EIO here
68 * because all we want to do is say readahead failed; there is no-one to report
69 * the error to, so this will distinguish it from a non-ra verifier failure.
70 * Changes to this readahead error behavour also need to be reflected in
71 * xfs_dquot_buf_readahead_verify().
72 */
73static void
74xfs_inode_buf_verify(
75 struct xfs_buf *bp,
76 bool readahead)
77{
78 struct xfs_mount *mp = bp->b_mount;
79 xfs_agnumber_t agno;
80 int i;
81 int ni;
82
83 /*
84 * Validate the magic number and version of every inode in the buffer
85 */
86 agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
87 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
88 for (i = 0; i < ni; i++) {
89 int di_ok;
90 xfs_dinode_t *dip;
91 xfs_agino_t unlinked_ino;
92
93 dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
94 unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
95 di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
96 xfs_dinode_good_version(mp, dip->di_version) &&
97 xfs_verify_agino_or_null(mp, agno, unlinked_ino);
98 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
99 XFS_ERRTAG_ITOBP_INOTOBP))) {
100 if (readahead) {
101 bp->b_flags &= ~XBF_DONE;
102 xfs_buf_ioerror(bp, -EIO);
103 return;
104 }
105
106#ifdef DEBUG
107 xfs_alert(mp,
108 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
109 (unsigned long long)bp->b_bn, i,
110 be16_to_cpu(dip->di_magic));
111#endif
112 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
113 __func__, dip, sizeof(*dip),
114 NULL);
115 return;
116 }
117 }
118}
119
120
121static void
122xfs_inode_buf_read_verify(
123 struct xfs_buf *bp)
124{
125 xfs_inode_buf_verify(bp, false);
126}
127
128static void
129xfs_inode_buf_readahead_verify(
130 struct xfs_buf *bp)
131{
132 xfs_inode_buf_verify(bp, true);
133}
134
135static void
136xfs_inode_buf_write_verify(
137 struct xfs_buf *bp)
138{
139 xfs_inode_buf_verify(bp, false);
140}
141
142const struct xfs_buf_ops xfs_inode_buf_ops = {
143 .name = "xfs_inode",
144 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
145 cpu_to_be16(XFS_DINODE_MAGIC) },
146 .verify_read = xfs_inode_buf_read_verify,
147 .verify_write = xfs_inode_buf_write_verify,
148};
149
150const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
151 .name = "xfs_inode_ra",
152 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
153 cpu_to_be16(XFS_DINODE_MAGIC) },
154 .verify_read = xfs_inode_buf_readahead_verify,
155 .verify_write = xfs_inode_buf_write_verify,
156};
157
158
159/*
160 * This routine is called to map an inode to the buffer containing the on-disk
161 * version of the inode. It returns a pointer to the buffer containing the
162 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a
163 * pointer to the on-disk inode within that buffer.
164 *
165 * If a non-zero error is returned, then the contents of bpp and dipp are
166 * undefined.
167 */
168int
169xfs_imap_to_bp(
170 struct xfs_mount *mp,
171 struct xfs_trans *tp,
172 struct xfs_imap *imap,
173 struct xfs_dinode **dipp,
174 struct xfs_buf **bpp,
175 uint buf_flags,
176 uint iget_flags)
177{
178 struct xfs_buf *bp;
179 int error;
180
181 buf_flags |= XBF_UNMAPPED;
182 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
183 (int)imap->im_len, buf_flags, &bp,
184 &xfs_inode_buf_ops);
185 if (error) {
186 if (error == -EAGAIN) {
187 ASSERT(buf_flags & XBF_TRYLOCK);
188 return error;
189 }
190 xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.",
191 __func__, error);
192 return error;
193 }
194
195 *bpp = bp;
196 *dipp = xfs_buf_offset(bp, imap->im_boffset);
197 return 0;
198}
199
200void
201xfs_inode_from_disk(
202 struct xfs_inode *ip,
203 struct xfs_dinode *from)
204{
205 struct xfs_icdinode *to = &ip->i_d;
206 struct inode *inode = VFS_I(ip);
207
208
209 /*
210 * Convert v1 inodes immediately to v2 inode format as this is the
211 * minimum inode version format we support in the rest of the code.
212 */
213 to->di_version = from->di_version;
214 if (to->di_version == 1) {
215 set_nlink(inode, be16_to_cpu(from->di_onlink));
216 to->di_projid_lo = 0;
217 to->di_projid_hi = 0;
218 to->di_version = 2;
219 } else {
220 set_nlink(inode, be32_to_cpu(from->di_nlink));
221 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
222 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
223 }
224
225 to->di_format = from->di_format;
226 to->di_uid = be32_to_cpu(from->di_uid);
227 to->di_gid = be32_to_cpu(from->di_gid);
228 to->di_flushiter = be16_to_cpu(from->di_flushiter);
229
230 /*
231 * Time is signed, so need to convert to signed 32 bit before
232 * storing in inode timestamp which may be 64 bit. Otherwise
233 * a time before epoch is converted to a time long after epoch
234 * on 64 bit systems.
235 */
236 inode->i_atime.tv_sec = (int)be32_to_cpu(from->di_atime.t_sec);
237 inode->i_atime.tv_nsec = (int)be32_to_cpu(from->di_atime.t_nsec);
238 inode->i_mtime.tv_sec = (int)be32_to_cpu(from->di_mtime.t_sec);
239 inode->i_mtime.tv_nsec = (int)be32_to_cpu(from->di_mtime.t_nsec);
240 inode->i_ctime.tv_sec = (int)be32_to_cpu(from->di_ctime.t_sec);
241 inode->i_ctime.tv_nsec = (int)be32_to_cpu(from->di_ctime.t_nsec);
242 inode->i_generation = be32_to_cpu(from->di_gen);
243 inode->i_mode = be16_to_cpu(from->di_mode);
244
245 to->di_size = be64_to_cpu(from->di_size);
246 to->di_nblocks = be64_to_cpu(from->di_nblocks);
247 to->di_extsize = be32_to_cpu(from->di_extsize);
248 to->di_nextents = be32_to_cpu(from->di_nextents);
249 to->di_anextents = be16_to_cpu(from->di_anextents);
250 to->di_forkoff = from->di_forkoff;
251 to->di_aformat = from->di_aformat;
252 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
253 to->di_dmstate = be16_to_cpu(from->di_dmstate);
254 to->di_flags = be16_to_cpu(from->di_flags);
255
256 if (to->di_version == 3) {
257 inode_set_iversion_queried(inode,
258 be64_to_cpu(from->di_changecount));
259 to->di_crtime.t_sec = be32_to_cpu(from->di_crtime.t_sec);
260 to->di_crtime.t_nsec = be32_to_cpu(from->di_crtime.t_nsec);
261 to->di_flags2 = be64_to_cpu(from->di_flags2);
262 to->di_cowextsize = be32_to_cpu(from->di_cowextsize);
263 }
264}
265
266void
267xfs_inode_to_disk(
268 struct xfs_inode *ip,
269 struct xfs_dinode *to,
270 xfs_lsn_t lsn)
271{
272 struct xfs_icdinode *from = &ip->i_d;
273 struct inode *inode = VFS_I(ip);
274
275 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
276 to->di_onlink = 0;
277
278 to->di_version = from->di_version;
279 to->di_format = from->di_format;
280 to->di_uid = cpu_to_be32(from->di_uid);
281 to->di_gid = cpu_to_be32(from->di_gid);
282 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
283 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
284
285 memset(to->di_pad, 0, sizeof(to->di_pad));
286 to->di_atime.t_sec = cpu_to_be32(inode->i_atime.tv_sec);
287 to->di_atime.t_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
288 to->di_mtime.t_sec = cpu_to_be32(inode->i_mtime.tv_sec);
289 to->di_mtime.t_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
290 to->di_ctime.t_sec = cpu_to_be32(inode->i_ctime.tv_sec);
291 to->di_ctime.t_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
292 to->di_nlink = cpu_to_be32(inode->i_nlink);
293 to->di_gen = cpu_to_be32(inode->i_generation);
294 to->di_mode = cpu_to_be16(inode->i_mode);
295
296 to->di_size = cpu_to_be64(from->di_size);
297 to->di_nblocks = cpu_to_be64(from->di_nblocks);
298 to->di_extsize = cpu_to_be32(from->di_extsize);
299 to->di_nextents = cpu_to_be32(from->di_nextents);
300 to->di_anextents = cpu_to_be16(from->di_anextents);
301 to->di_forkoff = from->di_forkoff;
302 to->di_aformat = from->di_aformat;
303 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
304 to->di_dmstate = cpu_to_be16(from->di_dmstate);
305 to->di_flags = cpu_to_be16(from->di_flags);
306
307 if (from->di_version == 3) {
308 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
309 to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
310 to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
311 to->di_flags2 = cpu_to_be64(from->di_flags2);
312 to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
313 to->di_ino = cpu_to_be64(ip->i_ino);
314 to->di_lsn = cpu_to_be64(lsn);
315 memset(to->di_pad2, 0, sizeof(to->di_pad2));
316 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
317 to->di_flushiter = 0;
318 } else {
319 to->di_flushiter = cpu_to_be16(from->di_flushiter);
320 }
321}
322
323void
324xfs_log_dinode_to_disk(
325 struct xfs_log_dinode *from,
326 struct xfs_dinode *to)
327{
328 to->di_magic = cpu_to_be16(from->di_magic);
329 to->di_mode = cpu_to_be16(from->di_mode);
330 to->di_version = from->di_version;
331 to->di_format = from->di_format;
332 to->di_onlink = 0;
333 to->di_uid = cpu_to_be32(from->di_uid);
334 to->di_gid = cpu_to_be32(from->di_gid);
335 to->di_nlink = cpu_to_be32(from->di_nlink);
336 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
337 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
338 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
339
340 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
341 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
342 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
343 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
344 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
345 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
346
347 to->di_size = cpu_to_be64(from->di_size);
348 to->di_nblocks = cpu_to_be64(from->di_nblocks);
349 to->di_extsize = cpu_to_be32(from->di_extsize);
350 to->di_nextents = cpu_to_be32(from->di_nextents);
351 to->di_anextents = cpu_to_be16(from->di_anextents);
352 to->di_forkoff = from->di_forkoff;
353 to->di_aformat = from->di_aformat;
354 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
355 to->di_dmstate = cpu_to_be16(from->di_dmstate);
356 to->di_flags = cpu_to_be16(from->di_flags);
357 to->di_gen = cpu_to_be32(from->di_gen);
358
359 if (from->di_version == 3) {
360 to->di_changecount = cpu_to_be64(from->di_changecount);
361 to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
362 to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
363 to->di_flags2 = cpu_to_be64(from->di_flags2);
364 to->di_cowextsize = cpu_to_be32(from->di_cowextsize);
365 to->di_ino = cpu_to_be64(from->di_ino);
366 to->di_lsn = cpu_to_be64(from->di_lsn);
367 memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
368 uuid_copy(&to->di_uuid, &from->di_uuid);
369 to->di_flushiter = 0;
370 } else {
371 to->di_flushiter = cpu_to_be16(from->di_flushiter);
372 }
373}
374
375static xfs_failaddr_t
376xfs_dinode_verify_fork(
377 struct xfs_dinode *dip,
378 struct xfs_mount *mp,
379 int whichfork)
380{
381 uint32_t di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
382
383 switch (XFS_DFORK_FORMAT(dip, whichfork)) {
384 case XFS_DINODE_FMT_LOCAL:
385 /*
386 * no local regular files yet
387 */
388 if (whichfork == XFS_DATA_FORK) {
389 if (S_ISREG(be16_to_cpu(dip->di_mode)))
390 return __this_address;
391 if (be64_to_cpu(dip->di_size) >
392 XFS_DFORK_SIZE(dip, mp, whichfork))
393 return __this_address;
394 }
395 if (di_nextents)
396 return __this_address;
397 break;
398 case XFS_DINODE_FMT_EXTENTS:
399 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
400 return __this_address;
401 break;
402 case XFS_DINODE_FMT_BTREE:
403 if (whichfork == XFS_ATTR_FORK) {
404 if (di_nextents > MAXAEXTNUM)
405 return __this_address;
406 } else if (di_nextents > MAXEXTNUM) {
407 return __this_address;
408 }
409 break;
410 default:
411 return __this_address;
412 }
413 return NULL;
414}
415
416static xfs_failaddr_t
417xfs_dinode_verify_forkoff(
418 struct xfs_dinode *dip,
419 struct xfs_mount *mp)
420{
421 if (!XFS_DFORK_Q(dip))
422 return NULL;
423
424 switch (dip->di_format) {
425 case XFS_DINODE_FMT_DEV:
426 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
427 return __this_address;
428 break;
429 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
430 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
431 case XFS_DINODE_FMT_BTREE:
432 if (dip->di_forkoff >= (XFS_LITINO(mp, dip->di_version) >> 3))
433 return __this_address;
434 break;
435 default:
436 return __this_address;
437 }
438 return NULL;
439}
440
441xfs_failaddr_t
442xfs_dinode_verify(
443 struct xfs_mount *mp,
444 xfs_ino_t ino,
445 struct xfs_dinode *dip)
446{
447 xfs_failaddr_t fa;
448 uint16_t mode;
449 uint16_t flags;
450 uint64_t flags2;
451 uint64_t di_size;
452
453 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
454 return __this_address;
455
456 /* Verify v3 integrity information first */
457 if (dip->di_version >= 3) {
458 if (!xfs_sb_version_hascrc(&mp->m_sb))
459 return __this_address;
460 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
461 XFS_DINODE_CRC_OFF))
462 return __this_address;
463 if (be64_to_cpu(dip->di_ino) != ino)
464 return __this_address;
465 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
466 return __this_address;
467 }
468
469 /* don't allow invalid i_size */
470 di_size = be64_to_cpu(dip->di_size);
471 if (di_size & (1ULL << 63))
472 return __this_address;
473
474 mode = be16_to_cpu(dip->di_mode);
475 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
476 return __this_address;
477
478 /* No zero-length symlinks/dirs. */
479 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
480 return __this_address;
481
482 /* Fork checks carried over from xfs_iformat_fork */
483 if (mode &&
484 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
485 be64_to_cpu(dip->di_nblocks))
486 return __this_address;
487
488 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
489 return __this_address;
490
491 flags = be16_to_cpu(dip->di_flags);
492
493 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
494 return __this_address;
495
496 /* check for illegal values of forkoff */
497 fa = xfs_dinode_verify_forkoff(dip, mp);
498 if (fa)
499 return fa;
500
501 /* Do we have appropriate data fork formats for the mode? */
502 switch (mode & S_IFMT) {
503 case S_IFIFO:
504 case S_IFCHR:
505 case S_IFBLK:
506 case S_IFSOCK:
507 if (dip->di_format != XFS_DINODE_FMT_DEV)
508 return __this_address;
509 break;
510 case S_IFREG:
511 case S_IFLNK:
512 case S_IFDIR:
513 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
514 if (fa)
515 return fa;
516 break;
517 case 0:
518 /* Uninitialized inode ok. */
519 break;
520 default:
521 return __this_address;
522 }
523
524 if (XFS_DFORK_Q(dip)) {
525 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
526 if (fa)
527 return fa;
528 } else {
529 /*
530 * If there is no fork offset, this may be a freshly-made inode
531 * in a new disk cluster, in which case di_aformat is zeroed.
532 * Otherwise, such an inode must be in EXTENTS format; this goes
533 * for freed inodes as well.
534 */
535 switch (dip->di_aformat) {
536 case 0:
537 case XFS_DINODE_FMT_EXTENTS:
538 break;
539 default:
540 return __this_address;
541 }
542 if (dip->di_anextents)
543 return __this_address;
544 }
545
546 /* extent size hint validation */
547 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
548 mode, flags);
549 if (fa)
550 return fa;
551
552 /* only version 3 or greater inodes are extensively verified here */
553 if (dip->di_version < 3)
554 return NULL;
555
556 flags2 = be64_to_cpu(dip->di_flags2);
557
558 /* don't allow reflink/cowextsize if we don't have reflink */
559 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
560 !xfs_sb_version_hasreflink(&mp->m_sb))
561 return __this_address;
562
563 /* only regular files get reflink */
564 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
565 return __this_address;
566
567 /* don't let reflink and realtime mix */
568 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
569 return __this_address;
570
571 /* don't let reflink and dax mix */
572 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags2 & XFS_DIFLAG2_DAX))
573 return __this_address;
574
575 /* COW extent size hint validation */
576 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
577 mode, flags, flags2);
578 if (fa)
579 return fa;
580
581 return NULL;
582}
583
584void
585xfs_dinode_calc_crc(
586 struct xfs_mount *mp,
587 struct xfs_dinode *dip)
588{
589 uint32_t crc;
590
591 if (dip->di_version < 3)
592 return;
593
594 ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
595 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
596 XFS_DINODE_CRC_OFF);
597 dip->di_crc = xfs_end_cksum(crc);
598}
599
600/*
601 * Read the disk inode attributes into the in-core inode structure.
602 *
603 * For version 5 superblocks, if we are initialising a new inode and we are not
604 * utilising the XFS_MOUNT_IKEEP inode cluster mode, we can simple build the new
605 * inode core with a random generation number. If we are keeping inodes around,
606 * we need to read the inode cluster to get the existing generation number off
607 * disk. Further, if we are using version 4 superblocks (i.e. v1/v2 inode
608 * format) then log recovery is dependent on the di_flushiter field being
609 * initialised from the current on-disk value and hence we must also read the
610 * inode off disk.
611 */
612int
613xfs_iread(
614 xfs_mount_t *mp,
615 xfs_trans_t *tp,
616 xfs_inode_t *ip,
617 uint iget_flags)
618{
619 xfs_buf_t *bp;
620 xfs_dinode_t *dip;
621 xfs_failaddr_t fa;
622 int error;
623
624 /*
625 * Fill in the location information in the in-core inode.
626 */
627 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
628 if (error)
629 return error;
630
631 /* shortcut IO on inode allocation if possible */
632 if ((iget_flags & XFS_IGET_CREATE) &&
633 xfs_sb_version_hascrc(&mp->m_sb) &&
634 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
635 /* initialise the on-disk inode core */
636 memset(&ip->i_d, 0, sizeof(ip->i_d));
637 VFS_I(ip)->i_generation = prandom_u32();
638 ip->i_d.di_version = 3;
639 return 0;
640 }
641
642 /*
643 * Get pointers to the on-disk inode and the buffer containing it.
644 */
645 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags);
646 if (error)
647 return error;
648
649 /* even unallocated inodes are verified */
650 fa = xfs_dinode_verify(mp, ip->i_ino, dip);
651 if (fa) {
652 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", dip,
653 sizeof(*dip), fa);
654 error = -EFSCORRUPTED;
655 goto out_brelse;
656 }
657
658 /*
659 * If the on-disk inode is already linked to a directory
660 * entry, copy all of the inode into the in-core inode.
661 * xfs_iformat_fork() handles copying in the inode format
662 * specific information.
663 * Otherwise, just get the truly permanent information.
664 */
665 if (dip->di_mode) {
666 xfs_inode_from_disk(ip, dip);
667 error = xfs_iformat_fork(ip, dip);
668 if (error) {
669#ifdef DEBUG
670 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
671 __func__, error);
672#endif /* DEBUG */
673 goto out_brelse;
674 }
675 } else {
676 /*
677 * Partial initialisation of the in-core inode. Just the bits
678 * that xfs_ialloc won't overwrite or relies on being correct.
679 */
680 ip->i_d.di_version = dip->di_version;
681 VFS_I(ip)->i_generation = be32_to_cpu(dip->di_gen);
682 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
683
684 /*
685 * Make sure to pull in the mode here as well in
686 * case the inode is released without being used.
687 * This ensures that xfs_inactive() will see that
688 * the inode is already free and not try to mess
689 * with the uninitialized part of it.
690 */
691 VFS_I(ip)->i_mode = 0;
692 }
693
694 ASSERT(ip->i_d.di_version >= 2);
695 ip->i_delayed_blks = 0;
696
697 /*
698 * Mark the buffer containing the inode as something to keep
699 * around for a while. This helps to keep recently accessed
700 * meta-data in-core longer.
701 */
702 xfs_buf_set_ref(bp, XFS_INO_REF);
703
704 /*
705 * Use xfs_trans_brelse() to release the buffer containing the on-disk
706 * inode, because it was acquired with xfs_trans_read_buf() in
707 * xfs_imap_to_bp() above. If tp is NULL, this is just a normal
708 * brelse(). If we're within a transaction, then xfs_trans_brelse()
709 * will only release the buffer if it is not dirty within the
710 * transaction. It will be OK to release the buffer in this case,
711 * because inodes on disk are never destroyed and we will be locking the
712 * new in-core inode before putting it in the cache where other
713 * processes can find it. Thus we don't have to worry about the inode
714 * being changed just because we released the buffer.
715 */
716 out_brelse:
717 xfs_trans_brelse(tp, bp);
718 return error;
719}
720
721/*
722 * Validate di_extsize hint.
723 *
724 * The rules are documented at xfs_ioctl_setattr_check_extsize().
725 * These functions must be kept in sync with each other.
726 */
727xfs_failaddr_t
728xfs_inode_validate_extsize(
729 struct xfs_mount *mp,
730 uint32_t extsize,
731 uint16_t mode,
732 uint16_t flags)
733{
734 bool rt_flag;
735 bool hint_flag;
736 bool inherit_flag;
737 uint32_t extsize_bytes;
738 uint32_t blocksize_bytes;
739
740 rt_flag = (flags & XFS_DIFLAG_REALTIME);
741 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
742 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
743 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
744
745 if (rt_flag)
746 blocksize_bytes = mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog;
747 else
748 blocksize_bytes = mp->m_sb.sb_blocksize;
749
750 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
751 return __this_address;
752
753 if (hint_flag && !S_ISREG(mode))
754 return __this_address;
755
756 if (inherit_flag && !S_ISDIR(mode))
757 return __this_address;
758
759 if ((hint_flag || inherit_flag) && extsize == 0)
760 return __this_address;
761
762 /* free inodes get flags set to zero but extsize remains */
763 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
764 return __this_address;
765
766 if (extsize_bytes % blocksize_bytes)
767 return __this_address;
768
769 if (extsize > MAXEXTLEN)
770 return __this_address;
771
772 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
773 return __this_address;
774
775 return NULL;
776}
777
778/*
779 * Validate di_cowextsize hint.
780 *
781 * The rules are documented at xfs_ioctl_setattr_check_cowextsize().
782 * These functions must be kept in sync with each other.
783 */
784xfs_failaddr_t
785xfs_inode_validate_cowextsize(
786 struct xfs_mount *mp,
787 uint32_t cowextsize,
788 uint16_t mode,
789 uint16_t flags,
790 uint64_t flags2)
791{
792 bool rt_flag;
793 bool hint_flag;
794 uint32_t cowextsize_bytes;
795
796 rt_flag = (flags & XFS_DIFLAG_REALTIME);
797 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
798 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
799
800 if (hint_flag && !xfs_sb_version_hasreflink(&mp->m_sb))
801 return __this_address;
802
803 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
804 return __this_address;
805
806 if (hint_flag && cowextsize == 0)
807 return __this_address;
808
809 /* free inodes get flags set to zero but cowextsize remains */
810 if (mode && !hint_flag && cowextsize != 0)
811 return __this_address;
812
813 if (hint_flag && rt_flag)
814 return __this_address;
815
816 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
817 return __this_address;
818
819 if (cowextsize > MAXEXTLEN)
820 return __this_address;
821
822 if (cowextsize > mp->m_sb.sb_agblocks / 2)
823 return __this_address;
824
825 return NULL;
826}