1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */
4 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * tmpfs vnode interface.
36 #include <sys/cdefs.h>
38 #include <sys/kernel.h>
39 #include <sys/kern_syscall.h>
40 #include <sys/param.h>
41 #include <sys/fcntl.h>
42 #include <sys/lockf.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/sfbuf.h>
49 #include <sys/systm.h>
50 #include <sys/unistd.h>
51 #include <sys/vfsops.h>
52 #include <sys/vnode.h>
54 #include <sys/mplock2.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
61 #include <vfs/fifofs/fifo.h>
62 #include <vfs/tmpfs/tmpfs_vnops.h>
63 #include <vfs/tmpfs/tmpfs.h>
65 MALLOC_DECLARE(M_TMPFS);
67 /* --------------------------------------------------------------------- */
70 tmpfs_nresolve(struct vop_nresolve_args *v)
72 struct vnode *dvp = v->a_dvp;
73 struct vnode *vp = NULL;
74 struct namecache *ncp = v->a_nch->ncp;
75 struct tmpfs_node *tnode;
78 struct tmpfs_dirent *de;
79 struct tmpfs_node *dnode;
81 dnode = VP_TO_TMPFS_DIR(dvp);
83 de = tmpfs_dir_lookup(dnode, NULL, ncp);
88 * Allocate a vnode for the node we found.
91 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
92 LK_EXCLUSIVE | LK_RETRY, &vp);
100 * Store the result of this lookup in the cache. Avoid this if the
101 * request was for creation, as it does not improve timings on
106 cache_setvp(v->a_nch, vp);
108 } else if (error == ENOENT) {
109 cache_setvp(v->a_nch, NULL);
115 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
117 struct vnode *dvp = v->a_dvp;
118 struct vnode **vpp = v->a_vpp;
119 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
120 struct ucred *cred = v->a_cred;
124 /* Check accessibility of requested node as a first step. */
125 error = VOP_ACCESS(dvp, VEXEC, cred);
129 if (dnode->tn_dir.tn_parent != NULL) {
130 /* Allocate a new vnode on the matching entry. */
131 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
132 LK_EXCLUSIVE | LK_RETRY, vpp);
138 return (*vpp == NULL) ? ENOENT : 0;
141 /* --------------------------------------------------------------------- */
144 tmpfs_ncreate(struct vop_ncreate_args *v)
146 struct vnode *dvp = v->a_dvp;
147 struct vnode **vpp = v->a_vpp;
148 struct namecache *ncp = v->a_nch->ncp;
149 struct vattr *vap = v->a_vap;
150 struct ucred *cred = v->a_cred;
153 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
155 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
157 cache_setunresolved(v->a_nch);
158 cache_setvp(v->a_nch, *vpp);
163 /* --------------------------------------------------------------------- */
166 tmpfs_nmknod(struct vop_nmknod_args *v)
168 struct vnode *dvp = v->a_dvp;
169 struct vnode **vpp = v->a_vpp;
170 struct namecache *ncp = v->a_nch->ncp;
171 struct vattr *vap = v->a_vap;
172 struct ucred *cred = v->a_cred;
175 if (vap->va_type != VBLK && vap->va_type != VCHR &&
176 vap->va_type != VFIFO)
179 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
181 cache_setunresolved(v->a_nch);
182 cache_setvp(v->a_nch, *vpp);
188 /* --------------------------------------------------------------------- */
191 tmpfs_open(struct vop_open_args *v)
193 struct vnode *vp = v->a_vp;
194 int mode = v->a_mode;
197 struct tmpfs_node *node;
199 node = VP_TO_TMPFS_NODE(vp);
201 /* The file is still active but all its names have been removed
202 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
203 * it is about to die. */
204 if (node->tn_links < 1)
207 /* If the file is marked append-only, deny write requests. */
208 if ((node->tn_flags & APPEND) &&
209 (mode & (FWRITE | O_APPEND)) == FWRITE) {
212 return (vop_stdopen(v));
217 /* --------------------------------------------------------------------- */
220 tmpfs_close(struct vop_close_args *v)
222 struct vnode *vp = v->a_vp;
223 struct tmpfs_node *node;
225 node = VP_TO_TMPFS_NODE(vp);
227 if (node->tn_links > 0) {
228 /* Update node times. No need to do it if the node has
229 * been deleted, because it will vanish after we return. */
233 return vop_stdclose(v);
236 /* --------------------------------------------------------------------- */
239 tmpfs_access(struct vop_access_args *v)
241 struct vnode *vp = v->a_vp;
243 struct tmpfs_node *node;
245 node = VP_TO_TMPFS_NODE(vp);
247 switch (vp->v_type) {
253 if (VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
273 if (VWRITE && node->tn_flags & IMMUTABLE) {
278 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
285 /* --------------------------------------------------------------------- */
288 tmpfs_getattr(struct vop_getattr_args *v)
290 struct vnode *vp = v->a_vp;
291 struct vattr *vap = v->a_vap;
292 struct tmpfs_node *node;
294 node = VP_TO_TMPFS_NODE(vp);
298 vap->va_type = vp->v_type;
299 vap->va_mode = node->tn_mode;
300 vap->va_nlink = node->tn_links;
301 vap->va_uid = node->tn_uid;
302 vap->va_gid = node->tn_gid;
303 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
304 vap->va_fileid = node->tn_id;
305 vap->va_size = node->tn_size;
306 vap->va_blocksize = PAGE_SIZE;
307 vap->va_atime.tv_sec = node->tn_atime;
308 vap->va_atime.tv_nsec = node->tn_atimensec;
309 vap->va_mtime.tv_sec = node->tn_mtime;
310 vap->va_mtime.tv_nsec = node->tn_mtimensec;
311 vap->va_ctime.tv_sec = node->tn_ctime;
312 vap->va_ctime.tv_nsec = node->tn_ctimensec;
313 vap->va_gen = node->tn_gen;
314 vap->va_flags = node->tn_flags;
315 if (vp->v_type == VBLK || vp->v_type == VCHR)
317 vap->va_rmajor = umajor(node->tn_rdev);
318 vap->va_rminor = uminor(node->tn_rdev);
320 vap->va_bytes = round_page(node->tn_size);
326 /* --------------------------------------------------------------------- */
329 tmpfs_setattr(struct vop_setattr_args *v)
331 struct vnode *vp = v->a_vp;
332 struct vattr *vap = v->a_vap;
333 struct ucred *cred = v->a_cred;
336 if (error == 0 && (vap->va_flags != VNOVAL))
337 error = tmpfs_chflags(vp, vap->va_flags, cred);
339 if (error == 0 && (vap->va_size != VNOVAL))
340 error = tmpfs_chsize(vp, vap->va_size, cred);
342 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
343 vap->va_gid != (gid_t)VNOVAL)) {
344 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
347 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
348 error = tmpfs_chmod(vp, vap->va_mode, cred);
350 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
351 vap->va_atime.tv_nsec != VNOVAL) ||
352 (vap->va_mtime.tv_sec != VNOVAL &&
353 vap->va_mtime.tv_nsec != VNOVAL) )) {
354 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
355 vap->va_vaflags, cred);
358 /* Update the node times. We give preference to the error codes
359 * generated by this function rather than the ones that may arise
360 * from tmpfs_update. */
366 /* --------------------------------------------------------------------- */
369 * fsync is usually a NOP, but we must take action when unmounting or
373 tmpfs_fsync(struct vop_fsync_args *v)
375 struct tmpfs_mount *tmp;
376 struct tmpfs_node *node;
377 struct vnode *vp = v->a_vp;
379 tmp = VFS_TO_TMPFS(vp->v_mount);
380 node = VP_TO_TMPFS_NODE(vp);
383 if (vp->v_type == VREG) {
384 if (tmp->tm_flags & TMPFS_FLAG_UNMOUNTING) {
385 tmpfs_truncate(vp, 0);
386 } else if (vp->v_flag & VRECLAIMED) {
387 if (node->tn_links == 0)
388 tmpfs_truncate(vp, 0);
390 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
396 /* --------------------------------------------------------------------- */
399 tmpfs_read (struct vop_read_args *ap)
402 struct vnode *vp = ap->a_vp;
403 struct uio *uio = ap->a_uio;
404 struct tmpfs_node *node;
412 if (uio->uio_resid == 0) {
416 node = VP_TO_TMPFS_NODE(vp);
418 if (uio->uio_offset < 0)
420 if (vp->v_type != VREG)
424 if(curthread->td_mpcount)
432 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
434 * Use buffer cache I/O (via tmpfs_strategy)
436 offset = (size_t)uio->uio_offset & BMASK;
437 base_offset = (off_t)uio->uio_offset - offset;
438 bp = getcacheblk(vp, base_offset);
441 if (got_mplock == 0) {
446 error = bread(vp, base_offset, BSIZE, &bp);
449 kprintf("tmpfs_read bread error %d\n", error);
454 if (got_mplock == 0) {
460 * Figure out how many bytes we can actually copy this loop.
462 len = BSIZE - offset;
463 if (len > uio->uio_resid)
464 len = uio->uio_resid;
465 if (len > node->tn_size - uio->uio_offset)
466 len = (size_t)(node->tn_size - uio->uio_offset);
468 error = uiomove((char *)bp->b_data + offset, len, uio);
471 kprintf("tmpfs_read uiomove error %d\n", error);
479 TMPFS_NODE_LOCK(node);
480 node->tn_status |= TMPFS_NODE_ACCESSED;
481 TMPFS_NODE_UNLOCK(node);
487 tmpfs_write (struct vop_write_args *ap)
490 struct vnode *vp = ap->a_vp;
491 struct uio *uio = ap->a_uio;
492 struct thread *td = uio->uio_td;
493 struct tmpfs_node *node;
505 if (uio->uio_resid == 0) {
509 node = VP_TO_TMPFS_NODE(vp);
511 if (vp->v_type != VREG)
514 oldsize = node->tn_size;
515 if (ap->a_ioflag & IO_APPEND)
516 uio->uio_offset = node->tn_size;
519 * Check for illegal write offsets.
521 if (uio->uio_offset + uio->uio_resid >
522 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
525 if (vp->v_type == VREG && td != NULL) {
526 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
529 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
530 ksignal(td->td_proc, SIGXFSZ);
537 * Extend the file's size if necessary
539 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
542 if (curthread->td_mpcount) {
551 while (uio->uio_resid > 0) {
553 * Use buffer cache I/O (via tmpfs_strategy)
555 offset = (size_t)uio->uio_offset & BMASK;
556 base_offset = (off_t)uio->uio_offset - offset;
557 len = BSIZE - offset;
558 if (len > uio->uio_resid)
559 len = uio->uio_resid;
561 if ((uio->uio_offset + len) > node->tn_size) {
562 trivial = (uio->uio_offset <= node->tn_size);
563 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
569 * Read to fill in any gaps. Theoretically we could
570 * optimize this if the write covers the entire buffer
571 * and is not a UIO_NOCOPY write, however this can lead
572 * to a security violation exposing random kernel memory
573 * (whatever junk was in the backing VM pages before).
575 * So just use bread() to do the right thing.
577 error = bread(vp, base_offset, BSIZE, &bp);
578 error = uiomove((char *)bp->b_data + offset, len, uio);
580 kprintf("tmpfs_write uiomove error %d\n", error);
585 if (uio->uio_offset > node->tn_size)
586 node->tn_size = uio->uio_offset;
589 * The data has been loaded into the buffer, write it out.
591 * We want tmpfs to be able to use all available ram, not
592 * just the buffer cache, so if not explicitly paging we
593 * use buwrite() to leave the buffer clean but mark all the
594 * VM pages valid+dirty.
596 * When the kernel is paging, either via normal pageout
597 * operation or when cleaning the object during a recycle,
598 * the underlying VM pages are going to get thrown away
599 * so we MUST write them to swap.
601 * XXX unfortunately this catches msync() system calls too
604 if (ap->a_ioflag & IO_SYNC) {
606 } else if ((ap->a_ioflag & IO_ASYNC) ||
607 (uio->uio_segflg == UIO_NOCOPY)) {
614 kprintf("tmpfs_write bwrite error %d\n", error);
624 (void)tmpfs_reg_resize(vp, oldsize, trivial);
628 TMPFS_NODE_LOCK(node);
629 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
630 (extended? TMPFS_NODE_CHANGED : 0);
632 if (node->tn_mode & (S_ISUID | S_ISGID)) {
633 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
634 node->tn_mode &= ~(S_ISUID | S_ISGID);
636 TMPFS_NODE_UNLOCK(node);
642 tmpfs_advlock (struct vop_advlock_args *ap)
644 struct tmpfs_node *node;
645 struct vnode *vp = ap->a_vp;
647 node = VP_TO_TMPFS_NODE(vp);
649 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
654 tmpfs_strategy(struct vop_strategy_args *ap)
656 struct bio *bio = ap->a_bio;
657 struct buf *bp = bio->bio_buf;
658 struct vnode *vp = ap->a_vp;
659 struct tmpfs_node *node;
662 if (vp->v_type != VREG) {
663 bp->b_resid = bp->b_bcount;
664 bp->b_flags |= B_ERROR | B_INVAL;
665 bp->b_error = EINVAL;
670 node = VP_TO_TMPFS_NODE(vp);
672 uobj = node->tn_reg.tn_aobj;
675 * Call swap_pager_strategy to read or write between the VM
676 * object and the buffer cache.
678 swap_pager_strategy(uobj, bio);
684 tmpfs_bmap(struct vop_bmap_args *ap)
686 if (ap->a_doffsetp != NULL)
687 *ap->a_doffsetp = ap->a_loffset;
688 if (ap->a_runp != NULL)
690 if (ap->a_runb != NULL)
696 /* --------------------------------------------------------------------- */
699 tmpfs_nremove(struct vop_nremove_args *v)
701 struct vnode *dvp = v->a_dvp;
702 struct namecache *ncp = v->a_nch->ncp;
705 struct tmpfs_dirent *de;
706 struct tmpfs_mount *tmp;
707 struct tmpfs_node *dnode;
708 struct tmpfs_node *node;
711 * We have to acquire the vp from v->a_nch because
712 * we will likely unresolve the namecache entry, and
713 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
714 * sequence to recover space from the file.
716 error = cache_vref(v->a_nch, v->a_cred, &vp);
717 KKASSERT(error == 0);
719 if (vp->v_type == VDIR) {
724 dnode = VP_TO_TMPFS_DIR(dvp);
725 node = VP_TO_TMPFS_NODE(vp);
726 tmp = VFS_TO_TMPFS(vp->v_mount);
727 de = tmpfs_dir_lookup(dnode, node, ncp);
733 /* Files marked as immutable or append-only cannot be deleted. */
734 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
735 (dnode->tn_flags & APPEND)) {
740 /* Remove the entry from the directory; as it is a file, we do not
741 * have to change the number of hard links of the directory. */
742 tmpfs_dir_detach(dnode, de);
744 /* Free the directory entry we just deleted. Note that the node
745 * referred by it will not be removed until the vnode is really
747 tmpfs_free_dirent(tmp, de);
749 if (node->tn_links > 0) {
750 TMPFS_NODE_LOCK(node);
751 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
753 TMPFS_NODE_UNLOCK(node);
756 cache_setunresolved(v->a_nch);
757 cache_setvp(v->a_nch, NULL);
758 /*cache_inval_vp(vp, CINV_DESTROY);*/
767 /* --------------------------------------------------------------------- */
770 tmpfs_nlink(struct vop_nlink_args *v)
772 struct vnode *dvp = v->a_dvp;
773 struct vnode *vp = v->a_vp;
774 struct namecache *ncp = v->a_nch->ncp;
775 struct tmpfs_dirent *de;
776 struct tmpfs_node *node;
777 struct tmpfs_node *dnode;
780 KKASSERT(dvp != vp); /* XXX When can this be false? */
782 node = VP_TO_TMPFS_NODE(vp);
783 dnode = VP_TO_TMPFS_NODE(dvp);
785 /* XXX: Why aren't the following two tests done by the caller? */
787 /* Hard links of directories are forbidden. */
788 if (vp->v_type == VDIR) {
793 /* Cannot create cross-device links. */
794 if (dvp->v_mount != vp->v_mount) {
799 /* Ensure that we do not overflow the maximum number of links imposed
801 KKASSERT(node->tn_links <= LINK_MAX);
802 if (node->tn_links == LINK_MAX) {
807 /* We cannot create links of files marked immutable or append-only. */
808 if (node->tn_flags & (IMMUTABLE | APPEND)) {
813 /* Allocate a new directory entry to represent the node. */
814 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
815 ncp->nc_name, ncp->nc_nlen, &de);
819 /* Insert the new directory entry into the appropriate directory. */
820 tmpfs_dir_attach(dnode, de);
822 /* vp link count has changed, so update node times. */
824 TMPFS_NODE_LOCK(node);
825 node->tn_status |= TMPFS_NODE_CHANGED;
826 TMPFS_NODE_UNLOCK(node);
829 cache_setunresolved(v->a_nch);
830 cache_setvp(v->a_nch, vp);
837 /* --------------------------------------------------------------------- */
840 tmpfs_nrename(struct vop_nrename_args *v)
842 struct vnode *fdvp = v->a_fdvp;
843 struct namecache *fncp = v->a_fnch->ncp;
844 struct vnode *fvp = fncp->nc_vp;
845 struct vnode *tdvp = v->a_tdvp;
846 struct namecache *tncp = v->a_tnch->ncp;
847 struct vnode *tvp = tncp->nc_vp;
848 struct tmpfs_dirent *de;
849 struct tmpfs_mount *tmp;
850 struct tmpfs_node *fdnode;
851 struct tmpfs_node *fnode;
852 struct tmpfs_node *tnode;
853 struct tmpfs_node *tdnode;
858 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
860 /* Disallow cross-device renames.
861 * XXX Why isn't this done by the caller? */
862 if (fvp->v_mount != tdvp->v_mount ||
863 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
868 tmp = VFS_TO_TMPFS(tdvp->v_mount);
869 tdnode = VP_TO_TMPFS_DIR(tdvp);
871 /* If source and target are the same file, there is nothing to do. */
877 fdnode = VP_TO_TMPFS_DIR(fdvp);
878 fnode = VP_TO_TMPFS_NODE(fvp);
879 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
881 /* Avoid manipulating '.' and '..' entries. */
886 KKASSERT(de->td_node == fnode);
889 * If replacing an entry in the target directory and that entry
890 * is a directory, it must be empty.
892 * Kern_rename gurantees the destination to be a directory
893 * if the source is one (it does?).
896 KKASSERT(tnode != NULL);
898 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
899 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
904 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
905 if (tnode->tn_size > 0) {
909 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
912 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
916 KKASSERT(fnode->tn_type != VDIR &&
917 tnode->tn_type != VDIR);
921 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
922 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
928 * Ensure that we have enough memory to hold the new name, if it
931 if (fncp->nc_nlen != tncp->nc_nlen ||
932 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
933 newname = kmalloc(tncp->nc_nlen + 1, M_TMPFSNAME, M_WAITOK);
934 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
935 newname[tncp->nc_nlen] = '\0';
941 * Unlink entry from source directory. Note that the kernel has
942 * already checked for illegal recursion cases (renaming a directory
943 * into a subdirectory of itself).
945 if (fdnode != tdnode)
946 tmpfs_dir_detach(fdnode, de);
949 * Handle any name change. Swap with newname, we will
950 * deallocate it at the end.
952 if (newname != NULL) {
954 TMPFS_NODE_LOCK(fnode);
955 fnode->tn_status |= TMPFS_NODE_CHANGED;
956 TMPFS_NODE_UNLOCK(fnode);
958 oldname = de->td_name;
959 de->td_name = newname;
960 de->td_namelen = (uint16_t)tncp->nc_nlen;
965 * Link entry to target directory. If the entry
966 * represents a directory move the parent linkage
969 if (fdnode != tdnode) {
970 if (de->td_node->tn_type == VDIR) {
971 TMPFS_VALIDATE_DIR(fnode);
973 TMPFS_NODE_LOCK(tdnode);
975 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
976 TMPFS_NODE_UNLOCK(tdnode);
978 TMPFS_NODE_LOCK(fnode);
979 fnode->tn_dir.tn_parent = tdnode;
980 fnode->tn_status |= TMPFS_NODE_CHANGED;
981 TMPFS_NODE_UNLOCK(fnode);
983 TMPFS_NODE_LOCK(fdnode);
985 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
986 TMPFS_NODE_UNLOCK(fdnode);
988 tmpfs_dir_attach(tdnode, de);
990 TMPFS_NODE_LOCK(tdnode);
991 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
992 TMPFS_NODE_UNLOCK(tdnode);
996 * If we are overwriting an entry, we have to remove the old one
997 * from the target directory.
1000 /* Remove the old entry from the target directory. */
1001 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1002 tmpfs_dir_detach(tdnode, de);
1005 * Free the directory entry we just deleted. Note that the
1006 * node referred by it will not be removed until the vnode is
1009 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1010 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1017 kfree(newname, M_TMPFSNAME);
1020 cache_rename(v->a_fnch, v->a_tnch);
1027 /* Release target nodes. */
1028 /* XXX: I don't understand when tdvp can be the same as tvp, but
1029 * other code takes care of this... */
1036 /* --------------------------------------------------------------------- */
1039 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1041 struct vnode *dvp = v->a_dvp;
1042 struct vnode **vpp = v->a_vpp;
1043 struct namecache *ncp = v->a_nch->ncp;
1044 struct vattr *vap = v->a_vap;
1045 struct ucred *cred = v->a_cred;
1048 KKASSERT(vap->va_type == VDIR);
1050 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1052 cache_setunresolved(v->a_nch);
1053 cache_setvp(v->a_nch, *vpp);
1059 /* --------------------------------------------------------------------- */
1062 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1064 struct vnode *dvp = v->a_dvp;
1065 struct namecache *ncp = v->a_nch->ncp;
1069 struct tmpfs_dirent *de;
1070 struct tmpfs_mount *tmp;
1071 struct tmpfs_node *dnode;
1072 struct tmpfs_node *node;
1075 * We have to acquire the vp from v->a_nch because
1076 * we will likely unresolve the namecache entry, and
1077 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1080 error = cache_vref(v->a_nch, v->a_cred, &vp);
1081 KKASSERT(error == 0);
1083 tmp = VFS_TO_TMPFS(dvp->v_mount);
1084 dnode = VP_TO_TMPFS_DIR(dvp);
1085 node = VP_TO_TMPFS_DIR(vp);
1087 /* Directories with more than two entries ('.' and '..') cannot be
1089 if (node->tn_size > 0) {
1094 if ((dnode->tn_flags & APPEND)
1095 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1100 /* This invariant holds only if we are not trying to remove "..".
1101 * We checked for that above so this is safe now. */
1102 KKASSERT(node->tn_dir.tn_parent == dnode);
1104 /* Get the directory entry associated with node (vp). This was
1105 * filled by tmpfs_lookup while looking up the entry. */
1106 de = tmpfs_dir_lookup(dnode, node, ncp);
1107 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1111 /* Check flags to see if we are allowed to remove the directory. */
1112 if ((dnode->tn_flags & APPEND) ||
1113 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1119 /* Detach the directory entry from the directory (dnode). */
1120 tmpfs_dir_detach(dnode, de);
1122 /* No vnode should be allocated for this entry from this point */
1123 TMPFS_NODE_LOCK(node);
1124 TMPFS_ASSERT_ELOCKED(node);
1125 TMPFS_NODE_LOCK(dnode);
1126 TMPFS_ASSERT_ELOCKED(dnode);
1129 /* handled by tmpfs_free_node */
1130 KKASSERT(node->tn_links > 0);
1132 node->tn_dir.tn_parent = NULL;
1134 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1135 TMPFS_NODE_MODIFIED;
1138 /* handled by tmpfs_free_node */
1139 KKASSERT(dnode->tn_links > 0);
1142 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1143 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1145 TMPFS_NODE_UNLOCK(dnode);
1146 TMPFS_NODE_UNLOCK(node);
1148 /* Free the directory entry we just deleted. Note that the node
1149 * referred by it will not be removed until the vnode is really
1151 tmpfs_free_dirent(tmp, de);
1153 /* Release the deleted vnode (will destroy the node, notify
1154 * interested parties and clean it from the cache). */
1156 TMPFS_NODE_LOCK(dnode);
1157 dnode->tn_status |= TMPFS_NODE_CHANGED;
1158 TMPFS_NODE_UNLOCK(dnode);
1161 cache_setunresolved(v->a_nch);
1162 cache_setvp(v->a_nch, NULL);
1163 /*cache_inval_vp(vp, CINV_DESTROY);*/
1172 /* --------------------------------------------------------------------- */
1175 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1177 struct vnode *dvp = v->a_dvp;
1178 struct vnode **vpp = v->a_vpp;
1179 struct namecache *ncp = v->a_nch->ncp;
1180 struct vattr *vap = v->a_vap;
1181 struct ucred *cred = v->a_cred;
1182 char *target = v->a_target;
1185 vap->va_type = VLNK;
1186 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1188 cache_setunresolved(v->a_nch);
1189 cache_setvp(v->a_nch, *vpp);
1195 /* --------------------------------------------------------------------- */
1198 tmpfs_readdir(struct vop_readdir_args *v)
1200 struct vnode *vp = v->a_vp;
1201 struct uio *uio = v->a_uio;
1202 int *eofflag = v->a_eofflag;
1203 off_t **cookies = v->a_cookies;
1204 int *ncookies = v->a_ncookies;
1205 struct tmpfs_mount *tmp;
1209 struct tmpfs_node *node;
1211 /* This operation only makes sense on directory nodes. */
1212 if (vp->v_type != VDIR)
1215 tmp = VFS_TO_TMPFS(vp->v_mount);
1216 node = VP_TO_TMPFS_DIR(vp);
1217 startoff = uio->uio_offset;
1219 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1220 error = tmpfs_dir_getdotdent(node, uio);
1226 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1227 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1233 error = tmpfs_dir_getdents(node, uio, &cnt);
1236 KKASSERT(error >= -1);
1241 if (eofflag != NULL)
1243 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1245 /* Update NFS-related variables. */
1246 if (error == 0 && cookies != NULL && ncookies != NULL) {
1248 off_t off = startoff;
1249 struct tmpfs_dirent *de = NULL;
1252 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1254 for (i = 0; i < cnt; i++) {
1255 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1256 if (off == TMPFS_DIRCOOKIE_DOT) {
1257 off = TMPFS_DIRCOOKIE_DOTDOT;
1259 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1260 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1261 } else if (de != NULL) {
1262 de = TAILQ_NEXT(de, td_entries);
1264 de = tmpfs_dir_lookupbycookie(node,
1266 KKASSERT(de != NULL);
1267 de = TAILQ_NEXT(de, td_entries);
1270 off = TMPFS_DIRCOOKIE_EOF;
1272 off = tmpfs_dircookie(de);
1275 (*cookies)[i] = off;
1277 KKASSERT(uio->uio_offset == off);
1283 /* --------------------------------------------------------------------- */
1286 tmpfs_readlink(struct vop_readlink_args *v)
1288 struct vnode *vp = v->a_vp;
1289 struct uio *uio = v->a_uio;
1292 struct tmpfs_node *node;
1294 KKASSERT(uio->uio_offset == 0);
1295 KKASSERT(vp->v_type == VLNK);
1297 node = VP_TO_TMPFS_NODE(vp);
1299 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1301 TMPFS_NODE_LOCK(node);
1302 node->tn_status |= TMPFS_NODE_ACCESSED;
1303 TMPFS_NODE_UNLOCK(node);
1308 /* --------------------------------------------------------------------- */
1311 tmpfs_inactive(struct vop_inactive_args *v)
1313 struct vnode *vp = v->a_vp;
1315 struct tmpfs_node *node;
1317 node = VP_TO_TMPFS_NODE(vp);
1320 * Get rid of unreferenced deleted vnodes sooner rather than
1321 * later so the data memory can be recovered immediately.
1323 TMPFS_NODE_LOCK(node);
1324 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1325 (node->tn_links == 0 ||
1326 (node->tn_links == 1 && node->tn_type == VDIR &&
1327 node->tn_dir.tn_parent)))
1329 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1330 TMPFS_NODE_UNLOCK(node);
1333 TMPFS_NODE_UNLOCK(node);
1339 /* --------------------------------------------------------------------- */
1342 tmpfs_reclaim(struct vop_reclaim_args *v)
1344 struct vnode *vp = v->a_vp;
1345 struct tmpfs_mount *tmp;
1346 struct tmpfs_node *node;
1348 node = VP_TO_TMPFS_NODE(vp);
1349 tmp = VFS_TO_TMPFS(vp->v_mount);
1354 * If the node referenced by this vnode was deleted by the
1355 * user, we must free its associated data structures now that
1356 * the vnode is being reclaimed.
1358 * Directories have an extra link ref.
1360 TMPFS_NODE_LOCK(node);
1361 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1362 (node->tn_links == 0 ||
1363 (node->tn_links == 1 && node->tn_type == VDIR &&
1364 node->tn_dir.tn_parent)))
1366 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1367 tmpfs_free_node(tmp, node);
1370 TMPFS_NODE_UNLOCK(node);
1373 KKASSERT(vp->v_data == NULL);
1377 /* --------------------------------------------------------------------- */
1380 tmpfs_print(struct vop_print_args *v)
1382 struct vnode *vp = v->a_vp;
1384 struct tmpfs_node *node;
1386 node = VP_TO_TMPFS_NODE(vp);
1388 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1389 node, node->tn_flags, node->tn_links);
1390 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1391 node->tn_mode, node->tn_uid, node->tn_gid,
1392 (uintmax_t)node->tn_size, node->tn_status);
1394 if (vp->v_type == VFIFO)
1402 /* --------------------------------------------------------------------- */
1405 tmpfs_pathconf(struct vop_pathconf_args *v)
1407 int name = v->a_name;
1408 register_t *retval = v->a_retval;
1431 case _PC_CHOWN_RESTRICTED:
1443 case _PC_FILESIZEBITS:
1444 *retval = 0; /* XXX Don't know which value should I return. */
1454 /* --------------------------------------------------------------------- */
1457 * vnode operations vector used for files stored in a tmpfs file system.
1459 struct vop_ops tmpfs_vnode_vops = {
1460 .vop_default = vop_defaultop,
1461 .vop_getpages = vop_stdgetpages,
1462 .vop_putpages = vop_stdputpages,
1463 .vop_ncreate = tmpfs_ncreate,
1464 .vop_nresolve = tmpfs_nresolve,
1465 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1466 .vop_nmknod = tmpfs_nmknod,
1467 .vop_open = tmpfs_open,
1468 .vop_close = tmpfs_close,
1469 .vop_access = tmpfs_access,
1470 .vop_getattr = tmpfs_getattr,
1471 .vop_setattr = tmpfs_setattr,
1472 .vop_read = tmpfs_read,
1473 .vop_write = tmpfs_write,
1474 .vop_fsync = tmpfs_fsync,
1475 .vop_nremove = tmpfs_nremove,
1476 .vop_nlink = tmpfs_nlink,
1477 .vop_nrename = tmpfs_nrename,
1478 .vop_nmkdir = tmpfs_nmkdir,
1479 .vop_nrmdir = tmpfs_nrmdir,
1480 .vop_nsymlink = tmpfs_nsymlink,
1481 .vop_readdir = tmpfs_readdir,
1482 .vop_readlink = tmpfs_readlink,
1483 .vop_inactive = tmpfs_inactive,
1484 .vop_reclaim = tmpfs_reclaim,
1485 .vop_print = tmpfs_print,
1486 .vop_pathconf = tmpfs_pathconf,
1487 .vop_bmap = tmpfs_bmap,
1488 .vop_bmap = (void *)vop_eopnotsupp,
1489 .vop_strategy = tmpfs_strategy,
1490 .vop_advlock = tmpfs_advlock,