2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
53 #include <linux/log2.h>
55 kmem_zone_t *xfs_ifork_zone;
56 kmem_zone_t *xfs_inode_zone;
57 kmem_zone_t *xfs_chashlist_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
73 * Make sure that the extents in the given memory buffer
88 for (i = 0; i < nrecs; i++) {
89 ep = xfs_iext_get_ext(ifp, i);
90 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
91 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
93 xfs_bmbt_disk_get_all(&rec, &irec);
95 xfs_bmbt_get_all(&rec, &irec);
96 if (fmt == XFS_EXTFMT_NOSTATE)
97 ASSERT(irec.br_state == XFS_EXT_NORM);
101 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
118 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120 for (i = 0; i < j; i++) {
121 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
122 i * mp->m_sb.sb_inodesize);
123 if (!dip->di_next_unlinked) {
124 xfs_fs_cmn_err(CE_ALERT, mp,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 ASSERT(dip->di_next_unlinked);
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
162 * Call the space management code to find the location of the
166 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error, mp->m_fsname);
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
179 if ((imap.im_blkno + imap.im_len) >
180 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
183 "of the file system %s. Returning EINVAL.",
184 (unsigned long long)imap.im_blkno,
185 imap.im_len, mp->m_fsname);
186 return XFS_ERROR(EINVAL);
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
193 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
194 (int)imap.im_len, XFS_BUF_LOCK, &bp);
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error, mp->m_fsname);
202 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
204 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
206 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
207 XFS_RANDOM_ITOBP_INOTOBP))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
209 xfs_trans_brelse(tp, bp);
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
213 return XFS_ERROR(EFSCORRUPTED);
216 xfs_inobp_check(mp, bp);
219 * Set *dipp to point to the on-disk inode in the buffer.
221 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
223 *offset = imap.im_boffset;
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
264 if (ip->i_blkno == (xfs_daddr_t)0) {
266 * Call the space management code to find the location of the
270 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
271 XFS_IMAP_LOOKUP | imap_flags)))
275 * If the inode number maps to a block outside the bounds
276 * of the file system then return NULL rather than calling
277 * read_buf and panicing when we get an error from the
280 if ((imap.im_blkno + imap.im_len) >
281 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
283 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
284 "(imap.im_blkno (0x%llx) "
285 "+ imap.im_len (0x%llx)) > "
286 " XFS_FSB_TO_BB(mp, "
287 "mp->m_sb.sb_dblocks) (0x%llx)",
288 (unsigned long long) imap.im_blkno,
289 (unsigned long long) imap.im_len,
290 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
292 return XFS_ERROR(EINVAL);
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
311 ASSERT(bno == 0 || bno == imap.im_blkno);
314 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
315 * default to just a read_buf() call.
317 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
318 (int)imap.im_len, XFS_BUF_LOCK, &bp);
321 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
322 "xfs_trans_read_buf() returned error %d, "
323 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
324 error, (unsigned long long) imap.im_blkno,
325 (unsigned long long) imap.im_len);
331 * Validate the magic number and version of every inode in the buffer
332 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
333 * No validation is done here in userspace (xfs_repair).
335 #if !defined(__KERNEL__)
338 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
339 #else /* usual case */
343 for (i = 0; i < ni; i++) {
347 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
348 (i << mp->m_sb.sb_inodelog));
349 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
350 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
351 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
352 XFS_ERRTAG_ITOBP_INOTOBP,
353 XFS_RANDOM_ITOBP_INOTOBP))) {
354 if (imap_flags & XFS_IMAP_BULKSTAT) {
355 xfs_trans_brelse(tp, bp);
356 return XFS_ERROR(EINVAL);
360 "Device %s - bad inode magic/vsn "
361 "daddr %lld #%d (magic=%x)",
362 XFS_BUFTARG_NAME(mp->m_ddev_targp),
363 (unsigned long long)imap.im_blkno, i,
364 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
366 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
368 xfs_trans_brelse(tp, bp);
369 return XFS_ERROR(EFSCORRUPTED);
373 xfs_inobp_check(mp, bp);
376 * Mark the buffer as an inode buffer now that it looks good
378 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
381 * Set *dipp to point to the on-disk inode in the buffer.
383 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
389 * Move inode type and inode format specific information from the
390 * on-disk inode to the in-core inode. For fifos, devs, and sockets
391 * this means set if_rdev to the proper value. For files, directories,
392 * and symlinks this means to bring in the in-line data or extent
393 * pointers. For a file in B-tree format, only the root is immediately
394 * brought in-core. The rest will be in-lined in if_extents when it
395 * is first referenced (see xfs_iread_extents()).
402 xfs_attr_shortform_t *atp;
406 ip->i_df.if_ext_max =
407 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
411 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
412 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
413 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
414 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
415 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
416 (unsigned long long)ip->i_ino,
417 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
418 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
420 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
421 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
423 return XFS_ERROR(EFSCORRUPTED);
426 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
427 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
428 "corrupt dinode %Lu, forkoff = 0x%x.",
429 (unsigned long long)ip->i_ino,
430 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
431 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
433 return XFS_ERROR(EFSCORRUPTED);
436 switch (ip->i_d.di_mode & S_IFMT) {
441 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
442 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
444 return XFS_ERROR(EFSCORRUPTED);
448 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
454 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
455 case XFS_DINODE_FMT_LOCAL:
457 * no local regular files yet
459 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
460 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
462 "(local format for regular file).",
463 (unsigned long long) ip->i_ino);
464 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
467 return XFS_ERROR(EFSCORRUPTED);
470 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
471 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
472 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
474 "(bad size %Ld for local inode).",
475 (unsigned long long) ip->i_ino,
476 (long long) di_size);
477 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
480 return XFS_ERROR(EFSCORRUPTED);
484 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
486 case XFS_DINODE_FMT_EXTENTS:
487 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
489 case XFS_DINODE_FMT_BTREE:
490 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
493 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
495 return XFS_ERROR(EFSCORRUPTED);
500 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
501 return XFS_ERROR(EFSCORRUPTED);
506 if (!XFS_DFORK_Q(dip))
508 ASSERT(ip->i_afp == NULL);
509 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
510 ip->i_afp->if_ext_max =
511 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
512 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
513 case XFS_DINODE_FMT_LOCAL:
514 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
515 size = be16_to_cpu(atp->hdr.totsize);
516 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
518 case XFS_DINODE_FMT_EXTENTS:
519 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
521 case XFS_DINODE_FMT_BTREE:
522 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
525 error = XFS_ERROR(EFSCORRUPTED);
529 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
531 xfs_idestroy_fork(ip, XFS_DATA_FORK);
537 * The file is in-lined in the on-disk inode.
538 * If it fits into if_inline_data, then copy
539 * it there, otherwise allocate a buffer for it
540 * and copy the data there. Either way, set
541 * if_data to point at the data.
542 * If we allocate a buffer for the data, make
543 * sure that its size is a multiple of 4 and
544 * record the real size in i_real_bytes.
557 * If the size is unreasonable, then something
558 * is wrong and we just bail out rather than crash in
559 * kmem_alloc() or memcpy() below.
561 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
562 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
564 "(bad size %d for local fork, size = %d).",
565 (unsigned long long) ip->i_ino, size,
566 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
567 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
569 return XFS_ERROR(EFSCORRUPTED);
571 ifp = XFS_IFORK_PTR(ip, whichfork);
574 ifp->if_u1.if_data = NULL;
575 else if (size <= sizeof(ifp->if_u2.if_inline_data))
576 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
578 real_size = roundup(size, 4);
579 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
581 ifp->if_bytes = size;
582 ifp->if_real_bytes = real_size;
584 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
585 ifp->if_flags &= ~XFS_IFEXTENTS;
586 ifp->if_flags |= XFS_IFINLINE;
591 * The file consists of a set of extents all
592 * of which fit into the on-disk inode.
593 * If there are few enough extents to fit into
594 * the if_inline_ext, then copy them there.
595 * Otherwise allocate a buffer for them and copy
596 * them into it. Either way, set if_extents
597 * to point at the extents.
605 xfs_bmbt_rec_t *ep, *dp;
611 ifp = XFS_IFORK_PTR(ip, whichfork);
612 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
613 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
616 * If the number of extents is unreasonable, then something
617 * is wrong and we just bail out rather than crash in
618 * kmem_alloc() or memcpy() below.
620 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
621 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
622 "corrupt inode %Lu ((a)extents = %d).",
623 (unsigned long long) ip->i_ino, nex);
624 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
626 return XFS_ERROR(EFSCORRUPTED);
629 ifp->if_real_bytes = 0;
631 ifp->if_u1.if_extents = NULL;
632 else if (nex <= XFS_INLINE_EXTS)
633 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
635 xfs_iext_add(ifp, 0, nex);
637 ifp->if_bytes = size;
639 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
640 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
641 for (i = 0; i < nex; i++, dp++) {
642 ep = xfs_iext_get_ext(ifp, i);
643 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
645 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
648 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
650 if (whichfork != XFS_DATA_FORK ||
651 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
652 if (unlikely(xfs_check_nostate_extents(
654 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
657 return XFS_ERROR(EFSCORRUPTED);
660 ifp->if_flags |= XFS_IFEXTENTS;
665 * The file has too many extents to fit into
666 * the inode, so they are in B-tree format.
667 * Allocate a buffer for the root of the B-tree
668 * and copy the root into it. The i_extents
669 * field will remain NULL until all of the
670 * extents are read in (when they are needed).
678 xfs_bmdr_block_t *dfp;
684 ifp = XFS_IFORK_PTR(ip, whichfork);
685 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
686 size = XFS_BMAP_BROOT_SPACE(dfp);
687 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
690 * blow out if -- fork has less extents than can fit in
691 * fork (fork shouldn't be a btree format), root btree
692 * block has more records than can fit into the fork,
693 * or the number of extents is greater than the number of
696 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
697 || XFS_BMDR_SPACE_CALC(nrecs) >
698 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
699 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
700 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
701 "corrupt inode %Lu (btree).",
702 (unsigned long long) ip->i_ino);
703 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
705 return XFS_ERROR(EFSCORRUPTED);
708 ifp->if_broot_bytes = size;
709 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
710 ASSERT(ifp->if_broot != NULL);
712 * Copy and convert from the on-disk structure
713 * to the in-memory structure.
715 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
716 ifp->if_broot, size);
717 ifp->if_flags &= ~XFS_IFEXTENTS;
718 ifp->if_flags |= XFS_IFBROOT;
724 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
727 * buf = on-disk representation
728 * dip = native representation
729 * dir = direction - +ve -> disk to native
730 * -ve -> native to disk
733 xfs_xlate_dinode_core(
735 xfs_dinode_core_t *dip,
738 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
739 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
740 xfs_arch_t arch = ARCH_CONVERT;
744 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
745 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
746 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
747 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
748 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
749 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
750 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
751 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
752 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
755 memcpy(mem_core->di_pad, buf_core->di_pad,
756 sizeof(buf_core->di_pad));
758 memcpy(buf_core->di_pad, mem_core->di_pad,
759 sizeof(buf_core->di_pad));
762 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
764 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
766 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
768 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
770 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
772 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
774 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
776 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
777 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
778 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
779 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
780 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
781 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
782 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
783 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
784 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
785 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
786 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
795 if (di_flags & XFS_DIFLAG_ANY) {
796 if (di_flags & XFS_DIFLAG_REALTIME)
797 flags |= XFS_XFLAG_REALTIME;
798 if (di_flags & XFS_DIFLAG_PREALLOC)
799 flags |= XFS_XFLAG_PREALLOC;
800 if (di_flags & XFS_DIFLAG_IMMUTABLE)
801 flags |= XFS_XFLAG_IMMUTABLE;
802 if (di_flags & XFS_DIFLAG_APPEND)
803 flags |= XFS_XFLAG_APPEND;
804 if (di_flags & XFS_DIFLAG_SYNC)
805 flags |= XFS_XFLAG_SYNC;
806 if (di_flags & XFS_DIFLAG_NOATIME)
807 flags |= XFS_XFLAG_NOATIME;
808 if (di_flags & XFS_DIFLAG_NODUMP)
809 flags |= XFS_XFLAG_NODUMP;
810 if (di_flags & XFS_DIFLAG_RTINHERIT)
811 flags |= XFS_XFLAG_RTINHERIT;
812 if (di_flags & XFS_DIFLAG_PROJINHERIT)
813 flags |= XFS_XFLAG_PROJINHERIT;
814 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
815 flags |= XFS_XFLAG_NOSYMLINKS;
816 if (di_flags & XFS_DIFLAG_EXTSIZE)
817 flags |= XFS_XFLAG_EXTSIZE;
818 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
819 flags |= XFS_XFLAG_EXTSZINHERIT;
820 if (di_flags & XFS_DIFLAG_NODEFRAG)
821 flags |= XFS_XFLAG_NODEFRAG;
822 if (di_flags & XFS_DIFLAG_FILESTREAM)
823 flags |= XFS_XFLAG_FILESTREAM;
833 xfs_dinode_core_t *dic = &ip->i_d;
835 return _xfs_dic2xflags(dic->di_flags) |
836 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
841 xfs_dinode_core_t *dic)
843 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
844 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
848 * Given a mount structure and an inode number, return a pointer
849 * to a newly allocated in-core inode corresponding to the given
852 * Initialize the inode's attributes and extent pointers if it
853 * already has them (it will not if the inode has no links).
869 ASSERT(xfs_inode_zone != NULL);
871 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
874 spin_lock_init(&ip->i_flags_lock);
877 * Get pointer's to the on-disk inode and the buffer containing it.
878 * If the inode number refers to a block outside the file system
879 * then xfs_itobp() will return NULL. In this case we should
880 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
881 * know that this is a new incore inode.
883 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
885 kmem_zone_free(xfs_inode_zone, ip);
890 * Initialize inode's trace buffers.
891 * Do this before xfs_iformat in case it adds entries.
893 #ifdef XFS_BMAP_TRACE
894 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
896 #ifdef XFS_BMBT_TRACE
897 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
900 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
902 #ifdef XFS_ILOCK_TRACE
903 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
905 #ifdef XFS_DIR2_TRACE
906 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
910 * If we got something that isn't an inode it means someone
911 * (nfs or dmi) has a stale handle.
913 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
914 kmem_zone_free(xfs_inode_zone, ip);
915 xfs_trans_brelse(tp, bp);
917 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
918 "dip->di_core.di_magic (0x%x) != "
919 "XFS_DINODE_MAGIC (0x%x)",
920 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
923 return XFS_ERROR(EINVAL);
927 * If the on-disk inode is already linked to a directory
928 * entry, copy all of the inode into the in-core inode.
929 * xfs_iformat() handles copying in the inode format
930 * specific information.
931 * Otherwise, just get the truly permanent information.
933 if (dip->di_core.di_mode) {
934 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
936 error = xfs_iformat(ip, dip);
938 kmem_zone_free(xfs_inode_zone, ip);
939 xfs_trans_brelse(tp, bp);
941 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
942 "xfs_iformat() returned error %d",
948 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
949 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
950 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
951 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
953 * Make sure to pull in the mode here as well in
954 * case the inode is released without being used.
955 * This ensures that xfs_inactive() will see that
956 * the inode is already free and not try to mess
957 * with the uninitialized part of it.
961 * Initialize the per-fork minima and maxima for a new
962 * inode here. xfs_iformat will do it for old inodes.
964 ip->i_df.if_ext_max =
965 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
968 INIT_LIST_HEAD(&ip->i_reclaim);
971 * The inode format changed when we moved the link count and
972 * made it 32 bits long. If this is an old format inode,
973 * convert it in memory to look like a new one. If it gets
974 * flushed to disk we will convert back before flushing or
975 * logging it. We zero out the new projid field and the old link
976 * count field. We'll handle clearing the pad field (the remains
977 * of the old uuid field) when we actually convert the inode to
978 * the new format. We don't change the version number so that we
979 * can distinguish this from a real new format inode.
981 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
982 ip->i_d.di_nlink = ip->i_d.di_onlink;
983 ip->i_d.di_onlink = 0;
984 ip->i_d.di_projid = 0;
987 ip->i_delayed_blks = 0;
988 ip->i_size = ip->i_d.di_size;
991 * Mark the buffer containing the inode as something to keep
992 * around for a while. This helps to keep recently accessed
993 * meta-data in-core longer.
995 XFS_BUF_SET_REF(bp, XFS_INO_REF);
998 * Use xfs_trans_brelse() to release the buffer containing the
999 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1000 * in xfs_itobp() above. If tp is NULL, this is just a normal
1001 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1002 * will only release the buffer if it is not dirty within the
1003 * transaction. It will be OK to release the buffer in this case,
1004 * because inodes on disk are never destroyed and we will be
1005 * locking the new in-core inode before putting it in the hash
1006 * table where other processes can find it. Thus we don't have
1007 * to worry about the inode being changed just because we released
1010 xfs_trans_brelse(tp, bp);
1016 * Read in extents from a btree-format inode.
1017 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1027 xfs_extnum_t nextents;
1030 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1031 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1033 return XFS_ERROR(EFSCORRUPTED);
1035 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1036 size = nextents * sizeof(xfs_bmbt_rec_t);
1037 ifp = XFS_IFORK_PTR(ip, whichfork);
1040 * We know that the size is valid (it's checked in iformat_btree)
1042 ifp->if_lastex = NULLEXTNUM;
1043 ifp->if_bytes = ifp->if_real_bytes = 0;
1044 ifp->if_flags |= XFS_IFEXTENTS;
1045 xfs_iext_add(ifp, 0, nextents);
1046 error = xfs_bmap_read_extents(tp, ip, whichfork);
1048 xfs_iext_destroy(ifp);
1049 ifp->if_flags &= ~XFS_IFEXTENTS;
1052 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1057 * Allocate an inode on disk and return a copy of its in-core version.
1058 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1059 * appropriately within the inode. The uid and gid for the inode are
1060 * set according to the contents of the given cred structure.
1062 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1063 * has a free inode available, call xfs_iget()
1064 * to obtain the in-core version of the allocated inode. Finally,
1065 * fill in the inode and log its initial contents. In this case,
1066 * ialloc_context would be set to NULL and call_again set to false.
1068 * If xfs_dialloc() does not have an available inode,
1069 * it will replenish its supply by doing an allocation. Since we can
1070 * only do one allocation within a transaction without deadlocks, we
1071 * must commit the current transaction before returning the inode itself.
1072 * In this case, therefore, we will set call_again to true and return.
1073 * The caller should then commit the current transaction, start a new
1074 * transaction, and call xfs_ialloc() again to actually get the inode.
1076 * To ensure that some other process does not grab the inode that
1077 * was allocated during the first call to xfs_ialloc(), this routine
1078 * also returns the [locked] bp pointing to the head of the freelist
1079 * as ialloc_context. The caller should hold this buffer across
1080 * the commit and pass it back into this routine on the second call.
1092 xfs_buf_t **ialloc_context,
1093 boolean_t *call_again,
1103 * Call the space management code to pick
1104 * the on-disk inode to be allocated.
1106 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1107 ialloc_context, call_again, &ino);
1111 if (*call_again || ino == NULLFSINO) {
1115 ASSERT(*ialloc_context == NULL);
1118 * Get the in-core inode with the lock held exclusively.
1119 * This is because we're setting fields here we need
1120 * to prevent others from looking at until we're done.
1122 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1123 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1130 ip->i_d.di_mode = (__uint16_t)mode;
1131 ip->i_d.di_onlink = 0;
1132 ip->i_d.di_nlink = nlink;
1133 ASSERT(ip->i_d.di_nlink == nlink);
1134 ip->i_d.di_uid = current_fsuid(cr);
1135 ip->i_d.di_gid = current_fsgid(cr);
1136 ip->i_d.di_projid = prid;
1137 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1140 * If the superblock version is up to where we support new format
1141 * inodes and this is currently an old format inode, then change
1142 * the inode version number now. This way we only do the conversion
1143 * here rather than here and in the flush/logging code.
1145 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1146 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1147 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1149 * We've already zeroed the old link count, the projid field,
1150 * and the pad field.
1155 * Project ids won't be stored on disk if we are using a version 1 inode.
1157 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1158 xfs_bump_ino_vers2(tp, ip);
1160 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1161 ip->i_d.di_gid = pip->i_d.di_gid;
1162 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1163 ip->i_d.di_mode |= S_ISGID;
1168 * If the group ID of the new file does not match the effective group
1169 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1170 * (and only if the irix_sgid_inherit compatibility variable is set).
1172 if ((irix_sgid_inherit) &&
1173 (ip->i_d.di_mode & S_ISGID) &&
1174 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1175 ip->i_d.di_mode &= ~S_ISGID;
1178 ip->i_d.di_size = 0;
1180 ip->i_d.di_nextents = 0;
1181 ASSERT(ip->i_d.di_nblocks == 0);
1182 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1184 * di_gen will have been taken care of in xfs_iread.
1186 ip->i_d.di_extsize = 0;
1187 ip->i_d.di_dmevmask = 0;
1188 ip->i_d.di_dmstate = 0;
1189 ip->i_d.di_flags = 0;
1190 flags = XFS_ILOG_CORE;
1191 switch (mode & S_IFMT) {
1196 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1197 ip->i_df.if_u2.if_rdev = rdev;
1198 ip->i_df.if_flags = 0;
1199 flags |= XFS_ILOG_DEV;
1202 if (xfs_inode_is_filestream(pip)) {
1203 error = xfs_filestream_associate(pip, ip);
1207 xfs_iflags_set(ip, XFS_IFILESTREAM);
1211 if (pip->i_d.di_flags & XFS_DIFLAG_ANY) {
1214 if ((mode & S_IFMT) == S_IFDIR) {
1215 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1216 di_flags |= XFS_DIFLAG_RTINHERIT;
1217 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1218 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1219 ip->i_d.di_extsize = pip->i_d.di_extsize;
1221 } else if ((mode & S_IFMT) == S_IFREG) {
1222 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1223 di_flags |= XFS_DIFLAG_REALTIME;
1224 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1226 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1227 di_flags |= XFS_DIFLAG_EXTSIZE;
1228 ip->i_d.di_extsize = pip->i_d.di_extsize;
1231 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1232 xfs_inherit_noatime)
1233 di_flags |= XFS_DIFLAG_NOATIME;
1234 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1236 di_flags |= XFS_DIFLAG_NODUMP;
1237 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1239 di_flags |= XFS_DIFLAG_SYNC;
1240 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1241 xfs_inherit_nosymlinks)
1242 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1243 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1244 di_flags |= XFS_DIFLAG_PROJINHERIT;
1245 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1246 xfs_inherit_nodefrag)
1247 di_flags |= XFS_DIFLAG_NODEFRAG;
1248 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1249 di_flags |= XFS_DIFLAG_FILESTREAM;
1250 ip->i_d.di_flags |= di_flags;
1254 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1255 ip->i_df.if_flags = XFS_IFEXTENTS;
1256 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1257 ip->i_df.if_u1.if_extents = NULL;
1263 * Attribute fork settings for new inode.
1265 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1266 ip->i_d.di_anextents = 0;
1269 * Log the new values stuffed into the inode.
1271 xfs_trans_log_inode(tp, ip, flags);
1273 /* now that we have an i_mode we can setup inode ops and unlock */
1274 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1281 * Check to make sure that there are no blocks allocated to the
1282 * file beyond the size of the file. We don't check this for
1283 * files with fixed size extents or real time extents, but we
1284 * at least do it for regular files.
1293 xfs_fileoff_t map_first;
1295 xfs_bmbt_irec_t imaps[2];
1297 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1300 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1304 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1306 * The filesystem could be shutting down, so bmapi may return
1309 if (xfs_bmapi(NULL, ip, map_first,
1311 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1313 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1316 ASSERT(nimaps == 1);
1317 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1322 * Calculate the last possible buffered byte in a file. This must
1323 * include data that was buffered beyond the EOF by the write code.
1324 * This also needs to deal with overflowing the xfs_fsize_t type
1325 * which can happen for sizes near the limit.
1327 * We also need to take into account any blocks beyond the EOF. It
1328 * may be the case that they were buffered by a write which failed.
1329 * In that case the pages will still be in memory, but the inode size
1330 * will never have been updated.
1337 xfs_fsize_t last_byte;
1338 xfs_fileoff_t last_block;
1339 xfs_fileoff_t size_last_block;
1342 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1346 * Only check for blocks beyond the EOF if the extents have
1347 * been read in. This eliminates the need for the inode lock,
1348 * and it also saves us from looking when it really isn't
1351 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1352 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1360 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1361 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1363 last_byte = XFS_FSB_TO_B(mp, last_block);
1364 if (last_byte < 0) {
1365 return XFS_MAXIOFFSET(mp);
1367 last_byte += (1 << mp->m_writeio_log);
1368 if (last_byte < 0) {
1369 return XFS_MAXIOFFSET(mp);
1374 #if defined(XFS_RW_TRACE)
1380 xfs_fsize_t new_size,
1381 xfs_off_t toss_start,
1382 xfs_off_t toss_finish)
1384 if (ip->i_rwtrace == NULL) {
1388 ktrace_enter(ip->i_rwtrace,
1391 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1392 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1393 (void*)((long)flag),
1394 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1395 (void*)(unsigned long)(new_size & 0xffffffff),
1396 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1397 (void*)(unsigned long)(toss_start & 0xffffffff),
1398 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1399 (void*)(unsigned long)(toss_finish & 0xffffffff),
1400 (void*)(unsigned long)current_cpu(),
1401 (void*)(unsigned long)current_pid(),
1407 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1411 * Start the truncation of the file to new_size. The new size
1412 * must be smaller than the current size. This routine will
1413 * clear the buffer and page caches of file data in the removed
1414 * range, and xfs_itruncate_finish() will remove the underlying
1417 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1418 * must NOT have the inode lock held at all. This is because we're
1419 * calling into the buffer/page cache code and we can't hold the
1420 * inode lock when we do so.
1422 * We need to wait for any direct I/Os in flight to complete before we
1423 * proceed with the truncate. This is needed to prevent the extents
1424 * being read or written by the direct I/Os from being removed while the
1425 * I/O is in flight as there is no other method of synchronising
1426 * direct I/O with the truncate operation. Also, because we hold
1427 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1428 * started until the truncate completes and drops the lock. Essentially,
1429 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1430 * between direct I/Os and the truncate operation.
1432 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1433 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1434 * in the case that the caller is locking things out of order and
1435 * may not be able to call xfs_itruncate_finish() with the inode lock
1436 * held without dropping the I/O lock. If the caller must drop the
1437 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1438 * must be called again with all the same restrictions as the initial
1442 xfs_itruncate_start(
1445 xfs_fsize_t new_size)
1447 xfs_fsize_t last_byte;
1448 xfs_off_t toss_start;
1453 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1454 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1455 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1456 (flags == XFS_ITRUNC_MAYBE));
1461 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1464 * Call toss_pages or flushinval_pages to get rid of pages
1465 * overlapping the region being removed. We have to use
1466 * the less efficient flushinval_pages in the case that the
1467 * caller may not be able to finish the truncate without
1468 * dropping the inode's I/O lock. Make sure
1469 * to catch any pages brought in by buffers overlapping
1470 * the EOF by searching out beyond the isize by our
1471 * block size. We round new_size up to a block boundary
1472 * so that we don't toss things on the same block as
1473 * new_size but before it.
1475 * Before calling toss_page or flushinval_pages, make sure to
1476 * call remapf() over the same region if the file is mapped.
1477 * This frees up mapped file references to the pages in the
1478 * given range and for the flushinval_pages case it ensures
1479 * that we get the latest mapped changes flushed out.
1481 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1482 toss_start = XFS_FSB_TO_B(mp, toss_start);
1483 if (toss_start < 0) {
1485 * The place to start tossing is beyond our maximum
1486 * file size, so there is no way that the data extended
1491 last_byte = xfs_file_last_byte(ip);
1492 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1494 if (last_byte > toss_start) {
1495 if (flags & XFS_ITRUNC_DEFINITE) {
1496 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1498 error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1503 if (new_size == 0) {
1504 ASSERT(VN_CACHED(vp) == 0);
1511 * Shrink the file to the given new_size. The new
1512 * size must be smaller than the current size.
1513 * This will free up the underlying blocks
1514 * in the removed range after a call to xfs_itruncate_start()
1515 * or xfs_atruncate_start().
1517 * The transaction passed to this routine must have made
1518 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1519 * This routine may commit the given transaction and
1520 * start new ones, so make sure everything involved in
1521 * the transaction is tidy before calling here.
1522 * Some transaction will be returned to the caller to be
1523 * committed. The incoming transaction must already include
1524 * the inode, and both inode locks must be held exclusively.
1525 * The inode must also be "held" within the transaction. On
1526 * return the inode will be "held" within the returned transaction.
1527 * This routine does NOT require any disk space to be reserved
1528 * for it within the transaction.
1530 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1531 * and it indicates the fork which is to be truncated. For the
1532 * attribute fork we only support truncation to size 0.
1534 * We use the sync parameter to indicate whether or not the first
1535 * transaction we perform might have to be synchronous. For the attr fork,
1536 * it needs to be so if the unlink of the inode is not yet known to be
1537 * permanent in the log. This keeps us from freeing and reusing the
1538 * blocks of the attribute fork before the unlink of the inode becomes
1541 * For the data fork, we normally have to run synchronously if we're
1542 * being called out of the inactive path or we're being called
1543 * out of the create path where we're truncating an existing file.
1544 * Either way, the truncate needs to be sync so blocks don't reappear
1545 * in the file with altered data in case of a crash. wsync filesystems
1546 * can run the first case async because anything that shrinks the inode
1547 * has to run sync so by the time we're called here from inactive, the
1548 * inode size is permanently set to 0.
1550 * Calls from the truncate path always need to be sync unless we're
1551 * in a wsync filesystem and the file has already been unlinked.
1553 * The caller is responsible for correctly setting the sync parameter.
1554 * It gets too hard for us to guess here which path we're being called
1555 * out of just based on inode state.
1558 xfs_itruncate_finish(
1561 xfs_fsize_t new_size,
1565 xfs_fsblock_t first_block;
1566 xfs_fileoff_t first_unmap_block;
1567 xfs_fileoff_t last_block;
1568 xfs_filblks_t unmap_len=0;
1573 xfs_bmap_free_t free_list;
1576 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1577 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1578 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1579 ASSERT(*tp != NULL);
1580 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1581 ASSERT(ip->i_transp == *tp);
1582 ASSERT(ip->i_itemp != NULL);
1583 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1587 mp = (ntp)->t_mountp;
1588 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1591 * We only support truncating the entire attribute fork.
1593 if (fork == XFS_ATTR_FORK) {
1596 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1597 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1599 * The first thing we do is set the size to new_size permanently
1600 * on disk. This way we don't have to worry about anyone ever
1601 * being able to look at the data being freed even in the face
1602 * of a crash. What we're getting around here is the case where
1603 * we free a block, it is allocated to another file, it is written
1604 * to, and then we crash. If the new data gets written to the
1605 * file but the log buffers containing the free and reallocation
1606 * don't, then we'd end up with garbage in the blocks being freed.
1607 * As long as we make the new_size permanent before actually
1608 * freeing any blocks it doesn't matter if they get writtten to.
1610 * The callers must signal into us whether or not the size
1611 * setting here must be synchronous. There are a few cases
1612 * where it doesn't have to be synchronous. Those cases
1613 * occur if the file is unlinked and we know the unlink is
1614 * permanent or if the blocks being truncated are guaranteed
1615 * to be beyond the inode eof (regardless of the link count)
1616 * and the eof value is permanent. Both of these cases occur
1617 * only on wsync-mounted filesystems. In those cases, we're
1618 * guaranteed that no user will ever see the data in the blocks
1619 * that are being truncated so the truncate can run async.
1620 * In the free beyond eof case, the file may wind up with
1621 * more blocks allocated to it than it needs if we crash
1622 * and that won't get fixed until the next time the file
1623 * is re-opened and closed but that's ok as that shouldn't
1624 * be too many blocks.
1626 * However, we can't just make all wsync xactions run async
1627 * because there's one call out of the create path that needs
1628 * to run sync where it's truncating an existing file to size
1629 * 0 whose size is > 0.
1631 * It's probably possible to come up with a test in this
1632 * routine that would correctly distinguish all the above
1633 * cases from the values of the function parameters and the
1634 * inode state but for sanity's sake, I've decided to let the
1635 * layers above just tell us. It's simpler to correctly figure
1636 * out in the layer above exactly under what conditions we
1637 * can run async and I think it's easier for others read and
1638 * follow the logic in case something has to be changed.
1639 * cscope is your friend -- rcc.
1641 * The attribute fork is much simpler.
1643 * For the attribute fork we allow the caller to tell us whether
1644 * the unlink of the inode that led to this call is yet permanent
1645 * in the on disk log. If it is not and we will be freeing extents
1646 * in this inode then we make the first transaction synchronous
1647 * to make sure that the unlink is permanent by the time we free
1650 if (fork == XFS_DATA_FORK) {
1651 if (ip->i_d.di_nextents > 0) {
1653 * If we are not changing the file size then do
1654 * not update the on-disk file size - we may be
1655 * called from xfs_inactive_free_eofblocks(). If we
1656 * update the on-disk file size and then the system
1657 * crashes before the contents of the file are
1658 * flushed to disk then the files may be full of
1659 * holes (ie NULL files bug).
1661 if (ip->i_size != new_size) {
1662 ip->i_d.di_size = new_size;
1663 ip->i_size = new_size;
1664 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1668 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1669 if (ip->i_d.di_anextents > 0)
1670 xfs_trans_set_sync(ntp);
1672 ASSERT(fork == XFS_DATA_FORK ||
1673 (fork == XFS_ATTR_FORK &&
1674 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1675 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1678 * Since it is possible for space to become allocated beyond
1679 * the end of the file (in a crash where the space is allocated
1680 * but the inode size is not yet updated), simply remove any
1681 * blocks which show up between the new EOF and the maximum
1682 * possible file size. If the first block to be removed is
1683 * beyond the maximum file size (ie it is the same as last_block),
1684 * then there is nothing to do.
1686 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1687 ASSERT(first_unmap_block <= last_block);
1689 if (last_block == first_unmap_block) {
1692 unmap_len = last_block - first_unmap_block + 1;
1696 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1697 * will tell us whether it freed the entire range or
1698 * not. If this is a synchronous mount (wsync),
1699 * then we can tell bunmapi to keep all the
1700 * transactions asynchronous since the unlink
1701 * transaction that made this inode inactive has
1702 * already hit the disk. There's no danger of
1703 * the freed blocks being reused, there being a
1704 * crash, and the reused blocks suddenly reappearing
1705 * in this file with garbage in them once recovery
1708 XFS_BMAP_INIT(&free_list, &first_block);
1709 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1710 first_unmap_block, unmap_len,
1711 XFS_BMAPI_AFLAG(fork) |
1712 (sync ? 0 : XFS_BMAPI_ASYNC),
1713 XFS_ITRUNC_MAX_EXTENTS,
1714 &first_block, &free_list,
1718 * If the bunmapi call encounters an error,
1719 * return to the caller where the transaction
1720 * can be properly aborted. We just need to
1721 * make sure we're not holding any resources
1722 * that we were not when we came in.
1724 xfs_bmap_cancel(&free_list);
1729 * Duplicate the transaction that has the permanent
1730 * reservation and commit the old transaction.
1732 error = xfs_bmap_finish(tp, &free_list, &committed);
1736 * If the bmap finish call encounters an error,
1737 * return to the caller where the transaction
1738 * can be properly aborted. We just need to
1739 * make sure we're not holding any resources
1740 * that we were not when we came in.
1742 * Aborting from this point might lose some
1743 * blocks in the file system, but oh well.
1745 xfs_bmap_cancel(&free_list);
1748 * If the passed in transaction committed
1749 * in xfs_bmap_finish(), then we want to
1750 * add the inode to this one before returning.
1751 * This keeps things simple for the higher
1752 * level code, because it always knows that
1753 * the inode is locked and held in the
1754 * transaction that returns to it whether
1755 * errors occur or not. We don't mark the
1756 * inode dirty so that this transaction can
1757 * be easily aborted if possible.
1759 xfs_trans_ijoin(ntp, ip,
1760 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1761 xfs_trans_ihold(ntp, ip);
1768 * The first xact was committed,
1769 * so add the inode to the new one.
1770 * Mark it dirty so it will be logged
1771 * and moved forward in the log as
1772 * part of every commit.
1774 xfs_trans_ijoin(ntp, ip,
1775 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1776 xfs_trans_ihold(ntp, ip);
1777 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1779 ntp = xfs_trans_dup(ntp);
1780 (void) xfs_trans_commit(*tp, 0);
1782 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1783 XFS_TRANS_PERM_LOG_RES,
1784 XFS_ITRUNCATE_LOG_COUNT);
1786 * Add the inode being truncated to the next chained
1789 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1790 xfs_trans_ihold(ntp, ip);
1795 * Only update the size in the case of the data fork, but
1796 * always re-log the inode so that our permanent transaction
1797 * can keep on rolling it forward in the log.
1799 if (fork == XFS_DATA_FORK) {
1800 xfs_isize_check(mp, ip, new_size);
1802 * If we are not changing the file size then do
1803 * not update the on-disk file size - we may be
1804 * called from xfs_inactive_free_eofblocks(). If we
1805 * update the on-disk file size and then the system
1806 * crashes before the contents of the file are
1807 * flushed to disk then the files may be full of
1808 * holes (ie NULL files bug).
1810 if (ip->i_size != new_size) {
1811 ip->i_d.di_size = new_size;
1812 ip->i_size = new_size;
1815 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1816 ASSERT((new_size != 0) ||
1817 (fork == XFS_ATTR_FORK) ||
1818 (ip->i_delayed_blks == 0));
1819 ASSERT((new_size != 0) ||
1820 (fork == XFS_ATTR_FORK) ||
1821 (ip->i_d.di_nextents == 0));
1822 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1830 * Do the first part of growing a file: zero any data in the last
1831 * block that is beyond the old EOF. We need to do this before
1832 * the inode is joined to the transaction to modify the i_size.
1833 * That way we can drop the inode lock and call into the buffer
1834 * cache to get the buffer mapping the EOF.
1839 xfs_fsize_t new_size,
1844 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1845 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1846 ASSERT(new_size > ip->i_size);
1849 * Zero any pages that may have been created by
1850 * xfs_write_file() beyond the end of the file
1851 * and any blocks between the old and new file sizes.
1853 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1861 * This routine is called to extend the size of a file.
1862 * The inode must have both the iolock and the ilock locked
1863 * for update and it must be a part of the current transaction.
1864 * The xfs_igrow_start() function must have been called previously.
1865 * If the change_flag is not zero, the inode change timestamp will
1872 xfs_fsize_t new_size,
1875 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1876 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1877 ASSERT(ip->i_transp == tp);
1878 ASSERT(new_size > ip->i_size);
1881 * Update the file size. Update the inode change timestamp
1882 * if change_flag set.
1884 ip->i_d.di_size = new_size;
1885 ip->i_size = new_size;
1887 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1888 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1894 * This is called when the inode's link count goes to 0.
1895 * We place the on-disk inode on a list in the AGI. It
1896 * will be pulled from this list when the inode is freed.
1908 xfs_agnumber_t agno;
1909 xfs_daddr_t agdaddr;
1916 ASSERT(ip->i_d.di_nlink == 0);
1917 ASSERT(ip->i_d.di_mode != 0);
1918 ASSERT(ip->i_transp == tp);
1922 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1923 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1926 * Get the agi buffer first. It ensures lock ordering
1929 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1930 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1935 * Validate the magic number of the agi block.
1937 agi = XFS_BUF_TO_AGI(agibp);
1939 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1940 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1941 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1942 XFS_RANDOM_IUNLINK))) {
1943 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1944 xfs_trans_brelse(tp, agibp);
1945 return XFS_ERROR(EFSCORRUPTED);
1948 * Get the index into the agi hash table for the
1949 * list this inode will go on.
1951 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1953 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1954 ASSERT(agi->agi_unlinked[bucket_index]);
1955 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1957 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1959 * There is already another inode in the bucket we need
1960 * to add ourselves to. Add us at the front of the list.
1961 * Here we put the head pointer into our next pointer,
1962 * and then we fall through to point the head at us.
1964 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1968 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1969 ASSERT(dip->di_next_unlinked);
1970 /* both on-disk, don't endian flip twice */
1971 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1972 offset = ip->i_boffset +
1973 offsetof(xfs_dinode_t, di_next_unlinked);
1974 xfs_trans_inode_buf(tp, ibp);
1975 xfs_trans_log_buf(tp, ibp, offset,
1976 (offset + sizeof(xfs_agino_t) - 1));
1977 xfs_inobp_check(mp, ibp);
1981 * Point the bucket head pointer at the inode being inserted.
1984 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1985 offset = offsetof(xfs_agi_t, agi_unlinked) +
1986 (sizeof(xfs_agino_t) * bucket_index);
1987 xfs_trans_log_buf(tp, agibp, offset,
1988 (offset + sizeof(xfs_agino_t) - 1));
1993 * Pull the on-disk inode from the AGI unlinked list.
2006 xfs_agnumber_t agno;
2007 xfs_daddr_t agdaddr;
2009 xfs_agino_t next_agino;
2010 xfs_buf_t *last_ibp;
2011 xfs_dinode_t *last_dip = NULL;
2013 int offset, last_offset = 0;
2018 * First pull the on-disk inode from the AGI unlinked list.
2022 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2023 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2026 * Get the agi buffer first. It ensures lock ordering
2029 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2030 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2033 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2034 error, mp->m_fsname);
2038 * Validate the magic number of the agi block.
2040 agi = XFS_BUF_TO_AGI(agibp);
2042 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2043 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2044 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2045 XFS_RANDOM_IUNLINK_REMOVE))) {
2046 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2048 xfs_trans_brelse(tp, agibp);
2050 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2052 return XFS_ERROR(EFSCORRUPTED);
2055 * Get the index into the agi hash table for the
2056 * list this inode will go on.
2058 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2060 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2061 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2062 ASSERT(agi->agi_unlinked[bucket_index]);
2064 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2066 * We're at the head of the list. Get the inode's
2067 * on-disk buffer to see if there is anyone after us
2068 * on the list. Only modify our next pointer if it
2069 * is not already NULLAGINO. This saves us the overhead
2070 * of dealing with the buffer when there is no need to
2073 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2076 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2077 error, mp->m_fsname);
2080 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2081 ASSERT(next_agino != 0);
2082 if (next_agino != NULLAGINO) {
2083 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2084 offset = ip->i_boffset +
2085 offsetof(xfs_dinode_t, di_next_unlinked);
2086 xfs_trans_inode_buf(tp, ibp);
2087 xfs_trans_log_buf(tp, ibp, offset,
2088 (offset + sizeof(xfs_agino_t) - 1));
2089 xfs_inobp_check(mp, ibp);
2091 xfs_trans_brelse(tp, ibp);
2094 * Point the bucket head pointer at the next inode.
2096 ASSERT(next_agino != 0);
2097 ASSERT(next_agino != agino);
2098 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2099 offset = offsetof(xfs_agi_t, agi_unlinked) +
2100 (sizeof(xfs_agino_t) * bucket_index);
2101 xfs_trans_log_buf(tp, agibp, offset,
2102 (offset + sizeof(xfs_agino_t) - 1));
2105 * We need to search the list for the inode being freed.
2107 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2109 while (next_agino != agino) {
2111 * If the last inode wasn't the one pointing to
2112 * us, then release its buffer since we're not
2113 * going to do anything with it.
2115 if (last_ibp != NULL) {
2116 xfs_trans_brelse(tp, last_ibp);
2118 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2119 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2120 &last_ibp, &last_offset);
2123 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2124 error, mp->m_fsname);
2127 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2128 ASSERT(next_agino != NULLAGINO);
2129 ASSERT(next_agino != 0);
2132 * Now last_ibp points to the buffer previous to us on
2133 * the unlinked list. Pull us from the list.
2135 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2138 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2139 error, mp->m_fsname);
2142 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2143 ASSERT(next_agino != 0);
2144 ASSERT(next_agino != agino);
2145 if (next_agino != NULLAGINO) {
2146 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2147 offset = ip->i_boffset +
2148 offsetof(xfs_dinode_t, di_next_unlinked);
2149 xfs_trans_inode_buf(tp, ibp);
2150 xfs_trans_log_buf(tp, ibp, offset,
2151 (offset + sizeof(xfs_agino_t) - 1));
2152 xfs_inobp_check(mp, ibp);
2154 xfs_trans_brelse(tp, ibp);
2157 * Point the previous inode on the list to the next inode.
2159 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2160 ASSERT(next_agino != 0);
2161 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2162 xfs_trans_inode_buf(tp, last_ibp);
2163 xfs_trans_log_buf(tp, last_ibp, offset,
2164 (offset + sizeof(xfs_agino_t) - 1));
2165 xfs_inobp_check(mp, last_ibp);
2170 STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
2172 return (((ip->i_itemp == NULL) ||
2173 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2174 (ip->i_update_core == 0));
2179 xfs_inode_t *free_ip,
2183 xfs_mount_t *mp = free_ip->i_mount;
2184 int blks_per_cluster;
2187 int i, j, found, pre_flushed;
2191 xfs_inode_t *ip, **ip_found;
2192 xfs_inode_log_item_t *iip;
2193 xfs_log_item_t *lip;
2196 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2197 blks_per_cluster = 1;
2198 ninodes = mp->m_sb.sb_inopblock;
2199 nbufs = XFS_IALLOC_BLOCKS(mp);
2201 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2202 mp->m_sb.sb_blocksize;
2203 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2204 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2207 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2209 for (j = 0; j < nbufs; j++, inum += ninodes) {
2210 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2211 XFS_INO_TO_AGBNO(mp, inum));
2215 * Look for each inode in memory and attempt to lock it,
2216 * we can be racing with flush and tail pushing here.
2217 * any inode we get the locks on, add to an array of
2218 * inode items to process later.
2220 * The get the buffer lock, we could beat a flush
2221 * or tail pushing thread to the lock here, in which
2222 * case they will go looking for the inode buffer
2223 * and fail, we need some other form of interlock
2227 for (i = 0; i < ninodes; i++) {
2228 ih = XFS_IHASH(mp, inum + i);
2229 read_lock(&ih->ih_lock);
2230 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2231 if (ip->i_ino == inum + i)
2235 /* Inode not in memory or we found it already,
2238 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2239 read_unlock(&ih->ih_lock);
2243 if (xfs_inode_clean(ip)) {
2244 read_unlock(&ih->ih_lock);
2248 /* If we can get the locks then add it to the
2249 * list, otherwise by the time we get the bp lock
2250 * below it will already be attached to the
2254 /* This inode will already be locked - by us, lets
2258 if (ip == free_ip) {
2259 if (xfs_iflock_nowait(ip)) {
2260 xfs_iflags_set(ip, XFS_ISTALE);
2261 if (xfs_inode_clean(ip)) {
2264 ip_found[found++] = ip;
2267 read_unlock(&ih->ih_lock);
2271 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2272 if (xfs_iflock_nowait(ip)) {
2273 xfs_iflags_set(ip, XFS_ISTALE);
2275 if (xfs_inode_clean(ip)) {
2277 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2279 ip_found[found++] = ip;
2282 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2286 read_unlock(&ih->ih_lock);
2289 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2290 mp->m_bsize * blks_per_cluster,
2294 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2296 if (lip->li_type == XFS_LI_INODE) {
2297 iip = (xfs_inode_log_item_t *)lip;
2298 ASSERT(iip->ili_logged == 1);
2299 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2301 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2303 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2306 lip = lip->li_bio_list;
2309 for (i = 0; i < found; i++) {
2314 ip->i_update_core = 0;
2316 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2320 iip->ili_last_fields = iip->ili_format.ilf_fields;
2321 iip->ili_format.ilf_fields = 0;
2322 iip->ili_logged = 1;
2324 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2327 xfs_buf_attach_iodone(bp,
2328 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2329 xfs_istale_done, (xfs_log_item_t *)iip);
2330 if (ip != free_ip) {
2331 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2335 if (found || pre_flushed)
2336 xfs_trans_stale_inode_buf(tp, bp);
2337 xfs_trans_binval(tp, bp);
2340 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2344 * This is called to return an inode to the inode free list.
2345 * The inode should already be truncated to 0 length and have
2346 * no pages associated with it. This routine also assumes that
2347 * the inode is already a part of the transaction.
2349 * The on-disk copy of the inode will have been added to the list
2350 * of unlinked inodes in the AGI. We need to remove the inode from
2351 * that list atomically with respect to freeing it here.
2357 xfs_bmap_free_t *flist)
2361 xfs_ino_t first_ino;
2363 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2364 ASSERT(ip->i_transp == tp);
2365 ASSERT(ip->i_d.di_nlink == 0);
2366 ASSERT(ip->i_d.di_nextents == 0);
2367 ASSERT(ip->i_d.di_anextents == 0);
2368 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2369 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2370 ASSERT(ip->i_d.di_nblocks == 0);
2373 * Pull the on-disk inode from the AGI unlinked list.
2375 error = xfs_iunlink_remove(tp, ip);
2380 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2384 ip->i_d.di_mode = 0; /* mark incore inode as free */
2385 ip->i_d.di_flags = 0;
2386 ip->i_d.di_dmevmask = 0;
2387 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2388 ip->i_df.if_ext_max =
2389 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2390 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2391 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2393 * Bump the generation count so no one will be confused
2394 * by reincarnations of this inode.
2397 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2400 xfs_ifree_cluster(ip, tp, first_ino);
2407 * Reallocate the space for if_broot based on the number of records
2408 * being added or deleted as indicated in rec_diff. Move the records
2409 * and pointers in if_broot to fit the new size. When shrinking this
2410 * will eliminate holes between the records and pointers created by
2411 * the caller. When growing this will create holes to be filled in
2414 * The caller must not request to add more records than would fit in
2415 * the on-disk inode root. If the if_broot is currently NULL, then
2416 * if we adding records one will be allocated. The caller must also
2417 * not request that the number of records go below zero, although
2418 * it can go to zero.
2420 * ip -- the inode whose if_broot area is changing
2421 * ext_diff -- the change in the number of records, positive or negative,
2422 * requested for the if_broot array.
2432 xfs_bmbt_block_t *new_broot;
2439 * Handle the degenerate case quietly.
2441 if (rec_diff == 0) {
2445 ifp = XFS_IFORK_PTR(ip, whichfork);
2448 * If there wasn't any memory allocated before, just
2449 * allocate it now and get out.
2451 if (ifp->if_broot_bytes == 0) {
2452 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2453 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2455 ifp->if_broot_bytes = (int)new_size;
2460 * If there is already an existing if_broot, then we need
2461 * to realloc() it and shift the pointers to their new
2462 * location. The records don't change location because
2463 * they are kept butted up against the btree block header.
2465 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2466 new_max = cur_max + rec_diff;
2467 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2468 ifp->if_broot = (xfs_bmbt_block_t *)
2469 kmem_realloc(ifp->if_broot,
2471 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2473 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2474 ifp->if_broot_bytes);
2475 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2477 ifp->if_broot_bytes = (int)new_size;
2478 ASSERT(ifp->if_broot_bytes <=
2479 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2480 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2485 * rec_diff is less than 0. In this case, we are shrinking the
2486 * if_broot buffer. It must already exist. If we go to zero
2487 * records, just get rid of the root and clear the status bit.
2489 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2490 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2491 new_max = cur_max + rec_diff;
2492 ASSERT(new_max >= 0);
2494 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2498 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2500 * First copy over the btree block header.
2502 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2505 ifp->if_flags &= ~XFS_IFBROOT;
2509 * Only copy the records and pointers if there are any.
2513 * First copy the records.
2515 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2516 ifp->if_broot_bytes);
2517 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2519 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2522 * Then copy the pointers.
2524 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2525 ifp->if_broot_bytes);
2526 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2528 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2530 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2531 ifp->if_broot = new_broot;
2532 ifp->if_broot_bytes = (int)new_size;
2533 ASSERT(ifp->if_broot_bytes <=
2534 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2540 * This is called when the amount of space needed for if_data
2541 * is increased or decreased. The change in size is indicated by
2542 * the number of bytes that need to be added or deleted in the
2543 * byte_diff parameter.
2545 * If the amount of space needed has decreased below the size of the
2546 * inline buffer, then switch to using the inline buffer. Otherwise,
2547 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2548 * to what is needed.
2550 * ip -- the inode whose if_data area is changing
2551 * byte_diff -- the change in the number of bytes, positive or negative,
2552 * requested for the if_data array.
2564 if (byte_diff == 0) {
2568 ifp = XFS_IFORK_PTR(ip, whichfork);
2569 new_size = (int)ifp->if_bytes + byte_diff;
2570 ASSERT(new_size >= 0);
2572 if (new_size == 0) {
2573 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2574 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2576 ifp->if_u1.if_data = NULL;
2578 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2580 * If the valid extents/data can fit in if_inline_ext/data,
2581 * copy them from the malloc'd vector and free it.
2583 if (ifp->if_u1.if_data == NULL) {
2584 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2585 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2586 ASSERT(ifp->if_real_bytes != 0);
2587 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2589 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2590 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2595 * Stuck with malloc/realloc.
2596 * For inline data, the underlying buffer must be
2597 * a multiple of 4 bytes in size so that it can be
2598 * logged and stay on word boundaries. We enforce
2601 real_size = roundup(new_size, 4);
2602 if (ifp->if_u1.if_data == NULL) {
2603 ASSERT(ifp->if_real_bytes == 0);
2604 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2605 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2607 * Only do the realloc if the underlying size
2608 * is really changing.
2610 if (ifp->if_real_bytes != real_size) {
2611 ifp->if_u1.if_data =
2612 kmem_realloc(ifp->if_u1.if_data,
2618 ASSERT(ifp->if_real_bytes == 0);
2619 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2620 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2624 ifp->if_real_bytes = real_size;
2625 ifp->if_bytes = new_size;
2626 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2633 * Map inode to disk block and offset.
2635 * mp -- the mount point structure for the current file system
2636 * tp -- the current transaction
2637 * ino -- the inode number of the inode to be located
2638 * imap -- this structure is filled in with the information necessary
2639 * to retrieve the given inode from disk
2640 * flags -- flags to pass to xfs_dilocate indicating whether or not
2641 * lookups in the inode btree were OK or not
2651 xfs_fsblock_t fsbno;
2656 fsbno = imap->im_blkno ?
2657 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2658 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2662 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2663 imap->im_len = XFS_FSB_TO_BB(mp, len);
2664 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2665 imap->im_ioffset = (ushort)off;
2666 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2677 ifp = XFS_IFORK_PTR(ip, whichfork);
2678 if (ifp->if_broot != NULL) {
2679 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2680 ifp->if_broot = NULL;
2684 * If the format is local, then we can't have an extents
2685 * array so just look for an inline data array. If we're
2686 * not local then we may or may not have an extents list,
2687 * so check and free it up if we do.
2689 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2690 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2691 (ifp->if_u1.if_data != NULL)) {
2692 ASSERT(ifp->if_real_bytes != 0);
2693 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2694 ifp->if_u1.if_data = NULL;
2695 ifp->if_real_bytes = 0;
2697 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2698 ((ifp->if_flags & XFS_IFEXTIREC) ||
2699 ((ifp->if_u1.if_extents != NULL) &&
2700 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2701 ASSERT(ifp->if_real_bytes != 0);
2702 xfs_iext_destroy(ifp);
2704 ASSERT(ifp->if_u1.if_extents == NULL ||
2705 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2706 ASSERT(ifp->if_real_bytes == 0);
2707 if (whichfork == XFS_ATTR_FORK) {
2708 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2714 * This is called free all the memory associated with an inode.
2715 * It must free the inode itself and any buffers allocated for
2716 * if_extents/if_data and if_broot. It must also free the lock
2717 * associated with the inode.
2724 switch (ip->i_d.di_mode & S_IFMT) {
2728 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2732 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2733 mrfree(&ip->i_lock);
2734 mrfree(&ip->i_iolock);
2735 freesema(&ip->i_flock);
2736 #ifdef XFS_BMAP_TRACE
2737 ktrace_free(ip->i_xtrace);
2739 #ifdef XFS_BMBT_TRACE
2740 ktrace_free(ip->i_btrace);
2743 ktrace_free(ip->i_rwtrace);
2745 #ifdef XFS_ILOCK_TRACE
2746 ktrace_free(ip->i_lock_trace);
2748 #ifdef XFS_DIR2_TRACE
2749 ktrace_free(ip->i_dir_trace);
2753 * Only if we are shutting down the fs will we see an
2754 * inode still in the AIL. If it is there, we should remove
2755 * it to prevent a use-after-free from occurring.
2757 xfs_mount_t *mp = ip->i_mount;
2758 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2761 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2762 XFS_FORCED_SHUTDOWN(ip->i_mount));
2763 if (lip->li_flags & XFS_LI_IN_AIL) {
2765 if (lip->li_flags & XFS_LI_IN_AIL)
2766 xfs_trans_delete_ail(mp, lip, s);
2770 xfs_inode_item_destroy(ip);
2772 kmem_zone_free(xfs_inode_zone, ip);
2777 * Increment the pin count of the given buffer.
2778 * This value is protected by ipinlock spinlock in the mount structure.
2784 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2786 atomic_inc(&ip->i_pincount);
2790 * Decrement the pin count of the given inode, and wake up
2791 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2792 * inode must have been previously pinned with a call to xfs_ipin().
2798 ASSERT(atomic_read(&ip->i_pincount) > 0);
2800 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
2803 * If the inode is currently being reclaimed, the link between
2804 * the bhv_vnode and the xfs_inode will be broken after the
2805 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2806 * set, then we can move forward and mark the linux inode dirty
2807 * knowing that it is still valid as it won't freed until after
2808 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2809 * i_flags_lock is used to synchronise the setting of the
2810 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2811 * can execute atomically w.r.t to reclaim by holding this lock
2814 * However, we still need to issue the unpin wakeup call as the
2815 * inode reclaim may be blocked waiting for the inode to become
2819 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
2820 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2821 struct inode *inode = NULL;
2824 inode = vn_to_inode(vp);
2825 BUG_ON(inode->i_state & I_CLEAR);
2827 /* make sync come back and flush this inode */
2828 if (!(inode->i_state & (I_NEW|I_FREEING)))
2829 mark_inode_dirty_sync(inode);
2831 spin_unlock(&ip->i_flags_lock);
2832 wake_up(&ip->i_ipin_wait);
2837 * This is called to wait for the given inode to be unpinned.
2838 * It will sleep until this happens. The caller must have the
2839 * inode locked in at least shared mode so that the buffer cannot
2840 * be subsequently pinned once someone is waiting for it to be
2847 xfs_inode_log_item_t *iip;
2850 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2852 if (atomic_read(&ip->i_pincount) == 0) {
2857 if (iip && iip->ili_last_lsn) {
2858 lsn = iip->ili_last_lsn;
2864 * Give the log a push so we don't wait here too long.
2866 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2868 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2873 * xfs_iextents_copy()
2875 * This is called to copy the REAL extents (as opposed to the delayed
2876 * allocation extents) from the inode into the given buffer. It
2877 * returns the number of bytes copied into the buffer.
2879 * If there are no delayed allocation extents, then we can just
2880 * memcpy() the extents into the buffer. Otherwise, we need to
2881 * examine each extent in turn and skip those which are delayed.
2886 xfs_bmbt_rec_t *buffer,
2890 xfs_bmbt_rec_t *dest_ep;
2892 #ifdef XFS_BMAP_TRACE
2893 static char fname[] = "xfs_iextents_copy";
2898 xfs_fsblock_t start_block;
2900 ifp = XFS_IFORK_PTR(ip, whichfork);
2901 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2902 ASSERT(ifp->if_bytes > 0);
2904 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2905 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2909 * There are some delayed allocation extents in the
2910 * inode, so copy the extents one at a time and skip
2911 * the delayed ones. There must be at least one
2912 * non-delayed extent.
2916 for (i = 0; i < nrecs; i++) {
2917 ep = xfs_iext_get_ext(ifp, i);
2918 start_block = xfs_bmbt_get_startblock(ep);
2919 if (ISNULLSTARTBLOCK(start_block)) {
2921 * It's a delayed allocation extent, so skip it.
2926 /* Translate to on disk format */
2927 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2928 (__uint64_t*)&dest_ep->l0);
2929 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2930 (__uint64_t*)&dest_ep->l1);
2934 ASSERT(copied != 0);
2935 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2937 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2941 * Each of the following cases stores data into the same region
2942 * of the on-disk inode, so only one of them can be valid at
2943 * any given time. While it is possible to have conflicting formats
2944 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2945 * in EXTENTS format, this can only happen when the fork has
2946 * changed formats after being modified but before being flushed.
2947 * In these cases, the format always takes precedence, because the
2948 * format indicates the current state of the fork.
2955 xfs_inode_log_item_t *iip,
2962 #ifdef XFS_TRANS_DEBUG
2965 static const short brootflag[2] =
2966 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2967 static const short dataflag[2] =
2968 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2969 static const short extflag[2] =
2970 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2974 ifp = XFS_IFORK_PTR(ip, whichfork);
2976 * This can happen if we gave up in iformat in an error path,
2977 * for the attribute fork.
2980 ASSERT(whichfork == XFS_ATTR_FORK);
2983 cp = XFS_DFORK_PTR(dip, whichfork);
2985 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2986 case XFS_DINODE_FMT_LOCAL:
2987 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2988 (ifp->if_bytes > 0)) {
2989 ASSERT(ifp->if_u1.if_data != NULL);
2990 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2991 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2995 case XFS_DINODE_FMT_EXTENTS:
2996 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2997 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2998 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2999 (ifp->if_bytes == 0));
3000 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
3001 (ifp->if_bytes > 0));
3002 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
3003 (ifp->if_bytes > 0)) {
3004 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
3005 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3010 case XFS_DINODE_FMT_BTREE:
3011 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3012 (ifp->if_broot_bytes > 0)) {
3013 ASSERT(ifp->if_broot != NULL);
3014 ASSERT(ifp->if_broot_bytes <=
3015 (XFS_IFORK_SIZE(ip, whichfork) +
3016 XFS_BROOT_SIZE_ADJ));
3017 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3018 (xfs_bmdr_block_t *)cp,
3019 XFS_DFORK_SIZE(dip, mp, whichfork));
3023 case XFS_DINODE_FMT_DEV:
3024 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3025 ASSERT(whichfork == XFS_DATA_FORK);
3026 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
3030 case XFS_DINODE_FMT_UUID:
3031 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3032 ASSERT(whichfork == XFS_DATA_FORK);
3033 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3047 * xfs_iflush() will write a modified inode's changes out to the
3048 * inode's on disk home. The caller must have the inode lock held
3049 * in at least shared mode and the inode flush semaphore must be
3050 * held as well. The inode lock will still be held upon return from
3051 * the call and the caller is free to unlock it.
3052 * The inode flush lock will be unlocked when the inode reaches the disk.
3053 * The flags indicate how the inode's buffer should be written out.
3060 xfs_inode_log_item_t *iip;
3068 int clcount; /* count of inodes clustered */
3070 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3073 XFS_STATS_INC(xs_iflush_count);
3075 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3076 ASSERT(issemalocked(&(ip->i_flock)));
3077 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3078 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3084 * If the inode isn't dirty, then just release the inode
3085 * flush lock and do nothing.
3087 if ((ip->i_update_core == 0) &&
3088 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3089 ASSERT((iip != NULL) ?
3090 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3096 * We can't flush the inode until it is unpinned, so
3097 * wait for it. We know noone new can pin it, because
3098 * we are holding the inode lock shared and you need
3099 * to hold it exclusively to pin the inode.
3101 xfs_iunpin_wait(ip);
3104 * This may have been unpinned because the filesystem is shutting
3105 * down forcibly. If that's the case we must not write this inode
3106 * to disk, because the log record didn't make it to disk!
3108 if (XFS_FORCED_SHUTDOWN(mp)) {
3109 ip->i_update_core = 0;
3111 iip->ili_format.ilf_fields = 0;
3113 return XFS_ERROR(EIO);
3117 * Get the buffer containing the on-disk inode.
3119 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3126 * Decide how buffer will be flushed out. This is done before
3127 * the call to xfs_iflush_int because this field is zeroed by it.
3129 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3131 * Flush out the inode buffer according to the directions
3132 * of the caller. In the cases where the caller has given
3133 * us a choice choose the non-delwri case. This is because
3134 * the inode is in the AIL and we need to get it out soon.
3137 case XFS_IFLUSH_SYNC:
3138 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3141 case XFS_IFLUSH_ASYNC:
3142 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3145 case XFS_IFLUSH_DELWRI:
3155 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3156 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3157 case XFS_IFLUSH_DELWRI:
3160 case XFS_IFLUSH_ASYNC:
3163 case XFS_IFLUSH_SYNC:
3174 * First flush out the inode that xfs_iflush was called with.
3176 error = xfs_iflush_int(ip, bp);
3183 * see if other inodes can be gathered into this write
3186 ip->i_chash->chl_buf = bp;
3188 ch = XFS_CHASH(mp, ip->i_blkno);
3189 s = mutex_spinlock(&ch->ch_lock);
3192 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3194 * Do an un-protected check to see if the inode is dirty and
3195 * is a candidate for flushing. These checks will be repeated
3196 * later after the appropriate locks are acquired.
3199 if ((iq->i_update_core == 0) &&
3201 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3202 xfs_ipincount(iq) == 0) {
3207 * Try to get locks. If any are unavailable,
3208 * then this inode cannot be flushed and is skipped.
3211 /* get inode locks (just i_lock) */
3212 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3213 /* get inode flush lock */
3214 if (xfs_iflock_nowait(iq)) {
3215 /* check if pinned */
3216 if (xfs_ipincount(iq) == 0) {
3217 /* arriving here means that
3218 * this inode can be flushed.
3219 * first re-check that it's
3223 if ((iq->i_update_core != 0)||
3225 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3227 error = xfs_iflush_int(iq, bp);
3231 goto cluster_corrupt_out;
3240 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3243 mutex_spinunlock(&ch->ch_lock, s);
3246 XFS_STATS_INC(xs_icluster_flushcnt);
3247 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3251 * If the buffer is pinned then push on the log so we won't
3252 * get stuck waiting in the write for too long.
3254 if (XFS_BUF_ISPINNED(bp)){
3255 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3258 if (flags & INT_DELWRI) {
3259 xfs_bdwrite(mp, bp);
3260 } else if (flags & INT_ASYNC) {
3261 xfs_bawrite(mp, bp);
3263 error = xfs_bwrite(mp, bp);
3269 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3270 xfs_iflush_abort(ip);
3272 * Unlocks the flush lock
3274 return XFS_ERROR(EFSCORRUPTED);
3276 cluster_corrupt_out:
3277 /* Corruption detected in the clustering loop. Invalidate the
3278 * inode buffer and shut down the filesystem.
3280 mutex_spinunlock(&ch->ch_lock, s);
3283 * Clean up the buffer. If it was B_DELWRI, just release it --
3284 * brelse can handle it with no problems. If not, shut down the
3285 * filesystem before releasing the buffer.
3287 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3291 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3295 * Just like incore_relse: if we have b_iodone functions,
3296 * mark the buffer as an error and call them. Otherwise
3297 * mark it as stale and brelse.
3299 if (XFS_BUF_IODONE_FUNC(bp)) {
3300 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3304 XFS_BUF_ERROR(bp,EIO);
3312 xfs_iflush_abort(iq);
3314 * Unlocks the flush lock
3316 return XFS_ERROR(EFSCORRUPTED);
3325 xfs_inode_log_item_t *iip;
3328 #ifdef XFS_TRANS_DEBUG
3333 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3334 ASSERT(issemalocked(&(ip->i_flock)));
3335 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3336 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3343 * If the inode isn't dirty, then just release the inode
3344 * flush lock and do nothing.
3346 if ((ip->i_update_core == 0) &&
3347 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3352 /* set *dip = inode's place in the buffer */
3353 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3356 * Clear i_update_core before copying out the data.
3357 * This is for coordination with our timestamp updates
3358 * that don't hold the inode lock. They will always
3359 * update the timestamps BEFORE setting i_update_core,
3360 * so if we clear i_update_core after they set it we
3361 * are guaranteed to see their updates to the timestamps.
3362 * I believe that this depends on strongly ordered memory
3363 * semantics, but we have that. We use the SYNCHRONIZE
3364 * macro to make sure that the compiler does not reorder
3365 * the i_update_core access below the data copy below.
3367 ip->i_update_core = 0;
3371 * Make sure to get the latest atime from the Linux inode.
3373 xfs_synchronize_atime(ip);
3375 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3376 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3377 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3378 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3379 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3382 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3383 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3384 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3385 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3386 ip->i_ino, ip, ip->i_d.di_magic);
3389 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3391 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3392 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3393 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3394 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3395 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3399 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3401 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3402 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3403 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3404 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3405 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3406 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3411 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3412 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3413 XFS_RANDOM_IFLUSH_5)) {
3414 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3415 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3417 ip->i_d.di_nextents + ip->i_d.di_anextents,
3422 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3423 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3424 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3425 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3426 ip->i_ino, ip->i_d.di_forkoff, ip);
3430 * bump the flush iteration count, used to detect flushes which
3431 * postdate a log record during recovery.
3434 ip->i_d.di_flushiter++;
3437 * Copy the dirty parts of the inode into the on-disk
3438 * inode. We always copy out the core of the inode,
3439 * because if the inode is dirty at all the core must
3442 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3444 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3445 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3446 ip->i_d.di_flushiter = 0;
3449 * If this is really an old format inode and the superblock version
3450 * has not been updated to support only new format inodes, then
3451 * convert back to the old inode format. If the superblock version
3452 * has been updated, then make the conversion permanent.
3454 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3455 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3456 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3457 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3461 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3462 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3465 * The superblock version has already been bumped,
3466 * so just make the conversion to the new inode
3469 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3470 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3471 ip->i_d.di_onlink = 0;
3472 dip->di_core.di_onlink = 0;
3473 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3474 memset(&(dip->di_core.di_pad[0]), 0,
3475 sizeof(dip->di_core.di_pad));
3476 ASSERT(ip->i_d.di_projid == 0);
3480 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3484 if (XFS_IFORK_Q(ip)) {
3486 * The only error from xfs_iflush_fork is on the data fork.
3488 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3490 xfs_inobp_check(mp, bp);
3493 * We've recorded everything logged in the inode, so we'd
3494 * like to clear the ilf_fields bits so we don't log and
3495 * flush things unnecessarily. However, we can't stop
3496 * logging all this information until the data we've copied
3497 * into the disk buffer is written to disk. If we did we might
3498 * overwrite the copy of the inode in the log with all the
3499 * data after re-logging only part of it, and in the face of
3500 * a crash we wouldn't have all the data we need to recover.
3502 * What we do is move the bits to the ili_last_fields field.
3503 * When logging the inode, these bits are moved back to the
3504 * ilf_fields field. In the xfs_iflush_done() routine we
3505 * clear ili_last_fields, since we know that the information
3506 * those bits represent is permanently on disk. As long as
3507 * the flush completes before the inode is logged again, then
3508 * both ilf_fields and ili_last_fields will be cleared.
3510 * We can play with the ilf_fields bits here, because the inode
3511 * lock must be held exclusively in order to set bits there
3512 * and the flush lock protects the ili_last_fields bits.
3513 * Set ili_logged so the flush done
3514 * routine can tell whether or not to look in the AIL.
3515 * Also, store the current LSN of the inode so that we can tell
3516 * whether the item has moved in the AIL from xfs_iflush_done().
3517 * In order to read the lsn we need the AIL lock, because
3518 * it is a 64 bit value that cannot be read atomically.
3520 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3521 iip->ili_last_fields = iip->ili_format.ilf_fields;
3522 iip->ili_format.ilf_fields = 0;
3523 iip->ili_logged = 1;
3525 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3527 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3531 * Attach the function xfs_iflush_done to the inode's
3532 * buffer. This will remove the inode from the AIL
3533 * and unlock the inode's flush lock when the inode is
3534 * completely written to disk.
3536 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3537 xfs_iflush_done, (xfs_log_item_t *)iip);
3539 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3540 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3543 * We're flushing an inode which is not in the AIL and has
3544 * not been logged but has i_update_core set. For this
3545 * case we can use a B_DELWRI flush and immediately drop
3546 * the inode flush lock because we can avoid the whole
3547 * AIL state thing. It's OK to drop the flush lock now,
3548 * because we've already locked the buffer and to do anything
3549 * you really need both.
3552 ASSERT(iip->ili_logged == 0);
3553 ASSERT(iip->ili_last_fields == 0);
3554 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3562 return XFS_ERROR(EFSCORRUPTED);
3567 * Flush all inactive inodes in mp.
3577 XFS_MOUNT_ILOCK(mp);
3583 /* Make sure we skip markers inserted by sync */
3584 if (ip->i_mount == NULL) {
3589 vp = XFS_ITOV_NULL(ip);
3591 XFS_MOUNT_IUNLOCK(mp);
3592 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3596 ASSERT(vn_count(vp) == 0);
3599 } while (ip != mp->m_inodes);
3601 XFS_MOUNT_IUNLOCK(mp);
3605 * xfs_iaccess: check accessibility of inode for mode.
3614 mode_t orgmode = mode;
3615 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3617 if (mode & S_IWUSR) {
3618 umode_t imode = inode->i_mode;
3620 if (IS_RDONLY(inode) &&
3621 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3622 return XFS_ERROR(EROFS);
3624 if (IS_IMMUTABLE(inode))
3625 return XFS_ERROR(EACCES);
3629 * If there's an Access Control List it's used instead of
3632 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3633 return error ? XFS_ERROR(error) : 0;
3635 if (current_fsuid(cr) != ip->i_d.di_uid) {
3637 if (!in_group_p((gid_t)ip->i_d.di_gid))
3642 * If the DACs are ok we don't need any capability check.
3644 if ((ip->i_d.di_mode & mode) == mode)
3647 * Read/write DACs are always overridable.
3648 * Executable DACs are overridable if at least one exec bit is set.
3650 if (!(orgmode & S_IXUSR) ||
3651 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3652 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3655 if ((orgmode == S_IRUSR) ||
3656 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3657 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3660 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3662 return XFS_ERROR(EACCES);
3664 return XFS_ERROR(EACCES);
3668 * xfs_iroundup: round up argument to next power of two
3677 if ((v & (v - 1)) == 0)
3679 ASSERT((v & 0x80000000) == 0);
3680 if ((v & (v + 1)) == 0)
3682 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3686 if ((v & (v + 1)) == 0)
3693 #ifdef XFS_ILOCK_TRACE
3694 ktrace_t *xfs_ilock_trace_buf;
3697 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3699 ktrace_enter(ip->i_lock_trace,
3701 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3702 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3703 (void *)ra, /* caller of ilock */
3704 (void *)(unsigned long)current_cpu(),
3705 (void *)(unsigned long)current_pid(),
3706 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3711 * Return a pointer to the extent record at file index idx.
3715 xfs_ifork_t *ifp, /* inode fork pointer */
3716 xfs_extnum_t idx) /* index of target extent */
3719 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3720 return ifp->if_u1.if_ext_irec->er_extbuf;
3721 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3722 xfs_ext_irec_t *erp; /* irec pointer */
3723 int erp_idx = 0; /* irec index */
3724 xfs_extnum_t page_idx = idx; /* ext index in target list */
3726 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3727 return &erp->er_extbuf[page_idx];
3728 } else if (ifp->if_bytes) {
3729 return &ifp->if_u1.if_extents[idx];
3736 * Insert new item(s) into the extent records for incore inode
3737 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3741 xfs_ifork_t *ifp, /* inode fork pointer */
3742 xfs_extnum_t idx, /* starting index of new items */
3743 xfs_extnum_t count, /* number of inserted items */
3744 xfs_bmbt_irec_t *new) /* items to insert */
3746 xfs_bmbt_rec_t *ep; /* extent record pointer */
3747 xfs_extnum_t i; /* extent record index */
3749 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3750 xfs_iext_add(ifp, idx, count);
3751 for (i = idx; i < idx + count; i++, new++) {
3752 ep = xfs_iext_get_ext(ifp, i);
3753 xfs_bmbt_set_all(ep, new);
3758 * This is called when the amount of space required for incore file
3759 * extents needs to be increased. The ext_diff parameter stores the
3760 * number of new extents being added and the idx parameter contains
3761 * the extent index where the new extents will be added. If the new
3762 * extents are being appended, then we just need to (re)allocate and
3763 * initialize the space. Otherwise, if the new extents are being
3764 * inserted into the middle of the existing entries, a bit more work
3765 * is required to make room for the new extents to be inserted. The
3766 * caller is responsible for filling in the new extent entries upon
3771 xfs_ifork_t *ifp, /* inode fork pointer */
3772 xfs_extnum_t idx, /* index to begin adding exts */
3773 int ext_diff) /* number of extents to add */
3775 int byte_diff; /* new bytes being added */
3776 int new_size; /* size of extents after adding */
3777 xfs_extnum_t nextents; /* number of extents in file */
3779 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3780 ASSERT((idx >= 0) && (idx <= nextents));
3781 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3782 new_size = ifp->if_bytes + byte_diff;
3784 * If the new number of extents (nextents + ext_diff)
3785 * fits inside the inode, then continue to use the inline
3788 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3789 if (idx < nextents) {
3790 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3791 &ifp->if_u2.if_inline_ext[idx],
3792 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3793 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3795 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3796 ifp->if_real_bytes = 0;
3797 ifp->if_lastex = nextents + ext_diff;
3800 * Otherwise use a linear (direct) extent list.
3801 * If the extents are currently inside the inode,
3802 * xfs_iext_realloc_direct will switch us from
3803 * inline to direct extent allocation mode.
3805 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3806 xfs_iext_realloc_direct(ifp, new_size);
3807 if (idx < nextents) {
3808 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3809 &ifp->if_u1.if_extents[idx],
3810 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3811 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3814 /* Indirection array */
3816 xfs_ext_irec_t *erp;
3820 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3821 if (ifp->if_flags & XFS_IFEXTIREC) {
3822 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3824 xfs_iext_irec_init(ifp);
3825 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3826 erp = ifp->if_u1.if_ext_irec;
3828 /* Extents fit in target extent page */
3829 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3830 if (page_idx < erp->er_extcount) {
3831 memmove(&erp->er_extbuf[page_idx + ext_diff],
3832 &erp->er_extbuf[page_idx],
3833 (erp->er_extcount - page_idx) *
3834 sizeof(xfs_bmbt_rec_t));
3835 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3837 erp->er_extcount += ext_diff;
3838 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3840 /* Insert a new extent page */
3842 xfs_iext_add_indirect_multi(ifp,
3843 erp_idx, page_idx, ext_diff);
3846 * If extent(s) are being appended to the last page in
3847 * the indirection array and the new extent(s) don't fit
3848 * in the page, then erp is NULL and erp_idx is set to
3849 * the next index needed in the indirection array.
3852 int count = ext_diff;
3855 erp = xfs_iext_irec_new(ifp, erp_idx);
3856 erp->er_extcount = count;
3857 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3864 ifp->if_bytes = new_size;
3868 * This is called when incore extents are being added to the indirection
3869 * array and the new extents do not fit in the target extent list. The
3870 * erp_idx parameter contains the irec index for the target extent list
3871 * in the indirection array, and the idx parameter contains the extent
3872 * index within the list. The number of extents being added is stored
3873 * in the count parameter.
3875 * |-------| |-------|
3876 * | | | | idx - number of extents before idx
3878 * | | | | count - number of extents being inserted at idx
3879 * |-------| |-------|
3880 * | count | | nex2 | nex2 - number of extents after idx + count
3881 * |-------| |-------|
3884 xfs_iext_add_indirect_multi(
3885 xfs_ifork_t *ifp, /* inode fork pointer */
3886 int erp_idx, /* target extent irec index */
3887 xfs_extnum_t idx, /* index within target list */
3888 int count) /* new extents being added */
3890 int byte_diff; /* new bytes being added */
3891 xfs_ext_irec_t *erp; /* pointer to irec entry */
3892 xfs_extnum_t ext_diff; /* number of extents to add */
3893 xfs_extnum_t ext_cnt; /* new extents still needed */
3894 xfs_extnum_t nex2; /* extents after idx + count */
3895 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3896 int nlists; /* number of irec's (lists) */
3898 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3899 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3900 nex2 = erp->er_extcount - idx;
3901 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3904 * Save second part of target extent list
3905 * (all extents past */
3907 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3908 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3909 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3910 erp->er_extcount -= nex2;
3911 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3912 memset(&erp->er_extbuf[idx], 0, byte_diff);
3916 * Add the new extents to the end of the target
3917 * list, then allocate new irec record(s) and
3918 * extent buffer(s) as needed to store the rest
3919 * of the new extents.
3922 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3924 erp->er_extcount += ext_diff;
3925 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3926 ext_cnt -= ext_diff;
3930 erp = xfs_iext_irec_new(ifp, erp_idx);
3931 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3932 erp->er_extcount = ext_diff;
3933 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3934 ext_cnt -= ext_diff;
3937 /* Add nex2 extents back to indirection array */
3939 xfs_extnum_t ext_avail;
3942 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3943 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3946 * If nex2 extents fit in the current page, append
3947 * nex2_ep after the new extents.
3949 if (nex2 <= ext_avail) {
3950 i = erp->er_extcount;
3953 * Otherwise, check if space is available in the
3956 else if ((erp_idx < nlists - 1) &&
3957 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3958 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3961 /* Create a hole for nex2 extents */
3962 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3963 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3966 * Final choice, create a new extent page for
3971 erp = xfs_iext_irec_new(ifp, erp_idx);
3973 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3974 kmem_free(nex2_ep, byte_diff);
3975 erp->er_extcount += nex2;
3976 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3981 * This is called when the amount of space required for incore file
3982 * extents needs to be decreased. The ext_diff parameter stores the
3983 * number of extents to be removed and the idx parameter contains
3984 * the extent index where the extents will be removed from.
3986 * If the amount of space needed has decreased below the linear
3987 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3988 * extent array. Otherwise, use kmem_realloc() to adjust the
3989 * size to what is needed.
3993 xfs_ifork_t *ifp, /* inode fork pointer */
3994 xfs_extnum_t idx, /* index to begin removing exts */
3995 int ext_diff) /* number of extents to remove */
3997 xfs_extnum_t nextents; /* number of extents in file */
3998 int new_size; /* size of extents after removal */
4000 ASSERT(ext_diff > 0);
4001 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4002 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
4004 if (new_size == 0) {
4005 xfs_iext_destroy(ifp);
4006 } else if (ifp->if_flags & XFS_IFEXTIREC) {
4007 xfs_iext_remove_indirect(ifp, idx, ext_diff);
4008 } else if (ifp->if_real_bytes) {
4009 xfs_iext_remove_direct(ifp, idx, ext_diff);
4011 xfs_iext_remove_inline(ifp, idx, ext_diff);
4013 ifp->if_bytes = new_size;
4017 * This removes ext_diff extents from the inline buffer, beginning
4018 * at extent index idx.
4021 xfs_iext_remove_inline(
4022 xfs_ifork_t *ifp, /* inode fork pointer */
4023 xfs_extnum_t idx, /* index to begin removing exts */
4024 int ext_diff) /* number of extents to remove */
4026 int nextents; /* number of extents in file */
4028 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4029 ASSERT(idx < XFS_INLINE_EXTS);
4030 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4031 ASSERT(((nextents - ext_diff) > 0) &&
4032 (nextents - ext_diff) < XFS_INLINE_EXTS);
4034 if (idx + ext_diff < nextents) {
4035 memmove(&ifp->if_u2.if_inline_ext[idx],
4036 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4037 (nextents - (idx + ext_diff)) *
4038 sizeof(xfs_bmbt_rec_t));
4039 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4040 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4042 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4043 ext_diff * sizeof(xfs_bmbt_rec_t));
4048 * This removes ext_diff extents from a linear (direct) extent list,
4049 * beginning at extent index idx. If the extents are being removed
4050 * from the end of the list (ie. truncate) then we just need to re-
4051 * allocate the list to remove the extra space. Otherwise, if the
4052 * extents are being removed from the middle of the existing extent
4053 * entries, then we first need to move the extent records beginning
4054 * at idx + ext_diff up in the list to overwrite the records being
4055 * removed, then remove the extra space via kmem_realloc.
4058 xfs_iext_remove_direct(
4059 xfs_ifork_t *ifp, /* inode fork pointer */
4060 xfs_extnum_t idx, /* index to begin removing exts */
4061 int ext_diff) /* number of extents to remove */
4063 xfs_extnum_t nextents; /* number of extents in file */
4064 int new_size; /* size of extents after removal */
4066 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4067 new_size = ifp->if_bytes -
4068 (ext_diff * sizeof(xfs_bmbt_rec_t));
4069 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4071 if (new_size == 0) {
4072 xfs_iext_destroy(ifp);
4075 /* Move extents up in the list (if needed) */
4076 if (idx + ext_diff < nextents) {
4077 memmove(&ifp->if_u1.if_extents[idx],
4078 &ifp->if_u1.if_extents[idx + ext_diff],
4079 (nextents - (idx + ext_diff)) *
4080 sizeof(xfs_bmbt_rec_t));
4082 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4083 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4085 * Reallocate the direct extent list. If the extents
4086 * will fit inside the inode then xfs_iext_realloc_direct
4087 * will switch from direct to inline extent allocation
4090 xfs_iext_realloc_direct(ifp, new_size);
4091 ifp->if_bytes = new_size;
4095 * This is called when incore extents are being removed from the
4096 * indirection array and the extents being removed span multiple extent
4097 * buffers. The idx parameter contains the file extent index where we
4098 * want to begin removing extents, and the count parameter contains
4099 * how many extents need to be removed.
4101 * |-------| |-------|
4102 * | nex1 | | | nex1 - number of extents before idx
4103 * |-------| | count |
4104 * | | | | count - number of extents being removed at idx
4105 * | count | |-------|
4106 * | | | nex2 | nex2 - number of extents after idx + count
4107 * |-------| |-------|
4110 xfs_iext_remove_indirect(
4111 xfs_ifork_t *ifp, /* inode fork pointer */
4112 xfs_extnum_t idx, /* index to begin removing extents */
4113 int count) /* number of extents to remove */
4115 xfs_ext_irec_t *erp; /* indirection array pointer */
4116 int erp_idx = 0; /* indirection array index */
4117 xfs_extnum_t ext_cnt; /* extents left to remove */
4118 xfs_extnum_t ext_diff; /* extents to remove in current list */
4119 xfs_extnum_t nex1; /* number of extents before idx */
4120 xfs_extnum_t nex2; /* extents after idx + count */
4121 int nlists; /* entries in indirection array */
4122 int page_idx = idx; /* index in target extent list */
4124 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4125 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4126 ASSERT(erp != NULL);
4127 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4131 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4132 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4134 * Check for deletion of entire list;
4135 * xfs_iext_irec_remove() updates extent offsets.
4137 if (ext_diff == erp->er_extcount) {
4138 xfs_iext_irec_remove(ifp, erp_idx);
4139 ext_cnt -= ext_diff;
4142 ASSERT(erp_idx < ifp->if_real_bytes /
4144 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4151 /* Move extents up (if needed) */
4153 memmove(&erp->er_extbuf[nex1],
4154 &erp->er_extbuf[nex1 + ext_diff],
4155 nex2 * sizeof(xfs_bmbt_rec_t));
4157 /* Zero out rest of page */
4158 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4159 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4160 /* Update remaining counters */
4161 erp->er_extcount -= ext_diff;
4162 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4163 ext_cnt -= ext_diff;
4168 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4169 xfs_iext_irec_compact(ifp);
4173 * Create, destroy, or resize a linear (direct) block of extents.
4176 xfs_iext_realloc_direct(
4177 xfs_ifork_t *ifp, /* inode fork pointer */
4178 int new_size) /* new size of extents */
4180 int rnew_size; /* real new size of extents */
4182 rnew_size = new_size;
4184 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4185 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4186 (new_size != ifp->if_real_bytes)));
4188 /* Free extent records */
4189 if (new_size == 0) {
4190 xfs_iext_destroy(ifp);
4192 /* Resize direct extent list and zero any new bytes */
4193 else if (ifp->if_real_bytes) {
4194 /* Check if extents will fit inside the inode */
4195 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4196 xfs_iext_direct_to_inline(ifp, new_size /
4197 (uint)sizeof(xfs_bmbt_rec_t));
4198 ifp->if_bytes = new_size;
4201 if (!is_power_of_2(new_size)){
4202 rnew_size = xfs_iroundup(new_size);
4204 if (rnew_size != ifp->if_real_bytes) {
4205 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4206 kmem_realloc(ifp->if_u1.if_extents,
4211 if (rnew_size > ifp->if_real_bytes) {
4212 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4213 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4214 rnew_size - ifp->if_real_bytes);
4218 * Switch from the inline extent buffer to a direct
4219 * extent list. Be sure to include the inline extent
4220 * bytes in new_size.
4223 new_size += ifp->if_bytes;
4224 if (!is_power_of_2(new_size)) {
4225 rnew_size = xfs_iroundup(new_size);
4227 xfs_iext_inline_to_direct(ifp, rnew_size);
4229 ifp->if_real_bytes = rnew_size;
4230 ifp->if_bytes = new_size;
4234 * Switch from linear (direct) extent records to inline buffer.
4237 xfs_iext_direct_to_inline(
4238 xfs_ifork_t *ifp, /* inode fork pointer */
4239 xfs_extnum_t nextents) /* number of extents in file */
4241 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4242 ASSERT(nextents <= XFS_INLINE_EXTS);
4244 * The inline buffer was zeroed when we switched
4245 * from inline to direct extent allocation mode,
4246 * so we don't need to clear it here.
4248 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4249 nextents * sizeof(xfs_bmbt_rec_t));
4250 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4251 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4252 ifp->if_real_bytes = 0;
4256 * Switch from inline buffer to linear (direct) extent records.
4257 * new_size should already be rounded up to the next power of 2
4258 * by the caller (when appropriate), so use new_size as it is.
4259 * However, since new_size may be rounded up, we can't update
4260 * if_bytes here. It is the caller's responsibility to update
4261 * if_bytes upon return.
4264 xfs_iext_inline_to_direct(
4265 xfs_ifork_t *ifp, /* inode fork pointer */
4266 int new_size) /* number of extents in file */
4268 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4269 kmem_alloc(new_size, KM_SLEEP);
4270 memset(ifp->if_u1.if_extents, 0, new_size);
4271 if (ifp->if_bytes) {
4272 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4274 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4275 sizeof(xfs_bmbt_rec_t));
4277 ifp->if_real_bytes = new_size;
4281 * Resize an extent indirection array to new_size bytes.
4284 xfs_iext_realloc_indirect(
4285 xfs_ifork_t *ifp, /* inode fork pointer */
4286 int new_size) /* new indirection array size */
4288 int nlists; /* number of irec's (ex lists) */
4289 int size; /* current indirection array size */
4291 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4292 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4293 size = nlists * sizeof(xfs_ext_irec_t);
4294 ASSERT(ifp->if_real_bytes);
4295 ASSERT((new_size >= 0) && (new_size != size));
4296 if (new_size == 0) {
4297 xfs_iext_destroy(ifp);
4299 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4300 kmem_realloc(ifp->if_u1.if_ext_irec,
4301 new_size, size, KM_SLEEP);
4306 * Switch from indirection array to linear (direct) extent allocations.
4309 xfs_iext_indirect_to_direct(
4310 xfs_ifork_t *ifp) /* inode fork pointer */
4312 xfs_bmbt_rec_t *ep; /* extent record pointer */
4313 xfs_extnum_t nextents; /* number of extents in file */
4314 int size; /* size of file extents */
4316 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4317 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4318 ASSERT(nextents <= XFS_LINEAR_EXTS);
4319 size = nextents * sizeof(xfs_bmbt_rec_t);
4321 xfs_iext_irec_compact_full(ifp);
4322 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4324 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4325 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4326 ifp->if_flags &= ~XFS_IFEXTIREC;
4327 ifp->if_u1.if_extents = ep;
4328 ifp->if_bytes = size;
4329 if (nextents < XFS_LINEAR_EXTS) {
4330 xfs_iext_realloc_direct(ifp, size);
4335 * Free incore file extents.
4339 xfs_ifork_t *ifp) /* inode fork pointer */
4341 if (ifp->if_flags & XFS_IFEXTIREC) {
4345 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4346 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4347 xfs_iext_irec_remove(ifp, erp_idx);
4349 ifp->if_flags &= ~XFS_IFEXTIREC;
4350 } else if (ifp->if_real_bytes) {
4351 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4352 } else if (ifp->if_bytes) {
4353 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4354 sizeof(xfs_bmbt_rec_t));
4356 ifp->if_u1.if_extents = NULL;
4357 ifp->if_real_bytes = 0;
4362 * Return a pointer to the extent record for file system block bno.
4364 xfs_bmbt_rec_t * /* pointer to found extent record */
4365 xfs_iext_bno_to_ext(
4366 xfs_ifork_t *ifp, /* inode fork pointer */
4367 xfs_fileoff_t bno, /* block number to search for */
4368 xfs_extnum_t *idxp) /* index of target extent */
4370 xfs_bmbt_rec_t *base; /* pointer to first extent */
4371 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4372 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4373 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4374 int high; /* upper boundary in search */
4375 xfs_extnum_t idx = 0; /* index of target extent */
4376 int low; /* lower boundary in search */
4377 xfs_extnum_t nextents; /* number of file extents */
4378 xfs_fileoff_t startoff = 0; /* start offset of extent */
4380 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4381 if (nextents == 0) {
4386 if (ifp->if_flags & XFS_IFEXTIREC) {
4387 /* Find target extent list */
4389 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4390 base = erp->er_extbuf;
4391 high = erp->er_extcount - 1;
4393 base = ifp->if_u1.if_extents;
4394 high = nextents - 1;
4396 /* Binary search extent records */
4397 while (low <= high) {
4398 idx = (low + high) >> 1;
4400 startoff = xfs_bmbt_get_startoff(ep);
4401 blockcount = xfs_bmbt_get_blockcount(ep);
4402 if (bno < startoff) {
4404 } else if (bno >= startoff + blockcount) {
4407 /* Convert back to file-based extent index */
4408 if (ifp->if_flags & XFS_IFEXTIREC) {
4409 idx += erp->er_extoff;
4415 /* Convert back to file-based extent index */
4416 if (ifp->if_flags & XFS_IFEXTIREC) {
4417 idx += erp->er_extoff;
4419 if (bno >= startoff + blockcount) {
4420 if (++idx == nextents) {
4423 ep = xfs_iext_get_ext(ifp, idx);
4431 * Return a pointer to the indirection array entry containing the
4432 * extent record for filesystem block bno. Store the index of the
4433 * target irec in *erp_idxp.
4435 xfs_ext_irec_t * /* pointer to found extent record */
4436 xfs_iext_bno_to_irec(
4437 xfs_ifork_t *ifp, /* inode fork pointer */
4438 xfs_fileoff_t bno, /* block number to search for */
4439 int *erp_idxp) /* irec index of target ext list */
4441 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4442 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4443 int erp_idx; /* indirection array index */
4444 int nlists; /* number of extent irec's (lists) */
4445 int high; /* binary search upper limit */
4446 int low; /* binary search lower limit */
4448 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4449 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4453 while (low <= high) {
4454 erp_idx = (low + high) >> 1;
4455 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4456 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4457 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4459 } else if (erp_next && bno >=
4460 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4466 *erp_idxp = erp_idx;
4471 * Return a pointer to the indirection array entry containing the
4472 * extent record at file extent index *idxp. Store the index of the
4473 * target irec in *erp_idxp and store the page index of the target
4474 * extent record in *idxp.
4477 xfs_iext_idx_to_irec(
4478 xfs_ifork_t *ifp, /* inode fork pointer */
4479 xfs_extnum_t *idxp, /* extent index (file -> page) */
4480 int *erp_idxp, /* pointer to target irec */
4481 int realloc) /* new bytes were just added */
4483 xfs_ext_irec_t *prev; /* pointer to previous irec */
4484 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4485 int erp_idx; /* indirection array index */
4486 int nlists; /* number of irec's (ex lists) */
4487 int high; /* binary search upper limit */
4488 int low; /* binary search lower limit */
4489 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4491 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4492 ASSERT(page_idx >= 0 && page_idx <=
4493 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4494 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4499 /* Binary search extent irec's */
4500 while (low <= high) {
4501 erp_idx = (low + high) >> 1;
4502 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4503 prev = erp_idx > 0 ? erp - 1 : NULL;
4504 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4505 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4507 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4508 (page_idx == erp->er_extoff + erp->er_extcount &&
4511 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4512 erp->er_extcount == XFS_LINEAR_EXTS) {
4516 erp = erp_idx < nlists ? erp + 1 : NULL;
4519 page_idx -= erp->er_extoff;
4524 *erp_idxp = erp_idx;
4529 * Allocate and initialize an indirection array once the space needed
4530 * for incore extents increases above XFS_IEXT_BUFSZ.
4534 xfs_ifork_t *ifp) /* inode fork pointer */
4536 xfs_ext_irec_t *erp; /* indirection array pointer */
4537 xfs_extnum_t nextents; /* number of extents in file */
4539 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4540 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4541 ASSERT(nextents <= XFS_LINEAR_EXTS);
4543 erp = (xfs_ext_irec_t *)
4544 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4546 if (nextents == 0) {
4547 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4548 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4549 } else if (!ifp->if_real_bytes) {
4550 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4551 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4552 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4554 erp->er_extbuf = ifp->if_u1.if_extents;
4555 erp->er_extcount = nextents;
4558 ifp->if_flags |= XFS_IFEXTIREC;
4559 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4560 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4561 ifp->if_u1.if_ext_irec = erp;
4567 * Allocate and initialize a new entry in the indirection array.
4571 xfs_ifork_t *ifp, /* inode fork pointer */
4572 int erp_idx) /* index for new irec */
4574 xfs_ext_irec_t *erp; /* indirection array pointer */
4575 int i; /* loop counter */
4576 int nlists; /* number of irec's (ex lists) */
4578 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4579 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4581 /* Resize indirection array */
4582 xfs_iext_realloc_indirect(ifp, ++nlists *
4583 sizeof(xfs_ext_irec_t));
4585 * Move records down in the array so the
4586 * new page can use erp_idx.
4588 erp = ifp->if_u1.if_ext_irec;
4589 for (i = nlists - 1; i > erp_idx; i--) {
4590 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4592 ASSERT(i == erp_idx);
4594 /* Initialize new extent record */
4595 erp = ifp->if_u1.if_ext_irec;
4596 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4597 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4598 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4599 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4600 erp[erp_idx].er_extcount = 0;
4601 erp[erp_idx].er_extoff = erp_idx > 0 ?
4602 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4603 return (&erp[erp_idx]);
4607 * Remove a record from the indirection array.
4610 xfs_iext_irec_remove(
4611 xfs_ifork_t *ifp, /* inode fork pointer */
4612 int erp_idx) /* irec index to remove */
4614 xfs_ext_irec_t *erp; /* indirection array pointer */
4615 int i; /* loop counter */
4616 int nlists; /* number of irec's (ex lists) */
4618 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4619 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4620 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4621 if (erp->er_extbuf) {
4622 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4624 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4626 /* Compact extent records */
4627 erp = ifp->if_u1.if_ext_irec;
4628 for (i = erp_idx; i < nlists - 1; i++) {
4629 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4632 * Manually free the last extent record from the indirection
4633 * array. A call to xfs_iext_realloc_indirect() with a size
4634 * of zero would result in a call to xfs_iext_destroy() which
4635 * would in turn call this function again, creating a nasty
4639 xfs_iext_realloc_indirect(ifp,
4640 nlists * sizeof(xfs_ext_irec_t));
4642 kmem_free(ifp->if_u1.if_ext_irec,
4643 sizeof(xfs_ext_irec_t));
4645 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4649 * This is called to clean up large amounts of unused memory allocated
4650 * by the indirection array. Before compacting anything though, verify
4651 * that the indirection array is still needed and switch back to the
4652 * linear extent list (or even the inline buffer) if possible. The
4653 * compaction policy is as follows:
4655 * Full Compaction: Extents fit into a single page (or inline buffer)
4656 * Full Compaction: Extents occupy less than 10% of allocated space
4657 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4658 * No Compaction: Extents occupy at least 50% of allocated space
4661 xfs_iext_irec_compact(
4662 xfs_ifork_t *ifp) /* inode fork pointer */
4664 xfs_extnum_t nextents; /* number of extents in file */
4665 int nlists; /* number of irec's (ex lists) */
4667 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4668 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4669 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4671 if (nextents == 0) {
4672 xfs_iext_destroy(ifp);
4673 } else if (nextents <= XFS_INLINE_EXTS) {
4674 xfs_iext_indirect_to_direct(ifp);
4675 xfs_iext_direct_to_inline(ifp, nextents);
4676 } else if (nextents <= XFS_LINEAR_EXTS) {
4677 xfs_iext_indirect_to_direct(ifp);
4678 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4679 xfs_iext_irec_compact_full(ifp);
4680 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4681 xfs_iext_irec_compact_pages(ifp);
4686 * Combine extents from neighboring extent pages.
4689 xfs_iext_irec_compact_pages(
4690 xfs_ifork_t *ifp) /* inode fork pointer */
4692 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4693 int erp_idx = 0; /* indirection array index */
4694 int nlists; /* number of irec's (ex lists) */
4696 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4697 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4698 while (erp_idx < nlists - 1) {
4699 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4701 if (erp_next->er_extcount <=
4702 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4703 memmove(&erp->er_extbuf[erp->er_extcount],
4704 erp_next->er_extbuf, erp_next->er_extcount *
4705 sizeof(xfs_bmbt_rec_t));
4706 erp->er_extcount += erp_next->er_extcount;
4708 * Free page before removing extent record
4709 * so er_extoffs don't get modified in
4710 * xfs_iext_irec_remove.
4712 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4713 erp_next->er_extbuf = NULL;
4714 xfs_iext_irec_remove(ifp, erp_idx + 1);
4715 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4723 * Fully compact the extent records managed by the indirection array.
4726 xfs_iext_irec_compact_full(
4727 xfs_ifork_t *ifp) /* inode fork pointer */
4729 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4730 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4731 int erp_idx = 0; /* extent irec index */
4732 int ext_avail; /* empty entries in ex list */
4733 int ext_diff; /* number of exts to add */
4734 int nlists; /* number of irec's (ex lists) */
4736 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4737 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4738 erp = ifp->if_u1.if_ext_irec;
4739 ep = &erp->er_extbuf[erp->er_extcount];
4741 ep_next = erp_next->er_extbuf;
4742 while (erp_idx < nlists - 1) {
4743 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4744 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4745 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4746 erp->er_extcount += ext_diff;
4747 erp_next->er_extcount -= ext_diff;
4748 /* Remove next page */
4749 if (erp_next->er_extcount == 0) {
4751 * Free page before removing extent record
4752 * so er_extoffs don't get modified in
4753 * xfs_iext_irec_remove.
4755 kmem_free(erp_next->er_extbuf,
4756 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4757 erp_next->er_extbuf = NULL;
4758 xfs_iext_irec_remove(ifp, erp_idx + 1);
4759 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4760 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4761 /* Update next page */
4763 /* Move rest of page up to become next new page */
4764 memmove(erp_next->er_extbuf, ep_next,
4765 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4766 ep_next = erp_next->er_extbuf;
4767 memset(&ep_next[erp_next->er_extcount], 0,
4768 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4769 sizeof(xfs_bmbt_rec_t));
4771 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4773 if (erp_idx < nlists)
4774 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4778 ep = &erp->er_extbuf[erp->er_extcount];
4780 ep_next = erp_next->er_extbuf;
4785 * This is called to update the er_extoff field in the indirection
4786 * array when extents have been added or removed from one of the
4787 * extent lists. erp_idx contains the irec index to begin updating
4788 * at and ext_diff contains the number of extents that were added
4792 xfs_iext_irec_update_extoffs(
4793 xfs_ifork_t *ifp, /* inode fork pointer */
4794 int erp_idx, /* irec index to update */
4795 int ext_diff) /* number of new extents */
4797 int i; /* loop counter */
4798 int nlists; /* number of irec's (ex lists */
4800 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4801 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4802 for (i = erp_idx; i < nlists; i++) {
4803 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;