2 * Copyright (c) 2000-2003,2005 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"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
58 kmem_zone_t *xfs_chashlist_zone;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
74 * Make sure that the extents in the given memory buffer
89 for (i = 0; i < nrecs; i++) {
90 ep = xfs_iext_get_ext(ifp, i);
91 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
92 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
94 xfs_bmbt_disk_get_all(&rec, &irec);
96 xfs_bmbt_get_all(&rec, &irec);
97 if (fmt == XFS_EXTFMT_NOSTATE)
98 ASSERT(irec.br_state == XFS_EXT_NORM);
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
121 for (i = 0; i < j; i++) {
122 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
123 i * mp->m_sb.sb_inodesize);
124 if (!dip->di_next_unlinked) {
125 xfs_fs_cmn_err(CE_ALERT, mp,
126 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
128 ASSERT(dip->di_next_unlinked);
135 * This routine is called to map an inode number within a file
136 * system to the buffer containing the on-disk version of the
137 * inode. It returns a pointer to the buffer containing the
138 * on-disk inode in the bpp parameter, and in the dip parameter
139 * it returns a pointer to the on-disk inode within that buffer.
141 * If a non-zero error is returned, then the contents of bpp and
142 * dipp are undefined.
144 * Use xfs_imap() to determine the size and location of the
145 * buffer to read from disk.
163 * Call the space managment code to find the location of the
167 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
170 "xfs_inotobp: xfs_imap() returned an "
171 "error %d on %s. Returning error.", error, mp->m_fsname);
176 * If the inode number maps to a block outside the bounds of the
177 * file system then return NULL rather than calling read_buf
178 * and panicing when we get an error from the driver.
180 if ((imap.im_blkno + imap.im_len) >
181 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
183 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
184 "of the file system %s. Returning EINVAL.",
185 (unsigned long long)imap.im_blkno,
186 imap.im_len, mp->m_fsname);
187 return XFS_ERROR(EINVAL);
191 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
192 * default to just a read_buf() call.
194 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
195 (int)imap.im_len, XFS_BUF_LOCK, &bp);
199 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 "error %d on %s. Returning error.", error, mp->m_fsname);
203 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
205 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
206 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
207 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
208 XFS_RANDOM_ITOBP_INOTOBP))) {
209 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
210 xfs_trans_brelse(tp, bp);
212 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
214 return XFS_ERROR(EFSCORRUPTED);
217 xfs_inobp_check(mp, bp);
220 * Set *dipp to point to the on-disk inode in the buffer.
222 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
224 *offset = imap.im_boffset;
230 * This routine is called to map an inode to the buffer containing
231 * the on-disk version of the inode. It returns a pointer to the
232 * buffer containing the on-disk inode in the bpp parameter, and in
233 * the dip parameter it returns a pointer to the on-disk inode within
236 * If a non-zero error is returned, then the contents of bpp and
237 * dipp are undefined.
239 * If the inode is new and has not yet been initialized, use xfs_imap()
240 * to determine the size and location of the buffer to read from disk.
241 * If the inode has already been mapped to its buffer and read in once,
242 * then use the mapping information stored in the inode rather than
243 * calling xfs_imap(). This allows us to avoid the overhead of looking
244 * at the inode btree for small block file systems (see xfs_dilocate()).
245 * We can tell whether the inode has been mapped in before by comparing
246 * its disk block address to 0. Only uninitialized inodes will have
247 * 0 for the disk block address.
266 if (ip->i_blkno == (xfs_daddr_t)0) {
268 * Call the space management code to find the location of the
272 error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP);
278 * If the inode number maps to a block outside the bounds
279 * of the file system then return NULL rather than calling
280 * read_buf and panicing when we get an error from the
283 if ((imap.im_blkno + imap.im_len) >
284 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
286 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
287 "(imap.im_blkno (0x%llx) "
288 "+ imap.im_len (0x%llx)) > "
289 " XFS_FSB_TO_BB(mp, "
290 "mp->m_sb.sb_dblocks) (0x%llx)",
291 (unsigned long long) imap.im_blkno,
292 (unsigned long long) imap.im_len,
293 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
295 return XFS_ERROR(EINVAL);
299 * Fill in the fields in the inode that will be used to
300 * map the inode to its buffer from now on.
302 ip->i_blkno = imap.im_blkno;
303 ip->i_len = imap.im_len;
304 ip->i_boffset = imap.im_boffset;
307 * We've already mapped the inode once, so just use the
308 * mapping that we saved the first time.
310 imap.im_blkno = ip->i_blkno;
311 imap.im_len = ip->i_len;
312 imap.im_boffset = ip->i_boffset;
314 ASSERT(bno == 0 || bno == imap.im_blkno);
317 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
318 * default to just a read_buf() call.
320 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
321 (int)imap.im_len, XFS_BUF_LOCK, &bp);
325 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
326 "xfs_trans_read_buf() returned error %d, "
327 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
328 error, (unsigned long long) imap.im_blkno,
329 (unsigned long long) imap.im_len);
335 * Validate the magic number and version of every inode in the buffer
336 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
339 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
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, XFS_ERRTAG_ITOBP_INOTOBP,
352 XFS_RANDOM_ITOBP_INOTOBP))) {
354 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
356 (unsigned long long)imap.im_blkno, i,
357 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
359 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
361 xfs_trans_brelse(tp, bp);
362 return XFS_ERROR(EFSCORRUPTED);
365 #endif /* __KERNEL__ */
367 xfs_inobp_check(mp, bp);
370 * Mark the buffer as an inode buffer now that it looks good
372 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
375 * Set *dipp to point to the on-disk inode in the buffer.
377 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
383 * Move inode type and inode format specific information from the
384 * on-disk inode to the in-core inode. For fifos, devs, and sockets
385 * this means set if_rdev to the proper value. For files, directories,
386 * and symlinks this means to bring in the in-line data or extent
387 * pointers. For a file in B-tree format, only the root is immediately
388 * brought in-core. The rest will be in-lined in if_extents when it
389 * is first referenced (see xfs_iread_extents()).
396 xfs_attr_shortform_t *atp;
400 ip->i_df.if_ext_max =
401 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
405 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
406 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
407 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
408 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
409 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
410 (unsigned long long)ip->i_ino,
411 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
412 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
414 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
415 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
417 return XFS_ERROR(EFSCORRUPTED);
420 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
421 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
422 "corrupt dinode %Lu, forkoff = 0x%x.",
423 (unsigned long long)ip->i_ino,
424 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
425 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
427 return XFS_ERROR(EFSCORRUPTED);
430 switch (ip->i_d.di_mode & S_IFMT) {
435 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
436 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
438 return XFS_ERROR(EFSCORRUPTED);
441 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
447 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
448 case XFS_DINODE_FMT_LOCAL:
450 * no local regular files yet
452 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
453 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
455 "(local format for regular file).",
456 (unsigned long long) ip->i_ino);
457 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
460 return XFS_ERROR(EFSCORRUPTED);
463 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
464 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
465 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
467 "(bad size %Ld for local inode).",
468 (unsigned long long) ip->i_ino,
469 (long long) di_size);
470 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
473 return XFS_ERROR(EFSCORRUPTED);
477 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
479 case XFS_DINODE_FMT_EXTENTS:
480 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
482 case XFS_DINODE_FMT_BTREE:
483 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
486 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
488 return XFS_ERROR(EFSCORRUPTED);
493 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
494 return XFS_ERROR(EFSCORRUPTED);
499 if (!XFS_DFORK_Q(dip))
501 ASSERT(ip->i_afp == NULL);
502 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
503 ip->i_afp->if_ext_max =
504 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
505 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
506 case XFS_DINODE_FMT_LOCAL:
507 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
508 size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT);
509 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
511 case XFS_DINODE_FMT_EXTENTS:
512 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
514 case XFS_DINODE_FMT_BTREE:
515 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
518 error = XFS_ERROR(EFSCORRUPTED);
522 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
524 xfs_idestroy_fork(ip, XFS_DATA_FORK);
530 * The file is in-lined in the on-disk inode.
531 * If it fits into if_inline_data, then copy
532 * it there, otherwise allocate a buffer for it
533 * and copy the data there. Either way, set
534 * if_data to point at the data.
535 * If we allocate a buffer for the data, make
536 * sure that its size is a multiple of 4 and
537 * record the real size in i_real_bytes.
550 * If the size is unreasonable, then something
551 * is wrong and we just bail out rather than crash in
552 * kmem_alloc() or memcpy() below.
554 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
555 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
557 "(bad size %d for local fork, size = %d).",
558 (unsigned long long) ip->i_ino, size,
559 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
560 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
562 return XFS_ERROR(EFSCORRUPTED);
564 ifp = XFS_IFORK_PTR(ip, whichfork);
567 ifp->if_u1.if_data = NULL;
568 else if (size <= sizeof(ifp->if_u2.if_inline_data))
569 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
571 real_size = roundup(size, 4);
572 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
574 ifp->if_bytes = size;
575 ifp->if_real_bytes = real_size;
577 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
578 ifp->if_flags &= ~XFS_IFEXTENTS;
579 ifp->if_flags |= XFS_IFINLINE;
584 * The file consists of a set of extents all
585 * of which fit into the on-disk inode.
586 * If there are few enough extents to fit into
587 * the if_inline_ext, then copy them there.
588 * Otherwise allocate a buffer for them and copy
589 * them into it. Either way, set if_extents
590 * to point at the extents.
598 xfs_bmbt_rec_t *ep, *dp;
604 ifp = XFS_IFORK_PTR(ip, whichfork);
605 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
606 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
609 * If the number of extents is unreasonable, then something
610 * is wrong and we just bail out rather than crash in
611 * kmem_alloc() or memcpy() below.
613 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
614 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
615 "corrupt inode %Lu ((a)extents = %d).",
616 (unsigned long long) ip->i_ino, nex);
617 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
619 return XFS_ERROR(EFSCORRUPTED);
622 ifp->if_real_bytes = 0;
624 ifp->if_u1.if_extents = NULL;
625 else if (nex <= XFS_INLINE_EXTS)
626 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
628 xfs_iext_add(ifp, 0, nex);
630 ifp->if_bytes = size;
632 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
633 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
634 for (i = 0; i < nex; i++, dp++) {
635 ep = xfs_iext_get_ext(ifp, i);
636 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
638 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
641 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
643 if (whichfork != XFS_DATA_FORK ||
644 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
645 if (unlikely(xfs_check_nostate_extents(
647 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
650 return XFS_ERROR(EFSCORRUPTED);
653 ifp->if_flags |= XFS_IFEXTENTS;
658 * The file has too many extents to fit into
659 * the inode, so they are in B-tree format.
660 * Allocate a buffer for the root of the B-tree
661 * and copy the root into it. The i_extents
662 * field will remain NULL until all of the
663 * extents are read in (when they are needed).
671 xfs_bmdr_block_t *dfp;
677 ifp = XFS_IFORK_PTR(ip, whichfork);
678 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
679 size = XFS_BMAP_BROOT_SPACE(dfp);
680 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
683 * blow out if -- fork has less extents than can fit in
684 * fork (fork shouldn't be a btree format), root btree
685 * block has more records than can fit into the fork,
686 * or the number of extents is greater than the number of
689 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
690 || XFS_BMDR_SPACE_CALC(nrecs) >
691 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
692 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
693 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
694 "corrupt inode %Lu (btree).",
695 (unsigned long long) ip->i_ino);
696 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
698 return XFS_ERROR(EFSCORRUPTED);
701 ifp->if_broot_bytes = size;
702 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
703 ASSERT(ifp->if_broot != NULL);
705 * Copy and convert from the on-disk structure
706 * to the in-memory structure.
708 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
709 ifp->if_broot, size);
710 ifp->if_flags &= ~XFS_IFEXTENTS;
711 ifp->if_flags |= XFS_IFBROOT;
717 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
720 * buf = on-disk representation
721 * dip = native representation
722 * dir = direction - +ve -> disk to native
723 * -ve -> native to disk
726 xfs_xlate_dinode_core(
728 xfs_dinode_core_t *dip,
731 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
732 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
733 xfs_arch_t arch = ARCH_CONVERT;
737 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
738 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
739 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
740 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
741 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
742 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
743 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
744 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
745 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
748 memcpy(mem_core->di_pad, buf_core->di_pad,
749 sizeof(buf_core->di_pad));
751 memcpy(buf_core->di_pad, mem_core->di_pad,
752 sizeof(buf_core->di_pad));
755 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
757 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
759 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
761 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
763 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
765 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
767 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
769 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
770 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
771 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
772 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
773 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
774 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
775 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
776 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
777 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
778 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
779 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
784 xfs_dinode_core_t *dic,
789 if (di_flags & XFS_DIFLAG_ANY) {
790 if (di_flags & XFS_DIFLAG_REALTIME)
791 flags |= XFS_XFLAG_REALTIME;
792 if (di_flags & XFS_DIFLAG_PREALLOC)
793 flags |= XFS_XFLAG_PREALLOC;
794 if (di_flags & XFS_DIFLAG_IMMUTABLE)
795 flags |= XFS_XFLAG_IMMUTABLE;
796 if (di_flags & XFS_DIFLAG_APPEND)
797 flags |= XFS_XFLAG_APPEND;
798 if (di_flags & XFS_DIFLAG_SYNC)
799 flags |= XFS_XFLAG_SYNC;
800 if (di_flags & XFS_DIFLAG_NOATIME)
801 flags |= XFS_XFLAG_NOATIME;
802 if (di_flags & XFS_DIFLAG_NODUMP)
803 flags |= XFS_XFLAG_NODUMP;
804 if (di_flags & XFS_DIFLAG_RTINHERIT)
805 flags |= XFS_XFLAG_RTINHERIT;
806 if (di_flags & XFS_DIFLAG_PROJINHERIT)
807 flags |= XFS_XFLAG_PROJINHERIT;
808 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
809 flags |= XFS_XFLAG_NOSYMLINKS;
810 if (di_flags & XFS_DIFLAG_EXTSIZE)
811 flags |= XFS_XFLAG_EXTSIZE;
812 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
813 flags |= XFS_XFLAG_EXTSZINHERIT;
823 xfs_dinode_core_t *dic = &ip->i_d;
825 return _xfs_dic2xflags(dic, dic->di_flags) |
826 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
831 xfs_dinode_core_t *dic)
833 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
834 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
838 * Given a mount structure and an inode number, return a pointer
839 * to a newly allocated in-core inode coresponding to the given
842 * Initialize the inode's attributes and extent pointers if it
843 * already has them (it will not if the inode has no links).
858 ASSERT(xfs_inode_zone != NULL);
860 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
865 * Get pointer's to the on-disk inode and the buffer containing it.
866 * If the inode number refers to a block outside the file system
867 * then xfs_itobp() will return NULL. In this case we should
868 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
869 * know that this is a new incore inode.
871 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno);
874 kmem_zone_free(xfs_inode_zone, ip);
879 * Initialize inode's trace buffers.
880 * Do this before xfs_iformat in case it adds entries.
882 #ifdef XFS_BMAP_TRACE
883 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
885 #ifdef XFS_BMBT_TRACE
886 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
889 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
891 #ifdef XFS_ILOCK_TRACE
892 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
894 #ifdef XFS_DIR2_TRACE
895 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
899 * If we got something that isn't an inode it means someone
900 * (nfs or dmi) has a stale handle.
902 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
903 kmem_zone_free(xfs_inode_zone, ip);
904 xfs_trans_brelse(tp, bp);
906 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
907 "dip->di_core.di_magic (0x%x) != "
908 "XFS_DINODE_MAGIC (0x%x)",
909 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
912 return XFS_ERROR(EINVAL);
916 * If the on-disk inode is already linked to a directory
917 * entry, copy all of the inode into the in-core inode.
918 * xfs_iformat() handles copying in the inode format
919 * specific information.
920 * Otherwise, just get the truly permanent information.
922 if (dip->di_core.di_mode) {
923 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
925 error = xfs_iformat(ip, dip);
927 kmem_zone_free(xfs_inode_zone, ip);
928 xfs_trans_brelse(tp, bp);
930 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
931 "xfs_iformat() returned error %d",
937 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
938 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
939 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
940 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
953 ip->i_df.if_ext_max =
954 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
957 INIT_LIST_HEAD(&ip->i_reclaim);
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
970 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
971 ip->i_d.di_nlink = ip->i_d.di_onlink;
972 ip->i_d.di_onlink = 0;
973 ip->i_d.di_projid = 0;
976 ip->i_delayed_blks = 0;
979 * Mark the buffer containing the inode as something to keep
980 * around for a while. This helps to keep recently accessed
981 * meta-data in-core longer.
983 XFS_BUF_SET_REF(bp, XFS_INO_REF);
986 * Use xfs_trans_brelse() to release the buffer containing the
987 * on-disk inode, because it was acquired with xfs_trans_read_buf()
988 * in xfs_itobp() above. If tp is NULL, this is just a normal
989 * brelse(). If we're within a transaction, then xfs_trans_brelse()
990 * will only release the buffer if it is not dirty within the
991 * transaction. It will be OK to release the buffer in this case,
992 * because inodes on disk are never destroyed and we will be
993 * locking the new in-core inode before putting it in the hash
994 * table where other processes can find it. Thus we don't have
995 * to worry about the inode being changed just because we released
998 xfs_trans_brelse(tp, bp);
1004 * Read in extents from a btree-format inode.
1005 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1015 xfs_extnum_t nextents;
1018 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1019 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1021 return XFS_ERROR(EFSCORRUPTED);
1023 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1024 size = nextents * sizeof(xfs_bmbt_rec_t);
1025 ifp = XFS_IFORK_PTR(ip, whichfork);
1028 * We know that the size is valid (it's checked in iformat_btree)
1030 ifp->if_lastex = NULLEXTNUM;
1031 ifp->if_bytes = ifp->if_real_bytes = 0;
1032 ifp->if_flags |= XFS_IFEXTENTS;
1033 xfs_iext_add(ifp, 0, nextents);
1034 error = xfs_bmap_read_extents(tp, ip, whichfork);
1036 xfs_iext_destroy(ifp);
1037 ifp->if_flags &= ~XFS_IFEXTENTS;
1040 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1045 * Allocate an inode on disk and return a copy of its in-core version.
1046 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1047 * appropriately within the inode. The uid and gid for the inode are
1048 * set according to the contents of the given cred structure.
1050 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1051 * has a free inode available, call xfs_iget()
1052 * to obtain the in-core version of the allocated inode. Finally,
1053 * fill in the inode and log its initial contents. In this case,
1054 * ialloc_context would be set to NULL and call_again set to false.
1056 * If xfs_dialloc() does not have an available inode,
1057 * it will replenish its supply by doing an allocation. Since we can
1058 * only do one allocation within a transaction without deadlocks, we
1059 * must commit the current transaction before returning the inode itself.
1060 * In this case, therefore, we will set call_again to true and return.
1061 * The caller should then commit the current transaction, start a new
1062 * transaction, and call xfs_ialloc() again to actually get the inode.
1064 * To ensure that some other process does not grab the inode that
1065 * was allocated during the first call to xfs_ialloc(), this routine
1066 * also returns the [locked] bp pointing to the head of the freelist
1067 * as ialloc_context. The caller should hold this buffer across
1068 * the commit and pass it back into this routine on the second call.
1080 xfs_buf_t **ialloc_context,
1081 boolean_t *call_again,
1091 * Call the space management code to pick
1092 * the on-disk inode to be allocated.
1094 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1095 ialloc_context, call_again, &ino);
1099 if (*call_again || ino == NULLFSINO) {
1103 ASSERT(*ialloc_context == NULL);
1106 * Get the in-core inode with the lock held exclusively.
1107 * This is because we're setting fields here we need
1108 * to prevent others from looking at until we're done.
1110 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1111 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1118 ip->i_d.di_mode = (__uint16_t)mode;
1119 ip->i_d.di_onlink = 0;
1120 ip->i_d.di_nlink = nlink;
1121 ASSERT(ip->i_d.di_nlink == nlink);
1122 ip->i_d.di_uid = current_fsuid(cr);
1123 ip->i_d.di_gid = current_fsgid(cr);
1124 ip->i_d.di_projid = prid;
1125 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1128 * If the superblock version is up to where we support new format
1129 * inodes and this is currently an old format inode, then change
1130 * the inode version number now. This way we only do the conversion
1131 * here rather than here and in the flush/logging code.
1133 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1134 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1135 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1137 * We've already zeroed the old link count, the projid field,
1138 * and the pad field.
1143 * Project ids won't be stored on disk if we are using a version 1 inode.
1145 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1146 xfs_bump_ino_vers2(tp, ip);
1148 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1149 ip->i_d.di_gid = pip->i_d.di_gid;
1150 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1151 ip->i_d.di_mode |= S_ISGID;
1156 * If the group ID of the new file does not match the effective group
1157 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1158 * (and only if the irix_sgid_inherit compatibility variable is set).
1160 if ((irix_sgid_inherit) &&
1161 (ip->i_d.di_mode & S_ISGID) &&
1162 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1163 ip->i_d.di_mode &= ~S_ISGID;
1166 ip->i_d.di_size = 0;
1167 ip->i_d.di_nextents = 0;
1168 ASSERT(ip->i_d.di_nblocks == 0);
1169 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1171 * di_gen will have been taken care of in xfs_iread.
1173 ip->i_d.di_extsize = 0;
1174 ip->i_d.di_dmevmask = 0;
1175 ip->i_d.di_dmstate = 0;
1176 ip->i_d.di_flags = 0;
1177 flags = XFS_ILOG_CORE;
1178 switch (mode & S_IFMT) {
1183 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1184 ip->i_df.if_u2.if_rdev = rdev;
1185 ip->i_df.if_flags = 0;
1186 flags |= XFS_ILOG_DEV;
1190 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1193 if ((mode & S_IFMT) == S_IFDIR) {
1194 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1195 di_flags |= XFS_DIFLAG_RTINHERIT;
1196 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1197 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1198 ip->i_d.di_extsize = pip->i_d.di_extsize;
1200 } else if ((mode & S_IFMT) == S_IFREG) {
1201 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1202 di_flags |= XFS_DIFLAG_REALTIME;
1203 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1205 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1206 di_flags |= XFS_DIFLAG_EXTSIZE;
1207 ip->i_d.di_extsize = pip->i_d.di_extsize;
1210 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1211 xfs_inherit_noatime)
1212 di_flags |= XFS_DIFLAG_NOATIME;
1213 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1215 di_flags |= XFS_DIFLAG_NODUMP;
1216 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1218 di_flags |= XFS_DIFLAG_SYNC;
1219 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1220 xfs_inherit_nosymlinks)
1221 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1222 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1223 di_flags |= XFS_DIFLAG_PROJINHERIT;
1224 ip->i_d.di_flags |= di_flags;
1228 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1229 ip->i_df.if_flags = XFS_IFEXTENTS;
1230 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1231 ip->i_df.if_u1.if_extents = NULL;
1237 * Attribute fork settings for new inode.
1239 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1240 ip->i_d.di_anextents = 0;
1243 * Log the new values stuffed into the inode.
1245 xfs_trans_log_inode(tp, ip, flags);
1247 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1248 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1255 * Check to make sure that there are no blocks allocated to the
1256 * file beyond the size of the file. We don't check this for
1257 * files with fixed size extents or real time extents, but we
1258 * at least do it for regular files.
1267 xfs_fileoff_t map_first;
1269 xfs_bmbt_irec_t imaps[2];
1271 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1274 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1278 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1280 * The filesystem could be shutting down, so bmapi may return
1283 if (xfs_bmapi(NULL, ip, map_first,
1285 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1287 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1290 ASSERT(nimaps == 1);
1291 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1296 * Calculate the last possible buffered byte in a file. This must
1297 * include data that was buffered beyond the EOF by the write code.
1298 * This also needs to deal with overflowing the xfs_fsize_t type
1299 * which can happen for sizes near the limit.
1301 * We also need to take into account any blocks beyond the EOF. It
1302 * may be the case that they were buffered by a write which failed.
1303 * In that case the pages will still be in memory, but the inode size
1304 * will never have been updated.
1311 xfs_fsize_t last_byte;
1312 xfs_fileoff_t last_block;
1313 xfs_fileoff_t size_last_block;
1316 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1320 * Only check for blocks beyond the EOF if the extents have
1321 * been read in. This eliminates the need for the inode lock,
1322 * and it also saves us from looking when it really isn't
1325 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1326 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1334 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1335 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1337 last_byte = XFS_FSB_TO_B(mp, last_block);
1338 if (last_byte < 0) {
1339 return XFS_MAXIOFFSET(mp);
1341 last_byte += (1 << mp->m_writeio_log);
1342 if (last_byte < 0) {
1343 return XFS_MAXIOFFSET(mp);
1348 #if defined(XFS_RW_TRACE)
1354 xfs_fsize_t new_size,
1355 xfs_off_t toss_start,
1356 xfs_off_t toss_finish)
1358 if (ip->i_rwtrace == NULL) {
1362 ktrace_enter(ip->i_rwtrace,
1365 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1366 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1367 (void*)((long)flag),
1368 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1369 (void*)(unsigned long)(new_size & 0xffffffff),
1370 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1371 (void*)(unsigned long)(toss_start & 0xffffffff),
1372 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1373 (void*)(unsigned long)(toss_finish & 0xffffffff),
1374 (void*)(unsigned long)current_cpu(),
1381 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1385 * Start the truncation of the file to new_size. The new size
1386 * must be smaller than the current size. This routine will
1387 * clear the buffer and page caches of file data in the removed
1388 * range, and xfs_itruncate_finish() will remove the underlying
1391 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1392 * must NOT have the inode lock held at all. This is because we're
1393 * calling into the buffer/page cache code and we can't hold the
1394 * inode lock when we do so.
1396 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1397 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1398 * in the case that the caller is locking things out of order and
1399 * may not be able to call xfs_itruncate_finish() with the inode lock
1400 * held without dropping the I/O lock. If the caller must drop the
1401 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1402 * must be called again with all the same restrictions as the initial
1406 xfs_itruncate_start(
1409 xfs_fsize_t new_size)
1411 xfs_fsize_t last_byte;
1412 xfs_off_t toss_start;
1416 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1417 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1418 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1419 (flags == XFS_ITRUNC_MAYBE));
1424 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1425 * overlapping the region being removed. We have to use
1426 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1427 * caller may not be able to finish the truncate without
1428 * dropping the inode's I/O lock. Make sure
1429 * to catch any pages brought in by buffers overlapping
1430 * the EOF by searching out beyond the isize by our
1431 * block size. We round new_size up to a block boundary
1432 * so that we don't toss things on the same block as
1433 * new_size but before it.
1435 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1436 * call remapf() over the same region if the file is mapped.
1437 * This frees up mapped file references to the pages in the
1438 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1439 * that we get the latest mapped changes flushed out.
1441 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1442 toss_start = XFS_FSB_TO_B(mp, toss_start);
1443 if (toss_start < 0) {
1445 * The place to start tossing is beyond our maximum
1446 * file size, so there is no way that the data extended
1451 last_byte = xfs_file_last_byte(ip);
1452 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1454 if (last_byte > toss_start) {
1455 if (flags & XFS_ITRUNC_DEFINITE) {
1456 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1458 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1463 if (new_size == 0) {
1464 ASSERT(VN_CACHED(vp) == 0);
1470 * Shrink the file to the given new_size. The new
1471 * size must be smaller than the current size.
1472 * This will free up the underlying blocks
1473 * in the removed range after a call to xfs_itruncate_start()
1474 * or xfs_atruncate_start().
1476 * The transaction passed to this routine must have made
1477 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1478 * This routine may commit the given transaction and
1479 * start new ones, so make sure everything involved in
1480 * the transaction is tidy before calling here.
1481 * Some transaction will be returned to the caller to be
1482 * committed. The incoming transaction must already include
1483 * the inode, and both inode locks must be held exclusively.
1484 * The inode must also be "held" within the transaction. On
1485 * return the inode will be "held" within the returned transaction.
1486 * This routine does NOT require any disk space to be reserved
1487 * for it within the transaction.
1489 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1490 * and it indicates the fork which is to be truncated. For the
1491 * attribute fork we only support truncation to size 0.
1493 * We use the sync parameter to indicate whether or not the first
1494 * transaction we perform might have to be synchronous. For the attr fork,
1495 * it needs to be so if the unlink of the inode is not yet known to be
1496 * permanent in the log. This keeps us from freeing and reusing the
1497 * blocks of the attribute fork before the unlink of the inode becomes
1500 * For the data fork, we normally have to run synchronously if we're
1501 * being called out of the inactive path or we're being called
1502 * out of the create path where we're truncating an existing file.
1503 * Either way, the truncate needs to be sync so blocks don't reappear
1504 * in the file with altered data in case of a crash. wsync filesystems
1505 * can run the first case async because anything that shrinks the inode
1506 * has to run sync so by the time we're called here from inactive, the
1507 * inode size is permanently set to 0.
1509 * Calls from the truncate path always need to be sync unless we're
1510 * in a wsync filesystem and the file has already been unlinked.
1512 * The caller is responsible for correctly setting the sync parameter.
1513 * It gets too hard for us to guess here which path we're being called
1514 * out of just based on inode state.
1517 xfs_itruncate_finish(
1520 xfs_fsize_t new_size,
1524 xfs_fsblock_t first_block;
1525 xfs_fileoff_t first_unmap_block;
1526 xfs_fileoff_t last_block;
1527 xfs_filblks_t unmap_len=0;
1532 xfs_bmap_free_t free_list;
1535 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1536 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1537 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1538 ASSERT(*tp != NULL);
1539 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1540 ASSERT(ip->i_transp == *tp);
1541 ASSERT(ip->i_itemp != NULL);
1542 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1546 mp = (ntp)->t_mountp;
1547 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1550 * We only support truncating the entire attribute fork.
1552 if (fork == XFS_ATTR_FORK) {
1555 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1556 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1558 * The first thing we do is set the size to new_size permanently
1559 * on disk. This way we don't have to worry about anyone ever
1560 * being able to look at the data being freed even in the face
1561 * of a crash. What we're getting around here is the case where
1562 * we free a block, it is allocated to another file, it is written
1563 * to, and then we crash. If the new data gets written to the
1564 * file but the log buffers containing the free and reallocation
1565 * don't, then we'd end up with garbage in the blocks being freed.
1566 * As long as we make the new_size permanent before actually
1567 * freeing any blocks it doesn't matter if they get writtten to.
1569 * The callers must signal into us whether or not the size
1570 * setting here must be synchronous. There are a few cases
1571 * where it doesn't have to be synchronous. Those cases
1572 * occur if the file is unlinked and we know the unlink is
1573 * permanent or if the blocks being truncated are guaranteed
1574 * to be beyond the inode eof (regardless of the link count)
1575 * and the eof value is permanent. Both of these cases occur
1576 * only on wsync-mounted filesystems. In those cases, we're
1577 * guaranteed that no user will ever see the data in the blocks
1578 * that are being truncated so the truncate can run async.
1579 * In the free beyond eof case, the file may wind up with
1580 * more blocks allocated to it than it needs if we crash
1581 * and that won't get fixed until the next time the file
1582 * is re-opened and closed but that's ok as that shouldn't
1583 * be too many blocks.
1585 * However, we can't just make all wsync xactions run async
1586 * because there's one call out of the create path that needs
1587 * to run sync where it's truncating an existing file to size
1588 * 0 whose size is > 0.
1590 * It's probably possible to come up with a test in this
1591 * routine that would correctly distinguish all the above
1592 * cases from the values of the function parameters and the
1593 * inode state but for sanity's sake, I've decided to let the
1594 * layers above just tell us. It's simpler to correctly figure
1595 * out in the layer above exactly under what conditions we
1596 * can run async and I think it's easier for others read and
1597 * follow the logic in case something has to be changed.
1598 * cscope is your friend -- rcc.
1600 * The attribute fork is much simpler.
1602 * For the attribute fork we allow the caller to tell us whether
1603 * the unlink of the inode that led to this call is yet permanent
1604 * in the on disk log. If it is not and we will be freeing extents
1605 * in this inode then we make the first transaction synchronous
1606 * to make sure that the unlink is permanent by the time we free
1609 if (fork == XFS_DATA_FORK) {
1610 if (ip->i_d.di_nextents > 0) {
1611 ip->i_d.di_size = new_size;
1612 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1615 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1616 if (ip->i_d.di_anextents > 0)
1617 xfs_trans_set_sync(ntp);
1619 ASSERT(fork == XFS_DATA_FORK ||
1620 (fork == XFS_ATTR_FORK &&
1621 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1622 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1625 * Since it is possible for space to become allocated beyond
1626 * the end of the file (in a crash where the space is allocated
1627 * but the inode size is not yet updated), simply remove any
1628 * blocks which show up between the new EOF and the maximum
1629 * possible file size. If the first block to be removed is
1630 * beyond the maximum file size (ie it is the same as last_block),
1631 * then there is nothing to do.
1633 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1634 ASSERT(first_unmap_block <= last_block);
1636 if (last_block == first_unmap_block) {
1639 unmap_len = last_block - first_unmap_block + 1;
1643 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1644 * will tell us whether it freed the entire range or
1645 * not. If this is a synchronous mount (wsync),
1646 * then we can tell bunmapi to keep all the
1647 * transactions asynchronous since the unlink
1648 * transaction that made this inode inactive has
1649 * already hit the disk. There's no danger of
1650 * the freed blocks being reused, there being a
1651 * crash, and the reused blocks suddenly reappearing
1652 * in this file with garbage in them once recovery
1655 XFS_BMAP_INIT(&free_list, &first_block);
1656 error = xfs_bunmapi(ntp, ip, first_unmap_block,
1658 XFS_BMAPI_AFLAG(fork) |
1659 (sync ? 0 : XFS_BMAPI_ASYNC),
1660 XFS_ITRUNC_MAX_EXTENTS,
1661 &first_block, &free_list, &done);
1664 * If the bunmapi call encounters an error,
1665 * return to the caller where the transaction
1666 * can be properly aborted. We just need to
1667 * make sure we're not holding any resources
1668 * that we were not when we came in.
1670 xfs_bmap_cancel(&free_list);
1675 * Duplicate the transaction that has the permanent
1676 * reservation and commit the old transaction.
1678 error = xfs_bmap_finish(tp, &free_list, first_block,
1683 * If the bmap finish call encounters an error,
1684 * return to the caller where the transaction
1685 * can be properly aborted. We just need to
1686 * make sure we're not holding any resources
1687 * that we were not when we came in.
1689 * Aborting from this point might lose some
1690 * blocks in the file system, but oh well.
1692 xfs_bmap_cancel(&free_list);
1695 * If the passed in transaction committed
1696 * in xfs_bmap_finish(), then we want to
1697 * add the inode to this one before returning.
1698 * This keeps things simple for the higher
1699 * level code, because it always knows that
1700 * the inode is locked and held in the
1701 * transaction that returns to it whether
1702 * errors occur or not. We don't mark the
1703 * inode dirty so that this transaction can
1704 * be easily aborted if possible.
1706 xfs_trans_ijoin(ntp, ip,
1707 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1708 xfs_trans_ihold(ntp, ip);
1715 * The first xact was committed,
1716 * so add the inode to the new one.
1717 * Mark it dirty so it will be logged
1718 * and moved forward in the log as
1719 * part of every commit.
1721 xfs_trans_ijoin(ntp, ip,
1722 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1723 xfs_trans_ihold(ntp, ip);
1724 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1726 ntp = xfs_trans_dup(ntp);
1727 (void) xfs_trans_commit(*tp, 0, NULL);
1729 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1730 XFS_TRANS_PERM_LOG_RES,
1731 XFS_ITRUNCATE_LOG_COUNT);
1733 * Add the inode being truncated to the next chained
1736 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1737 xfs_trans_ihold(ntp, ip);
1742 * Only update the size in the case of the data fork, but
1743 * always re-log the inode so that our permanent transaction
1744 * can keep on rolling it forward in the log.
1746 if (fork == XFS_DATA_FORK) {
1747 xfs_isize_check(mp, ip, new_size);
1748 ip->i_d.di_size = new_size;
1750 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1751 ASSERT((new_size != 0) ||
1752 (fork == XFS_ATTR_FORK) ||
1753 (ip->i_delayed_blks == 0));
1754 ASSERT((new_size != 0) ||
1755 (fork == XFS_ATTR_FORK) ||
1756 (ip->i_d.di_nextents == 0));
1757 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1765 * Do the first part of growing a file: zero any data in the last
1766 * block that is beyond the old EOF. We need to do this before
1767 * the inode is joined to the transaction to modify the i_size.
1768 * That way we can drop the inode lock and call into the buffer
1769 * cache to get the buffer mapping the EOF.
1774 xfs_fsize_t new_size,
1779 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1780 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1781 ASSERT(new_size > ip->i_d.di_size);
1784 * Zero any pages that may have been created by
1785 * xfs_write_file() beyond the end of the file
1786 * and any blocks between the old and new file sizes.
1788 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1789 ip->i_d.di_size, new_size);
1796 * This routine is called to extend the size of a file.
1797 * The inode must have both the iolock and the ilock locked
1798 * for update and it must be a part of the current transaction.
1799 * The xfs_igrow_start() function must have been called previously.
1800 * If the change_flag is not zero, the inode change timestamp will
1807 xfs_fsize_t new_size,
1810 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1811 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1812 ASSERT(ip->i_transp == tp);
1813 ASSERT(new_size > ip->i_d.di_size);
1816 * Update the file size. Update the inode change timestamp
1817 * if change_flag set.
1819 ip->i_d.di_size = new_size;
1821 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1822 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1828 * This is called when the inode's link count goes to 0.
1829 * We place the on-disk inode on a list in the AGI. It
1830 * will be pulled from this list when the inode is freed.
1842 xfs_agnumber_t agno;
1843 xfs_daddr_t agdaddr;
1850 ASSERT(ip->i_d.di_nlink == 0);
1851 ASSERT(ip->i_d.di_mode != 0);
1852 ASSERT(ip->i_transp == tp);
1856 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1857 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1860 * Get the agi buffer first. It ensures lock ordering
1863 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1864 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1869 * Validate the magic number of the agi block.
1871 agi = XFS_BUF_TO_AGI(agibp);
1873 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1874 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1875 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1876 XFS_RANDOM_IUNLINK))) {
1877 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1878 xfs_trans_brelse(tp, agibp);
1879 return XFS_ERROR(EFSCORRUPTED);
1882 * Get the index into the agi hash table for the
1883 * list this inode will go on.
1885 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1887 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1888 ASSERT(agi->agi_unlinked[bucket_index]);
1889 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1891 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1893 * There is already another inode in the bucket we need
1894 * to add ourselves to. Add us at the front of the list.
1895 * Here we put the head pointer into our next pointer,
1896 * and then we fall through to point the head at us.
1898 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1902 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1903 ASSERT(dip->di_next_unlinked);
1904 /* both on-disk, don't endian flip twice */
1905 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1906 offset = ip->i_boffset +
1907 offsetof(xfs_dinode_t, di_next_unlinked);
1908 xfs_trans_inode_buf(tp, ibp);
1909 xfs_trans_log_buf(tp, ibp, offset,
1910 (offset + sizeof(xfs_agino_t) - 1));
1911 xfs_inobp_check(mp, ibp);
1915 * Point the bucket head pointer at the inode being inserted.
1918 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1919 offset = offsetof(xfs_agi_t, agi_unlinked) +
1920 (sizeof(xfs_agino_t) * bucket_index);
1921 xfs_trans_log_buf(tp, agibp, offset,
1922 (offset + sizeof(xfs_agino_t) - 1));
1927 * Pull the on-disk inode from the AGI unlinked list.
1940 xfs_agnumber_t agno;
1941 xfs_daddr_t agdaddr;
1943 xfs_agino_t next_agino;
1944 xfs_buf_t *last_ibp;
1945 xfs_dinode_t *last_dip;
1947 int offset, last_offset;
1952 * First pull the on-disk inode from the AGI unlinked list.
1956 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1957 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1960 * Get the agi buffer first. It ensures lock ordering
1963 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1964 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1967 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1968 error, mp->m_fsname);
1972 * Validate the magic number of the agi block.
1974 agi = XFS_BUF_TO_AGI(agibp);
1976 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1977 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1978 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1979 XFS_RANDOM_IUNLINK_REMOVE))) {
1980 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1982 xfs_trans_brelse(tp, agibp);
1984 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1986 return XFS_ERROR(EFSCORRUPTED);
1989 * Get the index into the agi hash table for the
1990 * list this inode will go on.
1992 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1994 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1995 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1996 ASSERT(agi->agi_unlinked[bucket_index]);
1998 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2000 * We're at the head of the list. Get the inode's
2001 * on-disk buffer to see if there is anyone after us
2002 * on the list. Only modify our next pointer if it
2003 * is not already NULLAGINO. This saves us the overhead
2004 * of dealing with the buffer when there is no need to
2007 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2010 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2011 error, mp->m_fsname);
2014 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2015 ASSERT(next_agino != 0);
2016 if (next_agino != NULLAGINO) {
2017 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2018 offset = ip->i_boffset +
2019 offsetof(xfs_dinode_t, di_next_unlinked);
2020 xfs_trans_inode_buf(tp, ibp);
2021 xfs_trans_log_buf(tp, ibp, offset,
2022 (offset + sizeof(xfs_agino_t) - 1));
2023 xfs_inobp_check(mp, ibp);
2025 xfs_trans_brelse(tp, ibp);
2028 * Point the bucket head pointer at the next inode.
2030 ASSERT(next_agino != 0);
2031 ASSERT(next_agino != agino);
2032 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2033 offset = offsetof(xfs_agi_t, agi_unlinked) +
2034 (sizeof(xfs_agino_t) * bucket_index);
2035 xfs_trans_log_buf(tp, agibp, offset,
2036 (offset + sizeof(xfs_agino_t) - 1));
2039 * We need to search the list for the inode being freed.
2041 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2043 while (next_agino != agino) {
2045 * If the last inode wasn't the one pointing to
2046 * us, then release its buffer since we're not
2047 * going to do anything with it.
2049 if (last_ibp != NULL) {
2050 xfs_trans_brelse(tp, last_ibp);
2052 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2053 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2054 &last_ibp, &last_offset);
2057 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2058 error, mp->m_fsname);
2061 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2062 ASSERT(next_agino != NULLAGINO);
2063 ASSERT(next_agino != 0);
2066 * Now last_ibp points to the buffer previous to us on
2067 * the unlinked list. Pull us from the list.
2069 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2072 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2073 error, mp->m_fsname);
2076 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2077 ASSERT(next_agino != 0);
2078 ASSERT(next_agino != agino);
2079 if (next_agino != NULLAGINO) {
2080 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2081 offset = ip->i_boffset +
2082 offsetof(xfs_dinode_t, di_next_unlinked);
2083 xfs_trans_inode_buf(tp, ibp);
2084 xfs_trans_log_buf(tp, ibp, offset,
2085 (offset + sizeof(xfs_agino_t) - 1));
2086 xfs_inobp_check(mp, ibp);
2088 xfs_trans_brelse(tp, ibp);
2091 * Point the previous inode on the list to the next inode.
2093 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2094 ASSERT(next_agino != 0);
2095 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2096 xfs_trans_inode_buf(tp, last_ibp);
2097 xfs_trans_log_buf(tp, last_ibp, offset,
2098 (offset + sizeof(xfs_agino_t) - 1));
2099 xfs_inobp_check(mp, last_ibp);
2104 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2106 return (((ip->i_itemp == NULL) ||
2107 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2108 (ip->i_update_core == 0));
2113 xfs_inode_t *free_ip,
2117 xfs_mount_t *mp = free_ip->i_mount;
2118 int blks_per_cluster;
2121 int i, j, found, pre_flushed;
2125 xfs_inode_t *ip, **ip_found;
2126 xfs_inode_log_item_t *iip;
2127 xfs_log_item_t *lip;
2130 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2131 blks_per_cluster = 1;
2132 ninodes = mp->m_sb.sb_inopblock;
2133 nbufs = XFS_IALLOC_BLOCKS(mp);
2135 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2136 mp->m_sb.sb_blocksize;
2137 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2138 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2141 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2143 for (j = 0; j < nbufs; j++, inum += ninodes) {
2144 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2145 XFS_INO_TO_AGBNO(mp, inum));
2149 * Look for each inode in memory and attempt to lock it,
2150 * we can be racing with flush and tail pushing here.
2151 * any inode we get the locks on, add to an array of
2152 * inode items to process later.
2154 * The get the buffer lock, we could beat a flush
2155 * or tail pushing thread to the lock here, in which
2156 * case they will go looking for the inode buffer
2157 * and fail, we need some other form of interlock
2161 for (i = 0; i < ninodes; i++) {
2162 ih = XFS_IHASH(mp, inum + i);
2163 read_lock(&ih->ih_lock);
2164 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2165 if (ip->i_ino == inum + i)
2169 /* Inode not in memory or we found it already,
2172 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2173 read_unlock(&ih->ih_lock);
2177 if (xfs_inode_clean(ip)) {
2178 read_unlock(&ih->ih_lock);
2182 /* If we can get the locks then add it to the
2183 * list, otherwise by the time we get the bp lock
2184 * below it will already be attached to the
2188 /* This inode will already be locked - by us, lets
2192 if (ip == free_ip) {
2193 if (xfs_iflock_nowait(ip)) {
2194 ip->i_flags |= XFS_ISTALE;
2196 if (xfs_inode_clean(ip)) {
2199 ip_found[found++] = ip;
2202 read_unlock(&ih->ih_lock);
2206 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2207 if (xfs_iflock_nowait(ip)) {
2208 ip->i_flags |= XFS_ISTALE;
2210 if (xfs_inode_clean(ip)) {
2212 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2214 ip_found[found++] = ip;
2217 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2221 read_unlock(&ih->ih_lock);
2224 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2225 mp->m_bsize * blks_per_cluster,
2229 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2231 if (lip->li_type == XFS_LI_INODE) {
2232 iip = (xfs_inode_log_item_t *)lip;
2233 ASSERT(iip->ili_logged == 1);
2234 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2236 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2238 iip->ili_inode->i_flags |= XFS_ISTALE;
2241 lip = lip->li_bio_list;
2244 for (i = 0; i < found; i++) {
2249 ip->i_update_core = 0;
2251 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2255 iip->ili_last_fields = iip->ili_format.ilf_fields;
2256 iip->ili_format.ilf_fields = 0;
2257 iip->ili_logged = 1;
2259 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2262 xfs_buf_attach_iodone(bp,
2263 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2264 xfs_istale_done, (xfs_log_item_t *)iip);
2265 if (ip != free_ip) {
2266 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2270 if (found || pre_flushed)
2271 xfs_trans_stale_inode_buf(tp, bp);
2272 xfs_trans_binval(tp, bp);
2275 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2279 * This is called to return an inode to the inode free list.
2280 * The inode should already be truncated to 0 length and have
2281 * no pages associated with it. This routine also assumes that
2282 * the inode is already a part of the transaction.
2284 * The on-disk copy of the inode will have been added to the list
2285 * of unlinked inodes in the AGI. We need to remove the inode from
2286 * that list atomically with respect to freeing it here.
2292 xfs_bmap_free_t *flist)
2296 xfs_ino_t first_ino;
2298 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2299 ASSERT(ip->i_transp == tp);
2300 ASSERT(ip->i_d.di_nlink == 0);
2301 ASSERT(ip->i_d.di_nextents == 0);
2302 ASSERT(ip->i_d.di_anextents == 0);
2303 ASSERT((ip->i_d.di_size == 0) ||
2304 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2305 ASSERT(ip->i_d.di_nblocks == 0);
2308 * Pull the on-disk inode from the AGI unlinked list.
2310 error = xfs_iunlink_remove(tp, ip);
2315 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2319 ip->i_d.di_mode = 0; /* mark incore inode as free */
2320 ip->i_d.di_flags = 0;
2321 ip->i_d.di_dmevmask = 0;
2322 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2323 ip->i_df.if_ext_max =
2324 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2325 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2326 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2328 * Bump the generation count so no one will be confused
2329 * by reincarnations of this inode.
2332 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2335 xfs_ifree_cluster(ip, tp, first_ino);
2342 * Reallocate the space for if_broot based on the number of records
2343 * being added or deleted as indicated in rec_diff. Move the records
2344 * and pointers in if_broot to fit the new size. When shrinking this
2345 * will eliminate holes between the records and pointers created by
2346 * the caller. When growing this will create holes to be filled in
2349 * The caller must not request to add more records than would fit in
2350 * the on-disk inode root. If the if_broot is currently NULL, then
2351 * if we adding records one will be allocated. The caller must also
2352 * not request that the number of records go below zero, although
2353 * it can go to zero.
2355 * ip -- the inode whose if_broot area is changing
2356 * ext_diff -- the change in the number of records, positive or negative,
2357 * requested for the if_broot array.
2367 xfs_bmbt_block_t *new_broot;
2374 * Handle the degenerate case quietly.
2376 if (rec_diff == 0) {
2380 ifp = XFS_IFORK_PTR(ip, whichfork);
2383 * If there wasn't any memory allocated before, just
2384 * allocate it now and get out.
2386 if (ifp->if_broot_bytes == 0) {
2387 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2388 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2390 ifp->if_broot_bytes = (int)new_size;
2395 * If there is already an existing if_broot, then we need
2396 * to realloc() it and shift the pointers to their new
2397 * location. The records don't change location because
2398 * they are kept butted up against the btree block header.
2400 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2401 new_max = cur_max + rec_diff;
2402 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2403 ifp->if_broot = (xfs_bmbt_block_t *)
2404 kmem_realloc(ifp->if_broot,
2406 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2408 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2409 ifp->if_broot_bytes);
2410 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2412 ifp->if_broot_bytes = (int)new_size;
2413 ASSERT(ifp->if_broot_bytes <=
2414 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2415 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2420 * rec_diff is less than 0. In this case, we are shrinking the
2421 * if_broot buffer. It must already exist. If we go to zero
2422 * records, just get rid of the root and clear the status bit.
2424 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2425 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2426 new_max = cur_max + rec_diff;
2427 ASSERT(new_max >= 0);
2429 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2433 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2435 * First copy over the btree block header.
2437 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2440 ifp->if_flags &= ~XFS_IFBROOT;
2444 * Only copy the records and pointers if there are any.
2448 * First copy the records.
2450 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2451 ifp->if_broot_bytes);
2452 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2454 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2457 * Then copy the pointers.
2459 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2460 ifp->if_broot_bytes);
2461 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2463 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2465 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2466 ifp->if_broot = new_broot;
2467 ifp->if_broot_bytes = (int)new_size;
2468 ASSERT(ifp->if_broot_bytes <=
2469 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2475 * This is called when the amount of space needed for if_data
2476 * is increased or decreased. The change in size is indicated by
2477 * the number of bytes that need to be added or deleted in the
2478 * byte_diff parameter.
2480 * If the amount of space needed has decreased below the size of the
2481 * inline buffer, then switch to using the inline buffer. Otherwise,
2482 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2483 * to what is needed.
2485 * ip -- the inode whose if_data area is changing
2486 * byte_diff -- the change in the number of bytes, positive or negative,
2487 * requested for the if_data array.
2499 if (byte_diff == 0) {
2503 ifp = XFS_IFORK_PTR(ip, whichfork);
2504 new_size = (int)ifp->if_bytes + byte_diff;
2505 ASSERT(new_size >= 0);
2507 if (new_size == 0) {
2508 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2509 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2511 ifp->if_u1.if_data = NULL;
2513 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2515 * If the valid extents/data can fit in if_inline_ext/data,
2516 * copy them from the malloc'd vector and free it.
2518 if (ifp->if_u1.if_data == NULL) {
2519 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2520 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2521 ASSERT(ifp->if_real_bytes != 0);
2522 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2524 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2525 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2530 * Stuck with malloc/realloc.
2531 * For inline data, the underlying buffer must be
2532 * a multiple of 4 bytes in size so that it can be
2533 * logged and stay on word boundaries. We enforce
2536 real_size = roundup(new_size, 4);
2537 if (ifp->if_u1.if_data == NULL) {
2538 ASSERT(ifp->if_real_bytes == 0);
2539 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2540 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2542 * Only do the realloc if the underlying size
2543 * is really changing.
2545 if (ifp->if_real_bytes != real_size) {
2546 ifp->if_u1.if_data =
2547 kmem_realloc(ifp->if_u1.if_data,
2553 ASSERT(ifp->if_real_bytes == 0);
2554 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2555 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2559 ifp->if_real_bytes = real_size;
2560 ifp->if_bytes = new_size;
2561 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2568 * Map inode to disk block and offset.
2570 * mp -- the mount point structure for the current file system
2571 * tp -- the current transaction
2572 * ino -- the inode number of the inode to be located
2573 * imap -- this structure is filled in with the information necessary
2574 * to retrieve the given inode from disk
2575 * flags -- flags to pass to xfs_dilocate indicating whether or not
2576 * lookups in the inode btree were OK or not
2586 xfs_fsblock_t fsbno;
2591 fsbno = imap->im_blkno ?
2592 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2593 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2597 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2598 imap->im_len = XFS_FSB_TO_BB(mp, len);
2599 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2600 imap->im_ioffset = (ushort)off;
2601 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2612 ifp = XFS_IFORK_PTR(ip, whichfork);
2613 if (ifp->if_broot != NULL) {
2614 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2615 ifp->if_broot = NULL;
2619 * If the format is local, then we can't have an extents
2620 * array so just look for an inline data array. If we're
2621 * not local then we may or may not have an extents list,
2622 * so check and free it up if we do.
2624 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2625 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2626 (ifp->if_u1.if_data != NULL)) {
2627 ASSERT(ifp->if_real_bytes != 0);
2628 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2629 ifp->if_u1.if_data = NULL;
2630 ifp->if_real_bytes = 0;
2632 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2633 ((ifp->if_flags & XFS_IFEXTIREC) ||
2634 ((ifp->if_u1.if_extents != NULL) &&
2635 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2636 ASSERT(ifp->if_real_bytes != 0);
2637 xfs_iext_destroy(ifp);
2639 ASSERT(ifp->if_u1.if_extents == NULL ||
2640 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2641 ASSERT(ifp->if_real_bytes == 0);
2642 if (whichfork == XFS_ATTR_FORK) {
2643 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2649 * This is called free all the memory associated with an inode.
2650 * It must free the inode itself and any buffers allocated for
2651 * if_extents/if_data and if_broot. It must also free the lock
2652 * associated with the inode.
2659 switch (ip->i_d.di_mode & S_IFMT) {
2663 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2667 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2668 mrfree(&ip->i_lock);
2669 mrfree(&ip->i_iolock);
2670 freesema(&ip->i_flock);
2671 #ifdef XFS_BMAP_TRACE
2672 ktrace_free(ip->i_xtrace);
2674 #ifdef XFS_BMBT_TRACE
2675 ktrace_free(ip->i_btrace);
2678 ktrace_free(ip->i_rwtrace);
2680 #ifdef XFS_ILOCK_TRACE
2681 ktrace_free(ip->i_lock_trace);
2683 #ifdef XFS_DIR2_TRACE
2684 ktrace_free(ip->i_dir_trace);
2687 /* XXXdpd should be able to assert this but shutdown
2688 * is leaving the AIL behind. */
2689 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2690 XFS_FORCED_SHUTDOWN(ip->i_mount));
2691 xfs_inode_item_destroy(ip);
2693 kmem_zone_free(xfs_inode_zone, ip);
2698 * Increment the pin count of the given buffer.
2699 * This value is protected by ipinlock spinlock in the mount structure.
2705 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2707 atomic_inc(&ip->i_pincount);
2711 * Decrement the pin count of the given inode, and wake up
2712 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2713 * inode must have been previoulsy pinned with a call to xfs_ipin().
2719 ASSERT(atomic_read(&ip->i_pincount) > 0);
2721 if (atomic_dec_and_test(&ip->i_pincount)) {
2722 vnode_t *vp = XFS_ITOV_NULL(ip);
2724 /* make sync come back and flush this inode */
2726 struct inode *inode = vn_to_inode(vp);
2728 if (!(inode->i_state & I_NEW))
2729 mark_inode_dirty_sync(inode);
2732 wake_up(&ip->i_ipin_wait);
2737 * This is called to wait for the given inode to be unpinned.
2738 * It will sleep until this happens. The caller must have the
2739 * inode locked in at least shared mode so that the buffer cannot
2740 * be subsequently pinned once someone is waiting for it to be
2747 xfs_inode_log_item_t *iip;
2750 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2752 if (atomic_read(&ip->i_pincount) == 0) {
2757 if (iip && iip->ili_last_lsn) {
2758 lsn = iip->ili_last_lsn;
2764 * Give the log a push so we don't wait here too long.
2766 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2768 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2773 * xfs_iextents_copy()
2775 * This is called to copy the REAL extents (as opposed to the delayed
2776 * allocation extents) from the inode into the given buffer. It
2777 * returns the number of bytes copied into the buffer.
2779 * If there are no delayed allocation extents, then we can just
2780 * memcpy() the extents into the buffer. Otherwise, we need to
2781 * examine each extent in turn and skip those which are delayed.
2786 xfs_bmbt_rec_t *buffer,
2790 xfs_bmbt_rec_t *dest_ep;
2792 #ifdef XFS_BMAP_TRACE
2793 static char fname[] = "xfs_iextents_copy";
2798 xfs_fsblock_t start_block;
2800 ifp = XFS_IFORK_PTR(ip, whichfork);
2801 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2802 ASSERT(ifp->if_bytes > 0);
2804 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2805 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2809 * There are some delayed allocation extents in the
2810 * inode, so copy the extents one at a time and skip
2811 * the delayed ones. There must be at least one
2812 * non-delayed extent.
2816 for (i = 0; i < nrecs; i++) {
2817 ep = xfs_iext_get_ext(ifp, i);
2818 start_block = xfs_bmbt_get_startblock(ep);
2819 if (ISNULLSTARTBLOCK(start_block)) {
2821 * It's a delayed allocation extent, so skip it.
2826 /* Translate to on disk format */
2827 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2828 (__uint64_t*)&dest_ep->l0);
2829 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2830 (__uint64_t*)&dest_ep->l1);
2834 ASSERT(copied != 0);
2835 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2837 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2841 * Each of the following cases stores data into the same region
2842 * of the on-disk inode, so only one of them can be valid at
2843 * any given time. While it is possible to have conflicting formats
2844 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2845 * in EXTENTS format, this can only happen when the fork has
2846 * changed formats after being modified but before being flushed.
2847 * In these cases, the format always takes precedence, because the
2848 * format indicates the current state of the fork.
2855 xfs_inode_log_item_t *iip,
2862 #ifdef XFS_TRANS_DEBUG
2865 static const short brootflag[2] =
2866 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2867 static const short dataflag[2] =
2868 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2869 static const short extflag[2] =
2870 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2874 ifp = XFS_IFORK_PTR(ip, whichfork);
2876 * This can happen if we gave up in iformat in an error path,
2877 * for the attribute fork.
2880 ASSERT(whichfork == XFS_ATTR_FORK);
2883 cp = XFS_DFORK_PTR(dip, whichfork);
2885 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2886 case XFS_DINODE_FMT_LOCAL:
2887 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2888 (ifp->if_bytes > 0)) {
2889 ASSERT(ifp->if_u1.if_data != NULL);
2890 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2891 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2893 if (whichfork == XFS_DATA_FORK) {
2894 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2895 XFS_ERROR_REPORT("xfs_iflush_fork",
2896 XFS_ERRLEVEL_LOW, mp);
2897 return XFS_ERROR(EFSCORRUPTED);
2902 case XFS_DINODE_FMT_EXTENTS:
2903 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2904 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2905 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2906 (ifp->if_bytes == 0));
2907 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2908 (ifp->if_bytes > 0));
2909 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2910 (ifp->if_bytes > 0)) {
2911 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2912 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2917 case XFS_DINODE_FMT_BTREE:
2918 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2919 (ifp->if_broot_bytes > 0)) {
2920 ASSERT(ifp->if_broot != NULL);
2921 ASSERT(ifp->if_broot_bytes <=
2922 (XFS_IFORK_SIZE(ip, whichfork) +
2923 XFS_BROOT_SIZE_ADJ));
2924 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2925 (xfs_bmdr_block_t *)cp,
2926 XFS_DFORK_SIZE(dip, mp, whichfork));
2930 case XFS_DINODE_FMT_DEV:
2931 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2932 ASSERT(whichfork == XFS_DATA_FORK);
2933 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2937 case XFS_DINODE_FMT_UUID:
2938 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2939 ASSERT(whichfork == XFS_DATA_FORK);
2940 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2954 * xfs_iflush() will write a modified inode's changes out to the
2955 * inode's on disk home. The caller must have the inode lock held
2956 * in at least shared mode and the inode flush semaphore must be
2957 * held as well. The inode lock will still be held upon return from
2958 * the call and the caller is free to unlock it.
2959 * The inode flush lock will be unlocked when the inode reaches the disk.
2960 * The flags indicate how the inode's buffer should be written out.
2967 xfs_inode_log_item_t *iip;
2975 int clcount; /* count of inodes clustered */
2977 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
2980 XFS_STATS_INC(xs_iflush_count);
2982 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2983 ASSERT(valusema(&ip->i_flock) <= 0);
2984 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2985 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2991 * If the inode isn't dirty, then just release the inode
2992 * flush lock and do nothing.
2994 if ((ip->i_update_core == 0) &&
2995 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
2996 ASSERT((iip != NULL) ?
2997 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3003 * We can't flush the inode until it is unpinned, so
3004 * wait for it. We know noone new can pin it, because
3005 * we are holding the inode lock shared and you need
3006 * to hold it exclusively to pin the inode.
3008 xfs_iunpin_wait(ip);
3011 * This may have been unpinned because the filesystem is shutting
3012 * down forcibly. If that's the case we must not write this inode
3013 * to disk, because the log record didn't make it to disk!
3015 if (XFS_FORCED_SHUTDOWN(mp)) {
3016 ip->i_update_core = 0;
3018 iip->ili_format.ilf_fields = 0;
3020 return XFS_ERROR(EIO);
3024 * Get the buffer containing the on-disk inode.
3026 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0);
3033 * Decide how buffer will be flushed out. This is done before
3034 * the call to xfs_iflush_int because this field is zeroed by it.
3036 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3038 * Flush out the inode buffer according to the directions
3039 * of the caller. In the cases where the caller has given
3040 * us a choice choose the non-delwri case. This is because
3041 * the inode is in the AIL and we need to get it out soon.
3044 case XFS_IFLUSH_SYNC:
3045 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3048 case XFS_IFLUSH_ASYNC:
3049 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3052 case XFS_IFLUSH_DELWRI:
3062 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3063 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3064 case XFS_IFLUSH_DELWRI:
3067 case XFS_IFLUSH_ASYNC:
3070 case XFS_IFLUSH_SYNC:
3081 * First flush out the inode that xfs_iflush was called with.
3083 error = xfs_iflush_int(ip, bp);
3090 * see if other inodes can be gathered into this write
3093 ip->i_chash->chl_buf = bp;
3095 ch = XFS_CHASH(mp, ip->i_blkno);
3096 s = mutex_spinlock(&ch->ch_lock);
3099 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3101 * Do an un-protected check to see if the inode is dirty and
3102 * is a candidate for flushing. These checks will be repeated
3103 * later after the appropriate locks are acquired.
3106 if ((iq->i_update_core == 0) &&
3108 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3109 xfs_ipincount(iq) == 0) {
3114 * Try to get locks. If any are unavailable,
3115 * then this inode cannot be flushed and is skipped.
3118 /* get inode locks (just i_lock) */
3119 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3120 /* get inode flush lock */
3121 if (xfs_iflock_nowait(iq)) {
3122 /* check if pinned */
3123 if (xfs_ipincount(iq) == 0) {
3124 /* arriving here means that
3125 * this inode can be flushed.
3126 * first re-check that it's
3130 if ((iq->i_update_core != 0)||
3132 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3134 error = xfs_iflush_int(iq, bp);
3138 goto cluster_corrupt_out;
3147 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3150 mutex_spinunlock(&ch->ch_lock, s);
3153 XFS_STATS_INC(xs_icluster_flushcnt);
3154 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3158 * If the buffer is pinned then push on the log so we won't
3159 * get stuck waiting in the write for too long.
3161 if (XFS_BUF_ISPINNED(bp)){
3162 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3165 if (flags & INT_DELWRI) {
3166 xfs_bdwrite(mp, bp);
3167 } else if (flags & INT_ASYNC) {
3168 xfs_bawrite(mp, bp);
3170 error = xfs_bwrite(mp, bp);
3176 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3177 xfs_iflush_abort(ip);
3179 * Unlocks the flush lock
3181 return XFS_ERROR(EFSCORRUPTED);
3183 cluster_corrupt_out:
3184 /* Corruption detected in the clustering loop. Invalidate the
3185 * inode buffer and shut down the filesystem.
3187 mutex_spinunlock(&ch->ch_lock, s);
3190 * Clean up the buffer. If it was B_DELWRI, just release it --
3191 * brelse can handle it with no problems. If not, shut down the
3192 * filesystem before releasing the buffer.
3194 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3198 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3202 * Just like incore_relse: if we have b_iodone functions,
3203 * mark the buffer as an error and call them. Otherwise
3204 * mark it as stale and brelse.
3206 if (XFS_BUF_IODONE_FUNC(bp)) {
3207 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3211 XFS_BUF_ERROR(bp,EIO);
3219 xfs_iflush_abort(iq);
3221 * Unlocks the flush lock
3223 return XFS_ERROR(EFSCORRUPTED);
3232 xfs_inode_log_item_t *iip;
3235 #ifdef XFS_TRANS_DEBUG
3240 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3241 ASSERT(valusema(&ip->i_flock) <= 0);
3242 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3243 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3250 * If the inode isn't dirty, then just release the inode
3251 * flush lock and do nothing.
3253 if ((ip->i_update_core == 0) &&
3254 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3259 /* set *dip = inode's place in the buffer */
3260 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3263 * Clear i_update_core before copying out the data.
3264 * This is for coordination with our timestamp updates
3265 * that don't hold the inode lock. They will always
3266 * update the timestamps BEFORE setting i_update_core,
3267 * so if we clear i_update_core after they set it we
3268 * are guaranteed to see their updates to the timestamps.
3269 * I believe that this depends on strongly ordered memory
3270 * semantics, but we have that. We use the SYNCHRONIZE
3271 * macro to make sure that the compiler does not reorder
3272 * the i_update_core access below the data copy below.
3274 ip->i_update_core = 0;
3278 * Make sure to get the latest atime from the Linux inode.
3280 xfs_synchronize_atime(ip);
3282 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3283 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3284 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3285 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3286 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3289 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3290 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3291 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3292 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3293 ip->i_ino, ip, ip->i_d.di_magic);
3296 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3298 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3299 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3300 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3301 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3302 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3306 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3308 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3309 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3310 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3311 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3312 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3313 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3318 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3319 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3320 XFS_RANDOM_IFLUSH_5)) {
3321 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3322 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3324 ip->i_d.di_nextents + ip->i_d.di_anextents,
3329 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3330 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3331 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3332 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3333 ip->i_ino, ip->i_d.di_forkoff, ip);
3337 * bump the flush iteration count, used to detect flushes which
3338 * postdate a log record during recovery.
3341 ip->i_d.di_flushiter++;
3344 * Copy the dirty parts of the inode into the on-disk
3345 * inode. We always copy out the core of the inode,
3346 * because if the inode is dirty at all the core must
3349 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3351 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3352 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3353 ip->i_d.di_flushiter = 0;
3356 * If this is really an old format inode and the superblock version
3357 * has not been updated to support only new format inodes, then
3358 * convert back to the old inode format. If the superblock version
3359 * has been updated, then make the conversion permanent.
3361 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3362 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3363 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3364 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3368 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3369 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3372 * The superblock version has already been bumped,
3373 * so just make the conversion to the new inode
3376 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3377 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3378 ip->i_d.di_onlink = 0;
3379 dip->di_core.di_onlink = 0;
3380 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3381 memset(&(dip->di_core.di_pad[0]), 0,
3382 sizeof(dip->di_core.di_pad));
3383 ASSERT(ip->i_d.di_projid == 0);
3387 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3391 if (XFS_IFORK_Q(ip)) {
3393 * The only error from xfs_iflush_fork is on the data fork.
3395 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3397 xfs_inobp_check(mp, bp);
3400 * We've recorded everything logged in the inode, so we'd
3401 * like to clear the ilf_fields bits so we don't log and
3402 * flush things unnecessarily. However, we can't stop
3403 * logging all this information until the data we've copied
3404 * into the disk buffer is written to disk. If we did we might
3405 * overwrite the copy of the inode in the log with all the
3406 * data after re-logging only part of it, and in the face of
3407 * a crash we wouldn't have all the data we need to recover.
3409 * What we do is move the bits to the ili_last_fields field.
3410 * When logging the inode, these bits are moved back to the
3411 * ilf_fields field. In the xfs_iflush_done() routine we
3412 * clear ili_last_fields, since we know that the information
3413 * those bits represent is permanently on disk. As long as
3414 * the flush completes before the inode is logged again, then
3415 * both ilf_fields and ili_last_fields will be cleared.
3417 * We can play with the ilf_fields bits here, because the inode
3418 * lock must be held exclusively in order to set bits there
3419 * and the flush lock protects the ili_last_fields bits.
3420 * Set ili_logged so the flush done
3421 * routine can tell whether or not to look in the AIL.
3422 * Also, store the current LSN of the inode so that we can tell
3423 * whether the item has moved in the AIL from xfs_iflush_done().
3424 * In order to read the lsn we need the AIL lock, because
3425 * it is a 64 bit value that cannot be read atomically.
3427 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3428 iip->ili_last_fields = iip->ili_format.ilf_fields;
3429 iip->ili_format.ilf_fields = 0;
3430 iip->ili_logged = 1;
3432 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3434 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3438 * Attach the function xfs_iflush_done to the inode's
3439 * buffer. This will remove the inode from the AIL
3440 * and unlock the inode's flush lock when the inode is
3441 * completely written to disk.
3443 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3444 xfs_iflush_done, (xfs_log_item_t *)iip);
3446 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3447 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3450 * We're flushing an inode which is not in the AIL and has
3451 * not been logged but has i_update_core set. For this
3452 * case we can use a B_DELWRI flush and immediately drop
3453 * the inode flush lock because we can avoid the whole
3454 * AIL state thing. It's OK to drop the flush lock now,
3455 * because we've already locked the buffer and to do anything
3456 * you really need both.
3459 ASSERT(iip->ili_logged == 0);
3460 ASSERT(iip->ili_last_fields == 0);
3461 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3469 return XFS_ERROR(EFSCORRUPTED);
3474 * Flush all inactive inodes in mp.
3484 XFS_MOUNT_ILOCK(mp);
3490 /* Make sure we skip markers inserted by sync */
3491 if (ip->i_mount == NULL) {
3496 vp = XFS_ITOV_NULL(ip);
3498 XFS_MOUNT_IUNLOCK(mp);
3499 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3503 ASSERT(vn_count(vp) == 0);
3506 } while (ip != mp->m_inodes);
3508 XFS_MOUNT_IUNLOCK(mp);
3512 * xfs_iaccess: check accessibility of inode for mode.
3521 mode_t orgmode = mode;
3522 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3524 if (mode & S_IWUSR) {
3525 umode_t imode = inode->i_mode;
3527 if (IS_RDONLY(inode) &&
3528 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3529 return XFS_ERROR(EROFS);
3531 if (IS_IMMUTABLE(inode))
3532 return XFS_ERROR(EACCES);
3536 * If there's an Access Control List it's used instead of
3539 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3540 return error ? XFS_ERROR(error) : 0;
3542 if (current_fsuid(cr) != ip->i_d.di_uid) {
3544 if (!in_group_p((gid_t)ip->i_d.di_gid))
3549 * If the DACs are ok we don't need any capability check.
3551 if ((ip->i_d.di_mode & mode) == mode)
3554 * Read/write DACs are always overridable.
3555 * Executable DACs are overridable if at least one exec bit is set.
3557 if (!(orgmode & S_IXUSR) ||
3558 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3559 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3562 if ((orgmode == S_IRUSR) ||
3563 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3564 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3567 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3569 return XFS_ERROR(EACCES);
3571 return XFS_ERROR(EACCES);
3575 * xfs_iroundup: round up argument to next power of two
3584 if ((v & (v - 1)) == 0)
3586 ASSERT((v & 0x80000000) == 0);
3587 if ((v & (v + 1)) == 0)
3589 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3593 if ((v & (v + 1)) == 0)
3600 #ifdef XFS_ILOCK_TRACE
3601 ktrace_t *xfs_ilock_trace_buf;
3604 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3606 ktrace_enter(ip->i_lock_trace,
3608 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3609 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3610 (void *)ra, /* caller of ilock */
3611 (void *)(unsigned long)current_cpu(),
3612 (void *)(unsigned long)current_pid(),
3613 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3618 * Return a pointer to the extent record at file index idx.
3622 xfs_ifork_t *ifp, /* inode fork pointer */
3623 xfs_extnum_t idx) /* index of target extent */
3626 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3627 return ifp->if_u1.if_ext_irec->er_extbuf;
3628 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3629 xfs_ext_irec_t *erp; /* irec pointer */
3630 int erp_idx = 0; /* irec index */
3631 xfs_extnum_t page_idx = idx; /* ext index in target list */
3633 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3634 return &erp->er_extbuf[page_idx];
3635 } else if (ifp->if_bytes) {
3636 return &ifp->if_u1.if_extents[idx];
3643 * Insert new item(s) into the extent records for incore inode
3644 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3648 xfs_ifork_t *ifp, /* inode fork pointer */
3649 xfs_extnum_t idx, /* starting index of new items */
3650 xfs_extnum_t count, /* number of inserted items */
3651 xfs_bmbt_irec_t *new) /* items to insert */
3653 xfs_bmbt_rec_t *ep; /* extent record pointer */
3654 xfs_extnum_t i; /* extent record index */
3656 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3657 xfs_iext_add(ifp, idx, count);
3658 for (i = idx; i < idx + count; i++, new++) {
3659 ep = xfs_iext_get_ext(ifp, i);
3660 xfs_bmbt_set_all(ep, new);
3665 * This is called when the amount of space required for incore file
3666 * extents needs to be increased. The ext_diff parameter stores the
3667 * number of new extents being added and the idx parameter contains
3668 * the extent index where the new extents will be added. If the new
3669 * extents are being appended, then we just need to (re)allocate and
3670 * initialize the space. Otherwise, if the new extents are being
3671 * inserted into the middle of the existing entries, a bit more work
3672 * is required to make room for the new extents to be inserted. The
3673 * caller is responsible for filling in the new extent entries upon
3678 xfs_ifork_t *ifp, /* inode fork pointer */
3679 xfs_extnum_t idx, /* index to begin adding exts */
3680 int ext_diff) /* nubmer of extents to add */
3682 int byte_diff; /* new bytes being added */
3683 int new_size; /* size of extents after adding */
3684 xfs_extnum_t nextents; /* number of extents in file */
3686 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3687 ASSERT((idx >= 0) && (idx <= nextents));
3688 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3689 new_size = ifp->if_bytes + byte_diff;
3691 * If the new number of extents (nextents + ext_diff)
3692 * fits inside the inode, then continue to use the inline
3695 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3696 if (idx < nextents) {
3697 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3698 &ifp->if_u2.if_inline_ext[idx],
3699 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3700 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3702 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3703 ifp->if_real_bytes = 0;
3704 ifp->if_lastex = nextents + ext_diff;
3707 * Otherwise use a linear (direct) extent list.
3708 * If the extents are currently inside the inode,
3709 * xfs_iext_realloc_direct will switch us from
3710 * inline to direct extent allocation mode.
3712 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3713 xfs_iext_realloc_direct(ifp, new_size);
3714 if (idx < nextents) {
3715 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3716 &ifp->if_u1.if_extents[idx],
3717 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3718 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3721 /* Indirection array */
3723 xfs_ext_irec_t *erp;
3727 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3728 if (ifp->if_flags & XFS_IFEXTIREC) {
3729 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3731 xfs_iext_irec_init(ifp);
3732 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3733 erp = ifp->if_u1.if_ext_irec;
3735 /* Extents fit in target extent page */
3736 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3737 if (page_idx < erp->er_extcount) {
3738 memmove(&erp->er_extbuf[page_idx + ext_diff],
3739 &erp->er_extbuf[page_idx],
3740 (erp->er_extcount - page_idx) *
3741 sizeof(xfs_bmbt_rec_t));
3742 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3744 erp->er_extcount += ext_diff;
3745 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3747 /* Insert a new extent page */
3749 xfs_iext_add_indirect_multi(ifp,
3750 erp_idx, page_idx, ext_diff);
3753 * If extent(s) are being appended to the last page in
3754 * the indirection array and the new extent(s) don't fit
3755 * in the page, then erp is NULL and erp_idx is set to
3756 * the next index needed in the indirection array.
3759 int count = ext_diff;
3762 erp = xfs_iext_irec_new(ifp, erp_idx);
3763 erp->er_extcount = count;
3764 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3771 ifp->if_bytes = new_size;
3775 * This is called when incore extents are being added to the indirection
3776 * array and the new extents do not fit in the target extent list. The
3777 * erp_idx parameter contains the irec index for the target extent list
3778 * in the indirection array, and the idx parameter contains the extent
3779 * index within the list. The number of extents being added is stored
3780 * in the count parameter.
3782 * |-------| |-------|
3783 * | | | | idx - number of extents before idx
3785 * | | | | count - number of extents being inserted at idx
3786 * |-------| |-------|
3787 * | count | | nex2 | nex2 - number of extents after idx + count
3788 * |-------| |-------|
3791 xfs_iext_add_indirect_multi(
3792 xfs_ifork_t *ifp, /* inode fork pointer */
3793 int erp_idx, /* target extent irec index */
3794 xfs_extnum_t idx, /* index within target list */
3795 int count) /* new extents being added */
3797 int byte_diff; /* new bytes being added */
3798 xfs_ext_irec_t *erp; /* pointer to irec entry */
3799 xfs_extnum_t ext_diff; /* number of extents to add */
3800 xfs_extnum_t ext_cnt; /* new extents still needed */
3801 xfs_extnum_t nex2; /* extents after idx + count */
3802 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3803 int nlists; /* number of irec's (lists) */
3805 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3806 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3807 nex2 = erp->er_extcount - idx;
3808 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3811 * Save second part of target extent list
3812 * (all extents past */
3814 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3815 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3816 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3817 erp->er_extcount -= nex2;
3818 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3819 memset(&erp->er_extbuf[idx], 0, byte_diff);
3823 * Add the new extents to the end of the target
3824 * list, then allocate new irec record(s) and
3825 * extent buffer(s) as needed to store the rest
3826 * of the new extents.
3829 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3831 erp->er_extcount += ext_diff;
3832 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3833 ext_cnt -= ext_diff;
3837 erp = xfs_iext_irec_new(ifp, erp_idx);
3838 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3839 erp->er_extcount = ext_diff;
3840 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3841 ext_cnt -= ext_diff;
3844 /* Add nex2 extents back to indirection array */
3846 xfs_extnum_t ext_avail;
3849 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3850 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3853 * If nex2 extents fit in the current page, append
3854 * nex2_ep after the new extents.
3856 if (nex2 <= ext_avail) {
3857 i = erp->er_extcount;
3860 * Otherwise, check if space is available in the
3863 else if ((erp_idx < nlists - 1) &&
3864 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3865 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3868 /* Create a hole for nex2 extents */
3869 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3870 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3873 * Final choice, create a new extent page for
3878 erp = xfs_iext_irec_new(ifp, erp_idx);
3880 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3881 kmem_free(nex2_ep, byte_diff);
3882 erp->er_extcount += nex2;
3883 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3888 * This is called when the amount of space required for incore file
3889 * extents needs to be decreased. The ext_diff parameter stores the
3890 * number of extents to be removed and the idx parameter contains
3891 * the extent index where the extents will be removed from.
3893 * If the amount of space needed has decreased below the linear
3894 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3895 * extent array. Otherwise, use kmem_realloc() to adjust the
3896 * size to what is needed.
3900 xfs_ifork_t *ifp, /* inode fork pointer */
3901 xfs_extnum_t idx, /* index to begin removing exts */
3902 int ext_diff) /* number of extents to remove */
3904 xfs_extnum_t nextents; /* number of extents in file */
3905 int new_size; /* size of extents after removal */
3907 ASSERT(ext_diff > 0);
3908 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3909 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3911 if (new_size == 0) {
3912 xfs_iext_destroy(ifp);
3913 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3914 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3915 } else if (ifp->if_real_bytes) {
3916 xfs_iext_remove_direct(ifp, idx, ext_diff);
3918 xfs_iext_remove_inline(ifp, idx, ext_diff);
3920 ifp->if_bytes = new_size;
3924 * This removes ext_diff extents from the inline buffer, beginning
3925 * at extent index idx.
3928 xfs_iext_remove_inline(
3929 xfs_ifork_t *ifp, /* inode fork pointer */
3930 xfs_extnum_t idx, /* index to begin removing exts */
3931 int ext_diff) /* number of extents to remove */
3933 int nextents; /* number of extents in file */
3935 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3936 ASSERT(idx < XFS_INLINE_EXTS);
3937 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3938 ASSERT(((nextents - ext_diff) > 0) &&
3939 (nextents - ext_diff) < XFS_INLINE_EXTS);
3941 if (idx + ext_diff < nextents) {
3942 memmove(&ifp->if_u2.if_inline_ext[idx],
3943 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3944 (nextents - (idx + ext_diff)) *
3945 sizeof(xfs_bmbt_rec_t));
3946 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3947 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3949 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3950 ext_diff * sizeof(xfs_bmbt_rec_t));
3955 * This removes ext_diff extents from a linear (direct) extent list,
3956 * beginning at extent index idx. If the extents are being removed
3957 * from the end of the list (ie. truncate) then we just need to re-
3958 * allocate the list to remove the extra space. Otherwise, if the
3959 * extents are being removed from the middle of the existing extent
3960 * entries, then we first need to move the extent records beginning
3961 * at idx + ext_diff up in the list to overwrite the records being
3962 * removed, then remove the extra space via kmem_realloc.
3965 xfs_iext_remove_direct(
3966 xfs_ifork_t *ifp, /* inode fork pointer */
3967 xfs_extnum_t idx, /* index to begin removing exts */
3968 int ext_diff) /* number of extents to remove */
3970 xfs_extnum_t nextents; /* number of extents in file */
3971 int new_size; /* size of extents after removal */
3973 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3974 new_size = ifp->if_bytes -
3975 (ext_diff * sizeof(xfs_bmbt_rec_t));
3976 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3978 if (new_size == 0) {
3979 xfs_iext_destroy(ifp);
3982 /* Move extents up in the list (if needed) */
3983 if (idx + ext_diff < nextents) {
3984 memmove(&ifp->if_u1.if_extents[idx],
3985 &ifp->if_u1.if_extents[idx + ext_diff],
3986 (nextents - (idx + ext_diff)) *
3987 sizeof(xfs_bmbt_rec_t));
3989 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3990 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3992 * Reallocate the direct extent list. If the extents
3993 * will fit inside the inode then xfs_iext_realloc_direct
3994 * will switch from direct to inline extent allocation
3997 xfs_iext_realloc_direct(ifp, new_size);
3998 ifp->if_bytes = new_size;
4002 * This is called when incore extents are being removed from the
4003 * indirection array and the extents being removed span multiple extent
4004 * buffers. The idx parameter contains the file extent index where we
4005 * want to begin removing extents, and the count parameter contains
4006 * how many extents need to be removed.
4008 * |-------| |-------|
4009 * | nex1 | | | nex1 - number of extents before idx
4010 * |-------| | count |
4011 * | | | | count - number of extents being removed at idx
4012 * | count | |-------|
4013 * | | | nex2 | nex2 - number of extents after idx + count
4014 * |-------| |-------|
4017 xfs_iext_remove_indirect(
4018 xfs_ifork_t *ifp, /* inode fork pointer */
4019 xfs_extnum_t idx, /* index to begin removing extents */
4020 int count) /* number of extents to remove */
4022 xfs_ext_irec_t *erp; /* indirection array pointer */
4023 int erp_idx = 0; /* indirection array index */
4024 xfs_extnum_t ext_cnt; /* extents left to remove */
4025 xfs_extnum_t ext_diff; /* extents to remove in current list */
4026 xfs_extnum_t nex1; /* number of extents before idx */
4027 xfs_extnum_t nex2; /* extents after idx + count */
4028 int nlists; /* entries in indirecton array */
4029 int page_idx = idx; /* index in target extent list */
4031 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4032 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4033 ASSERT(erp != NULL);
4034 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4038 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4039 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4041 * Check for deletion of entire list;
4042 * xfs_iext_irec_remove() updates extent offsets.
4044 if (ext_diff == erp->er_extcount) {
4045 xfs_iext_irec_remove(ifp, erp_idx);
4046 ext_cnt -= ext_diff;
4049 ASSERT(erp_idx < ifp->if_real_bytes /
4051 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4058 /* Move extents up (if needed) */
4060 memmove(&erp->er_extbuf[nex1],
4061 &erp->er_extbuf[nex1 + ext_diff],
4062 nex2 * sizeof(xfs_bmbt_rec_t));
4064 /* Zero out rest of page */
4065 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4066 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4067 /* Update remaining counters */
4068 erp->er_extcount -= ext_diff;
4069 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4070 ext_cnt -= ext_diff;
4075 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4076 xfs_iext_irec_compact(ifp);
4080 * Create, destroy, or resize a linear (direct) block of extents.
4083 xfs_iext_realloc_direct(
4084 xfs_ifork_t *ifp, /* inode fork pointer */
4085 int new_size) /* new size of extents */
4087 int rnew_size; /* real new size of extents */
4089 rnew_size = new_size;
4091 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4092 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4093 (new_size != ifp->if_real_bytes)));
4095 /* Free extent records */
4096 if (new_size == 0) {
4097 xfs_iext_destroy(ifp);
4099 /* Resize direct extent list and zero any new bytes */
4100 else if (ifp->if_real_bytes) {
4101 /* Check if extents will fit inside the inode */
4102 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4103 xfs_iext_direct_to_inline(ifp, new_size /
4104 (uint)sizeof(xfs_bmbt_rec_t));
4105 ifp->if_bytes = new_size;
4108 if ((new_size & (new_size - 1)) != 0) {
4109 rnew_size = xfs_iroundup(new_size);
4111 if (rnew_size != ifp->if_real_bytes) {
4112 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4113 kmem_realloc(ifp->if_u1.if_extents,
4118 if (rnew_size > ifp->if_real_bytes) {
4119 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4120 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4121 rnew_size - ifp->if_real_bytes);
4125 * Switch from the inline extent buffer to a direct
4126 * extent list. Be sure to include the inline extent
4127 * bytes in new_size.
4130 new_size += ifp->if_bytes;
4131 if ((new_size & (new_size - 1)) != 0) {
4132 rnew_size = xfs_iroundup(new_size);
4134 xfs_iext_inline_to_direct(ifp, rnew_size);
4136 ifp->if_real_bytes = rnew_size;
4137 ifp->if_bytes = new_size;
4141 * Switch from linear (direct) extent records to inline buffer.
4144 xfs_iext_direct_to_inline(
4145 xfs_ifork_t *ifp, /* inode fork pointer */
4146 xfs_extnum_t nextents) /* number of extents in file */
4148 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4149 ASSERT(nextents <= XFS_INLINE_EXTS);
4151 * The inline buffer was zeroed when we switched
4152 * from inline to direct extent allocation mode,
4153 * so we don't need to clear it here.
4155 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4156 nextents * sizeof(xfs_bmbt_rec_t));
4157 kmem_free(ifp->if_u1.if_extents, KM_SLEEP);
4158 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4159 ifp->if_real_bytes = 0;
4163 * Switch from inline buffer to linear (direct) extent records.
4164 * new_size should already be rounded up to the next power of 2
4165 * by the caller (when appropriate), so use new_size as it is.
4166 * However, since new_size may be rounded up, we can't update
4167 * if_bytes here. It is the caller's responsibility to update
4168 * if_bytes upon return.
4171 xfs_iext_inline_to_direct(
4172 xfs_ifork_t *ifp, /* inode fork pointer */
4173 int new_size) /* number of extents in file */
4175 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4176 kmem_alloc(new_size, KM_SLEEP);
4177 memset(ifp->if_u1.if_extents, 0, new_size);
4178 if (ifp->if_bytes) {
4179 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4181 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4182 sizeof(xfs_bmbt_rec_t));
4184 ifp->if_real_bytes = new_size;
4188 * Resize an extent indirection array to new_size bytes.
4191 xfs_iext_realloc_indirect(
4192 xfs_ifork_t *ifp, /* inode fork pointer */
4193 int new_size) /* new indirection array size */
4195 int nlists; /* number of irec's (ex lists) */
4196 int size; /* current indirection array size */
4198 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4199 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4200 size = nlists * sizeof(xfs_ext_irec_t);
4201 ASSERT(ifp->if_real_bytes);
4202 ASSERT((new_size >= 0) && (new_size != size));
4203 if (new_size == 0) {
4204 xfs_iext_destroy(ifp);
4206 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4207 kmem_realloc(ifp->if_u1.if_ext_irec,
4208 new_size, size, KM_SLEEP);
4213 * Switch from indirection array to linear (direct) extent allocations.
4216 xfs_iext_indirect_to_direct(
4217 xfs_ifork_t *ifp) /* inode fork pointer */
4219 xfs_bmbt_rec_t *ep; /* extent record pointer */
4220 xfs_extnum_t nextents; /* number of extents in file */
4221 int size; /* size of file extents */
4223 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4224 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4225 ASSERT(nextents <= XFS_LINEAR_EXTS);
4226 size = nextents * sizeof(xfs_bmbt_rec_t);
4228 xfs_iext_irec_compact_full(ifp);
4229 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4231 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4232 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4233 ifp->if_flags &= ~XFS_IFEXTIREC;
4234 ifp->if_u1.if_extents = ep;
4235 ifp->if_bytes = size;
4236 if (nextents < XFS_LINEAR_EXTS) {
4237 xfs_iext_realloc_direct(ifp, size);
4242 * Free incore file extents.
4246 xfs_ifork_t *ifp) /* inode fork pointer */
4248 if (ifp->if_flags & XFS_IFEXTIREC) {
4252 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4253 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4254 xfs_iext_irec_remove(ifp, erp_idx);
4256 ifp->if_flags &= ~XFS_IFEXTIREC;
4257 } else if (ifp->if_real_bytes) {
4258 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4259 } else if (ifp->if_bytes) {
4260 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4261 sizeof(xfs_bmbt_rec_t));
4263 ifp->if_u1.if_extents = NULL;
4264 ifp->if_real_bytes = 0;
4269 * Return a pointer to the extent record for file system block bno.
4271 xfs_bmbt_rec_t * /* pointer to found extent record */
4272 xfs_iext_bno_to_ext(
4273 xfs_ifork_t *ifp, /* inode fork pointer */
4274 xfs_fileoff_t bno, /* block number to search for */
4275 xfs_extnum_t *idxp) /* index of target extent */
4277 xfs_bmbt_rec_t *base; /* pointer to first extent */
4278 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4279 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4280 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4281 int high; /* upper boundry in search */
4282 xfs_extnum_t idx = 0; /* index of target extent */
4283 int low; /* lower boundry in search */
4284 xfs_extnum_t nextents; /* number of file extents */
4285 xfs_fileoff_t startoff = 0; /* start offset of extent */
4287 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4288 if (nextents == 0) {
4293 if (ifp->if_flags & XFS_IFEXTIREC) {
4294 /* Find target extent list */
4296 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4297 base = erp->er_extbuf;
4298 high = erp->er_extcount - 1;
4300 base = ifp->if_u1.if_extents;
4301 high = nextents - 1;
4303 /* Binary search extent records */
4304 while (low <= high) {
4305 idx = (low + high) >> 1;
4307 startoff = xfs_bmbt_get_startoff(ep);
4308 blockcount = xfs_bmbt_get_blockcount(ep);
4309 if (bno < startoff) {
4311 } else if (bno >= startoff + blockcount) {
4314 /* Convert back to file-based extent index */
4315 if (ifp->if_flags & XFS_IFEXTIREC) {
4316 idx += erp->er_extoff;
4322 /* Convert back to file-based extent index */
4323 if (ifp->if_flags & XFS_IFEXTIREC) {
4324 idx += erp->er_extoff;
4326 if (bno >= startoff + blockcount) {
4327 if (++idx == nextents) {
4330 ep = xfs_iext_get_ext(ifp, idx);
4338 * Return a pointer to the indirection array entry containing the
4339 * extent record for filesystem block bno. Store the index of the
4340 * target irec in *erp_idxp.
4342 xfs_ext_irec_t * /* pointer to found extent record */
4343 xfs_iext_bno_to_irec(
4344 xfs_ifork_t *ifp, /* inode fork pointer */
4345 xfs_fileoff_t bno, /* block number to search for */
4346 int *erp_idxp) /* irec index of target ext list */
4348 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4349 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4350 int erp_idx; /* indirection array index */
4351 int nlists; /* number of extent irec's (lists) */
4352 int high; /* binary search upper limit */
4353 int low; /* binary search lower limit */
4355 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4356 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4360 while (low <= high) {
4361 erp_idx = (low + high) >> 1;
4362 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4363 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4364 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4366 } else if (erp_next && bno >=
4367 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4373 *erp_idxp = erp_idx;
4378 * Return a pointer to the indirection array entry containing the
4379 * extent record at file extent index *idxp. Store the index of the
4380 * target irec in *erp_idxp and store the page index of the target
4381 * extent record in *idxp.
4384 xfs_iext_idx_to_irec(
4385 xfs_ifork_t *ifp, /* inode fork pointer */
4386 xfs_extnum_t *idxp, /* extent index (file -> page) */
4387 int *erp_idxp, /* pointer to target irec */
4388 int realloc) /* new bytes were just added */
4390 xfs_ext_irec_t *prev; /* pointer to previous irec */
4391 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4392 int erp_idx; /* indirection array index */
4393 int nlists; /* number of irec's (ex lists) */
4394 int high; /* binary search upper limit */
4395 int low; /* binary search lower limit */
4396 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4398 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4399 ASSERT(page_idx >= 0 && page_idx <=
4400 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4401 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4406 /* Binary search extent irec's */
4407 while (low <= high) {
4408 erp_idx = (low + high) >> 1;
4409 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4410 prev = erp_idx > 0 ? erp - 1 : NULL;
4411 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4412 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4414 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4415 (page_idx == erp->er_extoff + erp->er_extcount &&
4418 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4419 erp->er_extcount == XFS_LINEAR_EXTS) {
4423 erp = erp_idx < nlists ? erp + 1 : NULL;
4426 page_idx -= erp->er_extoff;
4431 *erp_idxp = erp_idx;
4436 * Allocate and initialize an indirection array once the space needed
4437 * for incore extents increases above XFS_IEXT_BUFSZ.
4441 xfs_ifork_t *ifp) /* inode fork pointer */
4443 xfs_ext_irec_t *erp; /* indirection array pointer */
4444 xfs_extnum_t nextents; /* number of extents in file */
4446 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4447 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4448 ASSERT(nextents <= XFS_LINEAR_EXTS);
4450 erp = (xfs_ext_irec_t *)
4451 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4453 if (nextents == 0) {
4454 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4455 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4456 } else if (!ifp->if_real_bytes) {
4457 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4458 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4459 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4461 erp->er_extbuf = ifp->if_u1.if_extents;
4462 erp->er_extcount = nextents;
4465 ifp->if_flags |= XFS_IFEXTIREC;
4466 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4467 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4468 ifp->if_u1.if_ext_irec = erp;
4474 * Allocate and initialize a new entry in the indirection array.
4478 xfs_ifork_t *ifp, /* inode fork pointer */
4479 int erp_idx) /* index for new irec */
4481 xfs_ext_irec_t *erp; /* indirection array pointer */
4482 int i; /* loop counter */
4483 int nlists; /* number of irec's (ex lists) */
4485 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4486 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4488 /* Resize indirection array */
4489 xfs_iext_realloc_indirect(ifp, ++nlists *
4490 sizeof(xfs_ext_irec_t));
4492 * Move records down in the array so the
4493 * new page can use erp_idx.
4495 erp = ifp->if_u1.if_ext_irec;
4496 for (i = nlists - 1; i > erp_idx; i--) {
4497 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4499 ASSERT(i == erp_idx);
4501 /* Initialize new extent record */
4502 erp = ifp->if_u1.if_ext_irec;
4503 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4504 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4505 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4506 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4507 erp[erp_idx].er_extcount = 0;
4508 erp[erp_idx].er_extoff = erp_idx > 0 ?
4509 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4510 return (&erp[erp_idx]);
4514 * Remove a record from the indirection array.
4517 xfs_iext_irec_remove(
4518 xfs_ifork_t *ifp, /* inode fork pointer */
4519 int erp_idx) /* irec index to remove */
4521 xfs_ext_irec_t *erp; /* indirection array pointer */
4522 int i; /* loop counter */
4523 int nlists; /* number of irec's (ex lists) */
4525 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4526 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4527 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4528 if (erp->er_extbuf) {
4529 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4531 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4533 /* Compact extent records */
4534 erp = ifp->if_u1.if_ext_irec;
4535 for (i = erp_idx; i < nlists - 1; i++) {
4536 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4539 * Manually free the last extent record from the indirection
4540 * array. A call to xfs_iext_realloc_indirect() with a size
4541 * of zero would result in a call to xfs_iext_destroy() which
4542 * would in turn call this function again, creating a nasty
4546 xfs_iext_realloc_indirect(ifp,
4547 nlists * sizeof(xfs_ext_irec_t));
4549 kmem_free(ifp->if_u1.if_ext_irec,
4550 sizeof(xfs_ext_irec_t));
4552 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4556 * This is called to clean up large amounts of unused memory allocated
4557 * by the indirection array. Before compacting anything though, verify
4558 * that the indirection array is still needed and switch back to the
4559 * linear extent list (or even the inline buffer) if possible. The
4560 * compaction policy is as follows:
4562 * Full Compaction: Extents fit into a single page (or inline buffer)
4563 * Full Compaction: Extents occupy less than 10% of allocated space
4564 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4565 * No Compaction: Extents occupy at least 50% of allocated space
4568 xfs_iext_irec_compact(
4569 xfs_ifork_t *ifp) /* inode fork pointer */
4571 xfs_extnum_t nextents; /* number of extents in file */
4572 int nlists; /* number of irec's (ex lists) */
4574 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4575 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4576 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4578 if (nextents == 0) {
4579 xfs_iext_destroy(ifp);
4580 } else if (nextents <= XFS_INLINE_EXTS) {
4581 xfs_iext_indirect_to_direct(ifp);
4582 xfs_iext_direct_to_inline(ifp, nextents);
4583 } else if (nextents <= XFS_LINEAR_EXTS) {
4584 xfs_iext_indirect_to_direct(ifp);
4585 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4586 xfs_iext_irec_compact_full(ifp);
4587 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4588 xfs_iext_irec_compact_pages(ifp);
4593 * Combine extents from neighboring extent pages.
4596 xfs_iext_irec_compact_pages(
4597 xfs_ifork_t *ifp) /* inode fork pointer */
4599 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4600 int erp_idx = 0; /* indirection array index */
4601 int nlists; /* number of irec's (ex lists) */
4603 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4604 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4605 while (erp_idx < nlists - 1) {
4606 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4608 if (erp_next->er_extcount <=
4609 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4610 memmove(&erp->er_extbuf[erp->er_extcount],
4611 erp_next->er_extbuf, erp_next->er_extcount *
4612 sizeof(xfs_bmbt_rec_t));
4613 erp->er_extcount += erp_next->er_extcount;
4615 * Free page before removing extent record
4616 * so er_extoffs don't get modified in
4617 * xfs_iext_irec_remove.
4619 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4620 erp_next->er_extbuf = NULL;
4621 xfs_iext_irec_remove(ifp, erp_idx + 1);
4622 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4630 * Fully compact the extent records managed by the indirection array.
4633 xfs_iext_irec_compact_full(
4634 xfs_ifork_t *ifp) /* inode fork pointer */
4636 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4637 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4638 int erp_idx = 0; /* extent irec index */
4639 int ext_avail; /* empty entries in ex list */
4640 int ext_diff; /* number of exts to add */
4641 int nlists; /* number of irec's (ex lists) */
4643 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4644 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4645 erp = ifp->if_u1.if_ext_irec;
4646 ep = &erp->er_extbuf[erp->er_extcount];
4648 ep_next = erp_next->er_extbuf;
4649 while (erp_idx < nlists - 1) {
4650 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4651 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4652 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4653 erp->er_extcount += ext_diff;
4654 erp_next->er_extcount -= ext_diff;
4655 /* Remove next page */
4656 if (erp_next->er_extcount == 0) {
4658 * Free page before removing extent record
4659 * so er_extoffs don't get modified in
4660 * xfs_iext_irec_remove.
4662 kmem_free(erp_next->er_extbuf,
4663 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4664 erp_next->er_extbuf = NULL;
4665 xfs_iext_irec_remove(ifp, erp_idx + 1);
4666 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4667 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4668 /* Update next page */
4670 /* Move rest of page up to become next new page */
4671 memmove(erp_next->er_extbuf, ep_next,
4672 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4673 ep_next = erp_next->er_extbuf;
4674 memset(&ep_next[erp_next->er_extcount], 0,
4675 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4676 sizeof(xfs_bmbt_rec_t));
4678 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4680 if (erp_idx < nlists)
4681 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4685 ep = &erp->er_extbuf[erp->er_extcount];
4687 ep_next = erp_next->er_extbuf;
4692 * This is called to update the er_extoff field in the indirection
4693 * array when extents have been added or removed from one of the
4694 * extent lists. erp_idx contains the irec index to begin updating
4695 * at and ext_diff contains the number of extents that were added
4699 xfs_iext_irec_update_extoffs(
4700 xfs_ifork_t *ifp, /* inode fork pointer */
4701 int erp_idx, /* irec index to update */
4702 int ext_diff) /* number of new extents */
4704 int i; /* loop counter */
4705 int nlists; /* number of irec's (ex lists */
4707 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4708 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4709 for (i = erp_idx; i < nlists; i++) {
4710 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;