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#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}