1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */
2
3 /*-
4 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 /*
34 * tmpfs vnode interface.
35 */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/sched.h>
46 #include <sys/sf_buf.h>
47 #include <sys/stat.h>
48 #include <sys/systm.h>
49 #include <sys/sysctl.h>
50 #include <sys/unistd.h>
51 #include <sys/vnode.h>
52
53 #include <vm/vm.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57
58 #include <machine/_inttypes.h>
59
60 #include <fs/fifofs/fifo.h>
61 #include <fs/tmpfs/tmpfs_vnops.h>
62 #include <fs/tmpfs/tmpfs.h>
63
64 SYSCTL_DECL(_vfs_tmpfs);
65
66 static volatile int tmpfs_rename_restarts;
67 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
68 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
69 "Times rename had to restart due to lock contention");
70
71 /* --------------------------------------------------------------------- */
72
73 static int
74 tmpfs_lookup(struct vop_cachedlookup_args *v)
75 {
76 struct vnode *dvp = v->a_dvp;
77 struct vnode **vpp = v->a_vpp;
78 struct componentname *cnp = v->a_cnp;
79
80 int error;
81 struct tmpfs_dirent *de;
82 struct tmpfs_node *dnode;
83
84 dnode = VP_TO_TMPFS_DIR(dvp);
85 *vpp = NULLVP;
86
87 /* Check accessibility of requested node as a first step. */
88 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
89 if (error != 0)
90 goto out;
91
92 /* We cannot be requesting the parent directory of the root node. */
93 MPASS(IMPLIES(dnode->tn_type == VDIR &&
94 dnode->tn_dir.tn_parent == dnode,
95 !(cnp->cn_flags & ISDOTDOT)));
96
97 TMPFS_ASSERT_LOCKED(dnode);
98 if (dnode->tn_dir.tn_parent == NULL) {
99 error = ENOENT;
100 goto out;
101 }
102 if (cnp->cn_flags & ISDOTDOT) {
103 int ltype = 0;
104
105 ltype = VOP_ISLOCKED(dvp);
106 vhold(dvp);
107 VOP_UNLOCK(dvp, 0);
108 /* Allocate a new vnode on the matching entry. */
109 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
110 cnp->cn_lkflags, vpp);
111
112 vn_lock(dvp, ltype | LK_RETRY);
113 vdrop(dvp);
114 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
115 VREF(dvp);
116 *vpp = dvp;
117 error = 0;
118 } else {
119 de = tmpfs_dir_lookup(dnode, NULL, cnp);
120 if (de != NULL && de->td_node == NULL)
121 cnp->cn_flags |= ISWHITEOUT;
122 if (de == NULL || de->td_node == NULL) {
123 /* The entry was not found in the directory.
124 * This is OK if we are creating or renaming an
125 * entry and are working on the last component of
126 * the path name. */
127 if ((cnp->cn_flags & ISLASTCN) &&
128 (cnp->cn_nameiop == CREATE || \
129 cnp->cn_nameiop == RENAME ||
130 (cnp->cn_nameiop == DELETE &&
131 cnp->cn_flags & DOWHITEOUT &&
132 cnp->cn_flags & ISWHITEOUT))) {
133 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
134 cnp->cn_thread);
135 if (error != 0)
136 goto out;
137
138 /* Keep the component name in the buffer for
139 * future uses. */
140 cnp->cn_flags |= SAVENAME;
141
142 error = EJUSTRETURN;
143 } else
144 error = ENOENT;
145 } else {
146 struct tmpfs_node *tnode;
147
148 /* The entry was found, so get its associated
149 * tmpfs_node. */
150 tnode = de->td_node;
151
152 /* If we are not at the last path component and
153 * found a non-directory or non-link entry (which
154 * may itself be pointing to a directory), raise
155 * an error. */
156 if ((tnode->tn_type != VDIR &&
157 tnode->tn_type != VLNK) &&
158 !(cnp->cn_flags & ISLASTCN)) {
159 error = ENOTDIR;
160 goto out;
161 }
162
163 /* If we are deleting or renaming the entry, keep
164 * track of its tmpfs_dirent so that it can be
165 * easily deleted later. */
166 if ((cnp->cn_flags & ISLASTCN) &&
167 (cnp->cn_nameiop == DELETE ||
168 cnp->cn_nameiop == RENAME)) {
169 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
170 cnp->cn_thread);
171 if (error != 0)
172 goto out;
173
174 /* Allocate a new vnode on the matching entry. */
175 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
176 cnp->cn_lkflags, vpp);
177 if (error != 0)
178 goto out;
179
180 if ((dnode->tn_mode & S_ISTXT) &&
181 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
182 VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
183 error = EPERM;
184 vput(*vpp);
185 *vpp = NULL;
186 goto out;
187 }
188 cnp->cn_flags |= SAVENAME;
189 } else {
190 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
191 cnp->cn_lkflags, vpp);
192 }
193 }
194 }
195
196 /* Store the result of this lookup in the cache. Avoid this if the
197 * request was for creation, as it does not improve timings on
198 * emprical tests. */
199 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
200 cache_enter(dvp, *vpp, cnp);
201
202 out:
203 /* If there were no errors, *vpp cannot be null and it must be
204 * locked. */
205 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
206
207 return error;
208 }
209
210 /* --------------------------------------------------------------------- */
211
212 static int
213 tmpfs_create(struct vop_create_args *v)
214 {
215 struct vnode *dvp = v->a_dvp;
216 struct vnode **vpp = v->a_vpp;
217 struct componentname *cnp = v->a_cnp;
218 struct vattr *vap = v->a_vap;
219
220 MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
221
222 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
223 }
224 /* --------------------------------------------------------------------- */
225
226 static int
227 tmpfs_mknod(struct vop_mknod_args *v)
228 {
229 struct vnode *dvp = v->a_dvp;
230 struct vnode **vpp = v->a_vpp;
231 struct componentname *cnp = v->a_cnp;
232 struct vattr *vap = v->a_vap;
233
234 if (vap->va_type != VBLK && vap->va_type != VCHR &&
235 vap->va_type != VFIFO)
236 return EINVAL;
237
238 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
239 }
240
241 /* --------------------------------------------------------------------- */
242
243 static int
244 tmpfs_open(struct vop_open_args *v)
245 {
246 struct vnode *vp = v->a_vp;
247 int mode = v->a_mode;
248
249 int error;
250 struct tmpfs_node *node;
251
252 MPASS(VOP_ISLOCKED(vp));
253
254 node = VP_TO_TMPFS_NODE(vp);
255
256 /* The file is still active but all its names have been removed
257 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
258 * it is about to die. */
259 if (node->tn_links < 1)
260 return (ENOENT);
261
262 /* If the file is marked append-only, deny write requests. */
263 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
264 error = EPERM;
265 else {
266 error = 0;
267 vnode_create_vobject(vp, node->tn_size, v->a_td);
268 }
269
270 MPASS(VOP_ISLOCKED(vp));
271 return error;
272 }
273
274 /* --------------------------------------------------------------------- */
275
276 static int
277 tmpfs_close(struct vop_close_args *v)
278 {
279 struct vnode *vp = v->a_vp;
280
281 MPASS(VOP_ISLOCKED(vp));
282
283 /* Update node times. */
284 tmpfs_update(vp);
285
286 return (0);
287 }
288
289 /* --------------------------------------------------------------------- */
290
291 int
292 tmpfs_access(struct vop_access_args *v)
293 {
294 struct vnode *vp = v->a_vp;
295 accmode_t accmode = v->a_accmode;
296 struct ucred *cred = v->a_cred;
297
298 int error;
299 struct tmpfs_node *node;
300
301 MPASS(VOP_ISLOCKED(vp));
302
303 node = VP_TO_TMPFS_NODE(vp);
304
305 switch (vp->v_type) {
306 case VDIR:
307 /* FALLTHROUGH */
308 case VLNK:
309 /* FALLTHROUGH */
310 case VREG:
311 if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
312 error = EROFS;
313 goto out;
314 }
315 break;
316
317 case VBLK:
318 /* FALLTHROUGH */
319 case VCHR:
320 /* FALLTHROUGH */
321 case VSOCK:
322 /* FALLTHROUGH */
323 case VFIFO:
324 break;
325
326 default:
327 error = EINVAL;
328 goto out;
329 }
330
331 if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
332 error = EPERM;
333 goto out;
334 }
335
336 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
337 node->tn_gid, accmode, cred, NULL);
338
339 out:
340 MPASS(VOP_ISLOCKED(vp));
341
342 return error;
343 }
344
345 /* --------------------------------------------------------------------- */
346
347 int
348 tmpfs_getattr(struct vop_getattr_args *v)
349 {
350 struct vnode *vp = v->a_vp;
351 struct vattr *vap = v->a_vap;
352
353 struct tmpfs_node *node;
354
355 node = VP_TO_TMPFS_NODE(vp);
356
357 tmpfs_update(vp);
358
359 vap->va_type = vp->v_type;
360 vap->va_mode = node->tn_mode;
361 vap->va_nlink = node->tn_links;
362 vap->va_uid = node->tn_uid;
363 vap->va_gid = node->tn_gid;
364 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
365 vap->va_fileid = node->tn_id;
366 vap->va_size = node->tn_size;
367 vap->va_blocksize = PAGE_SIZE;
368 vap->va_atime = node->tn_atime;
369 vap->va_mtime = node->tn_mtime;
370 vap->va_ctime = node->tn_ctime;
371 vap->va_birthtime = node->tn_birthtime;
372 vap->va_gen = node->tn_gen;
373 vap->va_flags = node->tn_flags;
374 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
375 node->tn_rdev : NODEV;
376 vap->va_bytes = round_page(node->tn_size);
377 vap->va_filerev = 0;
378
379 return 0;
380 }
381
382 /* --------------------------------------------------------------------- */
383
384 /* XXX Should this operation be atomic? I think it should, but code in
385 * XXX other places (e.g., ufs) doesn't seem to be... */
386 int
387 tmpfs_setattr(struct vop_setattr_args *v)
388 {
389 struct vnode *vp = v->a_vp;
390 struct vattr *vap = v->a_vap;
391 struct ucred *cred = v->a_cred;
392 struct thread *td = curthread;
393
394 int error;
395
396 MPASS(VOP_ISLOCKED(vp));
397
398 error = 0;
399
400 /* Abort if any unsettable attribute is given. */
401 if (vap->va_type != VNON ||
402 vap->va_nlink != VNOVAL ||
403 vap->va_fsid != VNOVAL ||
404 vap->va_fileid != VNOVAL ||
405 vap->va_blocksize != VNOVAL ||
406 vap->va_gen != VNOVAL ||
407 vap->va_rdev != VNOVAL ||
408 vap->va_bytes != VNOVAL)
409 error = EINVAL;
410
411 if (error == 0 && (vap->va_flags != VNOVAL))
412 error = tmpfs_chflags(vp, vap->va_flags, cred, td);
413
414 if (error == 0 && (vap->va_size != VNOVAL))
415 error = tmpfs_chsize(vp, vap->va_size, cred, td);
416
417 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
418 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
419
420 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
421 error = tmpfs_chmod(vp, vap->va_mode, cred, td);
422
423 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
424 vap->va_atime.tv_nsec != VNOVAL) ||
425 (vap->va_mtime.tv_sec != VNOVAL &&
426 vap->va_mtime.tv_nsec != VNOVAL) ||
427 (vap->va_birthtime.tv_sec != VNOVAL &&
428 vap->va_birthtime.tv_nsec != VNOVAL)))
429 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
430 &vap->va_birthtime, vap->va_vaflags, cred, td);
431
432 /* Update the node times. We give preference to the error codes
433 * generated by this function rather than the ones that may arise
434 * from tmpfs_update. */
435 tmpfs_update(vp);
436
437 MPASS(VOP_ISLOCKED(vp));
438
439 return error;
440 }
441
442 /* --------------------------------------------------------------------- */
443 static int
444 tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
445 vm_offset_t offset, size_t tlen, struct uio *uio)
446 {
447 vm_page_t m;
448 int error;
449
450 VM_OBJECT_LOCK(tobj);
451 vm_object_pip_add(tobj, 1);
452 m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
453 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
454 if (m->valid != VM_PAGE_BITS_ALL) {
455 if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
456 error = vm_pager_get_pages(tobj, &m, 1, 0);
457 if (error != 0) {
458 printf("tmpfs get pages from pager error [read]\n");
459 goto out;
460 }
461 } else
462 vm_page_zero_invalid(m, TRUE);
463 }
464 VM_OBJECT_UNLOCK(tobj);
465 error = uiomove_fromphys(&m, offset, tlen, uio);
466 VM_OBJECT_LOCK(tobj);
467 out:
468 vm_page_lock_queues();
469 vm_page_unwire(m, TRUE);
470 vm_page_unlock_queues();
471 vm_page_wakeup(m);
472 vm_object_pip_subtract(tobj, 1);
473 VM_OBJECT_UNLOCK(tobj);
474
475 return (error);
476 }
477
478 static __inline int
479 tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
480 vm_offset_t offset, size_t tlen, void *buf)
481 {
482 struct uio uio;
483 struct iovec iov;
484
485 uio.uio_iovcnt = 1;
486 uio.uio_iov = &iov;
487 iov.iov_base = buf;
488 iov.iov_len = tlen;
489
490 uio.uio_offset = 0;
491 uio.uio_resid = tlen;
492 uio.uio_rw = UIO_READ;
493 uio.uio_segflg = UIO_SYSSPACE;
494 uio.uio_td = curthread;
495
496 return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
497 }
498
499 static int
500 tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
501 {
502 struct sf_buf *sf;
503 vm_pindex_t idx;
504 vm_page_t m;
505 vm_offset_t offset;
506 off_t addr;
507 size_t tlen;
508 char *ma;
509 int error;
510
511 addr = uio->uio_offset;
512 idx = OFF_TO_IDX(addr);
513 offset = addr & PAGE_MASK;
514 tlen = MIN(PAGE_SIZE - offset, len);
515
516 if ((vobj == NULL) || (vobj->resident_page_count == 0))
517 goto nocache;
518
519 VM_OBJECT_LOCK(vobj);
520 lookupvpg:
521 if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
522 vm_page_is_valid(m, offset, tlen)) {
523 if (vm_page_sleep_if_busy(m, FALSE, "tmfsmr"))
524 goto lookupvpg;
525 vm_page_busy(m);
526 VM_OBJECT_UNLOCK(vobj);
527 error = uiomove_fromphys(&m, offset, tlen, uio);
528 VM_OBJECT_LOCK(vobj);
529 vm_page_wakeup(m);
530 VM_OBJECT_UNLOCK(vobj);
531 return (error);
532 } else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
533 KASSERT(offset == 0,
534 ("unexpected offset in tmpfs_mappedread for sendfile"));
535 if (vm_page_sleep_if_busy(m, FALSE, "tmfsmr"))
536 goto lookupvpg;
537 vm_page_busy(m);
538 VM_OBJECT_UNLOCK(vobj);
539 sched_pin();
540 sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
541 ma = (char *)sf_buf_kva(sf);
542 error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
543 if (error == 0) {
544 if (tlen != PAGE_SIZE)
545 bzero(ma + tlen, PAGE_SIZE - tlen);
546 uio->uio_offset += tlen;
547 uio->uio_resid -= tlen;
548 }
549 sf_buf_free(sf);
550 sched_unpin();
551 VM_OBJECT_LOCK(vobj);
552 if (error == 0)
553 m->valid = VM_PAGE_BITS_ALL;
554 vm_page_wakeup(m);
555 VM_OBJECT_UNLOCK(vobj);
556 return (error);
557 }
558 VM_OBJECT_UNLOCK(vobj);
559 nocache:
560 error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
561
562 return (error);
563 }
564
565 static int
566 tmpfs_read(struct vop_read_args *v)
567 {
568 struct vnode *vp = v->a_vp;
569 struct uio *uio = v->a_uio;
570
571 struct tmpfs_node *node;
572 vm_object_t uobj;
573 size_t len;
574 int resid;
575
576 int error = 0;
577
578 node = VP_TO_TMPFS_NODE(vp);
579
580 if (vp->v_type != VREG) {
581 error = EISDIR;
582 goto out;
583 }
584
585 if (uio->uio_offset < 0) {
586 error = EINVAL;
587 goto out;
588 }
589
590 node->tn_status |= TMPFS_NODE_ACCESSED;
591
592 uobj = node->tn_reg.tn_aobj;
593 while ((resid = uio->uio_resid) > 0) {
594 error = 0;
595 if (node->tn_size <= uio->uio_offset)
596 break;
597 len = MIN(node->tn_size - uio->uio_offset, resid);
598 if (len == 0)
599 break;
600 error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
601 if ((error != 0) || (resid == uio->uio_resid))
602 break;
603 }
604
605 out:
606
607 return error;
608 }
609
610 /* --------------------------------------------------------------------- */
611
612 static int
613 tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
614 {
615 vm_pindex_t idx;
616 vm_page_t vpg, tpg;
617 vm_offset_t offset;
618 off_t addr;
619 size_t tlen;
620 int error;
621
622 error = 0;
623
624 addr = uio->uio_offset;
625 idx = OFF_TO_IDX(addr);
626 offset = addr & PAGE_MASK;
627 tlen = MIN(PAGE_SIZE - offset, len);
628
629 if ((vobj == NULL) || (vobj->resident_page_count == 0)) {
630 vpg = NULL;
631 goto nocache;
632 }
633
634 VM_OBJECT_LOCK(vobj);
635 lookupvpg:
636 if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
637 vm_page_is_valid(vpg, offset, tlen)) {
638 if (vm_page_sleep_if_busy(vpg, FALSE, "tmfsmw"))
639 goto lookupvpg;
640 vm_page_busy(vpg);
641 vm_page_lock_queues();
642 vm_page_undirty(vpg);
643 vm_page_unlock_queues();
644 VM_OBJECT_UNLOCK(vobj);
645 error = uiomove_fromphys(&vpg, offset, tlen, uio);
646 } else {
647 VM_OBJECT_UNLOCK(vobj);
648 vpg = NULL;
649 }
650 nocache:
651 VM_OBJECT_LOCK(tobj);
652 vm_object_pip_add(tobj, 1);
653 tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
654 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
655 if (tpg->valid != VM_PAGE_BITS_ALL) {
656 if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
657 error = vm_pager_get_pages(tobj, &tpg, 1, 0);
658 if (error != 0) {
659 printf("tmpfs get pages from pager error [write]\n");
660 goto out;
661 }
662 } else
663 vm_page_zero_invalid(tpg, TRUE);
664 }
665 VM_OBJECT_UNLOCK(tobj);
666 if (vpg == NULL)
667 error = uiomove_fromphys(&tpg, offset, tlen, uio);
668 else {
669 KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
670 pmap_copy_page(vpg, tpg);
671 }
672 VM_OBJECT_LOCK(tobj);
673 out:
674 if (vobj != NULL)
675 VM_OBJECT_LOCK(vobj);
676 vm_page_lock_queues();
677 if (error == 0) {
678 KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
679 ("parts of tpg invalid"));
680 vm_page_dirty(tpg);
681 }
682 vm_page_unwire(tpg, TRUE);
683 vm_page_unlock_queues();
684 vm_page_wakeup(tpg);
685 if (vpg != NULL)
686 vm_page_wakeup(vpg);
687 if (vobj != NULL)
688 VM_OBJECT_UNLOCK(vobj);
689 vm_object_pip_subtract(tobj, 1);
690 VM_OBJECT_UNLOCK(tobj);
691
692 return (error);
693 }
694
695 static int
696 tmpfs_write(struct vop_write_args *v)
697 {
698 struct vnode *vp = v->a_vp;
699 struct uio *uio = v->a_uio;
700 int ioflag = v->a_ioflag;
701 struct thread *td = uio->uio_td;
702
703 boolean_t extended;
704 int error = 0;
705 off_t oldsize;
706 struct tmpfs_node *node;
707 vm_object_t uobj;
708 size_t len;
709 int resid;
710
711 node = VP_TO_TMPFS_NODE(vp);
712 oldsize = node->tn_size;
713
714 if (uio->uio_offset < 0 || vp->v_type != VREG) {
715 error = EINVAL;
716 goto out;
717 }
718
719 if (uio->uio_resid == 0) {
720 error = 0;
721 goto out;
722 }
723
724 if (ioflag & IO_APPEND)
725 uio->uio_offset = node->tn_size;
726
727 if (uio->uio_offset + uio->uio_resid >
728 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
729 return (EFBIG);
730
731 if (vn_rlimit_fsize(vp, uio, td))
732 return (EFBIG);
733
734 extended = uio->uio_offset + uio->uio_resid > node->tn_size;
735 if (extended) {
736 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
737 if (error != 0)
738 goto out;
739 }
740
741 uobj = node->tn_reg.tn_aobj;
742 while ((resid = uio->uio_resid) > 0) {
743 if (node->tn_size <= uio->uio_offset)
744 break;
745 len = MIN(node->tn_size - uio->uio_offset, resid);
746 if (len == 0)
747 break;
748 error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
749 if ((error != 0) || (resid == uio->uio_resid))
750 break;
751 }
752
753 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
754 (extended ? TMPFS_NODE_CHANGED : 0);
755
756 if (node->tn_mode & (S_ISUID | S_ISGID)) {
757 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
758 node->tn_mode &= ~(S_ISUID | S_ISGID);
759 }
760
761 if (error != 0)
762 (void)tmpfs_reg_resize(vp, oldsize);
763
764 out:
765 MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
766 MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
767
768 return error;
769 }
770
771 /* --------------------------------------------------------------------- */
772
773 static int
774 tmpfs_fsync(struct vop_fsync_args *v)
775 {
776 struct vnode *vp = v->a_vp;
777
778 MPASS(VOP_ISLOCKED(vp));
779
780 tmpfs_update(vp);
781
782 return 0;
783 }
784
785 /* --------------------------------------------------------------------- */
786
787 static int
788 tmpfs_remove(struct vop_remove_args *v)
789 {
790 struct vnode *dvp = v->a_dvp;
791 struct vnode *vp = v->a_vp;
792
793 int error;
794 struct tmpfs_dirent *de;
795 struct tmpfs_mount *tmp;
796 struct tmpfs_node *dnode;
797 struct tmpfs_node *node;
798
799 MPASS(VOP_ISLOCKED(dvp));
800 MPASS(VOP_ISLOCKED(vp));
801
802 if (vp->v_type == VDIR) {
803 error = EISDIR;
804 goto out;
805 }
806
807 dnode = VP_TO_TMPFS_DIR(dvp);
808 node = VP_TO_TMPFS_NODE(vp);
809 tmp = VFS_TO_TMPFS(vp->v_mount);
810 de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
811 MPASS(de != NULL);
812
813 /* Files marked as immutable or append-only cannot be deleted. */
814 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
815 (dnode->tn_flags & APPEND)) {
816 error = EPERM;
817 goto out;
818 }
819
820 /* Remove the entry from the directory; as it is a file, we do not
821 * have to change the number of hard links of the directory. */
822 tmpfs_dir_detach(dvp, de);
823 if (v->a_cnp->cn_flags & DOWHITEOUT)
824 tmpfs_dir_whiteout_add(dvp, v->a_cnp);
825
826 /* Free the directory entry we just deleted. Note that the node
827 * referred by it will not be removed until the vnode is really
828 * reclaimed. */
829 tmpfs_free_dirent(tmp, de, TRUE);
830
831 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
832 error = 0;
833
834 out:
835
836 return error;
837 }
838
839 /* --------------------------------------------------------------------- */
840
841 static int
842 tmpfs_link(struct vop_link_args *v)
843 {
844 struct vnode *dvp = v->a_tdvp;
845 struct vnode *vp = v->a_vp;
846 struct componentname *cnp = v->a_cnp;
847
848 int error;
849 struct tmpfs_dirent *de;
850 struct tmpfs_node *node;
851
852 MPASS(VOP_ISLOCKED(dvp));
853 MPASS(cnp->cn_flags & HASBUF);
854 MPASS(dvp != vp); /* XXX When can this be false? */
855
856 node = VP_TO_TMPFS_NODE(vp);
857
858 /* XXX: Why aren't the following two tests done by the caller? */
859
860 /* Hard links of directories are forbidden. */
861 if (vp->v_type == VDIR) {
862 error = EPERM;
863 goto out;
864 }
865
866 /* Cannot create cross-device links. */
867 if (dvp->v_mount != vp->v_mount) {
868 error = EXDEV;
869 goto out;
870 }
871
872 /* Ensure that we do not overflow the maximum number of links imposed
873 * by the system. */
874 MPASS(node->tn_links <= LINK_MAX);
875 if (node->tn_links == LINK_MAX) {
876 error = EMLINK;
877 goto out;
878 }
879
880 /* We cannot create links of files marked immutable or append-only. */
881 if (node->tn_flags & (IMMUTABLE | APPEND)) {
882 error = EPERM;
883 goto out;
884 }
885
886 /* Allocate a new directory entry to represent the node. */
887 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
888 cnp->cn_nameptr, cnp->cn_namelen, &de);
889 if (error != 0)
890 goto out;
891
892 /* Insert the new directory entry into the appropriate directory. */
893 if (cnp->cn_flags & ISWHITEOUT)
894 tmpfs_dir_whiteout_remove(dvp, cnp);
895 tmpfs_dir_attach(dvp, de);
896
897 /* vp link count has changed, so update node times. */
898 node->tn_status |= TMPFS_NODE_CHANGED;
899 tmpfs_update(vp);
900
901 error = 0;
902
903 out:
904 return error;
905 }
906
907 /* --------------------------------------------------------------------- */
908
909 /*
910 * We acquire all but fdvp locks using non-blocking acquisitions. If we
911 * fail to acquire any lock in the path we will drop all held locks,
912 * acquire the new lock in a blocking fashion, and then release it and
913 * restart the rename. This acquire/release step ensures that we do not
914 * spin on a lock waiting for release. On error release all vnode locks
915 * and decrement references the way tmpfs_rename() would do.
916 */
917 static int
918 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
919 struct vnode *tdvp, struct vnode **tvpp,
920 struct componentname *fcnp, struct componentname *tcnp)
921 {
922 struct vnode *nvp;
923 struct mount *mp;
924 struct tmpfs_dirent *de;
925 int error, restarts = 0;
926
927 VOP_UNLOCK(tdvp, 0);
928 if (*tvpp != NULL && *tvpp != tdvp)
929 VOP_UNLOCK(*tvpp, 0);
930 mp = fdvp->v_mount;
931
932 relock:
933 restarts += 1;
934 error = vn_lock(fdvp, LK_EXCLUSIVE);
935 if (error)
936 goto releout;
937 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
938 VOP_UNLOCK(fdvp, 0);
939 error = vn_lock(tdvp, LK_EXCLUSIVE);
940 if (error)
941 goto releout;
942 VOP_UNLOCK(tdvp, 0);
943 goto relock;
944 }
945 /*
946 * Re-resolve fvp to be certain it still exists and fetch the
947 * correct vnode.
948 */
949 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
950 if (de == NULL) {
951 VOP_UNLOCK(fdvp, 0);
952 VOP_UNLOCK(tdvp, 0);
953 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
954 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
955 error = EINVAL;
956 else
957 error = ENOENT;
958 goto releout;
959 }
960 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
961 if (error != 0) {
962 VOP_UNLOCK(fdvp, 0);
963 VOP_UNLOCK(tdvp, 0);
964 if (error != EBUSY)
965 goto releout;
966 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
967 if (error != 0)
968 goto releout;
969 VOP_UNLOCK(nvp, 0);
970 /*
971 * Concurrent rename race.
972 */
973 if (nvp == tdvp) {
974 vrele(nvp);
975 error = EINVAL;
976 goto releout;
977 }
978 vrele(*fvpp);
979 *fvpp = nvp;
980 goto relock;
981 }
982 vrele(*fvpp);
983 *fvpp = nvp;
984 VOP_UNLOCK(*fvpp, 0);
985 /*
986 * Re-resolve tvp and acquire the vnode lock if present.
987 */
988 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
989 /*
990 * If tvp disappeared we just carry on.
991 */
992 if (de == NULL && *tvpp != NULL) {
993 vrele(*tvpp);
994 *tvpp = NULL;
995 }
996 /*
997 * Get the tvp ino if the lookup succeeded. We may have to restart
998 * if the non-blocking acquire fails.
999 */
1000 if (de != NULL) {
1001 nvp = NULL;
1002 error = tmpfs_alloc_vp(mp, de->td_node,
1003 LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1004 if (*tvpp != NULL)
1005 vrele(*tvpp);
1006 *tvpp = nvp;
1007 if (error != 0) {
1008 VOP_UNLOCK(fdvp, 0);
1009 VOP_UNLOCK(tdvp, 0);
1010 if (error != EBUSY)
1011 goto releout;
1012 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
1013 &nvp);
1014 if (error != 0)
1015 goto releout;
1016 VOP_UNLOCK(nvp, 0);
1017 /*
1018 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
1019 */
1020 if (nvp == fdvp) {
1021 error = ENOTEMPTY;
1022 goto releout;
1023 }
1024 goto relock;
1025 }
1026 }
1027 tmpfs_rename_restarts += restarts;
1028
1029 return (0);
1030
1031 releout:
1032 vrele(fdvp);
1033 vrele(*fvpp);
1034 vrele(tdvp);
1035 if (*tvpp != NULL)
1036 vrele(*tvpp);
1037 tmpfs_rename_restarts += restarts;
1038
1039 return (error);
1040 }
1041
1042 static int
1043 tmpfs_rename(struct vop_rename_args *v)
1044 {
1045 struct vnode *fdvp = v->a_fdvp;
1046 struct vnode *fvp = v->a_fvp;
1047 struct componentname *fcnp = v->a_fcnp;
1048 struct vnode *tdvp = v->a_tdvp;
1049 struct vnode *tvp = v->a_tvp;
1050 struct componentname *tcnp = v->a_tcnp;
1051 struct mount *mp = NULL;
1052
1053 char *newname;
1054 int error;
1055 struct tmpfs_dirent *de;
1056 struct tmpfs_mount *tmp;
1057 struct tmpfs_node *fdnode;
1058 struct tmpfs_node *fnode;
1059 struct tmpfs_node *tnode;
1060 struct tmpfs_node *tdnode;
1061
1062 MPASS(VOP_ISLOCKED(tdvp));
1063 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
1064 MPASS(fcnp->cn_flags & HASBUF);
1065 MPASS(tcnp->cn_flags & HASBUF);
1066
1067 /* Disallow cross-device renames.
1068 * XXX Why isn't this done by the caller? */
1069 if (fvp->v_mount != tdvp->v_mount ||
1070 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1071 error = EXDEV;
1072 goto out;
1073 }
1074
1075 /* If source and target are the same file, there is nothing to do. */
1076 if (fvp == tvp) {
1077 error = 0;
1078 goto out;
1079 }
1080
1081 /* If we need to move the directory between entries, lock the
1082 * source so that we can safely operate on it. */
1083 if (fdvp != tdvp && fdvp != tvp) {
1084 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1085 mp = tdvp->v_mount;
1086 error = vfs_busy(mp, 0);
1087 if (error != 0) {
1088 mp = NULL;
1089 goto out;
1090 }
1091 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1092 fcnp, tcnp);
1093 if (error != 0) {
1094 vfs_unbusy(mp);
1095 return (error);
1096 }
1097 ASSERT_VOP_ELOCKED(fdvp,
1098 "tmpfs_rename: fdvp not locked");
1099 ASSERT_VOP_ELOCKED(tdvp,
1100 "tmpfs_rename: tdvp not locked");
1101 if (tvp != NULL)
1102 ASSERT_VOP_ELOCKED(tvp,
1103 "tmpfs_rename: tvp not locked");
1104 if (fvp == tvp) {
1105 error = 0;
1106 goto out_locked;
1107 }
1108 }
1109 }
1110
1111 tmp = VFS_TO_TMPFS(tdvp->v_mount);
1112 tdnode = VP_TO_TMPFS_DIR(tdvp);
1113 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
1114 fdnode = VP_TO_TMPFS_DIR(fdvp);
1115 fnode = VP_TO_TMPFS_NODE(fvp);
1116 de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
1117
1118 /* Entry can disappear before we lock fdvp,
1119 * also avoid manipulating '.' and '..' entries. */
1120 if (de == NULL) {
1121 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1122 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1123 error = EINVAL;
1124 else
1125 error = ENOENT;
1126 goto out_locked;
1127 }
1128 MPASS(de->td_node == fnode);
1129
1130 /* If re-naming a directory to another preexisting directory
1131 * ensure that the target directory is empty so that its
1132 * removal causes no side effects.
1133 * Kern_rename gurantees the destination to be a directory
1134 * if the source is one. */
1135 if (tvp != NULL) {
1136 MPASS(tnode != NULL);
1137
1138 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1139 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1140 error = EPERM;
1141 goto out_locked;
1142 }
1143
1144 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1145 if (tnode->tn_size > 0) {
1146 error = ENOTEMPTY;
1147 goto out_locked;
1148 }
1149 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1150 error = ENOTDIR;
1151 goto out_locked;
1152 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1153 error = EISDIR;
1154 goto out_locked;
1155 } else {
1156 MPASS(fnode->tn_type != VDIR &&
1157 tnode->tn_type != VDIR);
1158 }
1159 }
1160
1161 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1162 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1163 error = EPERM;
1164 goto out_locked;
1165 }
1166
1167 /* Ensure that we have enough memory to hold the new name, if it
1168 * has to be changed. */
1169 if (fcnp->cn_namelen != tcnp->cn_namelen ||
1170 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1171 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1172 } else
1173 newname = NULL;
1174
1175 /* If the node is being moved to another directory, we have to do
1176 * the move. */
1177 if (fdnode != tdnode) {
1178 /* In case we are moving a directory, we have to adjust its
1179 * parent to point to the new parent. */
1180 if (de->td_node->tn_type == VDIR) {
1181 struct tmpfs_node *n;
1182
1183 /* Ensure the target directory is not a child of the
1184 * directory being moved. Otherwise, we'd end up
1185 * with stale nodes. */
1186 n = tdnode;
1187 /* TMPFS_LOCK garanties that no nodes are freed while
1188 * traversing the list. Nodes can only be marked as
1189 * removed: tn_parent == NULL. */
1190 TMPFS_LOCK(tmp);
1191 TMPFS_NODE_LOCK(n);
1192 while (n != n->tn_dir.tn_parent) {
1193 struct tmpfs_node *parent;
1194
1195 if (n == fnode) {
1196 TMPFS_NODE_UNLOCK(n);
1197 TMPFS_UNLOCK(tmp);
1198 error = EINVAL;
1199 if (newname != NULL)
1200 free(newname, M_TMPFSNAME);
1201 goto out_locked;
1202 }
1203 parent = n->tn_dir.tn_parent;
1204 TMPFS_NODE_UNLOCK(n);
1205 if (parent == NULL) {
1206 n = NULL;
1207 break;
1208 }
1209 TMPFS_NODE_LOCK(parent);
1210 if (parent->tn_dir.tn_parent == NULL) {
1211 TMPFS_NODE_UNLOCK(parent);
1212 n = NULL;
1213 break;
1214 }
1215 n = parent;
1216 }
1217 TMPFS_UNLOCK(tmp);
1218 if (n == NULL) {
1219 error = EINVAL;
1220 if (newname != NULL)
1221 free(newname, M_TMPFSNAME);
1222 goto out_locked;
1223 }
1224 TMPFS_NODE_UNLOCK(n);
1225
1226 /* Adjust the parent pointer. */
1227 TMPFS_VALIDATE_DIR(fnode);
1228 TMPFS_NODE_LOCK(de->td_node);
1229 de->td_node->tn_dir.tn_parent = tdnode;
1230 TMPFS_NODE_UNLOCK(de->td_node);
1231
1232 /* As a result of changing the target of the '..'
1233 * entry, the link count of the source and target
1234 * directories has to be adjusted. */
1235 TMPFS_NODE_LOCK(tdnode);
1236 TMPFS_ASSERT_LOCKED(tdnode);
1237 tdnode->tn_links++;
1238 TMPFS_NODE_UNLOCK(tdnode);
1239
1240 TMPFS_NODE_LOCK(fdnode);
1241 TMPFS_ASSERT_LOCKED(fdnode);
1242 fdnode->tn_links--;
1243 TMPFS_NODE_UNLOCK(fdnode);
1244 }
1245
1246 /* Do the move: just remove the entry from the source directory
1247 * and insert it into the target one. */
1248 tmpfs_dir_detach(fdvp, de);
1249 if (fcnp->cn_flags & DOWHITEOUT)
1250 tmpfs_dir_whiteout_add(fdvp, fcnp);
1251 if (tcnp->cn_flags & ISWHITEOUT)
1252 tmpfs_dir_whiteout_remove(tdvp, tcnp);
1253 tmpfs_dir_attach(tdvp, de);
1254 }
1255
1256 /* If the name has changed, we need to make it effective by changing
1257 * it in the directory entry. */
1258 if (newname != NULL) {
1259 MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1260
1261 free(de->td_name, M_TMPFSNAME);
1262 de->td_namelen = (uint16_t)tcnp->cn_namelen;
1263 memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1264 de->td_name = newname;
1265
1266 fnode->tn_status |= TMPFS_NODE_CHANGED;
1267 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1268 }
1269
1270 /* If we are overwriting an entry, we have to remove the old one
1271 * from the target directory. */
1272 if (tvp != NULL) {
1273 /* Remove the old entry from the target directory. */
1274 de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1275 tmpfs_dir_detach(tdvp, de);
1276
1277 /* Free the directory entry we just deleted. Note that the
1278 * node referred by it will not be removed until the vnode is
1279 * really reclaimed. */
1280 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1281 }
1282 cache_purge(fvp);
1283 if (tvp != NULL)
1284 cache_purge(tvp);
1285 cache_purge_negative(tdvp);
1286
1287 error = 0;
1288
1289 out_locked:
1290 if (fdvp != tdvp && fdvp != tvp)
1291 VOP_UNLOCK(fdvp, 0);
1292
1293 out:
1294 /* Release target nodes. */
1295 /* XXX: I don't understand when tdvp can be the same as tvp, but
1296 * other code takes care of this... */
1297 if (tdvp == tvp)
1298 vrele(tdvp);
1299 else
1300 vput(tdvp);
1301 if (tvp != NULL)
1302 vput(tvp);
1303
1304 /* Release source nodes. */
1305 vrele(fdvp);
1306 vrele(fvp);
1307
1308 if (mp != NULL)
1309 vfs_unbusy(mp);
1310
1311 return error;
1312 }
1313
1314 /* --------------------------------------------------------------------- */
1315
1316 static int
1317 tmpfs_mkdir(struct vop_mkdir_args *v)
1318 {
1319 struct vnode *dvp = v->a_dvp;
1320 struct vnode **vpp = v->a_vpp;
1321 struct componentname *cnp = v->a_cnp;
1322 struct vattr *vap = v->a_vap;
1323
1324 MPASS(vap->va_type == VDIR);
1325
1326 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1327 }
1328
1329 /* --------------------------------------------------------------------- */
1330
1331 static int
1332 tmpfs_rmdir(struct vop_rmdir_args *v)
1333 {
1334 struct vnode *dvp = v->a_dvp;
1335 struct vnode *vp = v->a_vp;
1336
1337 int error;
1338 struct tmpfs_dirent *de;
1339 struct tmpfs_mount *tmp;
1340 struct tmpfs_node *dnode;
1341 struct tmpfs_node *node;
1342
1343 MPASS(VOP_ISLOCKED(dvp));
1344 MPASS(VOP_ISLOCKED(vp));
1345
1346 tmp = VFS_TO_TMPFS(dvp->v_mount);
1347 dnode = VP_TO_TMPFS_DIR(dvp);
1348 node = VP_TO_TMPFS_DIR(vp);
1349
1350 /* Directories with more than two entries ('.' and '..') cannot be
1351 * removed. */
1352 if (node->tn_size > 0) {
1353 error = ENOTEMPTY;
1354 goto out;
1355 }
1356
1357 if ((dnode->tn_flags & APPEND)
1358 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1359 error = EPERM;
1360 goto out;
1361 }
1362
1363 /* This invariant holds only if we are not trying to remove "..".
1364 * We checked for that above so this is safe now. */
1365 MPASS(node->tn_dir.tn_parent == dnode);
1366
1367 /* Get the directory entry associated with node (vp). This was
1368 * filled by tmpfs_lookup while looking up the entry. */
1369 de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1370 MPASS(TMPFS_DIRENT_MATCHES(de,
1371 v->a_cnp->cn_nameptr,
1372 v->a_cnp->cn_namelen));
1373
1374 /* Check flags to see if we are allowed to remove the directory. */
1375 if (dnode->tn_flags & APPEND
1376 || node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1377 error = EPERM;
1378 goto out;
1379 }
1380
1381
1382 /* Detach the directory entry from the directory (dnode). */
1383 tmpfs_dir_detach(dvp, de);
1384 if (v->a_cnp->cn_flags & DOWHITEOUT)
1385 tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1386
1387 /* No vnode should be allocated for this entry from this point */
1388 TMPFS_NODE_LOCK(node);
1389 TMPFS_ASSERT_ELOCKED(node);
1390 node->tn_links--;
1391 node->tn_dir.tn_parent = NULL;
1392 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1393 TMPFS_NODE_MODIFIED;
1394
1395 TMPFS_NODE_UNLOCK(node);
1396
1397 TMPFS_NODE_LOCK(dnode);
1398 TMPFS_ASSERT_ELOCKED(dnode);
1399 dnode->tn_links--;
1400 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1401 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1402 TMPFS_NODE_UNLOCK(dnode);
1403
1404 cache_purge(dvp);
1405 cache_purge(vp);
1406
1407 /* Free the directory entry we just deleted. Note that the node
1408 * referred by it will not be removed until the vnode is really
1409 * reclaimed. */
1410 tmpfs_free_dirent(tmp, de, TRUE);
1411
1412 /* Release the deleted vnode (will destroy the node, notify
1413 * interested parties and clean it from the cache). */
1414
1415 dnode->tn_status |= TMPFS_NODE_CHANGED;
1416 tmpfs_update(dvp);
1417
1418 error = 0;
1419
1420 out:
1421 return error;
1422 }
1423
1424 /* --------------------------------------------------------------------- */
1425
1426 static int
1427 tmpfs_symlink(struct vop_symlink_args *v)
1428 {
1429 struct vnode *dvp = v->a_dvp;
1430 struct vnode **vpp = v->a_vpp;
1431 struct componentname *cnp = v->a_cnp;
1432 struct vattr *vap = v->a_vap;
1433 char *target = v->a_target;
1434
1435 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1436 MPASS(vap->va_type == VLNK);
1437 #else
1438 vap->va_type = VLNK;
1439 #endif
1440
1441 return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1442 }
1443
1444 /* --------------------------------------------------------------------- */
1445
1446 static int
1447 tmpfs_readdir(struct vop_readdir_args *v)
1448 {
1449 struct vnode *vp = v->a_vp;
1450 struct uio *uio = v->a_uio;
1451 int *eofflag = v->a_eofflag;
1452 u_long **cookies = v->a_cookies;
1453 int *ncookies = v->a_ncookies;
1454
1455 int error;
1456 off_t startoff;
1457 off_t cnt = 0;
1458 struct tmpfs_node *node;
1459
1460 /* This operation only makes sense on directory nodes. */
1461 if (vp->v_type != VDIR)
1462 return ENOTDIR;
1463
1464 node = VP_TO_TMPFS_DIR(vp);
1465
1466 startoff = uio->uio_offset;
1467
1468 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1469 error = tmpfs_dir_getdotdent(node, uio);
1470 if (error != 0)
1471 goto outok;
1472 cnt++;
1473 }
1474
1475 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1476 error = tmpfs_dir_getdotdotdent(node, uio);
1477 if (error != 0)
1478 goto outok;
1479 cnt++;
1480 }
1481
1482 error = tmpfs_dir_getdents(node, uio, &cnt);
1483
1484 outok:
1485 MPASS(error >= -1);
1486
1487 if (error == -1)
1488 error = (cnt != 0) ? 0 : EINVAL;
1489
1490 if (eofflag != NULL)
1491 *eofflag =
1492 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1493
1494 /* Update NFS-related variables. */
1495 if (error == 0 && cookies != NULL && ncookies != NULL) {
1496 off_t i;
1497 off_t off = startoff;
1498 struct tmpfs_dirent *de = NULL;
1499
1500 *ncookies = cnt;
1501 *cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1502
1503 for (i = 0; i < cnt; i++) {
1504 MPASS(off != TMPFS_DIRCOOKIE_EOF);
1505 if (off == TMPFS_DIRCOOKIE_DOT) {
1506 off = TMPFS_DIRCOOKIE_DOTDOT;
1507 } else {
1508 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1509 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1510 } else if (de != NULL) {
1511 de = TAILQ_NEXT(de, td_entries);
1512 } else {
1513 de = tmpfs_dir_lookupbycookie(node,
1514 off);
1515 MPASS(de != NULL);
1516 de = TAILQ_NEXT(de, td_entries);
1517 }
1518 if (de == NULL)
1519 off = TMPFS_DIRCOOKIE_EOF;
1520 else
1521 off = tmpfs_dircookie(de);
1522 }
1523
1524 (*cookies)[i] = off;
1525 }
1526 MPASS(uio->uio_offset == off);
1527 }
1528
1529 return error;
1530 }
1531
1532 /* --------------------------------------------------------------------- */
1533
1534 static int
1535 tmpfs_readlink(struct vop_readlink_args *v)
1536 {
1537 struct vnode *vp = v->a_vp;
1538 struct uio *uio = v->a_uio;
1539
1540 int error;
1541 struct tmpfs_node *node;
1542
1543 MPASS(uio->uio_offset == 0);
1544 MPASS(vp->v_type == VLNK);
1545
1546 node = VP_TO_TMPFS_NODE(vp);
1547
1548 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1549 uio);
1550 node->tn_status |= TMPFS_NODE_ACCESSED;
1551
1552 return error;
1553 }
1554
1555 /* --------------------------------------------------------------------- */
1556
1557 static int
1558 tmpfs_inactive(struct vop_inactive_args *v)
1559 {
1560 struct vnode *vp = v->a_vp;
1561 struct thread *l = v->a_td;
1562
1563 struct tmpfs_node *node;
1564
1565 MPASS(VOP_ISLOCKED(vp));
1566
1567 node = VP_TO_TMPFS_NODE(vp);
1568
1569 if (node->tn_links == 0)
1570 vrecycle(vp, l);
1571
1572 return 0;
1573 }
1574
1575 /* --------------------------------------------------------------------- */
1576
1577 int
1578 tmpfs_reclaim(struct vop_reclaim_args *v)
1579 {
1580 struct vnode *vp = v->a_vp;
1581
1582 struct tmpfs_mount *tmp;
1583 struct tmpfs_node *node;
1584
1585 node = VP_TO_TMPFS_NODE(vp);
1586 tmp = VFS_TO_TMPFS(vp->v_mount);
1587
1588 vnode_destroy_vobject(vp);
1589 cache_purge(vp);
1590
1591 TMPFS_NODE_LOCK(node);
1592 TMPFS_ASSERT_ELOCKED(node);
1593 tmpfs_free_vp(vp);
1594
1595 /* If the node referenced by this vnode was deleted by the user,
1596 * we must free its associated data structures (now that the vnode
1597 * is being reclaimed). */
1598 if (node->tn_links == 0 &&
1599 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1600 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1601 TMPFS_NODE_UNLOCK(node);
1602 tmpfs_free_node(tmp, node);
1603 } else
1604 TMPFS_NODE_UNLOCK(node);
1605
1606 MPASS(vp->v_data == NULL);
1607 return 0;
1608 }
1609
1610 /* --------------------------------------------------------------------- */
1611
1612 static int
1613 tmpfs_print(struct vop_print_args *v)
1614 {
1615 struct vnode *vp = v->a_vp;
1616
1617 struct tmpfs_node *node;
1618
1619 node = VP_TO_TMPFS_NODE(vp);
1620
1621 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1622 node, node->tn_flags, node->tn_links);
1623 printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1624 ", status 0x%x\n",
1625 node->tn_mode, node->tn_uid, node->tn_gid,
1626 (uintmax_t)node->tn_size, node->tn_status);
1627
1628 if (vp->v_type == VFIFO)
1629 fifo_printinfo(vp);
1630
1631 printf("\n");
1632
1633 return 0;
1634 }
1635
1636 /* --------------------------------------------------------------------- */
1637
1638 static int
1639 tmpfs_pathconf(struct vop_pathconf_args *v)
1640 {
1641 int name = v->a_name;
1642 register_t *retval = v->a_retval;
1643
1644 int error;
1645
1646 error = 0;
1647
1648 switch (name) {
1649 case _PC_LINK_MAX:
1650 *retval = LINK_MAX;
1651 break;
1652
1653 case _PC_NAME_MAX:
1654 *retval = NAME_MAX;
1655 break;
1656
1657 case _PC_PATH_MAX:
1658 *retval = PATH_MAX;
1659 break;
1660
1661 case _PC_PIPE_BUF:
1662 *retval = PIPE_BUF;
1663 break;
1664
1665 case _PC_CHOWN_RESTRICTED:
1666 *retval = 1;
1667 break;
1668
1669 case _PC_NO_TRUNC:
1670 *retval = 1;
1671 break;
1672
1673 case _PC_SYNC_IO:
1674 *retval = 1;
1675 break;
1676
1677 case _PC_FILESIZEBITS:
1678 *retval = 0; /* XXX Don't know which value should I return. */
1679 break;
1680
1681 default:
1682 error = EINVAL;
1683 }
1684
1685 return error;
1686 }
1687
1688 static int
1689 tmpfs_vptofh(struct vop_vptofh_args *ap)
1690 {
1691 struct tmpfs_fid *tfhp;
1692 struct tmpfs_node *node;
1693
1694 tfhp = (struct tmpfs_fid *)ap->a_fhp;
1695 node = VP_TO_TMPFS_NODE(ap->a_vp);
1696
1697 tfhp->tf_len = sizeof(struct tmpfs_fid);
1698 tfhp->tf_id = node->tn_id;
1699 tfhp->tf_gen = node->tn_gen;
1700
1701 return (0);
1702 }
1703
1704 static int
1705 tmpfs_whiteout(struct vop_whiteout_args *ap)
1706 {
1707 struct vnode *dvp = ap->a_dvp;
1708 struct componentname *cnp = ap->a_cnp;
1709 struct tmpfs_dirent *de;
1710
1711 switch (ap->a_flags) {
1712 case LOOKUP:
1713 return (0);
1714 case CREATE:
1715 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1716 if (de != NULL)
1717 return (de->td_node == NULL ? 0 : EEXIST);
1718 return (tmpfs_dir_whiteout_add(dvp, cnp));
1719 case DELETE:
1720 tmpfs_dir_whiteout_remove(dvp, cnp);
1721 return (0);
1722 default:
1723 panic("tmpfs_whiteout: unknown op");
1724 }
1725 }
1726
1727 /* --------------------------------------------------------------------- */
1728
1729 /*
1730 * vnode operations vector used for files stored in a tmpfs file system.
1731 */
1732 struct vop_vector tmpfs_vnodeop_entries = {
1733 .vop_default = &default_vnodeops,
1734 .vop_lookup = vfs_cache_lookup,
1735 .vop_cachedlookup = tmpfs_lookup,
1736 .vop_create = tmpfs_create,
1737 .vop_mknod = tmpfs_mknod,
1738 .vop_open = tmpfs_open,
1739 .vop_close = tmpfs_close,
1740 .vop_access = tmpfs_access,
1741 .vop_getattr = tmpfs_getattr,
1742 .vop_setattr = tmpfs_setattr,
1743 .vop_read = tmpfs_read,
1744 .vop_write = tmpfs_write,
1745 .vop_fsync = tmpfs_fsync,
1746 .vop_remove = tmpfs_remove,
1747 .vop_link = tmpfs_link,
1748 .vop_rename = tmpfs_rename,
1749 .vop_mkdir = tmpfs_mkdir,
1750 .vop_rmdir = tmpfs_rmdir,
1751 .vop_symlink = tmpfs_symlink,
1752 .vop_readdir = tmpfs_readdir,
1753 .vop_readlink = tmpfs_readlink,
1754 .vop_inactive = tmpfs_inactive,
1755 .vop_reclaim = tmpfs_reclaim,
1756 .vop_print = tmpfs_print,
1757 .vop_pathconf = tmpfs_pathconf,
1758 .vop_vptofh = tmpfs_vptofh,
1759 .vop_whiteout = tmpfs_whiteout,
1760 .vop_bmap = VOP_EOPNOTSUPP,
1761 };
1762
Cache object: 68d639d0aca6bcf8c83ab2d155050289
|