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