hammer2 - create new branch, sync working trees from dillon & vsrinivas
[dragonfly.git] / sys / vfs / hammer2 / hammer2_vnops.c
CommitLineData
703720e4
MD
1#include <sys/param.h>
2#include <sys/systm.h>
3#include <sys/kernel.h>
4#include <sys/fcntl.h>
5#include <sys/buf.h>
6#include <sys/proc.h>
7#include <sys/namei.h>
8#include <sys/mount.h>
9#include <sys/vnode.h>
10
11#include "hammer2.h"
12
13
14/*
15 * Last reference to a vnode is going away but it is still cached.
16 */
17int
18hammer2_inactive(struct vop_inactive_args *ap)
19{
20 struct vnode *vp;
21 struct hammer2_inode *ip;
22 struct hammer2_mount *hmp;
23
24 kprintf("hammer2_inactive\n");
25
26 vp = ap->a_vp;
27 ip = VTOI(vp);
28 // hmp = ip->hi_mp;
29
30 return (0);
31}
32
33/*
34 * Reclaim a vnode so that it can be reused; after the inode is
35 * disassociated, the filesystem must manage it alone.
36 */
37int
38hammer2_reclaim(struct vop_reclaim_args *ap)
39{
40 struct vnode *vp;
41 struct hammer2_inode *ip;
42 struct hammer2_mount *hmp;
43
44 kprintf("hammer2_reclaim\n");
45
46 /* Is the vnode locked? Must it be on exit? */
47
48 vp = ap->a_vp;
49 ip = VTOI(vp);
50 hmp = ip->hi_mp;
51
52 vp->v_data = NULL;
53 ip->hi_vnode = NULL;
54
55 return (0);
56}
57
58
59int
60hammer2_fsync(struct vop_fsync_args *ap)
61{
62 kprintf("hammer2_fsync\n");
63 return (EOPNOTSUPP);
64}
65
66int
67hammer2_access(struct vop_access_args *ap)
68{
69 kprintf("hammer2_access\n");
70 return (0);
71}
72
73int
74hammer2_getattr(struct vop_getattr_args *ap)
75{
76 struct vnode *vp;
77 struct vattr *vap;
78 struct hammer2_inode *ip;
79
80 vp = ap->a_vp;
81 vap = ap->a_vap;
82
83 kprintf("hammer2_getattr\n");
84
85 ip = VTOI(vp);
86 hammer2_inode_lock_sh(ip);
87
88 vap->va_type = vp->v_type;
89 vap->va_mode = 0777;
90 vap->va_nlink = 1;
91 vap->va_uid = 0;
92 vap->va_gid = 0;
93 vap->va_size = 0;
94 vap->va_blocksize = PAGE_SIZE;
95 vap->va_flags = 0;
96
97 hammer2_inode_unlock_sh(ip);
98
99 return (0);
100}
101
102int
103hammer2_readdir(struct vop_readdir_args *ap)
104{
105 kprintf("hammer2_readdir\n");
106 return (EOPNOTSUPP);
107}
108
109int
110hammer2_read(struct vop_read_args *ap)
111{
112
113}
114
115int
116hammer2_nresolve(struct vop_nresolve_args *ap)
117{
118 kprintf("hammer2_nresolve\n");
119 return EOPNOTSUPP;
120}
121
122int
123hammer2_bmap(struct vop_bmap_args *ap)
124{
125 kprintf("hammer2_bmap\n");
126 return (EOPNOTSUPP);
127}
128
129int
130hammer2_open(struct vop_open_args *ap)
131{
132 kprintf("hammer2_open\n");
133 return vop_stdopen(ap);
134}
135
136int
137hammer2_strategy(struct vop_strategy_args *ap)
138{
139 struct vnode *vp;
140 struct bio *biop;
141 struct buf *bp;
142 struct hammer2_inode *ip;
143 int error;
144
145 vp = ap->a_vp;
146 biop = ap->a_bio;
147 bp = biop->bio_buf;
148 ip = VTOI(vp);
149
150 switch(bp->b_cmd) {
151 case (BUF_CMD_READ):
152 case (BUF_CMD_WRITE):
153 default:
154 bp->b_error = error = EINVAL;
155 bp->b_flags |= B_ERROR;
156 biodone(biop);
157 break;
158 }
159
160 return (error);
161}
162
163struct vop_ops hammer2_vnode_vops = {
164 .vop_default = vop_defaultop,
165 .vop_fsync = hammer2_fsync,
166 .vop_getpages = vop_stdgetpages,
167 .vop_putpages = vop_stdputpages,
168 .vop_access = hammer2_access,
169 .vop_getattr = hammer2_getattr,
170//// .vop_readdir = hammer2_readdir,
171// .vop_read = hammer2_read,
172// .vop_write = hammer2_write,
173 .vop_open = hammer2_open,
174 .vop_inactive = hammer2_inactive,
175 .vop_reclaim = hammer2_reclaim,
176 .vop_nresolve = hammer2_nresolve,
177
178// .vop_bmap = hammer2_bmap,
179// .vop_strategy = hammer2_strategy,
180};
181
182struct vop_ops hammer2_spec_vops = {
183
184};
185
186struct vop_ops hammer2_fifo_vops = {
187
188};
189/*-
190 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
191 * All rights reserved.
192 *
193 * This code is derived from software contributed to The NetBSD Foundation
194 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
195 * 2005 program.
196 *
197 * Redistribution and use in source and binary forms, with or without
198 * modification, are permitted provided that the following conditions
199 * are met:
200 * 1. Redistributions of source code must retain the above copyright
201 * notice, this list of conditions and the following disclaimer.
202 * 2. Redistributions in binary form must reproduce the above copyright
203 * notice, this list of conditions and the following disclaimer in the
204 * documentation and/or other materials provided with the distribution.
205 *
206 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
207 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
208 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
209 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
210 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
211 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
212 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
213 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
214 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
215 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
216 * POSSIBILITY OF SUCH DAMAGE.
217 *
218 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
219 */
220
221/*
222 * tmpfs vnode interface.
223 */
224
225#include <sys/kernel.h>
226#include <sys/kern_syscall.h>
227#include <sys/param.h>
228#include <sys/fcntl.h>
229#include <sys/lockf.h>
230#include <sys/priv.h>
231#include <sys/proc.h>
232#include <sys/resourcevar.h>
233#include <sys/sched.h>
234#include <sys/stat.h>
235#include <sys/systm.h>
236#include <sys/unistd.h>
237#include <sys/vfsops.h>
238#include <sys/vnode.h>
239
240#include <vm/vm.h>
241#include <vm/vm_object.h>
242#include <vm/vm_page.h>
243#include <vm/vm_pager.h>
244#include <vm/swap_pager.h>
245
246#include <sys/buf2.h>
247
248#include <vfs/fifofs/fifo.h>
249#include "hammer2.h"
250
251MALLOC_DECLARE(M_TMPFS);
252
253static void tmpfs_strategy_done(struct bio *bio);
254
255static __inline
256void
257tmpfs_knote(struct vnode *vp, int flags)
258{
259 if (flags)
260 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
261}
262
263
264/* --------------------------------------------------------------------- */
265
266static int
267tmpfs_nresolve(struct vop_nresolve_args *v)
268{
269 struct vnode *dvp = v->a_dvp;
270 struct vnode *vp = NULL;
271 struct namecache *ncp = v->a_nch->ncp;
272 struct tmpfs_node *tnode;
273
274 int error;
275 struct tmpfs_dirent *de;
276 struct tmpfs_node *dnode;
277
278 dnode = VP_TO_TMPFS_DIR(dvp);
279
280 de = tmpfs_dir_lookup(dnode, NULL, ncp);
281 if (de == NULL) {
282 error = ENOENT;
283 } else {
284 /*
285 * Allocate a vnode for the node we found.
286 */
287 tnode = de->td_node;
288 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
289 LK_EXCLUSIVE | LK_RETRY, &vp);
290 if (error)
291 goto out;
292 KKASSERT(vp);
293 }
294
295out:
296 /*
297 * Store the result of this lookup in the cache. Avoid this if the
298 * request was for creation, as it does not improve timings on
299 * emprical tests.
300 */
301 if (vp) {
302 vn_unlock(vp);
303 cache_setvp(v->a_nch, vp);
304 vrele(vp);
305 } else if (error == ENOENT) {
306 cache_setvp(v->a_nch, NULL);
307 }
308 return error;
309}
310
311static int
312tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
313{
314 struct vnode *dvp = v->a_dvp;
315 struct vnode **vpp = v->a_vpp;
316 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
317 struct ucred *cred = v->a_cred;
318 int error;
319
320 *vpp = NULL;
321 /* Check accessibility of requested node as a first step. */
322 error = VOP_ACCESS(dvp, VEXEC, cred);
323 if (error != 0)
324 return error;
325
326 if (dnode->tn_dir.tn_parent != NULL) {
327 /* Allocate a new vnode on the matching entry. */
328 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
329 LK_EXCLUSIVE | LK_RETRY, vpp);
330
331 if (*vpp)
332 vn_unlock(*vpp);
333 }
334
335 return (*vpp == NULL) ? ENOENT : 0;
336}
337
338/* --------------------------------------------------------------------- */
339
340static int
341tmpfs_ncreate(struct vop_ncreate_args *v)
342{
343 struct vnode *dvp = v->a_dvp;
344 struct vnode **vpp = v->a_vpp;
345 struct namecache *ncp = v->a_nch->ncp;
346 struct vattr *vap = v->a_vap;
347 struct ucred *cred = v->a_cred;
348 int error;
349
350 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
351
352 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
353 if (error == 0) {
354 cache_setunresolved(v->a_nch);
355 cache_setvp(v->a_nch, *vpp);
356 tmpfs_knote(dvp, NOTE_WRITE);
357 }
358
359 return error;
360}
361/* --------------------------------------------------------------------- */
362
363static int
364tmpfs_nmknod(struct vop_nmknod_args *v)
365{
366 struct vnode *dvp = v->a_dvp;
367 struct vnode **vpp = v->a_vpp;
368 struct namecache *ncp = v->a_nch->ncp;
369 struct vattr *vap = v->a_vap;
370 struct ucred *cred = v->a_cred;
371 int error;
372
373 if (vap->va_type != VBLK && vap->va_type != VCHR &&
374 vap->va_type != VFIFO)
375 return EINVAL;
376
377 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
378 if (error == 0) {
379 cache_setunresolved(v->a_nch);
380 cache_setvp(v->a_nch, *vpp);
381 tmpfs_knote(dvp, NOTE_WRITE);
382 }
383
384 return error;
385}
386
387/* --------------------------------------------------------------------- */
388
389static int
390tmpfs_open(struct vop_open_args *v)
391{
392 struct vnode *vp = v->a_vp;
393 int mode = v->a_mode;
394
395 int error;
396 struct tmpfs_node *node;
397
398 node = VP_TO_TMPFS_NODE(vp);
399
400 /* The file is still active but all its names have been removed
401 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
402 * it is about to die. */
403 if (node->tn_links < 1)
404 return (ENOENT);
405
406 /* If the file is marked append-only, deny write requests. */
407 if ((node->tn_flags & APPEND) &&
408 (mode & (FWRITE | O_APPEND)) == FWRITE) {
409 error = EPERM;
410 } else {
411 return (vop_stdopen(v));
412 }
413 return error;
414}
415
416/* --------------------------------------------------------------------- */
417
418static int
419tmpfs_close(struct vop_close_args *v)
420{
421 struct vnode *vp = v->a_vp;
422 struct tmpfs_node *node;
423
424 node = VP_TO_TMPFS_NODE(vp);
425
426 if (node->tn_links > 0) {
427 /* Update node times. No need to do it if the node has
428 * been deleted, because it will vanish after we return. */
429 tmpfs_update(vp);
430 }
431
432 return vop_stdclose(v);
433}
434
435/* --------------------------------------------------------------------- */
436
437int
438tmpfs_access(struct vop_access_args *v)
439{
440 struct vnode *vp = v->a_vp;
441 int error;
442 struct tmpfs_node *node;
443
444 node = VP_TO_TMPFS_NODE(vp);
445
446 switch (vp->v_type) {
447 case VDIR:
448 /* FALLTHROUGH */
449 case VLNK:
450 /* FALLTHROUGH */
451 case VREG:
452 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
453 error = EROFS;
454 goto out;
455 }
456 break;
457
458 case VBLK:
459 /* FALLTHROUGH */
460 case VCHR:
461 /* FALLTHROUGH */
462 case VSOCK:
463 /* FALLTHROUGH */
464 case VFIFO:
465 break;
466
467 default:
468 error = EINVAL;
469 goto out;
470 }
471
472 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
473 error = EPERM;
474 goto out;
475 }
476
477 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
478
479out:
480
481 return error;
482}
483
484/* --------------------------------------------------------------------- */
485
486int
487tmpfs_getattr(struct vop_getattr_args *v)
488{
489 struct vnode *vp = v->a_vp;
490 struct vattr *vap = v->a_vap;
491 struct tmpfs_node *node;
492
493 node = VP_TO_TMPFS_NODE(vp);
494
495 lwkt_gettoken(&vp->v_mount->mnt_token);
496 tmpfs_update(vp);
497
498 vap->va_type = vp->v_type;
499 vap->va_mode = node->tn_mode;
500 vap->va_nlink = node->tn_links;
501 vap->va_uid = node->tn_uid;
502 vap->va_gid = node->tn_gid;
503 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
504 vap->va_fileid = node->tn_id;
505 vap->va_size = node->tn_size;
506 vap->va_blocksize = PAGE_SIZE;
507 vap->va_atime.tv_sec = node->tn_atime;
508 vap->va_atime.tv_nsec = node->tn_atimensec;
509 vap->va_mtime.tv_sec = node->tn_mtime;
510 vap->va_mtime.tv_nsec = node->tn_mtimensec;
511 vap->va_ctime.tv_sec = node->tn_ctime;
512 vap->va_ctime.tv_nsec = node->tn_ctimensec;
513 vap->va_gen = node->tn_gen;
514 vap->va_flags = node->tn_flags;
515 if (vp->v_type == VBLK || vp->v_type == VCHR)
516 {
517 vap->va_rmajor = umajor(node->tn_rdev);
518 vap->va_rminor = uminor(node->tn_rdev);
519 }
520 vap->va_bytes = round_page(node->tn_size);
521 vap->va_filerev = 0;
522
523 lwkt_reltoken(&vp->v_mount->mnt_token);
524
525 return 0;
526}
527
528/* --------------------------------------------------------------------- */
529
530int
531tmpfs_setattr(struct vop_setattr_args *v)
532{
533 struct vnode *vp = v->a_vp;
534 struct vattr *vap = v->a_vap;
535 struct ucred *cred = v->a_cred;
536 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
537 int error = 0;
538 int kflags = 0;
539
540 if (error == 0 && (vap->va_flags != VNOVAL)) {
541 error = tmpfs_chflags(vp, vap->va_flags, cred);
542 kflags |= NOTE_ATTRIB;
543 }
544
545 if (error == 0 && (vap->va_size != VNOVAL)) {
546 if (vap->va_size > node->tn_size)
547 kflags |= NOTE_WRITE | NOTE_EXTEND;
548 else
549 kflags |= NOTE_WRITE;
550 error = tmpfs_chsize(vp, vap->va_size, cred);
551 }
552
553 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
554 vap->va_gid != (gid_t)VNOVAL)) {
555 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
556 kflags |= NOTE_ATTRIB;
557 }
558
559 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
560 error = tmpfs_chmod(vp, vap->va_mode, cred);
561 kflags |= NOTE_ATTRIB;
562 }
563
564 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
565 vap->va_atime.tv_nsec != VNOVAL) ||
566 (vap->va_mtime.tv_sec != VNOVAL &&
567 vap->va_mtime.tv_nsec != VNOVAL) )) {
568 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
569 vap->va_vaflags, cred);
570 kflags |= NOTE_ATTRIB;
571 }
572
573 /* Update the node times. We give preference to the error codes
574 * generated by this function rather than the ones that may arise
575 * from tmpfs_update. */
576 tmpfs_update(vp);
577 tmpfs_knote(vp, kflags);
578
579 return error;
580}
581
582/* --------------------------------------------------------------------- */
583
584/*
585 * fsync is usually a NOP, but we must take action when unmounting or
586 * when recycling.
587 */
588static int
589tmpfs_fsync(struct vop_fsync_args *v)
590{
591 struct hammer2_mount *tmp;
592 struct tmpfs_node *node;
593 struct vnode *vp = v->a_vp;
594
595 tmp = VFS_TO_TMPFS(vp->v_mount);
596 node = VP_TO_TMPFS_NODE(vp);
597
598 tmpfs_update(vp);
599 if (vp->v_type == VREG) {
600 if (vp->v_flag & VRECLAIMED) {
601 if (node->tn_links == 0)
602 tmpfs_truncate(vp, 0);
603 else
604 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
605 }
606 }
607 return 0;
608}
609
610/* --------------------------------------------------------------------- */
611
612static int
613tmpfs_read (struct vop_read_args *ap)
614{
615 struct buf *bp;
616 struct vnode *vp = ap->a_vp;
617 struct uio *uio = ap->a_uio;
618 struct tmpfs_node *node;
619 off_t base_offset;
620 size_t offset;
621 size_t len;
622 int error;
623
624 error = 0;
625 if (uio->uio_resid == 0) {
626 return error;
627 }
628
629 node = VP_TO_TMPFS_NODE(vp);
630
631 if (uio->uio_offset < 0)
632 return (EINVAL);
633 if (vp->v_type != VREG)
634 return (EINVAL);
635
636 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
637 /*
638 * Use buffer cache I/O (via tmpfs_strategy)
639 */
640 offset = (size_t)uio->uio_offset & BMASK;
641 base_offset = (off_t)uio->uio_offset - offset;
642 bp = getcacheblk(vp, base_offset, BSIZE);
643 if (bp == NULL)
644 {
645 lwkt_gettoken(&vp->v_mount->mnt_token);
646 error = bread(vp, base_offset, BSIZE, &bp);
647 if (error) {
648 brelse(bp);
649 lwkt_reltoken(&vp->v_mount->mnt_token);
650 kprintf("tmpfs_read bread error %d\n", error);
651 break;
652 }
653 lwkt_reltoken(&vp->v_mount->mnt_token);
654 }
655
656 /*
657 * Figure out how many bytes we can actually copy this loop.
658 */
659 len = BSIZE - offset;
660 if (len > uio->uio_resid)
661 len = uio->uio_resid;
662 if (len > node->tn_size - uio->uio_offset)
663 len = (size_t)(node->tn_size - uio->uio_offset);
664
665 error = uiomove((char *)bp->b_data + offset, len, uio);
666 bqrelse(bp);
667 if (error) {
668 kprintf("tmpfs_read uiomove error %d\n", error);
669 break;
670 }
671 }
672
673 TMPFS_NODE_LOCK(node);
674 node->tn_status |= TMPFS_NODE_ACCESSED;
675 TMPFS_NODE_UNLOCK(node);
676
677 return(error);
678}
679
680static int
681hammer2_write(struct vop_write_args *ap)
682{
683 struct vnode *vp = ap->a_vp;
684 struct uio *uio = ap->a_uio;
685 int rc;
686
687
688 return (rc);
689}
690
691static int
692tmpfs_write (struct vop_write_args *ap)
693{
694 struct buf *bp;
695 struct vnode *vp = ap->a_vp;
696 struct uio *uio = ap->a_uio;
697 struct thread *td = uio->uio_td;
698 struct tmpfs_node *node;
699 boolean_t extended;
700 off_t oldsize;
701 int error;
702 off_t base_offset;
703 size_t offset;
704 size_t len;
705 struct rlimit limit;
706 int trivial = 0;
707 int kflags = 0;
708
709 error = 0;
710 if (uio->uio_resid == 0) {
711 return error;
712 }
713
714 node = VP_TO_TMPFS_NODE(vp);
715
716 if (vp->v_type != VREG)
717 return (EINVAL);
718
719 lwkt_gettoken(&vp->v_mount->mnt_token);
720
721 oldsize = node->tn_size;
722 if (ap->a_ioflag & IO_APPEND)
723 uio->uio_offset = node->tn_size;
724
725 /*
726 * Check for illegal write offsets.
727 */
728 if (uio->uio_offset + uio->uio_resid >
729 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
730 lwkt_reltoken(&vp->v_mount->mnt_token);
731 return (EFBIG);
732 }
733
734 if (vp->v_type == VREG && td != NULL) {
735 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
736 if (error != 0) {
737 lwkt_reltoken(&vp->v_mount->mnt_token);
738 return error;
739 }
740 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
741 ksignal(td->td_proc, SIGXFSZ);
742 lwkt_reltoken(&vp->v_mount->mnt_token);
743 return (EFBIG);
744 }
745 }
746
747
748 /*
749 * Extend the file's size if necessary
750 */
751 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
752
753 while (uio->uio_resid > 0) {
754 /*
755 * Use buffer cache I/O (via tmpfs_strategy)
756 */
757 offset = (size_t)uio->uio_offset & BMASK;
758 base_offset = (off_t)uio->uio_offset - offset;
759 len = BSIZE - offset;
760 if (len > uio->uio_resid)
761 len = uio->uio_resid;
762
763 if ((uio->uio_offset + len) > node->tn_size) {
764 trivial = (uio->uio_offset <= node->tn_size);
765 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
766 if (error)
767 break;
768 }
769
770 /*
771 * Read to fill in any gaps. Theoretically we could
772 * optimize this if the write covers the entire buffer
773 * and is not a UIO_NOCOPY write, however this can lead
774 * to a security violation exposing random kernel memory
775 * (whatever junk was in the backing VM pages before).
776 *
777 * So just use bread() to do the right thing.
778 */
779 error = bread(vp, base_offset, BSIZE, &bp);
780 error = uiomove((char *)bp->b_data + offset, len, uio);
781 if (error) {
782 kprintf("tmpfs_write uiomove error %d\n", error);
783 brelse(bp);
784 break;
785 }
786
787 if (uio->uio_offset > node->tn_size) {
788 node->tn_size = uio->uio_offset;
789 kflags |= NOTE_EXTEND;
790 }
791 kflags |= NOTE_WRITE;
792
793 /*
794 * The data has been loaded into the buffer, write it out.
795 *
796 * We want tmpfs to be able to use all available ram, not
797 * just the buffer cache, so if not explicitly paging we
798 * use buwrite() to leave the buffer clean but mark all the
799 * VM pages valid+dirty.
800 *
801 * When the kernel is paging, either via normal pageout
802 * operation or when cleaning the object during a recycle,
803 * the underlying VM pages are going to get thrown away
804 * so we MUST write them to swap.
805 *
806 * XXX unfortunately this catches msync() system calls too
807 * for the moment.
808 */
809 if (vm_swap_size == 0) {
810 /*
811 * if swap isn't configured yet, force a buwrite() to
812 * avoid problems further down the line, due to flushing
813 * to swap.
814 */
815 buwrite(bp);
816 } else {
817 if (ap->a_ioflag & IO_SYNC) {
818 bwrite(bp);
819 } else if ((ap->a_ioflag & IO_ASYNC) ||
820 (uio->uio_segflg == UIO_NOCOPY)) {
821 bawrite(bp);
822 } else {
823 buwrite(bp);
824 }
825 }
826
827 if (bp->b_error) {
828 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
829 break;
830 }
831 }
832
833 if (error) {
834 if (extended) {
835 (void)tmpfs_reg_resize(vp, oldsize, trivial);
836 kflags &= ~NOTE_EXTEND;
837 }
838 goto done;
839 }
840
841 TMPFS_NODE_LOCK(node);
842 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
843 (extended? TMPFS_NODE_CHANGED : 0);
844
845 if (node->tn_mode & (S_ISUID | S_ISGID)) {
846 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
847 node->tn_mode &= ~(S_ISUID | S_ISGID);
848 }
849 TMPFS_NODE_UNLOCK(node);
850done:
851
852 tmpfs_knote(vp, kflags);
853
854
855 lwkt_reltoken(&vp->v_mount->mnt_token);
856 return(error);
857}
858
859static int
860tmpfs_advlock (struct vop_advlock_args *ap)
861{
862 struct tmpfs_node *node;
863 struct vnode *vp = ap->a_vp;
864
865 node = VP_TO_TMPFS_NODE(vp);
866
867 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
868}
869
870static int
871tmpfs_strategy(struct vop_strategy_args *ap)
872{
873 struct bio *bio = ap->a_bio;
874 struct bio *nbio;
875 struct buf *bp = bio->bio_buf;
876 struct vnode *vp = ap->a_vp;
877 struct tmpfs_node *node;
878 vm_object_t uobj;
879 vm_page_t m;
880 int i;
881
882 if (vp->v_type != VREG) {
883 bp->b_resid = bp->b_bcount;
884 bp->b_flags |= B_ERROR | B_INVAL;
885 bp->b_error = EINVAL;
886 biodone(bio);
887 return(0);
888 }
889
890 lwkt_gettoken(&vp->v_mount->mnt_token);
891 node = VP_TO_TMPFS_NODE(vp);
892
893 uobj = node->tn_reg.tn_aobj;
894
895 /*
896 * Certain operations such as nvtruncbuf() can result in a
897 * bdwrite() of one or more buffers related to the file,
898 * leading to the possibility of our strategy function
899 * being called for writing even when there is no swap space.
900 *
901 * When this case occurs we mark the underlying pages as valid
902 * and dirty and complete the I/O manually.
903 *
904 * Otherwise just call swap_pager_strategy to read or write,
905 * potentially assigning swap on write. We push a BIO to catch
906 * any swap allocation errors.
907 */
908 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
909 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
910 m = bp->b_xio.xio_pages[i];
911 vm_page_set_validdirty(m, 0, PAGE_SIZE);
912 }
913 bp->b_resid = 0;
914 bp->b_error = 0;
915 biodone(bio);
916 } else {
917 nbio = push_bio(bio);
918 nbio->bio_done = tmpfs_strategy_done;
919 nbio->bio_offset = bio->bio_offset;
920 swap_pager_strategy(uobj, nbio);
921 }
922
923 lwkt_reltoken(&vp->v_mount->mnt_token);
924 return 0;
925}
926
927/*
928 * bio finished. If we ran out of sap just mark the pages valid
929 * and dirty and make it appear that the I/O has completed successfully.
930 */
931static void
932tmpfs_strategy_done(struct bio *bio)
933{
934 struct buf *bp;
935 vm_page_t m;
936 int i;
937
938 bp = bio->bio_buf;
939
940 if ((bp->b_flags & B_ERROR) && bp->b_error == ENOMEM) {
941 bp->b_flags &= ~B_ERROR;
942 bp->b_error = 0;
943 bp->b_resid = 0;
944 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
945 m = bp->b_xio.xio_pages[i];
946 vm_page_set_validdirty(m, 0, PAGE_SIZE);
947 }
948 }
949 bio = pop_bio(bio);
950 biodone(bio);
951}
952
953static int
954tmpfs_bmap(struct vop_bmap_args *ap)
955{
956 if (ap->a_doffsetp != NULL)
957 *ap->a_doffsetp = ap->a_loffset;
958 if (ap->a_runp != NULL)
959 *ap->a_runp = 0;
960 if (ap->a_runb != NULL)
961 *ap->a_runb = 0;
962
963 return 0;
964}
965
966/* --------------------------------------------------------------------- */
967
968static int
969tmpfs_nremove(struct vop_nremove_args *v)
970{
971 struct vnode *dvp = v->a_dvp;
972 struct namecache *ncp = v->a_nch->ncp;
973 struct vnode *vp;
974 int error;
975 struct tmpfs_dirent *de;
976 struct hammer2_mount *tmp;
977 struct tmpfs_node *dnode;
978 struct tmpfs_node *node;
979
980 /*
981 * We have to acquire the vp from v->a_nch because we will likely
982 * unresolve the namecache entry, and a vrele/vput is needed to
983 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
984 *
985 * We have to use vget to clear any inactive state on the vnode,
986 * otherwise the vnode may remain inactive and thus tmpfs_inactive
987 * will not get called when we release it.
988 */
989 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
990 KKASSERT(error == 0);
991 vn_unlock(vp);
992
993 if (vp->v_type == VDIR) {
994 error = EISDIR;
995 goto out;
996 }
997
998 dnode = VP_TO_TMPFS_DIR(dvp);
999 node = VP_TO_TMPFS_NODE(vp);
1000 tmp = VFS_TO_TMPFS(vp->v_mount);
1001 de = tmpfs_dir_lookup(dnode, node, ncp);
1002 if (de == NULL) {
1003 error = ENOENT;
1004 goto out;
1005 }
1006
1007 /* Files marked as immutable or append-only cannot be deleted. */
1008 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
1009 (dnode->tn_flags & APPEND)) {
1010 error = EPERM;
1011 goto out;
1012 }
1013
1014 /* Remove the entry from the directory; as it is a file, we do not
1015 * have to change the number of hard links of the directory. */
1016 tmpfs_dir_detach(dnode, de);
1017
1018 /* Free the directory entry we just deleted. Note that the node
1019 * referred by it will not be removed until the vnode is really
1020 * reclaimed. */
1021 tmpfs_free_dirent(tmp, de);
1022
1023 if (node->tn_links > 0) {
1024 TMPFS_NODE_LOCK(node);
1025 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1026 TMPFS_NODE_MODIFIED;
1027 TMPFS_NODE_UNLOCK(node);
1028 }
1029
1030 cache_setunresolved(v->a_nch);
1031 cache_setvp(v->a_nch, NULL);
1032 tmpfs_knote(vp, NOTE_DELETE);
1033 /*cache_inval_vp(vp, CINV_DESTROY);*/
1034 tmpfs_knote(dvp, NOTE_WRITE);
1035 error = 0;
1036
1037out:
1038 vrele(vp);
1039
1040 return error;
1041}
1042
1043/* --------------------------------------------------------------------- */
1044
1045static int
1046tmpfs_nlink(struct vop_nlink_args *v)
1047{
1048 struct vnode *dvp = v->a_dvp;
1049 struct vnode *vp = v->a_vp;
1050 struct namecache *ncp = v->a_nch->ncp;
1051 struct tmpfs_dirent *de;
1052 struct tmpfs_node *node;
1053 struct tmpfs_node *dnode;
1054 int error;
1055
1056 KKASSERT(dvp != vp); /* XXX When can this be false? */
1057
1058 node = VP_TO_TMPFS_NODE(vp);
1059 dnode = VP_TO_TMPFS_NODE(dvp);
1060
1061 /* XXX: Why aren't the following two tests done by the caller? */
1062
1063 /* Hard links of directories are forbidden. */
1064 if (vp->v_type == VDIR) {
1065 error = EPERM;
1066 goto out;
1067 }
1068
1069 /* Cannot create cross-device links. */
1070 if (dvp->v_mount != vp->v_mount) {
1071 error = EXDEV;
1072 goto out;
1073 }
1074
1075 /* Ensure that we do not overflow the maximum number of links imposed
1076 * by the system. */
1077 KKASSERT(node->tn_links <= LINK_MAX);
1078 if (node->tn_links == LINK_MAX) {
1079 error = EMLINK;
1080 goto out;
1081 }
1082
1083 /* We cannot create links of files marked immutable or append-only. */
1084 if (node->tn_flags & (IMMUTABLE | APPEND)) {
1085 error = EPERM;
1086 goto out;
1087 }
1088
1089 /* Allocate a new directory entry to represent the node. */
1090 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
1091 ncp->nc_name, ncp->nc_nlen, &de);
1092 if (error != 0)
1093 goto out;
1094
1095 /* Insert the new directory entry into the appropriate directory. */
1096 tmpfs_dir_attach(dnode, de);
1097
1098 /* vp link count has changed, so update node times. */
1099
1100 TMPFS_NODE_LOCK(node);
1101 node->tn_status |= TMPFS_NODE_CHANGED;
1102 TMPFS_NODE_UNLOCK(node);
1103 tmpfs_update(vp);
1104
1105 tmpfs_knote(vp, NOTE_LINK);
1106 cache_setunresolved(v->a_nch);
1107 cache_setvp(v->a_nch, vp);
1108 tmpfs_knote(dvp, NOTE_WRITE);
1109 error = 0;
1110
1111out:
1112 return error;
1113}
1114
1115/* --------------------------------------------------------------------- */
1116
1117static int
1118tmpfs_nrename(struct vop_nrename_args *v)
1119{
1120 struct vnode *fdvp = v->a_fdvp;
1121 struct namecache *fncp = v->a_fnch->ncp;
1122 struct vnode *fvp = fncp->nc_vp;
1123 struct vnode *tdvp = v->a_tdvp;
1124 struct namecache *tncp = v->a_tnch->ncp;
1125 struct vnode *tvp;
1126 struct tmpfs_dirent *de;
1127 struct hammer2_mount *tmp;
1128 struct tmpfs_node *fdnode;
1129 struct tmpfs_node *fnode;
1130 struct tmpfs_node *tnode;
1131 struct tmpfs_node *tdnode;
1132 char *newname;
1133 char *oldname;
1134 int error;
1135
1136 /*
1137 * Because tvp can get overwritten we have to vget it instead of
1138 * just vref or use it, otherwise it's VINACTIVE flag may not get
1139 * cleared and the node won't get destroyed.
1140 */
1141 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
1142 if (error == 0) {
1143 tnode = VP_TO_TMPFS_NODE(tvp);
1144 vn_unlock(tvp);
1145 } else {
1146 tnode = NULL;
1147 }
1148
1149 /* Disallow cross-device renames.
1150 * XXX Why isn't this done by the caller? */
1151 if (fvp->v_mount != tdvp->v_mount ||
1152 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1153 error = EXDEV;
1154 goto out;
1155 }
1156
1157 tmp = VFS_TO_TMPFS(tdvp->v_mount);
1158 tdnode = VP_TO_TMPFS_DIR(tdvp);
1159
1160 /* If source and target are the same file, there is nothing to do. */
1161 if (fvp == tvp) {
1162 error = 0;
1163 goto out;
1164 }
1165
1166 fdnode = VP_TO_TMPFS_DIR(fdvp);
1167 fnode = VP_TO_TMPFS_NODE(fvp);
1168 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
1169
1170 /* Avoid manipulating '.' and '..' entries. */
1171 if (de == NULL) {
1172 error = ENOENT;
1173 goto out_locked;
1174 }
1175 KKASSERT(de->td_node == fnode);
1176
1177 /*
1178 * If replacing an entry in the target directory and that entry
1179 * is a directory, it must be empty.
1180 *
1181 * Kern_rename gurantees the destination to be a directory
1182 * if the source is one (it does?).
1183 */
1184 if (tvp != NULL) {
1185 KKASSERT(tnode != NULL);
1186
1187 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1188 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1189 error = EPERM;
1190 goto out_locked;
1191 }
1192
1193 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1194 if (tnode->tn_size > 0) {
1195 error = ENOTEMPTY;
1196 goto out_locked;
1197 }
1198 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1199 error = ENOTDIR;
1200 goto out_locked;
1201 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1202 error = EISDIR;
1203 goto out_locked;
1204 } else {
1205 KKASSERT(fnode->tn_type != VDIR &&
1206 tnode->tn_type != VDIR);
1207 }
1208 }
1209
1210 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1211 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1212 error = EPERM;
1213 goto out_locked;
1214 }
1215
1216 /*
1217 * Ensure that we have enough memory to hold the new name, if it
1218 * has to be changed.
1219 */
1220 if (fncp->nc_nlen != tncp->nc_nlen ||
1221 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
1222 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
1223 M_WAITOK | M_NULLOK);
1224 if (newname == NULL) {
1225 error = ENOSPC;
1226 goto out_locked;
1227 }
1228 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1229 newname[tncp->nc_nlen] = '\0';
1230 } else {
1231 newname = NULL;
1232 }
1233
1234 /*
1235 * Unlink entry from source directory. Note that the kernel has
1236 * already checked for illegal recursion cases (renaming a directory
1237 * into a subdirectory of itself).
1238 */
1239 if (fdnode != tdnode)
1240 tmpfs_dir_detach(fdnode, de);
1241
1242 /*
1243 * Handle any name change. Swap with newname, we will
1244 * deallocate it at the end.
1245 */
1246 if (newname != NULL) {
1247#if 0
1248 TMPFS_NODE_LOCK(fnode);
1249 fnode->tn_status |= TMPFS_NODE_CHANGED;
1250 TMPFS_NODE_UNLOCK(fnode);
1251#endif
1252 oldname = de->td_name;
1253 de->td_name = newname;
1254 de->td_namelen = (uint16_t)tncp->nc_nlen;
1255 newname = oldname;
1256 }
1257
1258 /*
1259 * Link entry to target directory. If the entry
1260 * represents a directory move the parent linkage
1261 * as well.
1262 */
1263 if (fdnode != tdnode) {
1264 if (de->td_node->tn_type == VDIR) {
1265 TMPFS_VALIDATE_DIR(fnode);
1266
1267 TMPFS_NODE_LOCK(tdnode);
1268 tdnode->tn_links++;
1269 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1270 TMPFS_NODE_UNLOCK(tdnode);
1271
1272 TMPFS_NODE_LOCK(fnode);
1273 fnode->tn_dir.tn_parent = tdnode;
1274 fnode->tn_status |= TMPFS_NODE_CHANGED;
1275 TMPFS_NODE_UNLOCK(fnode);
1276
1277 TMPFS_NODE_LOCK(fdnode);
1278 fdnode->tn_links--;
1279 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1280 TMPFS_NODE_UNLOCK(fdnode);
1281 }
1282 tmpfs_dir_attach(tdnode, de);
1283 } else {
1284 TMPFS_NODE_LOCK(tdnode);
1285 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1286 TMPFS_NODE_UNLOCK(tdnode);
1287 }
1288
1289 /*
1290 * If we are overwriting an entry, we have to remove the old one
1291 * from the target directory.
1292 */
1293 if (tvp != NULL) {
1294 /* Remove the old entry from the target directory. */
1295 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1296 tmpfs_dir_detach(tdnode, de);
1297 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1298
1299 /*
1300 * Free the directory entry we just deleted. Note that the
1301 * node referred by it will not be removed until the vnode is
1302 * really reclaimed.
1303 */
1304 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1305 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1306 }
1307
1308 /*
1309 * Finish up
1310 */
1311 if (newname) {
1312 kfree(newname, tmp->tm_name_zone);
1313 newname = NULL;
1314 }
1315 cache_rename(v->a_fnch, v->a_tnch);
1316 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1317 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1318 if (fnode->tn_vnode)
1319 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1320 error = 0;
1321
1322out_locked:
1323 ;
1324
1325out:
1326 if (tvp)
1327 vrele(tvp);
1328
1329 return error;
1330}
1331
1332/* --------------------------------------------------------------------- */
1333
1334static int
1335tmpfs_nmkdir(struct vop_nmkdir_args *v)
1336{
1337 struct vnode *dvp = v->a_dvp;
1338 struct vnode **vpp = v->a_vpp;
1339 struct namecache *ncp = v->a_nch->ncp;
1340 struct vattr *vap = v->a_vap;
1341 struct ucred *cred = v->a_cred;
1342 int error;
1343
1344 KKASSERT(vap->va_type == VDIR);
1345
1346 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1347 if (error == 0) {
1348 cache_setunresolved(v->a_nch);
1349 cache_setvp(v->a_nch, *vpp);
1350 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1351 }
1352
1353 return error;
1354}
1355
1356/* --------------------------------------------------------------------- */
1357
1358static int
1359tmpfs_nrmdir(struct vop_nrmdir_args *v)
1360{
1361 struct vnode *dvp = v->a_dvp;
1362 struct namecache *ncp = v->a_nch->ncp;
1363 struct vnode *vp;
1364 struct tmpfs_dirent *de;
1365 struct hammer2_mount *tmp;
1366 struct tmpfs_node *dnode;
1367 struct tmpfs_node *node;
1368 int error;
1369
1370 /*
1371 * We have to acquire the vp from v->a_nch because we will likely
1372 * unresolve the namecache entry, and a vrele/vput is needed to
1373 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1374 *
1375 * We have to use vget to clear any inactive state on the vnode,
1376 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1377 * will not get called when we release it.
1378 */
1379 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
1380 KKASSERT(error == 0);
1381 vn_unlock(vp);
1382
1383 /*
1384 * Prevalidate so we don't hit an assertion later
1385 */
1386 if (vp->v_type != VDIR) {
1387 error = ENOTDIR;
1388 goto out;
1389 }
1390
1391 tmp = VFS_TO_TMPFS(dvp->v_mount);
1392 dnode = VP_TO_TMPFS_DIR(dvp);
1393 node = VP_TO_TMPFS_DIR(vp);
1394
1395 /* Directories with more than two entries ('.' and '..') cannot be
1396 * removed. */
1397 if (node->tn_size > 0) {
1398 error = ENOTEMPTY;
1399 goto out;
1400 }
1401
1402 if ((dnode->tn_flags & APPEND)
1403 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1404 error = EPERM;
1405 goto out;
1406 }
1407
1408 /* This invariant holds only if we are not trying to remove "..".
1409 * We checked for that above so this is safe now. */
1410 KKASSERT(node->tn_dir.tn_parent == dnode);
1411
1412 /* Get the directory entry associated with node (vp). This was
1413 * filled by tmpfs_lookup while looking up the entry. */
1414 de = tmpfs_dir_lookup(dnode, node, ncp);
1415 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1416 ncp->nc_name,
1417 ncp->nc_nlen));
1418
1419 /* Check flags to see if we are allowed to remove the directory. */
1420 if ((dnode->tn_flags & APPEND) ||
1421 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1422 error = EPERM;
1423 goto out;
1424 }
1425
1426
1427 /* Detach the directory entry from the directory (dnode). */
1428 tmpfs_dir_detach(dnode, de);
1429
1430 /* No vnode should be allocated for this entry from this point */
1431 TMPFS_NODE_LOCK(node);
1432 TMPFS_ASSERT_ELOCKED(node);
1433 TMPFS_NODE_LOCK(dnode);
1434 TMPFS_ASSERT_ELOCKED(dnode);
1435
1436#if 0
1437 /* handled by tmpfs_free_node */
1438 KKASSERT(node->tn_links > 0);
1439 node->tn_links--;
1440 node->tn_dir.tn_parent = NULL;
1441#endif
1442 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1443 TMPFS_NODE_MODIFIED;
1444
1445#if 0
1446 /* handled by tmpfs_free_node */
1447 KKASSERT(dnode->tn_links > 0);
1448 dnode->tn_links--;
1449#endif
1450 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1451 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1452
1453 TMPFS_NODE_UNLOCK(dnode);
1454 TMPFS_NODE_UNLOCK(node);
1455
1456 /* Free the directory entry we just deleted. Note that the node
1457 * referred by it will not be removed until the vnode is really
1458 * reclaimed. */
1459 tmpfs_free_dirent(tmp, de);
1460
1461 /* Release the deleted vnode (will destroy the node, notify
1462 * interested parties and clean it from the cache). */
1463
1464 TMPFS_NODE_LOCK(dnode);
1465 dnode->tn_status |= TMPFS_NODE_CHANGED;
1466 TMPFS_NODE_UNLOCK(dnode);
1467 tmpfs_update(dvp);
1468
1469 cache_setunresolved(v->a_nch);
1470 cache_setvp(v->a_nch, NULL);
1471 /*cache_inval_vp(vp, CINV_DESTROY);*/
1472 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1473 error = 0;
1474
1475out:
1476 vrele(vp);
1477
1478 return error;
1479}
1480
1481/* --------------------------------------------------------------------- */
1482
1483static int
1484tmpfs_nsymlink(struct vop_nsymlink_args *v)
1485{
1486 struct vnode *dvp = v->a_dvp;
1487 struct vnode **vpp = v->a_vpp;
1488 struct namecache *ncp = v->a_nch->ncp;
1489 struct vattr *vap = v->a_vap;
1490 struct ucred *cred = v->a_cred;
1491 char *target = v->a_target;
1492 int error;
1493
1494 vap->va_type = VLNK;
1495 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1496 if (error == 0) {
1497 tmpfs_knote(*vpp, NOTE_WRITE);
1498 cache_setunresolved(v->a_nch);
1499 cache_setvp(v->a_nch, *vpp);
1500 }
1501
1502 return error;
1503}
1504
1505/* --------------------------------------------------------------------- */
1506
1507static int
1508tmpfs_readdir(struct vop_readdir_args *v)
1509{
1510 struct vnode *vp = v->a_vp;
1511 struct uio *uio = v->a_uio;
1512 int *eofflag = v->a_eofflag;
1513 off_t **cookies = v->a_cookies;
1514 int *ncookies = v->a_ncookies;
1515 struct hammer2_mount *tmp;
1516 int error;
1517 off_t startoff;
1518 off_t cnt = 0;
1519 struct tmpfs_node *node;
1520
1521 /* This operation only makes sense on directory nodes. */
1522 if (vp->v_type != VDIR)
1523 return ENOTDIR;
1524
1525 tmp = VFS_TO_TMPFS(vp->v_mount);
1526 node = VP_TO_TMPFS_DIR(vp);
1527 startoff = uio->uio_offset;
1528
1529 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1530 error = tmpfs_dir_getdotdent(node, uio);
1531 if (error != 0)
1532 goto outok;
1533 cnt++;
1534 }
1535
1536 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1537 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1538 if (error != 0)
1539 goto outok;
1540 cnt++;
1541 }
1542
1543 error = tmpfs_dir_getdents(node, uio, &cnt);
1544
1545outok:
1546 KKASSERT(error >= -1);
1547
1548 if (error == -1)
1549 error = 0;
1550
1551 if (eofflag != NULL)
1552 *eofflag =
1553 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1554
1555 /* Update NFS-related variables. */
1556 if (error == 0 && cookies != NULL && ncookies != NULL) {
1557 off_t i;
1558 off_t off = startoff;
1559 struct tmpfs_dirent *de = NULL;
1560
1561 *ncookies = cnt;
1562 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1563
1564 for (i = 0; i < cnt; i++) {
1565 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1566 if (off == TMPFS_DIRCOOKIE_DOT) {
1567 off = TMPFS_DIRCOOKIE_DOTDOT;
1568 } else {
1569 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1570 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1571 } else if (de != NULL) {
1572 de = TAILQ_NEXT(de, td_entries);
1573 } else {
1574 de = tmpfs_dir_lookupbycookie(node,
1575 off);
1576 KKASSERT(de != NULL);
1577 de = TAILQ_NEXT(de, td_entries);
1578 }
1579 if (de == NULL)
1580 off = TMPFS_DIRCOOKIE_EOF;
1581 else
1582 off = tmpfs_dircookie(de);
1583 }
1584
1585 (*cookies)[i] = off;
1586 }
1587 KKASSERT(uio->uio_offset == off);
1588 }
1589
1590 return error;
1591}
1592
1593/* --------------------------------------------------------------------- */
1594
1595static int
1596tmpfs_readlink(struct vop_readlink_args *v)
1597{
1598 struct vnode *vp = v->a_vp;
1599 struct uio *uio = v->a_uio;
1600
1601 int error;
1602 struct tmpfs_node *node;
1603
1604 KKASSERT(uio->uio_offset == 0);
1605 KKASSERT(vp->v_type == VLNK);
1606
1607 node = VP_TO_TMPFS_NODE(vp);
1608
1609 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1610 uio);
1611 TMPFS_NODE_LOCK(node);
1612 node->tn_status |= TMPFS_NODE_ACCESSED;
1613 TMPFS_NODE_UNLOCK(node);
1614
1615 return error;
1616}
1617
1618/* --------------------------------------------------------------------- */
1619
1620static int
1621tmpfs_inactive(struct vop_inactive_args *v)
1622{
1623 struct vnode *vp = v->a_vp;
1624 struct tmpfs_node *node;
1625
1626 node = VP_TO_TMPFS_NODE(vp);
1627
1628 /*
1629 * Degenerate case
1630 */
1631 if (node == NULL) {
1632 vrecycle(vp);
1633 return(0);
1634 }
1635
1636 /*
1637 * Get rid of unreferenced deleted vnodes sooner rather than
1638 * later so the data memory can be recovered immediately.
1639 *
1640 * We must truncate the vnode to prevent the normal reclamation
1641 * path from flushing the data for the removed file to disk.
1642 */
1643 TMPFS_NODE_LOCK(node);
1644 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1645 (node->tn_links == 0 ||
1646 (node->tn_links == 1 && node->tn_type == VDIR &&
1647 node->tn_dir.tn_parent)))
1648 {
1649 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1650 TMPFS_NODE_UNLOCK(node);
1651 if (node->tn_type == VREG)
1652 tmpfs_truncate(vp, 0);
1653 vrecycle(vp);
1654 } else {
1655 TMPFS_NODE_UNLOCK(node);
1656 }
1657
1658 return 0;
1659}
1660
1661/* --------------------------------------------------------------------- */
1662
1663int
1664tmpfs_reclaim(struct vop_reclaim_args *v)
1665{
1666 struct vnode *vp = v->a_vp;
1667 struct hammer2_mount *tmp;
1668 struct tmpfs_node *node;
1669
1670 node = VP_TO_TMPFS_NODE(vp);
1671 tmp = VFS_TO_TMPFS(vp->v_mount);
1672
1673 tmpfs_free_vp(vp);
1674
1675 /*
1676 * If the node referenced by this vnode was deleted by the
1677 * user, we must free its associated data structures now that
1678 * the vnode is being reclaimed.
1679 *
1680 * Directories have an extra link ref.
1681 */
1682 TMPFS_NODE_LOCK(node);
1683 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1684 (node->tn_links == 0 ||
1685 (node->tn_links == 1 && node->tn_type == VDIR &&
1686 node->tn_dir.tn_parent)))
1687 {
1688 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1689 tmpfs_free_node(tmp, node);
1690 /* eats the lock */
1691 } else {
1692 TMPFS_NODE_UNLOCK(node);
1693 }
1694
1695 KKASSERT(vp->v_data == NULL);
1696 return 0;
1697}
1698
1699/* --------------------------------------------------------------------- */
1700
1701static int
1702tmpfs_print(struct vop_print_args *v)
1703{
1704 struct vnode *vp = v->a_vp;
1705
1706 struct tmpfs_node *node;
1707
1708 node = VP_TO_TMPFS_NODE(vp);
1709
1710 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1711 node, node->tn_flags, node->tn_links);
1712 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1713 node->tn_mode, node->tn_uid, node->tn_gid,
1714 (uintmax_t)node->tn_size, node->tn_status);
1715
1716 if (vp->v_type == VFIFO)
1717 fifo_printinfo(vp);
1718
1719 kprintf("\n");
1720
1721 return 0;
1722}
1723
1724/* --------------------------------------------------------------------- */
1725
1726static int
1727tmpfs_pathconf(struct vop_pathconf_args *v)
1728{
1729 int name = v->a_name;
1730 register_t *retval = v->a_retval;
1731
1732 int error;
1733
1734 error = 0;
1735
1736 switch (name) {
1737 case _PC_LINK_MAX:
1738 *retval = LINK_MAX;
1739 break;
1740
1741 case _PC_NAME_MAX:
1742 *retval = NAME_MAX;
1743 break;
1744
1745 case _PC_PATH_MAX:
1746 *retval = PATH_MAX;
1747 break;
1748
1749 case _PC_PIPE_BUF:
1750 *retval = PIPE_BUF;
1751 break;
1752
1753 case _PC_CHOWN_RESTRICTED:
1754 *retval = 1;
1755 break;
1756
1757 case _PC_NO_TRUNC:
1758 *retval = 1;
1759 break;
1760
1761 case _PC_SYNC_IO:
1762 *retval = 1;
1763 break;
1764
1765 case _PC_FILESIZEBITS:
1766 *retval = 0; /* XXX Don't know which value should I return. */
1767 break;
1768
1769 default:
1770 error = EINVAL;
1771 }
1772
1773 return error;
1774}
1775
1776/************************************************************************
1777 * KQFILTER OPS *
1778 ************************************************************************/
1779
1780static void filt_tmpfsdetach(struct knote *kn);
1781static int filt_tmpfsread(struct knote *kn, long hint);
1782static int filt_tmpfswrite(struct knote *kn, long hint);
1783static int filt_tmpfsvnode(struct knote *kn, long hint);
1784
1785static struct filterops tmpfsread_filtops =
1786 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1787static struct filterops tmpfswrite_filtops =
1788 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1789static struct filterops tmpfsvnode_filtops =
1790 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1791
1792static int
1793tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1794{
1795 struct vnode *vp = ap->a_vp;
1796 struct knote *kn = ap->a_kn;
1797
1798 switch (kn->kn_filter) {
1799 case EVFILT_READ:
1800 kn->kn_fop = &tmpfsread_filtops;
1801 break;
1802 case EVFILT_WRITE:
1803 kn->kn_fop = &tmpfswrite_filtops;
1804 break;
1805 case EVFILT_VNODE:
1806 kn->kn_fop = &tmpfsvnode_filtops;
1807 break;
1808 default:
1809 return (EOPNOTSUPP);
1810 }
1811
1812 kn->kn_hook = (caddr_t)vp;
1813
1814 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1815
1816 return(0);
1817}
1818
1819static void
1820filt_tmpfsdetach(struct knote *kn)
1821{
1822 struct vnode *vp = (void *)kn->kn_hook;
1823
1824 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1825}
1826
1827static int
1828filt_tmpfsread(struct knote *kn, long hint)
1829{
1830 struct vnode *vp = (void *)kn->kn_hook;
1831 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1832 off_t off;
1833
1834 if (hint == NOTE_REVOKE) {
1835 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1836 return(1);
1837 }
1838
1839 /*
1840 * Interlock against MP races when performing this function.
1841 */
1842 lwkt_gettoken(&vp->v_mount->mnt_token);
1843 off = node->tn_size - kn->kn_fp->f_offset;
1844 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1845 if (kn->kn_sfflags & NOTE_OLDAPI) {
1846 lwkt_reltoken(&vp->v_mount->mnt_token);
1847 return(1);
1848 }
1849
1850 if (kn->kn_data == 0) {
1851 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1852 }
1853 lwkt_reltoken(&vp->v_mount->mnt_token);
1854 return (kn->kn_data != 0);
1855}
1856
1857static int
1858filt_tmpfswrite(struct knote *kn, long hint)
1859{
1860 if (hint == NOTE_REVOKE)
1861 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
1862 kn->kn_data = 0;
1863 return (1);
1864}
1865
1866static int
1867filt_tmpfsvnode(struct knote *kn, long hint)
1868{
1869 if (kn->kn_sfflags & hint)
1870 kn->kn_fflags |= hint;
1871 if (hint == NOTE_REVOKE) {
1872 kn->kn_flags |= (EV_EOF | EV_NODATA);
1873 return (1);
1874 }
1875 return (kn->kn_fflags != 0);
1876}
1877
1878
1879/* --------------------------------------------------------------------- */
1880
1881/*
1882 * vnode operations vector used for files stored in a tmpfs file system.
1883 */
1884struct vop_ops tmpfs_vnode_vops = {
1885 .vop_default = vop_defaultop,
1886 .vop_getpages = vop_stdgetpages,
1887 .vop_putpages = vop_stdputpages,
1888 .vop_ncreate = tmpfs_ncreate,
1889 .vop_nresolve = tmpfs_nresolve,
1890 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1891 .vop_nmknod = tmpfs_nmknod,
1892 .vop_open = tmpfs_open,
1893 .vop_close = tmpfs_close,
1894 .vop_access = tmpfs_access,
1895 .vop_getattr = tmpfs_getattr,
1896 .vop_setattr = tmpfs_setattr,
1897 .vop_read = tmpfs_read,
1898 .vop_write = tmpfs_write,
1899 .vop_fsync = tmpfs_fsync,
1900 .vop_nremove = tmpfs_nremove,
1901 .vop_nlink = tmpfs_nlink,
1902 .vop_nrename = tmpfs_nrename,
1903 .vop_nmkdir = tmpfs_nmkdir,
1904 .vop_nrmdir = tmpfs_nrmdir,
1905 .vop_nsymlink = tmpfs_nsymlink,
1906 .vop_readdir = tmpfs_readdir,
1907 .vop_readlink = tmpfs_readlink,
1908 .vop_inactive = tmpfs_inactive,
1909 .vop_reclaim = tmpfs_reclaim,
1910 .vop_print = tmpfs_print,
1911 .vop_pathconf = tmpfs_pathconf,
1912 .vop_bmap = tmpfs_bmap,
1913 .vop_strategy = tmpfs_strategy,
1914 .vop_advlock = tmpfs_advlock,
1915 .vop_kqfilter = tmpfs_kqfilter
1916};
1917/* $NetBSD: tmpfs_fifoops.c,v 1.5 2005/12/11 12:24:29 christos Exp $ */
1918
1919/*-
1920 * Copyright (c) 2005 The NetBSD Foundation, Inc.
1921 * All rights reserved.
1922 *
1923 * This code is derived from software contributed to The NetBSD Foundation
1924 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
1925 * 2005 program.
1926 *
1927 * Redistribution and use in source and binary forms, with or without
1928 * modification, are permitted provided that the following conditions
1929 * are met:
1930 * 1. Redistributions of source code must retain the above copyright
1931 * notice, this list of conditions and the following disclaimer.
1932 * 2. Redistributions in binary form must reproduce the above copyright
1933 * notice, this list of conditions and the following disclaimer in the
1934 * documentation and/or other materials provided with the distribution.
1935 *
1936 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
1937 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
1938 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
1939 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
1940 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
1941 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
1942 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
1943 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
1944 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
1945 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
1946 * POSSIBILITY OF SUCH DAMAGE.
1947 */
1948
1949/*
1950 * tmpfs vnode interface for named pipes.
1951 */
1952
1953#include <sys/kernel.h>
1954#include <sys/param.h>
1955#include <sys/filedesc.h>
1956#include <sys/proc.h>
1957#include <sys/vnode.h>
1958
1959#include <vm/vm.h>
1960#include <vm/vm_object.h>
1961
1962#include <vfs/fifofs/fifo.h>
1963#include "hammer2.h"
1964
1965/* --------------------------------------------------------------------- */
1966
1967static int
1968tmpfs_fifo_kqfilter(struct vop_kqfilter_args *ap)
1969{
1970 struct vnode *vp;
1971 struct tmpfs_node *node;
1972
1973 vp = ap->a_vp;
1974 node = VP_TO_TMPFS_NODE(vp);
1975
1976 switch (ap->a_kn->kn_filter){
1977 case EVFILT_READ:
1978 node->tn_status |= TMPFS_NODE_ACCESSED;
1979 break;
1980 case EVFILT_WRITE:
1981 node->tn_status |= TMPFS_NODE_MODIFIED;
1982 break;
1983 }
1984
1985 return fifo_vnode_vops.vop_kqfilter(ap);
1986}
1987
1988/* --------------------------------------------------------------------- */
1989
1990static int
1991tmpfs_fifo_close(struct vop_close_args *v)
1992{
1993 struct tmpfs_node *node;
1994 node = VP_TO_TMPFS_NODE(v->a_vp);
1995 node->tn_status |= TMPFS_NODE_ACCESSED;
1996
1997 tmpfs_update(v->a_vp);
1998 return fifo_vnode_vops.vop_close(v);
1999}
2000
2001/*
2002 * vnode operations vector used for fifos stored in a tmpfs file system.
2003 */
2004struct vop_ops tmpfs_fifo_vops = {
2005 .vop_default = fifo_vnoperate,
2006 .vop_close = tmpfs_fifo_close,
2007 .vop_reclaim = tmpfs_reclaim,
2008 .vop_access = tmpfs_access,
2009 .vop_getattr = tmpfs_getattr,
2010 .vop_setattr = tmpfs_setattr,
2011 .vop_kqfilter = tmpfs_fifo_kqfilter,
2012};