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_mount *tmp;
406 struct tmpfs_node *node;
407 struct vnode *vp = v->a_vp;
409 tmp = VFS_TO_TMPFS(vp->v_mount);
410 node = VP_TO_TMPFS_NODE(vp);
413 if (vp->v_type == VREG) {
414 if (vp->v_flag & VRECLAIMED) {
415 if (node->tn_links == 0)
416 tmpfs_truncate(vp, 0);
418 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
424 /* --------------------------------------------------------------------- */
427 tmpfs_read (struct vop_read_args *ap)
430 struct vnode *vp = ap->a_vp;
431 struct uio *uio = ap->a_uio;
432 struct tmpfs_node *node;
439 if (uio->uio_resid == 0) {
443 node = VP_TO_TMPFS_NODE(vp);
445 if (uio->uio_offset < 0)
447 if (vp->v_type != VREG)
450 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
452 * Use buffer cache I/O (via tmpfs_strategy)
454 offset = (size_t)uio->uio_offset & BMASK;
455 base_offset = (off_t)uio->uio_offset - offset;
456 bp = getcacheblk(vp, base_offset, BSIZE);
459 lwkt_gettoken(&vp->v_mount->mnt_token);
460 error = bread(vp, base_offset, BSIZE, &bp);
463 lwkt_reltoken(&vp->v_mount->mnt_token);
464 kprintf("tmpfs_read bread error %d\n", error);
467 lwkt_reltoken(&vp->v_mount->mnt_token);
471 * Figure out how many bytes we can actually copy this loop.
473 len = BSIZE - offset;
474 if (len > uio->uio_resid)
475 len = uio->uio_resid;
476 if (len > node->tn_size - uio->uio_offset)
477 len = (size_t)(node->tn_size - uio->uio_offset);
479 error = uiomove((char *)bp->b_data + offset, len, uio);
482 kprintf("tmpfs_read uiomove error %d\n", error);
487 TMPFS_NODE_LOCK(node);
488 node->tn_status |= TMPFS_NODE_ACCESSED;
489 TMPFS_NODE_UNLOCK(node);
495 tmpfs_write (struct vop_write_args *ap)
498 struct vnode *vp = ap->a_vp;
499 struct uio *uio = ap->a_uio;
500 struct thread *td = uio->uio_td;
501 struct tmpfs_node *node;
513 if (uio->uio_resid == 0) {
517 node = VP_TO_TMPFS_NODE(vp);
519 if (vp->v_type != VREG)
522 lwkt_gettoken(&vp->v_mount->mnt_token);
524 oldsize = node->tn_size;
525 if (ap->a_ioflag & IO_APPEND)
526 uio->uio_offset = node->tn_size;
529 * Check for illegal write offsets.
531 if (uio->uio_offset + uio->uio_resid >
532 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
533 lwkt_reltoken(&vp->v_mount->mnt_token);
537 if (vp->v_type == VREG && td != NULL) {
538 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
540 lwkt_reltoken(&vp->v_mount->mnt_token);
543 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
544 ksignal(td->td_proc, SIGXFSZ);
545 lwkt_reltoken(&vp->v_mount->mnt_token);
552 * Extend the file's size if necessary
554 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
556 while (uio->uio_resid > 0) {
558 * Use buffer cache I/O (via tmpfs_strategy)
560 offset = (size_t)uio->uio_offset & BMASK;
561 base_offset = (off_t)uio->uio_offset - offset;
562 len = BSIZE - offset;
563 if (len > uio->uio_resid)
564 len = uio->uio_resid;
566 if ((uio->uio_offset + len) > node->tn_size) {
567 trivial = (uio->uio_offset <= node->tn_size);
568 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
574 * Read to fill in any gaps. Theoretically we could
575 * optimize this if the write covers the entire buffer
576 * and is not a UIO_NOCOPY write, however this can lead
577 * to a security violation exposing random kernel memory
578 * (whatever junk was in the backing VM pages before).
580 * So just use bread() to do the right thing.
582 error = bread(vp, base_offset, BSIZE, &bp);
583 error = uiomove((char *)bp->b_data + offset, len, uio);
585 kprintf("tmpfs_write uiomove error %d\n", error);
590 if (uio->uio_offset > node->tn_size) {
591 node->tn_size = uio->uio_offset;
592 kflags |= NOTE_EXTEND;
594 kflags |= NOTE_WRITE;
597 * Always try to flush the page if the request is coming
598 * from the pageout daemon (IO_ASYNC), else buwrite() the
601 * buwrite() dirties the underlying VM pages instead of
602 * dirtying the buffer, releasing the buffer as a clean
603 * buffer. This allows tmpfs to use essentially all
604 * available memory to cache file data. If we used bdwrite()
605 * the buffer cache would wind up flushing the data to
608 bp->b_flags |= B_AGE;
609 if (ap->a_ioflag & IO_ASYNC) {
616 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
623 (void)tmpfs_reg_resize(vp, oldsize, trivial);
624 kflags &= ~NOTE_EXTEND;
630 * Currently we don't set the mtime on files modified via mmap()
631 * because we can't tell the difference between those modifications
632 * and an attempt by the pageout daemon to flush tmpfs pages to
635 * This is because in order to defer flushes as long as possible
636 * buwrite() works by marking the underlying VM pages dirty in
637 * order to be able to dispose of the buffer cache buffer without
640 TMPFS_NODE_LOCK(node);
641 if (uio->uio_segflg != UIO_NOCOPY)
642 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
644 node->tn_status |= TMPFS_NODE_CHANGED;
646 if (node->tn_mode & (S_ISUID | S_ISGID)) {
647 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
648 node->tn_mode &= ~(S_ISUID | S_ISGID);
650 TMPFS_NODE_UNLOCK(node);
653 tmpfs_knote(vp, kflags);
656 lwkt_reltoken(&vp->v_mount->mnt_token);
661 tmpfs_advlock (struct vop_advlock_args *ap)
663 struct tmpfs_node *node;
664 struct vnode *vp = ap->a_vp;
666 node = VP_TO_TMPFS_NODE(vp);
668 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
672 * The strategy function is typically only called when memory pressure
673 * forces the system to attempt to pageout pages. It can also be called
674 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write
678 tmpfs_strategy(struct vop_strategy_args *ap)
680 struct bio *bio = ap->a_bio;
682 struct buf *bp = bio->bio_buf;
683 struct vnode *vp = ap->a_vp;
684 struct tmpfs_node *node;
689 if (vp->v_type != VREG) {
690 bp->b_resid = bp->b_bcount;
691 bp->b_flags |= B_ERROR | B_INVAL;
692 bp->b_error = EINVAL;
697 lwkt_gettoken(&vp->v_mount->mnt_token);
698 node = VP_TO_TMPFS_NODE(vp);
700 uobj = node->tn_reg.tn_aobj;
703 * Don't bother flushing to swap if there is no swap, just
704 * ensure that the pages are marked as needing a commit (still).
706 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
707 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
708 m = bp->b_xio.xio_pages[i];
709 vm_page_need_commit(m);
715 nbio = push_bio(bio);
716 nbio->bio_done = tmpfs_strategy_done;
717 nbio->bio_offset = bio->bio_offset;
718 swap_pager_strategy(uobj, nbio);
721 lwkt_reltoken(&vp->v_mount->mnt_token);
726 * If we were unable to commit the pages to swap make sure they are marked
727 * as needing a commit (again). If we were, clear the flag to allow the
731 tmpfs_strategy_done(struct bio *bio)
739 if (bp->b_flags & B_ERROR) {
740 bp->b_flags &= ~B_ERROR;
743 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
744 m = bp->b_xio.xio_pages[i];
745 vm_page_need_commit(m);
748 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
749 m = bp->b_xio.xio_pages[i];
750 vm_page_clear_commit(m);
758 tmpfs_bmap(struct vop_bmap_args *ap)
760 if (ap->a_doffsetp != NULL)
761 *ap->a_doffsetp = ap->a_loffset;
762 if (ap->a_runp != NULL)
764 if (ap->a_runb != NULL)
770 /* --------------------------------------------------------------------- */
773 tmpfs_nremove(struct vop_nremove_args *v)
775 struct vnode *dvp = v->a_dvp;
776 struct namecache *ncp = v->a_nch->ncp;
779 struct tmpfs_dirent *de;
780 struct tmpfs_mount *tmp;
781 struct tmpfs_node *dnode;
782 struct tmpfs_node *node;
785 * We have to acquire the vp from v->a_nch because we will likely
786 * unresolve the namecache entry, and a vrele/vput is needed to
787 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
789 * We have to use vget to clear any inactive state on the vnode,
790 * otherwise the vnode may remain inactive and thus tmpfs_inactive
791 * will not get called when we release it.
793 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
794 KKASSERT(error == 0);
797 if (vp->v_type == VDIR) {
802 dnode = VP_TO_TMPFS_DIR(dvp);
803 node = VP_TO_TMPFS_NODE(vp);
804 tmp = VFS_TO_TMPFS(vp->v_mount);
805 de = tmpfs_dir_lookup(dnode, node, ncp);
811 /* Files marked as immutable or append-only cannot be deleted. */
812 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
813 (dnode->tn_flags & APPEND)) {
818 /* Remove the entry from the directory; as it is a file, we do not
819 * have to change the number of hard links of the directory. */
820 tmpfs_dir_detach(dnode, de);
822 /* Free the directory entry we just deleted. Note that the node
823 * referred by it will not be removed until the vnode is really
825 tmpfs_free_dirent(tmp, de);
827 if (node->tn_links > 0) {
828 TMPFS_NODE_LOCK(node);
829 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
831 TMPFS_NODE_UNLOCK(node);
834 cache_setunresolved(v->a_nch);
835 cache_setvp(v->a_nch, NULL);
836 tmpfs_knote(vp, NOTE_DELETE);
837 /*cache_inval_vp(vp, CINV_DESTROY);*/
838 tmpfs_knote(dvp, NOTE_WRITE);
847 /* --------------------------------------------------------------------- */
850 tmpfs_nlink(struct vop_nlink_args *v)
852 struct vnode *dvp = v->a_dvp;
853 struct vnode *vp = v->a_vp;
854 struct namecache *ncp = v->a_nch->ncp;
855 struct tmpfs_dirent *de;
856 struct tmpfs_node *node;
857 struct tmpfs_node *dnode;
860 KKASSERT(dvp != vp); /* XXX When can this be false? */
862 node = VP_TO_TMPFS_NODE(vp);
863 dnode = VP_TO_TMPFS_NODE(dvp);
865 /* XXX: Why aren't the following two tests done by the caller? */
867 /* Hard links of directories are forbidden. */
868 if (vp->v_type == VDIR) {
873 /* Cannot create cross-device links. */
874 if (dvp->v_mount != vp->v_mount) {
879 /* Ensure that we do not overflow the maximum number of links imposed
881 KKASSERT(node->tn_links <= LINK_MAX);
882 if (node->tn_links == LINK_MAX) {
887 /* We cannot create links of files marked immutable or append-only. */
888 if (node->tn_flags & (IMMUTABLE | APPEND)) {
893 /* Allocate a new directory entry to represent the node. */
894 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
895 ncp->nc_name, ncp->nc_nlen, &de);
899 /* Insert the new directory entry into the appropriate directory. */
900 tmpfs_dir_attach(dnode, de);
902 /* vp link count has changed, so update node times. */
904 TMPFS_NODE_LOCK(node);
905 node->tn_status |= TMPFS_NODE_CHANGED;
906 TMPFS_NODE_UNLOCK(node);
909 tmpfs_knote(vp, NOTE_LINK);
910 cache_setunresolved(v->a_nch);
911 cache_setvp(v->a_nch, vp);
912 tmpfs_knote(dvp, NOTE_WRITE);
919 /* --------------------------------------------------------------------- */
922 tmpfs_nrename(struct vop_nrename_args *v)
924 struct vnode *fdvp = v->a_fdvp;
925 struct namecache *fncp = v->a_fnch->ncp;
926 struct vnode *fvp = fncp->nc_vp;
927 struct vnode *tdvp = v->a_tdvp;
928 struct namecache *tncp = v->a_tnch->ncp;
930 struct tmpfs_dirent *de;
931 struct tmpfs_mount *tmp;
932 struct tmpfs_node *fdnode;
933 struct tmpfs_node *fnode;
934 struct tmpfs_node *tnode;
935 struct tmpfs_node *tdnode;
941 * Because tvp can get overwritten we have to vget it instead of
942 * just vref or use it, otherwise it's VINACTIVE flag may not get
943 * cleared and the node won't get destroyed.
945 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
947 tnode = VP_TO_TMPFS_NODE(tvp);
953 /* Disallow cross-device renames.
954 * XXX Why isn't this done by the caller? */
955 if (fvp->v_mount != tdvp->v_mount ||
956 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
961 tmp = VFS_TO_TMPFS(tdvp->v_mount);
962 tdnode = VP_TO_TMPFS_DIR(tdvp);
964 /* If source and target are the same file, there is nothing to do. */
970 fdnode = VP_TO_TMPFS_DIR(fdvp);
971 fnode = VP_TO_TMPFS_NODE(fvp);
972 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
974 /* Avoid manipulating '.' and '..' entries. */
979 KKASSERT(de->td_node == fnode);
982 * If replacing an entry in the target directory and that entry
983 * is a directory, it must be empty.
985 * Kern_rename gurantees the destination to be a directory
986 * if the source is one (it does?).
989 KKASSERT(tnode != NULL);
991 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
992 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
997 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
998 if (tnode->tn_size > 0) {
1002 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1005 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1009 KKASSERT(fnode->tn_type != VDIR &&
1010 tnode->tn_type != VDIR);
1014 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1015 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1021 * Ensure that we have enough memory to hold the new name, if it
1022 * has to be changed.
1024 if (fncp->nc_nlen != tncp->nc_nlen ||
1025 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1026 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1027 M_WAITOK | M_NULLOK);
1028 if (newname == NULL) {
1032 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1033 newname[tncp->nc_nlen] = '\0';
1039 * Unlink entry from source directory. Note that the kernel has
1040 * already checked for illegal recursion cases (renaming a directory
1041 * into a subdirectory of itself).
1043 if (fdnode != tdnode)
1044 tmpfs_dir_detach(fdnode, 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 * Link entry to target directory. If the entry
1064 * represents a directory move the parent linkage
1067 if (fdnode != tdnode) {
1068 if (de->td_node->tn_type == VDIR) {
1069 TMPFS_VALIDATE_DIR(fnode);
1071 TMPFS_NODE_LOCK(tdnode);
1073 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1074 TMPFS_NODE_UNLOCK(tdnode);
1076 TMPFS_NODE_LOCK(fnode);
1077 fnode->tn_dir.tn_parent = tdnode;
1078 fnode->tn_status |= TMPFS_NODE_CHANGED;
1079 TMPFS_NODE_UNLOCK(fnode);
1081 TMPFS_NODE_LOCK(fdnode);
1083 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1084 TMPFS_NODE_UNLOCK(fdnode);
1086 tmpfs_dir_attach(tdnode, de);
1088 TMPFS_NODE_LOCK(tdnode);
1089 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1090 TMPFS_NODE_UNLOCK(tdnode);
1094 * If we are overwriting an entry, we have to remove the old one
1095 * from the target directory.
1098 /* Remove the old entry from the target directory. */
1099 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1100 tmpfs_dir_detach(tdnode, de);
1101 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1104 * Free the directory entry we just deleted. Note that the
1105 * node referred by it will not be removed until the vnode is
1108 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1109 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1116 kfree(newname, tmp->tm_name_zone);
1119 cache_rename(v->a_fnch, v->a_tnch);
1120 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1121 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1122 if (fnode->tn_vnode)
1123 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1136 /* --------------------------------------------------------------------- */
1139 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1141 struct vnode *dvp = v->a_dvp;
1142 struct vnode **vpp = v->a_vpp;
1143 struct namecache *ncp = v->a_nch->ncp;
1144 struct vattr *vap = v->a_vap;
1145 struct ucred *cred = v->a_cred;
1148 KKASSERT(vap->va_type == VDIR);
1150 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1152 cache_setunresolved(v->a_nch);
1153 cache_setvp(v->a_nch, *vpp);
1154 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1160 /* --------------------------------------------------------------------- */
1163 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1165 struct vnode *dvp = v->a_dvp;
1166 struct namecache *ncp = v->a_nch->ncp;
1168 struct tmpfs_dirent *de;
1169 struct tmpfs_mount *tmp;
1170 struct tmpfs_node *dnode;
1171 struct tmpfs_node *node;
1175 * We have to acquire the vp from v->a_nch because we will likely
1176 * unresolve the namecache entry, and a vrele/vput is needed to
1177 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1179 * We have to use vget to clear any inactive state on the vnode,
1180 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1181 * will not get called when we release it.
1183 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1184 KKASSERT(error == 0);
1188 * Prevalidate so we don't hit an assertion later
1190 if (vp->v_type != VDIR) {
1195 tmp = VFS_TO_TMPFS(dvp->v_mount);
1196 dnode = VP_TO_TMPFS_DIR(dvp);
1197 node = VP_TO_TMPFS_DIR(vp);
1199 /* Directories with more than two entries ('.' and '..') cannot be
1201 if (node->tn_size > 0) {
1206 if ((dnode->tn_flags & APPEND)
1207 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1212 /* This invariant holds only if we are not trying to remove "..".
1213 * We checked for that above so this is safe now. */
1214 KKASSERT(node->tn_dir.tn_parent == dnode);
1216 /* Get the directory entry associated with node (vp). This was
1217 * filled by tmpfs_lookup while looking up the entry. */
1218 de = tmpfs_dir_lookup(dnode, node, ncp);
1219 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1223 /* Check flags to see if we are allowed to remove the directory. */
1224 if ((dnode->tn_flags & APPEND) ||
1225 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1231 /* Detach the directory entry from the directory (dnode). */
1232 tmpfs_dir_detach(dnode, de);
1234 /* No vnode should be allocated for this entry from this point */
1235 TMPFS_NODE_LOCK(node);
1236 TMPFS_ASSERT_ELOCKED(node);
1237 TMPFS_NODE_LOCK(dnode);
1238 TMPFS_ASSERT_ELOCKED(dnode);
1241 /* handled by tmpfs_free_node */
1242 KKASSERT(node->tn_links > 0);
1244 node->tn_dir.tn_parent = NULL;
1246 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1247 TMPFS_NODE_MODIFIED;
1250 /* handled by tmpfs_free_node */
1251 KKASSERT(dnode->tn_links > 0);
1254 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1255 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1257 TMPFS_NODE_UNLOCK(dnode);
1258 TMPFS_NODE_UNLOCK(node);
1260 /* Free the directory entry we just deleted. Note that the node
1261 * referred by it will not be removed until the vnode is really
1263 tmpfs_free_dirent(tmp, de);
1265 /* Release the deleted vnode (will destroy the node, notify
1266 * interested parties and clean it from the cache). */
1268 TMPFS_NODE_LOCK(dnode);
1269 dnode->tn_status |= TMPFS_NODE_CHANGED;
1270 TMPFS_NODE_UNLOCK(dnode);
1273 cache_setunresolved(v->a_nch);
1274 cache_setvp(v->a_nch, NULL);
1275 /*cache_inval_vp(vp, CINV_DESTROY);*/
1276 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1285 /* --------------------------------------------------------------------- */
1288 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1290 struct vnode *dvp = v->a_dvp;
1291 struct vnode **vpp = v->a_vpp;
1292 struct namecache *ncp = v->a_nch->ncp;
1293 struct vattr *vap = v->a_vap;
1294 struct ucred *cred = v->a_cred;
1295 char *target = v->a_target;
1298 vap->va_type = VLNK;
1299 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1301 tmpfs_knote(*vpp, NOTE_WRITE);
1302 cache_setunresolved(v->a_nch);
1303 cache_setvp(v->a_nch, *vpp);
1309 /* --------------------------------------------------------------------- */
1312 tmpfs_readdir(struct vop_readdir_args *v)
1314 struct vnode *vp = v->a_vp;
1315 struct uio *uio = v->a_uio;
1316 int *eofflag = v->a_eofflag;
1317 off_t **cookies = v->a_cookies;
1318 int *ncookies = v->a_ncookies;
1319 struct tmpfs_mount *tmp;
1323 struct tmpfs_node *node;
1325 /* This operation only makes sense on directory nodes. */
1326 if (vp->v_type != VDIR)
1329 tmp = VFS_TO_TMPFS(vp->v_mount);
1330 node = VP_TO_TMPFS_DIR(vp);
1331 startoff = uio->uio_offset;
1333 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1334 error = tmpfs_dir_getdotdent(node, uio);
1340 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1341 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1347 error = tmpfs_dir_getdents(node, uio, &cnt);
1350 KKASSERT(error >= -1);
1355 if (eofflag != NULL)
1357 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1359 /* Update NFS-related variables. */
1360 if (error == 0 && cookies != NULL && ncookies != NULL) {
1362 off_t off = startoff;
1363 struct tmpfs_dirent *de = NULL;
1366 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1368 for (i = 0; i < cnt; i++) {
1369 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1370 if (off == TMPFS_DIRCOOKIE_DOT) {
1371 off = TMPFS_DIRCOOKIE_DOTDOT;
1373 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1374 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1375 } else if (de != NULL) {
1376 de = TAILQ_NEXT(de, td_entries);
1378 de = tmpfs_dir_lookupbycookie(node,
1380 KKASSERT(de != NULL);
1381 de = TAILQ_NEXT(de, td_entries);
1384 off = TMPFS_DIRCOOKIE_EOF;
1386 off = tmpfs_dircookie(de);
1389 (*cookies)[i] = off;
1391 KKASSERT(uio->uio_offset == off);
1397 /* --------------------------------------------------------------------- */
1400 tmpfs_readlink(struct vop_readlink_args *v)
1402 struct vnode *vp = v->a_vp;
1403 struct uio *uio = v->a_uio;
1406 struct tmpfs_node *node;
1408 KKASSERT(uio->uio_offset == 0);
1409 KKASSERT(vp->v_type == VLNK);
1411 node = VP_TO_TMPFS_NODE(vp);
1413 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1415 TMPFS_NODE_LOCK(node);
1416 node->tn_status |= TMPFS_NODE_ACCESSED;
1417 TMPFS_NODE_UNLOCK(node);
1422 /* --------------------------------------------------------------------- */
1425 tmpfs_inactive(struct vop_inactive_args *v)
1427 struct vnode *vp = v->a_vp;
1428 struct tmpfs_node *node;
1430 node = VP_TO_TMPFS_NODE(vp);
1441 * Get rid of unreferenced deleted vnodes sooner rather than
1442 * later so the data memory can be recovered immediately.
1444 * We must truncate the vnode to prevent the normal reclamation
1445 * path from flushing the data for the removed file to disk.
1447 TMPFS_NODE_LOCK(node);
1448 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1449 (node->tn_links == 0 ||
1450 (node->tn_links == 1 && node->tn_type == VDIR &&
1451 node->tn_dir.tn_parent)))
1453 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1454 TMPFS_NODE_UNLOCK(node);
1455 if (node->tn_type == VREG)
1456 tmpfs_truncate(vp, 0);
1459 TMPFS_NODE_UNLOCK(node);
1465 /* --------------------------------------------------------------------- */
1468 tmpfs_reclaim(struct vop_reclaim_args *v)
1470 struct vnode *vp = v->a_vp;
1471 struct tmpfs_mount *tmp;
1472 struct tmpfs_node *node;
1474 node = VP_TO_TMPFS_NODE(vp);
1475 tmp = VFS_TO_TMPFS(vp->v_mount);
1480 * If the node referenced by this vnode was deleted by the
1481 * user, we must free its associated data structures now that
1482 * the vnode is being reclaimed.
1484 * Directories have an extra link ref.
1486 TMPFS_NODE_LOCK(node);
1487 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1488 (node->tn_links == 0 ||
1489 (node->tn_links == 1 && node->tn_type == VDIR &&
1490 node->tn_dir.tn_parent)))
1492 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1493 tmpfs_free_node(tmp, node);
1496 TMPFS_NODE_UNLOCK(node);
1499 KKASSERT(vp->v_data == NULL);
1503 /* --------------------------------------------------------------------- */
1506 tmpfs_mountctl(struct vop_mountctl_args *ap)
1508 struct tmpfs_mount *tmp;
1513 case (MOUNTCTL_SET_EXPORT):
1514 mp = ap->a_head.a_ops->head.vv_mount;
1515 tmp = (struct tmpfs_mount *) mp->mnt_data;
1517 if (ap->a_ctllen != sizeof(struct export_args))
1520 rc = vfs_export(mp, &tmp->tm_export,
1521 (const struct export_args *) ap->a_ctl);
1524 rc = vop_stdmountctl(ap);
1530 /* --------------------------------------------------------------------- */
1533 tmpfs_print(struct vop_print_args *v)
1535 struct vnode *vp = v->a_vp;
1537 struct tmpfs_node *node;
1539 node = VP_TO_TMPFS_NODE(vp);
1541 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1542 node, node->tn_flags, node->tn_links);
1543 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1544 node->tn_mode, node->tn_uid, node->tn_gid,
1545 (uintmax_t)node->tn_size, node->tn_status);
1547 if (vp->v_type == VFIFO)
1555 /* --------------------------------------------------------------------- */
1558 tmpfs_pathconf(struct vop_pathconf_args *v)
1560 int name = v->a_name;
1561 register_t *retval = v->a_retval;
1584 case _PC_CHOWN_RESTRICTED:
1596 case _PC_FILESIZEBITS:
1597 *retval = 0; /* XXX Don't know which value should I return. */
1607 /************************************************************************
1609 ************************************************************************/
1611 static void filt_tmpfsdetach(struct knote *kn);
1612 static int filt_tmpfsread(struct knote *kn, long hint);
1613 static int filt_tmpfswrite(struct knote *kn, long hint);
1614 static int filt_tmpfsvnode(struct knote *kn, long hint);
1616 static struct filterops tmpfsread_filtops =
1617 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1618 static struct filterops tmpfswrite_filtops =
1619 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1620 static struct filterops tmpfsvnode_filtops =
1621 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1624 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1626 struct vnode *vp = ap->a_vp;
1627 struct knote *kn = ap->a_kn;
1629 switch (kn->kn_filter) {
1631 kn->kn_fop = &tmpfsread_filtops;
1634 kn->kn_fop = &tmpfswrite_filtops;
1637 kn->kn_fop = &tmpfsvnode_filtops;
1640 return (EOPNOTSUPP);
1643 kn->kn_hook = (caddr_t)vp;
1645 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1651 filt_tmpfsdetach(struct knote *kn)
1653 struct vnode *vp = (void *)kn->kn_hook;
1655 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1659 filt_tmpfsread(struct knote *kn, long hint)
1661 struct vnode *vp = (void *)kn->kn_hook;
1662 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1665 if (hint == NOTE_REVOKE) {
1666 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1671 * Interlock against MP races when performing this function.
1673 lwkt_gettoken(&vp->v_mount->mnt_token);
1674 off = node->tn_size - kn->kn_fp->f_offset;
1675 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1676 if (kn->kn_sfflags & NOTE_OLDAPI) {
1677 lwkt_reltoken(&vp->v_mount->mnt_token);
1681 if (kn->kn_data == 0) {
1682 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1684 lwkt_reltoken(&vp->v_mount->mnt_token);
1685 return (kn->kn_data != 0);
1689 filt_tmpfswrite(struct knote *kn, long hint)
1691 if (hint == NOTE_REVOKE)
1692 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1698 filt_tmpfsvnode(struct knote *kn, long hint)
1700 if (kn->kn_sfflags & hint)
1701 kn->kn_fflags |= hint;
1702 if (hint == NOTE_REVOKE) {
1703 kn->kn_flags |= (EV_EOF | EV_NODATA);
1706 return (kn->kn_fflags != 0);
1710 /* --------------------------------------------------------------------- */
1713 * vnode operations vector used for files stored in a tmpfs file system.
1715 struct vop_ops tmpfs_vnode_vops = {
1716 .vop_default = vop_defaultop,
1717 .vop_getpages = vop_stdgetpages,
1718 .vop_putpages = vop_stdputpages,
1719 .vop_ncreate = tmpfs_ncreate,
1720 .vop_nresolve = tmpfs_nresolve,
1721 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1722 .vop_nmknod = tmpfs_nmknod,
1723 .vop_open = tmpfs_open,
1724 .vop_close = tmpfs_close,
1725 .vop_access = tmpfs_access,
1726 .vop_getattr = tmpfs_getattr,
1727 .vop_setattr = tmpfs_setattr,
1728 .vop_read = tmpfs_read,
1729 .vop_write = tmpfs_write,
1730 .vop_fsync = tmpfs_fsync,
1731 .vop_mountctl = tmpfs_mountctl,
1732 .vop_nremove = tmpfs_nremove,
1733 .vop_nlink = tmpfs_nlink,
1734 .vop_nrename = tmpfs_nrename,
1735 .vop_nmkdir = tmpfs_nmkdir,
1736 .vop_nrmdir = tmpfs_nrmdir,
1737 .vop_nsymlink = tmpfs_nsymlink,
1738 .vop_readdir = tmpfs_readdir,
1739 .vop_readlink = tmpfs_readlink,
1740 .vop_inactive = tmpfs_inactive,
1741 .vop_reclaim = tmpfs_reclaim,
1742 .vop_print = tmpfs_print,
1743 .vop_pathconf = tmpfs_pathconf,
1744 .vop_bmap = tmpfs_bmap,
1745 .vop_strategy = tmpfs_strategy,
1746 .vop_advlock = tmpfs_advlock,
1747 .vop_kqfilter = tmpfs_kqfilter