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#include "xfs_types.h"
27#include "xfs_errortag.h"
28
29kmem_zone_t *xfs_ifork_zone;
30
31void
32xfs_init_local_fork(
33 struct xfs_inode *ip,
34 int whichfork,
35 const void *data,
36 int64_t size)
37{
38 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
39 int mem_size = size, real_size = 0;
40 bool zero_terminate;
41
42 /*
43 * If we are using the local fork to store a symlink body we need to
44 * zero-terminate it so that we can pass it back to the VFS directly.
45 * Overallocate the in-memory fork by one for that and add a zero
46 * to terminate it below.
47 */
48 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
49 if (zero_terminate)
50 mem_size++;
51
52 if (size) {
53 real_size = roundup(mem_size, 4);
54 ifp->if_u1.if_data = kmem_alloc(real_size, KM_NOFS);
55 memcpy(ifp->if_u1.if_data, data, size);
56 if (zero_terminate)
57 ifp->if_u1.if_data[size] = '\0';
58 } else {
59 ifp->if_u1.if_data = NULL;
60 }
61
62 ifp->if_bytes = size;
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 return 0;
153}
154
155/*
156 * The file has too many extents to fit into
157 * the inode, so they are in B-tree format.
158 * Allocate a buffer for the root of the B-tree
159 * and copy the root into it. The i_extents
160 * field will remain NULL until all of the
161 * extents are read in (when they are needed).
162 */
163STATIC int
164xfs_iformat_btree(
165 xfs_inode_t *ip,
166 xfs_dinode_t *dip,
167 int whichfork)
168{
169 struct xfs_mount *mp = ip->i_mount;
170 xfs_bmdr_block_t *dfp;
171 struct xfs_ifork *ifp;
172 /* REFERENCED */
173 int nrecs;
174 int size;
175 int level;
176
177 ifp = XFS_IFORK_PTR(ip, whichfork);
178 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
179 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
180 nrecs = be16_to_cpu(dfp->bb_numrecs);
181 level = be16_to_cpu(dfp->bb_level);
182
183 /*
184 * blow out if -- fork has less extents than can fit in
185 * fork (fork shouldn't be a btree format), root btree
186 * block has more records than can fit into the fork,
187 * or the number of extents is greater than the number of
188 * blocks.
189 */
190 if (unlikely(ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork) ||
191 nrecs == 0 ||
192 XFS_BMDR_SPACE_CALC(nrecs) >
193 XFS_DFORK_SIZE(dip, mp, whichfork) ||
194 ifp->if_nextents > ip->i_nblocks) ||
195 level == 0 || level > XFS_BM_MAXLEVELS(mp, whichfork)) {
196 xfs_warn(mp, "corrupt inode %Lu (btree).",
197 (unsigned long long) ip->i_ino);
198 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
199 "xfs_iformat_btree", dfp, size,
200 __this_address);
201 return -EFSCORRUPTED;
202 }
203
204 ifp->if_broot_bytes = size;
205 ifp->if_broot = kmem_alloc(size, KM_NOFS);
206 ASSERT(ifp->if_broot != NULL);
207 /*
208 * Copy and convert from the on-disk structure
209 * to the in-memory structure.
210 */
211 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
212 ifp->if_broot, size);
213
214 ifp->if_bytes = 0;
215 ifp->if_u1.if_root = NULL;
216 ifp->if_height = 0;
217 return 0;
218}
219
220int
221xfs_iformat_data_fork(
222 struct xfs_inode *ip,
223 struct xfs_dinode *dip)
224{
225 struct inode *inode = VFS_I(ip);
226 int error;
227
228 /*
229 * Initialize the extent count early, as the per-format routines may
230 * depend on it.
231 */
232 ip->i_df.if_format = dip->di_format;
233 ip->i_df.if_nextents = be32_to_cpu(dip->di_nextents);
234
235 switch (inode->i_mode & S_IFMT) {
236 case S_IFIFO:
237 case S_IFCHR:
238 case S_IFBLK:
239 case S_IFSOCK:
240 ip->i_disk_size = 0;
241 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
242 return 0;
243 case S_IFREG:
244 case S_IFLNK:
245 case S_IFDIR:
246 switch (ip->i_df.if_format) {
247 case XFS_DINODE_FMT_LOCAL:
248 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK,
249 be64_to_cpu(dip->di_size));
250 if (!error)
251 error = xfs_ifork_verify_local_data(ip);
252 return error;
253 case XFS_DINODE_FMT_EXTENTS:
254 return xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
255 case XFS_DINODE_FMT_BTREE:
256 return xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
257 default:
258 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
259 dip, sizeof(*dip), __this_address);
260 return -EFSCORRUPTED;
261 }
262 break;
263 default:
264 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
265 sizeof(*dip), __this_address);
266 return -EFSCORRUPTED;
267 }
268}
269
270static uint16_t
271xfs_dfork_attr_shortform_size(
272 struct xfs_dinode *dip)
273{
274 struct xfs_attr_shortform *atp =
275 (struct xfs_attr_shortform *)XFS_DFORK_APTR(dip);
276
277 return be16_to_cpu(atp->hdr.totsize);
278}
279
280struct xfs_ifork *
281xfs_ifork_alloc(
282 enum xfs_dinode_fmt format,
283 xfs_extnum_t nextents)
284{
285 struct xfs_ifork *ifp;
286
287 ifp = kmem_cache_zalloc(xfs_ifork_zone, GFP_NOFS | __GFP_NOFAIL);
288 ifp->if_format = format;
289 ifp->if_nextents = nextents;
290 return ifp;
291}
292
293int
294xfs_iformat_attr_fork(
295 struct xfs_inode *ip,
296 struct xfs_dinode *dip)
297{
298 int error = 0;
299
300 /*
301 * Initialize the extent count early, as the per-format routines may
302 * depend on it.
303 */
304 ip->i_afp = xfs_ifork_alloc(dip->di_aformat,
305 be16_to_cpu(dip->di_anextents));
306
307 switch (ip->i_afp->if_format) {
308 case XFS_DINODE_FMT_LOCAL:
309 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK,
310 xfs_dfork_attr_shortform_size(dip));
311 if (!error)
312 error = xfs_ifork_verify_local_attr(ip);
313 break;
314 case XFS_DINODE_FMT_EXTENTS:
315 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
316 break;
317 case XFS_DINODE_FMT_BTREE:
318 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
319 break;
320 default:
321 xfs_inode_verifier_error(ip, error, __func__, dip,
322 sizeof(*dip), __this_address);
323 error = -EFSCORRUPTED;
324 break;
325 }
326
327 if (error) {
328 kmem_cache_free(xfs_ifork_zone, ip->i_afp);
329 ip->i_afp = NULL;
330 }
331 return error;
332}
333
334/*
335 * Reallocate the space for if_broot based on the number of records
336 * being added or deleted as indicated in rec_diff. Move the records
337 * and pointers in if_broot to fit the new size. When shrinking this
338 * will eliminate holes between the records and pointers created by
339 * the caller. When growing this will create holes to be filled in
340 * by the caller.
341 *
342 * The caller must not request to add more records than would fit in
343 * the on-disk inode root. If the if_broot is currently NULL, then
344 * if we are adding records, one will be allocated. The caller must also
345 * not request that the number of records go below zero, although
346 * it can go to zero.
347 *
348 * ip -- the inode whose if_broot area is changing
349 * ext_diff -- the change in the number of records, positive or negative,
350 * requested for the if_broot array.
351 */
352void
353xfs_iroot_realloc(
354 xfs_inode_t *ip,
355 int rec_diff,
356 int whichfork)
357{
358 struct xfs_mount *mp = ip->i_mount;
359 int cur_max;
360 struct xfs_ifork *ifp;
361 struct xfs_btree_block *new_broot;
362 int new_max;
363 size_t new_size;
364 char *np;
365 char *op;
366
367 /*
368 * Handle the degenerate case quietly.
369 */
370 if (rec_diff == 0) {
371 return;
372 }
373
374 ifp = XFS_IFORK_PTR(ip, whichfork);
375 if (rec_diff > 0) {
376 /*
377 * If there wasn't any memory allocated before, just
378 * allocate it now and get out.
379 */
380 if (ifp->if_broot_bytes == 0) {
381 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
382 ifp->if_broot = kmem_alloc(new_size, KM_NOFS);
383 ifp->if_broot_bytes = (int)new_size;
384 return;
385 }
386
387 /*
388 * If there is already an existing if_broot, then we need
389 * to realloc() it and shift the pointers to their new
390 * location. The records don't change location because
391 * they are kept butted up against the btree block header.
392 */
393 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
394 new_max = cur_max + rec_diff;
395 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
396 ifp->if_broot = krealloc(ifp->if_broot, new_size,
397 GFP_NOFS | __GFP_NOFAIL);
398 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
399 ifp->if_broot_bytes);
400 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
401 (int)new_size);
402 ifp->if_broot_bytes = (int)new_size;
403 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
404 XFS_IFORK_SIZE(ip, whichfork));
405 memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
406 return;
407 }
408
409 /*
410 * rec_diff is less than 0. In this case, we are shrinking the
411 * if_broot buffer. It must already exist. If we go to zero
412 * records, just get rid of the root and clear the status bit.
413 */
414 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
415 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
416 new_max = cur_max + rec_diff;
417 ASSERT(new_max >= 0);
418 if (new_max > 0)
419 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
420 else
421 new_size = 0;
422 if (new_size > 0) {
423 new_broot = kmem_alloc(new_size, KM_NOFS);
424 /*
425 * First copy over the btree block header.
426 */
427 memcpy(new_broot, ifp->if_broot,
428 XFS_BMBT_BLOCK_LEN(ip->i_mount));
429 } else {
430 new_broot = NULL;
431 }
432
433 /*
434 * Only copy the records and pointers if there are any.
435 */
436 if (new_max > 0) {
437 /*
438 * First copy the records.
439 */
440 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
441 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
442 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
443
444 /*
445 * Then copy the pointers.
446 */
447 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
448 ifp->if_broot_bytes);
449 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
450 (int)new_size);
451 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
452 }
453 kmem_free(ifp->if_broot);
454 ifp->if_broot = new_broot;
455 ifp->if_broot_bytes = (int)new_size;
456 if (ifp->if_broot)
457 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
458 XFS_IFORK_SIZE(ip, whichfork));
459 return;
460}
461
462
463/*
464 * This is called when the amount of space needed for if_data
465 * is increased or decreased. The change in size is indicated by
466 * the number of bytes that need to be added or deleted in the
467 * byte_diff parameter.
468 *
469 * If the amount of space needed has decreased below the size of the
470 * inline buffer, then switch to using the inline buffer. Otherwise,
471 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
472 * to what is needed.
473 *
474 * ip -- the inode whose if_data area is changing
475 * byte_diff -- the change in the number of bytes, positive or negative,
476 * requested for the if_data array.
477 */
478void
479xfs_idata_realloc(
480 struct xfs_inode *ip,
481 int64_t byte_diff,
482 int whichfork)
483{
484 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
485 int64_t new_size = ifp->if_bytes + byte_diff;
486
487 ASSERT(new_size >= 0);
488 ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));
489
490 if (byte_diff == 0)
491 return;
492
493 if (new_size == 0) {
494 kmem_free(ifp->if_u1.if_data);
495 ifp->if_u1.if_data = NULL;
496 ifp->if_bytes = 0;
497 return;
498 }
499
500 /*
501 * For inline data, the underlying buffer must be a multiple of 4 bytes
502 * in size so that it can be logged and stay on word boundaries.
503 * We enforce that here.
504 */
505 ifp->if_u1.if_data = krealloc(ifp->if_u1.if_data, roundup(new_size, 4),
506 GFP_NOFS | __GFP_NOFAIL);
507 ifp->if_bytes = new_size;
508}
509
510void
511xfs_idestroy_fork(
512 struct xfs_ifork *ifp)
513{
514 if (ifp->if_broot != NULL) {
515 kmem_free(ifp->if_broot);
516 ifp->if_broot = NULL;
517 }
518
519 switch (ifp->if_format) {
520 case XFS_DINODE_FMT_LOCAL:
521 kmem_free(ifp->if_u1.if_data);
522 ifp->if_u1.if_data = NULL;
523 break;
524 case XFS_DINODE_FMT_EXTENTS:
525 case XFS_DINODE_FMT_BTREE:
526 if (ifp->if_height)
527 xfs_iext_destroy(ifp);
528 break;
529 }
530}
531
532/*
533 * Convert in-core extents to on-disk form
534 *
535 * In the case of the data fork, the in-core and on-disk fork sizes can be
536 * different due to delayed allocation extents. We only copy on-disk extents
537 * here, so callers must always use the physical fork size to determine the
538 * size of the buffer passed to this routine. We will return the size actually
539 * used.
540 */
541int
542xfs_iextents_copy(
543 struct xfs_inode *ip,
544 struct xfs_bmbt_rec *dp,
545 int whichfork)
546{
547 int state = xfs_bmap_fork_to_state(whichfork);
548 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
549 struct xfs_iext_cursor icur;
550 struct xfs_bmbt_irec rec;
551 int64_t copied = 0;
552
553 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
554 ASSERT(ifp->if_bytes > 0);
555
556 for_each_xfs_iext(ifp, &icur, &rec) {
557 if (isnullstartblock(rec.br_startblock))
558 continue;
559 ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
560 xfs_bmbt_disk_set_all(dp, &rec);
561 trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
562 copied += sizeof(struct xfs_bmbt_rec);
563 dp++;
564 }
565
566 ASSERT(copied > 0);
567 ASSERT(copied <= ifp->if_bytes);
568 return copied;
569}
570
571/*
572 * Each of the following cases stores data into the same region
573 * of the on-disk inode, so only one of them can be valid at
574 * any given time. While it is possible to have conflicting formats
575 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
576 * in EXTENTS format, this can only happen when the fork has
577 * changed formats after being modified but before being flushed.
578 * In these cases, the format always takes precedence, because the
579 * format indicates the current state of the fork.
580 */
581void
582xfs_iflush_fork(
583 xfs_inode_t *ip,
584 xfs_dinode_t *dip,
585 struct xfs_inode_log_item *iip,
586 int whichfork)
587{
588 char *cp;
589 struct xfs_ifork *ifp;
590 xfs_mount_t *mp;
591 static const short brootflag[2] =
592 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
593 static const short dataflag[2] =
594 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
595 static const short extflag[2] =
596 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
597
598 if (!iip)
599 return;
600 ifp = XFS_IFORK_PTR(ip, whichfork);
601 /*
602 * This can happen if we gave up in iformat in an error path,
603 * for the attribute fork.
604 */
605 if (!ifp) {
606 ASSERT(whichfork == XFS_ATTR_FORK);
607 return;
608 }
609 cp = XFS_DFORK_PTR(dip, whichfork);
610 mp = ip->i_mount;
611 switch (ifp->if_format) {
612 case XFS_DINODE_FMT_LOCAL:
613 if ((iip->ili_fields & dataflag[whichfork]) &&
614 (ifp->if_bytes > 0)) {
615 ASSERT(ifp->if_u1.if_data != NULL);
616 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
617 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
618 }
619 break;
620
621 case XFS_DINODE_FMT_EXTENTS:
622 if ((iip->ili_fields & extflag[whichfork]) &&
623 (ifp->if_bytes > 0)) {
624 ASSERT(ifp->if_nextents > 0);
625 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
626 whichfork);
627 }
628 break;
629
630 case XFS_DINODE_FMT_BTREE:
631 if ((iip->ili_fields & brootflag[whichfork]) &&
632 (ifp->if_broot_bytes > 0)) {
633 ASSERT(ifp->if_broot != NULL);
634 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
635 XFS_IFORK_SIZE(ip, whichfork));
636 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
637 (xfs_bmdr_block_t *)cp,
638 XFS_DFORK_SIZE(dip, mp, whichfork));
639 }
640 break;
641
642 case XFS_DINODE_FMT_DEV:
643 if (iip->ili_fields & XFS_ILOG_DEV) {
644 ASSERT(whichfork == XFS_DATA_FORK);
645 xfs_dinode_put_rdev(dip,
646 linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
647 }
648 break;
649
650 default:
651 ASSERT(0);
652 break;
653 }
654}
655
656/* Convert bmap state flags to an inode fork. */
657struct xfs_ifork *
658xfs_iext_state_to_fork(
659 struct xfs_inode *ip,
660 int state)
661{
662 if (state & BMAP_COWFORK)
663 return ip->i_cowfp;
664 else if (state & BMAP_ATTRFORK)
665 return ip->i_afp;
666 return &ip->i_df;
667}
668
669/*
670 * Initialize an inode's copy-on-write fork.
671 */
672void
673xfs_ifork_init_cow(
674 struct xfs_inode *ip)
675{
676 if (ip->i_cowfp)
677 return;
678
679 ip->i_cowfp = kmem_cache_zalloc(xfs_ifork_zone,
680 GFP_NOFS | __GFP_NOFAIL);
681 ip->i_cowfp->if_format = XFS_DINODE_FMT_EXTENTS;
682}
683
684/* Verify the inline contents of the data fork of an inode. */
685int
686xfs_ifork_verify_local_data(
687 struct xfs_inode *ip)
688{
689 xfs_failaddr_t fa = NULL;
690
691 switch (VFS_I(ip)->i_mode & S_IFMT) {
692 case S_IFDIR:
693 fa = xfs_dir2_sf_verify(ip);
694 break;
695 case S_IFLNK:
696 fa = xfs_symlink_shortform_verify(ip);
697 break;
698 default:
699 break;
700 }
701
702 if (fa) {
703 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
704 ip->i_df.if_u1.if_data, ip->i_df.if_bytes, fa);
705 return -EFSCORRUPTED;
706 }
707
708 return 0;
709}
710
711/* Verify the inline contents of the attr fork of an inode. */
712int
713xfs_ifork_verify_local_attr(
714 struct xfs_inode *ip)
715{
716 struct xfs_ifork *ifp = ip->i_afp;
717 xfs_failaddr_t fa;
718
719 if (!ifp)
720 fa = __this_address;
721 else
722 fa = xfs_attr_shortform_verify(ip);
723
724 if (fa) {
725 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
726 ifp ? ifp->if_u1.if_data : NULL,
727 ifp ? ifp->if_bytes : 0, fa);
728 return -EFSCORRUPTED;
729 }
730
731 return 0;
732}
733
734int
735xfs_iext_count_may_overflow(
736 struct xfs_inode *ip,
737 int whichfork,
738 int nr_to_add)
739{
740 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
741 uint64_t max_exts;
742 uint64_t nr_exts;
743
744 if (whichfork == XFS_COW_FORK)
745 return 0;
746
747 max_exts = (whichfork == XFS_ATTR_FORK) ? MAXAEXTNUM : MAXEXTNUM;
748
749 if (XFS_TEST_ERROR(false, ip->i_mount, XFS_ERRTAG_REDUCE_MAX_IEXTENTS))
750 max_exts = 10;
751
752 nr_exts = ifp->if_nextents + nr_to_add;
753 if (nr_exts < ifp->if_nextents || nr_exts > max_exts)
754 return -EFBIG;
755
756 return 0;
757}