2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
53 #include <linux/vserver/xid.h>
55 kmem_zone_t *xfs_ifork_zone;
56 kmem_zone_t *xfs_inode_zone;
57 kmem_zone_t *xfs_chashlist_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
73 * Make sure that the extents in the given memory buffer
88 for (i = 0; i < nrecs; i++) {
89 ep = xfs_iext_get_ext(ifp, i);
90 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
91 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
93 xfs_bmbt_disk_get_all(&rec, &irec);
95 xfs_bmbt_get_all(&rec, &irec);
96 if (fmt == XFS_EXTFMT_NOSTATE)
97 ASSERT(irec.br_state == XFS_EXT_NORM);
101 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
118 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120 for (i = 0; i < j; i++) {
121 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
122 i * mp->m_sb.sb_inodesize);
123 if (!dip->di_next_unlinked) {
124 xfs_fs_cmn_err(CE_ALERT, mp,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 ASSERT(dip->di_next_unlinked);
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
162 * Call the space management code to find the location of the
166 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error, mp->m_fsname);
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
179 if ((imap.im_blkno + imap.im_len) >
180 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
183 "of the file system %s. Returning EINVAL.",
184 (unsigned long long)imap.im_blkno,
185 imap.im_len, mp->m_fsname);
186 return XFS_ERROR(EINVAL);
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
193 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
194 (int)imap.im_len, XFS_BUF_LOCK, &bp);
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error, mp->m_fsname);
202 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
204 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
206 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
207 XFS_RANDOM_ITOBP_INOTOBP))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
209 xfs_trans_brelse(tp, bp);
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
213 return XFS_ERROR(EFSCORRUPTED);
216 xfs_inobp_check(mp, bp);
219 * Set *dipp to point to the on-disk inode in the buffer.
221 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
223 *offset = imap.im_boffset;
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
264 if (ip->i_blkno == (xfs_daddr_t)0) {
266 * Call the space management code to find the location of the
270 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
271 XFS_IMAP_LOOKUP | imap_flags)))
275 * If the inode number maps to a block outside the bounds
276 * of the file system then return NULL rather than calling
277 * read_buf and panicing when we get an error from the
280 if ((imap.im_blkno + imap.im_len) >
281 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
283 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
284 "(imap.im_blkno (0x%llx) "
285 "+ imap.im_len (0x%llx)) > "
286 " XFS_FSB_TO_BB(mp, "
287 "mp->m_sb.sb_dblocks) (0x%llx)",
288 (unsigned long long) imap.im_blkno,
289 (unsigned long long) imap.im_len,
290 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
292 return XFS_ERROR(EINVAL);
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
311 ASSERT(bno == 0 || bno == imap.im_blkno);
314 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
315 * default to just a read_buf() call.
317 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
318 (int)imap.im_len, XFS_BUF_LOCK, &bp);
321 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
322 "xfs_trans_read_buf() returned error %d, "
323 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
324 error, (unsigned long long) imap.im_blkno,
325 (unsigned long long) imap.im_len);
331 * Validate the magic number and version of every inode in the buffer
332 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
333 * No validation is done here in userspace (xfs_repair).
335 #if !defined(__KERNEL__)
338 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
339 #else /* usual case */
343 for (i = 0; i < ni; i++) {
347 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
348 (i << mp->m_sb.sb_inodelog));
349 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
350 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
351 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
352 XFS_ERRTAG_ITOBP_INOTOBP,
353 XFS_RANDOM_ITOBP_INOTOBP))) {
354 if (imap_flags & XFS_IMAP_BULKSTAT) {
355 xfs_trans_brelse(tp, bp);
356 return XFS_ERROR(EINVAL);
360 "Device %s - bad inode magic/vsn "
361 "daddr %lld #%d (magic=%x)",
362 XFS_BUFTARG_NAME(mp->m_ddev_targp),
363 (unsigned long long)imap.im_blkno, i,
364 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
366 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
368 xfs_trans_brelse(tp, bp);
369 return XFS_ERROR(EFSCORRUPTED);
373 xfs_inobp_check(mp, bp);
376 * Mark the buffer as an inode buffer now that it looks good
378 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
381 * Set *dipp to point to the on-disk inode in the buffer.
383 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
389 * Move inode type and inode format specific information from the
390 * on-disk inode to the in-core inode. For fifos, devs, and sockets
391 * this means set if_rdev to the proper value. For files, directories,
392 * and symlinks this means to bring in the in-line data or extent
393 * pointers. For a file in B-tree format, only the root is immediately
394 * brought in-core. The rest will be in-lined in if_extents when it
395 * is first referenced (see xfs_iread_extents()).
402 xfs_attr_shortform_t *atp;
406 ip->i_df.if_ext_max =
407 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
411 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
412 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
413 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
414 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
415 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
416 (unsigned long long)ip->i_ino,
417 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
418 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
420 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
421 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
423 return XFS_ERROR(EFSCORRUPTED);
426 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
427 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
428 "corrupt dinode %Lu, forkoff = 0x%x.",
429 (unsigned long long)ip->i_ino,
430 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
431 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
433 return XFS_ERROR(EFSCORRUPTED);
436 switch (ip->i_d.di_mode & S_IFMT) {
441 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
442 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
444 return XFS_ERROR(EFSCORRUPTED);
447 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
453 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
454 case XFS_DINODE_FMT_LOCAL:
456 * no local regular files yet
458 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
459 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
461 "(local format for regular file).",
462 (unsigned long long) ip->i_ino);
463 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
466 return XFS_ERROR(EFSCORRUPTED);
469 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
470 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
471 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
473 "(bad size %Ld for local inode).",
474 (unsigned long long) ip->i_ino,
475 (long long) di_size);
476 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
479 return XFS_ERROR(EFSCORRUPTED);
483 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
485 case XFS_DINODE_FMT_EXTENTS:
486 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
488 case XFS_DINODE_FMT_BTREE:
489 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
492 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
494 return XFS_ERROR(EFSCORRUPTED);
499 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
500 return XFS_ERROR(EFSCORRUPTED);
505 if (!XFS_DFORK_Q(dip))
507 ASSERT(ip->i_afp == NULL);
508 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
509 ip->i_afp->if_ext_max =
510 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
511 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
512 case XFS_DINODE_FMT_LOCAL:
513 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
514 size = be16_to_cpu(atp->hdr.totsize);
515 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
517 case XFS_DINODE_FMT_EXTENTS:
518 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
520 case XFS_DINODE_FMT_BTREE:
521 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
524 error = XFS_ERROR(EFSCORRUPTED);
528 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
530 xfs_idestroy_fork(ip, XFS_DATA_FORK);
536 * The file is in-lined in the on-disk inode.
537 * If it fits into if_inline_data, then copy
538 * it there, otherwise allocate a buffer for it
539 * and copy the data there. Either way, set
540 * if_data to point at the data.
541 * If we allocate a buffer for the data, make
542 * sure that its size is a multiple of 4 and
543 * record the real size in i_real_bytes.
556 * If the size is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
560 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
561 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
563 "(bad size %d for local fork, size = %d).",
564 (unsigned long long) ip->i_ino, size,
565 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
566 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
568 return XFS_ERROR(EFSCORRUPTED);
570 ifp = XFS_IFORK_PTR(ip, whichfork);
573 ifp->if_u1.if_data = NULL;
574 else if (size <= sizeof(ifp->if_u2.if_inline_data))
575 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
577 real_size = roundup(size, 4);
578 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
580 ifp->if_bytes = size;
581 ifp->if_real_bytes = real_size;
583 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
584 ifp->if_flags &= ~XFS_IFEXTENTS;
585 ifp->if_flags |= XFS_IFINLINE;
590 * The file consists of a set of extents all
591 * of which fit into the on-disk inode.
592 * If there are few enough extents to fit into
593 * the if_inline_ext, then copy them there.
594 * Otherwise allocate a buffer for them and copy
595 * them into it. Either way, set if_extents
596 * to point at the extents.
604 xfs_bmbt_rec_t *ep, *dp;
610 ifp = XFS_IFORK_PTR(ip, whichfork);
611 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
612 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
615 * If the number of extents is unreasonable, then something
616 * is wrong and we just bail out rather than crash in
617 * kmem_alloc() or memcpy() below.
619 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
620 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
621 "corrupt inode %Lu ((a)extents = %d).",
622 (unsigned long long) ip->i_ino, nex);
623 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
625 return XFS_ERROR(EFSCORRUPTED);
628 ifp->if_real_bytes = 0;
630 ifp->if_u1.if_extents = NULL;
631 else if (nex <= XFS_INLINE_EXTS)
632 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
634 xfs_iext_add(ifp, 0, nex);
636 ifp->if_bytes = size;
638 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
639 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
640 for (i = 0; i < nex; i++, dp++) {
641 ep = xfs_iext_get_ext(ifp, i);
642 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
644 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
647 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
649 if (whichfork != XFS_DATA_FORK ||
650 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
651 if (unlikely(xfs_check_nostate_extents(
653 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
656 return XFS_ERROR(EFSCORRUPTED);
659 ifp->if_flags |= XFS_IFEXTENTS;
664 * The file has too many extents to fit into
665 * the inode, so they are in B-tree format.
666 * Allocate a buffer for the root of the B-tree
667 * and copy the root into it. The i_extents
668 * field will remain NULL until all of the
669 * extents are read in (when they are needed).
677 xfs_bmdr_block_t *dfp;
683 ifp = XFS_IFORK_PTR(ip, whichfork);
684 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
685 size = XFS_BMAP_BROOT_SPACE(dfp);
686 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
689 * blow out if -- fork has less extents than can fit in
690 * fork (fork shouldn't be a btree format), root btree
691 * block has more records than can fit into the fork,
692 * or the number of extents is greater than the number of
695 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
696 || XFS_BMDR_SPACE_CALC(nrecs) >
697 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
698 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
699 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
700 "corrupt inode %Lu (btree).",
701 (unsigned long long) ip->i_ino);
702 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
704 return XFS_ERROR(EFSCORRUPTED);
707 ifp->if_broot_bytes = size;
708 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
709 ASSERT(ifp->if_broot != NULL);
711 * Copy and convert from the on-disk structure
712 * to the in-memory structure.
714 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
715 ifp->if_broot, size);
716 ifp->if_flags &= ~XFS_IFEXTENTS;
717 ifp->if_flags |= XFS_IFBROOT;
723 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
726 * buf = on-disk representation
727 * dip = native representation
728 * dir = direction - +ve -> disk to native
729 * -ve -> native to disk
732 xfs_xlate_dinode_core(
734 xfs_dinode_core_t *dip,
737 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
738 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
739 xfs_arch_t arch = ARCH_CONVERT;
740 uint32_t uid = 0, gid = 0;
746 xid = mem_core->di_xid;
747 /* FIXME: supposed to use superblock flag */
748 uid = XIDINO_UID(1, mem_core->di_uid, xid);
749 gid = XIDINO_GID(1, mem_core->di_gid, xid);
750 xid = XIDINO_XID(1, xid);
753 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
754 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
755 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
756 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
757 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
758 INT_XLATE(buf_core->di_uid, uid, dir, arch);
759 INT_XLATE(buf_core->di_gid, gid, dir, arch);
760 INT_XLATE(buf_core->di_xid, xid, dir, arch);
761 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
762 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
765 /* FIXME: supposed to use superblock flag */
766 mem_core->di_uid = INOXID_UID(1, uid, gid);
767 mem_core->di_gid = INOXID_GID(1, uid, gid);
768 mem_core->di_xid = INOXID_XID(1, uid, gid, xid);
769 memcpy(mem_core->di_pad, buf_core->di_pad,
770 sizeof(buf_core->di_pad));
772 memcpy(buf_core->di_pad, mem_core->di_pad,
773 sizeof(buf_core->di_pad));
776 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
778 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
780 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
782 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
784 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
786 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
788 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
790 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
791 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
792 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
793 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
794 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
795 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
796 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
797 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
798 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
799 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
800 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
809 if (di_flags & XFS_DIFLAG_ANY) {
810 if (di_flags & XFS_DIFLAG_REALTIME)
811 flags |= XFS_XFLAG_REALTIME;
812 if (di_flags & XFS_DIFLAG_PREALLOC)
813 flags |= XFS_XFLAG_PREALLOC;
814 if (di_flags & XFS_DIFLAG_IMMUTABLE)
815 flags |= XFS_XFLAG_IMMUTABLE;
816 if (di_flags & XFS_DIFLAG_IUNLINK)
817 flags |= XFS_XFLAG_IUNLINK;
818 if (di_flags & XFS_DIFLAG_BARRIER)
819 flags |= XFS_XFLAG_BARRIER;
820 if (di_flags & XFS_DIFLAG_APPEND)
821 flags |= XFS_XFLAG_APPEND;
822 if (di_flags & XFS_DIFLAG_SYNC)
823 flags |= XFS_XFLAG_SYNC;
824 if (di_flags & XFS_DIFLAG_NOATIME)
825 flags |= XFS_XFLAG_NOATIME;
826 if (di_flags & XFS_DIFLAG_NODUMP)
827 flags |= XFS_XFLAG_NODUMP;
828 if (di_flags & XFS_DIFLAG_RTINHERIT)
829 flags |= XFS_XFLAG_RTINHERIT;
830 if (di_flags & XFS_DIFLAG_PROJINHERIT)
831 flags |= XFS_XFLAG_PROJINHERIT;
832 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
833 flags |= XFS_XFLAG_NOSYMLINKS;
834 if (di_flags & XFS_DIFLAG_EXTSIZE)
835 flags |= XFS_XFLAG_EXTSIZE;
836 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
837 flags |= XFS_XFLAG_EXTSZINHERIT;
838 if (di_flags & XFS_DIFLAG_NODEFRAG)
839 flags |= XFS_XFLAG_NODEFRAG;
849 xfs_dinode_core_t *dic = &ip->i_d;
851 return _xfs_dic2xflags(dic->di_flags) |
852 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
857 xfs_dinode_core_t *dic)
859 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
860 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
864 * Given a mount structure and an inode number, return a pointer
865 * to a newly allocated in-core inode corresponding to the given
868 * Initialize the inode's attributes and extent pointers if it
869 * already has them (it will not if the inode has no links).
884 ASSERT(xfs_inode_zone != NULL);
886 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
891 * Get pointer's to the on-disk inode and the buffer containing it.
892 * If the inode number refers to a block outside the file system
893 * then xfs_itobp() will return NULL. In this case we should
894 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
895 * know that this is a new incore inode.
897 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
899 kmem_zone_free(xfs_inode_zone, ip);
904 * Initialize inode's trace buffers.
905 * Do this before xfs_iformat in case it adds entries.
907 #ifdef XFS_BMAP_TRACE
908 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
910 #ifdef XFS_BMBT_TRACE
911 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
914 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
916 #ifdef XFS_ILOCK_TRACE
917 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
919 #ifdef XFS_DIR2_TRACE
920 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
924 * If we got something that isn't an inode it means someone
925 * (nfs or dmi) has a stale handle.
927 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
928 kmem_zone_free(xfs_inode_zone, ip);
929 xfs_trans_brelse(tp, bp);
931 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
932 "dip->di_core.di_magic (0x%x) != "
933 "XFS_DINODE_MAGIC (0x%x)",
934 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
937 return XFS_ERROR(EINVAL);
941 * If the on-disk inode is already linked to a directory
942 * entry, copy all of the inode into the in-core inode.
943 * xfs_iformat() handles copying in the inode format
944 * specific information.
945 * Otherwise, just get the truly permanent information.
947 if (dip->di_core.di_mode) {
948 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
950 error = xfs_iformat(ip, dip);
952 kmem_zone_free(xfs_inode_zone, ip);
953 xfs_trans_brelse(tp, bp);
955 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
956 "xfs_iformat() returned error %d",
962 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
963 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
964 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
965 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
967 * Make sure to pull in the mode here as well in
968 * case the inode is released without being used.
969 * This ensures that xfs_inactive() will see that
970 * the inode is already free and not try to mess
971 * with the uninitialized part of it.
975 * Initialize the per-fork minima and maxima for a new
976 * inode here. xfs_iformat will do it for old inodes.
978 ip->i_df.if_ext_max =
979 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
982 INIT_LIST_HEAD(&ip->i_reclaim);
985 * The inode format changed when we moved the link count and
986 * made it 32 bits long. If this is an old format inode,
987 * convert it in memory to look like a new one. If it gets
988 * flushed to disk we will convert back before flushing or
989 * logging it. We zero out the new projid field and the old link
990 * count field. We'll handle clearing the pad field (the remains
991 * of the old uuid field) when we actually convert the inode to
992 * the new format. We don't change the version number so that we
993 * can distinguish this from a real new format inode.
995 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
996 ip->i_d.di_nlink = ip->i_d.di_onlink;
997 ip->i_d.di_onlink = 0;
998 ip->i_d.di_projid = 0;
1001 ip->i_delayed_blks = 0;
1004 * Mark the buffer containing the inode as something to keep
1005 * around for a while. This helps to keep recently accessed
1006 * meta-data in-core longer.
1008 XFS_BUF_SET_REF(bp, XFS_INO_REF);
1011 * Use xfs_trans_brelse() to release the buffer containing the
1012 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1013 * in xfs_itobp() above. If tp is NULL, this is just a normal
1014 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1015 * will only release the buffer if it is not dirty within the
1016 * transaction. It will be OK to release the buffer in this case,
1017 * because inodes on disk are never destroyed and we will be
1018 * locking the new in-core inode before putting it in the hash
1019 * table where other processes can find it. Thus we don't have
1020 * to worry about the inode being changed just because we released
1023 xfs_trans_brelse(tp, bp);
1029 * Read in extents from a btree-format inode.
1030 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1040 xfs_extnum_t nextents;
1043 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1044 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1046 return XFS_ERROR(EFSCORRUPTED);
1048 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1049 size = nextents * sizeof(xfs_bmbt_rec_t);
1050 ifp = XFS_IFORK_PTR(ip, whichfork);
1053 * We know that the size is valid (it's checked in iformat_btree)
1055 ifp->if_lastex = NULLEXTNUM;
1056 ifp->if_bytes = ifp->if_real_bytes = 0;
1057 ifp->if_flags |= XFS_IFEXTENTS;
1058 xfs_iext_add(ifp, 0, nextents);
1059 error = xfs_bmap_read_extents(tp, ip, whichfork);
1061 xfs_iext_destroy(ifp);
1062 ifp->if_flags &= ~XFS_IFEXTENTS;
1065 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1070 * Allocate an inode on disk and return a copy of its in-core version.
1071 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1072 * appropriately within the inode. The uid and gid for the inode are
1073 * set according to the contents of the given cred structure.
1075 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1076 * has a free inode available, call xfs_iget()
1077 * to obtain the in-core version of the allocated inode. Finally,
1078 * fill in the inode and log its initial contents. In this case,
1079 * ialloc_context would be set to NULL and call_again set to false.
1081 * If xfs_dialloc() does not have an available inode,
1082 * it will replenish its supply by doing an allocation. Since we can
1083 * only do one allocation within a transaction without deadlocks, we
1084 * must commit the current transaction before returning the inode itself.
1085 * In this case, therefore, we will set call_again to true and return.
1086 * The caller should then commit the current transaction, start a new
1087 * transaction, and call xfs_ialloc() again to actually get the inode.
1089 * To ensure that some other process does not grab the inode that
1090 * was allocated during the first call to xfs_ialloc(), this routine
1091 * also returns the [locked] bp pointing to the head of the freelist
1092 * as ialloc_context. The caller should hold this buffer across
1093 * the commit and pass it back into this routine on the second call.
1105 xfs_buf_t **ialloc_context,
1106 boolean_t *call_again,
1116 * Call the space management code to pick
1117 * the on-disk inode to be allocated.
1119 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1120 ialloc_context, call_again, &ino);
1124 if (*call_again || ino == NULLFSINO) {
1128 ASSERT(*ialloc_context == NULL);
1131 * Get the in-core inode with the lock held exclusively.
1132 * This is because we're setting fields here we need
1133 * to prevent others from looking at until we're done.
1135 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1136 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1143 ip->i_d.di_mode = (__uint16_t)mode;
1144 ip->i_d.di_onlink = 0;
1145 ip->i_d.di_nlink = nlink;
1146 ASSERT(ip->i_d.di_nlink == nlink);
1147 ip->i_d.di_uid = current_fsuid(cr);
1148 ip->i_d.di_gid = current_fsgid(cr);
1149 ip->i_d.di_xid = current_fsxid(cr, vp);
1150 ip->i_d.di_projid = prid;
1151 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1154 * If the superblock version is up to where we support new format
1155 * inodes and this is currently an old format inode, then change
1156 * the inode version number now. This way we only do the conversion
1157 * here rather than here and in the flush/logging code.
1159 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1160 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1161 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1163 * We've already zeroed the old link count, the projid field,
1164 * and the pad field.
1169 * Project ids won't be stored on disk if we are using a version 1 inode.
1171 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1172 xfs_bump_ino_vers2(tp, ip);
1174 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1175 ip->i_d.di_gid = pip->i_d.di_gid;
1176 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1177 ip->i_d.di_mode |= S_ISGID;
1182 * If the group ID of the new file does not match the effective group
1183 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1184 * (and only if the irix_sgid_inherit compatibility variable is set).
1186 if ((irix_sgid_inherit) &&
1187 (ip->i_d.di_mode & S_ISGID) &&
1188 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1189 ip->i_d.di_mode &= ~S_ISGID;
1192 ip->i_d.di_size = 0;
1193 ip->i_d.di_nextents = 0;
1194 ASSERT(ip->i_d.di_nblocks == 0);
1195 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1197 * di_gen will have been taken care of in xfs_iread.
1199 ip->i_d.di_extsize = 0;
1200 ip->i_d.di_dmevmask = 0;
1201 ip->i_d.di_dmstate = 0;
1202 ip->i_d.di_flags = 0;
1203 flags = XFS_ILOG_CORE;
1204 switch (mode & S_IFMT) {
1209 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1210 ip->i_df.if_u2.if_rdev = rdev;
1211 ip->i_df.if_flags = 0;
1212 flags |= XFS_ILOG_DEV;
1216 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1219 if ((mode & S_IFMT) == S_IFDIR) {
1220 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1221 di_flags |= XFS_DIFLAG_RTINHERIT;
1222 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1223 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1224 ip->i_d.di_extsize = pip->i_d.di_extsize;
1226 } else if ((mode & S_IFMT) == S_IFREG) {
1227 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1228 di_flags |= XFS_DIFLAG_REALTIME;
1229 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1231 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1232 di_flags |= XFS_DIFLAG_EXTSIZE;
1233 ip->i_d.di_extsize = pip->i_d.di_extsize;
1236 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1237 xfs_inherit_noatime)
1238 di_flags |= XFS_DIFLAG_NOATIME;
1239 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1241 di_flags |= XFS_DIFLAG_NODUMP;
1242 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1244 di_flags |= XFS_DIFLAG_SYNC;
1245 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1246 xfs_inherit_nosymlinks)
1247 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1248 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1249 di_flags |= XFS_DIFLAG_PROJINHERIT;
1250 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1251 xfs_inherit_nodefrag)
1252 di_flags |= XFS_DIFLAG_NODEFRAG;
1253 ip->i_d.di_flags |= di_flags;
1257 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1258 ip->i_df.if_flags = XFS_IFEXTENTS;
1259 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1260 ip->i_df.if_u1.if_extents = NULL;
1266 * Attribute fork settings for new inode.
1268 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1269 ip->i_d.di_anextents = 0;
1272 * Log the new values stuffed into the inode.
1274 xfs_trans_log_inode(tp, ip, flags);
1276 /* now that we have an i_mode we can setup inode ops and unlock */
1277 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1284 * Check to make sure that there are no blocks allocated to the
1285 * file beyond the size of the file. We don't check this for
1286 * files with fixed size extents or real time extents, but we
1287 * at least do it for regular files.
1296 xfs_fileoff_t map_first;
1298 xfs_bmbt_irec_t imaps[2];
1300 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1303 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1307 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1309 * The filesystem could be shutting down, so bmapi may return
1312 if (xfs_bmapi(NULL, ip, map_first,
1314 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1316 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1319 ASSERT(nimaps == 1);
1320 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1325 * Calculate the last possible buffered byte in a file. This must
1326 * include data that was buffered beyond the EOF by the write code.
1327 * This also needs to deal with overflowing the xfs_fsize_t type
1328 * which can happen for sizes near the limit.
1330 * We also need to take into account any blocks beyond the EOF. It
1331 * may be the case that they were buffered by a write which failed.
1332 * In that case the pages will still be in memory, but the inode size
1333 * will never have been updated.
1340 xfs_fsize_t last_byte;
1341 xfs_fileoff_t last_block;
1342 xfs_fileoff_t size_last_block;
1345 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1349 * Only check for blocks beyond the EOF if the extents have
1350 * been read in. This eliminates the need for the inode lock,
1351 * and it also saves us from looking when it really isn't
1354 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1355 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1363 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1364 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1366 last_byte = XFS_FSB_TO_B(mp, last_block);
1367 if (last_byte < 0) {
1368 return XFS_MAXIOFFSET(mp);
1370 last_byte += (1 << mp->m_writeio_log);
1371 if (last_byte < 0) {
1372 return XFS_MAXIOFFSET(mp);
1377 #if defined(XFS_RW_TRACE)
1383 xfs_fsize_t new_size,
1384 xfs_off_t toss_start,
1385 xfs_off_t toss_finish)
1387 if (ip->i_rwtrace == NULL) {
1391 ktrace_enter(ip->i_rwtrace,
1394 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1395 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1396 (void*)((long)flag),
1397 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1398 (void*)(unsigned long)(new_size & 0xffffffff),
1399 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1400 (void*)(unsigned long)(toss_start & 0xffffffff),
1401 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1402 (void*)(unsigned long)(toss_finish & 0xffffffff),
1403 (void*)(unsigned long)current_cpu(),
1404 (void*)(unsigned long)current_pid(),
1410 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1414 * Start the truncation of the file to new_size. The new size
1415 * must be smaller than the current size. This routine will
1416 * clear the buffer and page caches of file data in the removed
1417 * range, and xfs_itruncate_finish() will remove the underlying
1420 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1421 * must NOT have the inode lock held at all. This is because we're
1422 * calling into the buffer/page cache code and we can't hold the
1423 * inode lock when we do so.
1425 * We need to wait for any direct I/Os in flight to complete before we
1426 * proceed with the truncate. This is needed to prevent the extents
1427 * being read or written by the direct I/Os from being removed while the
1428 * I/O is in flight as there is no other method of synchronising
1429 * direct I/O with the truncate operation. Also, because we hold
1430 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1431 * started until the truncate completes and drops the lock. Essentially,
1432 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1433 * between direct I/Os and the truncate operation.
1435 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1436 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1437 * in the case that the caller is locking things out of order and
1438 * may not be able to call xfs_itruncate_finish() with the inode lock
1439 * held without dropping the I/O lock. If the caller must drop the
1440 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1441 * must be called again with all the same restrictions as the initial
1445 xfs_itruncate_start(
1448 xfs_fsize_t new_size)
1450 xfs_fsize_t last_byte;
1451 xfs_off_t toss_start;
1455 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1456 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1457 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1458 (flags == XFS_ITRUNC_MAYBE));
1463 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1466 * Call toss_pages or flushinval_pages to get rid of pages
1467 * overlapping the region being removed. We have to use
1468 * the less efficient flushinval_pages in the case that the
1469 * caller may not be able to finish the truncate without
1470 * dropping the inode's I/O lock. Make sure
1471 * to catch any pages brought in by buffers overlapping
1472 * the EOF by searching out beyond the isize by our
1473 * block size. We round new_size up to a block boundary
1474 * so that we don't toss things on the same block as
1475 * new_size but before it.
1477 * Before calling toss_page or flushinval_pages, make sure to
1478 * call remapf() over the same region if the file is mapped.
1479 * This frees up mapped file references to the pages in the
1480 * given range and for the flushinval_pages case it ensures
1481 * that we get the latest mapped changes flushed out.
1483 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1484 toss_start = XFS_FSB_TO_B(mp, toss_start);
1485 if (toss_start < 0) {
1487 * The place to start tossing is beyond our maximum
1488 * file size, so there is no way that the data extended
1493 last_byte = xfs_file_last_byte(ip);
1494 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1496 if (last_byte > toss_start) {
1497 if (flags & XFS_ITRUNC_DEFINITE) {
1498 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1500 bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1505 if (new_size == 0) {
1506 ASSERT(VN_CACHED(vp) == 0);
1512 * Shrink the file to the given new_size. The new
1513 * size must be smaller than the current size.
1514 * This will free up the underlying blocks
1515 * in the removed range after a call to xfs_itruncate_start()
1516 * or xfs_atruncate_start().
1518 * The transaction passed to this routine must have made
1519 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1520 * This routine may commit the given transaction and
1521 * start new ones, so make sure everything involved in
1522 * the transaction is tidy before calling here.
1523 * Some transaction will be returned to the caller to be
1524 * committed. The incoming transaction must already include
1525 * the inode, and both inode locks must be held exclusively.
1526 * The inode must also be "held" within the transaction. On
1527 * return the inode will be "held" within the returned transaction.
1528 * This routine does NOT require any disk space to be reserved
1529 * for it within the transaction.
1531 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1532 * and it indicates the fork which is to be truncated. For the
1533 * attribute fork we only support truncation to size 0.
1535 * We use the sync parameter to indicate whether or not the first
1536 * transaction we perform might have to be synchronous. For the attr fork,
1537 * it needs to be so if the unlink of the inode is not yet known to be
1538 * permanent in the log. This keeps us from freeing and reusing the
1539 * blocks of the attribute fork before the unlink of the inode becomes
1542 * For the data fork, we normally have to run synchronously if we're
1543 * being called out of the inactive path or we're being called
1544 * out of the create path where we're truncating an existing file.
1545 * Either way, the truncate needs to be sync so blocks don't reappear
1546 * in the file with altered data in case of a crash. wsync filesystems
1547 * can run the first case async because anything that shrinks the inode
1548 * has to run sync so by the time we're called here from inactive, the
1549 * inode size is permanently set to 0.
1551 * Calls from the truncate path always need to be sync unless we're
1552 * in a wsync filesystem and the file has already been unlinked.
1554 * The caller is responsible for correctly setting the sync parameter.
1555 * It gets too hard for us to guess here which path we're being called
1556 * out of just based on inode state.
1559 xfs_itruncate_finish(
1562 xfs_fsize_t new_size,
1566 xfs_fsblock_t first_block;
1567 xfs_fileoff_t first_unmap_block;
1568 xfs_fileoff_t last_block;
1569 xfs_filblks_t unmap_len=0;
1574 xfs_bmap_free_t free_list;
1577 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1578 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1579 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1580 ASSERT(*tp != NULL);
1581 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1582 ASSERT(ip->i_transp == *tp);
1583 ASSERT(ip->i_itemp != NULL);
1584 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1588 mp = (ntp)->t_mountp;
1589 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1592 * We only support truncating the entire attribute fork.
1594 if (fork == XFS_ATTR_FORK) {
1597 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1598 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1600 * The first thing we do is set the size to new_size permanently
1601 * on disk. This way we don't have to worry about anyone ever
1602 * being able to look at the data being freed even in the face
1603 * of a crash. What we're getting around here is the case where
1604 * we free a block, it is allocated to another file, it is written
1605 * to, and then we crash. If the new data gets written to the
1606 * file but the log buffers containing the free and reallocation
1607 * don't, then we'd end up with garbage in the blocks being freed.
1608 * As long as we make the new_size permanent before actually
1609 * freeing any blocks it doesn't matter if they get writtten to.
1611 * The callers must signal into us whether or not the size
1612 * setting here must be synchronous. There are a few cases
1613 * where it doesn't have to be synchronous. Those cases
1614 * occur if the file is unlinked and we know the unlink is
1615 * permanent or if the blocks being truncated are guaranteed
1616 * to be beyond the inode eof (regardless of the link count)
1617 * and the eof value is permanent. Both of these cases occur
1618 * only on wsync-mounted filesystems. In those cases, we're
1619 * guaranteed that no user will ever see the data in the blocks
1620 * that are being truncated so the truncate can run async.
1621 * In the free beyond eof case, the file may wind up with
1622 * more blocks allocated to it than it needs if we crash
1623 * and that won't get fixed until the next time the file
1624 * is re-opened and closed but that's ok as that shouldn't
1625 * be too many blocks.
1627 * However, we can't just make all wsync xactions run async
1628 * because there's one call out of the create path that needs
1629 * to run sync where it's truncating an existing file to size
1630 * 0 whose size is > 0.
1632 * It's probably possible to come up with a test in this
1633 * routine that would correctly distinguish all the above
1634 * cases from the values of the function parameters and the
1635 * inode state but for sanity's sake, I've decided to let the
1636 * layers above just tell us. It's simpler to correctly figure
1637 * out in the layer above exactly under what conditions we
1638 * can run async and I think it's easier for others read and
1639 * follow the logic in case something has to be changed.
1640 * cscope is your friend -- rcc.
1642 * The attribute fork is much simpler.
1644 * For the attribute fork we allow the caller to tell us whether
1645 * the unlink of the inode that led to this call is yet permanent
1646 * in the on disk log. If it is not and we will be freeing extents
1647 * in this inode then we make the first transaction synchronous
1648 * to make sure that the unlink is permanent by the time we free
1651 if (fork == XFS_DATA_FORK) {
1652 if (ip->i_d.di_nextents > 0) {
1653 ip->i_d.di_size = new_size;
1654 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1657 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1658 if (ip->i_d.di_anextents > 0)
1659 xfs_trans_set_sync(ntp);
1661 ASSERT(fork == XFS_DATA_FORK ||
1662 (fork == XFS_ATTR_FORK &&
1663 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1664 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1667 * Since it is possible for space to become allocated beyond
1668 * the end of the file (in a crash where the space is allocated
1669 * but the inode size is not yet updated), simply remove any
1670 * blocks which show up between the new EOF and the maximum
1671 * possible file size. If the first block to be removed is
1672 * beyond the maximum file size (ie it is the same as last_block),
1673 * then there is nothing to do.
1675 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1676 ASSERT(first_unmap_block <= last_block);
1678 if (last_block == first_unmap_block) {
1681 unmap_len = last_block - first_unmap_block + 1;
1685 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1686 * will tell us whether it freed the entire range or
1687 * not. If this is a synchronous mount (wsync),
1688 * then we can tell bunmapi to keep all the
1689 * transactions asynchronous since the unlink
1690 * transaction that made this inode inactive has
1691 * already hit the disk. There's no danger of
1692 * the freed blocks being reused, there being a
1693 * crash, and the reused blocks suddenly reappearing
1694 * in this file with garbage in them once recovery
1697 XFS_BMAP_INIT(&free_list, &first_block);
1698 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1699 first_unmap_block, unmap_len,
1700 XFS_BMAPI_AFLAG(fork) |
1701 (sync ? 0 : XFS_BMAPI_ASYNC),
1702 XFS_ITRUNC_MAX_EXTENTS,
1703 &first_block, &free_list,
1707 * If the bunmapi call encounters an error,
1708 * return to the caller where the transaction
1709 * can be properly aborted. We just need to
1710 * make sure we're not holding any resources
1711 * that we were not when we came in.
1713 xfs_bmap_cancel(&free_list);
1718 * Duplicate the transaction that has the permanent
1719 * reservation and commit the old transaction.
1721 error = xfs_bmap_finish(tp, &free_list, first_block,
1726 * If the bmap finish call encounters an error,
1727 * return to the caller where the transaction
1728 * can be properly aborted. We just need to
1729 * make sure we're not holding any resources
1730 * that we were not when we came in.
1732 * Aborting from this point might lose some
1733 * blocks in the file system, but oh well.
1735 xfs_bmap_cancel(&free_list);
1738 * If the passed in transaction committed
1739 * in xfs_bmap_finish(), then we want to
1740 * add the inode to this one before returning.
1741 * This keeps things simple for the higher
1742 * level code, because it always knows that
1743 * the inode is locked and held in the
1744 * transaction that returns to it whether
1745 * errors occur or not. We don't mark the
1746 * inode dirty so that this transaction can
1747 * be easily aborted if possible.
1749 xfs_trans_ijoin(ntp, ip,
1750 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1751 xfs_trans_ihold(ntp, ip);
1758 * The first xact was committed,
1759 * so add the inode to the new one.
1760 * Mark it dirty so it will be logged
1761 * and moved forward in the log as
1762 * part of every commit.
1764 xfs_trans_ijoin(ntp, ip,
1765 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1766 xfs_trans_ihold(ntp, ip);
1767 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1769 ntp = xfs_trans_dup(ntp);
1770 (void) xfs_trans_commit(*tp, 0, NULL);
1772 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1773 XFS_TRANS_PERM_LOG_RES,
1774 XFS_ITRUNCATE_LOG_COUNT);
1776 * Add the inode being truncated to the next chained
1779 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1780 xfs_trans_ihold(ntp, ip);
1785 * Only update the size in the case of the data fork, but
1786 * always re-log the inode so that our permanent transaction
1787 * can keep on rolling it forward in the log.
1789 if (fork == XFS_DATA_FORK) {
1790 xfs_isize_check(mp, ip, new_size);
1791 ip->i_d.di_size = new_size;
1793 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1794 ASSERT((new_size != 0) ||
1795 (fork == XFS_ATTR_FORK) ||
1796 (ip->i_delayed_blks == 0));
1797 ASSERT((new_size != 0) ||
1798 (fork == XFS_ATTR_FORK) ||
1799 (ip->i_d.di_nextents == 0));
1800 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1808 * Do the first part of growing a file: zero any data in the last
1809 * block that is beyond the old EOF. We need to do this before
1810 * the inode is joined to the transaction to modify the i_size.
1811 * That way we can drop the inode lock and call into the buffer
1812 * cache to get the buffer mapping the EOF.
1817 xfs_fsize_t new_size,
1822 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1823 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1824 ASSERT(new_size > ip->i_d.di_size);
1827 * Zero any pages that may have been created by
1828 * xfs_write_file() beyond the end of the file
1829 * and any blocks between the old and new file sizes.
1831 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1832 ip->i_d.di_size, new_size);
1839 * This routine is called to extend the size of a file.
1840 * The inode must have both the iolock and the ilock locked
1841 * for update and it must be a part of the current transaction.
1842 * The xfs_igrow_start() function must have been called previously.
1843 * If the change_flag is not zero, the inode change timestamp will
1850 xfs_fsize_t new_size,
1853 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1854 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1855 ASSERT(ip->i_transp == tp);
1856 ASSERT(new_size > ip->i_d.di_size);
1859 * Update the file size. Update the inode change timestamp
1860 * if change_flag set.
1862 ip->i_d.di_size = new_size;
1864 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1865 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1871 * This is called when the inode's link count goes to 0.
1872 * We place the on-disk inode on a list in the AGI. It
1873 * will be pulled from this list when the inode is freed.
1885 xfs_agnumber_t agno;
1886 xfs_daddr_t agdaddr;
1893 ASSERT(ip->i_d.di_nlink == 0);
1894 ASSERT(ip->i_d.di_mode != 0);
1895 ASSERT(ip->i_transp == tp);
1899 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1900 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1903 * Get the agi buffer first. It ensures lock ordering
1906 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1907 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1912 * Validate the magic number of the agi block.
1914 agi = XFS_BUF_TO_AGI(agibp);
1916 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1917 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1918 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1919 XFS_RANDOM_IUNLINK))) {
1920 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1921 xfs_trans_brelse(tp, agibp);
1922 return XFS_ERROR(EFSCORRUPTED);
1925 * Get the index into the agi hash table for the
1926 * list this inode will go on.
1928 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1930 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1931 ASSERT(agi->agi_unlinked[bucket_index]);
1932 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1934 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1936 * There is already another inode in the bucket we need
1937 * to add ourselves to. Add us at the front of the list.
1938 * Here we put the head pointer into our next pointer,
1939 * and then we fall through to point the head at us.
1941 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1945 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1946 ASSERT(dip->di_next_unlinked);
1947 /* both on-disk, don't endian flip twice */
1948 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1949 offset = ip->i_boffset +
1950 offsetof(xfs_dinode_t, di_next_unlinked);
1951 xfs_trans_inode_buf(tp, ibp);
1952 xfs_trans_log_buf(tp, ibp, offset,
1953 (offset + sizeof(xfs_agino_t) - 1));
1954 xfs_inobp_check(mp, ibp);
1958 * Point the bucket head pointer at the inode being inserted.
1961 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1962 offset = offsetof(xfs_agi_t, agi_unlinked) +
1963 (sizeof(xfs_agino_t) * bucket_index);
1964 xfs_trans_log_buf(tp, agibp, offset,
1965 (offset + sizeof(xfs_agino_t) - 1));
1970 * Pull the on-disk inode from the AGI unlinked list.
1983 xfs_agnumber_t agno;
1984 xfs_daddr_t agdaddr;
1986 xfs_agino_t next_agino;
1987 xfs_buf_t *last_ibp;
1988 xfs_dinode_t *last_dip = NULL;
1990 int offset, last_offset = 0;
1995 * First pull the on-disk inode from the AGI unlinked list.
1999 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2000 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2003 * Get the agi buffer first. It ensures lock ordering
2006 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2007 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2010 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2011 error, mp->m_fsname);
2015 * Validate the magic number of the agi block.
2017 agi = XFS_BUF_TO_AGI(agibp);
2019 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2020 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2021 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2022 XFS_RANDOM_IUNLINK_REMOVE))) {
2023 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2025 xfs_trans_brelse(tp, agibp);
2027 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2029 return XFS_ERROR(EFSCORRUPTED);
2032 * Get the index into the agi hash table for the
2033 * list this inode will go on.
2035 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2037 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2038 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2039 ASSERT(agi->agi_unlinked[bucket_index]);
2041 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2043 * We're at the head of the list. Get the inode's
2044 * on-disk buffer to see if there is anyone after us
2045 * on the list. Only modify our next pointer if it
2046 * is not already NULLAGINO. This saves us the overhead
2047 * of dealing with the buffer when there is no need to
2050 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2053 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2054 error, mp->m_fsname);
2057 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2058 ASSERT(next_agino != 0);
2059 if (next_agino != NULLAGINO) {
2060 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2061 offset = ip->i_boffset +
2062 offsetof(xfs_dinode_t, di_next_unlinked);
2063 xfs_trans_inode_buf(tp, ibp);
2064 xfs_trans_log_buf(tp, ibp, offset,
2065 (offset + sizeof(xfs_agino_t) - 1));
2066 xfs_inobp_check(mp, ibp);
2068 xfs_trans_brelse(tp, ibp);
2071 * Point the bucket head pointer at the next inode.
2073 ASSERT(next_agino != 0);
2074 ASSERT(next_agino != agino);
2075 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2076 offset = offsetof(xfs_agi_t, agi_unlinked) +
2077 (sizeof(xfs_agino_t) * bucket_index);
2078 xfs_trans_log_buf(tp, agibp, offset,
2079 (offset + sizeof(xfs_agino_t) - 1));
2082 * We need to search the list for the inode being freed.
2084 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2086 while (next_agino != agino) {
2088 * If the last inode wasn't the one pointing to
2089 * us, then release its buffer since we're not
2090 * going to do anything with it.
2092 if (last_ibp != NULL) {
2093 xfs_trans_brelse(tp, last_ibp);
2095 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2096 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2097 &last_ibp, &last_offset);
2100 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2101 error, mp->m_fsname);
2104 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2105 ASSERT(next_agino != NULLAGINO);
2106 ASSERT(next_agino != 0);
2109 * Now last_ibp points to the buffer previous to us on
2110 * the unlinked list. Pull us from the list.
2112 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2115 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2116 error, mp->m_fsname);
2119 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2120 ASSERT(next_agino != 0);
2121 ASSERT(next_agino != agino);
2122 if (next_agino != NULLAGINO) {
2123 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2124 offset = ip->i_boffset +
2125 offsetof(xfs_dinode_t, di_next_unlinked);
2126 xfs_trans_inode_buf(tp, ibp);
2127 xfs_trans_log_buf(tp, ibp, offset,
2128 (offset + sizeof(xfs_agino_t) - 1));
2129 xfs_inobp_check(mp, ibp);
2131 xfs_trans_brelse(tp, ibp);
2134 * Point the previous inode on the list to the next inode.
2136 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2137 ASSERT(next_agino != 0);
2138 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2139 xfs_trans_inode_buf(tp, last_ibp);
2140 xfs_trans_log_buf(tp, last_ibp, offset,
2141 (offset + sizeof(xfs_agino_t) - 1));
2142 xfs_inobp_check(mp, last_ibp);
2147 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2149 return (((ip->i_itemp == NULL) ||
2150 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2151 (ip->i_update_core == 0));
2156 xfs_inode_t *free_ip,
2160 xfs_mount_t *mp = free_ip->i_mount;
2161 int blks_per_cluster;
2164 int i, j, found, pre_flushed;
2168 xfs_inode_t *ip, **ip_found;
2169 xfs_inode_log_item_t *iip;
2170 xfs_log_item_t *lip;
2173 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2174 blks_per_cluster = 1;
2175 ninodes = mp->m_sb.sb_inopblock;
2176 nbufs = XFS_IALLOC_BLOCKS(mp);
2178 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2179 mp->m_sb.sb_blocksize;
2180 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2181 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2184 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2186 for (j = 0; j < nbufs; j++, inum += ninodes) {
2187 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2188 XFS_INO_TO_AGBNO(mp, inum));
2192 * Look for each inode in memory and attempt to lock it,
2193 * we can be racing with flush and tail pushing here.
2194 * any inode we get the locks on, add to an array of
2195 * inode items to process later.
2197 * The get the buffer lock, we could beat a flush
2198 * or tail pushing thread to the lock here, in which
2199 * case they will go looking for the inode buffer
2200 * and fail, we need some other form of interlock
2204 for (i = 0; i < ninodes; i++) {
2205 ih = XFS_IHASH(mp, inum + i);
2206 read_lock(&ih->ih_lock);
2207 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2208 if (ip->i_ino == inum + i)
2212 /* Inode not in memory or we found it already,
2215 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2216 read_unlock(&ih->ih_lock);
2220 if (xfs_inode_clean(ip)) {
2221 read_unlock(&ih->ih_lock);
2225 /* If we can get the locks then add it to the
2226 * list, otherwise by the time we get the bp lock
2227 * below it will already be attached to the
2231 /* This inode will already be locked - by us, lets
2235 if (ip == free_ip) {
2236 if (xfs_iflock_nowait(ip)) {
2237 ip->i_flags |= XFS_ISTALE;
2239 if (xfs_inode_clean(ip)) {
2242 ip_found[found++] = ip;
2245 read_unlock(&ih->ih_lock);
2249 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2250 if (xfs_iflock_nowait(ip)) {
2251 ip->i_flags |= XFS_ISTALE;
2253 if (xfs_inode_clean(ip)) {
2255 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2257 ip_found[found++] = ip;
2260 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2264 read_unlock(&ih->ih_lock);
2267 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2268 mp->m_bsize * blks_per_cluster,
2272 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2274 if (lip->li_type == XFS_LI_INODE) {
2275 iip = (xfs_inode_log_item_t *)lip;
2276 ASSERT(iip->ili_logged == 1);
2277 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2279 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2281 iip->ili_inode->i_flags |= XFS_ISTALE;
2284 lip = lip->li_bio_list;
2287 for (i = 0; i < found; i++) {
2292 ip->i_update_core = 0;
2294 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2298 iip->ili_last_fields = iip->ili_format.ilf_fields;
2299 iip->ili_format.ilf_fields = 0;
2300 iip->ili_logged = 1;
2302 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2305 xfs_buf_attach_iodone(bp,
2306 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2307 xfs_istale_done, (xfs_log_item_t *)iip);
2308 if (ip != free_ip) {
2309 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2313 if (found || pre_flushed)
2314 xfs_trans_stale_inode_buf(tp, bp);
2315 xfs_trans_binval(tp, bp);
2318 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2322 * This is called to return an inode to the inode free list.
2323 * The inode should already be truncated to 0 length and have
2324 * no pages associated with it. This routine also assumes that
2325 * the inode is already a part of the transaction.
2327 * The on-disk copy of the inode will have been added to the list
2328 * of unlinked inodes in the AGI. We need to remove the inode from
2329 * that list atomically with respect to freeing it here.
2335 xfs_bmap_free_t *flist)
2339 xfs_ino_t first_ino;
2341 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2342 ASSERT(ip->i_transp == tp);
2343 ASSERT(ip->i_d.di_nlink == 0);
2344 ASSERT(ip->i_d.di_nextents == 0);
2345 ASSERT(ip->i_d.di_anextents == 0);
2346 ASSERT((ip->i_d.di_size == 0) ||
2347 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2348 ASSERT(ip->i_d.di_nblocks == 0);
2351 * Pull the on-disk inode from the AGI unlinked list.
2353 error = xfs_iunlink_remove(tp, ip);
2358 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2362 ip->i_d.di_mode = 0; /* mark incore inode as free */
2363 ip->i_d.di_flags = 0;
2364 ip->i_d.di_dmevmask = 0;
2365 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2366 ip->i_df.if_ext_max =
2367 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2368 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2369 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2371 * Bump the generation count so no one will be confused
2372 * by reincarnations of this inode.
2375 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2378 xfs_ifree_cluster(ip, tp, first_ino);
2385 * Reallocate the space for if_broot based on the number of records
2386 * being added or deleted as indicated in rec_diff. Move the records
2387 * and pointers in if_broot to fit the new size. When shrinking this
2388 * will eliminate holes between the records and pointers created by
2389 * the caller. When growing this will create holes to be filled in
2392 * The caller must not request to add more records than would fit in
2393 * the on-disk inode root. If the if_broot is currently NULL, then
2394 * if we adding records one will be allocated. The caller must also
2395 * not request that the number of records go below zero, although
2396 * it can go to zero.
2398 * ip -- the inode whose if_broot area is changing
2399 * ext_diff -- the change in the number of records, positive or negative,
2400 * requested for the if_broot array.
2410 xfs_bmbt_block_t *new_broot;
2417 * Handle the degenerate case quietly.
2419 if (rec_diff == 0) {
2423 ifp = XFS_IFORK_PTR(ip, whichfork);
2426 * If there wasn't any memory allocated before, just
2427 * allocate it now and get out.
2429 if (ifp->if_broot_bytes == 0) {
2430 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2431 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2433 ifp->if_broot_bytes = (int)new_size;
2438 * If there is already an existing if_broot, then we need
2439 * to realloc() it and shift the pointers to their new
2440 * location. The records don't change location because
2441 * they are kept butted up against the btree block header.
2443 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2444 new_max = cur_max + rec_diff;
2445 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2446 ifp->if_broot = (xfs_bmbt_block_t *)
2447 kmem_realloc(ifp->if_broot,
2449 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2451 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2452 ifp->if_broot_bytes);
2453 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2455 ifp->if_broot_bytes = (int)new_size;
2456 ASSERT(ifp->if_broot_bytes <=
2457 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2458 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2463 * rec_diff is less than 0. In this case, we are shrinking the
2464 * if_broot buffer. It must already exist. If we go to zero
2465 * records, just get rid of the root and clear the status bit.
2467 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2468 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2469 new_max = cur_max + rec_diff;
2470 ASSERT(new_max >= 0);
2472 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2476 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2478 * First copy over the btree block header.
2480 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2483 ifp->if_flags &= ~XFS_IFBROOT;
2487 * Only copy the records and pointers if there are any.
2491 * First copy the records.
2493 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2494 ifp->if_broot_bytes);
2495 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2497 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2500 * Then copy the pointers.
2502 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2503 ifp->if_broot_bytes);
2504 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2506 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2508 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2509 ifp->if_broot = new_broot;
2510 ifp->if_broot_bytes = (int)new_size;
2511 ASSERT(ifp->if_broot_bytes <=
2512 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2518 * This is called when the amount of space needed for if_data
2519 * is increased or decreased. The change in size is indicated by
2520 * the number of bytes that need to be added or deleted in the
2521 * byte_diff parameter.
2523 * If the amount of space needed has decreased below the size of the
2524 * inline buffer, then switch to using the inline buffer. Otherwise,
2525 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2526 * to what is needed.
2528 * ip -- the inode whose if_data area is changing
2529 * byte_diff -- the change in the number of bytes, positive or negative,
2530 * requested for the if_data array.
2542 if (byte_diff == 0) {
2546 ifp = XFS_IFORK_PTR(ip, whichfork);
2547 new_size = (int)ifp->if_bytes + byte_diff;
2548 ASSERT(new_size >= 0);
2550 if (new_size == 0) {
2551 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2552 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2554 ifp->if_u1.if_data = NULL;
2556 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2558 * If the valid extents/data can fit in if_inline_ext/data,
2559 * copy them from the malloc'd vector and free it.
2561 if (ifp->if_u1.if_data == NULL) {
2562 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2563 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2564 ASSERT(ifp->if_real_bytes != 0);
2565 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2567 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2568 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2573 * Stuck with malloc/realloc.
2574 * For inline data, the underlying buffer must be
2575 * a multiple of 4 bytes in size so that it can be
2576 * logged and stay on word boundaries. We enforce
2579 real_size = roundup(new_size, 4);
2580 if (ifp->if_u1.if_data == NULL) {
2581 ASSERT(ifp->if_real_bytes == 0);
2582 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2583 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2585 * Only do the realloc if the underlying size
2586 * is really changing.
2588 if (ifp->if_real_bytes != real_size) {
2589 ifp->if_u1.if_data =
2590 kmem_realloc(ifp->if_u1.if_data,
2596 ASSERT(ifp->if_real_bytes == 0);
2597 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2598 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2602 ifp->if_real_bytes = real_size;
2603 ifp->if_bytes = new_size;
2604 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2611 * Map inode to disk block and offset.
2613 * mp -- the mount point structure for the current file system
2614 * tp -- the current transaction
2615 * ino -- the inode number of the inode to be located
2616 * imap -- this structure is filled in with the information necessary
2617 * to retrieve the given inode from disk
2618 * flags -- flags to pass to xfs_dilocate indicating whether or not
2619 * lookups in the inode btree were OK or not
2629 xfs_fsblock_t fsbno;
2634 fsbno = imap->im_blkno ?
2635 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2636 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2640 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2641 imap->im_len = XFS_FSB_TO_BB(mp, len);
2642 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2643 imap->im_ioffset = (ushort)off;
2644 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2655 ifp = XFS_IFORK_PTR(ip, whichfork);
2656 if (ifp->if_broot != NULL) {
2657 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2658 ifp->if_broot = NULL;
2662 * If the format is local, then we can't have an extents
2663 * array so just look for an inline data array. If we're
2664 * not local then we may or may not have an extents list,
2665 * so check and free it up if we do.
2667 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2668 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2669 (ifp->if_u1.if_data != NULL)) {
2670 ASSERT(ifp->if_real_bytes != 0);
2671 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2672 ifp->if_u1.if_data = NULL;
2673 ifp->if_real_bytes = 0;
2675 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2676 ((ifp->if_flags & XFS_IFEXTIREC) ||
2677 ((ifp->if_u1.if_extents != NULL) &&
2678 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2679 ASSERT(ifp->if_real_bytes != 0);
2680 xfs_iext_destroy(ifp);
2682 ASSERT(ifp->if_u1.if_extents == NULL ||
2683 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2684 ASSERT(ifp->if_real_bytes == 0);
2685 if (whichfork == XFS_ATTR_FORK) {
2686 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2692 * This is called free all the memory associated with an inode.
2693 * It must free the inode itself and any buffers allocated for
2694 * if_extents/if_data and if_broot. It must also free the lock
2695 * associated with the inode.
2702 switch (ip->i_d.di_mode & S_IFMT) {
2706 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2710 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2711 mrfree(&ip->i_lock);
2712 mrfree(&ip->i_iolock);
2713 freesema(&ip->i_flock);
2714 #ifdef XFS_BMAP_TRACE
2715 ktrace_free(ip->i_xtrace);
2717 #ifdef XFS_BMBT_TRACE
2718 ktrace_free(ip->i_btrace);
2721 ktrace_free(ip->i_rwtrace);
2723 #ifdef XFS_ILOCK_TRACE
2724 ktrace_free(ip->i_lock_trace);
2726 #ifdef XFS_DIR2_TRACE
2727 ktrace_free(ip->i_dir_trace);
2730 /* XXXdpd should be able to assert this but shutdown
2731 * is leaving the AIL behind. */
2732 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2733 XFS_FORCED_SHUTDOWN(ip->i_mount));
2734 xfs_inode_item_destroy(ip);
2736 kmem_zone_free(xfs_inode_zone, ip);
2741 * Increment the pin count of the given buffer.
2742 * This value is protected by ipinlock spinlock in the mount structure.
2748 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2750 atomic_inc(&ip->i_pincount);
2754 * Decrement the pin count of the given inode, and wake up
2755 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2756 * inode must have been previously pinned with a call to xfs_ipin().
2762 ASSERT(atomic_read(&ip->i_pincount) > 0);
2764 if (atomic_dec_and_test(&ip->i_pincount)) {
2766 * If the inode is currently being reclaimed, the
2767 * linux inode _and_ the xfs vnode may have been
2768 * freed so we cannot reference either of them safely.
2769 * Hence we should not try to do anything to them
2770 * if the xfs inode is currently in the reclaim
2773 * However, we still need to issue the unpin wakeup
2774 * call as the inode reclaim may be blocked waiting for
2775 * the inode to become unpinned.
2777 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
2778 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2780 /* make sync come back and flush this inode */
2782 struct inode *inode = vn_to_inode(vp);
2784 if (!(inode->i_state &
2785 (I_NEW|I_FREEING|I_CLEAR)))
2786 mark_inode_dirty_sync(inode);
2789 wake_up(&ip->i_ipin_wait);
2794 * This is called to wait for the given inode to be unpinned.
2795 * It will sleep until this happens. The caller must have the
2796 * inode locked in at least shared mode so that the buffer cannot
2797 * be subsequently pinned once someone is waiting for it to be
2804 xfs_inode_log_item_t *iip;
2807 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2809 if (atomic_read(&ip->i_pincount) == 0) {
2814 if (iip && iip->ili_last_lsn) {
2815 lsn = iip->ili_last_lsn;
2821 * Give the log a push so we don't wait here too long.
2823 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2825 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2830 * xfs_iextents_copy()
2832 * This is called to copy the REAL extents (as opposed to the delayed
2833 * allocation extents) from the inode into the given buffer. It
2834 * returns the number of bytes copied into the buffer.
2836 * If there are no delayed allocation extents, then we can just
2837 * memcpy() the extents into the buffer. Otherwise, we need to
2838 * examine each extent in turn and skip those which are delayed.
2843 xfs_bmbt_rec_t *buffer,
2847 xfs_bmbt_rec_t *dest_ep;
2849 #ifdef XFS_BMAP_TRACE
2850 static char fname[] = "xfs_iextents_copy";
2855 xfs_fsblock_t start_block;
2857 ifp = XFS_IFORK_PTR(ip, whichfork);
2858 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2859 ASSERT(ifp->if_bytes > 0);
2861 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2862 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2866 * There are some delayed allocation extents in the
2867 * inode, so copy the extents one at a time and skip
2868 * the delayed ones. There must be at least one
2869 * non-delayed extent.
2873 for (i = 0; i < nrecs; i++) {
2874 ep = xfs_iext_get_ext(ifp, i);
2875 start_block = xfs_bmbt_get_startblock(ep);
2876 if (ISNULLSTARTBLOCK(start_block)) {
2878 * It's a delayed allocation extent, so skip it.
2883 /* Translate to on disk format */
2884 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2885 (__uint64_t*)&dest_ep->l0);
2886 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2887 (__uint64_t*)&dest_ep->l1);
2891 ASSERT(copied != 0);
2892 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2894 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2898 * Each of the following cases stores data into the same region
2899 * of the on-disk inode, so only one of them can be valid at
2900 * any given time. While it is possible to have conflicting formats
2901 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2902 * in EXTENTS format, this can only happen when the fork has
2903 * changed formats after being modified but before being flushed.
2904 * In these cases, the format always takes precedence, because the
2905 * format indicates the current state of the fork.
2912 xfs_inode_log_item_t *iip,
2919 #ifdef XFS_TRANS_DEBUG
2922 static const short brootflag[2] =
2923 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2924 static const short dataflag[2] =
2925 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2926 static const short extflag[2] =
2927 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2931 ifp = XFS_IFORK_PTR(ip, whichfork);
2933 * This can happen if we gave up in iformat in an error path,
2934 * for the attribute fork.
2937 ASSERT(whichfork == XFS_ATTR_FORK);
2940 cp = XFS_DFORK_PTR(dip, whichfork);
2942 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2943 case XFS_DINODE_FMT_LOCAL:
2944 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2945 (ifp->if_bytes > 0)) {
2946 ASSERT(ifp->if_u1.if_data != NULL);
2947 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2948 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2952 case XFS_DINODE_FMT_EXTENTS:
2953 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2954 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2955 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2956 (ifp->if_bytes == 0));
2957 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2958 (ifp->if_bytes > 0));
2959 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2960 (ifp->if_bytes > 0)) {
2961 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2962 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2967 case XFS_DINODE_FMT_BTREE:
2968 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2969 (ifp->if_broot_bytes > 0)) {
2970 ASSERT(ifp->if_broot != NULL);
2971 ASSERT(ifp->if_broot_bytes <=
2972 (XFS_IFORK_SIZE(ip, whichfork) +
2973 XFS_BROOT_SIZE_ADJ));
2974 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2975 (xfs_bmdr_block_t *)cp,
2976 XFS_DFORK_SIZE(dip, mp, whichfork));
2980 case XFS_DINODE_FMT_DEV:
2981 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2982 ASSERT(whichfork == XFS_DATA_FORK);
2983 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2987 case XFS_DINODE_FMT_UUID:
2988 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2989 ASSERT(whichfork == XFS_DATA_FORK);
2990 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3004 * xfs_iflush() will write a modified inode's changes out to the
3005 * inode's on disk home. The caller must have the inode lock held
3006 * in at least shared mode and the inode flush semaphore must be
3007 * held as well. The inode lock will still be held upon return from
3008 * the call and the caller is free to unlock it.
3009 * The inode flush lock will be unlocked when the inode reaches the disk.
3010 * The flags indicate how the inode's buffer should be written out.
3017 xfs_inode_log_item_t *iip;
3025 int clcount; /* count of inodes clustered */
3027 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3030 XFS_STATS_INC(xs_iflush_count);
3032 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3033 ASSERT(issemalocked(&(ip->i_flock)));
3034 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3035 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3041 * If the inode isn't dirty, then just release the inode
3042 * flush lock and do nothing.
3044 if ((ip->i_update_core == 0) &&
3045 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3046 ASSERT((iip != NULL) ?
3047 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3053 * We can't flush the inode until it is unpinned, so
3054 * wait for it. We know noone new can pin it, because
3055 * we are holding the inode lock shared and you need
3056 * to hold it exclusively to pin the inode.
3058 xfs_iunpin_wait(ip);
3061 * This may have been unpinned because the filesystem is shutting
3062 * down forcibly. If that's the case we must not write this inode
3063 * to disk, because the log record didn't make it to disk!
3065 if (XFS_FORCED_SHUTDOWN(mp)) {
3066 ip->i_update_core = 0;
3068 iip->ili_format.ilf_fields = 0;
3070 return XFS_ERROR(EIO);
3074 * Get the buffer containing the on-disk inode.
3076 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3083 * Decide how buffer will be flushed out. This is done before
3084 * the call to xfs_iflush_int because this field is zeroed by it.
3086 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3088 * Flush out the inode buffer according to the directions
3089 * of the caller. In the cases where the caller has given
3090 * us a choice choose the non-delwri case. This is because
3091 * the inode is in the AIL and we need to get it out soon.
3094 case XFS_IFLUSH_SYNC:
3095 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3098 case XFS_IFLUSH_ASYNC:
3099 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3102 case XFS_IFLUSH_DELWRI:
3112 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3113 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3114 case XFS_IFLUSH_DELWRI:
3117 case XFS_IFLUSH_ASYNC:
3120 case XFS_IFLUSH_SYNC:
3131 * First flush out the inode that xfs_iflush was called with.
3133 error = xfs_iflush_int(ip, bp);
3140 * see if other inodes can be gathered into this write
3143 ip->i_chash->chl_buf = bp;
3145 ch = XFS_CHASH(mp, ip->i_blkno);
3146 s = mutex_spinlock(&ch->ch_lock);
3149 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3151 * Do an un-protected check to see if the inode is dirty and
3152 * is a candidate for flushing. These checks will be repeated
3153 * later after the appropriate locks are acquired.
3156 if ((iq->i_update_core == 0) &&
3158 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3159 xfs_ipincount(iq) == 0) {
3164 * Try to get locks. If any are unavailable,
3165 * then this inode cannot be flushed and is skipped.
3168 /* get inode locks (just i_lock) */
3169 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3170 /* get inode flush lock */
3171 if (xfs_iflock_nowait(iq)) {
3172 /* check if pinned */
3173 if (xfs_ipincount(iq) == 0) {
3174 /* arriving here means that
3175 * this inode can be flushed.
3176 * first re-check that it's
3180 if ((iq->i_update_core != 0)||
3182 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3184 error = xfs_iflush_int(iq, bp);
3188 goto cluster_corrupt_out;
3197 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3200 mutex_spinunlock(&ch->ch_lock, s);
3203 XFS_STATS_INC(xs_icluster_flushcnt);
3204 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3208 * If the buffer is pinned then push on the log so we won't
3209 * get stuck waiting in the write for too long.
3211 if (XFS_BUF_ISPINNED(bp)){
3212 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3215 if (flags & INT_DELWRI) {
3216 xfs_bdwrite(mp, bp);
3217 } else if (flags & INT_ASYNC) {
3218 xfs_bawrite(mp, bp);
3220 error = xfs_bwrite(mp, bp);
3226 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3227 xfs_iflush_abort(ip);
3229 * Unlocks the flush lock
3231 return XFS_ERROR(EFSCORRUPTED);
3233 cluster_corrupt_out:
3234 /* Corruption detected in the clustering loop. Invalidate the
3235 * inode buffer and shut down the filesystem.
3237 mutex_spinunlock(&ch->ch_lock, s);
3240 * Clean up the buffer. If it was B_DELWRI, just release it --
3241 * brelse can handle it with no problems. If not, shut down the
3242 * filesystem before releasing the buffer.
3244 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3248 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3252 * Just like incore_relse: if we have b_iodone functions,
3253 * mark the buffer as an error and call them. Otherwise
3254 * mark it as stale and brelse.
3256 if (XFS_BUF_IODONE_FUNC(bp)) {
3257 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3261 XFS_BUF_ERROR(bp,EIO);
3269 xfs_iflush_abort(iq);
3271 * Unlocks the flush lock
3273 return XFS_ERROR(EFSCORRUPTED);
3282 xfs_inode_log_item_t *iip;
3285 #ifdef XFS_TRANS_DEBUG
3290 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3291 ASSERT(issemalocked(&(ip->i_flock)));
3292 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3293 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3300 * If the inode isn't dirty, then just release the inode
3301 * flush lock and do nothing.
3303 if ((ip->i_update_core == 0) &&
3304 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3309 /* set *dip = inode's place in the buffer */
3310 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3313 * Clear i_update_core before copying out the data.
3314 * This is for coordination with our timestamp updates
3315 * that don't hold the inode lock. They will always
3316 * update the timestamps BEFORE setting i_update_core,
3317 * so if we clear i_update_core after they set it we
3318 * are guaranteed to see their updates to the timestamps.
3319 * I believe that this depends on strongly ordered memory
3320 * semantics, but we have that. We use the SYNCHRONIZE
3321 * macro to make sure that the compiler does not reorder
3322 * the i_update_core access below the data copy below.
3324 ip->i_update_core = 0;
3328 * Make sure to get the latest atime from the Linux inode.
3330 xfs_synchronize_atime(ip);
3332 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3333 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3334 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3335 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3336 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3339 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3340 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3341 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3342 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3343 ip->i_ino, ip, ip->i_d.di_magic);
3346 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3348 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3349 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3350 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3351 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3352 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3356 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3358 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3359 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3360 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3361 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3362 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3363 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3368 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3369 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3370 XFS_RANDOM_IFLUSH_5)) {
3371 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3372 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3374 ip->i_d.di_nextents + ip->i_d.di_anextents,
3379 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3380 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3381 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3382 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3383 ip->i_ino, ip->i_d.di_forkoff, ip);
3387 * bump the flush iteration count, used to detect flushes which
3388 * postdate a log record during recovery.
3391 ip->i_d.di_flushiter++;
3394 * Copy the dirty parts of the inode into the on-disk
3395 * inode. We always copy out the core of the inode,
3396 * because if the inode is dirty at all the core must
3399 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3401 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3402 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3403 ip->i_d.di_flushiter = 0;
3406 * If this is really an old format inode and the superblock version
3407 * has not been updated to support only new format inodes, then
3408 * convert back to the old inode format. If the superblock version
3409 * has been updated, then make the conversion permanent.
3411 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3412 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3413 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3414 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3418 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3419 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3422 * The superblock version has already been bumped,
3423 * so just make the conversion to the new inode
3426 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3427 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3428 ip->i_d.di_onlink = 0;
3429 dip->di_core.di_onlink = 0;
3430 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3431 memset(&(dip->di_core.di_pad[0]), 0,
3432 sizeof(dip->di_core.di_pad));
3433 ASSERT(ip->i_d.di_projid == 0);
3437 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3441 if (XFS_IFORK_Q(ip)) {
3443 * The only error from xfs_iflush_fork is on the data fork.
3445 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3447 xfs_inobp_check(mp, bp);
3450 * We've recorded everything logged in the inode, so we'd
3451 * like to clear the ilf_fields bits so we don't log and
3452 * flush things unnecessarily. However, we can't stop
3453 * logging all this information until the data we've copied
3454 * into the disk buffer is written to disk. If we did we might
3455 * overwrite the copy of the inode in the log with all the
3456 * data after re-logging only part of it, and in the face of
3457 * a crash we wouldn't have all the data we need to recover.
3459 * What we do is move the bits to the ili_last_fields field.
3460 * When logging the inode, these bits are moved back to the
3461 * ilf_fields field. In the xfs_iflush_done() routine we
3462 * clear ili_last_fields, since we know that the information
3463 * those bits represent is permanently on disk. As long as
3464 * the flush completes before the inode is logged again, then
3465 * both ilf_fields and ili_last_fields will be cleared.
3467 * We can play with the ilf_fields bits here, because the inode
3468 * lock must be held exclusively in order to set bits there
3469 * and the flush lock protects the ili_last_fields bits.
3470 * Set ili_logged so the flush done
3471 * routine can tell whether or not to look in the AIL.
3472 * Also, store the current LSN of the inode so that we can tell
3473 * whether the item has moved in the AIL from xfs_iflush_done().
3474 * In order to read the lsn we need the AIL lock, because
3475 * it is a 64 bit value that cannot be read atomically.
3477 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3478 iip->ili_last_fields = iip->ili_format.ilf_fields;
3479 iip->ili_format.ilf_fields = 0;
3480 iip->ili_logged = 1;
3482 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3484 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3488 * Attach the function xfs_iflush_done to the inode's
3489 * buffer. This will remove the inode from the AIL
3490 * and unlock the inode's flush lock when the inode is
3491 * completely written to disk.
3493 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3494 xfs_iflush_done, (xfs_log_item_t *)iip);
3496 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3497 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3500 * We're flushing an inode which is not in the AIL and has
3501 * not been logged but has i_update_core set. For this
3502 * case we can use a B_DELWRI flush and immediately drop
3503 * the inode flush lock because we can avoid the whole
3504 * AIL state thing. It's OK to drop the flush lock now,
3505 * because we've already locked the buffer and to do anything
3506 * you really need both.
3509 ASSERT(iip->ili_logged == 0);
3510 ASSERT(iip->ili_last_fields == 0);
3511 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3519 return XFS_ERROR(EFSCORRUPTED);
3524 * Flush all inactive inodes in mp.
3534 XFS_MOUNT_ILOCK(mp);
3540 /* Make sure we skip markers inserted by sync */
3541 if (ip->i_mount == NULL) {
3546 vp = XFS_ITOV_NULL(ip);
3548 XFS_MOUNT_IUNLOCK(mp);
3549 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3553 ASSERT(vn_count(vp) == 0);
3556 } while (ip != mp->m_inodes);
3558 XFS_MOUNT_IUNLOCK(mp);
3562 * xfs_iaccess: check accessibility of inode for mode.
3571 mode_t orgmode = mode;
3572 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3574 if (mode & S_IWUSR) {
3575 umode_t imode = inode->i_mode;
3577 if (IS_RDONLY(inode) &&
3578 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3579 return XFS_ERROR(EROFS);
3581 if (IS_IMMUTABLE(inode))
3582 return XFS_ERROR(EACCES);
3586 * If there's an Access Control List it's used instead of
3589 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3590 return error ? XFS_ERROR(error) : 0;
3592 if (current_fsuid(cr) != ip->i_d.di_uid) {
3594 if (!in_group_p((gid_t)ip->i_d.di_gid))
3599 * If the DACs are ok we don't need any capability check.
3601 if ((ip->i_d.di_mode & mode) == mode)
3604 * Read/write DACs are always overridable.
3605 * Executable DACs are overridable if at least one exec bit is set.
3607 if (!(orgmode & S_IXUSR) ||
3608 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3609 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3612 if ((orgmode == S_IRUSR) ||
3613 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3614 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3617 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3619 return XFS_ERROR(EACCES);
3621 return XFS_ERROR(EACCES);
3625 * xfs_iroundup: round up argument to next power of two
3634 if ((v & (v - 1)) == 0)
3636 ASSERT((v & 0x80000000) == 0);
3637 if ((v & (v + 1)) == 0)
3639 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3643 if ((v & (v + 1)) == 0)
3650 #ifdef XFS_ILOCK_TRACE
3651 ktrace_t *xfs_ilock_trace_buf;
3654 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3656 ktrace_enter(ip->i_lock_trace,
3658 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3659 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3660 (void *)ra, /* caller of ilock */
3661 (void *)(unsigned long)current_cpu(),
3662 (void *)(unsigned long)current_pid(),
3663 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3668 * Return a pointer to the extent record at file index idx.
3672 xfs_ifork_t *ifp, /* inode fork pointer */
3673 xfs_extnum_t idx) /* index of target extent */
3676 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3677 return ifp->if_u1.if_ext_irec->er_extbuf;
3678 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3679 xfs_ext_irec_t *erp; /* irec pointer */
3680 int erp_idx = 0; /* irec index */
3681 xfs_extnum_t page_idx = idx; /* ext index in target list */
3683 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3684 return &erp->er_extbuf[page_idx];
3685 } else if (ifp->if_bytes) {
3686 return &ifp->if_u1.if_extents[idx];
3693 * Insert new item(s) into the extent records for incore inode
3694 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3698 xfs_ifork_t *ifp, /* inode fork pointer */
3699 xfs_extnum_t idx, /* starting index of new items */
3700 xfs_extnum_t count, /* number of inserted items */
3701 xfs_bmbt_irec_t *new) /* items to insert */
3703 xfs_bmbt_rec_t *ep; /* extent record pointer */
3704 xfs_extnum_t i; /* extent record index */
3706 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3707 xfs_iext_add(ifp, idx, count);
3708 for (i = idx; i < idx + count; i++, new++) {
3709 ep = xfs_iext_get_ext(ifp, i);
3710 xfs_bmbt_set_all(ep, new);
3715 * This is called when the amount of space required for incore file
3716 * extents needs to be increased. The ext_diff parameter stores the
3717 * number of new extents being added and the idx parameter contains
3718 * the extent index where the new extents will be added. If the new
3719 * extents are being appended, then we just need to (re)allocate and
3720 * initialize the space. Otherwise, if the new extents are being
3721 * inserted into the middle of the existing entries, a bit more work
3722 * is required to make room for the new extents to be inserted. The
3723 * caller is responsible for filling in the new extent entries upon
3728 xfs_ifork_t *ifp, /* inode fork pointer */
3729 xfs_extnum_t idx, /* index to begin adding exts */
3730 int ext_diff) /* number of extents to add */
3732 int byte_diff; /* new bytes being added */
3733 int new_size; /* size of extents after adding */
3734 xfs_extnum_t nextents; /* number of extents in file */
3736 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3737 ASSERT((idx >= 0) && (idx <= nextents));
3738 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3739 new_size = ifp->if_bytes + byte_diff;
3741 * If the new number of extents (nextents + ext_diff)
3742 * fits inside the inode, then continue to use the inline
3745 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3746 if (idx < nextents) {
3747 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3748 &ifp->if_u2.if_inline_ext[idx],
3749 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3750 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3752 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3753 ifp->if_real_bytes = 0;
3754 ifp->if_lastex = nextents + ext_diff;
3757 * Otherwise use a linear (direct) extent list.
3758 * If the extents are currently inside the inode,
3759 * xfs_iext_realloc_direct will switch us from
3760 * inline to direct extent allocation mode.
3762 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3763 xfs_iext_realloc_direct(ifp, new_size);
3764 if (idx < nextents) {
3765 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3766 &ifp->if_u1.if_extents[idx],
3767 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3768 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3771 /* Indirection array */
3773 xfs_ext_irec_t *erp;
3777 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3778 if (ifp->if_flags & XFS_IFEXTIREC) {
3779 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3781 xfs_iext_irec_init(ifp);
3782 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3783 erp = ifp->if_u1.if_ext_irec;
3785 /* Extents fit in target extent page */
3786 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3787 if (page_idx < erp->er_extcount) {
3788 memmove(&erp->er_extbuf[page_idx + ext_diff],
3789 &erp->er_extbuf[page_idx],
3790 (erp->er_extcount - page_idx) *
3791 sizeof(xfs_bmbt_rec_t));
3792 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3794 erp->er_extcount += ext_diff;
3795 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3797 /* Insert a new extent page */
3799 xfs_iext_add_indirect_multi(ifp,
3800 erp_idx, page_idx, ext_diff);
3803 * If extent(s) are being appended to the last page in
3804 * the indirection array and the new extent(s) don't fit
3805 * in the page, then erp is NULL and erp_idx is set to
3806 * the next index needed in the indirection array.
3809 int count = ext_diff;
3812 erp = xfs_iext_irec_new(ifp, erp_idx);
3813 erp->er_extcount = count;
3814 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3821 ifp->if_bytes = new_size;
3825 * This is called when incore extents are being added to the indirection
3826 * array and the new extents do not fit in the target extent list. The
3827 * erp_idx parameter contains the irec index for the target extent list
3828 * in the indirection array, and the idx parameter contains the extent
3829 * index within the list. The number of extents being added is stored
3830 * in the count parameter.
3832 * |-------| |-------|
3833 * | | | | idx - number of extents before idx
3835 * | | | | count - number of extents being inserted at idx
3836 * |-------| |-------|
3837 * | count | | nex2 | nex2 - number of extents after idx + count
3838 * |-------| |-------|
3841 xfs_iext_add_indirect_multi(
3842 xfs_ifork_t *ifp, /* inode fork pointer */
3843 int erp_idx, /* target extent irec index */
3844 xfs_extnum_t idx, /* index within target list */
3845 int count) /* new extents being added */
3847 int byte_diff; /* new bytes being added */
3848 xfs_ext_irec_t *erp; /* pointer to irec entry */
3849 xfs_extnum_t ext_diff; /* number of extents to add */
3850 xfs_extnum_t ext_cnt; /* new extents still needed */
3851 xfs_extnum_t nex2; /* extents after idx + count */
3852 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3853 int nlists; /* number of irec's (lists) */
3855 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3856 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3857 nex2 = erp->er_extcount - idx;
3858 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3861 * Save second part of target extent list
3862 * (all extents past */
3864 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3865 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3866 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3867 erp->er_extcount -= nex2;
3868 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3869 memset(&erp->er_extbuf[idx], 0, byte_diff);
3873 * Add the new extents to the end of the target
3874 * list, then allocate new irec record(s) and
3875 * extent buffer(s) as needed to store the rest
3876 * of the new extents.
3879 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3881 erp->er_extcount += ext_diff;
3882 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3883 ext_cnt -= ext_diff;
3887 erp = xfs_iext_irec_new(ifp, erp_idx);
3888 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3889 erp->er_extcount = ext_diff;
3890 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3891 ext_cnt -= ext_diff;
3894 /* Add nex2 extents back to indirection array */
3896 xfs_extnum_t ext_avail;
3899 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3900 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3903 * If nex2 extents fit in the current page, append
3904 * nex2_ep after the new extents.
3906 if (nex2 <= ext_avail) {
3907 i = erp->er_extcount;
3910 * Otherwise, check if space is available in the
3913 else if ((erp_idx < nlists - 1) &&
3914 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3915 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3918 /* Create a hole for nex2 extents */
3919 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3920 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3923 * Final choice, create a new extent page for
3928 erp = xfs_iext_irec_new(ifp, erp_idx);
3930 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3931 kmem_free(nex2_ep, byte_diff);
3932 erp->er_extcount += nex2;
3933 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3938 * This is called when the amount of space required for incore file
3939 * extents needs to be decreased. The ext_diff parameter stores the
3940 * number of extents to be removed and the idx parameter contains
3941 * the extent index where the extents will be removed from.
3943 * If the amount of space needed has decreased below the linear
3944 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3945 * extent array. Otherwise, use kmem_realloc() to adjust the
3946 * size to what is needed.
3950 xfs_ifork_t *ifp, /* inode fork pointer */
3951 xfs_extnum_t idx, /* index to begin removing exts */
3952 int ext_diff) /* number of extents to remove */
3954 xfs_extnum_t nextents; /* number of extents in file */
3955 int new_size; /* size of extents after removal */
3957 ASSERT(ext_diff > 0);
3958 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3959 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3961 if (new_size == 0) {
3962 xfs_iext_destroy(ifp);
3963 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3964 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3965 } else if (ifp->if_real_bytes) {
3966 xfs_iext_remove_direct(ifp, idx, ext_diff);
3968 xfs_iext_remove_inline(ifp, idx, ext_diff);
3970 ifp->if_bytes = new_size;
3974 * This removes ext_diff extents from the inline buffer, beginning
3975 * at extent index idx.
3978 xfs_iext_remove_inline(
3979 xfs_ifork_t *ifp, /* inode fork pointer */
3980 xfs_extnum_t idx, /* index to begin removing exts */
3981 int ext_diff) /* number of extents to remove */
3983 int nextents; /* number of extents in file */
3985 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3986 ASSERT(idx < XFS_INLINE_EXTS);
3987 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3988 ASSERT(((nextents - ext_diff) > 0) &&
3989 (nextents - ext_diff) < XFS_INLINE_EXTS);
3991 if (idx + ext_diff < nextents) {
3992 memmove(&ifp->if_u2.if_inline_ext[idx],
3993 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3994 (nextents - (idx + ext_diff)) *
3995 sizeof(xfs_bmbt_rec_t));
3996 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3997 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3999 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4000 ext_diff * sizeof(xfs_bmbt_rec_t));
4005 * This removes ext_diff extents from a linear (direct) extent list,
4006 * beginning at extent index idx. If the extents are being removed
4007 * from the end of the list (ie. truncate) then we just need to re-
4008 * allocate the list to remove the extra space. Otherwise, if the
4009 * extents are being removed from the middle of the existing extent
4010 * entries, then we first need to move the extent records beginning
4011 * at idx + ext_diff up in the list to overwrite the records being
4012 * removed, then remove the extra space via kmem_realloc.
4015 xfs_iext_remove_direct(
4016 xfs_ifork_t *ifp, /* inode fork pointer */
4017 xfs_extnum_t idx, /* index to begin removing exts */
4018 int ext_diff) /* number of extents to remove */
4020 xfs_extnum_t nextents; /* number of extents in file */
4021 int new_size; /* size of extents after removal */
4023 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4024 new_size = ifp->if_bytes -
4025 (ext_diff * sizeof(xfs_bmbt_rec_t));
4026 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4028 if (new_size == 0) {
4029 xfs_iext_destroy(ifp);
4032 /* Move extents up in the list (if needed) */
4033 if (idx + ext_diff < nextents) {
4034 memmove(&ifp->if_u1.if_extents[idx],
4035 &ifp->if_u1.if_extents[idx + ext_diff],
4036 (nextents - (idx + ext_diff)) *
4037 sizeof(xfs_bmbt_rec_t));
4039 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4040 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4042 * Reallocate the direct extent list. If the extents
4043 * will fit inside the inode then xfs_iext_realloc_direct
4044 * will switch from direct to inline extent allocation
4047 xfs_iext_realloc_direct(ifp, new_size);
4048 ifp->if_bytes = new_size;
4052 * This is called when incore extents are being removed from the
4053 * indirection array and the extents being removed span multiple extent
4054 * buffers. The idx parameter contains the file extent index where we
4055 * want to begin removing extents, and the count parameter contains
4056 * how many extents need to be removed.
4058 * |-------| |-------|
4059 * | nex1 | | | nex1 - number of extents before idx
4060 * |-------| | count |
4061 * | | | | count - number of extents being removed at idx
4062 * | count | |-------|
4063 * | | | nex2 | nex2 - number of extents after idx + count
4064 * |-------| |-------|
4067 xfs_iext_remove_indirect(
4068 xfs_ifork_t *ifp, /* inode fork pointer */
4069 xfs_extnum_t idx, /* index to begin removing extents */
4070 int count) /* number of extents to remove */
4072 xfs_ext_irec_t *erp; /* indirection array pointer */
4073 int erp_idx = 0; /* indirection array index */
4074 xfs_extnum_t ext_cnt; /* extents left to remove */
4075 xfs_extnum_t ext_diff; /* extents to remove in current list */
4076 xfs_extnum_t nex1; /* number of extents before idx */
4077 xfs_extnum_t nex2; /* extents after idx + count */
4078 int nlists; /* entries in indirection array */
4079 int page_idx = idx; /* index in target extent list */
4081 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4082 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4083 ASSERT(erp != NULL);
4084 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4088 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4089 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4091 * Check for deletion of entire list;
4092 * xfs_iext_irec_remove() updates extent offsets.
4094 if (ext_diff == erp->er_extcount) {
4095 xfs_iext_irec_remove(ifp, erp_idx);
4096 ext_cnt -= ext_diff;
4099 ASSERT(erp_idx < ifp->if_real_bytes /
4101 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4108 /* Move extents up (if needed) */
4110 memmove(&erp->er_extbuf[nex1],
4111 &erp->er_extbuf[nex1 + ext_diff],
4112 nex2 * sizeof(xfs_bmbt_rec_t));
4114 /* Zero out rest of page */
4115 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4116 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4117 /* Update remaining counters */
4118 erp->er_extcount -= ext_diff;
4119 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4120 ext_cnt -= ext_diff;
4125 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4126 xfs_iext_irec_compact(ifp);
4130 * Create, destroy, or resize a linear (direct) block of extents.
4133 xfs_iext_realloc_direct(
4134 xfs_ifork_t *ifp, /* inode fork pointer */
4135 int new_size) /* new size of extents */
4137 int rnew_size; /* real new size of extents */
4139 rnew_size = new_size;
4141 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4142 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4143 (new_size != ifp->if_real_bytes)));
4145 /* Free extent records */
4146 if (new_size == 0) {
4147 xfs_iext_destroy(ifp);
4149 /* Resize direct extent list and zero any new bytes */
4150 else if (ifp->if_real_bytes) {
4151 /* Check if extents will fit inside the inode */
4152 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4153 xfs_iext_direct_to_inline(ifp, new_size /
4154 (uint)sizeof(xfs_bmbt_rec_t));
4155 ifp->if_bytes = new_size;
4158 if ((new_size & (new_size - 1)) != 0) {
4159 rnew_size = xfs_iroundup(new_size);
4161 if (rnew_size != ifp->if_real_bytes) {
4162 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4163 kmem_realloc(ifp->if_u1.if_extents,
4168 if (rnew_size > ifp->if_real_bytes) {
4169 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4170 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4171 rnew_size - ifp->if_real_bytes);
4175 * Switch from the inline extent buffer to a direct
4176 * extent list. Be sure to include the inline extent
4177 * bytes in new_size.
4180 new_size += ifp->if_bytes;
4181 if ((new_size & (new_size - 1)) != 0) {
4182 rnew_size = xfs_iroundup(new_size);
4184 xfs_iext_inline_to_direct(ifp, rnew_size);
4186 ifp->if_real_bytes = rnew_size;
4187 ifp->if_bytes = new_size;
4191 * Switch from linear (direct) extent records to inline buffer.
4194 xfs_iext_direct_to_inline(
4195 xfs_ifork_t *ifp, /* inode fork pointer */
4196 xfs_extnum_t nextents) /* number of extents in file */
4198 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4199 ASSERT(nextents <= XFS_INLINE_EXTS);
4201 * The inline buffer was zeroed when we switched
4202 * from inline to direct extent allocation mode,
4203 * so we don't need to clear it here.
4205 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4206 nextents * sizeof(xfs_bmbt_rec_t));
4207 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4208 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4209 ifp->if_real_bytes = 0;
4213 * Switch from inline buffer to linear (direct) extent records.
4214 * new_size should already be rounded up to the next power of 2
4215 * by the caller (when appropriate), so use new_size as it is.
4216 * However, since new_size may be rounded up, we can't update
4217 * if_bytes here. It is the caller's responsibility to update
4218 * if_bytes upon return.
4221 xfs_iext_inline_to_direct(
4222 xfs_ifork_t *ifp, /* inode fork pointer */
4223 int new_size) /* number of extents in file */
4225 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4226 kmem_alloc(new_size, KM_SLEEP);
4227 memset(ifp->if_u1.if_extents, 0, new_size);
4228 if (ifp->if_bytes) {
4229 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4231 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4232 sizeof(xfs_bmbt_rec_t));
4234 ifp->if_real_bytes = new_size;
4238 * Resize an extent indirection array to new_size bytes.
4241 xfs_iext_realloc_indirect(
4242 xfs_ifork_t *ifp, /* inode fork pointer */
4243 int new_size) /* new indirection array size */
4245 int nlists; /* number of irec's (ex lists) */
4246 int size; /* current indirection array size */
4248 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4249 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4250 size = nlists * sizeof(xfs_ext_irec_t);
4251 ASSERT(ifp->if_real_bytes);
4252 ASSERT((new_size >= 0) && (new_size != size));
4253 if (new_size == 0) {
4254 xfs_iext_destroy(ifp);
4256 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4257 kmem_realloc(ifp->if_u1.if_ext_irec,
4258 new_size, size, KM_SLEEP);
4263 * Switch from indirection array to linear (direct) extent allocations.
4266 xfs_iext_indirect_to_direct(
4267 xfs_ifork_t *ifp) /* inode fork pointer */
4269 xfs_bmbt_rec_t *ep; /* extent record pointer */
4270 xfs_extnum_t nextents; /* number of extents in file */
4271 int size; /* size of file extents */
4273 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4274 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4275 ASSERT(nextents <= XFS_LINEAR_EXTS);
4276 size = nextents * sizeof(xfs_bmbt_rec_t);
4278 xfs_iext_irec_compact_full(ifp);
4279 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4281 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4282 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4283 ifp->if_flags &= ~XFS_IFEXTIREC;
4284 ifp->if_u1.if_extents = ep;
4285 ifp->if_bytes = size;
4286 if (nextents < XFS_LINEAR_EXTS) {
4287 xfs_iext_realloc_direct(ifp, size);
4292 * Free incore file extents.
4296 xfs_ifork_t *ifp) /* inode fork pointer */
4298 if (ifp->if_flags & XFS_IFEXTIREC) {
4302 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4303 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4304 xfs_iext_irec_remove(ifp, erp_idx);
4306 ifp->if_flags &= ~XFS_IFEXTIREC;
4307 } else if (ifp->if_real_bytes) {
4308 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4309 } else if (ifp->if_bytes) {
4310 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4311 sizeof(xfs_bmbt_rec_t));
4313 ifp->if_u1.if_extents = NULL;
4314 ifp->if_real_bytes = 0;
4319 * Return a pointer to the extent record for file system block bno.
4321 xfs_bmbt_rec_t * /* pointer to found extent record */
4322 xfs_iext_bno_to_ext(
4323 xfs_ifork_t *ifp, /* inode fork pointer */
4324 xfs_fileoff_t bno, /* block number to search for */
4325 xfs_extnum_t *idxp) /* index of target extent */
4327 xfs_bmbt_rec_t *base; /* pointer to first extent */
4328 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4329 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4330 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4331 int high; /* upper boundary in search */
4332 xfs_extnum_t idx = 0; /* index of target extent */
4333 int low; /* lower boundary in search */
4334 xfs_extnum_t nextents; /* number of file extents */
4335 xfs_fileoff_t startoff = 0; /* start offset of extent */
4337 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4338 if (nextents == 0) {
4343 if (ifp->if_flags & XFS_IFEXTIREC) {
4344 /* Find target extent list */
4346 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4347 base = erp->er_extbuf;
4348 high = erp->er_extcount - 1;
4350 base = ifp->if_u1.if_extents;
4351 high = nextents - 1;
4353 /* Binary search extent records */
4354 while (low <= high) {
4355 idx = (low + high) >> 1;
4357 startoff = xfs_bmbt_get_startoff(ep);
4358 blockcount = xfs_bmbt_get_blockcount(ep);
4359 if (bno < startoff) {
4361 } else if (bno >= startoff + blockcount) {
4364 /* Convert back to file-based extent index */
4365 if (ifp->if_flags & XFS_IFEXTIREC) {
4366 idx += erp->er_extoff;
4372 /* Convert back to file-based extent index */
4373 if (ifp->if_flags & XFS_IFEXTIREC) {
4374 idx += erp->er_extoff;
4376 if (bno >= startoff + blockcount) {
4377 if (++idx == nextents) {
4380 ep = xfs_iext_get_ext(ifp, idx);
4388 * Return a pointer to the indirection array entry containing the
4389 * extent record for filesystem block bno. Store the index of the
4390 * target irec in *erp_idxp.
4392 xfs_ext_irec_t * /* pointer to found extent record */
4393 xfs_iext_bno_to_irec(
4394 xfs_ifork_t *ifp, /* inode fork pointer */
4395 xfs_fileoff_t bno, /* block number to search for */
4396 int *erp_idxp) /* irec index of target ext list */
4398 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4399 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4400 int erp_idx; /* indirection array index */
4401 int nlists; /* number of extent irec's (lists) */
4402 int high; /* binary search upper limit */
4403 int low; /* binary search lower limit */
4405 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4406 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4410 while (low <= high) {
4411 erp_idx = (low + high) >> 1;
4412 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4413 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4414 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4416 } else if (erp_next && bno >=
4417 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4423 *erp_idxp = erp_idx;
4428 * Return a pointer to the indirection array entry containing the
4429 * extent record at file extent index *idxp. Store the index of the
4430 * target irec in *erp_idxp and store the page index of the target
4431 * extent record in *idxp.
4434 xfs_iext_idx_to_irec(
4435 xfs_ifork_t *ifp, /* inode fork pointer */
4436 xfs_extnum_t *idxp, /* extent index (file -> page) */
4437 int *erp_idxp, /* pointer to target irec */
4438 int realloc) /* new bytes were just added */
4440 xfs_ext_irec_t *prev; /* pointer to previous irec */
4441 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4442 int erp_idx; /* indirection array index */
4443 int nlists; /* number of irec's (ex lists) */
4444 int high; /* binary search upper limit */
4445 int low; /* binary search lower limit */
4446 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4448 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4449 ASSERT(page_idx >= 0 && page_idx <=
4450 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4451 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4456 /* Binary search extent irec's */
4457 while (low <= high) {
4458 erp_idx = (low + high) >> 1;
4459 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4460 prev = erp_idx > 0 ? erp - 1 : NULL;
4461 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4462 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4464 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4465 (page_idx == erp->er_extoff + erp->er_extcount &&
4468 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4469 erp->er_extcount == XFS_LINEAR_EXTS) {
4473 erp = erp_idx < nlists ? erp + 1 : NULL;
4476 page_idx -= erp->er_extoff;
4481 *erp_idxp = erp_idx;
4486 * Allocate and initialize an indirection array once the space needed
4487 * for incore extents increases above XFS_IEXT_BUFSZ.
4491 xfs_ifork_t *ifp) /* inode fork pointer */
4493 xfs_ext_irec_t *erp; /* indirection array pointer */
4494 xfs_extnum_t nextents; /* number of extents in file */
4496 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4497 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4498 ASSERT(nextents <= XFS_LINEAR_EXTS);
4500 erp = (xfs_ext_irec_t *)
4501 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4503 if (nextents == 0) {
4504 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4505 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4506 } else if (!ifp->if_real_bytes) {
4507 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4508 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4509 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4511 erp->er_extbuf = ifp->if_u1.if_extents;
4512 erp->er_extcount = nextents;
4515 ifp->if_flags |= XFS_IFEXTIREC;
4516 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4517 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4518 ifp->if_u1.if_ext_irec = erp;
4524 * Allocate and initialize a new entry in the indirection array.
4528 xfs_ifork_t *ifp, /* inode fork pointer */
4529 int erp_idx) /* index for new irec */
4531 xfs_ext_irec_t *erp; /* indirection array pointer */
4532 int i; /* loop counter */
4533 int nlists; /* number of irec's (ex lists) */
4535 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4536 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4538 /* Resize indirection array */
4539 xfs_iext_realloc_indirect(ifp, ++nlists *
4540 sizeof(xfs_ext_irec_t));
4542 * Move records down in the array so the
4543 * new page can use erp_idx.
4545 erp = ifp->if_u1.if_ext_irec;
4546 for (i = nlists - 1; i > erp_idx; i--) {
4547 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4549 ASSERT(i == erp_idx);
4551 /* Initialize new extent record */
4552 erp = ifp->if_u1.if_ext_irec;
4553 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4554 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4555 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4556 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4557 erp[erp_idx].er_extcount = 0;
4558 erp[erp_idx].er_extoff = erp_idx > 0 ?
4559 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4560 return (&erp[erp_idx]);
4564 * Remove a record from the indirection array.
4567 xfs_iext_irec_remove(
4568 xfs_ifork_t *ifp, /* inode fork pointer */
4569 int erp_idx) /* irec index to remove */
4571 xfs_ext_irec_t *erp; /* indirection array pointer */
4572 int i; /* loop counter */
4573 int nlists; /* number of irec's (ex lists) */
4575 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4576 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4577 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4578 if (erp->er_extbuf) {
4579 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4581 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4583 /* Compact extent records */
4584 erp = ifp->if_u1.if_ext_irec;
4585 for (i = erp_idx; i < nlists - 1; i++) {
4586 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4589 * Manually free the last extent record from the indirection
4590 * array. A call to xfs_iext_realloc_indirect() with a size
4591 * of zero would result in a call to xfs_iext_destroy() which
4592 * would in turn call this function again, creating a nasty
4596 xfs_iext_realloc_indirect(ifp,
4597 nlists * sizeof(xfs_ext_irec_t));
4599 kmem_free(ifp->if_u1.if_ext_irec,
4600 sizeof(xfs_ext_irec_t));
4602 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4606 * This is called to clean up large amounts of unused memory allocated
4607 * by the indirection array. Before compacting anything though, verify
4608 * that the indirection array is still needed and switch back to the
4609 * linear extent list (or even the inline buffer) if possible. The
4610 * compaction policy is as follows:
4612 * Full Compaction: Extents fit into a single page (or inline buffer)
4613 * Full Compaction: Extents occupy less than 10% of allocated space
4614 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4615 * No Compaction: Extents occupy at least 50% of allocated space
4618 xfs_iext_irec_compact(
4619 xfs_ifork_t *ifp) /* inode fork pointer */
4621 xfs_extnum_t nextents; /* number of extents in file */
4622 int nlists; /* number of irec's (ex lists) */
4624 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4625 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4626 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4628 if (nextents == 0) {
4629 xfs_iext_destroy(ifp);
4630 } else if (nextents <= XFS_INLINE_EXTS) {
4631 xfs_iext_indirect_to_direct(ifp);
4632 xfs_iext_direct_to_inline(ifp, nextents);
4633 } else if (nextents <= XFS_LINEAR_EXTS) {
4634 xfs_iext_indirect_to_direct(ifp);
4635 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4636 xfs_iext_irec_compact_full(ifp);
4637 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4638 xfs_iext_irec_compact_pages(ifp);
4643 * Combine extents from neighboring extent pages.
4646 xfs_iext_irec_compact_pages(
4647 xfs_ifork_t *ifp) /* inode fork pointer */
4649 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4650 int erp_idx = 0; /* indirection array index */
4651 int nlists; /* number of irec's (ex lists) */
4653 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4654 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4655 while (erp_idx < nlists - 1) {
4656 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4658 if (erp_next->er_extcount <=
4659 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4660 memmove(&erp->er_extbuf[erp->er_extcount],
4661 erp_next->er_extbuf, erp_next->er_extcount *
4662 sizeof(xfs_bmbt_rec_t));
4663 erp->er_extcount += erp_next->er_extcount;
4665 * Free page before removing extent record
4666 * so er_extoffs don't get modified in
4667 * xfs_iext_irec_remove.
4669 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4670 erp_next->er_extbuf = NULL;
4671 xfs_iext_irec_remove(ifp, erp_idx + 1);
4672 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4680 * Fully compact the extent records managed by the indirection array.
4683 xfs_iext_irec_compact_full(
4684 xfs_ifork_t *ifp) /* inode fork pointer */
4686 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4687 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4688 int erp_idx = 0; /* extent irec index */
4689 int ext_avail; /* empty entries in ex list */
4690 int ext_diff; /* number of exts to add */
4691 int nlists; /* number of irec's (ex lists) */
4693 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4694 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4695 erp = ifp->if_u1.if_ext_irec;
4696 ep = &erp->er_extbuf[erp->er_extcount];
4698 ep_next = erp_next->er_extbuf;
4699 while (erp_idx < nlists - 1) {
4700 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4701 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4702 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4703 erp->er_extcount += ext_diff;
4704 erp_next->er_extcount -= ext_diff;
4705 /* Remove next page */
4706 if (erp_next->er_extcount == 0) {
4708 * Free page before removing extent record
4709 * so er_extoffs don't get modified in
4710 * xfs_iext_irec_remove.
4712 kmem_free(erp_next->er_extbuf,
4713 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4714 erp_next->er_extbuf = NULL;
4715 xfs_iext_irec_remove(ifp, erp_idx + 1);
4716 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4717 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4718 /* Update next page */
4720 /* Move rest of page up to become next new page */
4721 memmove(erp_next->er_extbuf, ep_next,
4722 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4723 ep_next = erp_next->er_extbuf;
4724 memset(&ep_next[erp_next->er_extcount], 0,
4725 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4726 sizeof(xfs_bmbt_rec_t));
4728 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4730 if (erp_idx < nlists)
4731 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4735 ep = &erp->er_extbuf[erp->er_extcount];
4737 ep_next = erp_next->er_extbuf;
4742 * This is called to update the er_extoff field in the indirection
4743 * array when extents have been added or removed from one of the
4744 * extent lists. erp_idx contains the irec index to begin updating
4745 * at and ext_diff contains the number of extents that were added
4749 xfs_iext_irec_update_extoffs(
4750 xfs_ifork_t *ifp, /* inode fork pointer */
4751 int erp_idx, /* irec index to update */
4752 int ext_diff) /* number of new extents */
4754 int i; /* loop counter */
4755 int nlists; /* number of irec's (ex lists */
4757 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4758 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4759 for (i = erp_idx; i < nlists; i++) {
4760 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;