Loading...
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include <linux/log2.h>
19
20#include "xfs.h"
21#include "xfs_fs.h"
22#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_mount.h"
26#include "xfs_inode.h"
27#include "xfs_trans.h"
28#include "xfs_inode_item.h"
29#include "xfs_btree.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_bmap.h"
32#include "xfs_error.h"
33#include "xfs_trace.h"
34#include "xfs_attr_sf.h"
35#include "xfs_da_format.h"
36#include "xfs_da_btree.h"
37#include "xfs_dir2_priv.h"
38#include "xfs_attr_leaf.h"
39#include "xfs_shared.h"
40
41kmem_zone_t *xfs_ifork_zone;
42
43STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
44STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
45STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
46
47/*
48 * Copy inode type and data and attr format specific information from the
49 * on-disk inode to the in-core inode and fork structures. For fifos, devices,
50 * and sockets this means set i_rdev to the proper value. For files,
51 * directories, and symlinks this means to bring in the in-line data or extent
52 * pointers as well as the attribute fork. For a fork in B-tree format, only
53 * the root is immediately brought in-core. The rest will be read in later when
54 * first referenced (see xfs_iread_extents()).
55 */
56int
57xfs_iformat_fork(
58 struct xfs_inode *ip,
59 struct xfs_dinode *dip)
60{
61 struct inode *inode = VFS_I(ip);
62 struct xfs_attr_shortform *atp;
63 int size;
64 int error = 0;
65 xfs_fsize_t di_size;
66
67 switch (inode->i_mode & S_IFMT) {
68 case S_IFIFO:
69 case S_IFCHR:
70 case S_IFBLK:
71 case S_IFSOCK:
72 ip->i_d.di_size = 0;
73 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
74 break;
75
76 case S_IFREG:
77 case S_IFLNK:
78 case S_IFDIR:
79 switch (dip->di_format) {
80 case XFS_DINODE_FMT_LOCAL:
81 di_size = be64_to_cpu(dip->di_size);
82 size = (int)di_size;
83 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
84 break;
85 case XFS_DINODE_FMT_EXTENTS:
86 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
87 break;
88 case XFS_DINODE_FMT_BTREE:
89 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
90 break;
91 default:
92 return -EFSCORRUPTED;
93 }
94 break;
95
96 default:
97 return -EFSCORRUPTED;
98 }
99 if (error)
100 return error;
101
102 if (xfs_is_reflink_inode(ip)) {
103 ASSERT(ip->i_cowfp == NULL);
104 xfs_ifork_init_cow(ip);
105 }
106
107 if (!XFS_DFORK_Q(dip))
108 return 0;
109
110 ASSERT(ip->i_afp == NULL);
111 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
112
113 switch (dip->di_aformat) {
114 case XFS_DINODE_FMT_LOCAL:
115 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
116 size = be16_to_cpu(atp->hdr.totsize);
117
118 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
119 break;
120 case XFS_DINODE_FMT_EXTENTS:
121 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
122 break;
123 case XFS_DINODE_FMT_BTREE:
124 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
125 break;
126 default:
127 error = -EFSCORRUPTED;
128 break;
129 }
130 if (error) {
131 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
132 ip->i_afp = NULL;
133 if (ip->i_cowfp)
134 kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
135 ip->i_cowfp = NULL;
136 xfs_idestroy_fork(ip, XFS_DATA_FORK);
137 }
138 return error;
139}
140
141void
142xfs_init_local_fork(
143 struct xfs_inode *ip,
144 int whichfork,
145 const void *data,
146 int size)
147{
148 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
149 int mem_size = size, real_size = 0;
150 bool zero_terminate;
151
152 /*
153 * If we are using the local fork to store a symlink body we need to
154 * zero-terminate it so that we can pass it back to the VFS directly.
155 * Overallocate the in-memory fork by one for that and add a zero
156 * to terminate it below.
157 */
158 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
159 if (zero_terminate)
160 mem_size++;
161
162 if (size) {
163 real_size = roundup(mem_size, 4);
164 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
165 memcpy(ifp->if_u1.if_data, data, size);
166 if (zero_terminate)
167 ifp->if_u1.if_data[size] = '\0';
168 } else {
169 ifp->if_u1.if_data = NULL;
170 }
171
172 ifp->if_bytes = size;
173 ifp->if_real_bytes = real_size;
174 ifp->if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
175 ifp->if_flags |= XFS_IFINLINE;
176}
177
178/*
179 * The file is in-lined in the on-disk inode.
180 */
181STATIC int
182xfs_iformat_local(
183 xfs_inode_t *ip,
184 xfs_dinode_t *dip,
185 int whichfork,
186 int size)
187{
188 /*
189 * If the size is unreasonable, then something
190 * is wrong and we just bail out rather than crash in
191 * kmem_alloc() or memcpy() below.
192 */
193 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
194 xfs_warn(ip->i_mount,
195 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
196 (unsigned long long) ip->i_ino, size,
197 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
198 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
199 "xfs_iformat_local", dip, sizeof(*dip),
200 __this_address);
201 return -EFSCORRUPTED;
202 }
203
204 xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
205 return 0;
206}
207
208/*
209 * The file consists of a set of extents all of which fit into the on-disk
210 * inode.
211 */
212STATIC int
213xfs_iformat_extents(
214 struct xfs_inode *ip,
215 struct xfs_dinode *dip,
216 int whichfork)
217{
218 struct xfs_mount *mp = ip->i_mount;
219 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
220 int state = xfs_bmap_fork_to_state(whichfork);
221 int nex = XFS_DFORK_NEXTENTS(dip, whichfork);
222 int size = nex * sizeof(xfs_bmbt_rec_t);
223 struct xfs_iext_cursor icur;
224 struct xfs_bmbt_rec *dp;
225 struct xfs_bmbt_irec new;
226 int i;
227
228 /*
229 * If the number of extents is unreasonable, then something is wrong and
230 * we just bail out rather than crash in kmem_alloc() or memcpy() below.
231 */
232 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, mp, whichfork))) {
233 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
234 (unsigned long long) ip->i_ino, nex);
235 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
236 "xfs_iformat_extents(1)", dip, sizeof(*dip),
237 __this_address);
238 return -EFSCORRUPTED;
239 }
240
241 ifp->if_real_bytes = 0;
242 ifp->if_bytes = 0;
243 ifp->if_u1.if_root = NULL;
244 ifp->if_height = 0;
245 if (size) {
246 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
247
248 xfs_iext_first(ifp, &icur);
249 for (i = 0; i < nex; i++, dp++) {
250 xfs_failaddr_t fa;
251
252 xfs_bmbt_disk_get_all(dp, &new);
253 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
254 if (fa) {
255 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
256 "xfs_iformat_extents(2)",
257 dp, sizeof(*dp), fa);
258 return -EFSCORRUPTED;
259 }
260
261 xfs_iext_insert(ip, &icur, &new, state);
262 trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
263 xfs_iext_next(ifp, &icur);
264 }
265 }
266 ifp->if_flags |= XFS_IFEXTENTS;
267 return 0;
268}
269
270/*
271 * The file has too many extents to fit into
272 * the inode, so they are in B-tree format.
273 * Allocate a buffer for the root of the B-tree
274 * and copy the root into it. The i_extents
275 * field will remain NULL until all of the
276 * extents are read in (when they are needed).
277 */
278STATIC int
279xfs_iformat_btree(
280 xfs_inode_t *ip,
281 xfs_dinode_t *dip,
282 int whichfork)
283{
284 struct xfs_mount *mp = ip->i_mount;
285 xfs_bmdr_block_t *dfp;
286 xfs_ifork_t *ifp;
287 /* REFERENCED */
288 int nrecs;
289 int size;
290 int level;
291
292 ifp = XFS_IFORK_PTR(ip, whichfork);
293 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
294 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
295 nrecs = be16_to_cpu(dfp->bb_numrecs);
296 level = be16_to_cpu(dfp->bb_level);
297
298 /*
299 * blow out if -- fork has less extents than can fit in
300 * fork (fork shouldn't be a btree format), root btree
301 * block has more records than can fit into the fork,
302 * or the number of extents is greater than the number of
303 * blocks.
304 */
305 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
306 XFS_IFORK_MAXEXT(ip, whichfork) ||
307 nrecs == 0 ||
308 XFS_BMDR_SPACE_CALC(nrecs) >
309 XFS_DFORK_SIZE(dip, mp, whichfork) ||
310 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks) ||
311 level == 0 || level > XFS_BTREE_MAXLEVELS) {
312 xfs_warn(mp, "corrupt inode %Lu (btree).",
313 (unsigned long long) ip->i_ino);
314 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
315 "xfs_iformat_btree", dfp, size,
316 __this_address);
317 return -EFSCORRUPTED;
318 }
319
320 ifp->if_broot_bytes = size;
321 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
322 ASSERT(ifp->if_broot != NULL);
323 /*
324 * Copy and convert from the on-disk structure
325 * to the in-memory structure.
326 */
327 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
328 ifp->if_broot, size);
329 ifp->if_flags &= ~XFS_IFEXTENTS;
330 ifp->if_flags |= XFS_IFBROOT;
331
332 ifp->if_real_bytes = 0;
333 ifp->if_bytes = 0;
334 ifp->if_u1.if_root = NULL;
335 ifp->if_height = 0;
336 return 0;
337}
338
339/*
340 * Reallocate the space for if_broot based on the number of records
341 * being added or deleted as indicated in rec_diff. Move the records
342 * and pointers in if_broot to fit the new size. When shrinking this
343 * will eliminate holes between the records and pointers created by
344 * the caller. When growing this will create holes to be filled in
345 * by the caller.
346 *
347 * The caller must not request to add more records than would fit in
348 * the on-disk inode root. If the if_broot is currently NULL, then
349 * if we are adding records, one will be allocated. The caller must also
350 * not request that the number of records go below zero, although
351 * it can go to zero.
352 *
353 * ip -- the inode whose if_broot area is changing
354 * ext_diff -- the change in the number of records, positive or negative,
355 * requested for the if_broot array.
356 */
357void
358xfs_iroot_realloc(
359 xfs_inode_t *ip,
360 int rec_diff,
361 int whichfork)
362{
363 struct xfs_mount *mp = ip->i_mount;
364 int cur_max;
365 xfs_ifork_t *ifp;
366 struct xfs_btree_block *new_broot;
367 int new_max;
368 size_t new_size;
369 char *np;
370 char *op;
371
372 /*
373 * Handle the degenerate case quietly.
374 */
375 if (rec_diff == 0) {
376 return;
377 }
378
379 ifp = XFS_IFORK_PTR(ip, whichfork);
380 if (rec_diff > 0) {
381 /*
382 * If there wasn't any memory allocated before, just
383 * allocate it now and get out.
384 */
385 if (ifp->if_broot_bytes == 0) {
386 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
387 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
388 ifp->if_broot_bytes = (int)new_size;
389 return;
390 }
391
392 /*
393 * If there is already an existing if_broot, then we need
394 * to realloc() it and shift the pointers to their new
395 * location. The records don't change location because
396 * they are kept butted up against the btree block header.
397 */
398 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
399 new_max = cur_max + rec_diff;
400 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
401 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
402 KM_SLEEP | KM_NOFS);
403 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
404 ifp->if_broot_bytes);
405 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
406 (int)new_size);
407 ifp->if_broot_bytes = (int)new_size;
408 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
409 XFS_IFORK_SIZE(ip, whichfork));
410 memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
411 return;
412 }
413
414 /*
415 * rec_diff is less than 0. In this case, we are shrinking the
416 * if_broot buffer. It must already exist. If we go to zero
417 * records, just get rid of the root and clear the status bit.
418 */
419 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
420 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
421 new_max = cur_max + rec_diff;
422 ASSERT(new_max >= 0);
423 if (new_max > 0)
424 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
425 else
426 new_size = 0;
427 if (new_size > 0) {
428 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
429 /*
430 * First copy over the btree block header.
431 */
432 memcpy(new_broot, ifp->if_broot,
433 XFS_BMBT_BLOCK_LEN(ip->i_mount));
434 } else {
435 new_broot = NULL;
436 ifp->if_flags &= ~XFS_IFBROOT;
437 }
438
439 /*
440 * Only copy the records and pointers if there are any.
441 */
442 if (new_max > 0) {
443 /*
444 * First copy the records.
445 */
446 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
447 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
448 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
449
450 /*
451 * Then copy the pointers.
452 */
453 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
454 ifp->if_broot_bytes);
455 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
456 (int)new_size);
457 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
458 }
459 kmem_free(ifp->if_broot);
460 ifp->if_broot = new_broot;
461 ifp->if_broot_bytes = (int)new_size;
462 if (ifp->if_broot)
463 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
464 XFS_IFORK_SIZE(ip, whichfork));
465 return;
466}
467
468
469/*
470 * This is called when the amount of space needed for if_data
471 * is increased or decreased. The change in size is indicated by
472 * the number of bytes that need to be added or deleted in the
473 * byte_diff parameter.
474 *
475 * If the amount of space needed has decreased below the size of the
476 * inline buffer, then switch to using the inline buffer. Otherwise,
477 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
478 * to what is needed.
479 *
480 * ip -- the inode whose if_data area is changing
481 * byte_diff -- the change in the number of bytes, positive or negative,
482 * requested for the if_data array.
483 */
484void
485xfs_idata_realloc(
486 xfs_inode_t *ip,
487 int byte_diff,
488 int whichfork)
489{
490 xfs_ifork_t *ifp;
491 int new_size;
492 int real_size;
493
494 if (byte_diff == 0) {
495 return;
496 }
497
498 ifp = XFS_IFORK_PTR(ip, whichfork);
499 new_size = (int)ifp->if_bytes + byte_diff;
500 ASSERT(new_size >= 0);
501
502 if (new_size == 0) {
503 kmem_free(ifp->if_u1.if_data);
504 ifp->if_u1.if_data = NULL;
505 real_size = 0;
506 } else {
507 /*
508 * Stuck with malloc/realloc.
509 * For inline data, the underlying buffer must be
510 * a multiple of 4 bytes in size so that it can be
511 * logged and stay on word boundaries. We enforce
512 * that here.
513 */
514 real_size = roundup(new_size, 4);
515 if (ifp->if_u1.if_data == NULL) {
516 ASSERT(ifp->if_real_bytes == 0);
517 ifp->if_u1.if_data = kmem_alloc(real_size,
518 KM_SLEEP | KM_NOFS);
519 } else {
520 /*
521 * Only do the realloc if the underlying size
522 * is really changing.
523 */
524 if (ifp->if_real_bytes != real_size) {
525 ifp->if_u1.if_data =
526 kmem_realloc(ifp->if_u1.if_data,
527 real_size,
528 KM_SLEEP | KM_NOFS);
529 }
530 }
531 }
532 ifp->if_real_bytes = real_size;
533 ifp->if_bytes = new_size;
534 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
535}
536
537void
538xfs_idestroy_fork(
539 xfs_inode_t *ip,
540 int whichfork)
541{
542 xfs_ifork_t *ifp;
543
544 ifp = XFS_IFORK_PTR(ip, whichfork);
545 if (ifp->if_broot != NULL) {
546 kmem_free(ifp->if_broot);
547 ifp->if_broot = NULL;
548 }
549
550 /*
551 * If the format is local, then we can't have an extents
552 * array so just look for an inline data array. If we're
553 * not local then we may or may not have an extents list,
554 * so check and free it up if we do.
555 */
556 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
557 if (ifp->if_u1.if_data != NULL) {
558 ASSERT(ifp->if_real_bytes != 0);
559 kmem_free(ifp->if_u1.if_data);
560 ifp->if_u1.if_data = NULL;
561 ifp->if_real_bytes = 0;
562 }
563 } else if ((ifp->if_flags & XFS_IFEXTENTS) && ifp->if_height) {
564 xfs_iext_destroy(ifp);
565 }
566
567 ASSERT(ifp->if_real_bytes == 0);
568
569 if (whichfork == XFS_ATTR_FORK) {
570 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
571 ip->i_afp = NULL;
572 } else if (whichfork == XFS_COW_FORK) {
573 kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
574 ip->i_cowfp = NULL;
575 }
576}
577
578/*
579 * Convert in-core extents to on-disk form
580 *
581 * In the case of the data fork, the in-core and on-disk fork sizes can be
582 * different due to delayed allocation extents. We only copy on-disk extents
583 * here, so callers must always use the physical fork size to determine the
584 * size of the buffer passed to this routine. We will return the size actually
585 * used.
586 */
587int
588xfs_iextents_copy(
589 struct xfs_inode *ip,
590 struct xfs_bmbt_rec *dp,
591 int whichfork)
592{
593 int state = xfs_bmap_fork_to_state(whichfork);
594 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
595 struct xfs_iext_cursor icur;
596 struct xfs_bmbt_irec rec;
597 int copied = 0;
598
599 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
600 ASSERT(ifp->if_bytes > 0);
601
602 for_each_xfs_iext(ifp, &icur, &rec) {
603 if (isnullstartblock(rec.br_startblock))
604 continue;
605 ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
606 xfs_bmbt_disk_set_all(dp, &rec);
607 trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
608 copied += sizeof(struct xfs_bmbt_rec);
609 dp++;
610 }
611
612 ASSERT(copied > 0);
613 ASSERT(copied <= ifp->if_bytes);
614 return copied;
615}
616
617/*
618 * Each of the following cases stores data into the same region
619 * of the on-disk inode, so only one of them can be valid at
620 * any given time. While it is possible to have conflicting formats
621 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
622 * in EXTENTS format, this can only happen when the fork has
623 * changed formats after being modified but before being flushed.
624 * In these cases, the format always takes precedence, because the
625 * format indicates the current state of the fork.
626 */
627void
628xfs_iflush_fork(
629 xfs_inode_t *ip,
630 xfs_dinode_t *dip,
631 xfs_inode_log_item_t *iip,
632 int whichfork)
633{
634 char *cp;
635 xfs_ifork_t *ifp;
636 xfs_mount_t *mp;
637 static const short brootflag[2] =
638 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
639 static const short dataflag[2] =
640 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
641 static const short extflag[2] =
642 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
643
644 if (!iip)
645 return;
646 ifp = XFS_IFORK_PTR(ip, whichfork);
647 /*
648 * This can happen if we gave up in iformat in an error path,
649 * for the attribute fork.
650 */
651 if (!ifp) {
652 ASSERT(whichfork == XFS_ATTR_FORK);
653 return;
654 }
655 cp = XFS_DFORK_PTR(dip, whichfork);
656 mp = ip->i_mount;
657 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
658 case XFS_DINODE_FMT_LOCAL:
659 if ((iip->ili_fields & dataflag[whichfork]) &&
660 (ifp->if_bytes > 0)) {
661 ASSERT(ifp->if_u1.if_data != NULL);
662 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
663 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
664 }
665 break;
666
667 case XFS_DINODE_FMT_EXTENTS:
668 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
669 !(iip->ili_fields & extflag[whichfork]));
670 if ((iip->ili_fields & extflag[whichfork]) &&
671 (ifp->if_bytes > 0)) {
672 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
673 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
674 whichfork);
675 }
676 break;
677
678 case XFS_DINODE_FMT_BTREE:
679 if ((iip->ili_fields & brootflag[whichfork]) &&
680 (ifp->if_broot_bytes > 0)) {
681 ASSERT(ifp->if_broot != NULL);
682 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
683 XFS_IFORK_SIZE(ip, whichfork));
684 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
685 (xfs_bmdr_block_t *)cp,
686 XFS_DFORK_SIZE(dip, mp, whichfork));
687 }
688 break;
689
690 case XFS_DINODE_FMT_DEV:
691 if (iip->ili_fields & XFS_ILOG_DEV) {
692 ASSERT(whichfork == XFS_DATA_FORK);
693 xfs_dinode_put_rdev(dip,
694 linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
695 }
696 break;
697
698 default:
699 ASSERT(0);
700 break;
701 }
702}
703
704/* Convert bmap state flags to an inode fork. */
705struct xfs_ifork *
706xfs_iext_state_to_fork(
707 struct xfs_inode *ip,
708 int state)
709{
710 if (state & BMAP_COWFORK)
711 return ip->i_cowfp;
712 else if (state & BMAP_ATTRFORK)
713 return ip->i_afp;
714 return &ip->i_df;
715}
716
717/*
718 * Initialize an inode's copy-on-write fork.
719 */
720void
721xfs_ifork_init_cow(
722 struct xfs_inode *ip)
723{
724 if (ip->i_cowfp)
725 return;
726
727 ip->i_cowfp = kmem_zone_zalloc(xfs_ifork_zone,
728 KM_SLEEP | KM_NOFS);
729 ip->i_cowfp->if_flags = XFS_IFEXTENTS;
730 ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
731 ip->i_cnextents = 0;
732}
733
734/* Default fork content verifiers. */
735struct xfs_ifork_ops xfs_default_ifork_ops = {
736 .verify_attr = xfs_attr_shortform_verify,
737 .verify_dir = xfs_dir2_sf_verify,
738 .verify_symlink = xfs_symlink_shortform_verify,
739};
740
741/* Verify the inline contents of the data fork of an inode. */
742xfs_failaddr_t
743xfs_ifork_verify_data(
744 struct xfs_inode *ip,
745 struct xfs_ifork_ops *ops)
746{
747 /* Non-local data fork, we're done. */
748 if (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
749 return NULL;
750
751 /* Check the inline data fork if there is one. */
752 switch (VFS_I(ip)->i_mode & S_IFMT) {
753 case S_IFDIR:
754 return ops->verify_dir(ip);
755 case S_IFLNK:
756 return ops->verify_symlink(ip);
757 default:
758 return NULL;
759 }
760}
761
762/* Verify the inline contents of the attr fork of an inode. */
763xfs_failaddr_t
764xfs_ifork_verify_attr(
765 struct xfs_inode *ip,
766 struct xfs_ifork_ops *ops)
767{
768 /* There has to be an attr fork allocated if aformat is local. */
769 if (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
770 return NULL;
771 if (!XFS_IFORK_PTR(ip, XFS_ATTR_FORK))
772 return __this_address;
773 return ops->verify_attr(ip);
774}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_trans.h"
16#include "xfs_inode_item.h"
17#include "xfs_btree.h"
18#include "xfs_bmap_btree.h"
19#include "xfs_bmap.h"
20#include "xfs_error.h"
21#include "xfs_trace.h"
22#include "xfs_da_format.h"
23#include "xfs_da_btree.h"
24#include "xfs_dir2_priv.h"
25#include "xfs_attr_leaf.h"
26#include "xfs_types.h"
27#include "xfs_errortag.h"
28#include "xfs_health.h"
29#include "xfs_symlink_remote.h"
30
31struct kmem_cache *xfs_ifork_cache;
32
33void
34xfs_init_local_fork(
35 struct xfs_inode *ip,
36 int whichfork,
37 const void *data,
38 int64_t size)
39{
40 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
41 int mem_size = size;
42 bool zero_terminate;
43
44 /*
45 * If we are using the local fork to store a symlink body we need to
46 * zero-terminate it so that we can pass it back to the VFS directly.
47 * Overallocate the in-memory fork by one for that and add a zero
48 * to terminate it below.
49 */
50 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
51 if (zero_terminate)
52 mem_size++;
53
54 if (size) {
55 char *new_data = kmalloc(mem_size,
56 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
57
58 memcpy(new_data, data, size);
59 if (zero_terminate)
60 new_data[size] = '\0';
61
62 ifp->if_data = new_data;
63 } else {
64 ifp->if_data = NULL;
65 }
66
67 ifp->if_bytes = size;
68}
69
70/*
71 * The file is in-lined in the on-disk inode.
72 */
73STATIC int
74xfs_iformat_local(
75 struct xfs_inode *ip,
76 struct xfs_dinode *dip,
77 int whichfork,
78 int size)
79{
80 /*
81 * If the size is unreasonable, then something
82 * is wrong and we just bail out rather than crash in
83 * kmalloc() or memcpy() below.
84 */
85 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
86 xfs_warn(ip->i_mount,
87 "corrupt inode %llu (bad size %d for local fork, size = %zd).",
88 (unsigned long long) ip->i_ino, size,
89 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
90 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
91 "xfs_iformat_local", dip, sizeof(*dip),
92 __this_address);
93 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
94 return -EFSCORRUPTED;
95 }
96
97 xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
98 return 0;
99}
100
101/*
102 * The file consists of a set of extents all of which fit into the on-disk
103 * inode.
104 */
105STATIC int
106xfs_iformat_extents(
107 struct xfs_inode *ip,
108 struct xfs_dinode *dip,
109 int whichfork)
110{
111 struct xfs_mount *mp = ip->i_mount;
112 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
113 int state = xfs_bmap_fork_to_state(whichfork);
114 xfs_extnum_t nex = xfs_dfork_nextents(dip, whichfork);
115 int size = nex * sizeof(xfs_bmbt_rec_t);
116 struct xfs_iext_cursor icur;
117 struct xfs_bmbt_rec *dp;
118 struct xfs_bmbt_irec new;
119 int i;
120
121 /*
122 * If the number of extents is unreasonable, then something is wrong and
123 * we just bail out rather than crash in kmalloc() or memcpy() below.
124 */
125 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, mp, whichfork))) {
126 xfs_warn(ip->i_mount, "corrupt inode %llu ((a)extents = %llu).",
127 ip->i_ino, nex);
128 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
129 "xfs_iformat_extents(1)", dip, sizeof(*dip),
130 __this_address);
131 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
132 return -EFSCORRUPTED;
133 }
134
135 ifp->if_bytes = 0;
136 ifp->if_data = NULL;
137 ifp->if_height = 0;
138 if (size) {
139 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
140
141 xfs_iext_first(ifp, &icur);
142 for (i = 0; i < nex; i++, dp++) {
143 xfs_failaddr_t fa;
144
145 xfs_bmbt_disk_get_all(dp, &new);
146 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
147 if (fa) {
148 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
149 "xfs_iformat_extents(2)",
150 dp, sizeof(*dp), fa);
151 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
152 return xfs_bmap_complain_bad_rec(ip, whichfork,
153 fa, &new);
154 }
155
156 xfs_iext_insert(ip, &icur, &new, state);
157 trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
158 xfs_iext_next(ifp, &icur);
159 }
160 }
161 return 0;
162}
163
164/*
165 * The file has too many extents to fit into
166 * the inode, so they are in B-tree format.
167 * Allocate a buffer for the root of the B-tree
168 * and copy the root into it. The i_extents
169 * field will remain NULL until all of the
170 * extents are read in (when they are needed).
171 */
172STATIC int
173xfs_iformat_btree(
174 struct xfs_inode *ip,
175 struct xfs_dinode *dip,
176 int whichfork)
177{
178 struct xfs_mount *mp = ip->i_mount;
179 xfs_bmdr_block_t *dfp;
180 struct xfs_ifork *ifp;
181 /* REFERENCED */
182 int nrecs;
183 int size;
184 int level;
185
186 ifp = xfs_ifork_ptr(ip, whichfork);
187 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
188 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
189 nrecs = be16_to_cpu(dfp->bb_numrecs);
190 level = be16_to_cpu(dfp->bb_level);
191
192 /*
193 * blow out if -- fork has less extents than can fit in
194 * fork (fork shouldn't be a btree format), root btree
195 * block has more records than can fit into the fork,
196 * or the number of extents is greater than the number of
197 * blocks.
198 */
199 if (unlikely(ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork) ||
200 nrecs == 0 ||
201 XFS_BMDR_SPACE_CALC(nrecs) >
202 XFS_DFORK_SIZE(dip, mp, whichfork) ||
203 ifp->if_nextents > ip->i_nblocks) ||
204 level == 0 || level > XFS_BM_MAXLEVELS(mp, whichfork)) {
205 xfs_warn(mp, "corrupt inode %llu (btree).",
206 (unsigned long long) ip->i_ino);
207 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
208 "xfs_iformat_btree", dfp, size,
209 __this_address);
210 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
211 return -EFSCORRUPTED;
212 }
213
214 ifp->if_broot_bytes = size;
215 ifp->if_broot = kmalloc(size,
216 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
217 ASSERT(ifp->if_broot != NULL);
218 /*
219 * Copy and convert from the on-disk structure
220 * to the in-memory structure.
221 */
222 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
223 ifp->if_broot, size);
224
225 ifp->if_bytes = 0;
226 ifp->if_data = NULL;
227 ifp->if_height = 0;
228 return 0;
229}
230
231int
232xfs_iformat_data_fork(
233 struct xfs_inode *ip,
234 struct xfs_dinode *dip)
235{
236 struct inode *inode = VFS_I(ip);
237 int error;
238
239 /*
240 * Initialize the extent count early, as the per-format routines may
241 * depend on it. Use release semantics to set needextents /after/ we
242 * set the format. This ensures that we can use acquire semantics on
243 * needextents in xfs_need_iread_extents() and be guaranteed to see a
244 * valid format value after that load.
245 */
246 ip->i_df.if_format = dip->di_format;
247 ip->i_df.if_nextents = xfs_dfork_data_extents(dip);
248 smp_store_release(&ip->i_df.if_needextents,
249 ip->i_df.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
250
251 switch (inode->i_mode & S_IFMT) {
252 case S_IFIFO:
253 case S_IFCHR:
254 case S_IFBLK:
255 case S_IFSOCK:
256 ip->i_disk_size = 0;
257 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
258 return 0;
259 case S_IFREG:
260 case S_IFLNK:
261 case S_IFDIR:
262 switch (ip->i_df.if_format) {
263 case XFS_DINODE_FMT_LOCAL:
264 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK,
265 be64_to_cpu(dip->di_size));
266 if (!error)
267 error = xfs_ifork_verify_local_data(ip);
268 return error;
269 case XFS_DINODE_FMT_EXTENTS:
270 return xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
271 case XFS_DINODE_FMT_BTREE:
272 return xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
273 default:
274 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
275 dip, sizeof(*dip), __this_address);
276 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
277 return -EFSCORRUPTED;
278 }
279 break;
280 default:
281 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
282 sizeof(*dip), __this_address);
283 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
284 return -EFSCORRUPTED;
285 }
286}
287
288static uint16_t
289xfs_dfork_attr_shortform_size(
290 struct xfs_dinode *dip)
291{
292 struct xfs_attr_sf_hdr *sf = XFS_DFORK_APTR(dip);
293
294 return be16_to_cpu(sf->totsize);
295}
296
297void
298xfs_ifork_init_attr(
299 struct xfs_inode *ip,
300 enum xfs_dinode_fmt format,
301 xfs_extnum_t nextents)
302{
303 /*
304 * Initialize the extent count early, as the per-format routines may
305 * depend on it. Use release semantics to set needextents /after/ we
306 * set the format. This ensures that we can use acquire semantics on
307 * needextents in xfs_need_iread_extents() and be guaranteed to see a
308 * valid format value after that load.
309 */
310 ip->i_af.if_format = format;
311 ip->i_af.if_nextents = nextents;
312 smp_store_release(&ip->i_af.if_needextents,
313 ip->i_af.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
314}
315
316void
317xfs_ifork_zap_attr(
318 struct xfs_inode *ip)
319{
320 xfs_idestroy_fork(&ip->i_af);
321 memset(&ip->i_af, 0, sizeof(struct xfs_ifork));
322 ip->i_af.if_format = XFS_DINODE_FMT_EXTENTS;
323}
324
325int
326xfs_iformat_attr_fork(
327 struct xfs_inode *ip,
328 struct xfs_dinode *dip)
329{
330 xfs_extnum_t naextents = xfs_dfork_attr_extents(dip);
331 int error = 0;
332
333 /*
334 * Initialize the extent count early, as the per-format routines may
335 * depend on it.
336 */
337 xfs_ifork_init_attr(ip, dip->di_aformat, naextents);
338
339 switch (ip->i_af.if_format) {
340 case XFS_DINODE_FMT_LOCAL:
341 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK,
342 xfs_dfork_attr_shortform_size(dip));
343 if (!error)
344 error = xfs_ifork_verify_local_attr(ip);
345 break;
346 case XFS_DINODE_FMT_EXTENTS:
347 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
348 break;
349 case XFS_DINODE_FMT_BTREE:
350 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
351 break;
352 default:
353 xfs_inode_verifier_error(ip, error, __func__, dip,
354 sizeof(*dip), __this_address);
355 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
356 error = -EFSCORRUPTED;
357 break;
358 }
359
360 if (error)
361 xfs_ifork_zap_attr(ip);
362 return error;
363}
364
365/*
366 * Reallocate the space for if_broot based on the number of records
367 * being added or deleted as indicated in rec_diff. Move the records
368 * and pointers in if_broot to fit the new size. When shrinking this
369 * will eliminate holes between the records and pointers created by
370 * the caller. When growing this will create holes to be filled in
371 * by the caller.
372 *
373 * The caller must not request to add more records than would fit in
374 * the on-disk inode root. If the if_broot is currently NULL, then
375 * if we are adding records, one will be allocated. The caller must also
376 * not request that the number of records go below zero, although
377 * it can go to zero.
378 *
379 * ip -- the inode whose if_broot area is changing
380 * ext_diff -- the change in the number of records, positive or negative,
381 * requested for the if_broot array.
382 */
383void
384xfs_iroot_realloc(
385 xfs_inode_t *ip,
386 int rec_diff,
387 int whichfork)
388{
389 struct xfs_mount *mp = ip->i_mount;
390 int cur_max;
391 struct xfs_ifork *ifp;
392 struct xfs_btree_block *new_broot;
393 int new_max;
394 size_t new_size;
395 char *np;
396 char *op;
397
398 /*
399 * Handle the degenerate case quietly.
400 */
401 if (rec_diff == 0) {
402 return;
403 }
404
405 ifp = xfs_ifork_ptr(ip, whichfork);
406 if (rec_diff > 0) {
407 /*
408 * If there wasn't any memory allocated before, just
409 * allocate it now and get out.
410 */
411 if (ifp->if_broot_bytes == 0) {
412 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
413 ifp->if_broot = kmalloc(new_size,
414 GFP_KERNEL | __GFP_NOFAIL);
415 ifp->if_broot_bytes = (int)new_size;
416 return;
417 }
418
419 /*
420 * If there is already an existing if_broot, then we need
421 * to realloc() it and shift the pointers to their new
422 * location. The records don't change location because
423 * they are kept butted up against the btree block header.
424 */
425 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
426 new_max = cur_max + rec_diff;
427 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
428 ifp->if_broot = krealloc(ifp->if_broot, new_size,
429 GFP_KERNEL | __GFP_NOFAIL);
430 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
431 ifp->if_broot_bytes);
432 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
433 (int)new_size);
434 ifp->if_broot_bytes = (int)new_size;
435 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
436 xfs_inode_fork_size(ip, whichfork));
437 memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
438 return;
439 }
440
441 /*
442 * rec_diff is less than 0. In this case, we are shrinking the
443 * if_broot buffer. It must already exist. If we go to zero
444 * records, just get rid of the root and clear the status bit.
445 */
446 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
447 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
448 new_max = cur_max + rec_diff;
449 ASSERT(new_max >= 0);
450 if (new_max > 0)
451 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
452 else
453 new_size = 0;
454 if (new_size > 0) {
455 new_broot = kmalloc(new_size, GFP_KERNEL | __GFP_NOFAIL);
456 /*
457 * First copy over the btree block header.
458 */
459 memcpy(new_broot, ifp->if_broot,
460 XFS_BMBT_BLOCK_LEN(ip->i_mount));
461 } else {
462 new_broot = NULL;
463 }
464
465 /*
466 * Only copy the records and pointers if there are any.
467 */
468 if (new_max > 0) {
469 /*
470 * First copy the records.
471 */
472 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
473 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
474 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
475
476 /*
477 * Then copy the pointers.
478 */
479 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
480 ifp->if_broot_bytes);
481 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
482 (int)new_size);
483 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
484 }
485 kfree(ifp->if_broot);
486 ifp->if_broot = new_broot;
487 ifp->if_broot_bytes = (int)new_size;
488 if (ifp->if_broot)
489 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
490 xfs_inode_fork_size(ip, whichfork));
491 return;
492}
493
494
495/*
496 * This is called when the amount of space needed for if_data
497 * is increased or decreased. The change in size is indicated by
498 * the number of bytes that need to be added or deleted in the
499 * byte_diff parameter.
500 *
501 * If the amount of space needed has decreased below the size of the
502 * inline buffer, then switch to using the inline buffer. Otherwise,
503 * use krealloc() or kmalloc() to adjust the size of the buffer
504 * to what is needed.
505 *
506 * ip -- the inode whose if_data area is changing
507 * byte_diff -- the change in the number of bytes, positive or negative,
508 * requested for the if_data array.
509 */
510void *
511xfs_idata_realloc(
512 struct xfs_inode *ip,
513 int64_t byte_diff,
514 int whichfork)
515{
516 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
517 int64_t new_size = ifp->if_bytes + byte_diff;
518
519 ASSERT(new_size >= 0);
520 ASSERT(new_size <= xfs_inode_fork_size(ip, whichfork));
521
522 if (byte_diff) {
523 ifp->if_data = krealloc(ifp->if_data, new_size,
524 GFP_KERNEL | __GFP_NOFAIL);
525 if (new_size == 0)
526 ifp->if_data = NULL;
527 ifp->if_bytes = new_size;
528 }
529
530 return ifp->if_data;
531}
532
533/* Free all memory and reset a fork back to its initial state. */
534void
535xfs_idestroy_fork(
536 struct xfs_ifork *ifp)
537{
538 if (ifp->if_broot != NULL) {
539 kfree(ifp->if_broot);
540 ifp->if_broot = NULL;
541 }
542
543 switch (ifp->if_format) {
544 case XFS_DINODE_FMT_LOCAL:
545 kfree(ifp->if_data);
546 ifp->if_data = NULL;
547 break;
548 case XFS_DINODE_FMT_EXTENTS:
549 case XFS_DINODE_FMT_BTREE:
550 if (ifp->if_height)
551 xfs_iext_destroy(ifp);
552 break;
553 }
554}
555
556/*
557 * Convert in-core extents to on-disk form
558 *
559 * In the case of the data fork, the in-core and on-disk fork sizes can be
560 * different due to delayed allocation extents. We only copy on-disk extents
561 * here, so callers must always use the physical fork size to determine the
562 * size of the buffer passed to this routine. We will return the size actually
563 * used.
564 */
565int
566xfs_iextents_copy(
567 struct xfs_inode *ip,
568 struct xfs_bmbt_rec *dp,
569 int whichfork)
570{
571 int state = xfs_bmap_fork_to_state(whichfork);
572 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
573 struct xfs_iext_cursor icur;
574 struct xfs_bmbt_irec rec;
575 int64_t copied = 0;
576
577 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
578 ASSERT(ifp->if_bytes > 0);
579
580 for_each_xfs_iext(ifp, &icur, &rec) {
581 if (isnullstartblock(rec.br_startblock))
582 continue;
583 ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
584 xfs_bmbt_disk_set_all(dp, &rec);
585 trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
586 copied += sizeof(struct xfs_bmbt_rec);
587 dp++;
588 }
589
590 ASSERT(copied > 0);
591 ASSERT(copied <= ifp->if_bytes);
592 return copied;
593}
594
595/*
596 * Each of the following cases stores data into the same region
597 * of the on-disk inode, so only one of them can be valid at
598 * any given time. While it is possible to have conflicting formats
599 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
600 * in EXTENTS format, this can only happen when the fork has
601 * changed formats after being modified but before being flushed.
602 * In these cases, the format always takes precedence, because the
603 * format indicates the current state of the fork.
604 */
605void
606xfs_iflush_fork(
607 struct xfs_inode *ip,
608 struct xfs_dinode *dip,
609 struct xfs_inode_log_item *iip,
610 int whichfork)
611{
612 char *cp;
613 struct xfs_ifork *ifp;
614 xfs_mount_t *mp;
615 static const short brootflag[2] =
616 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
617 static const short dataflag[2] =
618 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
619 static const short extflag[2] =
620 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
621
622 if (!iip)
623 return;
624 ifp = xfs_ifork_ptr(ip, whichfork);
625 /*
626 * This can happen if we gave up in iformat in an error path,
627 * for the attribute fork.
628 */
629 if (!ifp) {
630 ASSERT(whichfork == XFS_ATTR_FORK);
631 return;
632 }
633 cp = XFS_DFORK_PTR(dip, whichfork);
634 mp = ip->i_mount;
635 switch (ifp->if_format) {
636 case XFS_DINODE_FMT_LOCAL:
637 if ((iip->ili_fields & dataflag[whichfork]) &&
638 (ifp->if_bytes > 0)) {
639 ASSERT(ifp->if_data != NULL);
640 ASSERT(ifp->if_bytes <= xfs_inode_fork_size(ip, whichfork));
641 memcpy(cp, ifp->if_data, ifp->if_bytes);
642 }
643 break;
644
645 case XFS_DINODE_FMT_EXTENTS:
646 if ((iip->ili_fields & extflag[whichfork]) &&
647 (ifp->if_bytes > 0)) {
648 ASSERT(ifp->if_nextents > 0);
649 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
650 whichfork);
651 }
652 break;
653
654 case XFS_DINODE_FMT_BTREE:
655 if ((iip->ili_fields & brootflag[whichfork]) &&
656 (ifp->if_broot_bytes > 0)) {
657 ASSERT(ifp->if_broot != NULL);
658 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
659 xfs_inode_fork_size(ip, whichfork));
660 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
661 (xfs_bmdr_block_t *)cp,
662 XFS_DFORK_SIZE(dip, mp, whichfork));
663 }
664 break;
665
666 case XFS_DINODE_FMT_DEV:
667 if (iip->ili_fields & XFS_ILOG_DEV) {
668 ASSERT(whichfork == XFS_DATA_FORK);
669 xfs_dinode_put_rdev(dip,
670 linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
671 }
672 break;
673
674 default:
675 ASSERT(0);
676 break;
677 }
678}
679
680/* Convert bmap state flags to an inode fork. */
681struct xfs_ifork *
682xfs_iext_state_to_fork(
683 struct xfs_inode *ip,
684 int state)
685{
686 if (state & BMAP_COWFORK)
687 return ip->i_cowfp;
688 else if (state & BMAP_ATTRFORK)
689 return &ip->i_af;
690 return &ip->i_df;
691}
692
693/*
694 * Initialize an inode's copy-on-write fork.
695 */
696void
697xfs_ifork_init_cow(
698 struct xfs_inode *ip)
699{
700 if (ip->i_cowfp)
701 return;
702
703 ip->i_cowfp = kmem_cache_zalloc(xfs_ifork_cache,
704 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
705 ip->i_cowfp->if_format = XFS_DINODE_FMT_EXTENTS;
706}
707
708/* Verify the inline contents of the data fork of an inode. */
709int
710xfs_ifork_verify_local_data(
711 struct xfs_inode *ip)
712{
713 xfs_failaddr_t fa = NULL;
714
715 switch (VFS_I(ip)->i_mode & S_IFMT) {
716 case S_IFDIR: {
717 struct xfs_mount *mp = ip->i_mount;
718 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
719 struct xfs_dir2_sf_hdr *sfp = ifp->if_data;
720
721 fa = xfs_dir2_sf_verify(mp, sfp, ifp->if_bytes);
722 break;
723 }
724 case S_IFLNK: {
725 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
726
727 fa = xfs_symlink_shortform_verify(ifp->if_data, ifp->if_bytes);
728 break;
729 }
730 default:
731 break;
732 }
733
734 if (fa) {
735 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
736 ip->i_df.if_data, ip->i_df.if_bytes, fa);
737 return -EFSCORRUPTED;
738 }
739
740 return 0;
741}
742
743/* Verify the inline contents of the attr fork of an inode. */
744int
745xfs_ifork_verify_local_attr(
746 struct xfs_inode *ip)
747{
748 struct xfs_ifork *ifp = &ip->i_af;
749 xfs_failaddr_t fa;
750
751 if (!xfs_inode_has_attr_fork(ip)) {
752 fa = __this_address;
753 } else {
754 struct xfs_ifork *ifp = &ip->i_af;
755
756 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
757 fa = xfs_attr_shortform_verify(ifp->if_data, ifp->if_bytes);
758 }
759 if (fa) {
760 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
761 ifp->if_data, ifp->if_bytes, fa);
762 return -EFSCORRUPTED;
763 }
764
765 return 0;
766}
767
768int
769xfs_iext_count_may_overflow(
770 struct xfs_inode *ip,
771 int whichfork,
772 int nr_to_add)
773{
774 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
775 uint64_t max_exts;
776 uint64_t nr_exts;
777
778 if (whichfork == XFS_COW_FORK)
779 return 0;
780
781 max_exts = xfs_iext_max_nextents(xfs_inode_has_large_extent_counts(ip),
782 whichfork);
783
784 if (XFS_TEST_ERROR(false, ip->i_mount, XFS_ERRTAG_REDUCE_MAX_IEXTENTS))
785 max_exts = 10;
786
787 nr_exts = ifp->if_nextents + nr_to_add;
788 if (nr_exts < ifp->if_nextents || nr_exts > max_exts)
789 return -EFBIG;
790
791 return 0;
792}
793
794/*
795 * Upgrade this inode's extent counter fields to be able to handle a potential
796 * increase in the extent count by nr_to_add. Normally this is the same
797 * quantity that caused xfs_iext_count_may_overflow() to return -EFBIG.
798 */
799int
800xfs_iext_count_upgrade(
801 struct xfs_trans *tp,
802 struct xfs_inode *ip,
803 uint nr_to_add)
804{
805 ASSERT(nr_to_add <= XFS_MAX_EXTCNT_UPGRADE_NR);
806
807 if (!xfs_has_large_extent_counts(ip->i_mount) ||
808 xfs_inode_has_large_extent_counts(ip) ||
809 XFS_TEST_ERROR(false, ip->i_mount, XFS_ERRTAG_REDUCE_MAX_IEXTENTS))
810 return -EFBIG;
811
812 ip->i_diflags2 |= XFS_DIFLAG2_NREXT64;
813 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
814
815 return 0;
816}
817
818/* Decide if a file mapping is on the realtime device or not. */
819bool
820xfs_ifork_is_realtime(
821 struct xfs_inode *ip,
822 int whichfork)
823{
824 return XFS_IS_REALTIME_INODE(ip) && whichfork != XFS_ATTR_FORK;
825}