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);
215 /* The file is still active but all its names have been removed
216 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
217 * it is about to die. */
218 if (node->tn_links < 1)
222 /* If the file is marked append-only, deny write requests. */
223 if ((node->tn_flags & APPEND) &&
224 (mode & (FWRITE | O_APPEND)) == FWRITE) {
227 return (vop_stdopen(v));
232 /* --------------------------------------------------------------------- */
235 tmpfs_close(struct vop_close_args *v)
237 struct vnode *vp = v->a_vp;
238 struct tmpfs_node *node;
240 node = VP_TO_TMPFS_NODE(vp);
242 if (node->tn_links > 0) {
244 * Update node times. No need to do it if the node has
245 * been deleted, because it will vanish after we return.
250 return vop_stdclose(v);
253 /* --------------------------------------------------------------------- */
256 tmpfs_access(struct vop_access_args *v)
258 struct vnode *vp = v->a_vp;
260 struct tmpfs_node *node;
262 node = VP_TO_TMPFS_NODE(vp);
264 switch (vp->v_type) {
270 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
290 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
295 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
302 /* --------------------------------------------------------------------- */
305 tmpfs_getattr(struct vop_getattr_args *v)
307 struct vnode *vp = v->a_vp;
308 struct vattr *vap = v->a_vap;
309 struct tmpfs_node *node;
311 node = VP_TO_TMPFS_NODE(vp);
313 lwkt_gettoken(&vp->v_mount->mnt_token);
316 vap->va_type = vp->v_type;
317 vap->va_mode = node->tn_mode;
318 vap->va_nlink = node->tn_links;
319 vap->va_uid = node->tn_uid;
320 vap->va_gid = node->tn_gid;
321 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
322 vap->va_fileid = node->tn_id;
323 vap->va_size = node->tn_size;
324 vap->va_blocksize = PAGE_SIZE;
325 vap->va_atime.tv_sec = node->tn_atime;
326 vap->va_atime.tv_nsec = node->tn_atimensec;
327 vap->va_mtime.tv_sec = node->tn_mtime;
328 vap->va_mtime.tv_nsec = node->tn_mtimensec;
329 vap->va_ctime.tv_sec = node->tn_ctime;
330 vap->va_ctime.tv_nsec = node->tn_ctimensec;
331 vap->va_gen = node->tn_gen;
332 vap->va_flags = node->tn_flags;
333 if (vp->v_type == VBLK || vp->v_type == VCHR)
335 vap->va_rmajor = umajor(node->tn_rdev);
336 vap->va_rminor = uminor(node->tn_rdev);
338 vap->va_bytes = round_page(node->tn_size);
341 lwkt_reltoken(&vp->v_mount->mnt_token);
346 /* --------------------------------------------------------------------- */
349 tmpfs_setattr(struct vop_setattr_args *v)
351 struct vnode *vp = v->a_vp;
352 struct vattr *vap = v->a_vap;
353 struct ucred *cred = v->a_cred;
354 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
358 if (error == 0 && (vap->va_flags != VNOVAL)) {
359 error = tmpfs_chflags(vp, vap->va_flags, cred);
360 kflags |= NOTE_ATTRIB;
363 if (error == 0 && (vap->va_size != VNOVAL)) {
364 if (vap->va_size > node->tn_size)
365 kflags |= NOTE_WRITE | NOTE_EXTEND;
367 kflags |= NOTE_WRITE;
368 error = tmpfs_chsize(vp, vap->va_size, cred);
371 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
372 vap->va_gid != (gid_t)VNOVAL)) {
373 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
374 kflags |= NOTE_ATTRIB;
377 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
378 error = tmpfs_chmod(vp, vap->va_mode, cred);
379 kflags |= NOTE_ATTRIB;
382 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
383 vap->va_atime.tv_nsec != VNOVAL) ||
384 (vap->va_mtime.tv_sec != VNOVAL &&
385 vap->va_mtime.tv_nsec != VNOVAL) )) {
386 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
387 vap->va_vaflags, cred);
388 kflags |= NOTE_ATTRIB;
391 /* Update the node times. We give preference to the error codes
392 * generated by this function rather than the ones that may arise
393 * from tmpfs_update. */
395 tmpfs_knote(vp, kflags);
400 /* --------------------------------------------------------------------- */
403 * fsync is usually a NOP, but we must take action when unmounting or
407 tmpfs_fsync(struct vop_fsync_args *v)
409 struct tmpfs_node *node;
410 struct vnode *vp = v->a_vp;
412 node = VP_TO_TMPFS_NODE(vp);
415 if (vp->v_type == VREG) {
416 if (vp->v_flag & VRECLAIMED) {
417 if (node->tn_links == 0)
418 tmpfs_truncate(vp, 0);
420 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
426 /* --------------------------------------------------------------------- */
429 tmpfs_read (struct vop_read_args *ap)
432 struct vnode *vp = ap->a_vp;
433 struct uio *uio = ap->a_uio;
434 struct tmpfs_node *node;
441 if (uio->uio_resid == 0) {
445 node = VP_TO_TMPFS_NODE(vp);
447 if (uio->uio_offset < 0)
449 if (vp->v_type != VREG)
452 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
454 * Use buffer cache I/O (via tmpfs_strategy)
456 offset = (size_t)uio->uio_offset & BMASK;
457 base_offset = (off_t)uio->uio_offset - offset;
458 bp = getcacheblk(vp, base_offset, BSIZE, 0);
460 lwkt_gettoken(&vp->v_mount->mnt_token);
461 error = bread(vp, base_offset, BSIZE, &bp);
464 lwkt_reltoken(&vp->v_mount->mnt_token);
465 kprintf("tmpfs_read bread error %d\n", error);
468 lwkt_reltoken(&vp->v_mount->mnt_token);
472 * Figure out how many bytes we can actually copy this loop.
474 len = BSIZE - offset;
475 if (len > uio->uio_resid)
476 len = uio->uio_resid;
477 if (len > node->tn_size - uio->uio_offset)
478 len = (size_t)(node->tn_size - uio->uio_offset);
480 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
483 kprintf("tmpfs_read uiomove error %d\n", error);
488 TMPFS_NODE_LOCK(node);
489 node->tn_status |= TMPFS_NODE_ACCESSED;
490 TMPFS_NODE_UNLOCK(node);
496 tmpfs_write (struct vop_write_args *ap)
499 struct vnode *vp = ap->a_vp;
500 struct uio *uio = ap->a_uio;
501 struct thread *td = uio->uio_td;
502 struct tmpfs_node *node;
514 if (uio->uio_resid == 0) {
518 node = VP_TO_TMPFS_NODE(vp);
520 if (vp->v_type != VREG)
523 lwkt_gettoken(&vp->v_mount->mnt_token);
525 oldsize = node->tn_size;
526 if (ap->a_ioflag & IO_APPEND)
527 uio->uio_offset = node->tn_size;
530 * Check for illegal write offsets.
532 if (uio->uio_offset + uio->uio_resid >
533 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
534 lwkt_reltoken(&vp->v_mount->mnt_token);
539 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN).
541 if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) {
542 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
544 lwkt_reltoken(&vp->v_mount->mnt_token);
547 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
548 ksignal(td->td_proc, SIGXFSZ);
549 lwkt_reltoken(&vp->v_mount->mnt_token);
556 * Extend the file's size if necessary
558 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
560 while (uio->uio_resid > 0) {
562 * Use buffer cache I/O (via tmpfs_strategy)
564 offset = (size_t)uio->uio_offset & BMASK;
565 base_offset = (off_t)uio->uio_offset - offset;
566 len = BSIZE - offset;
567 if (len > uio->uio_resid)
568 len = uio->uio_resid;
570 if ((uio->uio_offset + len) > node->tn_size) {
571 trivial = (uio->uio_offset <= node->tn_size);
572 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
578 * Read to fill in any gaps. Theoretically we could
579 * optimize this if the write covers the entire buffer
580 * and is not a UIO_NOCOPY write, however this can lead
581 * to a security violation exposing random kernel memory
582 * (whatever junk was in the backing VM pages before).
584 * So just use bread() to do the right thing.
586 error = bread(vp, base_offset, BSIZE, &bp);
587 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
589 kprintf("tmpfs_write uiomove error %d\n", error);
594 if (uio->uio_offset > node->tn_size) {
595 node->tn_size = uio->uio_offset;
596 kflags |= NOTE_EXTEND;
598 kflags |= NOTE_WRITE;
601 * Always try to flush the page in the UIO_NOCOPY case. This
602 * can come from the pageout daemon or during vnode eviction.
603 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC.
605 * For the normal case we buwrite(), dirtying the underlying
606 * VM pages instead of dirtying the buffer and releasing the
607 * buffer as a clean buffer. This allows tmpfs to use
608 * essentially all available memory to cache file data.
609 * If we used bdwrite() the buffer cache would wind up
610 * flushing the data to swap too quickly.
612 bp->b_flags |= B_AGE;
613 if (uio->uio_segflg == UIO_NOCOPY) {
620 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
627 (void)tmpfs_reg_resize(vp, oldsize, trivial);
628 kflags &= ~NOTE_EXTEND;
634 * Currently we don't set the mtime on files modified via mmap()
635 * because we can't tell the difference between those modifications
636 * and an attempt by the pageout daemon to flush tmpfs pages to
639 * This is because in order to defer flushes as long as possible
640 * buwrite() works by marking the underlying VM pages dirty in
641 * order to be able to dispose of the buffer cache buffer without
644 TMPFS_NODE_LOCK(node);
645 if (uio->uio_segflg != UIO_NOCOPY)
646 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
648 node->tn_status |= TMPFS_NODE_CHANGED;
650 if (node->tn_mode & (S_ISUID | S_ISGID)) {
651 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
652 node->tn_mode &= ~(S_ISUID | S_ISGID);
654 TMPFS_NODE_UNLOCK(node);
657 tmpfs_knote(vp, kflags);
660 lwkt_reltoken(&vp->v_mount->mnt_token);
665 tmpfs_advlock (struct vop_advlock_args *ap)
667 struct tmpfs_node *node;
668 struct vnode *vp = ap->a_vp;
670 node = VP_TO_TMPFS_NODE(vp);
672 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
676 * The strategy function is typically only called when memory pressure
677 * forces the system to attempt to pageout pages. It can also be called
678 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write
682 tmpfs_strategy(struct vop_strategy_args *ap)
684 struct bio *bio = ap->a_bio;
686 struct buf *bp = bio->bio_buf;
687 struct vnode *vp = ap->a_vp;
688 struct tmpfs_node *node;
693 if (vp->v_type != VREG) {
694 bp->b_resid = bp->b_bcount;
695 bp->b_flags |= B_ERROR | B_INVAL;
696 bp->b_error = EINVAL;
701 lwkt_gettoken(&vp->v_mount->mnt_token);
702 node = VP_TO_TMPFS_NODE(vp);
704 uobj = node->tn_reg.tn_aobj;
707 * Don't bother flushing to swap if there is no swap, just
708 * ensure that the pages are marked as needing a commit (still).
710 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
711 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
712 m = bp->b_xio.xio_pages[i];
713 vm_page_need_commit(m);
719 nbio = push_bio(bio);
720 nbio->bio_done = tmpfs_strategy_done;
721 nbio->bio_offset = bio->bio_offset;
722 swap_pager_strategy(uobj, nbio);
725 lwkt_reltoken(&vp->v_mount->mnt_token);
730 * If we were unable to commit the pages to swap make sure they are marked
731 * as needing a commit (again). If we were, clear the flag to allow the
735 tmpfs_strategy_done(struct bio *bio)
743 if (bp->b_flags & B_ERROR) {
744 bp->b_flags &= ~B_ERROR;
747 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
748 m = bp->b_xio.xio_pages[i];
749 vm_page_need_commit(m);
752 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
753 m = bp->b_xio.xio_pages[i];
754 vm_page_clear_commit(m);
762 tmpfs_bmap(struct vop_bmap_args *ap)
764 if (ap->a_doffsetp != NULL)
765 *ap->a_doffsetp = ap->a_loffset;
766 if (ap->a_runp != NULL)
768 if (ap->a_runb != NULL)
774 /* --------------------------------------------------------------------- */
777 tmpfs_nremove(struct vop_nremove_args *v)
779 struct vnode *dvp = v->a_dvp;
780 struct namecache *ncp = v->a_nch->ncp;
783 struct tmpfs_dirent *de;
784 struct tmpfs_mount *tmp;
785 struct tmpfs_node *dnode;
786 struct tmpfs_node *node;
789 * We have to acquire the vp from v->a_nch because we will likely
790 * unresolve the namecache entry, and a vrele/vput is needed to
791 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
793 * We have to use vget to clear any inactive state on the vnode,
794 * otherwise the vnode may remain inactive and thus tmpfs_inactive
795 * will not get called when we release it.
797 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
798 KKASSERT(error == 0);
801 if (vp->v_type == VDIR) {
806 dnode = VP_TO_TMPFS_DIR(dvp);
807 node = VP_TO_TMPFS_NODE(vp);
808 tmp = VFS_TO_TMPFS(vp->v_mount);
809 de = tmpfs_dir_lookup(dnode, node, ncp);
815 /* Files marked as immutable or append-only cannot be deleted. */
816 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
817 (dnode->tn_flags & APPEND)) {
822 /* Remove the entry from the directory; as it is a file, we do not
823 * have to change the number of hard links of the directory. */
824 tmpfs_dir_detach(dnode, de);
826 /* Free the directory entry we just deleted. Note that the node
827 * referred by it will not be removed until the vnode is really
829 tmpfs_free_dirent(tmp, de);
831 if (node->tn_links > 0) {
832 TMPFS_NODE_LOCK(node);
833 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
835 TMPFS_NODE_UNLOCK(node);
838 cache_unlink(v->a_nch);
839 tmpfs_knote(vp, NOTE_DELETE);
840 tmpfs_knote(dvp, NOTE_WRITE);
849 /* --------------------------------------------------------------------- */
852 tmpfs_nlink(struct vop_nlink_args *v)
854 struct vnode *dvp = v->a_dvp;
855 struct vnode *vp = v->a_vp;
856 struct namecache *ncp = v->a_nch->ncp;
857 struct tmpfs_dirent *de;
858 struct tmpfs_node *node;
859 struct tmpfs_node *dnode;
862 KKASSERT(dvp != vp); /* XXX When can this be false? */
864 node = VP_TO_TMPFS_NODE(vp);
865 dnode = VP_TO_TMPFS_NODE(dvp);
867 /* XXX: Why aren't the following two tests done by the caller? */
869 /* Hard links of directories are forbidden. */
870 if (vp->v_type == VDIR) {
875 /* Cannot create cross-device links. */
876 if (dvp->v_mount != vp->v_mount) {
881 /* Ensure that we do not overflow the maximum number of links imposed
883 KKASSERT(node->tn_links <= LINK_MAX);
884 if (node->tn_links == LINK_MAX) {
889 /* We cannot create links of files marked immutable or append-only. */
890 if (node->tn_flags & (IMMUTABLE | APPEND)) {
895 /* Allocate a new directory entry to represent the node. */
896 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
897 ncp->nc_name, ncp->nc_nlen, &de);
901 /* Insert the new directory entry into the appropriate directory. */
902 tmpfs_dir_attach(dnode, de);
904 /* vp link count has changed, so update node times. */
906 TMPFS_NODE_LOCK(node);
907 node->tn_status |= TMPFS_NODE_CHANGED;
908 TMPFS_NODE_UNLOCK(node);
911 tmpfs_knote(vp, NOTE_LINK);
912 cache_setunresolved(v->a_nch);
913 cache_setvp(v->a_nch, vp);
914 tmpfs_knote(dvp, NOTE_WRITE);
921 /* --------------------------------------------------------------------- */
924 tmpfs_nrename(struct vop_nrename_args *v)
926 struct vnode *fdvp = v->a_fdvp;
927 struct namecache *fncp = v->a_fnch->ncp;
928 struct vnode *fvp = fncp->nc_vp;
929 struct vnode *tdvp = v->a_tdvp;
930 struct namecache *tncp = v->a_tnch->ncp;
932 struct tmpfs_dirent *de, *tde;
933 struct tmpfs_mount *tmp;
934 struct tmpfs_node *fdnode;
935 struct tmpfs_node *fnode;
936 struct tmpfs_node *tnode;
937 struct tmpfs_node *tdnode;
943 * Because tvp can get overwritten we have to vget it instead of
944 * just vref or use it, otherwise it's VINACTIVE flag may not get
945 * cleared and the node won't get destroyed.
947 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
949 tnode = VP_TO_TMPFS_NODE(tvp);
955 /* Disallow cross-device renames.
956 * XXX Why isn't this done by the caller? */
957 if (fvp->v_mount != tdvp->v_mount ||
958 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
963 tmp = VFS_TO_TMPFS(tdvp->v_mount);
964 tdnode = VP_TO_TMPFS_DIR(tdvp);
966 /* If source and target are the same file, there is nothing to do. */
972 fdnode = VP_TO_TMPFS_DIR(fdvp);
973 fnode = VP_TO_TMPFS_NODE(fvp);
974 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
976 /* Avoid manipulating '.' and '..' entries. */
981 KKASSERT(de->td_node == fnode);
984 * If replacing an entry in the target directory and that entry
985 * is a directory, it must be empty.
987 * Kern_rename gurantees the destination to be a directory
988 * if the source is one (it does?).
991 KKASSERT(tnode != NULL);
993 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
994 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
999 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1000 if (tnode->tn_size > 0) {
1004 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1007 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1011 KKASSERT(fnode->tn_type != VDIR &&
1012 tnode->tn_type != VDIR);
1016 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1017 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1023 * Ensure that we have enough memory to hold the new name, if it
1024 * has to be changed.
1026 if (fncp->nc_nlen != tncp->nc_nlen ||
1027 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1028 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1029 M_WAITOK | M_NULLOK);
1030 if (newname == NULL) {
1034 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1035 newname[tncp->nc_nlen] = '\0';
1041 * Unlink entry from source directory. Note that the kernel has
1042 * already checked for illegal recursion cases (renaming a directory
1043 * into a subdirectory of itself).
1045 if (fdnode != tdnode) {
1046 tmpfs_dir_detach(fdnode, de);
1048 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1052 * Handle any name change. Swap with newname, we will
1053 * deallocate it at the end.
1055 if (newname != NULL) {
1057 TMPFS_NODE_LOCK(fnode);
1058 fnode->tn_status |= TMPFS_NODE_CHANGED;
1059 TMPFS_NODE_UNLOCK(fnode);
1061 oldname = de->td_name;
1062 de->td_name = newname;
1063 de->td_namelen = (uint16_t)tncp->nc_nlen;
1068 * If we are overwriting an entry, we have to remove the old one
1069 * from the target directory.
1072 /* Remove the old entry from the target directory. */
1073 tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1074 tmpfs_dir_detach(tdnode, tde);
1075 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1078 * Free the directory entry we just deleted. Note that the
1079 * node referred by it will not be removed until the vnode is
1082 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1083 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1087 * Link entry to target directory. If the entry
1088 * represents a directory move the parent linkage
1091 if (fdnode != tdnode) {
1092 if (de->td_node->tn_type == VDIR) {
1093 TMPFS_VALIDATE_DIR(fnode);
1095 tmpfs_dir_attach(tdnode, de);
1097 TMPFS_NODE_LOCK(tdnode);
1098 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1099 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
1100 TMPFS_NODE_UNLOCK(tdnode);
1107 kfree(newname, tmp->tm_name_zone);
1110 cache_rename(v->a_fnch, v->a_tnch);
1111 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1112 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1113 if (fnode->tn_vnode)
1114 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1127 /* --------------------------------------------------------------------- */
1130 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1132 struct vnode *dvp = v->a_dvp;
1133 struct vnode **vpp = v->a_vpp;
1134 struct namecache *ncp = v->a_nch->ncp;
1135 struct vattr *vap = v->a_vap;
1136 struct ucred *cred = v->a_cred;
1139 KKASSERT(vap->va_type == VDIR);
1141 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1143 cache_setunresolved(v->a_nch);
1144 cache_setvp(v->a_nch, *vpp);
1145 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1151 /* --------------------------------------------------------------------- */
1154 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1156 struct vnode *dvp = v->a_dvp;
1157 struct namecache *ncp = v->a_nch->ncp;
1159 struct tmpfs_dirent *de;
1160 struct tmpfs_mount *tmp;
1161 struct tmpfs_node *dnode;
1162 struct tmpfs_node *node;
1166 * We have to acquire the vp from v->a_nch because we will likely
1167 * unresolve the namecache entry, and a vrele/vput is needed to
1168 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1170 * We have to use vget to clear any inactive state on the vnode,
1171 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1172 * will not get called when we release it.
1174 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1175 KKASSERT(error == 0);
1179 * Prevalidate so we don't hit an assertion later
1181 if (vp->v_type != VDIR) {
1186 tmp = VFS_TO_TMPFS(dvp->v_mount);
1187 dnode = VP_TO_TMPFS_DIR(dvp);
1188 node = VP_TO_TMPFS_DIR(vp);
1190 /* Directories with more than two entries ('.' and '..') cannot be
1192 if (node->tn_size > 0) {
1197 if ((dnode->tn_flags & APPEND)
1198 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1203 /* This invariant holds only if we are not trying to remove "..".
1204 * We checked for that above so this is safe now. */
1205 KKASSERT(node->tn_dir.tn_parent == dnode);
1207 /* Get the directory entry associated with node (vp). This was
1208 * filled by tmpfs_lookup while looking up the entry. */
1209 de = tmpfs_dir_lookup(dnode, node, ncp);
1210 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1214 /* Check flags to see if we are allowed to remove the directory. */
1215 if ((dnode->tn_flags & APPEND) ||
1216 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1222 /* Detach the directory entry from the directory (dnode). */
1223 tmpfs_dir_detach(dnode, de);
1225 /* No vnode should be allocated for this entry from this point */
1226 TMPFS_NODE_LOCK(node);
1227 TMPFS_ASSERT_ELOCKED(node);
1228 TMPFS_NODE_LOCK(dnode);
1229 TMPFS_ASSERT_ELOCKED(dnode);
1232 * Must set parent linkage to NULL (tested by ncreate to disallow
1233 * the creation of new files/dirs in a deleted directory)
1235 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1236 TMPFS_NODE_MODIFIED;
1238 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1239 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1241 TMPFS_NODE_UNLOCK(dnode);
1242 TMPFS_NODE_UNLOCK(node);
1244 /* Free the directory entry we just deleted. Note that the node
1245 * referred by it will not be removed until the vnode is really
1247 tmpfs_free_dirent(tmp, de);
1249 /* Release the deleted vnode (will destroy the node, notify
1250 * interested parties and clean it from the cache). */
1252 TMPFS_NODE_LOCK(dnode);
1253 dnode->tn_status |= TMPFS_NODE_CHANGED;
1254 TMPFS_NODE_UNLOCK(dnode);
1257 cache_unlink(v->a_nch);
1258 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1267 /* --------------------------------------------------------------------- */
1270 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1272 struct vnode *dvp = v->a_dvp;
1273 struct vnode **vpp = v->a_vpp;
1274 struct namecache *ncp = v->a_nch->ncp;
1275 struct vattr *vap = v->a_vap;
1276 struct ucred *cred = v->a_cred;
1277 char *target = v->a_target;
1280 vap->va_type = VLNK;
1281 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1283 tmpfs_knote(*vpp, NOTE_WRITE);
1284 cache_setunresolved(v->a_nch);
1285 cache_setvp(v->a_nch, *vpp);
1291 /* --------------------------------------------------------------------- */
1294 tmpfs_readdir(struct vop_readdir_args *v)
1296 struct vnode *vp = v->a_vp;
1297 struct uio *uio = v->a_uio;
1298 int *eofflag = v->a_eofflag;
1299 off_t **cookies = v->a_cookies;
1300 int *ncookies = v->a_ncookies;
1301 struct tmpfs_mount *tmp;
1305 struct tmpfs_node *node;
1307 /* This operation only makes sense on directory nodes. */
1308 if (vp->v_type != VDIR)
1311 tmp = VFS_TO_TMPFS(vp->v_mount);
1312 node = VP_TO_TMPFS_DIR(vp);
1313 startoff = uio->uio_offset;
1315 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1316 error = tmpfs_dir_getdotdent(node, uio);
1322 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1323 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1329 error = tmpfs_dir_getdents(node, uio, &cnt);
1332 KKASSERT(error >= -1);
1337 if (eofflag != NULL)
1339 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1341 /* Update NFS-related variables. */
1342 if (error == 0 && cookies != NULL && ncookies != NULL) {
1344 off_t off = startoff;
1345 struct tmpfs_dirent *de = NULL;
1348 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1350 for (i = 0; i < cnt; i++) {
1351 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1352 if (off == TMPFS_DIRCOOKIE_DOT) {
1353 off = TMPFS_DIRCOOKIE_DOTDOT;
1355 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1356 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
1357 } else if (de != NULL) {
1358 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1360 de = tmpfs_dir_lookupbycookie(node,
1362 KKASSERT(de != NULL);
1363 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
1366 off = TMPFS_DIRCOOKIE_EOF;
1368 off = tmpfs_dircookie(de);
1371 (*cookies)[i] = off;
1373 KKASSERT(uio->uio_offset == off);
1379 /* --------------------------------------------------------------------- */
1382 tmpfs_readlink(struct vop_readlink_args *v)
1384 struct vnode *vp = v->a_vp;
1385 struct uio *uio = v->a_uio;
1388 struct tmpfs_node *node;
1390 KKASSERT(uio->uio_offset == 0);
1391 KKASSERT(vp->v_type == VLNK);
1393 node = VP_TO_TMPFS_NODE(vp);
1395 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1397 TMPFS_NODE_LOCK(node);
1398 node->tn_status |= TMPFS_NODE_ACCESSED;
1399 TMPFS_NODE_UNLOCK(node);
1404 /* --------------------------------------------------------------------- */
1407 tmpfs_inactive(struct vop_inactive_args *v)
1409 struct vnode *vp = v->a_vp;
1410 struct tmpfs_node *node;
1412 node = VP_TO_TMPFS_NODE(vp);
1423 * Get rid of unreferenced deleted vnodes sooner rather than
1424 * later so the data memory can be recovered immediately.
1426 * We must truncate the vnode to prevent the normal reclamation
1427 * path from flushing the data for the removed file to disk.
1429 TMPFS_NODE_LOCK(node);
1430 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1431 node->tn_links == 0)
1433 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1434 TMPFS_NODE_UNLOCK(node);
1435 if (node->tn_type == VREG)
1436 tmpfs_truncate(vp, 0);
1439 TMPFS_NODE_UNLOCK(node);
1445 /* --------------------------------------------------------------------- */
1446 #include <vm/vm_object.h>
1447 #include <vm/vm_page.h>
1450 tmpfs_reclaim(struct vop_reclaim_args *v)
1452 struct vnode *vp = v->a_vp;
1453 struct tmpfs_mount *tmp;
1454 struct tmpfs_node *node;
1456 node = VP_TO_TMPFS_NODE(vp);
1457 tmp = VFS_TO_TMPFS(vp->v_mount);
1462 * If the node referenced by this vnode was deleted by the
1463 * user, we must free its associated data structures now that
1464 * the vnode is being reclaimed.
1466 * Directories have an extra link ref.
1468 TMPFS_NODE_LOCK(node);
1469 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1470 node->tn_links == 0) {
1471 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1472 tmpfs_free_node(tmp, node);
1475 TMPFS_NODE_UNLOCK(node);
1478 KKASSERT(vp->v_data == NULL);
1482 /* --------------------------------------------------------------------- */
1485 tmpfs_mountctl(struct vop_mountctl_args *ap)
1487 struct tmpfs_mount *tmp;
1492 case (MOUNTCTL_SET_EXPORT):
1493 mp = ap->a_head.a_ops->head.vv_mount;
1494 tmp = (struct tmpfs_mount *) mp->mnt_data;
1496 if (ap->a_ctllen != sizeof(struct export_args))
1499 rc = vfs_export(mp, &tmp->tm_export,
1500 (const struct export_args *) ap->a_ctl);
1503 rc = vop_stdmountctl(ap);
1509 /* --------------------------------------------------------------------- */
1512 tmpfs_print(struct vop_print_args *v)
1514 struct vnode *vp = v->a_vp;
1516 struct tmpfs_node *node;
1518 node = VP_TO_TMPFS_NODE(vp);
1520 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1521 node, node->tn_flags, node->tn_links);
1522 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1523 node->tn_mode, node->tn_uid, node->tn_gid,
1524 (uintmax_t)node->tn_size, node->tn_status);
1526 if (vp->v_type == VFIFO)
1534 /* --------------------------------------------------------------------- */
1537 tmpfs_pathconf(struct vop_pathconf_args *v)
1539 int name = v->a_name;
1540 register_t *retval = v->a_retval;
1563 case _PC_CHOWN_RESTRICTED:
1575 case _PC_FILESIZEBITS:
1576 *retval = 0; /* XXX Don't know which value should I return. */
1586 /************************************************************************
1588 ************************************************************************/
1590 static void filt_tmpfsdetach(struct knote *kn);
1591 static int filt_tmpfsread(struct knote *kn, long hint);
1592 static int filt_tmpfswrite(struct knote *kn, long hint);
1593 static int filt_tmpfsvnode(struct knote *kn, long hint);
1595 static struct filterops tmpfsread_filtops =
1596 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1597 static struct filterops tmpfswrite_filtops =
1598 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1599 static struct filterops tmpfsvnode_filtops =
1600 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1603 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1605 struct vnode *vp = ap->a_vp;
1606 struct knote *kn = ap->a_kn;
1608 switch (kn->kn_filter) {
1610 kn->kn_fop = &tmpfsread_filtops;
1613 kn->kn_fop = &tmpfswrite_filtops;
1616 kn->kn_fop = &tmpfsvnode_filtops;
1619 return (EOPNOTSUPP);
1622 kn->kn_hook = (caddr_t)vp;
1624 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1630 filt_tmpfsdetach(struct knote *kn)
1632 struct vnode *vp = (void *)kn->kn_hook;
1634 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1638 filt_tmpfsread(struct knote *kn, long hint)
1640 struct vnode *vp = (void *)kn->kn_hook;
1641 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1644 if (hint == NOTE_REVOKE) {
1645 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1650 * Interlock against MP races when performing this function.
1652 lwkt_gettoken(&vp->v_mount->mnt_token);
1653 off = node->tn_size - kn->kn_fp->f_offset;
1654 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1655 if (kn->kn_sfflags & NOTE_OLDAPI) {
1656 lwkt_reltoken(&vp->v_mount->mnt_token);
1660 if (kn->kn_data == 0) {
1661 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1663 lwkt_reltoken(&vp->v_mount->mnt_token);
1664 return (kn->kn_data != 0);
1668 filt_tmpfswrite(struct knote *kn, long hint)
1670 if (hint == NOTE_REVOKE)
1671 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1677 filt_tmpfsvnode(struct knote *kn, long hint)
1679 if (kn->kn_sfflags & hint)
1680 kn->kn_fflags |= hint;
1681 if (hint == NOTE_REVOKE) {
1682 kn->kn_flags |= (EV_EOF | EV_NODATA);
1685 return (kn->kn_fflags != 0);
1689 /* --------------------------------------------------------------------- */
1692 * vnode operations vector used for files stored in a tmpfs file system.
1694 struct vop_ops tmpfs_vnode_vops = {
1695 .vop_default = vop_defaultop,
1696 .vop_getpages = vop_stdgetpages,
1697 .vop_putpages = vop_stdputpages,
1698 .vop_ncreate = tmpfs_ncreate,
1699 .vop_nresolve = tmpfs_nresolve,
1700 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1701 .vop_nmknod = tmpfs_nmknod,
1702 .vop_open = tmpfs_open,
1703 .vop_close = tmpfs_close,
1704 .vop_access = tmpfs_access,
1705 .vop_getattr = tmpfs_getattr,
1706 .vop_setattr = tmpfs_setattr,
1707 .vop_read = tmpfs_read,
1708 .vop_write = tmpfs_write,
1709 .vop_fsync = tmpfs_fsync,
1710 .vop_mountctl = tmpfs_mountctl,
1711 .vop_nremove = tmpfs_nremove,
1712 .vop_nlink = tmpfs_nlink,
1713 .vop_nrename = tmpfs_nrename,
1714 .vop_nmkdir = tmpfs_nmkdir,
1715 .vop_nrmdir = tmpfs_nrmdir,
1716 .vop_nsymlink = tmpfs_nsymlink,
1717 .vop_readdir = tmpfs_readdir,
1718 .vop_readlink = tmpfs_readlink,
1719 .vop_inactive = tmpfs_inactive,
1720 .vop_reclaim = tmpfs_reclaim,
1721 .vop_print = tmpfs_print,
1722 .vop_pathconf = tmpfs_pathconf,
1723 .vop_bmap = tmpfs_bmap,
1724 .vop_strategy = tmpfs_strategy,
1725 .vop_advlock = tmpfs_advlock,
1726 .vop_kqfilter = tmpfs_kqfilter