2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)ufs_lookup.c 8.15 (Berkeley) 6/16/95
39 * $FreeBSD: src/sys/ufs/ufs/ufs_lookup.c,v 1.33.2.7 2001/09/22 19:22:13 iedowse Exp $
40 * $DragonFly: src/sys/vfs/ufs/ufs_lookup.c,v 1.6 2003/08/07 21:17:44 dillon Exp $
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
50 #include <sys/namei.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/sysctl.h>
57 #include <vm/vm_extern.h>
66 #include "ufs_extern.h"
74 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
76 /* true if old FS format...*/
77 #define OFSFMT(vp) ((vp)->v_mount->mnt_maxsymlinklen <= 0)
80 * Convert a component of a pathname into a pointer to a locked inode.
81 * This is a very central and rather complicated routine.
82 * If the file system is not maintained in a strict tree hierarchy,
83 * this can result in a deadlock situation (see comments in code below).
85 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
86 * on whether the name is to be looked up, created, renamed, or deleted.
87 * When CREATE, RENAME, or DELETE is specified, information usable in
88 * creating, renaming, or deleting a directory entry may be calculated.
89 * If flag has LOCKPARENT or'ed into it and the target of the pathname
90 * exists, lookup returns both the target and its parent directory locked.
91 * When creating or renaming and LOCKPARENT is specified, the target may
92 * not be ".". When deleting and LOCKPARENT is specified, the target may
93 * be "."., but the caller must check to ensure it does an vrele and vput
94 * instead of two vputs.
96 * This routine is actually used as VOP_CACHEDLOOKUP method, and the
97 * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
100 * vfs_cache_lookup() performs the following for us:
101 * check that it is a directory
102 * check accessibility of directory
103 * check for modification attempts on read-only mounts
104 * if name found in cache
105 * if at end of path and deleting or creating
109 * return VOP_CACHEDLOOKUP()
111 * Overall outline of ufs_lookup:
113 * search for name in directory, to found or notfound
115 * if creating, return locked directory, leaving info on available slots
118 * if at end of path and deleting, return information to allow delete
119 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
120 * inode and return info to allow rewrite
121 * if not at end, add name to cache; if at end and neither creating
122 * nor deleting, add name to cache
126 struct vop_cachedlookup_args /* {
128 struct vnode **a_vpp;
129 struct componentname *a_cnp;
132 struct vnode *vdp; /* vnode for directory being searched */
133 struct inode *dp; /* inode for directory being searched */
134 struct buf *bp; /* a buffer of directory entries */
135 struct direct *ep; /* the current directory entry */
136 int entryoffsetinblock; /* offset of ep in bp's buffer */
137 enum {NONE, COMPACT, FOUND} slotstatus;
138 doff_t slotoffset; /* offset of area with free space */
139 int slotsize; /* size of area at slotoffset */
140 int slotfreespace; /* amount of space free in slot */
141 int slotneeded; /* size of the entry we're seeking */
142 int numdirpasses; /* strategy for directory search */
143 doff_t endsearch; /* offset to end directory search */
144 doff_t prevoff; /* prev entry dp->i_offset */
145 struct vnode *pdp; /* saved dp during symlink work */
146 struct vnode *tdp; /* returned by VFS_VGET */
147 doff_t enduseful; /* pointer past last used dir slot */
148 u_long bmask; /* block offset mask */
149 int lockparent; /* 1 => lockparent flag is set */
150 int wantparent; /* 1 => wantparent or lockparent flag */
152 struct vnode **vpp = ap->a_vpp;
153 struct componentname *cnp = ap->a_cnp;
154 struct ucred *cred = cnp->cn_cred;
155 int flags = cnp->cn_flags;
156 int nameiop = cnp->cn_nameiop;
157 struct thread *td = cnp->cn_td;
161 cnp->cn_flags &= ~PDIRUNLOCK;
163 * XXX there was a soft-update diff about this I couldn't merge.
164 * I think this was the equiv.
170 lockparent = flags & LOCKPARENT;
171 wantparent = flags & (LOCKPARENT|WANTPARENT);
174 * We now have a segment name to search for, and a directory to search.
176 * Suppress search for slots unless creating
177 * file and at end of pathname, in which case
178 * we watch for a place to put the new file in
179 * case it doesn't already exist.
182 slotfreespace = slotsize = slotneeded = 0;
183 if ((nameiop == CREATE || nameiop == RENAME) &&
184 (flags & ISLASTCN)) {
186 slotneeded = DIRECTSIZ(cnp->cn_namelen);
188 bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
192 * Use dirhash for fast operations on large directories. The logic
193 * to determine whether to hash the directory is contained within
194 * ufsdirhash_build(); a zero return means that it decided to hash
195 * this directory and it successfully built up the hash table.
197 if (ufsdirhash_build(dp) == 0) {
198 /* Look for a free slot if needed. */
199 enduseful = dp->i_size;
200 if (slotstatus != FOUND) {
201 slotoffset = ufsdirhash_findfree(dp, slotneeded,
203 if (slotoffset >= 0) {
204 slotstatus = COMPACT;
205 enduseful = ufsdirhash_enduseful(dp);
207 enduseful = dp->i_size;
210 /* Look up the component. */
212 entryoffsetinblock = 0; /* silence compiler warning */
213 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
214 &dp->i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
216 ep = (struct direct *)((char *)bp->b_data +
217 (dp->i_offset & bmask));
220 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
223 /* Something failed; just do a linear search. */
227 #endif /* UFS_DIRHASH */
229 * If there is cached information on a previous search of
230 * this directory, pick up where we last left off.
231 * We cache only lookups as these are the most common
232 * and have the greatest payoff. Caching CREATE has little
233 * benefit as it usually must search the entire directory
234 * to determine that the entry does not exist. Caching the
235 * location of the last DELETE or RENAME has not reduced
236 * profiling time and hence has been removed in the interest
239 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
240 dp->i_diroff >= dp->i_size) {
241 entryoffsetinblock = 0;
245 dp->i_offset = dp->i_diroff;
246 if ((entryoffsetinblock = dp->i_offset & bmask) &&
247 (error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp)))
250 nchstats.ncs_2passes++;
252 prevoff = dp->i_offset;
253 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
257 while (dp->i_offset < endsearch) {
259 * If necessary, get the next directory block.
261 if ((dp->i_offset & bmask) == 0) {
265 UFS_BLKATOFF(vdp, (off_t)dp->i_offset, NULL, &bp);
268 entryoffsetinblock = 0;
271 * If still looking for a slot, and at a DIRBLKSIZE
272 * boundary, have to start looking for free space again.
274 if (slotstatus == NONE &&
275 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
280 * Get pointer to next entry.
281 * Full validation checks are slow, so we only check
282 * enough to insure forward progress through the
283 * directory. Complete checks can be run by patching
284 * "dirchk" to be true.
286 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
287 if (ep->d_reclen == 0 || ep->d_reclen >
288 DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
289 (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
292 ufs_dirbad(dp, dp->i_offset, "mangled entry");
293 i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
295 entryoffsetinblock += i;
300 * If an appropriate sized slot has not yet been found,
301 * check to see if one is available. Also accumulate space
302 * in the current block so that we can determine if
303 * compaction is viable.
305 if (slotstatus != FOUND) {
306 int size = ep->d_reclen;
309 size -= DIRSIZ(OFSFMT(vdp), ep);
311 if (size >= slotneeded) {
313 slotoffset = dp->i_offset;
314 slotsize = ep->d_reclen;
315 } else if (slotstatus == NONE) {
316 slotfreespace += size;
317 if (slotoffset == -1)
318 slotoffset = dp->i_offset;
319 if (slotfreespace >= slotneeded) {
320 slotstatus = COMPACT;
321 slotsize = dp->i_offset +
322 ep->d_reclen - slotoffset;
329 * Check for a name match.
332 # if (BYTE_ORDER == LITTLE_ENDIAN)
336 namlen = ep->d_namlen;
338 namlen = ep->d_namlen;
340 if (namlen == cnp->cn_namelen &&
341 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
342 !bcmp(cnp->cn_nameptr, ep->d_name,
348 * Save directory entry's inode number and
349 * reclen in ndp->ni_ufs area, and release
352 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
353 ep->d_type == DT_WHT) {
355 slotoffset = dp->i_offset;
356 slotsize = ep->d_reclen;
357 dp->i_reclen = slotsize;
358 enduseful = dp->i_size;
359 ap->a_cnp->cn_flags |= ISWHITEOUT;
363 dp->i_ino = ep->d_ino;
364 dp->i_reclen = ep->d_reclen;
368 prevoff = dp->i_offset;
369 dp->i_offset += ep->d_reclen;
370 entryoffsetinblock += ep->d_reclen;
372 enduseful = dp->i_offset;
376 * If we started in the middle of the directory and failed
377 * to find our target, we must check the beginning as well.
379 if (numdirpasses == 2) {
382 endsearch = dp->i_diroff;
388 * If creating, and at end of pathname and current
389 * directory has not been removed, then can consider
390 * allowing file to be created.
392 if ((nameiop == CREATE || nameiop == RENAME ||
393 (nameiop == DELETE &&
394 (ap->a_cnp->cn_flags & DOWHITEOUT) &&
395 (ap->a_cnp->cn_flags & ISWHITEOUT))) &&
396 (flags & ISLASTCN) && dp->i_effnlink != 0) {
398 * Access for write is interpreted as allowing
399 * creation of files in the directory.
401 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
405 * Return an indication of where the new directory
406 * entry should be put. If we didn't find a slot,
407 * then set dp->i_count to 0 indicating
408 * that the new slot belongs at the end of the
409 * directory. If we found a slot, then the new entry
410 * can be put in the range from dp->i_offset to
411 * dp->i_offset + dp->i_count.
413 if (slotstatus == NONE) {
414 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
416 enduseful = dp->i_offset;
417 } else if (nameiop == DELETE) {
418 dp->i_offset = slotoffset;
419 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
422 dp->i_count = dp->i_offset - prevoff;
424 dp->i_offset = slotoffset;
425 dp->i_count = slotsize;
426 if (enduseful < slotoffset + slotsize)
427 enduseful = slotoffset + slotsize;
429 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
430 dp->i_flag |= IN_CHANGE | IN_UPDATE;
432 * We return with the directory locked, so that
433 * the parameters we set up above will still be
434 * valid if we actually decide to do a direnter().
435 * We return ni_vp == NULL to indicate that the entry
436 * does not currently exist; we leave a pointer to
437 * the (locked) directory inode in ndp->ni_dvp.
438 * The pathname buffer is saved so that the name
439 * can be obtained later.
441 * NB - if the directory is unlocked, then this
442 * information cannot be used.
444 cnp->cn_flags |= SAVENAME;
446 VOP_UNLOCK(vdp, 0, td);
447 cnp->cn_flags |= PDIRUNLOCK;
449 return (EJUSTRETURN);
452 * Insert name into cache (as non-existent) if appropriate.
454 if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
455 cache_enter(vdp, *vpp, cnp);
459 if (numdirpasses == 2)
460 nchstats.ncs_pass2++;
462 * Check that directory length properly reflects presence
465 if (dp->i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
466 ufs_dirbad(dp, dp->i_offset, "i_size too small");
467 dp->i_size = dp->i_offset + DIRSIZ(OFSFMT(vdp), ep);
468 dp->i_flag |= IN_CHANGE | IN_UPDATE;
473 * Found component in pathname.
474 * If the final component of path name, save information
475 * in the cache as to where the entry was found.
477 if ((flags & ISLASTCN) && nameiop == LOOKUP)
478 dp->i_diroff = dp->i_offset &~ (DIRBLKSIZ - 1);
481 * If deleting, and at end of pathname, return
482 * parameters which can be used to remove file.
483 * If the wantparent flag isn't set, we return only
484 * the directory (in ndp->ni_dvp), otherwise we go
485 * on and lock the inode, being careful with ".".
487 if (nameiop == DELETE && (flags & ISLASTCN)) {
489 * Write access to directory required to delete files.
491 error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td);
495 * Return pointer to current entry in dp->i_offset,
496 * and distance past previous entry (if there
497 * is a previous entry in this block) in dp->i_count.
498 * Save directory inode pointer in ndp->ni_dvp for dirremove().
500 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
503 dp->i_count = dp->i_offset - prevoff;
504 if (dp->i_number == dp->i_ino) {
509 if (flags & ISDOTDOT)
510 VOP_UNLOCK(vdp, 0, td); /* race to get the inode */
511 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
512 if (flags & ISDOTDOT) {
513 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
514 cnp->cn_flags |= PDIRUNLOCK;
519 * If directory is "sticky", then user must own
520 * the directory, or the file in it, else she
521 * may not delete it (unless she's root). This
522 * implements append-only directories.
524 if ((dp->i_mode & ISVTX) &&
526 cred->cr_uid != dp->i_uid &&
527 VTOI(tdp)->i_uid != cred->cr_uid) {
533 VOP_UNLOCK(vdp, 0, td);
534 cnp->cn_flags |= PDIRUNLOCK;
540 * If rewriting (RENAME), return the inode and the
541 * information required to rewrite the present directory
542 * Must get inode of directory entry to verify it's a
543 * regular file, or empty directory.
545 if (nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
546 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_td)) != 0)
549 * Careful about locking second inode.
550 * This can only occur if the target is ".".
552 if (dp->i_number == dp->i_ino)
554 if (flags & ISDOTDOT)
555 VOP_UNLOCK(vdp, 0, td); /* race to get the inode */
556 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
557 if (flags & ISDOTDOT) {
558 if (vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY, td) != 0)
559 cnp->cn_flags |= PDIRUNLOCK;
564 cnp->cn_flags |= SAVENAME;
566 VOP_UNLOCK(vdp, 0, td);
567 cnp->cn_flags |= PDIRUNLOCK;
573 * Step through the translation in the name. We do not `vput' the
574 * directory because we may need it again if a symbolic link
575 * is relative to the current directory. Instead we save it
576 * unlocked as "pdp". We must get the target inode before unlocking
577 * the directory to insure that the inode will not be removed
578 * before we get it. We prevent deadlock by always fetching
579 * inodes from the root, moving down the directory tree. Thus
580 * when following backward pointers ".." we must unlock the
581 * parent directory before getting the requested directory.
582 * There is a potential race condition here if both the current
583 * and parent directories are removed before the VFS_VGET for the
584 * inode associated with ".." returns. We hope that this occurs
585 * infrequently since we cannot avoid this race condition without
586 * implementing a sophisticated deadlock detection algorithm.
587 * Note also that this simple deadlock detection scheme will not
588 * work if the file system has any hard links other than ".."
589 * that point backwards in the directory structure.
592 if (flags & ISDOTDOT) {
593 VOP_UNLOCK(pdp, 0, td); /* race to get the inode */
594 cnp->cn_flags |= PDIRUNLOCK;
595 if ((error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp)) != 0) {
596 if (vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, td) == 0)
597 cnp->cn_flags &= ~PDIRUNLOCK;
600 if (lockparent && (flags & ISLASTCN)) {
601 if ((error = vn_lock(pdp, LK_EXCLUSIVE, td)) != 0) {
605 cnp->cn_flags &= ~PDIRUNLOCK;
608 } else if (dp->i_number == dp->i_ino) {
609 VREF(vdp); /* we want ourself, ie "." */
612 error = VFS_VGET(vdp->v_mount, dp->i_ino, &tdp);
615 if (!lockparent || !(flags & ISLASTCN)) {
616 VOP_UNLOCK(pdp, 0, td);
617 cnp->cn_flags |= PDIRUNLOCK;
623 * Insert name into cache if appropriate.
625 if (cnp->cn_flags & MAKEENTRY)
626 cache_enter(vdp, *vpp, cnp);
631 ufs_dirbad(ip, offset, how)
638 mp = ITOV(ip)->v_mount;
639 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
640 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
641 if ((mp->mnt_flag & MNT_RDONLY) == 0)
642 panic("ufs_dirbad: bad dir");
646 * Do consistency checking on a directory entry:
647 * record length must be multiple of 4
648 * entry must fit in rest of its DIRBLKSIZ block
649 * record must be large enough to contain entry
650 * name is not longer than MAXNAMLEN
651 * name must be as long as advertised, and null terminated
654 ufs_dirbadentry(dp, ep, entryoffsetinblock)
657 int entryoffsetinblock;
662 # if (BYTE_ORDER == LITTLE_ENDIAN)
666 namlen = ep->d_namlen;
668 namlen = ep->d_namlen;
670 if ((ep->d_reclen & 0x3) != 0 ||
671 ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
672 ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > MAXNAMLEN) {
674 printf("First bad\n");
679 for (i = 0; i < namlen; i++)
680 if (ep->d_name[i] == '\0') {
682 printf("Second bad\n");
693 * Construct a new directory entry after a call to namei, using the
694 * parameters that it left in the componentname argument cnp. The
695 * argument ip is the inode to which the new directory entry will refer.
698 ufs_makedirentry(ip, cnp, newdirp)
700 struct componentname *cnp;
701 struct direct *newdirp;
705 if ((cnp->cn_flags & SAVENAME) == 0)
706 panic("ufs_makedirentry: missing name");
708 newdirp->d_ino = ip->i_number;
709 newdirp->d_namlen = cnp->cn_namelen;
710 bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
711 if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
712 newdirp->d_type = IFTODT(ip->i_mode);
715 # if (BYTE_ORDER == LITTLE_ENDIAN)
716 { u_char tmp = newdirp->d_namlen;
717 newdirp->d_namlen = newdirp->d_type;
718 newdirp->d_type = tmp; }
724 * Write a directory entry after a call to namei, using the parameters
725 * that it left in nameidata. The argument dirp is the new directory
726 * entry contents. Dvp is a pointer to the directory to be written,
727 * which was left locked by namei. Remaining parameters (dp->i_offset,
728 * dp->i_count) indicate how the space for the new entry is to be obtained.
729 * Non-null bp indicates that a directory is being created (for the
730 * soft dependency code).
733 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp)
737 struct componentname *cnp;
738 struct buf *newdirbp;
741 struct thread *td = curthread; /* XXX */
746 struct direct *ep, *nep;
747 int error, ret, blkoff, loc, spacefree, flags;
750 KKASSERT(td->td_proc); /* YYY use/require cred passed in cnp? */
751 cred = td->td_proc->p_ucred;
754 newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
756 if (dp->i_count == 0) {
758 * If dp->i_count is 0, then namei could find no
759 * space in the directory. Here, dp->i_offset will
760 * be on a directory block boundary and we will write the
761 * new entry into a fresh block.
763 if (dp->i_offset & (DIRBLKSIZ - 1))
764 panic("ufs_direnter: newblk");
766 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
768 if ((error = VOP_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
769 cred, flags, &bp)) != 0) {
770 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
774 dp->i_size = dp->i_offset + DIRBLKSIZ;
775 dp->i_flag |= IN_CHANGE | IN_UPDATE;
776 vnode_pager_setsize(dvp, (u_long)dp->i_size);
777 dirp->d_reclen = DIRBLKSIZ;
778 blkoff = dp->i_offset &
779 (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
780 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
782 if (dp->i_dirhash != NULL) {
783 ufsdirhash_newblk(dp, dp->i_offset);
784 ufsdirhash_add(dp, dirp, dp->i_offset);
785 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
789 if (DOINGSOFTDEP(dvp)) {
791 * Ensure that the entire newly allocated block is a
792 * valid directory so that future growth within the
793 * block does not have to ensure that the block is
794 * written before the inode.
797 while (blkoff < bp->b_bcount) {
799 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
802 softdep_setup_directory_add(bp, dp, dp->i_offset,
803 dirp->d_ino, newdirbp);
805 return (UFS_UPDATE(dvp, 0));
807 if (DOINGASYNC(dvp)) {
809 return (UFS_UPDATE(dvp, 0));
811 error = VOP_BWRITE(bp->b_vp, bp);
812 ret = UFS_UPDATE(dvp, 1);
819 * If dp->i_count is non-zero, then namei found space for the new
820 * entry in the range dp->i_offset to dp->i_offset + dp->i_count
821 * in the directory. To use this space, we may have to compact
822 * the entries located there, by copying them together towards the
823 * beginning of the block, leaving the free space in one usable
828 * Increase size of directory if entry eats into new space.
829 * This should never push the size past a new multiple of
832 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
834 if (dp->i_offset + dp->i_count > dp->i_size)
835 dp->i_size = dp->i_offset + dp->i_count;
837 * Get the block containing the space for the new directory entry.
839 error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
841 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
846 * Find space for the new entry. In the simple case, the entry at
847 * offset base will have the space. If it does not, then namei
848 * arranged that compacting the region dp->i_offset to
849 * dp->i_offset + dp->i_count would yield the space.
851 ep = (struct direct *)dirbuf;
852 dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
853 spacefree = ep->d_reclen - dsize;
854 for (loc = ep->d_reclen; loc < dp->i_count; ) {
855 nep = (struct direct *)(dirbuf + loc);
857 /* Trim the existing slot (NB: dsize may be zero). */
858 ep->d_reclen = dsize;
859 ep = (struct direct *)((char *)ep + dsize);
861 /* Read nep->d_reclen now as the bcopy() may clobber it. */
862 loc += nep->d_reclen;
863 if (nep->d_ino == 0) {
865 * A mid-block unused entry. Such entries are
866 * never created by the kernel, but fsck_ffs
867 * can create them (and it doesn't fix them).
869 * Add up the free space, and initialise the
870 * relocated entry since we don't bcopy it.
872 spacefree += nep->d_reclen;
877 dsize = DIRSIZ(OFSFMT(dvp), nep);
878 spacefree += nep->d_reclen - dsize;
880 if (dp->i_dirhash != NULL)
881 ufsdirhash_move(dp, nep,
882 dp->i_offset + ((char *)nep - dirbuf),
883 dp->i_offset + ((char *)ep - dirbuf));
885 if (DOINGSOFTDEP(dvp))
886 softdep_change_directoryentry_offset(dp, dirbuf,
887 (caddr_t)nep, (caddr_t)ep, dsize);
889 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
892 * Here, `ep' points to a directory entry containing `dsize' in-use
893 * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
894 * then the entry is completely unused (dsize == 0). The value
895 * of ep->d_reclen is always indeterminate.
897 * Update the pointer fields in the previous entry (if any),
898 * copy in the new entry, and write out the block.
900 if (ep->d_ino == 0 ||
901 (ep->d_ino == WINO &&
902 bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
903 if (spacefree + dsize < newentrysize)
904 panic("ufs_direnter: compact1");
905 dirp->d_reclen = spacefree + dsize;
907 if (spacefree < newentrysize)
908 panic("ufs_direnter: compact2");
909 dirp->d_reclen = spacefree;
910 ep->d_reclen = dsize;
911 ep = (struct direct *)((char *)ep + dsize);
914 if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
915 dirp->d_reclen == spacefree))
916 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
918 bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
920 if (dp->i_dirhash != NULL)
921 ufsdirhash_checkblock(dp, dirbuf -
922 (dp->i_offset & (DIRBLKSIZ - 1)),
923 dp->i_offset & ~(DIRBLKSIZ - 1));
926 if (DOINGSOFTDEP(dvp)) {
927 softdep_setup_directory_add(bp, dp,
928 dp->i_offset + (caddr_t)ep - dirbuf, dirp->d_ino, newdirbp);
931 if (DOINGASYNC(dvp)) {
938 dp->i_flag |= IN_CHANGE | IN_UPDATE;
940 * If all went well, and the directory can be shortened, proceed
941 * with the truncation. Note that we have to unlock the inode for
942 * the entry that we just entered, as the truncation may need to
943 * lock other inodes which can lead to deadlock if we also hold a
944 * lock on the newly entered node.
946 if (error == 0 && dp->i_endoff && dp->i_endoff < dp->i_size) {
948 VOP_UNLOCK(tvp, 0, td);
950 if (dp->i_dirhash != NULL)
951 ufsdirhash_dirtrunc(dp, dp->i_endoff);
953 (void) UFS_TRUNCATE(dvp, (off_t)dp->i_endoff, IO_SYNC, cred, td);
955 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, td);
961 * Remove a directory entry after a call to namei, using
962 * the parameters which it left in nameidata. The entry
963 * dp->i_offset contains the offset into the directory of the
964 * entry to be eliminated. The dp->i_count field contains the
965 * size of the previous record in the directory. If this
966 * is 0, the first entry is being deleted, so we need only
967 * zero the inode number to mark the entry as free. If the
968 * entry is not the first in the directory, we must reclaim
969 * the space of the now empty record by adding the record size
970 * to the size of the previous entry.
973 ufs_dirremove(dvp, ip, flags, isrmdir)
986 if (flags & DOWHITEOUT) {
988 * Whiteout entry: set d_ino to WINO.
991 UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
998 if ((error = UFS_BLKATOFF(dvp,
999 (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
1003 * Remove the dirhash entry. This is complicated by the fact
1004 * that `ep' is the previous entry when dp->i_count != 0.
1006 if (dp->i_dirhash != NULL)
1007 ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
1008 (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
1010 if (dp->i_count == 0) {
1012 * First entry in block: set d_ino to zero.
1017 * Collapse new free space into previous entry.
1019 ep->d_reclen += dp->i_reclen;
1022 if (dp->i_dirhash != NULL)
1023 ufsdirhash_checkblock(dp, (char *)ep -
1024 ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
1025 dp->i_offset & ~(DIRBLKSIZ - 1));
1028 if (DOINGSOFTDEP(dvp)) {
1031 softdep_change_linkcnt(ip);
1032 softdep_setup_remove(bp, dp, ip, isrmdir);
1034 if (softdep_slowdown(dvp)) {
1035 error = VOP_BWRITE(bp->b_vp, bp);
1044 ip->i_flag |= IN_CHANGE;
1046 if (flags & DOWHITEOUT)
1047 error = VOP_BWRITE(bp->b_vp, bp);
1048 else if (DOINGASYNC(dvp) && dp->i_count != 0) {
1052 error = bowrite(bp);
1054 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1059 * Rewrite an existing directory entry to point at the inode
1060 * supplied. The parameters describing the directory entry are
1061 * set up by a call to namei.
1064 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
1065 struct inode *dp, *oip;
1072 struct vnode *vdp = ITOV(dp);
1075 error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
1078 ep->d_ino = newinum;
1080 ep->d_type = newtype;
1082 if (DOINGSOFTDEP(vdp)) {
1083 softdep_change_linkcnt(oip);
1084 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
1088 oip->i_flag |= IN_CHANGE;
1089 if (DOINGASYNC(vdp)) {
1093 error = bowrite(bp);
1096 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1101 * Check if a directory is empty or not.
1102 * Inode supplied must be locked.
1104 * Using a struct dirtemplate here is not precisely
1105 * what we want, but better than using a struct direct.
1107 * NB: does not handle corrupted directories.
1110 ufs_dirempty(ip, parentino, cred)
1116 struct dirtemplate dbuf;
1117 struct direct *dp = (struct direct *)&dbuf;
1118 int error, count, namlen;
1119 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
1121 for (off = 0; off < ip->i_size; off += dp->d_reclen) {
1122 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
1123 UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
1125 * Since we read MINDIRSIZ, residual must
1126 * be 0 unless we're at end of file.
1128 if (error || count != 0)
1130 /* avoid infinite loops */
1131 if (dp->d_reclen == 0)
1133 /* skip empty entries */
1134 if (dp->d_ino == 0 || dp->d_ino == WINO)
1136 /* accept only "." and ".." */
1137 # if (BYTE_ORDER == LITTLE_ENDIAN)
1138 if (OFSFMT(ITOV(ip)))
1139 namlen = dp->d_type;
1141 namlen = dp->d_namlen;
1143 namlen = dp->d_namlen;
1147 if (dp->d_name[0] != '.')
1150 * At this point namlen must be 1 or 2.
1151 * 1 implies ".", 2 implies ".." if second
1154 if (namlen == 1 && dp->d_ino == ip->i_number)
1156 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
1164 * Check if source directory is in the path of the target directory.
1165 * Target is supplied locked, source is unlocked.
1166 * The target is always vput before returning.
1169 ufs_checkpath(source, target, cred)
1170 struct inode *source, *target;
1174 int error, rootino, namlen;
1175 struct dirtemplate dirbuf;
1178 if (target->i_number == source->i_number) {
1184 if (target->i_number == rootino)
1188 if (vp->v_type != VDIR) {
1192 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1193 sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1194 IO_NODELOCKED, cred, (int *)0, NULL);
1197 # if (BYTE_ORDER == LITTLE_ENDIAN)
1199 namlen = dirbuf.dotdot_type;
1201 namlen = dirbuf.dotdot_namlen;
1203 namlen = dirbuf.dotdot_namlen;
1206 dirbuf.dotdot_name[0] != '.' ||
1207 dirbuf.dotdot_name[1] != '.') {
1211 if (dirbuf.dotdot_ino == source->i_number) {
1215 if (dirbuf.dotdot_ino == rootino)
1218 error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino, &vp);
1226 if (error == ENOTDIR)
1227 printf("checkpath: .. not a directory\n");