tmpfs - Fix numerous races and adjust to use cache_unlink()
[dragonfly.git] / sys / vfs / tmpfs / tmpfs_vnops.c
CommitLineData
7a2de9a4
MD
1/*-
2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
3 * All rights reserved.
4 *
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
7 * 2005 program.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
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.
17 *
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.
80ae59d7
MD
29 *
30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
7a2de9a4
MD
31 */
32
33/*
34 * tmpfs vnode interface.
35 */
7a2de9a4
MD
36
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>
42#include <sys/priv.h>
43#include <sys/proc.h>
44#include <sys/resourcevar.h>
45#include <sys/sched.h>
7a2de9a4
MD
46#include <sys/stat.h>
47#include <sys/systm.h>
48#include <sys/unistd.h>
49#include <sys/vfsops.h>
50#include <sys/vnode.h>
66fa44e7 51#include <sys/mountctl.h>
7a2de9a4 52
7a2de9a4
MD
53#include <vm/vm.h>
54#include <vm/vm_object.h>
55#include <vm/vm_page.h>
56#include <vm/vm_pager.h>
b7545cb3 57#include <vm/swap_pager.h>
7a2de9a4 58
54341a3b
MD
59#include <sys/buf2.h>
60
7a2de9a4
MD
61#include <vfs/fifofs/fifo.h>
62#include <vfs/tmpfs/tmpfs_vnops.h>
63#include <vfs/tmpfs/tmpfs.h>
64
65MALLOC_DECLARE(M_TMPFS);
66
8f9ba07b
MD
67static void tmpfs_strategy_done(struct bio *bio);
68
80ae59d7
MD
69static __inline
70void
71tmpfs_knote(struct vnode *vp, int flags)
72{
73 if (flags)
74 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
75}
76
77
7a2de9a4
MD
78/* --------------------------------------------------------------------- */
79
80static int
81tmpfs_nresolve(struct vop_nresolve_args *v)
82{
83 struct vnode *dvp = v->a_dvp;
84 struct vnode *vp = NULL;
85 struct namecache *ncp = v->a_nch->ncp;
d89ce96a 86 struct tmpfs_node *tnode;
7a2de9a4
MD
87
88 int error;
89 struct tmpfs_dirent *de;
90 struct tmpfs_node *dnode;
91
92 dnode = VP_TO_TMPFS_DIR(dvp);
93
7a2de9a4
MD
94 de = tmpfs_dir_lookup(dnode, NULL, ncp);
95 if (de == NULL) {
d89ce96a 96 error = ENOENT;
7a2de9a4 97 } else {
d89ce96a
MD
98 /*
99 * Allocate a vnode for the node we found.
100 */
7a2de9a4 101 tnode = de->td_node;
7a2de9a4 102 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
d89ce96a
MD
103 LK_EXCLUSIVE | LK_RETRY, &vp);
104 if (error)
7a2de9a4 105 goto out;
d89ce96a 106 KKASSERT(vp);
7a2de9a4
MD
107 }
108
7a2de9a4 109out:
d89ce96a
MD
110 /*
111 * Store the result of this lookup in the cache. Avoid this if the
7a2de9a4 112 * request was for creation, as it does not improve timings on
d89ce96a
MD
113 * emprical tests.
114 */
7a2de9a4
MD
115 if (vp) {
116 vn_unlock(vp);
117 cache_setvp(v->a_nch, vp);
118 vrele(vp);
d89ce96a 119 } else if (error == ENOENT) {
7a2de9a4
MD
120 cache_setvp(v->a_nch, NULL);
121 }
122 return error;
123}
124
125static int
126tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
127{
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;
132 int error;
133
134 *vpp = NULL;
135 /* Check accessibility of requested node as a first step. */
136 error = VOP_ACCESS(dvp, VEXEC, cred);
137 if (error != 0)
138 return error;
139
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);
144
145 if (*vpp)
146 vn_unlock(*vpp);
147 }
148
149 return (*vpp == NULL) ? ENOENT : 0;
150}
151
152/* --------------------------------------------------------------------- */
153
154static int
155tmpfs_ncreate(struct vop_ncreate_args *v)
156{
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;
162 int error;
163
164 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
165
7a2de9a4
MD
166 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
167 if (error == 0) {
168 cache_setunresolved(v->a_nch);
169 cache_setvp(v->a_nch, *vpp);
80ae59d7 170 tmpfs_knote(dvp, NOTE_WRITE);
7a2de9a4 171 }
7a2de9a4
MD
172
173 return error;
174}
175/* --------------------------------------------------------------------- */
176
177static int
178tmpfs_nmknod(struct vop_nmknod_args *v)
179{
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;
185 int error;
186
187 if (vap->va_type != VBLK && vap->va_type != VCHR &&
188 vap->va_type != VFIFO)
189 return EINVAL;
190
7a2de9a4
MD
191 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
192 if (error == 0) {
193 cache_setunresolved(v->a_nch);
194 cache_setvp(v->a_nch, *vpp);
80ae59d7 195 tmpfs_knote(dvp, NOTE_WRITE);
7a2de9a4 196 }
7a2de9a4
MD
197
198 return error;
199}
200
201/* --------------------------------------------------------------------- */
202
203static int
204tmpfs_open(struct vop_open_args *v)
205{
206 struct vnode *vp = v->a_vp;
207 int mode = v->a_mode;
208
209 int error;
210 struct tmpfs_node *node;
211
7a2de9a4
MD
212 node = VP_TO_TMPFS_NODE(vp);
213
6e0c5aab 214#if 0
7a2de9a4
MD
215 /* The file is still active but all its names have been removed
216 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
217 * it is about to die. */
218 if (node->tn_links < 1)
219 return (ENOENT);
6e0c5aab 220#endif
7a2de9a4
MD
221
222 /* If the file is marked append-only, deny write requests. */
630e3a33
MD
223 if ((node->tn_flags & APPEND) &&
224 (mode & (FWRITE | O_APPEND)) == FWRITE) {
7a2de9a4 225 error = EPERM;
630e3a33 226 } else {
7a2de9a4
MD
227 return (vop_stdopen(v));
228 }
7a2de9a4
MD
229 return error;
230}
231
232/* --------------------------------------------------------------------- */
233
234static int
235tmpfs_close(struct vop_close_args *v)
236{
237 struct vnode *vp = v->a_vp;
238 struct tmpfs_node *node;
239
240 node = VP_TO_TMPFS_NODE(vp);
241
242 if (node->tn_links > 0) {
6e0c5aab
MD
243 /*
244 * Update node times. No need to do it if the node has
245 * been deleted, because it will vanish after we return.
246 */
7a2de9a4
MD
247 tmpfs_update(vp);
248 }
249
250 return vop_stdclose(v);
251}
252
253/* --------------------------------------------------------------------- */
254
255int
256tmpfs_access(struct vop_access_args *v)
257{
258 struct vnode *vp = v->a_vp;
259 int error;
260 struct tmpfs_node *node;
261
7a2de9a4
MD
262 node = VP_TO_TMPFS_NODE(vp);
263
264 switch (vp->v_type) {
265 case VDIR:
266 /* FALLTHROUGH */
267 case VLNK:
268 /* FALLTHROUGH */
269 case VREG:
5a9e9ac7 270 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
7a2de9a4
MD
271 error = EROFS;
272 goto out;
273 }
274 break;
275
276 case VBLK:
277 /* FALLTHROUGH */
278 case VCHR:
279 /* FALLTHROUGH */
280 case VSOCK:
281 /* FALLTHROUGH */
282 case VFIFO:
283 break;
284
285 default:
286 error = EINVAL;
287 goto out;
288 }
289
5a9e9ac7 290 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
7a2de9a4
MD
291 error = EPERM;
292 goto out;
293 }
294
295 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
296
297out:
298
299 return error;
300}
301
302/* --------------------------------------------------------------------- */
303
304int
305tmpfs_getattr(struct vop_getattr_args *v)
306{
307 struct vnode *vp = v->a_vp;
308 struct vattr *vap = v->a_vap;
7a2de9a4 309 struct tmpfs_node *node;
7a2de9a4
MD
310
311 node = VP_TO_TMPFS_NODE(vp);
312
e575e508 313 lwkt_gettoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
314 tmpfs_update(vp);
315
316 vap->va_type = vp->v_type;
317 vap->va_mode = node->tn_mode;
318 vap->va_nlink = node->tn_links;
319 vap->va_uid = node->tn_uid;
320 vap->va_gid = node->tn_gid;
321 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
322 vap->va_fileid = node->tn_id;
323 vap->va_size = node->tn_size;
324 vap->va_blocksize = PAGE_SIZE;
325 vap->va_atime.tv_sec = node->tn_atime;
326 vap->va_atime.tv_nsec = node->tn_atimensec;
327 vap->va_mtime.tv_sec = node->tn_mtime;
328 vap->va_mtime.tv_nsec = node->tn_mtimensec;
329 vap->va_ctime.tv_sec = node->tn_ctime;
330 vap->va_ctime.tv_nsec = node->tn_ctimensec;
331 vap->va_gen = node->tn_gen;
332 vap->va_flags = node->tn_flags;
333 if (vp->v_type == VBLK || vp->v_type == VCHR)
334 {
335 vap->va_rmajor = umajor(node->tn_rdev);
336 vap->va_rminor = uminor(node->tn_rdev);
337 }
338 vap->va_bytes = round_page(node->tn_size);
339 vap->va_filerev = 0;
340
e575e508
VS
341 lwkt_reltoken(&vp->v_mount->mnt_token);
342
7a2de9a4
MD
343 return 0;
344}
345
346/* --------------------------------------------------------------------- */
347
348int
349tmpfs_setattr(struct vop_setattr_args *v)
350{
351 struct vnode *vp = v->a_vp;
352 struct vattr *vap = v->a_vap;
353 struct ucred *cred = v->a_cred;
80ae59d7 354 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
7a2de9a4 355 int error = 0;
80ae59d7 356 int kflags = 0;
7a2de9a4 357
80ae59d7 358 if (error == 0 && (vap->va_flags != VNOVAL)) {
7a2de9a4 359 error = tmpfs_chflags(vp, vap->va_flags, cred);
80ae59d7
MD
360 kflags |= NOTE_ATTRIB;
361 }
7a2de9a4 362
80ae59d7
MD
363 if (error == 0 && (vap->va_size != VNOVAL)) {
364 if (vap->va_size > node->tn_size)
365 kflags |= NOTE_WRITE | NOTE_EXTEND;
366 else
367 kflags |= NOTE_WRITE;
7a2de9a4 368 error = tmpfs_chsize(vp, vap->va_size, cred);
80ae59d7 369 }
7a2de9a4 370
d89ce96a
MD
371 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
372 vap->va_gid != (gid_t)VNOVAL)) {
7a2de9a4 373 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
80ae59d7 374 kflags |= NOTE_ATTRIB;
d89ce96a 375 }
7a2de9a4 376
80ae59d7 377 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
7a2de9a4 378 error = tmpfs_chmod(vp, vap->va_mode, cred);
80ae59d7
MD
379 kflags |= NOTE_ATTRIB;
380 }
7a2de9a4
MD
381
382 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
383 vap->va_atime.tv_nsec != VNOVAL) ||
384 (vap->va_mtime.tv_sec != VNOVAL &&
d89ce96a 385 vap->va_mtime.tv_nsec != VNOVAL) )) {
7a2de9a4 386 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
d89ce96a 387 vap->va_vaflags, cred);
80ae59d7 388 kflags |= NOTE_ATTRIB;
d89ce96a 389 }
7a2de9a4
MD
390
391 /* Update the node times. We give preference to the error codes
392 * generated by this function rather than the ones that may arise
393 * from tmpfs_update. */
394 tmpfs_update(vp);
80ae59d7 395 tmpfs_knote(vp, kflags);
7a2de9a4 396
7a2de9a4
MD
397 return error;
398}
399
400/* --------------------------------------------------------------------- */
401
9fc94b5f 402/*
630e3a33
MD
403 * fsync is usually a NOP, but we must take action when unmounting or
404 * when recycling.
9fc94b5f 405 */
7a2de9a4
MD
406static int
407tmpfs_fsync(struct vop_fsync_args *v)
408{
630e3a33 409 struct tmpfs_node *node;
7a2de9a4
MD
410 struct vnode *vp = v->a_vp;
411
630e3a33
MD
412 node = VP_TO_TMPFS_NODE(vp);
413
414 tmpfs_update(vp);
415 if (vp->v_type == VREG) {
d4623db3 416 if (vp->v_flag & VRECLAIMED) {
630e3a33
MD
417 if (node->tn_links == 0)
418 tmpfs_truncate(vp, 0);
419 else
420 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
421 }
9fc94b5f 422 }
7a2de9a4
MD
423 return 0;
424}
425
426/* --------------------------------------------------------------------- */
427
428static int
429tmpfs_read (struct vop_read_args *ap)
430{
431 struct buf *bp;
432 struct vnode *vp = ap->a_vp;
433 struct uio *uio = ap->a_uio;
434 struct tmpfs_node *node;
7a2de9a4 435 off_t base_offset;
9fc94b5f 436 size_t offset;
7a2de9a4 437 size_t len;
9fc94b5f 438 int error;
7a2de9a4
MD
439
440 error = 0;
441 if (uio->uio_resid == 0) {
442 return error;
443 }
444
445 node = VP_TO_TMPFS_NODE(vp);
446
447 if (uio->uio_offset < 0)
448 return (EINVAL);
449 if (vp->v_type != VREG)
450 return (EINVAL);
451
7a2de9a4
MD
452 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
453 /*
454 * Use buffer cache I/O (via tmpfs_strategy)
455 */
9fc94b5f 456 offset = (size_t)uio->uio_offset & BMASK;
7a2de9a4 457 base_offset = (off_t)uio->uio_offset - offset;
9de13b88
MD
458 bp = getcacheblk(vp, base_offset, BSIZE, 0);
459 if (bp == NULL) {
b403e861 460 lwkt_gettoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
461 error = bread(vp, base_offset, BSIZE, &bp);
462 if (error) {
463 brelse(bp);
b403e861 464 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
465 kprintf("tmpfs_read bread error %d\n", error);
466 break;
467 }
b403e861 468 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
469 }
470
7a2de9a4
MD
471 /*
472 * Figure out how many bytes we can actually copy this loop.
473 */
474 len = BSIZE - offset;
475 if (len > uio->uio_resid)
476 len = uio->uio_resid;
477 if (len > node->tn_size - uio->uio_offset)
478 len = (size_t)(node->tn_size - uio->uio_offset);
479
44480e31 480 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
7a2de9a4
MD
481 bqrelse(bp);
482 if (error) {
483 kprintf("tmpfs_read uiomove error %d\n", error);
484 break;
485 }
486 }
487
7a2de9a4
MD
488 TMPFS_NODE_LOCK(node);
489 node->tn_status |= TMPFS_NODE_ACCESSED;
490 TMPFS_NODE_UNLOCK(node);
491
7a2de9a4
MD
492 return(error);
493}
494
495static int
496tmpfs_write (struct vop_write_args *ap)
497{
498 struct buf *bp;
499 struct vnode *vp = ap->a_vp;
500 struct uio *uio = ap->a_uio;
501 struct thread *td = uio->uio_td;
502 struct tmpfs_node *node;
503 boolean_t extended;
504 off_t oldsize;
505 int error;
7a2de9a4 506 off_t base_offset;
9fc94b5f 507 size_t offset;
7a2de9a4
MD
508 size_t len;
509 struct rlimit limit;
7a2de9a4 510 int trivial = 0;
80ae59d7 511 int kflags = 0;
7a2de9a4
MD
512
513 error = 0;
514 if (uio->uio_resid == 0) {
515 return error;
516 }
517
518 node = VP_TO_TMPFS_NODE(vp);
519
520 if (vp->v_type != VREG)
521 return (EINVAL);
522
1be4932c
VS
523 lwkt_gettoken(&vp->v_mount->mnt_token);
524
7a2de9a4
MD
525 oldsize = node->tn_size;
526 if (ap->a_ioflag & IO_APPEND)
527 uio->uio_offset = node->tn_size;
528
529 /*
530 * Check for illegal write offsets.
531 */
532 if (uio->uio_offset + uio->uio_resid >
1be4932c
VS
533 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
534 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4 535 return (EFBIG);
1be4932c 536 }
7a2de9a4
MD
537
538 if (vp->v_type == VREG && td != NULL) {
539 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
1be4932c
VS
540 if (error != 0) {
541 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4 542 return error;
1be4932c 543 }
7a2de9a4
MD
544 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
545 ksignal(td->td_proc, SIGXFSZ);
1be4932c 546 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
547 return (EFBIG);
548 }
549 }
550
551
552 /*
553 * Extend the file's size if necessary
554 */
9fc94b5f 555 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
7a2de9a4 556
7a2de9a4
MD
557 while (uio->uio_resid > 0) {
558 /*
559 * Use buffer cache I/O (via tmpfs_strategy)
560 */
9fc94b5f 561 offset = (size_t)uio->uio_offset & BMASK;
7a2de9a4
MD
562 base_offset = (off_t)uio->uio_offset - offset;
563 len = BSIZE - offset;
564 if (len > uio->uio_resid)
565 len = uio->uio_resid;
566
567 if ((uio->uio_offset + len) > node->tn_size) {
9fc94b5f 568 trivial = (uio->uio_offset <= node->tn_size);
7a2de9a4
MD
569 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
570 if (error)
571 break;
572 }
573
9fc94b5f
MD
574 /*
575 * Read to fill in any gaps. Theoretically we could
576 * optimize this if the write covers the entire buffer
577 * and is not a UIO_NOCOPY write, however this can lead
578 * to a security violation exposing random kernel memory
579 * (whatever junk was in the backing VM pages before).
580 *
581 * So just use bread() to do the right thing.
582 */
583 error = bread(vp, base_offset, BSIZE, &bp);
44480e31 584 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
7a2de9a4
MD
585 if (error) {
586 kprintf("tmpfs_write uiomove error %d\n", error);
587 brelse(bp);
588 break;
589 }
590
80ae59d7 591 if (uio->uio_offset > node->tn_size) {
7a2de9a4 592 node->tn_size = uio->uio_offset;
80ae59d7
MD
593 kflags |= NOTE_EXTEND;
594 }
595 kflags |= NOTE_WRITE;
7a2de9a4
MD
596
597 /*
fc1a9118
MD
598 * Always try to flush the page if the request is coming
599 * from the pageout daemon (IO_ASYNC), else buwrite() the
600 * buffer.
7a2de9a4 601 *
fc1a9118
MD
602 * buwrite() dirties the underlying VM pages instead of
603 * dirtying the buffer, releasing the buffer as a clean
604 * buffer. This allows tmpfs to use essentially all
605 * available memory to cache file data. If we used bdwrite()
606 * the buffer cache would wind up flushing the data to
607 * swap too quickly.
7a2de9a4 608 */
fc1a9118
MD
609 bp->b_flags |= B_AGE;
610 if (ap->a_ioflag & IO_ASYNC) {
611 bawrite(bp);
b7545cb3 612 } else {
fc1a9118 613 buwrite(bp);
d89ce96a 614 }
9fc94b5f 615
7a2de9a4 616 if (bp->b_error) {
2cd8c774 617 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
7a2de9a4
MD
618 break;
619 }
620 }
7a2de9a4 621
7a2de9a4 622 if (error) {
80ae59d7 623 if (extended) {
7a2de9a4 624 (void)tmpfs_reg_resize(vp, oldsize, trivial);
80ae59d7
MD
625 kflags &= ~NOTE_EXTEND;
626 }
627 goto done;
7a2de9a4
MD
628 }
629
fc1a9118
MD
630 /*
631 * Currently we don't set the mtime on files modified via mmap()
632 * because we can't tell the difference between those modifications
633 * and an attempt by the pageout daemon to flush tmpfs pages to
634 * swap.
635 *
636 * This is because in order to defer flushes as long as possible
637 * buwrite() works by marking the underlying VM pages dirty in
638 * order to be able to dispose of the buffer cache buffer without
639 * flushing it.
640 */
7a2de9a4 641 TMPFS_NODE_LOCK(node);
fc1a9118
MD
642 if (uio->uio_segflg != UIO_NOCOPY)
643 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
644 if (extended)
645 node->tn_status |= TMPFS_NODE_CHANGED;
7a2de9a4
MD
646
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);
650 }
651 TMPFS_NODE_UNLOCK(node);
80ae59d7 652done:
7a2de9a4 653
80ae59d7 654 tmpfs_knote(vp, kflags);
1be4932c
VS
655
656
657 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
658 return(error);
659}
660
661static int
662tmpfs_advlock (struct vop_advlock_args *ap)
663{
664 struct tmpfs_node *node;
665 struct vnode *vp = ap->a_vp;
666
667 node = VP_TO_TMPFS_NODE(vp);
668
669 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
670}
671
fc1a9118
MD
672/*
673 * The strategy function is typically only called when memory pressure
674 * forces the system to attempt to pageout pages. It can also be called
675 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write
676 * operations
677 */
7a2de9a4
MD
678static int
679tmpfs_strategy(struct vop_strategy_args *ap)
680{
681 struct bio *bio = ap->a_bio;
8f9ba07b 682 struct bio *nbio;
9fc94b5f 683 struct buf *bp = bio->bio_buf;
7a2de9a4
MD
684 struct vnode *vp = ap->a_vp;
685 struct tmpfs_node *node;
686 vm_object_t uobj;
8f9ba07b
MD
687 vm_page_t m;
688 int i;
7a2de9a4 689
9fc94b5f
MD
690 if (vp->v_type != VREG) {
691 bp->b_resid = bp->b_bcount;
692 bp->b_flags |= B_ERROR | B_INVAL;
693 bp->b_error = EINVAL;
694 biodone(bio);
695 return(0);
696 }
7a2de9a4 697
d1f61aa2 698 lwkt_gettoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
699 node = VP_TO_TMPFS_NODE(vp);
700
701 uobj = node->tn_reg.tn_aobj;
9fc94b5f 702
7a2de9a4 703 /*
fc1a9118
MD
704 * Don't bother flushing to swap if there is no swap, just
705 * ensure that the pages are marked as needing a commit (still).
7a2de9a4 706 */
8f9ba07b
MD
707 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) {
708 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
709 m = bp->b_xio.xio_pages[i];
fc1a9118 710 vm_page_need_commit(m);
8f9ba07b
MD
711 }
712 bp->b_resid = 0;
713 bp->b_error = 0;
714 biodone(bio);
715 } else {
716 nbio = push_bio(bio);
717 nbio->bio_done = tmpfs_strategy_done;
718 nbio->bio_offset = bio->bio_offset;
719 swap_pager_strategy(uobj, nbio);
720 }
7a2de9a4 721
d1f61aa2 722 lwkt_reltoken(&vp->v_mount->mnt_token);
7a2de9a4
MD
723 return 0;
724}
725
8f9ba07b 726/*
fc1a9118
MD
727 * If we were unable to commit the pages to swap make sure they are marked
728 * as needing a commit (again). If we were, clear the flag to allow the
729 * pages to be freed.
8f9ba07b
MD
730 */
731static void
732tmpfs_strategy_done(struct bio *bio)
733{
734 struct buf *bp;
735 vm_page_t m;
736 int i;
737
738 bp = bio->bio_buf;
739
fc1a9118 740 if (bp->b_flags & B_ERROR) {
8f9ba07b
MD
741 bp->b_flags &= ~B_ERROR;
742 bp->b_error = 0;
743 bp->b_resid = 0;
744 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
745 m = bp->b_xio.xio_pages[i];
fc1a9118
MD
746 vm_page_need_commit(m);
747 }
748 } else {
749 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
750 m = bp->b_xio.xio_pages[i];
751 vm_page_clear_commit(m);
8f9ba07b
MD
752 }
753 }
754 bio = pop_bio(bio);
755 biodone(bio);
756}
757
7a2de9a4
MD
758static int
759tmpfs_bmap(struct vop_bmap_args *ap)
760{
761 if (ap->a_doffsetp != NULL)
762 *ap->a_doffsetp = ap->a_loffset;
763 if (ap->a_runp != NULL)
764 *ap->a_runp = 0;
765 if (ap->a_runb != NULL)
766 *ap->a_runb = 0;
767
768 return 0;
769}
9fc94b5f 770
7a2de9a4
MD
771/* --------------------------------------------------------------------- */
772
773static int
774tmpfs_nremove(struct vop_nremove_args *v)
775{
776 struct vnode *dvp = v->a_dvp;
777 struct namecache *ncp = v->a_nch->ncp;
9fc94b5f 778 struct vnode *vp;
7a2de9a4
MD
779 int error;
780 struct tmpfs_dirent *de;
781 struct tmpfs_mount *tmp;
782 struct tmpfs_node *dnode;
783 struct tmpfs_node *node;
784
9fc94b5f 785 /*
a1fa5d8d
MD
786 * We have to acquire the vp from v->a_nch because we will likely
787 * unresolve the namecache entry, and a vrele/vput is needed to
788 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
789 *
790 * We have to use vget to clear any inactive state on the vnode,
791 * otherwise the vnode may remain inactive and thus tmpfs_inactive
792 * will not get called when we release it.
9fc94b5f 793 */
a1fa5d8d 794 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
9fc94b5f 795 KKASSERT(error == 0);
a1fa5d8d 796 vn_unlock(vp);
7a2de9a4
MD
797
798 if (vp->v_type == VDIR) {
799 error = EISDIR;
800 goto out;
801 }
802
803 dnode = VP_TO_TMPFS_DIR(dvp);
804 node = VP_TO_TMPFS_NODE(vp);
805 tmp = VFS_TO_TMPFS(vp->v_mount);
806 de = tmpfs_dir_lookup(dnode, node, ncp);
807 if (de == NULL) {
808 error = ENOENT;
809 goto out;
810 }
811
812 /* Files marked as immutable or append-only cannot be deleted. */
813 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
814 (dnode->tn_flags & APPEND)) {
815 error = EPERM;
816 goto out;
817 }
818
819 /* Remove the entry from the directory; as it is a file, we do not
820 * have to change the number of hard links of the directory. */
22d3b394 821 tmpfs_dir_detach(dnode, de);
7a2de9a4
MD
822
823 /* Free the directory entry we just deleted. Note that the node
824 * referred by it will not be removed until the vnode is really
825 * reclaimed. */
0786baf1 826 tmpfs_free_dirent(tmp, de);
7a2de9a4
MD
827
828 if (node->tn_links > 0) {
829 TMPFS_NODE_LOCK(node);
830 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
831 TMPFS_NODE_MODIFIED;
832 TMPFS_NODE_UNLOCK(node);
833 }
834
6e0c5aab 835 cache_unlink(v->a_nch);
80ae59d7 836 tmpfs_knote(vp, NOTE_DELETE);
80ae59d7 837 tmpfs_knote(dvp, NOTE_WRITE);
7a2de9a4
MD
838 error = 0;
839
7a2de9a4 840out:
9fc94b5f 841 vrele(vp);
7a2de9a4
MD
842
843 return error;
844}
845
846/* --------------------------------------------------------------------- */
847
848static int
849tmpfs_nlink(struct vop_nlink_args *v)
850{
851 struct vnode *dvp = v->a_dvp;
852 struct vnode *vp = v->a_vp;
853 struct namecache *ncp = v->a_nch->ncp;
7a2de9a4
MD
854 struct tmpfs_dirent *de;
855 struct tmpfs_node *node;
22d3b394
MD
856 struct tmpfs_node *dnode;
857 int error;
7a2de9a4 858
7a2de9a4
MD
859 KKASSERT(dvp != vp); /* XXX When can this be false? */
860
861 node = VP_TO_TMPFS_NODE(vp);
22d3b394 862 dnode = VP_TO_TMPFS_NODE(dvp);
7a2de9a4
MD
863
864 /* XXX: Why aren't the following two tests done by the caller? */
865
866 /* Hard links of directories are forbidden. */
867 if (vp->v_type == VDIR) {
868 error = EPERM;
869 goto out;
870 }
871
872 /* Cannot create cross-device links. */
873 if (dvp->v_mount != vp->v_mount) {
874 error = EXDEV;
875 goto out;
876 }
877
878 /* Ensure that we do not overflow the maximum number of links imposed
879 * by the system. */
880 KKASSERT(node->tn_links <= LINK_MAX);
881 if (node->tn_links == LINK_MAX) {
882 error = EMLINK;
883 goto out;
884 }
885
886 /* We cannot create links of files marked immutable or append-only. */
887 if (node->tn_flags & (IMMUTABLE | APPEND)) {
888 error = EPERM;
889 goto out;
890 }
891
892 /* Allocate a new directory entry to represent the node. */
893 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
894 ncp->nc_name, ncp->nc_nlen, &de);
895 if (error != 0)
896 goto out;
897
898 /* Insert the new directory entry into the appropriate directory. */
22d3b394 899 tmpfs_dir_attach(dnode, de);
7a2de9a4
MD
900
901 /* vp link count has changed, so update node times. */
902
903 TMPFS_NODE_LOCK(node);
904 node->tn_status |= TMPFS_NODE_CHANGED;
905 TMPFS_NODE_UNLOCK(node);
906 tmpfs_update(vp);
907
80ae59d7 908 tmpfs_knote(vp, NOTE_LINK);
7a2de9a4
MD
909 cache_setunresolved(v->a_nch);
910 cache_setvp(v->a_nch, vp);
80ae59d7 911 tmpfs_knote(dvp, NOTE_WRITE);
7a2de9a4
MD
912 error = 0;
913
914out:
7a2de9a4
MD
915 return error;
916}
917
918/* --------------------------------------------------------------------- */
919
920static int
921tmpfs_nrename(struct vop_nrename_args *v)
922{
923 struct vnode *fdvp = v->a_fdvp;
924 struct namecache *fncp = v->a_fnch->ncp;
925 struct vnode *fvp = fncp->nc_vp;
926 struct vnode *tdvp = v->a_tdvp;
927 struct namecache *tncp = v->a_tnch->ncp;
a1fa5d8d 928 struct vnode *tvp;
29ca4fd6 929 struct tmpfs_dirent *de, *tde;
7a2de9a4
MD
930 struct tmpfs_mount *tmp;
931 struct tmpfs_node *fdnode;
932 struct tmpfs_node *fnode;
933 struct tmpfs_node *tnode;
934 struct tmpfs_node *tdnode;
22d3b394 935 char *newname;
dca262fb 936 char *oldname;
22d3b394 937 int error;
7a2de9a4 938
a1fa5d8d
MD
939 /*
940 * Because tvp can get overwritten we have to vget it instead of
941 * just vref or use it, otherwise it's VINACTIVE flag may not get
942 * cleared and the node won't get destroyed.
943 */
944 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp);
945 if (error == 0) {
946 tnode = VP_TO_TMPFS_NODE(tvp);
947 vn_unlock(tvp);
948 } else {
949 tnode = NULL;
950 }
7a2de9a4
MD
951
952 /* Disallow cross-device renames.
953 * XXX Why isn't this done by the caller? */
954 if (fvp->v_mount != tdvp->v_mount ||
955 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
956 error = EXDEV;
957 goto out;
958 }
959
960 tmp = VFS_TO_TMPFS(tdvp->v_mount);
961 tdnode = VP_TO_TMPFS_DIR(tdvp);
962
963 /* If source and target are the same file, there is nothing to do. */
964 if (fvp == tvp) {
965 error = 0;
966 goto out;
967 }
968
7a2de9a4
MD
969 fdnode = VP_TO_TMPFS_DIR(fdvp);
970 fnode = VP_TO_TMPFS_NODE(fvp);
971 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
972
973 /* Avoid manipulating '.' and '..' entries. */
974 if (de == NULL) {
975 error = ENOENT;
976 goto out_locked;
977 }
978 KKASSERT(de->td_node == fnode);
979
dca262fb
MD
980 /*
981 * If replacing an entry in the target directory and that entry
982 * is a directory, it must be empty.
983 *
7a2de9a4 984 * Kern_rename gurantees the destination to be a directory
dca262fb
MD
985 * if the source is one (it does?).
986 */
7a2de9a4
MD
987 if (tvp != NULL) {
988 KKASSERT(tnode != NULL);
989
990 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
991 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
992 error = EPERM;
993 goto out_locked;
994 }
995
996 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
997 if (tnode->tn_size > 0) {
998 error = ENOTEMPTY;
999 goto out_locked;
1000 }
1001 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1002 error = ENOTDIR;
1003 goto out_locked;
1004 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1005 error = EISDIR;
1006 goto out_locked;
1007 } else {
1008 KKASSERT(fnode->tn_type != VDIR &&
1009 tnode->tn_type != VDIR);
1010 }
1011 }
1012
dca262fb
MD
1013 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1014 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
7a2de9a4
MD
1015 error = EPERM;
1016 goto out_locked;
1017 }
1018
dca262fb
MD
1019 /*
1020 * Ensure that we have enough memory to hold the new name, if it
1021 * has to be changed.
1022 */
7a2de9a4
MD
1023 if (fncp->nc_nlen != tncp->nc_nlen ||
1024 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
d00cd01c 1025 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
42f6f6b1
VS
1026 M_WAITOK | M_NULLOK);
1027 if (newname == NULL) {
1028 error = ENOSPC;
1029 goto out_locked;
1030 }
dca262fb
MD
1031 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
1032 newname[tncp->nc_nlen] = '\0';
1033 } else {
7a2de9a4 1034 newname = NULL;
dca262fb 1035 }
7a2de9a4 1036
dca262fb
MD
1037 /*
1038 * Unlink entry from source directory. Note that the kernel has
1039 * already checked for illegal recursion cases (renaming a directory
1040 * into a subdirectory of itself).
1041 */
6e0c5aab 1042 if (fdnode != tdnode) {
dca262fb 1043 tmpfs_dir_detach(fdnode, de);
6e0c5aab 1044 } else {
29ca4fd6
JH
1045 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de);
1046 }
dca262fb
MD
1047
1048 /*
1049 * Handle any name change. Swap with newname, we will
1050 * deallocate it at the end.
1051 */
1052 if (newname != NULL) {
1053#if 0
1054 TMPFS_NODE_LOCK(fnode);
1055 fnode->tn_status |= TMPFS_NODE_CHANGED;
1056 TMPFS_NODE_UNLOCK(fnode);
1057#endif
1058 oldname = de->td_name;
1059 de->td_name = newname;
1060 de->td_namelen = (uint16_t)tncp->nc_nlen;
1061 newname = oldname;
1062 }
1063
1064 /*
29ca4fd6
JH
1065 * If we are overwriting an entry, we have to remove the old one
1066 * from the target directory.
1067 */
1068 if (tvp != NULL) {
1069 /* Remove the old entry from the target directory. */
1070 tde = tmpfs_dir_lookup(tdnode, tnode, tncp);
1071 tmpfs_dir_detach(tdnode, tde);
1072 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1073
1074 /*
1075 * Free the directory entry we just deleted. Note that the
1076 * node referred by it will not be removed until the vnode is
1077 * really reclaimed.
1078 */
1079 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1080 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1081 }
1082
1083 /*
dca262fb
MD
1084 * Link entry to target directory. If the entry
1085 * represents a directory move the parent linkage
1086 * as well.
1087 */
7a2de9a4 1088 if (fdnode != tdnode) {
7a2de9a4 1089 if (de->td_node->tn_type == VDIR) {
7a2de9a4 1090 TMPFS_VALIDATE_DIR(fnode);
7a2de9a4 1091 }
22d3b394 1092 tmpfs_dir_attach(tdnode, de);
dca262fb 1093 } else {
7a2de9a4 1094 TMPFS_NODE_LOCK(tdnode);
7a2de9a4 1095 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
29ca4fd6 1096 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de);
7a2de9a4
MD
1097 TMPFS_NODE_UNLOCK(tdnode);
1098 }
1099
dca262fb 1100 /*
dca262fb
MD
1101 * Finish up
1102 */
1103 if (newname) {
d00cd01c 1104 kfree(newname, tmp->tm_name_zone);
dca262fb
MD
1105 newname = NULL;
1106 }
7a2de9a4 1107 cache_rename(v->a_fnch, v->a_tnch);
80ae59d7
MD
1108 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1109 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1110 if (fnode->tn_vnode)
1111 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
7a2de9a4
MD
1112 error = 0;
1113
1114out_locked:
630e3a33 1115 ;
7a2de9a4
MD
1116
1117out:
a1fa5d8d
MD
1118 if (tvp)
1119 vrele(tvp);
7a2de9a4
MD
1120
1121 return error;
1122}
1123
1124/* --------------------------------------------------------------------- */
1125
1126static int
1127tmpfs_nmkdir(struct vop_nmkdir_args *v)
1128{
1129 struct vnode *dvp = v->a_dvp;
1130 struct vnode **vpp = v->a_vpp;
1131 struct namecache *ncp = v->a_nch->ncp;
1132 struct vattr *vap = v->a_vap;
1133 struct ucred *cred = v->a_cred;
1134 int error;
1135
1136 KKASSERT(vap->va_type == VDIR);
1137
7a2de9a4
MD
1138 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1139 if (error == 0) {
1140 cache_setunresolved(v->a_nch);
1141 cache_setvp(v->a_nch, *vpp);
80ae59d7 1142 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
7a2de9a4 1143 }
7a2de9a4
MD
1144
1145 return error;
1146}
1147
1148/* --------------------------------------------------------------------- */
1149
1150static int
1151tmpfs_nrmdir(struct vop_nrmdir_args *v)
1152{
1153 struct vnode *dvp = v->a_dvp;
1154 struct namecache *ncp = v->a_nch->ncp;
9fc94b5f 1155 struct vnode *vp;
7a2de9a4
MD
1156 struct tmpfs_dirent *de;
1157 struct tmpfs_mount *tmp;
1158 struct tmpfs_node *dnode;
1159 struct tmpfs_node *node;
38e5e604
MD
1160 int error;
1161
1162 /*
a1fa5d8d
MD
1163 * We have to acquire the vp from v->a_nch because we will likely
1164 * unresolve the namecache entry, and a vrele/vput is needed to
1165 * trigger the tmpfs_inactive/tmpfs_reclaim sequence.
1166 *
1167 * We have to use vget to clear any inactive state on the vnode,
1168 * otherwise the vnode may remain inactive and thus tmpfs_inactive
1169 * will not get called when we release it.
9fc94b5f 1170 */
a1fa5d8d 1171 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp);
9fc94b5f 1172 KKASSERT(error == 0);
a1fa5d8d 1173 vn_unlock(vp);
7a2de9a4 1174
e527fb6b
MD
1175 /*
1176 * Prevalidate so we don't hit an assertion later
1177 */
1178 if (vp->v_type != VDIR) {
1179 error = ENOTDIR;
1180 goto out;
1181 }
1182
7a2de9a4
MD
1183 tmp = VFS_TO_TMPFS(dvp->v_mount);
1184 dnode = VP_TO_TMPFS_DIR(dvp);
1185 node = VP_TO_TMPFS_DIR(vp);
1186
1187 /* Directories with more than two entries ('.' and '..') cannot be
1188 * removed. */
1189 if (node->tn_size > 0) {
1190 error = ENOTEMPTY;
1191 goto out;
1192 }
1193
1194 if ((dnode->tn_flags & APPEND)
1195 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1196 error = EPERM;
1197 goto out;
1198 }
1199
1200 /* This invariant holds only if we are not trying to remove "..".
1201 * We checked for that above so this is safe now. */
1202 KKASSERT(node->tn_dir.tn_parent == dnode);
1203
1204 /* Get the directory entry associated with node (vp). This was
1205 * filled by tmpfs_lookup while looking up the entry. */
1206 de = tmpfs_dir_lookup(dnode, node, ncp);
1207 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1208 ncp->nc_name,
1209 ncp->nc_nlen));
1210
1211 /* Check flags to see if we are allowed to remove the directory. */
b7fe63af
MD
1212 if ((dnode->tn_flags & APPEND) ||
1213 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
7a2de9a4
MD
1214 error = EPERM;
1215 goto out;
1216 }
1217
1218
1219 /* Detach the directory entry from the directory (dnode). */
22d3b394 1220 tmpfs_dir_detach(dnode, de);
7a2de9a4
MD
1221
1222 /* No vnode should be allocated for this entry from this point */
1223 TMPFS_NODE_LOCK(node);
1224 TMPFS_ASSERT_ELOCKED(node);
1225 TMPFS_NODE_LOCK(dnode);
1226 TMPFS_ASSERT_ELOCKED(dnode);
1227
6e0c5aab
MD
1228 /*
1229 * Must set parent linkage to NULL (tested by ncreate to disallow
1230 * the creation of new files/dirs in a deleted directory)
1231 */
7a2de9a4
MD
1232 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1233 TMPFS_NODE_MODIFIED;
1234
7a2de9a4
MD
1235 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1236 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1237
1238 TMPFS_NODE_UNLOCK(dnode);
1239 TMPFS_NODE_UNLOCK(node);
1240
1241 /* Free the directory entry we just deleted. Note that the node
1242 * referred by it will not be removed until the vnode is really
1243 * reclaimed. */
0786baf1 1244 tmpfs_free_dirent(tmp, de);
7a2de9a4
MD
1245
1246 /* Release the deleted vnode (will destroy the node, notify
1247 * interested parties and clean it from the cache). */
1248
1249 TMPFS_NODE_LOCK(dnode);
1250 dnode->tn_status |= TMPFS_NODE_CHANGED;
1251 TMPFS_NODE_UNLOCK(dnode);
1252 tmpfs_update(dvp);
1253
6e0c5aab 1254 cache_unlink(v->a_nch);
80ae59d7 1255 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
7a2de9a4
MD
1256 error = 0;
1257
1258out:
9fc94b5f 1259 vrele(vp);
7a2de9a4
MD
1260
1261 return error;
1262}
1263
1264/* --------------------------------------------------------------------- */
1265
1266static int
1267tmpfs_nsymlink(struct vop_nsymlink_args *v)
1268{
1269 struct vnode *dvp = v->a_dvp;
1270 struct vnode **vpp = v->a_vpp;
1271 struct namecache *ncp = v->a_nch->ncp;
1272 struct vattr *vap = v->a_vap;
1273 struct ucred *cred = v->a_cred;
1274 char *target = v->a_target;
1275 int error;
1276
7a2de9a4
MD
1277 vap->va_type = VLNK;
1278 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1279 if (error == 0) {
80ae59d7 1280 tmpfs_knote(*vpp, NOTE_WRITE);
7a2de9a4
MD
1281 cache_setunresolved(v->a_nch);
1282 cache_setvp(v->a_nch, *vpp);
1283 }
7a2de9a4
MD
1284
1285 return error;
1286}
1287
1288/* --------------------------------------------------------------------- */
1289
1290static int
1291tmpfs_readdir(struct vop_readdir_args *v)
1292{
1293 struct vnode *vp = v->a_vp;
1294 struct uio *uio = v->a_uio;
1295 int *eofflag = v->a_eofflag;
1296 off_t **cookies = v->a_cookies;
1297 int *ncookies = v->a_ncookies;
22d3b394 1298 struct tmpfs_mount *tmp;
7a2de9a4
MD
1299 int error;
1300 off_t startoff;
1301 off_t cnt = 0;
1302 struct tmpfs_node *node;
1303
1304 /* This operation only makes sense on directory nodes. */
1305 if (vp->v_type != VDIR)
1306 return ENOTDIR;
1307
22d3b394 1308 tmp = VFS_TO_TMPFS(vp->v_mount);
7a2de9a4
MD
1309 node = VP_TO_TMPFS_DIR(vp);
1310 startoff = uio->uio_offset;
1311
1312 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1313 error = tmpfs_dir_getdotdent(node, uio);
1314 if (error != 0)
1315 goto outok;
1316 cnt++;
1317 }
1318
1319 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
22d3b394 1320 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
7a2de9a4
MD
1321 if (error != 0)
1322 goto outok;
1323 cnt++;
1324 }
1325
1326 error = tmpfs_dir_getdents(node, uio, &cnt);
1327
1328outok:
1329 KKASSERT(error >= -1);
1330
1331 if (error == -1)
1332 error = 0;
1333
1334 if (eofflag != NULL)
1335 *eofflag =
1336 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1337
1338 /* Update NFS-related variables. */
1339 if (error == 0 && cookies != NULL && ncookies != NULL) {
1340 off_t i;
1341 off_t off = startoff;
1342 struct tmpfs_dirent *de = NULL;
1343
1344 *ncookies = cnt;
1345 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1346
1347 for (i = 0; i < cnt; i++) {
1348 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1349 if (off == TMPFS_DIRCOOKIE_DOT) {
1350 off = TMPFS_DIRCOOKIE_DOTDOT;
1351 } else {
1352 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
29ca4fd6 1353 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
7a2de9a4 1354 } else if (de != NULL) {
29ca4fd6 1355 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
7a2de9a4
MD
1356 } else {
1357 de = tmpfs_dir_lookupbycookie(node,
1358 off);
1359 KKASSERT(de != NULL);
29ca4fd6 1360 de = RB_NEXT(tmpfs_dirtree, &node->tn_dir.tn_dirtree, de);
7a2de9a4
MD
1361 }
1362 if (de == NULL)
1363 off = TMPFS_DIRCOOKIE_EOF;
1364 else
1365 off = tmpfs_dircookie(de);
1366 }
1367
1368 (*cookies)[i] = off;
1369 }
1370 KKASSERT(uio->uio_offset == off);
1371 }
7a2de9a4
MD
1372
1373 return error;
1374}
1375
1376/* --------------------------------------------------------------------- */
1377
1378static int
1379tmpfs_readlink(struct vop_readlink_args *v)
1380{
1381 struct vnode *vp = v->a_vp;
1382 struct uio *uio = v->a_uio;
1383
1384 int error;
1385 struct tmpfs_node *node;
1386
1387 KKASSERT(uio->uio_offset == 0);
1388 KKASSERT(vp->v_type == VLNK);
1389
1390 node = VP_TO_TMPFS_NODE(vp);
1391
1392 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1393 uio);
1394 TMPFS_NODE_LOCK(node);
1395 node->tn_status |= TMPFS_NODE_ACCESSED;
1396 TMPFS_NODE_UNLOCK(node);
1397
1398 return error;
1399}
1400
1401/* --------------------------------------------------------------------- */
1402
1403static int
1404tmpfs_inactive(struct vop_inactive_args *v)
1405{
1406 struct vnode *vp = v->a_vp;
7a2de9a4
MD
1407 struct tmpfs_node *node;
1408
7a2de9a4
MD
1409 node = VP_TO_TMPFS_NODE(vp);
1410
9fc94b5f 1411 /*
a1fa5d8d
MD
1412 * Degenerate case
1413 */
1414 if (node == NULL) {
1415 vrecycle(vp);
1416 return(0);
1417 }
1418
1419 /*
9fc94b5f
MD
1420 * Get rid of unreferenced deleted vnodes sooner rather than
1421 * later so the data memory can be recovered immediately.
f96f2f39
MD
1422 *
1423 * We must truncate the vnode to prevent the normal reclamation
1424 * path from flushing the data for the removed file to disk.
9fc94b5f 1425 */
7a2de9a4 1426 TMPFS_NODE_LOCK(node);
b7fe63af 1427 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
6e0c5aab 1428 node->tn_links == 0)
b7fe63af 1429 {
9fc94b5f 1430 node->tn_vpstate = TMPFS_VNODE_DOOMED;
7a2de9a4 1431 TMPFS_NODE_UNLOCK(node);
f96f2f39
MD
1432 if (node->tn_type == VREG)
1433 tmpfs_truncate(vp, 0);
7a2de9a4 1434 vrecycle(vp);
9fc94b5f 1435 } else {
7a2de9a4 1436 TMPFS_NODE_UNLOCK(node);
9fc94b5f 1437 }
7a2de9a4
MD
1438
1439 return 0;
1440}
1441
1442/* --------------------------------------------------------------------- */
1443
1444int
1445tmpfs_reclaim(struct vop_reclaim_args *v)
1446{
1447 struct vnode *vp = v->a_vp;
7a2de9a4
MD
1448 struct tmpfs_mount *tmp;
1449 struct tmpfs_node *node;
1450
1451 node = VP_TO_TMPFS_NODE(vp);
1452 tmp = VFS_TO_TMPFS(vp->v_mount);
1453
7a2de9a4
MD
1454 tmpfs_free_vp(vp);
1455
b7fe63af
MD
1456 /*
1457 * If the node referenced by this vnode was deleted by the
1458 * user, we must free its associated data structures now that
1459 * the vnode is being reclaimed.
1460 *
1461 * Directories have an extra link ref.
1462 */
7a2de9a4 1463 TMPFS_NODE_LOCK(node);
b7fe63af 1464 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
6e0c5aab 1465 node->tn_links == 0) {
7a2de9a4 1466 node->tn_vpstate = TMPFS_VNODE_DOOMED;
7a2de9a4 1467 tmpfs_free_node(tmp, node);
0786baf1 1468 /* eats the lock */
9fc94b5f 1469 } else {
7a2de9a4 1470 TMPFS_NODE_UNLOCK(node);
9fc94b5f 1471 }
7a2de9a4
MD
1472
1473 KKASSERT(vp->v_data == NULL);
1474 return 0;
1475}
1476
66fa44e7
VS
1477/* --------------------------------------------------------------------- */
1478
1479static int
1480tmpfs_mountctl(struct vop_mountctl_args *ap)
1481{
1482 struct tmpfs_mount *tmp;
1483 struct mount *mp;
1484 int rc;
1485
1486 switch (ap->a_op) {
1487 case (MOUNTCTL_SET_EXPORT):
1488 mp = ap->a_head.a_ops->head.vv_mount;
1489 tmp = (struct tmpfs_mount *) mp->mnt_data;
1490
1491 if (ap->a_ctllen != sizeof(struct export_args))
1492 rc = (EINVAL);
1493 else
1494 rc = vfs_export(mp, &tmp->tm_export,
1495 (const struct export_args *) ap->a_ctl);
1496 break;
1497 default:
1498 rc = vop_stdmountctl(ap);
1499 break;
1500 }
1501 return (rc);
1502}
1503
7a2de9a4
MD
1504/* --------------------------------------------------------------------- */
1505
1506static int
1507tmpfs_print(struct vop_print_args *v)
1508{
1509 struct vnode *vp = v->a_vp;
1510
1511 struct tmpfs_node *node;
1512
1513 node = VP_TO_TMPFS_NODE(vp);
1514
1515 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1516 node, node->tn_flags, node->tn_links);
1517 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1518 node->tn_mode, node->tn_uid, node->tn_gid,
1519 (uintmax_t)node->tn_size, node->tn_status);
1520
1521 if (vp->v_type == VFIFO)
1522 fifo_printinfo(vp);
1523
1524 kprintf("\n");
1525
1526 return 0;
1527}
1528
1529/* --------------------------------------------------------------------- */
1530
1531static int
1532tmpfs_pathconf(struct vop_pathconf_args *v)
1533{
1534 int name = v->a_name;
1535 register_t *retval = v->a_retval;
1536
1537 int error;
1538
1539 error = 0;
1540
1541 switch (name) {
1542 case _PC_LINK_MAX:
1543 *retval = LINK_MAX;
1544 break;
1545
1546 case _PC_NAME_MAX:
1547 *retval = NAME_MAX;
1548 break;
1549
1550 case _PC_PATH_MAX:
1551 *retval = PATH_MAX;
1552 break;
1553
1554 case _PC_PIPE_BUF:
1555 *retval = PIPE_BUF;
1556 break;
1557
1558 case _PC_CHOWN_RESTRICTED:
1559 *retval = 1;
1560 break;
1561
1562 case _PC_NO_TRUNC:
1563 *retval = 1;
1564 break;
1565
1566 case _PC_SYNC_IO:
1567 *retval = 1;
1568 break;
1569
1570 case _PC_FILESIZEBITS:
1571 *retval = 0; /* XXX Don't know which value should I return. */
1572 break;
1573
1574 default:
1575 error = EINVAL;
1576 }
1577
1578 return error;
1579}
1580
80ae59d7
MD
1581/************************************************************************
1582 * KQFILTER OPS *
1583 ************************************************************************/
1584
1585static void filt_tmpfsdetach(struct knote *kn);
1586static int filt_tmpfsread(struct knote *kn, long hint);
1587static int filt_tmpfswrite(struct knote *kn, long hint);
1588static int filt_tmpfsvnode(struct knote *kn, long hint);
1589
1590static struct filterops tmpfsread_filtops =
1591 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1592static struct filterops tmpfswrite_filtops =
1593 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1594static struct filterops tmpfsvnode_filtops =
1595 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1596
1597static int
1598tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1599{
1600 struct vnode *vp = ap->a_vp;
1601 struct knote *kn = ap->a_kn;
1602
1603 switch (kn->kn_filter) {
1604 case EVFILT_READ:
1605 kn->kn_fop = &tmpfsread_filtops;
1606 break;
1607 case EVFILT_WRITE:
1608 kn->kn_fop = &tmpfswrite_filtops;
1609 break;
1610 case EVFILT_VNODE:
1611 kn->kn_fop = &tmpfsvnode_filtops;
1612 break;
1613 default:
1614 return (EOPNOTSUPP);
1615 }
1616
1617 kn->kn_hook = (caddr_t)vp;
1618
1619 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1620
1621 return(0);
1622}
1623
1624static void
1625filt_tmpfsdetach(struct knote *kn)
1626{
1627 struct vnode *vp = (void *)kn->kn_hook;
1628
1629 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1630}
1631
1632static int
1633filt_tmpfsread(struct knote *kn, long hint)
1634{
1635 struct vnode *vp = (void *)kn->kn_hook;
1636 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1637 off_t off;
1638
1639 if (hint == NOTE_REVOKE) {
3bcb6e5e 1640 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
80ae59d7
MD
1641 return(1);
1642 }
f79d9cc9
MD
1643
1644 /*
1645 * Interlock against MP races when performing this function.
1646 */
1647 lwkt_gettoken(&vp->v_mount->mnt_token);
80ae59d7
MD
1648 off = node->tn_size - kn->kn_fp->f_offset;
1649 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
f79d9cc9
MD
1650 if (kn->kn_sfflags & NOTE_OLDAPI) {
1651 lwkt_reltoken(&vp->v_mount->mnt_token);
80ae59d7 1652 return(1);
f79d9cc9 1653 }
80ae59d7 1654
80ae59d7 1655 if (kn->kn_data == 0) {
80ae59d7 1656 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
80ae59d7 1657 }
f79d9cc9 1658 lwkt_reltoken(&vp->v_mount->mnt_token);
80ae59d7
MD
1659 return (kn->kn_data != 0);
1660}
1661
1662static int
1663filt_tmpfswrite(struct knote *kn, long hint)
1664{
1665 if (hint == NOTE_REVOKE)
3bcb6e5e 1666 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
80ae59d7
MD
1667 kn->kn_data = 0;
1668 return (1);
1669}
1670
1671static int
1672filt_tmpfsvnode(struct knote *kn, long hint)
1673{
1674 if (kn->kn_sfflags & hint)
1675 kn->kn_fflags |= hint;
1676 if (hint == NOTE_REVOKE) {
3bcb6e5e 1677 kn->kn_flags |= (EV_EOF | EV_NODATA);
80ae59d7
MD
1678 return (1);
1679 }
1680 return (kn->kn_fflags != 0);
1681}
1682
1683
7a2de9a4
MD
1684/* --------------------------------------------------------------------- */
1685
1686/*
1687 * vnode operations vector used for files stored in a tmpfs file system.
1688 */
1689struct vop_ops tmpfs_vnode_vops = {
1690 .vop_default = vop_defaultop,
1691 .vop_getpages = vop_stdgetpages,
1692 .vop_putpages = vop_stdputpages,
1693 .vop_ncreate = tmpfs_ncreate,
1694 .vop_nresolve = tmpfs_nresolve,
1695 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1696 .vop_nmknod = tmpfs_nmknod,
1697 .vop_open = tmpfs_open,
1698 .vop_close = tmpfs_close,
1699 .vop_access = tmpfs_access,
1700 .vop_getattr = tmpfs_getattr,
1701 .vop_setattr = tmpfs_setattr,
1702 .vop_read = tmpfs_read,
1703 .vop_write = tmpfs_write,
1704 .vop_fsync = tmpfs_fsync,
66fa44e7 1705 .vop_mountctl = tmpfs_mountctl,
7a2de9a4
MD
1706 .vop_nremove = tmpfs_nremove,
1707 .vop_nlink = tmpfs_nlink,
1708 .vop_nrename = tmpfs_nrename,
1709 .vop_nmkdir = tmpfs_nmkdir,
1710 .vop_nrmdir = tmpfs_nrmdir,
1711 .vop_nsymlink = tmpfs_nsymlink,
1712 .vop_readdir = tmpfs_readdir,
1713 .vop_readlink = tmpfs_readlink,
1714 .vop_inactive = tmpfs_inactive,
1715 .vop_reclaim = tmpfs_reclaim,
1716 .vop_print = tmpfs_print,
1717 .vop_pathconf = tmpfs_pathconf,
9fc94b5f 1718 .vop_bmap = tmpfs_bmap,
7a2de9a4
MD
1719 .vop_strategy = tmpfs_strategy,
1720 .vop_advlock = tmpfs_advlock,
80ae59d7 1721 .vop_kqfilter = tmpfs_kqfilter
7a2de9a4 1722};