2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * This code is derived from software contributed to The NetBSD Foundation
6 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
34 * tmpfs vnode interface.
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.h>
50 #include <sys/vnode.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_pager.h>
56 #include <vm/swap_pager.h>
58 #include <vfs/fifofs/fifo.h>
59 #include <vfs/tmpfs/tmpfs_vnops.h>
60 #include <vfs/tmpfs/tmpfs.h>
62 MALLOC_DECLARE(M_TMPFS);
66 tmpfs_knote(struct vnode *vp, int flags)
69 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
73 /* --------------------------------------------------------------------- */
76 tmpfs_nresolve(struct vop_nresolve_args *v)
78 struct vnode *dvp = v->a_dvp;
79 struct vnode *vp = NULL;
80 struct namecache *ncp = v->a_nch->ncp;
81 struct tmpfs_node *tnode;
84 struct tmpfs_dirent *de;
85 struct tmpfs_node *dnode;
87 dnode = VP_TO_TMPFS_DIR(dvp);
89 de = tmpfs_dir_lookup(dnode, NULL, ncp);
94 * Allocate a vnode for the node we found.
97 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
98 LK_EXCLUSIVE | LK_RETRY, &vp);
106 * Store the result of this lookup in the cache. Avoid this if the
107 * request was for creation, as it does not improve timings on
112 cache_setvp(v->a_nch, vp);
114 } else if (error == ENOENT) {
115 cache_setvp(v->a_nch, NULL);
121 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
123 struct vnode *dvp = v->a_dvp;
124 struct vnode **vpp = v->a_vpp;
125 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
126 struct ucred *cred = v->a_cred;
130 /* Check accessibility of requested node as a first step. */
131 error = VOP_ACCESS(dvp, VEXEC, cred);
135 if (dnode->tn_dir.tn_parent != NULL) {
136 /* Allocate a new vnode on the matching entry. */
137 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
138 LK_EXCLUSIVE | LK_RETRY, vpp);
144 return (*vpp == NULL) ? ENOENT : 0;
147 /* --------------------------------------------------------------------- */
150 tmpfs_ncreate(struct vop_ncreate_args *v)
152 struct vnode *dvp = v->a_dvp;
153 struct vnode **vpp = v->a_vpp;
154 struct namecache *ncp = v->a_nch->ncp;
155 struct vattr *vap = v->a_vap;
156 struct ucred *cred = v->a_cred;
159 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
161 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
163 cache_setunresolved(v->a_nch);
164 cache_setvp(v->a_nch, *vpp);
165 tmpfs_knote(dvp, NOTE_WRITE);
170 /* --------------------------------------------------------------------- */
173 tmpfs_nmknod(struct vop_nmknod_args *v)
175 struct vnode *dvp = v->a_dvp;
176 struct vnode **vpp = v->a_vpp;
177 struct namecache *ncp = v->a_nch->ncp;
178 struct vattr *vap = v->a_vap;
179 struct ucred *cred = v->a_cred;
182 if (vap->va_type != VBLK && vap->va_type != VCHR &&
183 vap->va_type != VFIFO)
186 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
188 cache_setunresolved(v->a_nch);
189 cache_setvp(v->a_nch, *vpp);
190 tmpfs_knote(dvp, NOTE_WRITE);
196 /* --------------------------------------------------------------------- */
199 tmpfs_open(struct vop_open_args *v)
201 struct vnode *vp = v->a_vp;
202 int mode = v->a_mode;
205 struct tmpfs_node *node;
207 node = VP_TO_TMPFS_NODE(vp);
209 /* The file is still active but all its names have been removed
210 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
211 * it is about to die. */
212 if (node->tn_links < 1)
215 /* If the file is marked append-only, deny write requests. */
216 if ((node->tn_flags & APPEND) &&
217 (mode & (FWRITE | O_APPEND)) == FWRITE) {
220 return (vop_stdopen(v));
225 /* --------------------------------------------------------------------- */
228 tmpfs_close(struct vop_close_args *v)
230 struct vnode *vp = v->a_vp;
231 struct tmpfs_node *node;
233 node = VP_TO_TMPFS_NODE(vp);
235 if (node->tn_links > 0) {
236 /* Update node times. No need to do it if the node has
237 * been deleted, because it will vanish after we return. */
241 return vop_stdclose(v);
244 /* --------------------------------------------------------------------- */
247 tmpfs_access(struct vop_access_args *v)
249 struct vnode *vp = v->a_vp;
251 struct tmpfs_node *node;
253 node = VP_TO_TMPFS_NODE(vp);
255 switch (vp->v_type) {
261 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
281 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
286 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
293 /* --------------------------------------------------------------------- */
296 tmpfs_getattr(struct vop_getattr_args *v)
298 struct vnode *vp = v->a_vp;
299 struct vattr *vap = v->a_vap;
300 struct tmpfs_node *node;
302 node = VP_TO_TMPFS_NODE(vp);
304 lwkt_gettoken(&vp->v_mount->mnt_token);
307 vap->va_type = vp->v_type;
308 vap->va_mode = node->tn_mode;
309 vap->va_nlink = node->tn_links;
310 vap->va_uid = node->tn_uid;
311 vap->va_gid = node->tn_gid;
312 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
313 vap->va_fileid = node->tn_id;
314 vap->va_size = node->tn_size;
315 vap->va_blocksize = PAGE_SIZE;
316 vap->va_atime.tv_sec = node->tn_atime;
317 vap->va_atime.tv_nsec = node->tn_atimensec;
318 vap->va_mtime.tv_sec = node->tn_mtime;
319 vap->va_mtime.tv_nsec = node->tn_mtimensec;
320 vap->va_ctime.tv_sec = node->tn_ctime;
321 vap->va_ctime.tv_nsec = node->tn_ctimensec;
322 vap->va_gen = node->tn_gen;
323 vap->va_flags = node->tn_flags;
324 if (vp->v_type == VBLK || vp->v_type == VCHR)
326 vap->va_rmajor = umajor(node->tn_rdev);
327 vap->va_rminor = uminor(node->tn_rdev);
329 vap->va_bytes = round_page(node->tn_size);
332 lwkt_reltoken(&vp->v_mount->mnt_token);
337 /* --------------------------------------------------------------------- */
340 tmpfs_setattr(struct vop_setattr_args *v)
342 struct vnode *vp = v->a_vp;
343 struct vattr *vap = v->a_vap;
344 struct ucred *cred = v->a_cred;
345 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
349 if (error == 0 && (vap->va_flags != VNOVAL)) {
350 error = tmpfs_chflags(vp, vap->va_flags, cred);
351 kflags |= NOTE_ATTRIB;
354 if (error == 0 && (vap->va_size != VNOVAL)) {
355 if (vap->va_size > node->tn_size)
356 kflags |= NOTE_WRITE | NOTE_EXTEND;
358 kflags |= NOTE_WRITE;
359 error = tmpfs_chsize(vp, vap->va_size, cred);
362 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
363 vap->va_gid != (gid_t)VNOVAL)) {
364 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
365 kflags |= NOTE_ATTRIB;
368 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
369 error = tmpfs_chmod(vp, vap->va_mode, cred);
370 kflags |= NOTE_ATTRIB;
373 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
374 vap->va_atime.tv_nsec != VNOVAL) ||
375 (vap->va_mtime.tv_sec != VNOVAL &&
376 vap->va_mtime.tv_nsec != VNOVAL) )) {
377 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
378 vap->va_vaflags, cred);
379 kflags |= NOTE_ATTRIB;
382 /* Update the node times. We give preference to the error codes
383 * generated by this function rather than the ones that may arise
384 * from tmpfs_update. */
386 tmpfs_knote(vp, kflags);
391 /* --------------------------------------------------------------------- */
394 * fsync is usually a NOP, but we must take action when unmounting or
398 tmpfs_fsync(struct vop_fsync_args *v)
400 struct tmpfs_mount *tmp;
401 struct tmpfs_node *node;
402 struct vnode *vp = v->a_vp;
404 tmp = VFS_TO_TMPFS(vp->v_mount);
405 node = VP_TO_TMPFS_NODE(vp);
408 if (vp->v_type == VREG) {
409 if (vp->v_flag & VRECLAIMED) {
410 if (node->tn_links == 0)
411 tmpfs_truncate(vp, 0);
413 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
419 /* --------------------------------------------------------------------- */
422 tmpfs_read (struct vop_read_args *ap)
425 struct vnode *vp = ap->a_vp;
426 struct uio *uio = ap->a_uio;
427 struct tmpfs_node *node;
434 if (uio->uio_resid == 0) {
438 node = VP_TO_TMPFS_NODE(vp);
440 if (uio->uio_offset < 0)
442 if (vp->v_type != VREG)
445 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
447 * Use buffer cache I/O (via tmpfs_strategy)
449 offset = (size_t)uio->uio_offset & BMASK;
450 base_offset = (off_t)uio->uio_offset - offset;
451 bp = getcacheblk(vp, base_offset, BSIZE);
454 lwkt_gettoken(&vp->v_mount->mnt_token);
455 error = bread(vp, base_offset, BSIZE, &bp);
458 lwkt_reltoken(&vp->v_mount->mnt_token);
459 kprintf("tmpfs_read bread error %d\n", error);
462 lwkt_reltoken(&vp->v_mount->mnt_token);
466 * Figure out how many bytes we can actually copy this loop.
468 len = BSIZE - offset;
469 if (len > uio->uio_resid)
470 len = uio->uio_resid;
471 if (len > node->tn_size - uio->uio_offset)
472 len = (size_t)(node->tn_size - uio->uio_offset);
474 error = uiomove((char *)bp->b_data + offset, len, uio);
477 kprintf("tmpfs_read uiomove error %d\n", error);
482 TMPFS_NODE_LOCK(node);
483 node->tn_status |= TMPFS_NODE_ACCESSED;
484 TMPFS_NODE_UNLOCK(node);
490 tmpfs_write (struct vop_write_args *ap)
493 struct vnode *vp = ap->a_vp;
494 struct uio *uio = ap->a_uio;
495 struct thread *td = uio->uio_td;
496 struct tmpfs_node *node;
508 if (uio->uio_resid == 0) {
512 node = VP_TO_TMPFS_NODE(vp);
514 if (vp->v_type != VREG)
517 lwkt_gettoken(&vp->v_mount->mnt_token);
519 oldsize = node->tn_size;
520 if (ap->a_ioflag & IO_APPEND)
521 uio->uio_offset = node->tn_size;
524 * Check for illegal write offsets.
526 if (uio->uio_offset + uio->uio_resid >
527 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
528 lwkt_reltoken(&vp->v_mount->mnt_token);
532 if (vp->v_type == VREG && td != NULL) {
533 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
535 lwkt_reltoken(&vp->v_mount->mnt_token);
538 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
539 ksignal(td->td_proc, SIGXFSZ);
540 lwkt_reltoken(&vp->v_mount->mnt_token);
547 * Extend the file's size if necessary
549 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
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;
587 kflags |= NOTE_EXTEND;
589 kflags |= NOTE_WRITE;
592 * The data has been loaded into the buffer, write it out.
594 * We want tmpfs to be able to use all available ram, not
595 * just the buffer cache, so if not explicitly paging we
596 * use buwrite() to leave the buffer clean but mark all the
597 * VM pages valid+dirty.
599 * When the kernel is paging, either via normal pageout
600 * operation or when cleaning the object during a recycle,
601 * the underlying VM pages are going to get thrown away
602 * so we MUST write them to swap.
604 * XXX unfortunately this catches msync() system calls too
607 if (vm_swap_size == 0) {
609 * if swap isn't configured yet, force a buwrite() to
610 * avoid problems further down the line, due to flushing
615 if (ap->a_ioflag & IO_SYNC) {
617 } else if ((ap->a_ioflag & IO_ASYNC) ||
618 (uio->uio_segflg == UIO_NOCOPY)) {
626 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
633 (void)tmpfs_reg_resize(vp, oldsize, trivial);
634 kflags &= ~NOTE_EXTEND;
639 TMPFS_NODE_LOCK(node);
640 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
641 (extended? TMPFS_NODE_CHANGED : 0);
643 if (node->tn_mode & (S_ISUID | S_ISGID)) {
644 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
645 node->tn_mode &= ~(S_ISUID | S_ISGID);
647 TMPFS_NODE_UNLOCK(node);
650 tmpfs_knote(vp, kflags);
653 lwkt_reltoken(&vp->v_mount->mnt_token);
658 tmpfs_advlock (struct vop_advlock_args *ap)
660 struct tmpfs_node *node;
661 struct vnode *vp = ap->a_vp;
663 node = VP_TO_TMPFS_NODE(vp);
665 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
669 tmpfs_strategy(struct vop_strategy_args *ap)
671 struct bio *bio = ap->a_bio;
672 struct buf *bp = bio->bio_buf;
673 struct vnode *vp = ap->a_vp;
674 struct tmpfs_node *node;
677 if (vp->v_type != VREG) {
678 bp->b_resid = bp->b_bcount;
679 bp->b_flags |= B_ERROR | B_INVAL;
680 bp->b_error = EINVAL;
685 lwkt_gettoken(&vp->v_mount->mnt_token);
686 node = VP_TO_TMPFS_NODE(vp);
688 uobj = node->tn_reg.tn_aobj;
691 * Call swap_pager_strategy to read or write between the VM
692 * object and the buffer cache.
694 swap_pager_strategy(uobj, bio);
696 lwkt_reltoken(&vp->v_mount->mnt_token);
701 tmpfs_bmap(struct vop_bmap_args *ap)
703 if (ap->a_doffsetp != NULL)
704 *ap->a_doffsetp = ap->a_loffset;
705 if (ap->a_runp != NULL)
707 if (ap->a_runb != NULL)
713 /* --------------------------------------------------------------------- */
716 tmpfs_nremove(struct vop_nremove_args *v)
718 struct vnode *dvp = v->a_dvp;
719 struct namecache *ncp = v->a_nch->ncp;
722 struct tmpfs_dirent *de;
723 struct tmpfs_mount *tmp;
724 struct tmpfs_node *dnode;
725 struct tmpfs_node *node;
728 * We have to acquire the vp from v->a_nch because
729 * we will likely unresolve the namecache entry, and
730 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
731 * sequence to recover space from the file.
733 error = cache_vref(v->a_nch, v->a_cred, &vp);
734 KKASSERT(error == 0);
736 if (vp->v_type == VDIR) {
741 dnode = VP_TO_TMPFS_DIR(dvp);
742 node = VP_TO_TMPFS_NODE(vp);
743 tmp = VFS_TO_TMPFS(vp->v_mount);
744 de = tmpfs_dir_lookup(dnode, node, ncp);
750 /* Files marked as immutable or append-only cannot be deleted. */
751 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
752 (dnode->tn_flags & APPEND)) {
757 /* Remove the entry from the directory; as it is a file, we do not
758 * have to change the number of hard links of the directory. */
759 tmpfs_dir_detach(dnode, de);
761 /* Free the directory entry we just deleted. Note that the node
762 * referred by it will not be removed until the vnode is really
764 tmpfs_free_dirent(tmp, de);
766 if (node->tn_links > 0) {
767 TMPFS_NODE_LOCK(node);
768 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
770 TMPFS_NODE_UNLOCK(node);
773 cache_setunresolved(v->a_nch);
774 cache_setvp(v->a_nch, NULL);
775 tmpfs_knote(vp, NOTE_DELETE);
776 /*cache_inval_vp(vp, CINV_DESTROY);*/
777 tmpfs_knote(dvp, NOTE_WRITE);
786 /* --------------------------------------------------------------------- */
789 tmpfs_nlink(struct vop_nlink_args *v)
791 struct vnode *dvp = v->a_dvp;
792 struct vnode *vp = v->a_vp;
793 struct namecache *ncp = v->a_nch->ncp;
794 struct tmpfs_dirent *de;
795 struct tmpfs_node *node;
796 struct tmpfs_node *dnode;
799 KKASSERT(dvp != vp); /* XXX When can this be false? */
801 node = VP_TO_TMPFS_NODE(vp);
802 dnode = VP_TO_TMPFS_NODE(dvp);
804 /* XXX: Why aren't the following two tests done by the caller? */
806 /* Hard links of directories are forbidden. */
807 if (vp->v_type == VDIR) {
812 /* Cannot create cross-device links. */
813 if (dvp->v_mount != vp->v_mount) {
818 /* Ensure that we do not overflow the maximum number of links imposed
820 KKASSERT(node->tn_links <= LINK_MAX);
821 if (node->tn_links == LINK_MAX) {
826 /* We cannot create links of files marked immutable or append-only. */
827 if (node->tn_flags & (IMMUTABLE | APPEND)) {
832 /* Allocate a new directory entry to represent the node. */
833 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
834 ncp->nc_name, ncp->nc_nlen, &de);
838 /* Insert the new directory entry into the appropriate directory. */
839 tmpfs_dir_attach(dnode, de);
841 /* vp link count has changed, so update node times. */
843 TMPFS_NODE_LOCK(node);
844 node->tn_status |= TMPFS_NODE_CHANGED;
845 TMPFS_NODE_UNLOCK(node);
848 tmpfs_knote(vp, NOTE_LINK);
849 cache_setunresolved(v->a_nch);
850 cache_setvp(v->a_nch, vp);
851 tmpfs_knote(dvp, NOTE_WRITE);
858 /* --------------------------------------------------------------------- */
861 tmpfs_nrename(struct vop_nrename_args *v)
863 struct vnode *fdvp = v->a_fdvp;
864 struct namecache *fncp = v->a_fnch->ncp;
865 struct vnode *fvp = fncp->nc_vp;
866 struct vnode *tdvp = v->a_tdvp;
867 struct namecache *tncp = v->a_tnch->ncp;
868 struct vnode *tvp = tncp->nc_vp;
869 struct tmpfs_dirent *de;
870 struct tmpfs_mount *tmp;
871 struct tmpfs_node *fdnode;
872 struct tmpfs_node *fnode;
873 struct tmpfs_node *tnode;
874 struct tmpfs_node *tdnode;
879 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
881 /* Disallow cross-device renames.
882 * XXX Why isn't this done by the caller? */
883 if (fvp->v_mount != tdvp->v_mount ||
884 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
889 tmp = VFS_TO_TMPFS(tdvp->v_mount);
890 tdnode = VP_TO_TMPFS_DIR(tdvp);
892 /* If source and target are the same file, there is nothing to do. */
898 fdnode = VP_TO_TMPFS_DIR(fdvp);
899 fnode = VP_TO_TMPFS_NODE(fvp);
900 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
902 /* Avoid manipulating '.' and '..' entries. */
907 KKASSERT(de->td_node == fnode);
910 * If replacing an entry in the target directory and that entry
911 * is a directory, it must be empty.
913 * Kern_rename gurantees the destination to be a directory
914 * if the source is one (it does?).
917 KKASSERT(tnode != NULL);
919 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
920 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
925 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
926 if (tnode->tn_size > 0) {
930 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
933 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
937 KKASSERT(fnode->tn_type != VDIR &&
938 tnode->tn_type != VDIR);
942 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
943 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
949 * Ensure that we have enough memory to hold the new name, if it
952 if (fncp->nc_nlen != tncp->nc_nlen ||
953 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
954 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
955 M_WAITOK | M_NULLOK);
956 if (newname == NULL) {
960 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
961 newname[tncp->nc_nlen] = '\0';
967 * Unlink entry from source directory. Note that the kernel has
968 * already checked for illegal recursion cases (renaming a directory
969 * into a subdirectory of itself).
971 if (fdnode != tdnode)
972 tmpfs_dir_detach(fdnode, de);
975 * Handle any name change. Swap with newname, we will
976 * deallocate it at the end.
978 if (newname != NULL) {
980 TMPFS_NODE_LOCK(fnode);
981 fnode->tn_status |= TMPFS_NODE_CHANGED;
982 TMPFS_NODE_UNLOCK(fnode);
984 oldname = de->td_name;
985 de->td_name = newname;
986 de->td_namelen = (uint16_t)tncp->nc_nlen;
991 * Link entry to target directory. If the entry
992 * represents a directory move the parent linkage
995 if (fdnode != tdnode) {
996 if (de->td_node->tn_type == VDIR) {
997 TMPFS_VALIDATE_DIR(fnode);
999 TMPFS_NODE_LOCK(tdnode);
1001 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1002 TMPFS_NODE_UNLOCK(tdnode);
1004 TMPFS_NODE_LOCK(fnode);
1005 fnode->tn_dir.tn_parent = tdnode;
1006 fnode->tn_status |= TMPFS_NODE_CHANGED;
1007 TMPFS_NODE_UNLOCK(fnode);
1009 TMPFS_NODE_LOCK(fdnode);
1011 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1012 TMPFS_NODE_UNLOCK(fdnode);
1014 tmpfs_dir_attach(tdnode, de);
1016 TMPFS_NODE_LOCK(tdnode);
1017 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1018 TMPFS_NODE_UNLOCK(tdnode);
1022 * If we are overwriting an entry, we have to remove the old one
1023 * from the target directory.
1026 /* Remove the old entry from the target directory. */
1027 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1028 tmpfs_dir_detach(tdnode, de);
1029 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1032 * Free the directory entry we just deleted. Note that the
1033 * node referred by it will not be removed until the vnode is
1036 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1037 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1044 kfree(newname, tmp->tm_name_zone);
1047 cache_rename(v->a_fnch, v->a_tnch);
1048 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1049 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1050 if (fnode->tn_vnode)
1051 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1058 /* Release target nodes. */
1059 /* XXX: I don't understand when tdvp can be the same as tvp, but
1060 * other code takes care of this... */
1067 /* --------------------------------------------------------------------- */
1070 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1072 struct vnode *dvp = v->a_dvp;
1073 struct vnode **vpp = v->a_vpp;
1074 struct namecache *ncp = v->a_nch->ncp;
1075 struct vattr *vap = v->a_vap;
1076 struct ucred *cred = v->a_cred;
1079 KKASSERT(vap->va_type == VDIR);
1081 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1083 cache_setunresolved(v->a_nch);
1084 cache_setvp(v->a_nch, *vpp);
1085 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1091 /* --------------------------------------------------------------------- */
1094 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1096 struct vnode *dvp = v->a_dvp;
1097 struct namecache *ncp = v->a_nch->ncp;
1099 struct tmpfs_dirent *de;
1100 struct tmpfs_mount *tmp;
1101 struct tmpfs_node *dnode;
1102 struct tmpfs_node *node;
1106 * We have to acquire the vp from v->a_nch because
1107 * we will likely unresolve the namecache entry, and
1108 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1111 error = cache_vref(v->a_nch, v->a_cred, &vp);
1112 KKASSERT(error == 0);
1115 * Prevalidate so we don't hit an assertion later
1117 if (vp->v_type != VDIR) {
1122 tmp = VFS_TO_TMPFS(dvp->v_mount);
1123 dnode = VP_TO_TMPFS_DIR(dvp);
1124 node = VP_TO_TMPFS_DIR(vp);
1126 /* Directories with more than two entries ('.' and '..') cannot be
1128 if (node->tn_size > 0) {
1133 if ((dnode->tn_flags & APPEND)
1134 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1139 /* This invariant holds only if we are not trying to remove "..".
1140 * We checked for that above so this is safe now. */
1141 KKASSERT(node->tn_dir.tn_parent == dnode);
1143 /* Get the directory entry associated with node (vp). This was
1144 * filled by tmpfs_lookup while looking up the entry. */
1145 de = tmpfs_dir_lookup(dnode, node, ncp);
1146 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1150 /* Check flags to see if we are allowed to remove the directory. */
1151 if ((dnode->tn_flags & APPEND) ||
1152 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1158 /* Detach the directory entry from the directory (dnode). */
1159 tmpfs_dir_detach(dnode, de);
1161 /* No vnode should be allocated for this entry from this point */
1162 TMPFS_NODE_LOCK(node);
1163 TMPFS_ASSERT_ELOCKED(node);
1164 TMPFS_NODE_LOCK(dnode);
1165 TMPFS_ASSERT_ELOCKED(dnode);
1168 /* handled by tmpfs_free_node */
1169 KKASSERT(node->tn_links > 0);
1171 node->tn_dir.tn_parent = NULL;
1173 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1174 TMPFS_NODE_MODIFIED;
1177 /* handled by tmpfs_free_node */
1178 KKASSERT(dnode->tn_links > 0);
1181 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1182 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1184 TMPFS_NODE_UNLOCK(dnode);
1185 TMPFS_NODE_UNLOCK(node);
1187 /* Free the directory entry we just deleted. Note that the node
1188 * referred by it will not be removed until the vnode is really
1190 tmpfs_free_dirent(tmp, de);
1192 /* Release the deleted vnode (will destroy the node, notify
1193 * interested parties and clean it from the cache). */
1195 TMPFS_NODE_LOCK(dnode);
1196 dnode->tn_status |= TMPFS_NODE_CHANGED;
1197 TMPFS_NODE_UNLOCK(dnode);
1200 cache_setunresolved(v->a_nch);
1201 cache_setvp(v->a_nch, NULL);
1202 /*cache_inval_vp(vp, CINV_DESTROY);*/
1203 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1212 /* --------------------------------------------------------------------- */
1215 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1217 struct vnode *dvp = v->a_dvp;
1218 struct vnode **vpp = v->a_vpp;
1219 struct namecache *ncp = v->a_nch->ncp;
1220 struct vattr *vap = v->a_vap;
1221 struct ucred *cred = v->a_cred;
1222 char *target = v->a_target;
1225 vap->va_type = VLNK;
1226 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1228 tmpfs_knote(*vpp, NOTE_WRITE);
1229 cache_setunresolved(v->a_nch);
1230 cache_setvp(v->a_nch, *vpp);
1236 /* --------------------------------------------------------------------- */
1239 tmpfs_readdir(struct vop_readdir_args *v)
1241 struct vnode *vp = v->a_vp;
1242 struct uio *uio = v->a_uio;
1243 int *eofflag = v->a_eofflag;
1244 off_t **cookies = v->a_cookies;
1245 int *ncookies = v->a_ncookies;
1246 struct tmpfs_mount *tmp;
1250 struct tmpfs_node *node;
1252 /* This operation only makes sense on directory nodes. */
1253 if (vp->v_type != VDIR)
1256 tmp = VFS_TO_TMPFS(vp->v_mount);
1257 node = VP_TO_TMPFS_DIR(vp);
1258 startoff = uio->uio_offset;
1260 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1261 error = tmpfs_dir_getdotdent(node, uio);
1267 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1268 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1274 error = tmpfs_dir_getdents(node, uio, &cnt);
1277 KKASSERT(error >= -1);
1282 if (eofflag != NULL)
1284 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1286 /* Update NFS-related variables. */
1287 if (error == 0 && cookies != NULL && ncookies != NULL) {
1289 off_t off = startoff;
1290 struct tmpfs_dirent *de = NULL;
1293 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1295 for (i = 0; i < cnt; i++) {
1296 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1297 if (off == TMPFS_DIRCOOKIE_DOT) {
1298 off = TMPFS_DIRCOOKIE_DOTDOT;
1300 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1301 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1302 } else if (de != NULL) {
1303 de = TAILQ_NEXT(de, td_entries);
1305 de = tmpfs_dir_lookupbycookie(node,
1307 KKASSERT(de != NULL);
1308 de = TAILQ_NEXT(de, td_entries);
1311 off = TMPFS_DIRCOOKIE_EOF;
1313 off = tmpfs_dircookie(de);
1316 (*cookies)[i] = off;
1318 KKASSERT(uio->uio_offset == off);
1324 /* --------------------------------------------------------------------- */
1327 tmpfs_readlink(struct vop_readlink_args *v)
1329 struct vnode *vp = v->a_vp;
1330 struct uio *uio = v->a_uio;
1333 struct tmpfs_node *node;
1335 KKASSERT(uio->uio_offset == 0);
1336 KKASSERT(vp->v_type == VLNK);
1338 node = VP_TO_TMPFS_NODE(vp);
1340 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1342 TMPFS_NODE_LOCK(node);
1343 node->tn_status |= TMPFS_NODE_ACCESSED;
1344 TMPFS_NODE_UNLOCK(node);
1349 /* --------------------------------------------------------------------- */
1352 tmpfs_inactive(struct vop_inactive_args *v)
1354 struct vnode *vp = v->a_vp;
1356 struct tmpfs_node *node;
1358 node = VP_TO_TMPFS_NODE(vp);
1361 * Get rid of unreferenced deleted vnodes sooner rather than
1362 * later so the data memory can be recovered immediately.
1364 * We must truncate the vnode to prevent the normal reclamation
1365 * path from flushing the data for the removed file to disk.
1367 TMPFS_NODE_LOCK(node);
1368 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1369 (node->tn_links == 0 ||
1370 (node->tn_links == 1 && node->tn_type == VDIR &&
1371 node->tn_dir.tn_parent)))
1373 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1374 TMPFS_NODE_UNLOCK(node);
1375 if (node->tn_type == VREG)
1376 tmpfs_truncate(vp, 0);
1379 TMPFS_NODE_UNLOCK(node);
1385 /* --------------------------------------------------------------------- */
1388 tmpfs_reclaim(struct vop_reclaim_args *v)
1390 struct vnode *vp = v->a_vp;
1391 struct tmpfs_mount *tmp;
1392 struct tmpfs_node *node;
1394 node = VP_TO_TMPFS_NODE(vp);
1395 tmp = VFS_TO_TMPFS(vp->v_mount);
1400 * If the node referenced by this vnode was deleted by the
1401 * user, we must free its associated data structures now that
1402 * the vnode is being reclaimed.
1404 * Directories have an extra link ref.
1406 TMPFS_NODE_LOCK(node);
1407 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1408 (node->tn_links == 0 ||
1409 (node->tn_links == 1 && node->tn_type == VDIR &&
1410 node->tn_dir.tn_parent)))
1412 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1413 tmpfs_free_node(tmp, node);
1416 TMPFS_NODE_UNLOCK(node);
1419 KKASSERT(vp->v_data == NULL);
1423 /* --------------------------------------------------------------------- */
1426 tmpfs_print(struct vop_print_args *v)
1428 struct vnode *vp = v->a_vp;
1430 struct tmpfs_node *node;
1432 node = VP_TO_TMPFS_NODE(vp);
1434 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1435 node, node->tn_flags, node->tn_links);
1436 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1437 node->tn_mode, node->tn_uid, node->tn_gid,
1438 (uintmax_t)node->tn_size, node->tn_status);
1440 if (vp->v_type == VFIFO)
1448 /* --------------------------------------------------------------------- */
1451 tmpfs_pathconf(struct vop_pathconf_args *v)
1453 int name = v->a_name;
1454 register_t *retval = v->a_retval;
1477 case _PC_CHOWN_RESTRICTED:
1489 case _PC_FILESIZEBITS:
1490 *retval = 0; /* XXX Don't know which value should I return. */
1500 /************************************************************************
1502 ************************************************************************/
1504 static void filt_tmpfsdetach(struct knote *kn);
1505 static int filt_tmpfsread(struct knote *kn, long hint);
1506 static int filt_tmpfswrite(struct knote *kn, long hint);
1507 static int filt_tmpfsvnode(struct knote *kn, long hint);
1509 static struct filterops tmpfsread_filtops =
1510 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1511 static struct filterops tmpfswrite_filtops =
1512 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1513 static struct filterops tmpfsvnode_filtops =
1514 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1517 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1519 struct vnode *vp = ap->a_vp;
1520 struct knote *kn = ap->a_kn;
1522 switch (kn->kn_filter) {
1524 kn->kn_fop = &tmpfsread_filtops;
1527 kn->kn_fop = &tmpfswrite_filtops;
1530 kn->kn_fop = &tmpfsvnode_filtops;
1533 return (EOPNOTSUPP);
1536 kn->kn_hook = (caddr_t)vp;
1538 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1544 filt_tmpfsdetach(struct knote *kn)
1546 struct vnode *vp = (void *)kn->kn_hook;
1548 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1552 filt_tmpfsread(struct knote *kn, long hint)
1554 struct vnode *vp = (void *)kn->kn_hook;
1555 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1558 if (hint == NOTE_REVOKE) {
1559 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1564 * Interlock against MP races when performing this function.
1566 lwkt_gettoken(&vp->v_mount->mnt_token);
1567 off = node->tn_size - kn->kn_fp->f_offset;
1568 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1569 if (kn->kn_sfflags & NOTE_OLDAPI) {
1570 lwkt_reltoken(&vp->v_mount->mnt_token);
1574 if (kn->kn_data == 0) {
1575 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1577 lwkt_reltoken(&vp->v_mount->mnt_token);
1578 return (kn->kn_data != 0);
1582 filt_tmpfswrite(struct knote *kn, long hint)
1584 if (hint == NOTE_REVOKE)
1585 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1591 filt_tmpfsvnode(struct knote *kn, long hint)
1593 if (kn->kn_sfflags & hint)
1594 kn->kn_fflags |= hint;
1595 if (hint == NOTE_REVOKE) {
1596 kn->kn_flags |= EV_EOF;
1599 return (kn->kn_fflags != 0);
1603 /* --------------------------------------------------------------------- */
1606 * vnode operations vector used for files stored in a tmpfs file system.
1608 struct vop_ops tmpfs_vnode_vops = {
1609 .vop_default = vop_defaultop,
1610 .vop_getpages = vop_stdgetpages,
1611 .vop_putpages = vop_stdputpages,
1612 .vop_ncreate = tmpfs_ncreate,
1613 .vop_nresolve = tmpfs_nresolve,
1614 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1615 .vop_nmknod = tmpfs_nmknod,
1616 .vop_open = tmpfs_open,
1617 .vop_close = tmpfs_close,
1618 .vop_access = tmpfs_access,
1619 .vop_getattr = tmpfs_getattr,
1620 .vop_setattr = tmpfs_setattr,
1621 .vop_read = tmpfs_read,
1622 .vop_write = tmpfs_write,
1623 .vop_fsync = tmpfs_fsync,
1624 .vop_nremove = tmpfs_nremove,
1625 .vop_nlink = tmpfs_nlink,
1626 .vop_nrename = tmpfs_nrename,
1627 .vop_nmkdir = tmpfs_nmkdir,
1628 .vop_nrmdir = tmpfs_nrmdir,
1629 .vop_nsymlink = tmpfs_nsymlink,
1630 .vop_readdir = tmpfs_readdir,
1631 .vop_readlink = tmpfs_readlink,
1632 .vop_inactive = tmpfs_inactive,
1633 .vop_reclaim = tmpfs_reclaim,
1634 .vop_print = tmpfs_print,
1635 .vop_pathconf = tmpfs_pathconf,
1636 .vop_bmap = tmpfs_bmap,
1637 .vop_strategy = tmpfs_strategy,
1638 .vop_advlock = tmpfs_advlock,
1639 .vop_kqfilter = tmpfs_kqfilter