1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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"
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"
27 kmem_zone_t *xfs_ifork_zone;
29 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
30 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
31 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
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()).
45 struct xfs_dinode *dip)
47 struct inode *inode = VFS_I(ip);
48 struct xfs_attr_shortform *atp;
53 switch (inode->i_mode & S_IFMT) {
59 inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
65 switch (dip->di_format) {
66 case XFS_DINODE_FMT_LOCAL:
67 di_size = be64_to_cpu(dip->di_size);
69 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
71 case XFS_DINODE_FMT_EXTENTS:
72 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
74 case XFS_DINODE_FMT_BTREE:
75 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
88 if (xfs_is_reflink_inode(ip)) {
89 ASSERT(ip->i_cowfp == NULL);
90 xfs_ifork_init_cow(ip);
93 if (!XFS_DFORK_Q(dip))
96 ASSERT(ip->i_afp == NULL);
97 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_NOFS);
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);
104 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
106 case XFS_DINODE_FMT_EXTENTS:
107 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
109 case XFS_DINODE_FMT_BTREE:
110 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
113 error = -EFSCORRUPTED;
117 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
120 kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
122 xfs_idestroy_fork(ip, XFS_DATA_FORK);
129 struct xfs_inode *ip,
134 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
135 int mem_size = size, real_size = 0;
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.
144 zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
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);
153 ifp->if_u1.if_data[size] = '\0';
155 ifp->if_u1.if_data = NULL;
158 ifp->if_bytes = size;
159 ifp->if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
160 ifp->if_flags |= XFS_IFINLINE;
164 * The file is in-lined in the on-disk inode.
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.
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),
186 return -EFSCORRUPTED;
189 xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
194 * The file consists of a set of extents all of which fit into the on-disk
199 struct xfs_inode *ip,
200 struct xfs_dinode *dip,
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;
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.
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),
223 return -EFSCORRUPTED;
227 ifp->if_u1.if_root = NULL;
230 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
232 xfs_iext_first(ifp, &icur);
233 for (i = 0; i < nex; i++, dp++) {
236 xfs_bmbt_disk_get_all(dp, &new);
237 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
239 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
240 "xfs_iformat_extents(2)",
241 dp, sizeof(*dp), fa);
242 return -EFSCORRUPTED;
245 xfs_iext_insert(ip, &icur, &new, state);
246 trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
247 xfs_iext_next(ifp, &icur);
250 ifp->if_flags |= XFS_IFEXTENTS;
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).
268 struct xfs_mount *mp = ip->i_mount;
269 xfs_bmdr_block_t *dfp;
270 struct xfs_ifork *ifp;
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);
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
289 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
290 XFS_IFORK_MAXEXT(ip, whichfork) ||
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,
301 return -EFSCORRUPTED;
304 ifp->if_broot_bytes = size;
305 ifp->if_broot = kmem_alloc(size, KM_NOFS);
306 ASSERT(ifp->if_broot != NULL);
308 * Copy and convert from the on-disk structure
309 * to the in-memory structure.
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;
317 ifp->if_u1.if_root = NULL;
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
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
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.
346 struct xfs_mount *mp = ip->i_mount;
348 struct xfs_ifork *ifp;
349 struct xfs_btree_block *new_broot;
356 * Handle the degenerate case quietly.
362 ifp = XFS_IFORK_PTR(ip, whichfork);
365 * If there wasn't any memory allocated before, just
366 * allocate it now and get out.
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;
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.
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,
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,
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));
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.
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);
407 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
411 new_broot = kmem_alloc(new_size, KM_NOFS);
413 * First copy over the btree block header.
415 memcpy(new_broot, ifp->if_broot,
416 XFS_BMBT_BLOCK_LEN(ip->i_mount));
419 ifp->if_flags &= ~XFS_IFBROOT;
423 * Only copy the records and pointers if there are any.
427 * First copy the records.
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));
434 * Then copy the pointers.
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,
440 memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
442 kmem_free(ifp->if_broot);
443 ifp->if_broot = new_broot;
444 ifp->if_broot_bytes = (int)new_size;
446 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
447 XFS_IFORK_SIZE(ip, whichfork));
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.
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
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.
469 struct xfs_inode *ip,
473 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
474 int new_size = (int)ifp->if_bytes + byte_diff;
476 ASSERT(new_size >= 0);
477 ASSERT(new_size <= XFS_IFORK_SIZE(ip, whichfork));
483 kmem_free(ifp->if_u1.if_data);
484 ifp->if_u1.if_data = NULL;
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.
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;
504 struct xfs_ifork *ifp;
506 ifp = XFS_IFORK_PTR(ip, whichfork);
507 if (ifp->if_broot != NULL) {
508 kmem_free(ifp->if_broot);
509 ifp->if_broot = NULL;
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.
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;
523 } else if ((ifp->if_flags & XFS_IFEXTENTS) && ifp->if_height) {
524 xfs_iext_destroy(ifp);
527 if (whichfork == XFS_ATTR_FORK) {
528 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
530 } else if (whichfork == XFS_COW_FORK) {
531 kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
537 * Convert in-core extents to on-disk form
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
547 struct xfs_inode *ip,
548 struct xfs_bmbt_rec *dp,
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;
557 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
558 ASSERT(ifp->if_bytes > 0);
560 for_each_xfs_iext(ifp, &icur, &rec) {
561 if (isnullstartblock(rec.br_startblock))
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);
571 ASSERT(copied <= ifp->if_bytes);
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.
589 xfs_inode_log_item_t *iip,
593 struct xfs_ifork *ifp;
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 };
604 ifp = XFS_IFORK_PTR(ip, whichfork);
606 * This can happen if we gave up in iformat in an error path,
607 * for the attribute fork.
610 ASSERT(whichfork == XFS_ATTR_FORK);
613 cp = XFS_DFORK_PTR(dip, whichfork);
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);
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,
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));
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));
662 /* Convert bmap state flags to an inode fork. */
664 xfs_iext_state_to_fork(
665 struct xfs_inode *ip,
668 if (state & BMAP_COWFORK)
670 else if (state & BMAP_ATTRFORK)
676 * Initialize an inode's copy-on-write fork.
680 struct xfs_inode *ip)
685 ip->i_cowfp = kmem_zone_zalloc(xfs_ifork_zone,
687 ip->i_cowfp->if_flags = XFS_IFEXTENTS;
688 ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
692 /* Default fork content verifiers. */
693 struct 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,
699 /* Verify the inline contents of the data fork of an inode. */
701 xfs_ifork_verify_data(
702 struct xfs_inode *ip,
703 struct xfs_ifork_ops *ops)
705 /* Non-local data fork, we're done. */
706 if (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
709 /* Check the inline data fork if there is one. */
710 switch (VFS_I(ip)->i_mode & S_IFMT) {
712 return ops->verify_dir(ip);
714 return ops->verify_symlink(ip);
720 /* Verify the inline contents of the attr fork of an inode. */
722 xfs_ifork_verify_attr(
723 struct xfs_inode *ip,
724 struct xfs_ifork_ops *ops)
726 /* There has to be an attr fork allocated if aformat is local. */
727 if (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
729 if (!XFS_IFORK_PTR(ip, XFS_ATTR_FORK))
730 return __this_address;
731 return ops->verify_attr(ip);