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>
51 #include <sys/mountctl.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57 #include <vm/swap_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 static void tmpfs_strategy_done(struct bio *bio);
71 tmpfs_knote(struct vnode *vp, int flags)
74 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
78 /* --------------------------------------------------------------------- */
81 tmpfs_nresolve(struct vop_nresolve_args *v)
83 struct vnode *dvp = v->a_dvp;
84 struct vnode *vp = NULL;
85 struct namecache *ncp = v->a_nch->ncp;
86 struct tmpfs_node *tnode;
89 struct tmpfs_dirent *de;
90 struct tmpfs_node *dnode;
92 dnode = VP_TO_TMPFS_DIR(dvp);
94 de = tmpfs_dir_lookup(dnode, NULL, ncp);
99 * Allocate a vnode for the node we found.
102 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
103 LK_EXCLUSIVE | LK_RETRY, &vp);
111 * Store the result of this lookup in the cache. Avoid this if the
112 * request was for creation, as it does not improve timings on
117 cache_setvp(v->a_nch, vp);
119 } else if (error == ENOENT) {
120 cache_setvp(v->a_nch, NULL);
126 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
128 struct vnode *dvp = v->a_dvp;
129 struct vnode **vpp = v->a_vpp;
130 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
131 struct ucred *cred = v->a_cred;
135 /* Check accessibility of requested node as a first step. */
136 error = VOP_ACCESS(dvp, VEXEC, cred);
140 if (dnode->tn_dir.tn_parent != NULL) {
141 /* Allocate a new vnode on the matching entry. */
142 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
143 LK_EXCLUSIVE | LK_RETRY, vpp);
149 return (*vpp == NULL) ? ENOENT : 0;
152 /* --------------------------------------------------------------------- */
155 tmpfs_ncreate(struct vop_ncreate_args *v)
157 struct vnode *dvp = v->a_dvp;
158 struct vnode **vpp = v->a_vpp;
159 struct namecache *ncp = v->a_nch->ncp;
160 struct vattr *vap = v->a_vap;
161 struct ucred *cred = v->a_cred;
164 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
166 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
168 cache_setunresolved(v->a_nch);
169 cache_setvp(v->a_nch, *vpp);
170 tmpfs_knote(dvp, NOTE_WRITE);
175 /* --------------------------------------------------------------------- */
178 tmpfs_nmknod(struct vop_nmknod_args *v)
180 struct vnode *dvp = v->a_dvp;
181 struct vnode **vpp = v->a_vpp;
182 struct namecache *ncp = v->a_nch->ncp;
183 struct vattr *vap = v->a_vap;
184 struct ucred *cred = v->a_cred;
187 if (vap->va_type != VBLK && vap->va_type != VCHR &&
188 vap->va_type != VFIFO)
191 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
193 cache_setunresolved(v->a_nch);
194 cache_setvp(v->a_nch, *vpp);
195 tmpfs_knote(dvp, NOTE_WRITE);
201 /* --------------------------------------------------------------------- */
204 tmpfs_open(struct vop_open_args *v)
206 struct vnode *vp = v->a_vp;
207 int mode = v->a_mode;
210 struct tmpfs_node *node;
212 node = VP_TO_TMPFS_NODE(vp);
214 /* The file is still active but all its names have been removed
215 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
216 * it is about to die. */
217 if (node->tn_links < 1)
220 /* If the file is marked append-only, deny write requests. */
221 if ((node->tn_flags & APPEND) &&
222 (mode & (FWRITE | O_APPEND)) == FWRITE) {
225 return (vop_stdopen(v));
230 /* --------------------------------------------------------------------- */
233 tmpfs_close(struct vop_close_args *v)
235 struct vnode *vp = v->a_vp;
236 struct tmpfs_node *node;
238 node = VP_TO_TMPFS_NODE(vp);
240 if (node->tn_links > 0) {
241 /* Update node times. No need to do it if the node has
242 * been deleted, because it will vanish after we return. */
246 return vop_stdclose(v);
249 /* --------------------------------------------------------------------- */
252 tmpfs_access(struct vop_access_args *v)
254 struct vnode *vp = v->a_vp;
256 struct tmpfs_node *node;
258 node = VP_TO_TMPFS_NODE(vp);
260 switch (vp->v_type) {
266 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
286 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
291 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
298 /* --------------------------------------------------------------------- */
301 tmpfs_getattr(struct vop_getattr_args *v)
303 struct vnode *vp = v->a_vp;
304 struct vattr *vap = v->a_vap;
305 struct tmpfs_node *node;
307 node = VP_TO_TMPFS_NODE(vp);
309 lwkt_gettoken(&vp->v_mount->mnt_token);
312 vap->va_type = vp->v_type;
313 vap->va_mode = node->tn_mode;
314 vap->va_nlink = node->tn_links;
315 vap->va_uid = node->tn_uid;
316 vap->va_gid = node->tn_gid;
317 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
318 vap->va_fileid = node->tn_id;
319 vap->va_size = node->tn_size;
320 vap->va_blocksize = PAGE_SIZE;
321 vap->va_atime.tv_sec = node->tn_atime;
322 vap->va_atime.tv_nsec = node->tn_atimensec;
323 vap->va_mtime.tv_sec = node->tn_mtime;
324 vap->va_mtime.tv_nsec = node->tn_mtimensec;
325 vap->va_ctime.tv_sec = node->tn_ctime;
326 vap->va_ctime.tv_nsec = node->tn_ctimensec;
327 vap->va_gen = node->tn_gen;
328 vap->va_flags = node->tn_flags;
329 if (vp->v_type == VBLK || vp->v_type == VCHR)
331 vap->va_rmajor = umajor(node->tn_rdev);
332 vap->va_rminor = uminor(node->tn_rdev);
334 vap->va_bytes = round_page(node->tn_size);
337 lwkt_reltoken(&vp->v_mount->mnt_token);
342 /* --------------------------------------------------------------------- */
345 tmpfs_setattr(struct vop_setattr_args *v)
347 struct vnode *vp = v->a_vp;
348 struct vattr *vap = v->a_vap;
349 struct ucred *cred = v->a_cred;
350 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
354 if (error == 0 && (vap->va_flags != VNOVAL)) {
355 error = tmpfs_chflags(vp, vap->va_flags, cred);
356 kflags |= NOTE_ATTRIB;
359 if (error == 0 && (vap->va_size != VNOVAL)) {
360 if (vap->va_size > node->tn_size)
361 kflags |= NOTE_WRITE | NOTE_EXTEND;
363 kflags |= NOTE_WRITE;
364 error = tmpfs_chsize(vp, vap->va_size, cred);
367 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
368 vap->va_gid != (gid_t)VNOVAL)) {
369 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
370 kflags |= NOTE_ATTRIB;
373 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
374 error = tmpfs_chmod(vp, vap->va_mode, cred);
375 kflags |= NOTE_ATTRIB;
378 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
379 vap->va_atime.tv_nsec != VNOVAL) ||
380 (vap->va_mtime.tv_sec != VNOVAL &&
381 vap->va_mtime.tv_nsec != VNOVAL) )) {
382 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
383 vap->va_vaflags, cred);
384 kflags |= NOTE_ATTRIB;
387 /* Update the node times. We give preference to the error codes
388 * generated by this function rather than the ones that may arise
389 * from tmpfs_update. */
391 tmpfs_knote(vp, kflags);
396 /* --------------------------------------------------------------------- */
399 * fsync is usually a NOP, but we must take action when unmounting or
403 tmpfs_fsync(struct vop_fsync_args *v)
405 struct tmpfs_node *node;
406 struct vnode *vp = v->a_vp;
408 node = VP_TO_TMPFS_NODE(vp);
411 if (vp->v_type == VREG) {
412 if (vp->v_flag & VRECLAIMED) {
413 if (node->tn_links == 0)
414 tmpfs_truncate(vp, 0);
416 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
422 /* --------------------------------------------------------------------- */
425 tmpfs_read (struct vop_read_args *ap)
428 struct vnode *vp = ap->a_vp;
429 struct uio *uio = ap->a_uio;
430 struct tmpfs_node *node;
437 if (uio->uio_resid == 0) {
441 node = VP_TO_TMPFS_NODE(vp);
443 if (uio->uio_offset < 0)
445 if (vp->v_type != VREG)
448 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
450 * Use buffer cache I/O (via tmpfs_strategy)
452 offset = (size_t)uio->uio_offset & BMASK;
453 base_offset = (off_t)uio->uio_offset - offset;
454 bp = getcacheblk(vp, base_offset, BSIZE, 0);
456 lwkt_gettoken(&vp->v_mount->mnt_token);
457 error = bread(vp, base_offset, BSIZE, &bp);
460 lwkt_reltoken(&vp->v_mount->mnt_token);
461 kprintf("tmpfs_read bread error %d\n", error);
464 lwkt_reltoken(&vp->v_mount->mnt_token);
468 * Figure out how many bytes we can actually copy this loop.
470 len = BSIZE - offset;
471 if (len > uio->uio_resid)
472 len = uio->uio_resid;
473 if (len > node->tn_size - uio->uio_offset)
474 len = (size_t)(node->tn_size - uio->uio_offset);
476 error = uiomove((char *)bp->b_data + offset, len, uio);
479 kprintf("tmpfs_read uiomove error %d\n", error);
484 TMPFS_NODE_LOCK(node);
485 node->tn_status |= TMPFS_NODE_ACCESSED;
486 TMPFS_NODE_UNLOCK(node);
492 tmpfs_write (struct vop_write_args *ap)
495 struct vnode *vp = ap->a_vp;
496 struct uio *uio = ap->a_uio;
497 struct thread *td = uio->uio_td;
498 struct tmpfs_node *node;
510 if (uio->uio_resid == 0) {
514 node = VP_TO_TMPFS_NODE(vp);
516 if (vp->v_type != VREG)
519 lwkt_gettoken(&vp->v_mount->mnt_token);
521 oldsize = node->tn_size;
522 if (ap->a_ioflag & IO_APPEND)
523 uio->uio_offset = node->tn_size;
526 * Check for illegal write offsets.
528 if (uio->uio_offset + uio->uio_resid >
529 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
530 lwkt_reltoken(&vp->v_mount->mnt_token);
534 if (vp->v_type == VREG && td != NULL) {
535 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
537 lwkt_reltoken(&vp->v_mount->mnt_token);
540 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
541 ksignal(td->td_proc, SIGXFSZ);
542 lwkt_reltoken(&vp->v_mount->mnt_token);
549 * Extend the file's size if necessary
551 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
553 while (uio->uio_resid > 0) {
555 * Use buffer cache I/O (via tmpfs_strategy)
557 offset = (size_t)uio->uio_offset & BMASK;
558 base_offset = (off_t)uio->uio_offset - offset;
559 len = BSIZE - offset;
560 if (len > uio->uio_resid)
561 len = uio->uio_resid;
563 if ((uio->uio_offset + len) > node->tn_size) {
564 trivial = (uio->uio_offset <= node->tn_size);
565 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
571 * Read to fill in any gaps. Theoretically we could
572 * optimize this if the write covers the entire buffer
573 * and is not a UIO_NOCOPY write, however this can lead
574 * to a security violation exposing random kernel memory
575 * (whatever junk was in the backing VM pages before).
577 * So just use bread() to do the right thing.
579 error = bread(vp, base_offset, BSIZE, &bp);
580 error = uiomove((char *)bp->b_data + offset, len, uio);
582 kprintf("tmpfs_write uiomove error %d\n", error);
587 if (uio->uio_offset > node->tn_size) {
588 node->tn_size = uio->uio_offset;
589 kflags |= NOTE_EXTEND;
591 kflags |= NOTE_WRITE;
594 * Always try to flush the page if the request is coming
595 * from the pageout daemon (IO_ASYNC), else buwrite() the
598 * buwrite() dirties the underlying VM pages instead of
599 * dirtying the buffer, releasing the buffer as a clean
600 * buffer. This allows tmpfs to use essentially all
601 * available memory to cache file data. If we used bdwrite()
602 * the buffer cache would wind up flushing the data to
605 bp->b_flags |= B_AGE;
606 if (ap->a_ioflag & IO_ASYNC) {
613 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
620 (void)tmpfs_reg_resize(vp, oldsize, trivial);
621 kflags &= ~NOTE_EXTEND;
627 * Currently we don't set the mtime on files modified via mmap()
628 * because we can't tell the difference between those modifications
629 * and an attempt by the pageout daemon to flush tmpfs pages to
632 * This is because in order to defer flushes as long as possible
633 * buwrite() works by marking the underlying VM pages dirty in
634 * order to be able to dispose of the buffer cache buffer without
637 TMPFS_NODE_LOCK(node);
638 if (uio->uio_segflg != UIO_NOCOPY)
639 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
641 node->tn_status |= TMPFS_NODE_CHANGED;
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 * The strategy function is typically only called when memory pressure
670 * forces the system to attempt to pageout pages. It can also be called
671 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write
675 tmpfs_strategy(struct vop_strategy_args *ap)
677 struct bio *bio = ap->a_bio;
679 struct buf *bp = bio->bio_buf;
680 struct vnode *vp = ap->a_vp;
681 struct tmpfs_node *node;
686 if (vp->v_type != VREG) {
687 bp->b_resid = bp->b_bcount;
688 bp->b_flags |= B_ERROR | B_INVAL;
689 bp->b_error = EINVAL;
694 lwkt_gettoken(&vp->v_mount->mnt_token);
695 node = VP_TO_TMPFS_NODE(vp);
697 uobj = node->tn_reg.tn_aobj;
700 * Don't bother flushing to swap if there is no swap, just
701 * ensure that the pages are marked as needing a commit (still).
703 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
704 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
705 m = bp->b_xio.xio_pages[i];
706 vm_page_need_commit(m);
712 nbio = push_bio(bio);
713 nbio->bio_done = tmpfs_strategy_done;
714 nbio->bio_offset = bio->bio_offset;
715 swap_pager_strategy(uobj, nbio);
718 lwkt_reltoken(&vp->v_mount->mnt_token);
723 * If we were unable to commit the pages to swap make sure they are marked
724 * as needing a commit (again). If we were, clear the flag to allow the
728 tmpfs_strategy_done(struct bio *bio)
736 if (bp->b_flags & B_ERROR) {
737 bp->b_flags &= ~B_ERROR;
740 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
741 m = bp->b_xio.xio_pages[i];
742 vm_page_need_commit(m);
745 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
746 m = bp->b_xio.xio_pages[i];
747 vm_page_clear_commit(m);
755 tmpfs_bmap(struct vop_bmap_args *ap)
757 if (ap->a_doffsetp != NULL)
758 *ap->a_doffsetp = ap->a_loffset;
759 if (ap->a_runp != NULL)
761 if (ap->a_runb != NULL)
767 /* --------------------------------------------------------------------- */
770 tmpfs_nremove(struct vop_nremove_args *v)
772 struct vnode *dvp = v->a_dvp;
773 struct namecache *ncp = v->a_nch->ncp;
776 struct tmpfs_dirent *de;
777 struct tmpfs_mount *tmp;
778 struct tmpfs_node *dnode;
779 struct tmpfs_node *node;
782 * We have to acquire the vp from v->a_nch because we will likely
783 * unresolve the namecache entry, and a vrele/vput is needed to
784 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
786 * We have to use vget to clear any inactive state on the vnode,
787 * otherwise the vnode may remain inactive and thus tmpfs_inactive
788 * will not get called when we release it.
790 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
791 KKASSERT(error == 0);
794 if (vp->v_type == VDIR) {
799 dnode = VP_TO_TMPFS_DIR(dvp);
800 node = VP_TO_TMPFS_NODE(vp);
801 tmp = VFS_TO_TMPFS(vp->v_mount);
802 de = tmpfs_dir_lookup(dnode, node, ncp);
808 /* Files marked as immutable or append-only cannot be deleted. */
809 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
810 (dnode->tn_flags & APPEND)) {
815 /* Remove the entry from the directory; as it is a file, we do not
816 * have to change the number of hard links of the directory. */
817 tmpfs_dir_detach(dnode, de);
819 /* Free the directory entry we just deleted. Note that the node
820 * referred by it will not be removed until the vnode is really
822 tmpfs_free_dirent(tmp, de);
824 if (node->tn_links > 0) {
825 TMPFS_NODE_LOCK(node);
826 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
828 TMPFS_NODE_UNLOCK(node);
831 cache_setunresolved(v->a_nch);
832 cache_setvp(v->a_nch, NULL);
833 tmpfs_knote(vp, NOTE_DELETE);
834 /*cache_inval_vp(vp, CINV_DESTROY);*/
835 tmpfs_knote(dvp, NOTE_WRITE);
844 /* --------------------------------------------------------------------- */
847 tmpfs_nlink(struct vop_nlink_args *v)
849 struct vnode *dvp = v->a_dvp;
850 struct vnode *vp = v->a_vp;
851 struct namecache *ncp = v->a_nch->ncp;
852 struct tmpfs_dirent *de;
853 struct tmpfs_node *node;
854 struct tmpfs_node *dnode;
857 KKASSERT(dvp != vp); /* XXX When can this be false? */
859 node = VP_TO_TMPFS_NODE(vp);
860 dnode = VP_TO_TMPFS_NODE(dvp);
862 /* XXX: Why aren't the following two tests done by the caller? */
864 /* Hard links of directories are forbidden. */
865 if (vp->v_type == VDIR) {
870 /* Cannot create cross-device links. */
871 if (dvp->v_mount != vp->v_mount) {
876 /* Ensure that we do not overflow the maximum number of links imposed
878 KKASSERT(node->tn_links <= LINK_MAX);
879 if (node->tn_links == LINK_MAX) {
884 /* We cannot create links of files marked immutable or append-only. */
885 if (node->tn_flags & (IMMUTABLE | APPEND)) {
890 /* Allocate a new directory entry to represent the node. */
891 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
892 ncp->nc_name, ncp->nc_nlen, &de);
896 /* Insert the new directory entry into the appropriate directory. */
897 tmpfs_dir_attach(dnode, de);
899 /* vp link count has changed, so update node times. */
901 TMPFS_NODE_LOCK(node);
902 node->tn_status |= TMPFS_NODE_CHANGED;
903 TMPFS_NODE_UNLOCK(node);
906 tmpfs_knote(vp, NOTE_LINK);
907 cache_setunresolved(v->a_nch);
908 cache_setvp(v->a_nch, vp);
909 tmpfs_knote(dvp, NOTE_WRITE);
916 /* --------------------------------------------------------------------- */
919 tmpfs_nrename(struct vop_nrename_args *v)
921 struct vnode *fdvp = v->a_fdvp;
922 struct namecache *fncp = v->a_fnch->ncp;
923 struct vnode *fvp = fncp->nc_vp;
924 struct vnode *tdvp = v->a_tdvp;
925 struct namecache *tncp = v->a_tnch->ncp;
927 struct tmpfs_dirent *de, *tde;
928 struct tmpfs_mount *tmp;
929 struct tmpfs_node *fdnode;
930 struct tmpfs_node *fnode;
931 struct tmpfs_node *tnode;
932 struct tmpfs_node *tdnode;
938 * Because tvp can get overwritten we have to vget it instead of
939 * just vref or use it, otherwise it's VINACTIVE flag may not get
940 * cleared and the node won't get destroyed.
942 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
944 tnode = VP_TO_TMPFS_NODE(tvp);
950 /* Disallow cross-device renames.
951 * XXX Why isn't this done by the caller? */
952 if (fvp->v_mount != tdvp->v_mount ||
953 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
958 tmp = VFS_TO_TMPFS(tdvp->v_mount);
959 tdnode = VP_TO_TMPFS_DIR(tdvp);
961 /* If source and target are the same file, there is nothing to do. */
967 fdnode = VP_TO_TMPFS_DIR(fdvp);
968 fnode = VP_TO_TMPFS_NODE(fvp);
969 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
971 /* Avoid manipulating '.' and '..' entries. */
976 KKASSERT(de->td_node == fnode);
979 * If replacing an entry in the target directory and that entry
980 * is a directory, it must be empty.
982 * Kern_rename gurantees the destination to be a directory
983 * if the source is one (it does?).
986 KKASSERT(tnode != NULL);
988 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
989 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
994 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
995 if (tnode->tn_size > 0) {
999 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1002 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1006 KKASSERT(fnode->tn_type != VDIR &&
1007 tnode->tn_type != VDIR);
1011 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1012 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1018 * Ensure that we have enough memory to hold the new name, if it
1019 * has to be changed.
1021 if (fncp->nc_nlen != tncp->nc_nlen ||
1022 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1023 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1024 M_WAITOK | M_NULLOK);
1025 if (newname == NULL) {
1029 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1030 newname[tncp->nc_nlen] = '\0';
1036 * Unlink entry from source directory. Note that the kernel has
1037 * already checked for illegal recursion cases (renaming a directory
1038 * into a subdirectory of itself).
1040 if (fdnode != tdnode)
1041 tmpfs_dir_detach(fdnode, de);
1043 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1047 * Handle any name change. Swap with newname, we will
1048 * deallocate it at the end.
1050 if (newname != NULL) {
1052 TMPFS_NODE_LOCK(fnode);
1053 fnode->tn_status |= TMPFS_NODE_CHANGED;
1054 TMPFS_NODE_UNLOCK(fnode);
1056 oldname = de->td_name;
1057 de->td_name = newname;
1058 de->td_namelen = (uint16_t)tncp->nc_nlen;
1063 * If we are overwriting an entry, we have to remove the old one
1064 * from the target directory.
1067 /* Remove the old entry from the target directory. */
1068 tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1069 tmpfs_dir_detach(tdnode, tde);
1070 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1073 * Free the directory entry we just deleted. Note that the
1074 * node referred by it will not be removed until the vnode is
1077 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1078 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1082 * Link entry to target directory. If the entry
1083 * represents a directory move the parent linkage
1086 if (fdnode != tdnode) {
1087 if (de->td_node->tn_type == VDIR) {
1088 TMPFS_VALIDATE_DIR(fnode);
1090 TMPFS_NODE_LOCK(tdnode);
1092 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1093 TMPFS_NODE_UNLOCK(tdnode);
1095 TMPFS_NODE_LOCK(fnode);
1096 fnode->tn_dir.tn_parent = tdnode;
1097 fnode->tn_status |= TMPFS_NODE_CHANGED;
1098 TMPFS_NODE_UNLOCK(fnode);
1100 TMPFS_NODE_LOCK(fdnode);
1102 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1103 TMPFS_NODE_UNLOCK(fdnode);
1105 tmpfs_dir_attach(tdnode, de);
1107 TMPFS_NODE_LOCK(tdnode);
1108 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1109 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1110 TMPFS_NODE_UNLOCK(tdnode);
1117 kfree(newname, tmp->tm_name_zone);
1120 cache_rename(v->a_fnch, v->a_tnch);
1121 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1122 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1123 if (fnode->tn_vnode)
1124 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1137 /* --------------------------------------------------------------------- */
1140 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1142 struct vnode *dvp = v->a_dvp;
1143 struct vnode **vpp = v->a_vpp;
1144 struct namecache *ncp = v->a_nch->ncp;
1145 struct vattr *vap = v->a_vap;
1146 struct ucred *cred = v->a_cred;
1149 KKASSERT(vap->va_type == VDIR);
1151 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1153 cache_setunresolved(v->a_nch);
1154 cache_setvp(v->a_nch, *vpp);
1155 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1161 /* --------------------------------------------------------------------- */
1164 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1166 struct vnode *dvp = v->a_dvp;
1167 struct namecache *ncp = v->a_nch->ncp;
1169 struct tmpfs_dirent *de;
1170 struct tmpfs_mount *tmp;
1171 struct tmpfs_node *dnode;
1172 struct tmpfs_node *node;
1176 * We have to acquire the vp from v->a_nch because we will likely
1177 * unresolve the namecache entry, and a vrele/vput is needed to
1178 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1180 * We have to use vget to clear any inactive state on the vnode,
1181 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1182 * will not get called when we release it.
1184 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1185 KKASSERT(error == 0);
1189 * Prevalidate so we don't hit an assertion later
1191 if (vp->v_type != VDIR) {
1196 tmp = VFS_TO_TMPFS(dvp->v_mount);
1197 dnode = VP_TO_TMPFS_DIR(dvp);
1198 node = VP_TO_TMPFS_DIR(vp);
1200 /* Directories with more than two entries ('.' and '..') cannot be
1202 if (node->tn_size > 0) {
1207 if ((dnode->tn_flags & APPEND)
1208 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1213 /* This invariant holds only if we are not trying to remove "..".
1214 * We checked for that above so this is safe now. */
1215 KKASSERT(node->tn_dir.tn_parent == dnode);
1217 /* Get the directory entry associated with node (vp). This was
1218 * filled by tmpfs_lookup while looking up the entry. */
1219 de = tmpfs_dir_lookup(dnode, node, ncp);
1220 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1224 /* Check flags to see if we are allowed to remove the directory. */
1225 if ((dnode->tn_flags & APPEND) ||
1226 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1232 /* Detach the directory entry from the directory (dnode). */
1233 tmpfs_dir_detach(dnode, de);
1235 /* No vnode should be allocated for this entry from this point */
1236 TMPFS_NODE_LOCK(node);
1237 TMPFS_ASSERT_ELOCKED(node);
1238 TMPFS_NODE_LOCK(dnode);
1239 TMPFS_ASSERT_ELOCKED(dnode);
1242 /* handled by tmpfs_free_node */
1243 KKASSERT(node->tn_links > 0);
1245 node->tn_dir.tn_parent = NULL;
1247 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1248 TMPFS_NODE_MODIFIED;
1251 /* handled by tmpfs_free_node */
1252 KKASSERT(dnode->tn_links > 0);
1255 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1256 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1258 TMPFS_NODE_UNLOCK(dnode);
1259 TMPFS_NODE_UNLOCK(node);
1261 /* Free the directory entry we just deleted. Note that the node
1262 * referred by it will not be removed until the vnode is really
1264 tmpfs_free_dirent(tmp, de);
1266 /* Release the deleted vnode (will destroy the node, notify
1267 * interested parties and clean it from the cache). */
1269 TMPFS_NODE_LOCK(dnode);
1270 dnode->tn_status |= TMPFS_NODE_CHANGED;
1271 TMPFS_NODE_UNLOCK(dnode);
1274 cache_setunresolved(v->a_nch);
1275 cache_setvp(v->a_nch, NULL);
1276 /*cache_inval_vp(vp, CINV_DESTROY);*/
1277 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1286 /* --------------------------------------------------------------------- */
1289 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1291 struct vnode *dvp = v->a_dvp;
1292 struct vnode **vpp = v->a_vpp;
1293 struct namecache *ncp = v->a_nch->ncp;
1294 struct vattr *vap = v->a_vap;
1295 struct ucred *cred = v->a_cred;
1296 char *target = v->a_target;
1299 vap->va_type = VLNK;
1300 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1302 tmpfs_knote(*vpp, NOTE_WRITE);
1303 cache_setunresolved(v->a_nch);
1304 cache_setvp(v->a_nch, *vpp);
1310 /* --------------------------------------------------------------------- */
1313 tmpfs_readdir(struct vop_readdir_args *v)
1315 struct vnode *vp = v->a_vp;
1316 struct uio *uio = v->a_uio;
1317 int *eofflag = v->a_eofflag;
1318 off_t **cookies = v->a_cookies;
1319 int *ncookies = v->a_ncookies;
1320 struct tmpfs_mount *tmp;
1324 struct tmpfs_node *node;
1326 /* This operation only makes sense on directory nodes. */
1327 if (vp->v_type != VDIR)
1330 tmp = VFS_TO_TMPFS(vp->v_mount);
1331 node = VP_TO_TMPFS_DIR(vp);
1332 startoff = uio->uio_offset;
1334 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1335 error = tmpfs_dir_getdotdent(node, uio);
1341 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1342 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1348 error = tmpfs_dir_getdents(node, uio, &cnt);
1351 KKASSERT(error >= -1);
1356 if (eofflag != NULL)
1358 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1360 /* Update NFS-related variables. */
1361 if (error == 0 && cookies != NULL && ncookies != NULL) {
1363 off_t off = startoff;
1364 struct tmpfs_dirent *de = NULL;
1367 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1369 for (i = 0; i < cnt; i++) {
1370 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1371 if (off == TMPFS_DIRCOOKIE_DOT) {
1372 off = TMPFS_DIRCOOKIE_DOTDOT;
1374 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1375 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
1376 } else if (de != NULL) {
1377 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1379 de = tmpfs_dir_lookupbycookie(node,
1381 KKASSERT(de != NULL);
1382 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1385 off = TMPFS_DIRCOOKIE_EOF;
1387 off = tmpfs_dircookie(de);
1390 (*cookies)[i] = off;
1392 KKASSERT(uio->uio_offset == off);
1398 /* --------------------------------------------------------------------- */
1401 tmpfs_readlink(struct vop_readlink_args *v)
1403 struct vnode *vp = v->a_vp;
1404 struct uio *uio = v->a_uio;
1407 struct tmpfs_node *node;
1409 KKASSERT(uio->uio_offset == 0);
1410 KKASSERT(vp->v_type == VLNK);
1412 node = VP_TO_TMPFS_NODE(vp);
1414 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1416 TMPFS_NODE_LOCK(node);
1417 node->tn_status |= TMPFS_NODE_ACCESSED;
1418 TMPFS_NODE_UNLOCK(node);
1423 /* --------------------------------------------------------------------- */
1426 tmpfs_inactive(struct vop_inactive_args *v)
1428 struct vnode *vp = v->a_vp;
1429 struct tmpfs_node *node;
1431 node = VP_TO_TMPFS_NODE(vp);
1442 * Get rid of unreferenced deleted vnodes sooner rather than
1443 * later so the data memory can be recovered immediately.
1445 * We must truncate the vnode to prevent the normal reclamation
1446 * path from flushing the data for the removed file to disk.
1448 TMPFS_NODE_LOCK(node);
1449 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1450 (node->tn_links == 0 ||
1451 (node->tn_links == 1 && node->tn_type == VDIR &&
1452 node->tn_dir.tn_parent)))
1454 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1455 TMPFS_NODE_UNLOCK(node);
1456 if (node->tn_type == VREG)
1457 tmpfs_truncate(vp, 0);
1460 TMPFS_NODE_UNLOCK(node);
1466 /* --------------------------------------------------------------------- */
1469 tmpfs_reclaim(struct vop_reclaim_args *v)
1471 struct vnode *vp = v->a_vp;
1472 struct tmpfs_mount *tmp;
1473 struct tmpfs_node *node;
1475 node = VP_TO_TMPFS_NODE(vp);
1476 tmp = VFS_TO_TMPFS(vp->v_mount);
1481 * If the node referenced by this vnode was deleted by the
1482 * user, we must free its associated data structures now that
1483 * the vnode is being reclaimed.
1485 * Directories have an extra link ref.
1487 TMPFS_NODE_LOCK(node);
1488 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1489 (node->tn_links == 0 ||
1490 (node->tn_links == 1 && node->tn_type == VDIR &&
1491 node->tn_dir.tn_parent)))
1493 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1494 tmpfs_free_node(tmp, node);
1497 TMPFS_NODE_UNLOCK(node);
1500 KKASSERT(vp->v_data == NULL);
1504 /* --------------------------------------------------------------------- */
1507 tmpfs_mountctl(struct vop_mountctl_args *ap)
1509 struct tmpfs_mount *tmp;
1514 case (MOUNTCTL_SET_EXPORT):
1515 mp = ap->a_head.a_ops->head.vv_mount;
1516 tmp = (struct tmpfs_mount *) mp->mnt_data;
1518 if (ap->a_ctllen != sizeof(struct export_args))
1521 rc = vfs_export(mp, &tmp->tm_export,
1522 (const struct export_args *) ap->a_ctl);
1525 rc = vop_stdmountctl(ap);
1531 /* --------------------------------------------------------------------- */
1534 tmpfs_print(struct vop_print_args *v)
1536 struct vnode *vp = v->a_vp;
1538 struct tmpfs_node *node;
1540 node = VP_TO_TMPFS_NODE(vp);
1542 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1543 node, node->tn_flags, node->tn_links);
1544 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1545 node->tn_mode, node->tn_uid, node->tn_gid,
1546 (uintmax_t)node->tn_size, node->tn_status);
1548 if (vp->v_type == VFIFO)
1556 /* --------------------------------------------------------------------- */
1559 tmpfs_pathconf(struct vop_pathconf_args *v)
1561 int name = v->a_name;
1562 register_t *retval = v->a_retval;
1585 case _PC_CHOWN_RESTRICTED:
1597 case _PC_FILESIZEBITS:
1598 *retval = 0; /* XXX Don't know which value should I return. */
1608 /************************************************************************
1610 ************************************************************************/
1612 static void filt_tmpfsdetach(struct knote *kn);
1613 static int filt_tmpfsread(struct knote *kn, long hint);
1614 static int filt_tmpfswrite(struct knote *kn, long hint);
1615 static int filt_tmpfsvnode(struct knote *kn, long hint);
1617 static struct filterops tmpfsread_filtops =
1618 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1619 static struct filterops tmpfswrite_filtops =
1620 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1621 static struct filterops tmpfsvnode_filtops =
1622 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1625 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1627 struct vnode *vp = ap->a_vp;
1628 struct knote *kn = ap->a_kn;
1630 switch (kn->kn_filter) {
1632 kn->kn_fop = &tmpfsread_filtops;
1635 kn->kn_fop = &tmpfswrite_filtops;
1638 kn->kn_fop = &tmpfsvnode_filtops;
1641 return (EOPNOTSUPP);
1644 kn->kn_hook = (caddr_t)vp;
1646 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1652 filt_tmpfsdetach(struct knote *kn)
1654 struct vnode *vp = (void *)kn->kn_hook;
1656 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1660 filt_tmpfsread(struct knote *kn, long hint)
1662 struct vnode *vp = (void *)kn->kn_hook;
1663 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1666 if (hint == NOTE_REVOKE) {
1667 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1672 * Interlock against MP races when performing this function.
1674 lwkt_gettoken(&vp->v_mount->mnt_token);
1675 off = node->tn_size - kn->kn_fp->f_offset;
1676 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1677 if (kn->kn_sfflags & NOTE_OLDAPI) {
1678 lwkt_reltoken(&vp->v_mount->mnt_token);
1682 if (kn->kn_data == 0) {
1683 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1685 lwkt_reltoken(&vp->v_mount->mnt_token);
1686 return (kn->kn_data != 0);
1690 filt_tmpfswrite(struct knote *kn, long hint)
1692 if (hint == NOTE_REVOKE)
1693 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1699 filt_tmpfsvnode(struct knote *kn, long hint)
1701 if (kn->kn_sfflags & hint)
1702 kn->kn_fflags |= hint;
1703 if (hint == NOTE_REVOKE) {
1704 kn->kn_flags |= (EV_EOF | EV_NODATA);
1707 return (kn->kn_fflags != 0);
1711 /* --------------------------------------------------------------------- */
1714 * vnode operations vector used for files stored in a tmpfs file system.
1716 struct vop_ops tmpfs_vnode_vops = {
1717 .vop_default = vop_defaultop,
1718 .vop_getpages = vop_stdgetpages,
1719 .vop_putpages = vop_stdputpages,
1720 .vop_ncreate = tmpfs_ncreate,
1721 .vop_nresolve = tmpfs_nresolve,
1722 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1723 .vop_nmknod = tmpfs_nmknod,
1724 .vop_open = tmpfs_open,
1725 .vop_close = tmpfs_close,
1726 .vop_access = tmpfs_access,
1727 .vop_getattr = tmpfs_getattr,
1728 .vop_setattr = tmpfs_setattr,
1729 .vop_read = tmpfs_read,
1730 .vop_write = tmpfs_write,
1731 .vop_fsync = tmpfs_fsync,
1732 .vop_mountctl = tmpfs_mountctl,
1733 .vop_nremove = tmpfs_nremove,
1734 .vop_nlink = tmpfs_nlink,
1735 .vop_nrename = tmpfs_nrename,
1736 .vop_nmkdir = tmpfs_nmkdir,
1737 .vop_nrmdir = tmpfs_nrmdir,
1738 .vop_nsymlink = tmpfs_nsymlink,
1739 .vop_readdir = tmpfs_readdir,
1740 .vop_readlink = tmpfs_readlink,
1741 .vop_inactive = tmpfs_inactive,
1742 .vop_reclaim = tmpfs_reclaim,
1743 .vop_print = tmpfs_print,
1744 .vop_pathconf = tmpfs_pathconf,
1745 .vop_bmap = tmpfs_bmap,
1746 .vop_strategy = tmpfs_strategy,
1747 .vop_advlock = tmpfs_advlock,
1748 .vop_kqfilter = tmpfs_kqfilter