Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_trans.h"
16#include "xfs_inode_item.h"
17#include "xfs_btree.h"
18#include "xfs_bmap_btree.h"
19#include "xfs_bmap.h"
20#include "xfs_error.h"
21#include "xfs_trace.h"
22#include "xfs_da_format.h"
23#include "xfs_da_btree.h"
24#include "xfs_dir2_priv.h"
25#include "xfs_attr_leaf.h"
26#include "xfs_types.h"
27#include "xfs_errortag.h"
28#include "xfs_health.h"
29#include "xfs_symlink_remote.h"
30
31struct kmem_cache *xfs_ifork_cache;
32
33void
34xfs_init_local_fork(
35 struct xfs_inode *ip,
36 int whichfork,
37 const void *data,
38 int64_t size)
39{
40 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
41 int mem_size = size;
42 bool zero_terminate;
43
44 /*
45 * If we are using the local fork to store a symlink body we need to
46 * zero-terminate it so that we can pass it back to the VFS directly.
47 * Overallocate the in-memory fork by one for that and add a zero
48 * to terminate it below.
49 */
50 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
51 if (zero_terminate)
52 mem_size++;
53
54 if (size) {
55 char *new_data = kmalloc(mem_size,
56 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
57
58 memcpy(new_data, data, size);
59 if (zero_terminate)
60 new_data[size] = '\0';
61
62 ifp->if_data = new_data;
63 } else {
64 ifp->if_data = NULL;
65 }
66
67 ifp->if_bytes = size;
68}
69
70/*
71 * The file is in-lined in the on-disk inode.
72 */
73STATIC int
74xfs_iformat_local(
75 struct xfs_inode *ip,
76 struct xfs_dinode *dip,
77 int whichfork,
78 int size)
79{
80 /*
81 * If the size is unreasonable, then something
82 * is wrong and we just bail out rather than crash in
83 * kmalloc() or memcpy() below.
84 */
85 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
86 xfs_warn(ip->i_mount,
87 "corrupt inode %llu (bad size %d for local fork, size = %zd).",
88 (unsigned long long) ip->i_ino, size,
89 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
90 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
91 "xfs_iformat_local", dip, sizeof(*dip),
92 __this_address);
93 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
94 return -EFSCORRUPTED;
95 }
96
97 xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
98 return 0;
99}
100
101/*
102 * The file consists of a set of extents all of which fit into the on-disk
103 * inode.
104 */
105STATIC int
106xfs_iformat_extents(
107 struct xfs_inode *ip,
108 struct xfs_dinode *dip,
109 int whichfork)
110{
111 struct xfs_mount *mp = ip->i_mount;
112 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
113 int state = xfs_bmap_fork_to_state(whichfork);
114 xfs_extnum_t nex = xfs_dfork_nextents(dip, whichfork);
115 int size = nex * sizeof(xfs_bmbt_rec_t);
116 struct xfs_iext_cursor icur;
117 struct xfs_bmbt_rec *dp;
118 struct xfs_bmbt_irec new;
119 int i;
120
121 /*
122 * If the number of extents is unreasonable, then something is wrong and
123 * we just bail out rather than crash in kmalloc() or memcpy() below.
124 */
125 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, mp, whichfork))) {
126 xfs_warn(ip->i_mount, "corrupt inode %llu ((a)extents = %llu).",
127 ip->i_ino, nex);
128 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
129 "xfs_iformat_extents(1)", dip, sizeof(*dip),
130 __this_address);
131 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
132 return -EFSCORRUPTED;
133 }
134
135 ifp->if_bytes = 0;
136 ifp->if_data = NULL;
137 ifp->if_height = 0;
138 if (size) {
139 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
140
141 xfs_iext_first(ifp, &icur);
142 for (i = 0; i < nex; i++, dp++) {
143 xfs_failaddr_t fa;
144
145 xfs_bmbt_disk_get_all(dp, &new);
146 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
147 if (fa) {
148 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
149 "xfs_iformat_extents(2)",
150 dp, sizeof(*dp), fa);
151 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
152 return xfs_bmap_complain_bad_rec(ip, whichfork,
153 fa, &new);
154 }
155
156 xfs_iext_insert(ip, &icur, &new, state);
157 trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
158 xfs_iext_next(ifp, &icur);
159 }
160 }
161 return 0;
162}
163
164/*
165 * The file has too many extents to fit into
166 * the inode, so they are in B-tree format.
167 * Allocate a buffer for the root of the B-tree
168 * and copy the root into it. The i_extents
169 * field will remain NULL until all of the
170 * extents are read in (when they are needed).
171 */
172STATIC int
173xfs_iformat_btree(
174 struct xfs_inode *ip,
175 struct xfs_dinode *dip,
176 int whichfork)
177{
178 struct xfs_mount *mp = ip->i_mount;
179 xfs_bmdr_block_t *dfp;
180 struct xfs_ifork *ifp;
181 /* REFERENCED */
182 int nrecs;
183 int size;
184 int level;
185
186 ifp = xfs_ifork_ptr(ip, whichfork);
187 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
188 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
189 nrecs = be16_to_cpu(dfp->bb_numrecs);
190 level = be16_to_cpu(dfp->bb_level);
191
192 /*
193 * blow out if -- fork has less extents than can fit in
194 * fork (fork shouldn't be a btree format), root btree
195 * block has more records than can fit into the fork,
196 * or the number of extents is greater than the number of
197 * blocks.
198 */
199 if (unlikely(ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork) ||
200 nrecs == 0 ||
201 XFS_BMDR_SPACE_CALC(nrecs) >
202 XFS_DFORK_SIZE(dip, mp, whichfork) ||
203 ifp->if_nextents > ip->i_nblocks) ||
204 level == 0 || level > XFS_BM_MAXLEVELS(mp, whichfork)) {
205 xfs_warn(mp, "corrupt inode %llu (btree).",
206 (unsigned long long) ip->i_ino);
207 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
208 "xfs_iformat_btree", dfp, size,
209 __this_address);
210 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
211 return -EFSCORRUPTED;
212 }
213
214 ifp->if_broot_bytes = size;
215 ifp->if_broot = kmalloc(size,
216 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
217 ASSERT(ifp->if_broot != NULL);
218 /*
219 * Copy and convert from the on-disk structure
220 * to the in-memory structure.
221 */
222 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
223 ifp->if_broot, size);
224
225 ifp->if_bytes = 0;
226 ifp->if_data = NULL;
227 ifp->if_height = 0;
228 return 0;
229}
230
231int
232xfs_iformat_data_fork(
233 struct xfs_inode *ip,
234 struct xfs_dinode *dip)
235{
236 struct inode *inode = VFS_I(ip);
237 int error;
238
239 /*
240 * Initialize the extent count early, as the per-format routines may
241 * depend on it. Use release semantics to set needextents /after/ we
242 * set the format. This ensures that we can use acquire semantics on
243 * needextents in xfs_need_iread_extents() and be guaranteed to see a
244 * valid format value after that load.
245 */
246 ip->i_df.if_format = dip->di_format;
247 ip->i_df.if_nextents = xfs_dfork_data_extents(dip);
248 smp_store_release(&ip->i_df.if_needextents,
249 ip->i_df.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
250
251 switch (inode->i_mode & S_IFMT) {
252 case S_IFIFO:
253 case S_IFCHR:
254 case S_IFBLK:
255 case S_IFSOCK:
256 ip->i_disk_size = 0;
257 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
258 return 0;
259 case S_IFREG:
260 case S_IFLNK:
261 case S_IFDIR:
262 switch (ip->i_df.if_format) {
263 case XFS_DINODE_FMT_LOCAL:
264 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK,
265 be64_to_cpu(dip->di_size));
266 if (!error)
267 error = xfs_ifork_verify_local_data(ip);
268 return error;
269 case XFS_DINODE_FMT_EXTENTS:
270 return xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
271 case XFS_DINODE_FMT_BTREE:
272 return xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
273 default:
274 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
275 dip, sizeof(*dip), __this_address);
276 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
277 return -EFSCORRUPTED;
278 }
279 break;
280 default:
281 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
282 sizeof(*dip), __this_address);
283 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
284 return -EFSCORRUPTED;
285 }
286}
287
288static uint16_t
289xfs_dfork_attr_shortform_size(
290 struct xfs_dinode *dip)
291{
292 struct xfs_attr_sf_hdr *sf = XFS_DFORK_APTR(dip);
293
294 return be16_to_cpu(sf->totsize);
295}
296
297void
298xfs_ifork_init_attr(
299 struct xfs_inode *ip,
300 enum xfs_dinode_fmt format,
301 xfs_extnum_t nextents)
302{
303 /*
304 * Initialize the extent count early, as the per-format routines may
305 * depend on it. Use release semantics to set needextents /after/ we
306 * set the format. This ensures that we can use acquire semantics on
307 * needextents in xfs_need_iread_extents() and be guaranteed to see a
308 * valid format value after that load.
309 */
310 ip->i_af.if_format = format;
311 ip->i_af.if_nextents = nextents;
312 smp_store_release(&ip->i_af.if_needextents,
313 ip->i_af.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
314}
315
316void
317xfs_ifork_zap_attr(
318 struct xfs_inode *ip)
319{
320 xfs_idestroy_fork(&ip->i_af);
321 memset(&ip->i_af, 0, sizeof(struct xfs_ifork));
322 ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS;
323}
324
325int
326xfs_iformat_attr_fork(
327 struct xfs_inode *ip,
328 struct xfs_dinode *dip)
329{
330 xfs_extnum_t naextents = xfs_dfork_attr_extents(dip);
331 int error = 0;
332
333 /*
334 * Initialize the extent count early, as the per-format routines may
335 * depend on it.
336 */
337 xfs_ifork_init_attr(ip, dip->di_aformat, naextents);
338
339 switch (ip->i_af.if_format) {
340 case XFS_DINODE_FMT_LOCAL:
341 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK,
342 xfs_dfork_attr_shortform_size(dip));
343 if (!error)
344 error = xfs_ifork_verify_local_attr(ip);
345 break;
346 case XFS_DINODE_FMT_EXTENTS:
347 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
348 break;
349 case XFS_DINODE_FMT_BTREE:
350 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
351 break;
352 default:
353 xfs_inode_verifier_error(ip, error, __func__, dip,
354 sizeof(*dip), __this_address);
355 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
356 error = -EFSCORRUPTED;
357 break;
358 }
359
360 if (error)
361 xfs_ifork_zap_attr(ip);
362 return error;
363}
364
365/*
366 * Reallocate the space for if_broot based on the number of records
367 * being added or deleted as indicated in rec_diff. Move the records
368 * and pointers in if_broot to fit the new size. When shrinking this
369 * will eliminate holes between the records and pointers created by
370 * the caller. When growing this will create holes to be filled in
371 * by the caller.
372 *
373 * The caller must not request to add more records than would fit in
374 * the on-disk inode root. If the if_broot is currently NULL, then
375 * if we are adding records, one will be allocated. The caller must also
376 * not request that the number of records go below zero, although
377 * it can go to zero.
378 *
379 * ip -- the inode whose if_broot area is changing
380 * ext_diff -- the change in the number of records, positive or negative,
381 * requested for the if_broot array.
382 */
383void
384xfs_iroot_realloc(
385 xfs_inode_t *ip,
386 int rec_diff,
387 int whichfork)
388{
389 struct xfs_mount *mp = ip->i_mount;
390 int cur_max;
391 struct xfs_ifork *ifp;
392 struct xfs_btree_block *new_broot;
393 int new_max;
394 size_t new_size;
395 char *np;
396 char *op;
397
398 /*
399 * Handle the degenerate case quietly.
400 */
401 if (rec_diff == 0) {
402 return;
403 }
404
405 ifp = xfs_ifork_ptr(ip, whichfork);
406 if (rec_diff > 0) {
407 /*
408 * If there wasn't any memory allocated before, just
409 * allocate it now and get out.
410 */
411 if (ifp->if_broot_bytes == 0) {
412 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
413 ifp->if_broot = kmalloc(new_size,
414 GFP_KERNEL | __GFP_NOFAIL);
415 ifp->if_broot_bytes = (int)new_size;
416 return;
417 }
418
419 /*
420 * If there is already an existing if_broot, then we need
421 * to realloc() it and shift the pointers to their new
422 * location. The records don't change location because
423 * they are kept butted up against the btree block header.
424 */
425 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
426 new_max = cur_max + rec_diff;
427 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
428 ifp->if_broot = krealloc(ifp->if_broot, new_size,
429 GFP_KERNEL | __GFP_NOFAIL);
430 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
431 ifp->if_broot_bytes);
432 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
433 (int)new_size);
434 ifp->if_broot_bytes = (int)new_size;
435 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
436 xfs_inode_fork_size(ip, whichfork));
437 memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
438 return;
439 }
440
441 /*
442 * rec_diff is less than 0. In this case, we are shrinking the
443 * if_broot buffer. It must already exist. If we go to zero
444 * records, just get rid of the root and clear the status bit.
445 */
446 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
447 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
448 new_max = cur_max + rec_diff;
449 ASSERT(new_max >= 0);
450 if (new_max > 0)
451 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
452 else
453 new_size = 0;
454 if (new_size > 0) {
455 new_broot = kmalloc(new_size, GFP_KERNEL | __GFP_NOFAIL);
456 /*
457 * First copy over the btree block header.
458 */
459 memcpy(new_broot, ifp->if_broot,
460 XFS_BMBT_BLOCK_LEN(ip->i_mount));
461 } else {
462 new_broot = NULL;
463 }
464
465 /*
466 * Only copy the records and pointers if there are any.
467 */
468 if (new_max > 0) {
469 /*
470 * First copy the records.
471 */
472 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
473 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
474 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
475
476 /*
477 * Then copy the pointers.
478 */
479 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
480 ifp->if_broot_bytes);
481 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
482 (int)new_size);
483 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
484 }
485 kfree(ifp->if_broot);
486 ifp->if_broot = new_broot;
487 ifp->if_broot_bytes = (int)new_size;
488 if (ifp->if_broot)
489 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
490 xfs_inode_fork_size(ip, whichfork));
491 return;
492}
493
494
495/*
496 * This is called when the amount of space needed for if_data
497 * is increased or decreased. The change in size is indicated by
498 * the number of bytes that need to be added or deleted in the
499 * byte_diff parameter.
500 *
501 * If the amount of space needed has decreased below the size of the
502 * inline buffer, then switch to using the inline buffer. Otherwise,
503 * use krealloc() or kmalloc() to adjust the size of the buffer
504 * to what is needed.
505 *
506 * ip -- the inode whose if_data area is changing
507 * byte_diff -- the change in the number of bytes, positive or negative,
508 * requested for the if_data array.
509 */
510void *
511xfs_idata_realloc(
512 struct xfs_inode *ip,
513 int64_t byte_diff,
514 int whichfork)
515{
516 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
517 int64_t new_size = ifp->if_bytes + byte_diff;
518
519 ASSERT(new_size >= 0);
520 ASSERT(new_size <= xfs_inode_fork_size(ip, whichfork));
521
522 if (byte_diff) {
523 ifp->if_data = krealloc(ifp->if_data, new_size,
524 GFP_KERNEL | __GFP_NOFAIL);
525 if (new_size == 0)
526 ifp->if_data = NULL;
527 ifp->if_bytes = new_size;
528 }
529
530 return ifp->if_data;
531}
532
533/* Free all memory and reset a fork back to its initial state. */
534void
535xfs_idestroy_fork(
536 struct xfs_ifork *ifp)
537{
538 if (ifp->if_broot != NULL) {
539 kfree(ifp->if_broot);
540 ifp->if_broot = NULL;
541 }
542
543 switch (ifp->if_format) {
544 case XFS_DINODE_FMT_LOCAL:
545 kfree(ifp->if_data);
546 ifp->if_data = NULL;
547 break;
548 case XFS_DINODE_FMT_EXTENTS:
549 case XFS_DINODE_FMT_BTREE:
550 if (ifp->if_height)
551 xfs_iext_destroy(ifp);
552 break;
553 }
554}
555
556/*
557 * Convert in-core extents to on-disk form
558 *
559 * In the case of the data fork, the in-core and on-disk fork sizes can be
560 * different due to delayed allocation extents. We only copy on-disk extents
561 * here, so callers must always use the physical fork size to determine the
562 * size of the buffer passed to this routine. We will return the size actually
563 * used.
564 */
565int
566xfs_iextents_copy(
567 struct xfs_inode *ip,
568 struct xfs_bmbt_rec *dp,
569 int whichfork)
570{
571 int state = xfs_bmap_fork_to_state(whichfork);
572 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
573 struct xfs_iext_cursor icur;
574 struct xfs_bmbt_irec rec;
575 int64_t copied = 0;
576
577 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
578 ASSERT(ifp->if_bytes > 0);
579
580 for_each_xfs_iext(ifp, &icur, &rec) {
581 if (isnullstartblock(rec.br_startblock))
582 continue;
583 ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
584 xfs_bmbt_disk_set_all(dp, &rec);
585 trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
586 copied += sizeof(struct xfs_bmbt_rec);
587 dp++;
588 }
589
590 ASSERT(copied > 0);
591 ASSERT(copied <= ifp->if_bytes);
592 return copied;
593}
594
595/*
596 * Each of the following cases stores data into the same region
597 * of the on-disk inode, so only one of them can be valid at
598 * any given time. While it is possible to have conflicting formats
599 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
600 * in EXTENTS format, this can only happen when the fork has
601 * changed formats after being modified but before being flushed.
602 * In these cases, the format always takes precedence, because the
603 * format indicates the current state of the fork.
604 */
605void
606xfs_iflush_fork(
607 struct xfs_inode *ip,
608 struct xfs_dinode *dip,
609 struct xfs_inode_log_item *iip,
610 int whichfork)
611{
612 char *cp;
613 struct xfs_ifork *ifp;
614 xfs_mount_t *mp;
615 static const short brootflag[2] =
616 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
617 static const short dataflag[2] =
618 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
619 static const short extflag[2] =
620 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
621
622 if (!iip)
623 return;
624 ifp = xfs_ifork_ptr(ip, whichfork);
625 /*
626 * This can happen if we gave up in iformat in an error path,
627 * for the attribute fork.
628 */
629 if (!ifp) {
630 ASSERT(whichfork == XFS_ATTR_FORK);
631 return;
632 }
633 cp = XFS_DFORK_PTR(dip, whichfork);
634 mp = ip->i_mount;
635 switch (ifp->if_format) {
636 case XFS_DINODE_FMT_LOCAL:
637 if ((iip->ili_fields & dataflag[whichfork]) &&
638 (ifp->if_bytes > 0)) {
639 ASSERT(ifp->if_data != NULL);
640 ASSERT(ifp->if_bytes <= xfs_inode_fork_size(ip, whichfork));
641 memcpy(cp, ifp->if_data, ifp->if_bytes);
642 }
643 break;
644
645 case XFS_DINODE_FMT_EXTENTS:
646 if ((iip->ili_fields & extflag[whichfork]) &&
647 (ifp->if_bytes > 0)) {
648 ASSERT(ifp->if_nextents > 0);
649 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
650 whichfork);
651 }
652 break;
653
654 case XFS_DINODE_FMT_BTREE:
655 if ((iip->ili_fields & brootflag[whichfork]) &&
656 (ifp->if_broot_bytes > 0)) {
657 ASSERT(ifp->if_broot != NULL);
658 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
659 xfs_inode_fork_size(ip, whichfork));
660 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
661 (xfs_bmdr_block_t *)cp,
662 XFS_DFORK_SIZE(dip, mp, whichfork));
663 }
664 break;
665
666 case XFS_DINODE_FMT_DEV:
667 if (iip->ili_fields & XFS_ILOG_DEV) {
668 ASSERT(whichfork == XFS_DATA_FORK);
669 xfs_dinode_put_rdev(dip,
670 linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
671 }
672 break;
673
674 default:
675 ASSERT(0);
676 break;
677 }
678}
679
680/* Convert bmap state flags to an inode fork. */
681struct xfs_ifork *
682xfs_iext_state_to_fork(
683 struct xfs_inode *ip,
684 int state)
685{
686 if (state & BMAP_COWFORK)
687 return ip->i_cowfp;
688 else if (state & BMAP_ATTRFORK)
689 return &ip->i_af;
690 return &ip->i_df;
691}
692
693/*
694 * Initialize an inode's copy-on-write fork.
695 */
696void
697xfs_ifork_init_cow(
698 struct xfs_inode *ip)
699{
700 if (ip->i_cowfp)
701 return;
702
703 ip->i_cowfp = kmem_cache_zalloc(xfs_ifork_cache,
704 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
705 ip->i_cowfp->if_format = XFS_DINODE_FMT_EXTENTS;
706}
707
708/* Verify the inline contents of the data fork of an inode. */
709int
710xfs_ifork_verify_local_data(
711 struct xfs_inode *ip)
712{
713 xfs_failaddr_t fa = NULL;
714
715 switch (VFS_I(ip)->i_mode & S_IFMT) {
716 case S_IFDIR: {
717 struct xfs_mount *mp = ip->i_mount;
718 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
719 struct xfs_dir2_sf_hdr *sfp = ifp->if_data;
720
721 fa = xfs_dir2_sf_verify(mp, sfp, ifp->if_bytes);
722 break;
723 }
724 case S_IFLNK: {
725 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
726
727 fa = xfs_symlink_shortform_verify(ifp->if_data, ifp->if_bytes);
728 break;
729 }
730 default:
731 break;
732 }
733
734 if (fa) {
735 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
736 ip->i_df.if_data, ip->i_df.if_bytes, fa);
737 return -EFSCORRUPTED;
738 }
739
740 return 0;
741}
742
743/* Verify the inline contents of the attr fork of an inode. */
744int
745xfs_ifork_verify_local_attr(
746 struct xfs_inode *ip)
747{
748 struct xfs_ifork *ifp = &ip->i_af;
749 xfs_failaddr_t fa;
750
751 if (!xfs_inode_has_attr_fork(ip)) {
752 fa = __this_address;
753 } else {
754 struct xfs_ifork *ifp = &ip->i_af;
755
756 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
757 fa = xfs_attr_shortform_verify(ifp->if_data, ifp->if_bytes);
758 }
759 if (fa) {
760 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
761 ifp->if_data, ifp->if_bytes, fa);
762 return -EFSCORRUPTED;
763 }
764
765 return 0;
766}
767
768int
769xfs_iext_count_may_overflow(
770 struct xfs_inode *ip,
771 int whichfork,
772 int nr_to_add)
773{
774 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
775 uint64_t max_exts;
776 uint64_t nr_exts;
777
778 if (whichfork == XFS_COW_FORK)
779 return 0;
780
781 max_exts = xfs_iext_max_nextents(xfs_inode_has_large_extent_counts(ip),
782 whichfork);
783
784 if (XFS_TEST_ERROR(false, ip->i_mount, XFS_ERRTAG_REDUCE_MAX_IEXTENTS))
785 max_exts = 10;
786
787 nr_exts = ifp->if_nextents + nr_to_add;
788 if (nr_exts < ifp->if_nextents || nr_exts > max_exts)
789 return -EFBIG;
790
791 return 0;
792}
793
794/*
795 * Upgrade this inode's extent counter fields to be able to handle a potential
796 * increase in the extent count by nr_to_add. Normally this is the same
797 * quantity that caused xfs_iext_count_may_overflow() to return -EFBIG.
798 */
799int
800xfs_iext_count_upgrade(
801 struct xfs_trans *tp,
802 struct xfs_inode *ip,
803 uint nr_to_add)
804{
805 ASSERT(nr_to_add <= XFS_MAX_EXTCNT_UPGRADE_NR);
806
807 if (!xfs_has_large_extent_counts(ip->i_mount) ||
808 xfs_inode_has_large_extent_counts(ip) ||
809 XFS_TEST_ERROR(false, ip->i_mount, XFS_ERRTAG_REDUCE_MAX_IEXTENTS))
810 return -EFBIG;
811
812 ip->i_diflags2 |= XFS_DIFLAG2_NREXT64;
813 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
814
815 return 0;
816}
817
818/* Decide if a file mapping is on the realtime device or not. */
819bool
820xfs_ifork_is_realtime(
821 struct xfs_inode *ip,
822 int whichfork)
823{
824 return XFS_IS_REALTIME_INODE(ip) && whichfork != XFS_ATTR_FORK;
825}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_trans.h"
16#include "xfs_inode_item.h"
17#include "xfs_btree.h"
18#include "xfs_bmap_btree.h"
19#include "xfs_bmap.h"
20#include "xfs_error.h"
21#include "xfs_trace.h"
22#include "xfs_da_format.h"
23#include "xfs_da_btree.h"
24#include "xfs_dir2_priv.h"
25#include "xfs_attr_leaf.h"
26
27kmem_zone_t *xfs_ifork_zone;
28
29void
30xfs_init_local_fork(
31 struct xfs_inode *ip,
32 int whichfork,
33 const void *data,
34 int64_t size)
35{
36 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
37 int mem_size = size, real_size = 0;
38 bool zero_terminate;
39
40 /*
41 * If we are using the local fork to store a symlink body we need to
42 * zero-terminate it so that we can pass it back to the VFS directly.
43 * Overallocate the in-memory fork by one for that and add a zero
44 * to terminate it below.
45 */
46 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
47 if (zero_terminate)
48 mem_size++;
49
50 if (size) {
51 real_size = roundup(mem_size, 4);
52 ifp->if_u1.if_data = kmem_alloc(real_size, KM_NOFS);
53 memcpy(ifp->if_u1.if_data, data, size);
54 if (zero_terminate)
55 ifp->if_u1.if_data[size] = '\0';
56 } else {
57 ifp->if_u1.if_data = NULL;
58 }
59
60 ifp->if_bytes = size;
61 ifp->if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
62 ifp->if_flags |= XFS_IFINLINE;
63}
64
65/*
66 * The file is in-lined in the on-disk inode.
67 */
68STATIC int
69xfs_iformat_local(
70 xfs_inode_t *ip,
71 xfs_dinode_t *dip,
72 int whichfork,
73 int size)
74{
75 /*
76 * If the size is unreasonable, then something
77 * is wrong and we just bail out rather than crash in
78 * kmem_alloc() or memcpy() below.
79 */
80 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
81 xfs_warn(ip->i_mount,
82 "corrupt inode %Lu (bad size %d for local fork, size = %zd).",
83 (unsigned long long) ip->i_ino, size,
84 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
85 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
86 "xfs_iformat_local", dip, sizeof(*dip),
87 __this_address);
88 return -EFSCORRUPTED;
89 }
90
91 xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
92 return 0;
93}
94
95/*
96 * The file consists of a set of extents all of which fit into the on-disk
97 * inode.
98 */
99STATIC int
100xfs_iformat_extents(
101 struct xfs_inode *ip,
102 struct xfs_dinode *dip,
103 int whichfork)
104{
105 struct xfs_mount *mp = ip->i_mount;
106 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
107 int state = xfs_bmap_fork_to_state(whichfork);
108 int nex = XFS_DFORK_NEXTENTS(dip, whichfork);
109 int size = nex * sizeof(xfs_bmbt_rec_t);
110 struct xfs_iext_cursor icur;
111 struct xfs_bmbt_rec *dp;
112 struct xfs_bmbt_irec new;
113 int i;
114
115 /*
116 * If the number of extents is unreasonable, then something is wrong and
117 * we just bail out rather than crash in kmem_alloc() or memcpy() below.
118 */
119 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, mp, whichfork))) {
120 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
121 (unsigned long long) ip->i_ino, nex);
122 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
123 "xfs_iformat_extents(1)", dip, sizeof(*dip),
124 __this_address);
125 return -EFSCORRUPTED;
126 }
127
128 ifp->if_bytes = 0;
129 ifp->if_u1.if_root = NULL;
130 ifp->if_height = 0;
131 if (size) {
132 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
133
134 xfs_iext_first(ifp, &icur);
135 for (i = 0; i < nex; i++, dp++) {
136 xfs_failaddr_t fa;
137
138 xfs_bmbt_disk_get_all(dp, &new);
139 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
140 if (fa) {
141 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
142 "xfs_iformat_extents(2)",
143 dp, sizeof(*dp), fa);
144 return -EFSCORRUPTED;
145 }
146
147 xfs_iext_insert(ip, &icur, &new, state);
148 trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
149 xfs_iext_next(ifp, &icur);
150 }
151 }
152 ifp->if_flags |= XFS_IFEXTENTS;
153 return 0;
154}
155
156/*
157 * The file has too many extents to fit into
158 * the inode, so they are in B-tree format.
159 * Allocate a buffer for the root of the B-tree
160 * and copy the root into it. The i_extents
161 * field will remain NULL until all of the
162 * extents are read in (when they are needed).
163 */
164STATIC int
165xfs_iformat_btree(
166 xfs_inode_t *ip,
167 xfs_dinode_t *dip,
168 int whichfork)
169{
170 struct xfs_mount *mp = ip->i_mount;
171 xfs_bmdr_block_t *dfp;
172 struct xfs_ifork *ifp;
173 /* REFERENCED */
174 int nrecs;
175 int size;
176 int level;
177
178 ifp = XFS_IFORK_PTR(ip, whichfork);
179 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
180 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
181 nrecs = be16_to_cpu(dfp->bb_numrecs);
182 level = be16_to_cpu(dfp->bb_level);
183
184 /*
185 * blow out if -- fork has less extents than can fit in
186 * fork (fork shouldn't be a btree format), root btree
187 * block has more records than can fit into the fork,
188 * or the number of extents is greater than the number of
189 * blocks.
190 */
191 if (unlikely(ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork) ||
192 nrecs == 0 ||
193 XFS_BMDR_SPACE_CALC(nrecs) >
194 XFS_DFORK_SIZE(dip, mp, whichfork) ||
195 ifp->if_nextents > ip->i_d.di_nblocks) ||
196 level == 0 || level > XFS_BTREE_MAXLEVELS) {
197 xfs_warn(mp, "corrupt inode %Lu (btree).",
198 (unsigned long long) ip->i_ino);
199 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
200 "xfs_iformat_btree", dfp, size,
201 __this_address);
202 return -EFSCORRUPTED;
203 }
204
205 ifp->if_broot_bytes = size;
206 ifp->if_broot = kmem_alloc(size, KM_NOFS);
207 ASSERT(ifp->if_broot != NULL);
208 /*
209 * Copy and convert from the on-disk structure
210 * to the in-memory structure.
211 */
212 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
213 ifp->if_broot, size);
214 ifp->if_flags &= ~XFS_IFEXTENTS;
215 ifp->if_flags |= XFS_IFBROOT;
216
217 ifp->if_bytes = 0;
218 ifp->if_u1.if_root = NULL;
219 ifp->if_height = 0;
220 return 0;
221}
222
223int
224xfs_iformat_data_fork(
225 struct xfs_inode *ip,
226 struct xfs_dinode *dip)
227{
228 struct inode *inode = VFS_I(ip);
229 int error;
230
231 /*
232 * Initialize the extent count early, as the per-format routines may
233 * depend on it.
234 */
235 ip->i_df.if_format = dip->di_format;
236 ip->i_df.if_nextents = be32_to_cpu(dip->di_nextents);
237
238 switch (inode->i_mode & S_IFMT) {
239 case S_IFIFO:
240 case S_IFCHR:
241 case S_IFBLK:
242 case S_IFSOCK:
243 ip->i_d.di_size = 0;
244 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
245 return 0;
246 case S_IFREG:
247 case S_IFLNK:
248 case S_IFDIR:
249 switch (ip->i_df.if_format) {
250 case XFS_DINODE_FMT_LOCAL:
251 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK,
252 be64_to_cpu(dip->di_size));
253 if (!error)
254 error = xfs_ifork_verify_local_data(ip);
255 return error;
256 case XFS_DINODE_FMT_EXTENTS:
257 return xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
258 case XFS_DINODE_FMT_BTREE:
259 return xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
260 default:
261 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
262 dip, sizeof(*dip), __this_address);
263 return -EFSCORRUPTED;
264 }
265 break;
266 default:
267 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
268 sizeof(*dip), __this_address);
269 return -EFSCORRUPTED;
270 }
271}
272
273static uint16_t
274xfs_dfork_attr_shortform_size(
275 struct xfs_dinode *dip)
276{
277 struct xfs_attr_shortform *atp =
278 (struct xfs_attr_shortform *)XFS_DFORK_APTR(dip);
279
280 return be16_to_cpu(atp->hdr.totsize);
281}
282
283int
284xfs_iformat_attr_fork(
285 struct xfs_inode *ip,
286 struct xfs_dinode *dip)
287{
288 int error = 0;
289
290 /*
291 * Initialize the extent count early, as the per-format routines may
292 * depend on it.
293 */
294 ip->i_afp = kmem_cache_zalloc(xfs_ifork_zone, GFP_NOFS | __GFP_NOFAIL);
295 ip->i_afp->if_format = dip->di_aformat;
296 if (unlikely(ip->i_afp->if_format == 0)) /* pre IRIX 6.2 file system */
297 ip->i_afp->if_format = XFS_DINODE_FMT_EXTENTS;
298 ip->i_afp->if_nextents = be16_to_cpu(dip->di_anextents);
299
300 switch (ip->i_afp->if_format) {
301 case XFS_DINODE_FMT_LOCAL:
302 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK,
303 xfs_dfork_attr_shortform_size(dip));
304 if (!error)
305 error = xfs_ifork_verify_local_attr(ip);
306 break;
307 case XFS_DINODE_FMT_EXTENTS:
308 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
309 break;
310 case XFS_DINODE_FMT_BTREE:
311 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
312 break;
313 default:
314 xfs_inode_verifier_error(ip, error, __func__, dip,
315 sizeof(*dip), __this_address);
316 error = -EFSCORRUPTED;
317 break;
318 }
319
320 if (error) {
321 kmem_cache_free(xfs_ifork_zone, ip->i_afp);
322 ip->i_afp = NULL;
323 }
324 return error;
325}
326
327/*
328 * Reallocate the space for if_broot based on the number of records
329 * being added or deleted as indicated in rec_diff. Move the records
330 * and pointers in if_broot to fit the new size. When shrinking this
331 * will eliminate holes between the records and pointers created by
332 * the caller. When growing this will create holes to be filled in
333 * by the caller.
334 *
335 * The caller must not request to add more records than would fit in
336 * the on-disk inode root. If the if_broot is currently NULL, then
337 * if we are adding records, one will be allocated. The caller must also
338 * not request that the number of records go below zero, although
339 * it can go to zero.
340 *
341 * ip -- the inode whose if_broot area is changing
342 * ext_diff -- the change in the number of records, positive or negative,
343 * requested for the if_broot array.
344 */
345void
346xfs_iroot_realloc(
347 xfs_inode_t *ip,
348 int rec_diff,
349 int whichfork)
350{
351 struct xfs_mount *mp = ip->i_mount;
352 int cur_max;
353 struct xfs_ifork *ifp;
354 struct xfs_btree_block *new_broot;
355 int new_max;
356 size_t new_size;
357 char *np;
358 char *op;
359
360 /*
361 * Handle the degenerate case quietly.
362 */
363 if (rec_diff == 0) {
364 return;
365 }
366
367 ifp = XFS_IFORK_PTR(ip, whichfork);
368 if (rec_diff > 0) {
369 /*
370 * If there wasn't any memory allocated before, just
371 * allocate it now and get out.
372 */
373 if (ifp->if_broot_bytes == 0) {
374 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
375 ifp->if_broot = kmem_alloc(new_size, KM_NOFS);
376 ifp->if_broot_bytes = (int)new_size;
377 return;
378 }
379
380 /*
381 * If there is already an existing if_broot, then we need
382 * to realloc() it and shift the pointers to their new
383 * location. The records don't change location because
384 * they are kept butted up against the btree block header.
385 */
386 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
387 new_max = cur_max + rec_diff;
388 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
389 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
390 KM_NOFS);
391 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
392 ifp->if_broot_bytes);
393 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
394 (int)new_size);
395 ifp->if_broot_bytes = (int)new_size;
396 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
397 XFS_IFORK_SIZE(ip, whichfork));
398 memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
399 return;
400 }
401
402 /*
403 * rec_diff is less than 0. In this case, we are shrinking the
404 * if_broot buffer. It must already exist. If we go to zero
405 * records, just get rid of the root and clear the status bit.
406 */
407 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
408 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
409 new_max = cur_max + rec_diff;
410 ASSERT(new_max >= 0);
411 if (new_max > 0)
412 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
413 else
414 new_size = 0;
415 if (new_size > 0) {
416 new_broot = kmem_alloc(new_size, KM_NOFS);
417 /*
418 * First copy over the btree block header.
419 */
420 memcpy(new_broot, ifp->if_broot,
421 XFS_BMBT_BLOCK_LEN(ip->i_mount));
422 } else {
423 new_broot = NULL;
424 ifp->if_flags &= ~XFS_IFBROOT;
425 }
426
427 /*
428 * Only copy the records and pointers if there are any.
429 */
430 if (new_max > 0) {
431 /*
432 * First copy the records.
433 */
434 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
435 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
436 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
437
438 /*
439 * Then copy the pointers.
440 */
441 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
442 ifp->if_broot_bytes);
443 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
444 (int)new_size);
445 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
446 }
447 kmem_free(ifp->if_broot);
448 ifp->if_broot = new_broot;
449 ifp->if_broot_bytes = (int)new_size;
450 if (ifp->if_broot)
451 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
452 XFS_IFORK_SIZE(ip, whichfork));
453 return;
454}
455
456
457/*
458 * This is called when the amount of space needed for if_data
459 * is increased or decreased. The change in size is indicated by
460 * the number of bytes that need to be added or deleted in the
461 * byte_diff parameter.
462 *
463 * If the amount of space needed has decreased below the size of the
464 * inline buffer, then switch to using the inline buffer. Otherwise,
465 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
466 * to what is needed.
467 *
468 * ip -- the inode whose if_data area is changing
469 * byte_diff -- the change in the number of bytes, positive or negative,
470 * requested for the if_data array.
471 */
472void
473xfs_idata_realloc(
474 struct xfs_inode *ip,
475 int64_t byte_diff,
476 int whichfork)
477{
478 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
479 int64_t new_size = ifp->if_bytes + byte_diff;
480
481 ASSERT(new_size >= 0);
482 ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));
483
484 if (byte_diff == 0)
485 return;
486
487 if (new_size == 0) {
488 kmem_free(ifp->if_u1.if_data);
489 ifp->if_u1.if_data = NULL;
490 ifp->if_bytes = 0;
491 return;
492 }
493
494 /*
495 * For inline data, the underlying buffer must be a multiple of 4 bytes
496 * in size so that it can be logged and stay on word boundaries.
497 * We enforce that here.
498 */
499 ifp->if_u1.if_data = kmem_realloc(ifp->if_u1.if_data,
500 roundup(new_size, 4), KM_NOFS);
501 ifp->if_bytes = new_size;
502}
503
504void
505xfs_idestroy_fork(
506 struct xfs_ifork *ifp)
507{
508 if (ifp->if_broot != NULL) {
509 kmem_free(ifp->if_broot);
510 ifp->if_broot = NULL;
511 }
512
513 /*
514 * If the format is local, then we can't have an extents array so just
515 * look for an inline data array. If we're not local then we may or may
516 * not have an extents list, so check and free it up if we do.
517 */
518 if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
519 kmem_free(ifp->if_u1.if_data);
520 ifp->if_u1.if_data = NULL;
521 } else if (ifp->if_flags & XFS_IFEXTENTS) {
522 if (ifp->if_height)
523 xfs_iext_destroy(ifp);
524 }
525}
526
527/*
528 * Convert in-core extents to on-disk form
529 *
530 * In the case of the data fork, the in-core and on-disk fork sizes can be
531 * different due to delayed allocation extents. We only copy on-disk extents
532 * here, so callers must always use the physical fork size to determine the
533 * size of the buffer passed to this routine. We will return the size actually
534 * used.
535 */
536int
537xfs_iextents_copy(
538 struct xfs_inode *ip,
539 struct xfs_bmbt_rec *dp,
540 int whichfork)
541{
542 int state = xfs_bmap_fork_to_state(whichfork);
543 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
544 struct xfs_iext_cursor icur;
545 struct xfs_bmbt_irec rec;
546 int64_t copied = 0;
547
548 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
549 ASSERT(ifp->if_bytes > 0);
550
551 for_each_xfs_iext(ifp, &icur, &rec) {
552 if (isnullstartblock(rec.br_startblock))
553 continue;
554 ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
555 xfs_bmbt_disk_set_all(dp, &rec);
556 trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
557 copied += sizeof(struct xfs_bmbt_rec);
558 dp++;
559 }
560
561 ASSERT(copied > 0);
562 ASSERT(copied <= ifp->if_bytes);
563 return copied;
564}
565
566/*
567 * Each of the following cases stores data into the same region
568 * of the on-disk inode, so only one of them can be valid at
569 * any given time. While it is possible to have conflicting formats
570 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
571 * in EXTENTS format, this can only happen when the fork has
572 * changed formats after being modified but before being flushed.
573 * In these cases, the format always takes precedence, because the
574 * format indicates the current state of the fork.
575 */
576void
577xfs_iflush_fork(
578 xfs_inode_t *ip,
579 xfs_dinode_t *dip,
580 struct xfs_inode_log_item *iip,
581 int whichfork)
582{
583 char *cp;
584 struct xfs_ifork *ifp;
585 xfs_mount_t *mp;
586 static const short brootflag[2] =
587 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
588 static const short dataflag[2] =
589 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
590 static const short extflag[2] =
591 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
592
593 if (!iip)
594 return;
595 ifp = XFS_IFORK_PTR(ip, whichfork);
596 /*
597 * This can happen if we gave up in iformat in an error path,
598 * for the attribute fork.
599 */
600 if (!ifp) {
601 ASSERT(whichfork == XFS_ATTR_FORK);
602 return;
603 }
604 cp = XFS_DFORK_PTR(dip, whichfork);
605 mp = ip->i_mount;
606 switch (ifp->if_format) {
607 case XFS_DINODE_FMT_LOCAL:
608 if ((iip->ili_fields & dataflag[whichfork]) &&
609 (ifp->if_bytes > 0)) {
610 ASSERT(ifp->if_u1.if_data != NULL);
611 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
612 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
613 }
614 break;
615
616 case XFS_DINODE_FMT_EXTENTS:
617 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
618 !(iip->ili_fields & extflag[whichfork]));
619 if ((iip->ili_fields & extflag[whichfork]) &&
620 (ifp->if_bytes > 0)) {
621 ASSERT(ifp->if_nextents > 0);
622 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
623 whichfork);
624 }
625 break;
626
627 case XFS_DINODE_FMT_BTREE:
628 if ((iip->ili_fields & brootflag[whichfork]) &&
629 (ifp->if_broot_bytes > 0)) {
630 ASSERT(ifp->if_broot != NULL);
631 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
632 XFS_IFORK_SIZE(ip, whichfork));
633 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
634 (xfs_bmdr_block_t *)cp,
635 XFS_DFORK_SIZE(dip, mp, whichfork));
636 }
637 break;
638
639 case XFS_DINODE_FMT_DEV:
640 if (iip->ili_fields & XFS_ILOG_DEV) {
641 ASSERT(whichfork == XFS_DATA_FORK);
642 xfs_dinode_put_rdev(dip,
643 linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
644 }
645 break;
646
647 default:
648 ASSERT(0);
649 break;
650 }
651}
652
653/* Convert bmap state flags to an inode fork. */
654struct xfs_ifork *
655xfs_iext_state_to_fork(
656 struct xfs_inode *ip,
657 int state)
658{
659 if (state & BMAP_COWFORK)
660 return ip->i_cowfp;
661 else if (state & BMAP_ATTRFORK)
662 return ip->i_afp;
663 return &ip->i_df;
664}
665
666/*
667 * Initialize an inode's copy-on-write fork.
668 */
669void
670xfs_ifork_init_cow(
671 struct xfs_inode *ip)
672{
673 if (ip->i_cowfp)
674 return;
675
676 ip->i_cowfp = kmem_cache_zalloc(xfs_ifork_zone,
677 GFP_NOFS | __GFP_NOFAIL);
678 ip->i_cowfp->if_flags = XFS_IFEXTENTS;
679 ip->i_cowfp->if_format = XFS_DINODE_FMT_EXTENTS;
680}
681
682/* Verify the inline contents of the data fork of an inode. */
683int
684xfs_ifork_verify_local_data(
685 struct xfs_inode *ip)
686{
687 xfs_failaddr_t fa = NULL;
688
689 switch (VFS_I(ip)->i_mode & S_IFMT) {
690 case S_IFDIR:
691 fa = xfs_dir2_sf_verify(ip);
692 break;
693 case S_IFLNK:
694 fa = xfs_symlink_shortform_verify(ip);
695 break;
696 default:
697 break;
698 }
699
700 if (fa) {
701 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
702 ip->i_df.if_u1.if_data, ip->i_df.if_bytes, fa);
703 return -EFSCORRUPTED;
704 }
705
706 return 0;
707}
708
709/* Verify the inline contents of the attr fork of an inode. */
710int
711xfs_ifork_verify_local_attr(
712 struct xfs_inode *ip)
713{
714 struct xfs_ifork *ifp = ip->i_afp;
715 xfs_failaddr_t fa;
716
717 if (!ifp)
718 fa = __this_address;
719 else
720 fa = xfs_attr_shortform_verify(ip);
721
722 if (fa) {
723 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
724 ifp ? ifp->if_u1.if_data : NULL,
725 ifp ? ifp->if_bytes : 0, fa);
726 return -EFSCORRUPTED;
727 }
728
729 return 0;
730}