2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_inode_item.h"
39 #include "xfs_dir2_data.h"
40 #include "xfs_dir2_leaf.h"
41 #include "xfs_dir2_block.h"
42 #include "xfs_dir2_node.h"
43 #include "xfs_dir2_trace.h"
44 #include "xfs_error.h"
46 static int xfs_dir2_put_dirent64_direct(xfs_dir2_put_args_t *pa);
47 static int xfs_dir2_put_dirent64_uio(xfs_dir2_put_args_t *pa);
53 ASSERT(XFS_SB_VERSION_HASDIRV2(&mp->m_sb));
54 ASSERT((1 << (mp->m_sb.sb_blocklog + mp->m_sb.sb_dirblklog)) <=
56 mp->m_dirblksize = 1 << (mp->m_sb.sb_blocklog + mp->m_sb.sb_dirblklog);
57 mp->m_dirblkfsbs = 1 << mp->m_sb.sb_dirblklog;
58 mp->m_dirdatablk = xfs_dir2_db_to_da(mp, XFS_DIR2_DATA_FIRSTDB(mp));
59 mp->m_dirleafblk = xfs_dir2_db_to_da(mp, XFS_DIR2_LEAF_FIRSTDB(mp));
60 mp->m_dirfreeblk = xfs_dir2_db_to_da(mp, XFS_DIR2_FREE_FIRSTDB(mp));
61 mp->m_attr_node_ents =
62 (mp->m_sb.sb_blocksize - (uint)sizeof(xfs_da_node_hdr_t)) /
63 (uint)sizeof(xfs_da_node_entry_t);
65 (mp->m_dirblksize - (uint)sizeof(xfs_da_node_hdr_t)) /
66 (uint)sizeof(xfs_da_node_entry_t);
67 mp->m_dir_magicpct = (mp->m_dirblksize * 37) / 100;
71 * Return 1 if directory contains only "." and "..".
79 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
80 if (dp->i_d.di_size == 0) /* might happen during shutdown. */
82 if (dp->i_d.di_size > XFS_IFORK_DSIZE(dp))
84 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
85 return !sfp->hdr.count;
89 * Validate a given inode number.
96 xfs_agblock_t agblkno;
102 agno = XFS_INO_TO_AGNO(mp, ino);
103 agblkno = XFS_INO_TO_AGBNO(mp, ino);
104 ioff = XFS_INO_TO_OFFSET(mp, ino);
105 agino = XFS_OFFBNO_TO_AGINO(mp, agblkno, ioff);
107 agno < mp->m_sb.sb_agcount &&
108 agblkno < mp->m_sb.sb_agblocks &&
110 ioff < (1 << mp->m_sb.sb_inopblog) &&
111 XFS_AGINO_TO_INO(mp, agno, agino) == ino;
112 if (unlikely(XFS_TEST_ERROR(!ino_ok, mp, XFS_ERRTAG_DIR_INO_VALIDATE,
113 XFS_RANDOM_DIR_INO_VALIDATE))) {
114 xfs_fs_cmn_err(CE_WARN, mp, "Invalid inode number 0x%Lx",
115 (unsigned long long) ino);
116 XFS_ERROR_REPORT("xfs_dir_ino_validate", XFS_ERRLEVEL_LOW, mp);
117 return XFS_ERROR(EFSCORRUPTED);
123 * Initialize a directory with its "." and ".." entries.
134 memset((char *)&args, 0, sizeof(args));
137 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
138 if ((error = xfs_dir_ino_validate(tp->t_mountp, pdp->i_ino)))
140 return xfs_dir2_sf_create(&args, pdp->i_ino);
144 Enter a name in a directory.
152 xfs_ino_t inum, /* new entry inode number */
153 xfs_fsblock_t *first, /* bmap's firstblock */
154 xfs_bmap_free_t *flist, /* bmap's freeblock list */
155 xfs_extlen_t total) /* bmap's total block count */
159 int v; /* type-checking value */
161 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
162 if ((rval = xfs_dir_ino_validate(tp->t_mountp, inum)))
164 XFS_STATS_INC(xs_dir_create);
167 args.namelen = namelen;
168 args.hashval = xfs_da_hashname(name, namelen);
171 args.firstblock = first;
174 args.whichfork = XFS_DATA_FORK;
177 args.addname = args.oknoent = 1;
179 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
180 rval = xfs_dir2_sf_addname(&args);
181 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
184 rval = xfs_dir2_block_addname(&args);
185 else if ((rval = xfs_dir2_isleaf(tp, dp, &v)))
188 rval = xfs_dir2_leaf_addname(&args);
190 rval = xfs_dir2_node_addname(&args);
195 * Lookup a name in a directory, give back the inode number.
203 xfs_ino_t *inum) /* out: inode number */
207 int v; /* type-checking value */
209 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
210 XFS_STATS_INC(xs_dir_lookup);
213 args.namelen = namelen;
214 args.hashval = xfs_da_hashname(name, namelen);
217 args.firstblock = NULL;
220 args.whichfork = XFS_DATA_FORK;
222 args.justcheck = args.addname = 0;
225 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
226 rval = xfs_dir2_sf_lookup(&args);
227 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
230 rval = xfs_dir2_block_lookup(&args);
231 else if ((rval = xfs_dir2_isleaf(tp, dp, &v)))
234 rval = xfs_dir2_leaf_lookup(&args);
236 rval = xfs_dir2_node_lookup(&args);
240 *inum = args.inumber;
245 * Remove an entry from a directory.
254 xfs_fsblock_t *first, /* bmap's firstblock */
255 xfs_bmap_free_t *flist, /* bmap's freeblock list */
256 xfs_extlen_t total) /* bmap's total block count */
260 int v; /* type-checking value */
262 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
263 XFS_STATS_INC(xs_dir_remove);
266 args.namelen = namelen;
267 args.hashval = xfs_da_hashname(name, namelen);
270 args.firstblock = first;
273 args.whichfork = XFS_DATA_FORK;
275 args.justcheck = args.addname = args.oknoent = 0;
277 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
278 rval = xfs_dir2_sf_removename(&args);
279 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
282 rval = xfs_dir2_block_removename(&args);
283 else if ((rval = xfs_dir2_isleaf(tp, dp, &v)))
286 rval = xfs_dir2_leaf_removename(&args);
288 rval = xfs_dir2_node_removename(&args);
299 uio_t *uio, /* caller's buffer control */
300 int *eofp) /* out: eof reached */
302 int alignment; /* alignment required for ABI */
303 xfs_dirent_t *dbp; /* malloc'ed buffer */
304 xfs_dir2_put_t put; /* entry formatting routine */
305 int rval; /* return value */
306 int v; /* type-checking value */
308 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
309 XFS_STATS_INC(xs_dir_getdents);
311 * If our caller has given us a single contiguous aligned memory buffer,
312 * just work directly within that buffer. If it's in user memory,
313 * lock it down first.
315 alignment = sizeof(xfs_off_t) - 1;
316 if ((uio->uio_iovcnt == 1) &&
317 (((__psint_t)uio->uio_iov[0].iov_base & alignment) == 0) &&
318 ((uio->uio_iov[0].iov_len & alignment) == 0)) {
320 put = xfs_dir2_put_dirent64_direct;
322 dbp = kmem_alloc(sizeof(*dbp) + MAXNAMELEN, KM_SLEEP);
323 put = xfs_dir2_put_dirent64_uio;
327 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
328 rval = xfs_dir2_sf_getdents(dp, uio, eofp, dbp, put);
329 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
332 rval = xfs_dir2_block_getdents(tp, dp, uio, eofp, dbp, put);
334 rval = xfs_dir2_leaf_getdents(tp, dp, uio, eofp, dbp, put);
336 kmem_free(dbp, sizeof(*dbp) + MAXNAMELEN);
341 * Replace the inode number of a directory entry.
347 char *name, /* name of entry to replace */
349 xfs_ino_t inum, /* new inode number */
350 xfs_fsblock_t *first, /* bmap's firstblock */
351 xfs_bmap_free_t *flist, /* bmap's freeblock list */
352 xfs_extlen_t total) /* bmap's total block count */
356 int v; /* type-checking value */
358 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
360 if ((rval = xfs_dir_ino_validate(tp->t_mountp, inum)))
364 args.namelen = namelen;
365 args.hashval = xfs_da_hashname(name, namelen);
368 args.firstblock = first;
371 args.whichfork = XFS_DATA_FORK;
373 args.justcheck = args.addname = args.oknoent = 0;
375 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
376 rval = xfs_dir2_sf_replace(&args);
377 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
380 rval = xfs_dir2_block_replace(&args);
381 else if ((rval = xfs_dir2_isleaf(tp, dp, &v)))
384 rval = xfs_dir2_leaf_replace(&args);
386 rval = xfs_dir2_node_replace(&args);
391 * See if this entry can be added to the directory without allocating space.
397 char *name, /* name of entry to add */
402 int v; /* type-checking value */
404 ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
407 args.namelen = namelen;
408 args.hashval = xfs_da_hashname(name, namelen);
411 args.firstblock = NULL;
414 args.whichfork = XFS_DATA_FORK;
416 args.justcheck = args.addname = args.oknoent = 1;
418 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
419 rval = xfs_dir2_sf_addname(&args);
420 else if ((rval = xfs_dir2_isblock(tp, dp, &v)))
423 rval = xfs_dir2_block_addname(&args);
424 else if ((rval = xfs_dir2_isleaf(tp, dp, &v)))
427 rval = xfs_dir2_leaf_addname(&args);
429 rval = xfs_dir2_node_addname(&args);
438 * Add a block to the directory.
439 * This routine is for data and free blocks, not leaf/node blocks
440 * which are handled by xfs_da_grow_inode.
445 int space, /* v2 dir's space XFS_DIR2_xxx_SPACE */
446 xfs_dir2_db_t *dbp) /* out: block number added */
448 xfs_fileoff_t bno; /* directory offset of new block */
449 int count; /* count of filesystem blocks */
450 xfs_inode_t *dp; /* incore directory inode */
452 int got; /* blocks actually mapped */
454 xfs_bmbt_irec_t map; /* single structure for bmap */
455 int mapi; /* mapping index */
456 xfs_bmbt_irec_t *mapp; /* bmap mapping structure(s) */
458 int nmap; /* number of bmap entries */
461 xfs_dir2_trace_args_s("grow_inode", args, space);
466 * Set lowest possible block in the space requested.
468 bno = XFS_B_TO_FSBT(mp, space * XFS_DIR2_SPACE_SIZE);
469 count = mp->m_dirblkfsbs;
471 * Find the first hole for our block.
473 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, XFS_DATA_FORK)))
476 ASSERT(args->firstblock != NULL);
478 * Try mapping the new block contiguously (one extent).
480 if ((error = xfs_bmapi(tp, dp, bno, count,
481 XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
482 args->firstblock, args->total, &map, &nmap,
491 * Didn't work and this is a multiple-fsb directory block.
492 * Try again with contiguous flag turned on.
494 else if (nmap == 0 && count > 1) {
495 xfs_fileoff_t b; /* current file offset */
498 * Space for maximum number of mappings.
500 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
502 * Iterate until we get to the end of our block.
504 for (b = bno, mapi = 0; b < bno + count; ) {
505 int c; /* current fsb count */
508 * Can't map more than MAX_NMAP at once.
510 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
511 c = (int)(bno + count - b);
512 if ((error = xfs_bmapi(tp, dp, b, c,
513 XFS_BMAPI_WRITE|XFS_BMAPI_METADATA,
514 args->firstblock, args->total,
515 &mapp[mapi], &nmap, args->flist,
517 kmem_free(mapp, sizeof(*mapp) * count);
523 * Add this bunch into our table, go to the next offset.
526 b = mapp[mapi - 1].br_startoff +
527 mapp[mapi - 1].br_blockcount;
538 * See how many fsb's we got.
540 for (i = 0, got = 0; i < mapi; i++)
541 got += mapp[i].br_blockcount;
543 * Didn't get enough fsb's, or the first/last block's are wrong.
545 if (got != count || mapp[0].br_startoff != bno ||
546 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
549 kmem_free(mapp, sizeof(*mapp) * count);
550 return XFS_ERROR(ENOSPC);
553 * Done with the temporary mapping table.
556 kmem_free(mapp, sizeof(*mapp) * count);
557 *dbp = xfs_dir2_da_to_db(mp, (xfs_dablk_t)bno);
559 * Update file's size if this is the data space and it grew.
561 if (space == XFS_DIR2_DATA_SPACE) {
562 xfs_fsize_t size; /* directory file (data) size */
564 size = XFS_FSB_TO_B(mp, bno + count);
565 if (size > dp->i_d.di_size) {
566 dp->i_d.di_size = size;
567 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
574 * See if the directory is a single-block form directory.
580 int *vp) /* out: 1 is block, 0 is not block */
582 xfs_fileoff_t last; /* last file offset */
587 if ((rval = xfs_bmap_last_offset(tp, dp, &last, XFS_DATA_FORK)))
589 rval = XFS_FSB_TO_B(mp, last) == mp->m_dirblksize;
590 ASSERT(rval == 0 || dp->i_d.di_size == mp->m_dirblksize);
596 * See if the directory is a single-leaf form directory.
602 int *vp) /* out: 1 is leaf, 0 is not leaf */
604 xfs_fileoff_t last; /* last file offset */
609 if ((rval = xfs_bmap_last_offset(tp, dp, &last, XFS_DATA_FORK)))
611 *vp = last == mp->m_dirleafblk + (1 << mp->m_sb.sb_dirblklog);
616 * Getdents put routine for 64-bit ABI, direct form.
619 xfs_dir2_put_dirent64_direct(
620 xfs_dir2_put_args_t *pa)
622 xfs_dirent_t *idbp; /* dirent pointer */
623 iovec_t *iovp; /* io vector */
624 int namelen; /* entry name length */
625 int reclen; /* entry total length */
626 uio_t *uio; /* I/O control */
628 namelen = pa->namelen;
629 reclen = DIRENTSIZE(namelen);
632 * Won't fit in the remaining space.
634 if (reclen > uio->uio_resid) {
639 idbp = (xfs_dirent_t *)iovp->iov_base;
640 iovp->iov_base = (char *)idbp + reclen;
641 iovp->iov_len -= reclen;
642 uio->uio_resid -= reclen;
643 idbp->d_reclen = reclen;
644 idbp->d_ino = pa->ino;
645 idbp->d_off = pa->cook;
646 idbp->d_name[namelen] = '\0';
648 memcpy(idbp->d_name, pa->name, namelen);
653 * Getdents put routine for 64-bit ABI, uio form.
656 xfs_dir2_put_dirent64_uio(
657 xfs_dir2_put_args_t *pa)
659 xfs_dirent_t *idbp; /* dirent pointer */
660 int namelen; /* entry name length */
661 int reclen; /* entry total length */
662 int rval; /* return value */
663 uio_t *uio; /* I/O control */
665 namelen = pa->namelen;
666 reclen = DIRENTSIZE(namelen);
669 * Won't fit in the remaining space.
671 if (reclen > uio->uio_resid) {
676 idbp->d_reclen = reclen;
677 idbp->d_ino = pa->ino;
678 idbp->d_off = pa->cook;
679 idbp->d_name[namelen] = '\0';
680 memcpy(idbp->d_name, pa->name, namelen);
681 rval = xfs_uio_read((caddr_t)idbp, reclen, uio);
682 pa->done = (rval == 0);
687 * Remove the given block from the directory.
688 * This routine is used for data and free blocks, leaf/node are done
689 * by xfs_da_shrink_inode.
692 xfs_dir2_shrink_inode(
697 xfs_fileoff_t bno; /* directory file offset */
698 xfs_dablk_t da; /* directory file offset */
699 int done; /* bunmap is finished */
705 xfs_dir2_trace_args_db("shrink_inode", args, db, bp);
709 da = xfs_dir2_db_to_da(mp, db);
711 * Unmap the fsblock(s).
713 if ((error = xfs_bunmapi(tp, dp, da, mp->m_dirblkfsbs,
714 XFS_BMAPI_METADATA, 0, args->firstblock, args->flist,
717 * ENOSPC actually can happen if we're in a removename with
718 * no space reservation, and the resulting block removal
719 * would cause a bmap btree split or conversion from extents
720 * to btree. This can only happen for un-fragmented
721 * directory blocks, since you need to be punching out
722 * the middle of an extent.
723 * In this case we need to leave the block in the file,
725 * So the block has to be in a consistent empty state
726 * and appropriately logged.
727 * We don't free up the buffer, the caller can tell it
728 * hasn't happened since it got an error back.
734 * Invalidate the buffer from the transaction.
736 xfs_da_binval(tp, bp);
738 * If it's not a data block, we're done.
740 if (db >= XFS_DIR2_LEAF_FIRSTDB(mp))
743 * If the block isn't the last one in the directory, we're done.
745 if (dp->i_d.di_size > xfs_dir2_db_off_to_byte(mp, db + 1, 0))
748 if ((error = xfs_bmap_last_before(tp, dp, &bno, XFS_DATA_FORK))) {
750 * This can't really happen unless there's kernel corruption.
754 if (db == mp->m_dirdatablk)
759 * Set the size to the new last block.
761 dp->i_d.di_size = XFS_FSB_TO_B(mp, bno);
762 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);