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