2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * This code is derived from software contributed to The NetBSD Foundation
6 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
34 * tmpfs vnode interface.
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.h>
50 #include <sys/vnode.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_pager.h>
56 #include <vm/swap_pager.h>
60 #include <vfs/fifofs/fifo.h>
61 #include <vfs/tmpfs/tmpfs_vnops.h>
62 #include <vfs/tmpfs/tmpfs.h>
64 MALLOC_DECLARE(M_TMPFS);
66 static void tmpfs_strategy_done(struct bio *bio);
70 tmpfs_knote(struct vnode *vp, int flags)
73 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
77 /* --------------------------------------------------------------------- */
80 tmpfs_nresolve(struct vop_nresolve_args *v)
82 struct vnode *dvp = v->a_dvp;
83 struct vnode *vp = NULL;
84 struct namecache *ncp = v->a_nch->ncp;
85 struct tmpfs_node *tnode;
88 struct tmpfs_dirent *de;
89 struct tmpfs_node *dnode;
91 dnode = VP_TO_TMPFS_DIR(dvp);
93 de = tmpfs_dir_lookup(dnode, NULL, ncp);
98 * Allocate a vnode for the node we found.
101 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
102 LK_EXCLUSIVE | LK_RETRY, &vp);
110 * Store the result of this lookup in the cache. Avoid this if the
111 * request was for creation, as it does not improve timings on
116 cache_setvp(v->a_nch, vp);
118 } else if (error == ENOENT) {
119 cache_setvp(v->a_nch, NULL);
125 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
127 struct vnode *dvp = v->a_dvp;
128 struct vnode **vpp = v->a_vpp;
129 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
130 struct ucred *cred = v->a_cred;
134 /* Check accessibility of requested node as a first step. */
135 error = VOP_ACCESS(dvp, VEXEC, cred);
139 if (dnode->tn_dir.tn_parent != NULL) {
140 /* Allocate a new vnode on the matching entry. */
141 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
142 LK_EXCLUSIVE | LK_RETRY, vpp);
148 return (*vpp == NULL) ? ENOENT : 0;
151 /* --------------------------------------------------------------------- */
154 tmpfs_ncreate(struct vop_ncreate_args *v)
156 struct vnode *dvp = v->a_dvp;
157 struct vnode **vpp = v->a_vpp;
158 struct namecache *ncp = v->a_nch->ncp;
159 struct vattr *vap = v->a_vap;
160 struct ucred *cred = v->a_cred;
163 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
165 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
167 cache_setunresolved(v->a_nch);
168 cache_setvp(v->a_nch, *vpp);
169 tmpfs_knote(dvp, NOTE_WRITE);
174 /* --------------------------------------------------------------------- */
177 tmpfs_nmknod(struct vop_nmknod_args *v)
179 struct vnode *dvp = v->a_dvp;
180 struct vnode **vpp = v->a_vpp;
181 struct namecache *ncp = v->a_nch->ncp;
182 struct vattr *vap = v->a_vap;
183 struct ucred *cred = v->a_cred;
186 if (vap->va_type != VBLK && vap->va_type != VCHR &&
187 vap->va_type != VFIFO)
190 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
192 cache_setunresolved(v->a_nch);
193 cache_setvp(v->a_nch, *vpp);
194 tmpfs_knote(dvp, NOTE_WRITE);
200 /* --------------------------------------------------------------------- */
203 tmpfs_open(struct vop_open_args *v)
205 struct vnode *vp = v->a_vp;
206 int mode = v->a_mode;
209 struct tmpfs_node *node;
211 node = VP_TO_TMPFS_NODE(vp);
213 /* The file is still active but all its names have been removed
214 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
215 * it is about to die. */
216 if (node->tn_links < 1)
219 /* If the file is marked append-only, deny write requests. */
220 if ((node->tn_flags & APPEND) &&
221 (mode & (FWRITE | O_APPEND)) == FWRITE) {
224 return (vop_stdopen(v));
229 /* --------------------------------------------------------------------- */
232 tmpfs_close(struct vop_close_args *v)
234 struct vnode *vp = v->a_vp;
235 struct tmpfs_node *node;
237 node = VP_TO_TMPFS_NODE(vp);
239 if (node->tn_links > 0) {
240 /* Update node times. No need to do it if the node has
241 * been deleted, because it will vanish after we return. */
245 return vop_stdclose(v);
248 /* --------------------------------------------------------------------- */
251 tmpfs_access(struct vop_access_args *v)
253 struct vnode *vp = v->a_vp;
255 struct tmpfs_node *node;
257 node = VP_TO_TMPFS_NODE(vp);
259 switch (vp->v_type) {
265 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
285 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
290 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
297 /* --------------------------------------------------------------------- */
300 tmpfs_getattr(struct vop_getattr_args *v)
302 struct vnode *vp = v->a_vp;
303 struct vattr *vap = v->a_vap;
304 struct tmpfs_node *node;
306 node = VP_TO_TMPFS_NODE(vp);
308 lwkt_gettoken(&vp->v_mount->mnt_token);
311 vap->va_type = vp->v_type;
312 vap->va_mode = node->tn_mode;
313 vap->va_nlink = node->tn_links;
314 vap->va_uid = node->tn_uid;
315 vap->va_gid = node->tn_gid;
316 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
317 vap->va_fileid = node->tn_id;
318 vap->va_size = node->tn_size;
319 vap->va_blocksize = PAGE_SIZE;
320 vap->va_atime.tv_sec = node->tn_atime;
321 vap->va_atime.tv_nsec = node->tn_atimensec;
322 vap->va_mtime.tv_sec = node->tn_mtime;
323 vap->va_mtime.tv_nsec = node->tn_mtimensec;
324 vap->va_ctime.tv_sec = node->tn_ctime;
325 vap->va_ctime.tv_nsec = node->tn_ctimensec;
326 vap->va_gen = node->tn_gen;
327 vap->va_flags = node->tn_flags;
328 if (vp->v_type == VBLK || vp->v_type == VCHR)
330 vap->va_rmajor = umajor(node->tn_rdev);
331 vap->va_rminor = uminor(node->tn_rdev);
333 vap->va_bytes = round_page(node->tn_size);
336 lwkt_reltoken(&vp->v_mount->mnt_token);
341 /* --------------------------------------------------------------------- */
344 tmpfs_setattr(struct vop_setattr_args *v)
346 struct vnode *vp = v->a_vp;
347 struct vattr *vap = v->a_vap;
348 struct ucred *cred = v->a_cred;
349 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
353 if (error == 0 && (vap->va_flags != VNOVAL)) {
354 error = tmpfs_chflags(vp, vap->va_flags, cred);
355 kflags |= NOTE_ATTRIB;
358 if (error == 0 && (vap->va_size != VNOVAL)) {
359 if (vap->va_size > node->tn_size)
360 kflags |= NOTE_WRITE | NOTE_EXTEND;
362 kflags |= NOTE_WRITE;
363 error = tmpfs_chsize(vp, vap->va_size, cred);
366 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
367 vap->va_gid != (gid_t)VNOVAL)) {
368 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
369 kflags |= NOTE_ATTRIB;
372 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
373 error = tmpfs_chmod(vp, vap->va_mode, cred);
374 kflags |= NOTE_ATTRIB;
377 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
378 vap->va_atime.tv_nsec != VNOVAL) ||
379 (vap->va_mtime.tv_sec != VNOVAL &&
380 vap->va_mtime.tv_nsec != VNOVAL) )) {
381 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
382 vap->va_vaflags, cred);
383 kflags |= NOTE_ATTRIB;
386 /* Update the node times. We give preference to the error codes
387 * generated by this function rather than the ones that may arise
388 * from tmpfs_update. */
390 tmpfs_knote(vp, kflags);
395 /* --------------------------------------------------------------------- */
398 * fsync is usually a NOP, but we must take action when unmounting or
402 tmpfs_fsync(struct vop_fsync_args *v)
404 struct tmpfs_mount *tmp;
405 struct tmpfs_node *node;
406 struct vnode *vp = v->a_vp;
408 tmp = VFS_TO_TMPFS(vp->v_mount);
409 node = VP_TO_TMPFS_NODE(vp);
412 if (vp->v_type == VREG) {
413 if (vp->v_flag & VRECLAIMED) {
414 if (node->tn_links == 0)
415 tmpfs_truncate(vp, 0);
417 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
423 /* --------------------------------------------------------------------- */
426 tmpfs_read (struct vop_read_args *ap)
429 struct vnode *vp = ap->a_vp;
430 struct uio *uio = ap->a_uio;
431 struct tmpfs_node *node;
438 if (uio->uio_resid == 0) {
442 node = VP_TO_TMPFS_NODE(vp);
444 if (uio->uio_offset < 0)
446 if (vp->v_type != VREG)
449 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
451 * Use buffer cache I/O (via tmpfs_strategy)
453 offset = (size_t)uio->uio_offset & BMASK;
454 base_offset = (off_t)uio->uio_offset - offset;
455 bp = getcacheblk(vp, base_offset, BSIZE);
458 lwkt_gettoken(&vp->v_mount->mnt_token);
459 error = bread(vp, base_offset, BSIZE, &bp);
462 lwkt_reltoken(&vp->v_mount->mnt_token);
463 kprintf("tmpfs_read bread error %d\n", error);
466 lwkt_reltoken(&vp->v_mount->mnt_token);
470 * Figure out how many bytes we can actually copy this loop.
472 len = BSIZE - offset;
473 if (len > uio->uio_resid)
474 len = uio->uio_resid;
475 if (len > node->tn_size - uio->uio_offset)
476 len = (size_t)(node->tn_size - uio->uio_offset);
478 error = uiomove((char *)bp->b_data + offset, len, uio);
481 kprintf("tmpfs_read uiomove error %d\n", error);
486 TMPFS_NODE_LOCK(node);
487 node->tn_status |= TMPFS_NODE_ACCESSED;
488 TMPFS_NODE_UNLOCK(node);
494 tmpfs_write (struct vop_write_args *ap)
497 struct vnode *vp = ap->a_vp;
498 struct uio *uio = ap->a_uio;
499 struct thread *td = uio->uio_td;
500 struct tmpfs_node *node;
512 if (uio->uio_resid == 0) {
516 node = VP_TO_TMPFS_NODE(vp);
518 if (vp->v_type != VREG)
521 lwkt_gettoken(&vp->v_mount->mnt_token);
523 oldsize = node->tn_size;
524 if (ap->a_ioflag & IO_APPEND)
525 uio->uio_offset = node->tn_size;
528 * Check for illegal write offsets.
530 if (uio->uio_offset + uio->uio_resid >
531 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
532 lwkt_reltoken(&vp->v_mount->mnt_token);
536 if (vp->v_type == VREG && td != NULL) {
537 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
539 lwkt_reltoken(&vp->v_mount->mnt_token);
542 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
543 ksignal(td->td_proc, SIGXFSZ);
544 lwkt_reltoken(&vp->v_mount->mnt_token);
551 * Extend the file's size if necessary
553 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
555 while (uio->uio_resid > 0) {
557 * Use buffer cache I/O (via tmpfs_strategy)
559 offset = (size_t)uio->uio_offset & BMASK;
560 base_offset = (off_t)uio->uio_offset - offset;
561 len = BSIZE - offset;
562 if (len > uio->uio_resid)
563 len = uio->uio_resid;
565 if ((uio->uio_offset + len) > node->tn_size) {
566 trivial = (uio->uio_offset <= node->tn_size);
567 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
573 * Read to fill in any gaps. Theoretically we could
574 * optimize this if the write covers the entire buffer
575 * and is not a UIO_NOCOPY write, however this can lead
576 * to a security violation exposing random kernel memory
577 * (whatever junk was in the backing VM pages before).
579 * So just use bread() to do the right thing.
581 error = bread(vp, base_offset, BSIZE, &bp);
582 error = uiomove((char *)bp->b_data + offset, len, uio);
584 kprintf("tmpfs_write uiomove error %d\n", error);
589 if (uio->uio_offset > node->tn_size) {
590 node->tn_size = uio->uio_offset;
591 kflags |= NOTE_EXTEND;
593 kflags |= NOTE_WRITE;
596 * The data has been loaded into the buffer, write it out.
598 * We want tmpfs to be able to use all available ram, not
599 * just the buffer cache, so if not explicitly paging we
600 * use buwrite() to leave the buffer clean but mark all the
601 * VM pages valid+dirty.
603 * When the kernel is paging, either via normal pageout
604 * operation or when cleaning the object during a recycle,
605 * the underlying VM pages are going to get thrown away
606 * so we MUST write them to swap.
608 * XXX unfortunately this catches msync() system calls too
611 if (vm_swap_size == 0) {
613 * if swap isn't configured yet, force a buwrite() to
614 * avoid problems further down the line, due to flushing
619 if (ap->a_ioflag & IO_SYNC) {
621 } else if ((ap->a_ioflag & IO_ASYNC) ||
622 (uio->uio_segflg == UIO_NOCOPY)) {
630 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
637 (void)tmpfs_reg_resize(vp, oldsize, trivial);
638 kflags &= ~NOTE_EXTEND;
643 TMPFS_NODE_LOCK(node);
644 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
645 (extended? TMPFS_NODE_CHANGED : 0);
647 if (node->tn_mode & (S_ISUID | S_ISGID)) {
648 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
649 node->tn_mode &= ~(S_ISUID | S_ISGID);
651 TMPFS_NODE_UNLOCK(node);
654 tmpfs_knote(vp, kflags);
657 lwkt_reltoken(&vp->v_mount->mnt_token);
662 tmpfs_advlock (struct vop_advlock_args *ap)
664 struct tmpfs_node *node;
665 struct vnode *vp = ap->a_vp;
667 node = VP_TO_TMPFS_NODE(vp);
669 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
673 tmpfs_strategy(struct vop_strategy_args *ap)
675 struct bio *bio = ap->a_bio;
677 struct buf *bp = bio->bio_buf;
678 struct vnode *vp = ap->a_vp;
679 struct tmpfs_node *node;
684 if (vp->v_type != VREG) {
685 bp->b_resid = bp->b_bcount;
686 bp->b_flags |= B_ERROR | B_INVAL;
687 bp->b_error = EINVAL;
692 lwkt_gettoken(&vp->v_mount->mnt_token);
693 node = VP_TO_TMPFS_NODE(vp);
695 uobj = node->tn_reg.tn_aobj;
698 * Certain operations such as nvtruncbuf() can result in a
699 * bdwrite() of one or more buffers related to the file,
700 * leading to the possibility of our strategy function
701 * being called for writing even when there is no swap space.
703 * When this case occurs we mark the underlying pages as valid
704 * and dirty and complete the I/O manually.
706 * Otherwise just call swap_pager_strategy to read or write,
707 * potentially assigning swap on write. We push a BIO to catch
708 * any swap allocation errors.
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_set_validdirty(m, 0, PAGE_SIZE);
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 * bio finished. If we ran out of sap just mark the pages valid
731 * and dirty and make it appear that the I/O has completed successfully.
734 tmpfs_strategy_done(struct bio *bio)
742 if ((bp->b_flags & B_ERROR) && bp->b_error == ENOMEM) {
743 bp->b_flags &= ~B_ERROR;
746 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
747 m = bp->b_xio.xio_pages[i];
748 vm_page_set_validdirty(m, 0, PAGE_SIZE);
756 tmpfs_bmap(struct vop_bmap_args *ap)
758 if (ap->a_doffsetp != NULL)
759 *ap->a_doffsetp = ap->a_loffset;
760 if (ap->a_runp != NULL)
762 if (ap->a_runb != NULL)
768 /* --------------------------------------------------------------------- */
771 tmpfs_nremove(struct vop_nremove_args *v)
773 struct vnode *dvp = v->a_dvp;
774 struct namecache *ncp = v->a_nch->ncp;
777 struct tmpfs_dirent *de;
778 struct tmpfs_mount *tmp;
779 struct tmpfs_node *dnode;
780 struct tmpfs_node *node;
783 * We have to acquire the vp from v->a_nch because we will likely
784 * unresolve the namecache entry, and a vrele/vput is needed to
785 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
787 * We have to use vget to clear any inactive state on the vnode,
788 * otherwise the vnode may remain inactive and thus tmpfs_inactive
789 * will not get called when we release it.
791 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
792 KKASSERT(error == 0);
795 if (vp->v_type == VDIR) {
800 dnode = VP_TO_TMPFS_DIR(dvp);
801 node = VP_TO_TMPFS_NODE(vp);
802 tmp = VFS_TO_TMPFS(vp->v_mount);
803 de = tmpfs_dir_lookup(dnode, node, ncp);
809 /* Files marked as immutable or append-only cannot be deleted. */
810 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
811 (dnode->tn_flags & APPEND)) {
816 /* Remove the entry from the directory; as it is a file, we do not
817 * have to change the number of hard links of the directory. */
818 tmpfs_dir_detach(dnode, de);
820 /* Free the directory entry we just deleted. Note that the node
821 * referred by it will not be removed until the vnode is really
823 tmpfs_free_dirent(tmp, de);
825 if (node->tn_links > 0) {
826 TMPFS_NODE_LOCK(node);
827 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
829 TMPFS_NODE_UNLOCK(node);
832 cache_setunresolved(v->a_nch);
833 cache_setvp(v->a_nch, NULL);
834 tmpfs_knote(vp, NOTE_DELETE);
835 /*cache_inval_vp(vp, CINV_DESTROY);*/
836 tmpfs_knote(dvp, NOTE_WRITE);
845 /* --------------------------------------------------------------------- */
848 tmpfs_nlink(struct vop_nlink_args *v)
850 struct vnode *dvp = v->a_dvp;
851 struct vnode *vp = v->a_vp;
852 struct namecache *ncp = v->a_nch->ncp;
853 struct tmpfs_dirent *de;
854 struct tmpfs_node *node;
855 struct tmpfs_node *dnode;
858 KKASSERT(dvp != vp); /* XXX When can this be false? */
860 node = VP_TO_TMPFS_NODE(vp);
861 dnode = VP_TO_TMPFS_NODE(dvp);
863 /* XXX: Why aren't the following two tests done by the caller? */
865 /* Hard links of directories are forbidden. */
866 if (vp->v_type == VDIR) {
871 /* Cannot create cross-device links. */
872 if (dvp->v_mount != vp->v_mount) {
877 /* Ensure that we do not overflow the maximum number of links imposed
879 KKASSERT(node->tn_links <= LINK_MAX);
880 if (node->tn_links == LINK_MAX) {
885 /* We cannot create links of files marked immutable or append-only. */
886 if (node->tn_flags & (IMMUTABLE | APPEND)) {
891 /* Allocate a new directory entry to represent the node. */
892 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
893 ncp->nc_name, ncp->nc_nlen, &de);
897 /* Insert the new directory entry into the appropriate directory. */
898 tmpfs_dir_attach(dnode, de);
900 /* vp link count has changed, so update node times. */
902 TMPFS_NODE_LOCK(node);
903 node->tn_status |= TMPFS_NODE_CHANGED;
904 TMPFS_NODE_UNLOCK(node);
907 tmpfs_knote(vp, NOTE_LINK);
908 cache_setunresolved(v->a_nch);
909 cache_setvp(v->a_nch, vp);
910 tmpfs_knote(dvp, NOTE_WRITE);
917 /* --------------------------------------------------------------------- */
920 tmpfs_nrename(struct vop_nrename_args *v)
922 struct vnode *fdvp = v->a_fdvp;
923 struct namecache *fncp = v->a_fnch->ncp;
924 struct vnode *fvp = fncp->nc_vp;
925 struct vnode *tdvp = v->a_tdvp;
926 struct namecache *tncp = v->a_tnch->ncp;
928 struct tmpfs_dirent *de;
929 struct tmpfs_mount *tmp;
930 struct tmpfs_node *fdnode;
931 struct tmpfs_node *fnode;
932 struct tmpfs_node *tnode;
933 struct tmpfs_node *tdnode;
939 * Because tvp can get overwritten we have to vget it instead of
940 * just vref or use it, otherwise it's VINACTIVE flag may not get
941 * cleared and the node won't get destroyed.
943 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
945 tnode = VP_TO_TMPFS_NODE(tvp);
951 /* Disallow cross-device renames.
952 * XXX Why isn't this done by the caller? */
953 if (fvp->v_mount != tdvp->v_mount ||
954 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
959 tmp = VFS_TO_TMPFS(tdvp->v_mount);
960 tdnode = VP_TO_TMPFS_DIR(tdvp);
962 /* If source and target are the same file, there is nothing to do. */
968 fdnode = VP_TO_TMPFS_DIR(fdvp);
969 fnode = VP_TO_TMPFS_NODE(fvp);
970 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
972 /* Avoid manipulating '.' and '..' entries. */
977 KKASSERT(de->td_node == fnode);
980 * If replacing an entry in the target directory and that entry
981 * is a directory, it must be empty.
983 * Kern_rename gurantees the destination to be a directory
984 * if the source is one (it does?).
987 KKASSERT(tnode != NULL);
989 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
990 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
995 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
996 if (tnode->tn_size > 0) {
1000 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1003 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1007 KKASSERT(fnode->tn_type != VDIR &&
1008 tnode->tn_type != VDIR);
1012 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1013 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1019 * Ensure that we have enough memory to hold the new name, if it
1020 * has to be changed.
1022 if (fncp->nc_nlen != tncp->nc_nlen ||
1023 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1024 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1025 M_WAITOK | M_NULLOK);
1026 if (newname == NULL) {
1030 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1031 newname[tncp->nc_nlen] = '\0';
1037 * Unlink entry from source directory. Note that the kernel has
1038 * already checked for illegal recursion cases (renaming a directory
1039 * into a subdirectory of itself).
1041 if (fdnode != tdnode)
1042 tmpfs_dir_detach(fdnode, de);
1045 * Handle any name change. Swap with newname, we will
1046 * deallocate it at the end.
1048 if (newname != NULL) {
1050 TMPFS_NODE_LOCK(fnode);
1051 fnode->tn_status |= TMPFS_NODE_CHANGED;
1052 TMPFS_NODE_UNLOCK(fnode);
1054 oldname = de->td_name;
1055 de->td_name = newname;
1056 de->td_namelen = (uint16_t)tncp->nc_nlen;
1061 * Link entry to target directory. If the entry
1062 * represents a directory move the parent linkage
1065 if (fdnode != tdnode) {
1066 if (de->td_node->tn_type == VDIR) {
1067 TMPFS_VALIDATE_DIR(fnode);
1069 TMPFS_NODE_LOCK(tdnode);
1071 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1072 TMPFS_NODE_UNLOCK(tdnode);
1074 TMPFS_NODE_LOCK(fnode);
1075 fnode->tn_dir.tn_parent = tdnode;
1076 fnode->tn_status |= TMPFS_NODE_CHANGED;
1077 TMPFS_NODE_UNLOCK(fnode);
1079 TMPFS_NODE_LOCK(fdnode);
1081 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1082 TMPFS_NODE_UNLOCK(fdnode);
1084 tmpfs_dir_attach(tdnode, de);
1086 TMPFS_NODE_LOCK(tdnode);
1087 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1088 TMPFS_NODE_UNLOCK(tdnode);
1092 * If we are overwriting an entry, we have to remove the old one
1093 * from the target directory.
1096 /* Remove the old entry from the target directory. */
1097 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1098 tmpfs_dir_detach(tdnode, de);
1099 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1102 * Free the directory entry we just deleted. Note that the
1103 * node referred by it will not be removed until the vnode is
1106 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1107 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1114 kfree(newname, tmp->tm_name_zone);
1117 cache_rename(v->a_fnch, v->a_tnch);
1118 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1119 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1120 if (fnode->tn_vnode)
1121 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1134 /* --------------------------------------------------------------------- */
1137 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1139 struct vnode *dvp = v->a_dvp;
1140 struct vnode **vpp = v->a_vpp;
1141 struct namecache *ncp = v->a_nch->ncp;
1142 struct vattr *vap = v->a_vap;
1143 struct ucred *cred = v->a_cred;
1146 KKASSERT(vap->va_type == VDIR);
1148 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1150 cache_setunresolved(v->a_nch);
1151 cache_setvp(v->a_nch, *vpp);
1152 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1158 /* --------------------------------------------------------------------- */
1161 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1163 struct vnode *dvp = v->a_dvp;
1164 struct namecache *ncp = v->a_nch->ncp;
1166 struct tmpfs_dirent *de;
1167 struct tmpfs_mount *tmp;
1168 struct tmpfs_node *dnode;
1169 struct tmpfs_node *node;
1173 * We have to acquire the vp from v->a_nch because we will likely
1174 * unresolve the namecache entry, and a vrele/vput is needed to
1175 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1177 * We have to use vget to clear any inactive state on the vnode,
1178 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1179 * will not get called when we release it.
1181 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1182 KKASSERT(error == 0);
1186 * Prevalidate so we don't hit an assertion later
1188 if (vp->v_type != VDIR) {
1193 tmp = VFS_TO_TMPFS(dvp->v_mount);
1194 dnode = VP_TO_TMPFS_DIR(dvp);
1195 node = VP_TO_TMPFS_DIR(vp);
1197 /* Directories with more than two entries ('.' and '..') cannot be
1199 if (node->tn_size > 0) {
1204 if ((dnode->tn_flags & APPEND)
1205 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1210 /* This invariant holds only if we are not trying to remove "..".
1211 * We checked for that above so this is safe now. */
1212 KKASSERT(node->tn_dir.tn_parent == dnode);
1214 /* Get the directory entry associated with node (vp). This was
1215 * filled by tmpfs_lookup while looking up the entry. */
1216 de = tmpfs_dir_lookup(dnode, node, ncp);
1217 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1221 /* Check flags to see if we are allowed to remove the directory. */
1222 if ((dnode->tn_flags & APPEND) ||
1223 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1229 /* Detach the directory entry from the directory (dnode). */
1230 tmpfs_dir_detach(dnode, de);
1232 /* No vnode should be allocated for this entry from this point */
1233 TMPFS_NODE_LOCK(node);
1234 TMPFS_ASSERT_ELOCKED(node);
1235 TMPFS_NODE_LOCK(dnode);
1236 TMPFS_ASSERT_ELOCKED(dnode);
1239 /* handled by tmpfs_free_node */
1240 KKASSERT(node->tn_links > 0);
1242 node->tn_dir.tn_parent = NULL;
1244 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1245 TMPFS_NODE_MODIFIED;
1248 /* handled by tmpfs_free_node */
1249 KKASSERT(dnode->tn_links > 0);
1252 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1253 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1255 TMPFS_NODE_UNLOCK(dnode);
1256 TMPFS_NODE_UNLOCK(node);
1258 /* Free the directory entry we just deleted. Note that the node
1259 * referred by it will not be removed until the vnode is really
1261 tmpfs_free_dirent(tmp, de);
1263 /* Release the deleted vnode (will destroy the node, notify
1264 * interested parties and clean it from the cache). */
1266 TMPFS_NODE_LOCK(dnode);
1267 dnode->tn_status |= TMPFS_NODE_CHANGED;
1268 TMPFS_NODE_UNLOCK(dnode);
1271 cache_setunresolved(v->a_nch);
1272 cache_setvp(v->a_nch, NULL);
1273 /*cache_inval_vp(vp, CINV_DESTROY);*/
1274 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1283 /* --------------------------------------------------------------------- */
1286 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1288 struct vnode *dvp = v->a_dvp;
1289 struct vnode **vpp = v->a_vpp;
1290 struct namecache *ncp = v->a_nch->ncp;
1291 struct vattr *vap = v->a_vap;
1292 struct ucred *cred = v->a_cred;
1293 char *target = v->a_target;
1296 vap->va_type = VLNK;
1297 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1299 tmpfs_knote(*vpp, NOTE_WRITE);
1300 cache_setunresolved(v->a_nch);
1301 cache_setvp(v->a_nch, *vpp);
1307 /* --------------------------------------------------------------------- */
1310 tmpfs_readdir(struct vop_readdir_args *v)
1312 struct vnode *vp = v->a_vp;
1313 struct uio *uio = v->a_uio;
1314 int *eofflag = v->a_eofflag;
1315 off_t **cookies = v->a_cookies;
1316 int *ncookies = v->a_ncookies;
1317 struct tmpfs_mount *tmp;
1321 struct tmpfs_node *node;
1323 /* This operation only makes sense on directory nodes. */
1324 if (vp->v_type != VDIR)
1327 tmp = VFS_TO_TMPFS(vp->v_mount);
1328 node = VP_TO_TMPFS_DIR(vp);
1329 startoff = uio->uio_offset;
1331 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1332 error = tmpfs_dir_getdotdent(node, uio);
1338 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1339 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1345 error = tmpfs_dir_getdents(node, uio, &cnt);
1348 KKASSERT(error >= -1);
1353 if (eofflag != NULL)
1355 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1357 /* Update NFS-related variables. */
1358 if (error == 0 && cookies != NULL && ncookies != NULL) {
1360 off_t off = startoff;
1361 struct tmpfs_dirent *de = NULL;
1364 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1366 for (i = 0; i < cnt; i++) {
1367 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1368 if (off == TMPFS_DIRCOOKIE_DOT) {
1369 off = TMPFS_DIRCOOKIE_DOTDOT;
1371 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1372 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1373 } else if (de != NULL) {
1374 de = TAILQ_NEXT(de, td_entries);
1376 de = tmpfs_dir_lookupbycookie(node,
1378 KKASSERT(de != NULL);
1379 de = TAILQ_NEXT(de, td_entries);
1382 off = TMPFS_DIRCOOKIE_EOF;
1384 off = tmpfs_dircookie(de);
1387 (*cookies)[i] = off;
1389 KKASSERT(uio->uio_offset == off);
1395 /* --------------------------------------------------------------------- */
1398 tmpfs_readlink(struct vop_readlink_args *v)
1400 struct vnode *vp = v->a_vp;
1401 struct uio *uio = v->a_uio;
1404 struct tmpfs_node *node;
1406 KKASSERT(uio->uio_offset == 0);
1407 KKASSERT(vp->v_type == VLNK);
1409 node = VP_TO_TMPFS_NODE(vp);
1411 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1413 TMPFS_NODE_LOCK(node);
1414 node->tn_status |= TMPFS_NODE_ACCESSED;
1415 TMPFS_NODE_UNLOCK(node);
1420 /* --------------------------------------------------------------------- */
1423 tmpfs_inactive(struct vop_inactive_args *v)
1425 struct vnode *vp = v->a_vp;
1426 struct tmpfs_node *node;
1428 node = VP_TO_TMPFS_NODE(vp);
1439 * Get rid of unreferenced deleted vnodes sooner rather than
1440 * later so the data memory can be recovered immediately.
1442 * We must truncate the vnode to prevent the normal reclamation
1443 * path from flushing the data for the removed file to disk.
1445 TMPFS_NODE_LOCK(node);
1446 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1447 (node->tn_links == 0 ||
1448 (node->tn_links == 1 && node->tn_type == VDIR &&
1449 node->tn_dir.tn_parent)))
1451 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1452 TMPFS_NODE_UNLOCK(node);
1453 if (node->tn_type == VREG)
1454 tmpfs_truncate(vp, 0);
1457 TMPFS_NODE_UNLOCK(node);
1463 /* --------------------------------------------------------------------- */
1466 tmpfs_reclaim(struct vop_reclaim_args *v)
1468 struct vnode *vp = v->a_vp;
1469 struct tmpfs_mount *tmp;
1470 struct tmpfs_node *node;
1472 node = VP_TO_TMPFS_NODE(vp);
1473 tmp = VFS_TO_TMPFS(vp->v_mount);
1478 * If the node referenced by this vnode was deleted by the
1479 * user, we must free its associated data structures now that
1480 * the vnode is being reclaimed.
1482 * Directories have an extra link ref.
1484 TMPFS_NODE_LOCK(node);
1485 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1486 (node->tn_links == 0 ||
1487 (node->tn_links == 1 && node->tn_type == VDIR &&
1488 node->tn_dir.tn_parent)))
1490 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1491 tmpfs_free_node(tmp, node);
1494 TMPFS_NODE_UNLOCK(node);
1497 KKASSERT(vp->v_data == NULL);
1501 /* --------------------------------------------------------------------- */
1504 tmpfs_print(struct vop_print_args *v)
1506 struct vnode *vp = v->a_vp;
1508 struct tmpfs_node *node;
1510 node = VP_TO_TMPFS_NODE(vp);
1512 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1513 node, node->tn_flags, node->tn_links);
1514 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1515 node->tn_mode, node->tn_uid, node->tn_gid,
1516 (uintmax_t)node->tn_size, node->tn_status);
1518 if (vp->v_type == VFIFO)
1526 /* --------------------------------------------------------------------- */
1529 tmpfs_pathconf(struct vop_pathconf_args *v)
1531 int name = v->a_name;
1532 register_t *retval = v->a_retval;
1555 case _PC_CHOWN_RESTRICTED:
1567 case _PC_FILESIZEBITS:
1568 *retval = 0; /* XXX Don't know which value should I return. */
1578 /************************************************************************
1580 ************************************************************************/
1582 static void filt_tmpfsdetach(struct knote *kn);
1583 static int filt_tmpfsread(struct knote *kn, long hint);
1584 static int filt_tmpfswrite(struct knote *kn, long hint);
1585 static int filt_tmpfsvnode(struct knote *kn, long hint);
1587 static struct filterops tmpfsread_filtops =
1588 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1589 static struct filterops tmpfswrite_filtops =
1590 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1591 static struct filterops tmpfsvnode_filtops =
1592 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1595 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1597 struct vnode *vp = ap->a_vp;
1598 struct knote *kn = ap->a_kn;
1600 switch (kn->kn_filter) {
1602 kn->kn_fop = &tmpfsread_filtops;
1605 kn->kn_fop = &tmpfswrite_filtops;
1608 kn->kn_fop = &tmpfsvnode_filtops;
1611 return (EOPNOTSUPP);
1614 kn->kn_hook = (caddr_t)vp;
1616 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1622 filt_tmpfsdetach(struct knote *kn)
1624 struct vnode *vp = (void *)kn->kn_hook;
1626 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1630 filt_tmpfsread(struct knote *kn, long hint)
1632 struct vnode *vp = (void *)kn->kn_hook;
1633 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1636 if (hint == NOTE_REVOKE) {
1637 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1642 * Interlock against MP races when performing this function.
1644 lwkt_gettoken(&vp->v_mount->mnt_token);
1645 off = node->tn_size - kn->kn_fp->f_offset;
1646 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1647 if (kn->kn_sfflags & NOTE_OLDAPI) {
1648 lwkt_reltoken(&vp->v_mount->mnt_token);
1652 if (kn->kn_data == 0) {
1653 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1655 lwkt_reltoken(&vp->v_mount->mnt_token);
1656 return (kn->kn_data != 0);
1660 filt_tmpfswrite(struct knote *kn, long hint)
1662 if (hint == NOTE_REVOKE)
1663 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1669 filt_tmpfsvnode(struct knote *kn, long hint)
1671 if (kn->kn_sfflags & hint)
1672 kn->kn_fflags |= hint;
1673 if (hint == NOTE_REVOKE) {
1674 kn->kn_flags |= (EV_EOF | EV_NODATA);
1677 return (kn->kn_fflags != 0);
1681 /* --------------------------------------------------------------------- */
1684 * vnode operations vector used for files stored in a tmpfs file system.
1686 struct vop_ops tmpfs_vnode_vops = {
1687 .vop_default = vop_defaultop,
1688 .vop_getpages = vop_stdgetpages,
1689 .vop_putpages = vop_stdputpages,
1690 .vop_ncreate = tmpfs_ncreate,
1691 .vop_nresolve = tmpfs_nresolve,
1692 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1693 .vop_nmknod = tmpfs_nmknod,
1694 .vop_open = tmpfs_open,
1695 .vop_close = tmpfs_close,
1696 .vop_access = tmpfs_access,
1697 .vop_getattr = tmpfs_getattr,
1698 .vop_setattr = tmpfs_setattr,
1699 .vop_read = tmpfs_read,
1700 .vop_write = tmpfs_write,
1701 .vop_fsync = tmpfs_fsync,
1702 .vop_nremove = tmpfs_nremove,
1703 .vop_nlink = tmpfs_nlink,
1704 .vop_nrename = tmpfs_nrename,
1705 .vop_nmkdir = tmpfs_nmkdir,
1706 .vop_nrmdir = tmpfs_nrmdir,
1707 .vop_nsymlink = tmpfs_nsymlink,
1708 .vop_readdir = tmpfs_readdir,
1709 .vop_readlink = tmpfs_readlink,
1710 .vop_inactive = tmpfs_inactive,
1711 .vop_reclaim = tmpfs_reclaim,
1712 .vop_print = tmpfs_print,
1713 .vop_pathconf = tmpfs_pathconf,
1714 .vop_bmap = tmpfs_bmap,
1715 .vop_strategy = tmpfs_strategy,
1716 .vop_advlock = tmpfs_advlock,
1717 .vop_kqfilter = tmpfs_kqfilter