1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */
2
3 /*-
4 * SPDX-License-Identifier: BSD-2-Clause
5 *
6 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * 2005 program.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
33 */
34
35 /*
36 * tmpfs vnode interface.
37 */
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/dirent.h>
42 #include <sys/extattr.h>
43 #include <sys/fcntl.h>
44 #include <sys/file.h>
45 #include <sys/filio.h>
46 #include <sys/limits.h>
47 #include <sys/lockf.h>
48 #include <sys/lock.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/priv.h>
52 #include <sys/proc.h>
53 #include <sys/rwlock.h>
54 #include <sys/sched.h>
55 #include <sys/smr.h>
56 #include <sys/stat.h>
57 #include <sys/sysctl.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <security/audit/audit.h>
61 #include <security/mac/mac_framework.h>
62
63 #include <vm/vm.h>
64 #include <vm/vm_param.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pager.h>
68 #include <vm/swap_pager.h>
69
70 #include <fs/tmpfs/tmpfs_vnops.h>
71 #include <fs/tmpfs/tmpfs.h>
72
73 SYSCTL_DECL(_vfs_tmpfs);
74 VFS_SMR_DECLARE;
75
76 static volatile int tmpfs_rename_restarts;
77 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
78 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
79 "Times rename had to restart due to lock contention");
80
81 MALLOC_DEFINE(M_TMPFSEA, "tmpfs extattr", "tmpfs extattr structure");
82
83 static int
tmpfs_vn_get_ino_alloc(struct mount * mp,void * arg,int lkflags,struct vnode ** rvp)84 tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
85 struct vnode **rvp)
86 {
87
88 return (tmpfs_alloc_vp(mp, arg, lkflags, rvp));
89 }
90
91 static int
tmpfs_lookup1(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp)92 tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
93 {
94 struct tmpfs_dirent *de;
95 struct tmpfs_node *dnode, *pnode;
96 struct tmpfs_mount *tm;
97 int error;
98
99 /* Caller assumes responsibility for ensuring access (VEXEC). */
100 dnode = VP_TO_TMPFS_DIR(dvp);
101 *vpp = NULLVP;
102
103 /* We cannot be requesting the parent directory of the root node. */
104 MPASS(IMPLIES(dnode->tn_type == VDIR &&
105 dnode->tn_dir.tn_parent == dnode,
106 !(cnp->cn_flags & ISDOTDOT)));
107
108 TMPFS_ASSERT_LOCKED(dnode);
109 if (dnode->tn_dir.tn_parent == NULL) {
110 error = ENOENT;
111 goto out;
112 }
113 if (cnp->cn_flags & ISDOTDOT) {
114 tm = VFS_TO_TMPFS(dvp->v_mount);
115 pnode = dnode->tn_dir.tn_parent;
116 tmpfs_ref_node(pnode);
117 error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc,
118 pnode, cnp->cn_lkflags, vpp);
119 tmpfs_free_node(tm, pnode);
120 if (error != 0)
121 goto out;
122 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
123 VREF(dvp);
124 *vpp = dvp;
125 error = 0;
126 } else {
127 de = tmpfs_dir_lookup(dnode, NULL, cnp);
128 if (de != NULL && de->td_node == NULL)
129 cnp->cn_flags |= ISWHITEOUT;
130 if (de == NULL || de->td_node == NULL) {
131 /*
132 * The entry was not found in the directory.
133 * This is OK if we are creating or renaming an
134 * entry and are working on the last component of
135 * the path name.
136 */
137 if ((cnp->cn_flags & ISLASTCN) &&
138 (cnp->cn_nameiop == CREATE || \
139 cnp->cn_nameiop == RENAME ||
140 (cnp->cn_nameiop == DELETE &&
141 cnp->cn_flags & DOWHITEOUT &&
142 cnp->cn_flags & ISWHITEOUT))) {
143 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
144 curthread);
145 if (error != 0)
146 goto out;
147
148 error = EJUSTRETURN;
149 } else
150 error = ENOENT;
151 } else {
152 struct tmpfs_node *tnode;
153
154 /*
155 * The entry was found, so get its associated
156 * tmpfs_node.
157 */
158 tnode = de->td_node;
159
160 /*
161 * If we are not at the last path component and
162 * found a non-directory or non-link entry (which
163 * may itself be pointing to a directory), raise
164 * an error.
165 */
166 if ((tnode->tn_type != VDIR &&
167 tnode->tn_type != VLNK) &&
168 !(cnp->cn_flags & ISLASTCN)) {
169 error = ENOTDIR;
170 goto out;
171 }
172
173 /*
174 * If we are deleting or renaming the entry, keep
175 * track of its tmpfs_dirent so that it can be
176 * easily deleted later.
177 */
178 if ((cnp->cn_flags & ISLASTCN) &&
179 (cnp->cn_nameiop == DELETE ||
180 cnp->cn_nameiop == RENAME)) {
181 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
182 curthread);
183 if (error != 0)
184 goto out;
185
186 /* Allocate a new vnode on the matching entry. */
187 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
188 cnp->cn_lkflags, vpp);
189 if (error != 0)
190 goto out;
191
192 if ((dnode->tn_mode & S_ISTXT) &&
193 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred,
194 curthread) && VOP_ACCESS(*vpp, VADMIN,
195 cnp->cn_cred, curthread)) {
196 error = EPERM;
197 vput(*vpp);
198 *vpp = NULL;
199 goto out;
200 }
201 } else {
202 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
203 cnp->cn_lkflags, vpp);
204 if (error != 0)
205 goto out;
206 }
207 }
208 }
209
210 /*
211 * Store the result of this lookup in the cache. Avoid this if the
212 * request was for creation, as it does not improve timings on
213 * emprical tests.
214 */
215 if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
216 cache_enter(dvp, *vpp, cnp);
217
218 out:
219 #ifdef INVARIANTS
220 /*
221 * If there were no errors, *vpp cannot be null and it must be
222 * locked.
223 */
224 if (error == 0) {
225 MPASS(*vpp != NULLVP);
226 ASSERT_VOP_LOCKED(*vpp, __func__);
227 } else {
228 MPASS(*vpp == NULL);
229 }
230 #endif
231
232 return (error);
233 }
234
235 static int
tmpfs_cached_lookup(struct vop_cachedlookup_args * v)236 tmpfs_cached_lookup(struct vop_cachedlookup_args *v)
237 {
238
239 return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
240 }
241
242 static int
tmpfs_lookup(struct vop_lookup_args * v)243 tmpfs_lookup(struct vop_lookup_args *v)
244 {
245 struct vnode *dvp = v->a_dvp;
246 struct vnode **vpp = v->a_vpp;
247 struct componentname *cnp = v->a_cnp;
248 int error;
249
250 /* Check accessibility of requested node as a first step. */
251 error = vn_dir_check_exec(dvp, cnp);
252 if (error != 0)
253 return (error);
254
255 return (tmpfs_lookup1(dvp, vpp, cnp));
256 }
257
258 static int
tmpfs_create(struct vop_create_args * v)259 tmpfs_create(struct vop_create_args *v)
260 {
261 struct vnode *dvp = v->a_dvp;
262 struct vnode **vpp = v->a_vpp;
263 struct componentname *cnp = v->a_cnp;
264 struct vattr *vap = v->a_vap;
265 int error;
266
267 MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
268
269 error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
270 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
271 cache_enter(dvp, *vpp, cnp);
272 return (error);
273 }
274
275 static int
tmpfs_mknod(struct vop_mknod_args * v)276 tmpfs_mknod(struct vop_mknod_args *v)
277 {
278 struct vnode *dvp = v->a_dvp;
279 struct vnode **vpp = v->a_vpp;
280 struct componentname *cnp = v->a_cnp;
281 struct vattr *vap = v->a_vap;
282
283 if (!VATTR_ISDEV(vap) && vap->va_type != VFIFO)
284 return (EINVAL);
285
286 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL));
287 }
288
289 struct fileops tmpfs_fnops;
290
291 static int
tmpfs_open(struct vop_open_args * v)292 tmpfs_open(struct vop_open_args *v)
293 {
294 struct vnode *vp;
295 struct tmpfs_node *node;
296 struct file *fp;
297 int error, mode;
298
299 vp = v->a_vp;
300 mode = v->a_mode;
301 node = VP_TO_TMPFS_NODE(vp);
302
303 /*
304 * The file is still active but all its names have been removed
305 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
306 * it is about to die.
307 */
308 if (node->tn_links < 1)
309 return (ENOENT);
310
311 /* If the file is marked append-only, deny write requests. */
312 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
313 error = EPERM;
314 else {
315 error = 0;
316 /* For regular files, the call below is nop. */
317 KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags &
318 OBJ_DEAD) == 0, ("dead object"));
319 vnode_create_vobject(vp, node->tn_size, v->a_td);
320 }
321
322 fp = v->a_fp;
323 MPASS(fp == NULL || fp->f_data == NULL);
324 if (error == 0 && fp != NULL && vp->v_type == VREG) {
325 tmpfs_ref_node(node);
326 finit_vnode(fp, mode, node, &tmpfs_fnops);
327 }
328
329 return (error);
330 }
331
332 static int
tmpfs_close(struct vop_close_args * v)333 tmpfs_close(struct vop_close_args *v)
334 {
335 struct vnode *vp = v->a_vp;
336
337 /* Update node times. */
338 tmpfs_update(vp);
339
340 return (0);
341 }
342
343 int
tmpfs_fo_close(struct file * fp,struct thread * td)344 tmpfs_fo_close(struct file *fp, struct thread *td)
345 {
346 struct tmpfs_node *node;
347
348 node = fp->f_data;
349 if (node != NULL) {
350 MPASS(node->tn_type == VREG);
351 tmpfs_free_node(node->tn_reg.tn_tmp, node);
352 }
353 return (vnops.fo_close(fp, td));
354 }
355
356 /*
357 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
358 * the comment above cache_fplookup for details.
359 */
360 int
tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args * v)361 tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args *v)
362 {
363 struct vnode *vp;
364 struct tmpfs_node *node;
365 struct ucred *cred;
366 mode_t all_x, mode;
367
368 vp = v->a_vp;
369 node = VP_TO_TMPFS_NODE_SMR(vp);
370 if (__predict_false(node == NULL))
371 return (EAGAIN);
372
373 all_x = S_IXUSR | S_IXGRP | S_IXOTH;
374 mode = atomic_load_short(&node->tn_mode);
375 if (__predict_true((mode & all_x) == all_x))
376 return (0);
377
378 cred = v->a_cred;
379 return (vaccess_vexec_smr(mode, node->tn_uid, node->tn_gid, cred));
380 }
381
382 static int
tmpfs_access_locked(struct vnode * vp,struct tmpfs_node * node,accmode_t accmode,struct ucred * cred)383 tmpfs_access_locked(struct vnode *vp, struct tmpfs_node *node,
384 accmode_t accmode, struct ucred *cred)
385 {
386 #ifdef INVARIANTS
387 if (!mtx_owned(TMPFS_NODE_MTX(node))) {
388 ASSERT_VOP_LOCKED(vp,
389 "tmpfs_access_locked needs locked vnode or node");
390 }
391 #endif
392
393 if ((accmode & VWRITE) != 0 && (node->tn_flags & IMMUTABLE) != 0)
394 return (EPERM);
395 return (vaccess(vp->v_type, node->tn_mode, node->tn_uid, node->tn_gid,
396 accmode, cred));
397 }
398
399 int
tmpfs_access(struct vop_access_args * v)400 tmpfs_access(struct vop_access_args *v)
401 {
402 struct vnode *vp = v->a_vp;
403 struct ucred *cred = v->a_cred;
404 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
405 mode_t all_x = S_IXUSR | S_IXGRP | S_IXOTH;
406 accmode_t accmode = v->a_accmode;
407
408 /*
409 * Common case path lookup.
410 */
411 if (__predict_true(accmode == VEXEC &&
412 (node->tn_mode & all_x) == all_x))
413 return (0);
414
415 switch (vp->v_type) {
416 case VDIR:
417 /* FALLTHROUGH */
418 case VLNK:
419 /* FALLTHROUGH */
420 case VREG:
421 if ((accmode & VWRITE) != 0 &&
422 (vp->v_mount->mnt_flag & MNT_RDONLY) != 0)
423 return (EROFS);
424 break;
425
426 case VBLK:
427 /* FALLTHROUGH */
428 case VCHR:
429 /* FALLTHROUGH */
430 case VSOCK:
431 /* FALLTHROUGH */
432 case VFIFO:
433 break;
434
435 default:
436 return (EINVAL);
437 }
438
439 return (tmpfs_access_locked(vp, node, accmode, cred));
440 }
441
442 int
tmpfs_stat(struct vop_stat_args * v)443 tmpfs_stat(struct vop_stat_args *v)
444 {
445 struct vnode *vp = v->a_vp;
446 struct stat *sb = v->a_sb;
447 struct tmpfs_node *node;
448 int error;
449
450 node = VP_TO_TMPFS_NODE(vp);
451
452 tmpfs_update_getattr(vp);
453
454 error = vop_stat_helper_pre(v);
455 if (__predict_false(error))
456 return (error);
457
458 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
459 sb->st_ino = node->tn_id;
460 sb->st_mode = node->tn_mode | VTTOIF(vp->v_type);
461 sb->st_nlink = node->tn_links;
462 sb->st_uid = node->tn_uid;
463 sb->st_gid = node->tn_gid;
464 sb->st_rdev = VN_ISDEV(vp) ? node->tn_rdev : NODEV;
465 sb->st_size = node->tn_size;
466 sb->st_atim.tv_sec = node->tn_atime.tv_sec;
467 sb->st_atim.tv_nsec = node->tn_atime.tv_nsec;
468 sb->st_mtim.tv_sec = node->tn_mtime.tv_sec;
469 sb->st_mtim.tv_nsec = node->tn_mtime.tv_nsec;
470 sb->st_ctim.tv_sec = node->tn_ctime.tv_sec;
471 sb->st_ctim.tv_nsec = node->tn_ctime.tv_nsec;
472 sb->st_birthtim.tv_sec = node->tn_birthtime.tv_sec;
473 sb->st_birthtim.tv_nsec = node->tn_birthtime.tv_nsec;
474 sb->st_blksize = PAGE_SIZE;
475 sb->st_flags = node->tn_flags;
476 sb->st_gen = node->tn_gen;
477 sb->st_filerev = 0;
478 if (vp->v_type == VREG) {
479 #ifdef __ILP32__
480 vm_object_t obj = node->tn_reg.tn_aobj;
481
482 /* Handle torn read */
483 VM_OBJECT_RLOCK(obj);
484 #endif
485 sb->st_blocks = ptoa(node->tn_reg.tn_pages);
486 #ifdef __ILP32__
487 VM_OBJECT_RUNLOCK(obj);
488 #endif
489 } else {
490 sb->st_blocks = node->tn_size;
491 }
492 sb->st_blocks /= S_BLKSIZE;
493 return (vop_stat_helper_post(v, error));
494 }
495
496 int
tmpfs_getattr(struct vop_getattr_args * v)497 tmpfs_getattr(struct vop_getattr_args *v)
498 {
499 struct vnode *vp = v->a_vp;
500 struct vattr *vap = v->a_vap;
501 struct tmpfs_node *node;
502
503 node = VP_TO_TMPFS_NODE(vp);
504
505 tmpfs_update_getattr(vp);
506
507 vap->va_type = vp->v_type;
508 vap->va_mode = node->tn_mode;
509 vap->va_nlink = node->tn_links;
510 vap->va_uid = node->tn_uid;
511 vap->va_gid = node->tn_gid;
512 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
513 vap->va_fileid = node->tn_id;
514 vap->va_size = node->tn_size;
515 vap->va_blocksize = PAGE_SIZE;
516 vap->va_atime = node->tn_atime;
517 vap->va_mtime = node->tn_mtime;
518 vap->va_ctime = node->tn_ctime;
519 vap->va_birthtime = node->tn_birthtime;
520 vap->va_gen = node->tn_gen;
521 vap->va_flags = node->tn_flags;
522 vap->va_rdev = VN_ISDEV(vp) ? node->tn_rdev : NODEV;
523 if (vp->v_type == VREG) {
524 #ifdef __ILP32__
525 vm_object_t obj = node->tn_reg.tn_aobj;
526
527 VM_OBJECT_RLOCK(obj);
528 #endif
529 vap->va_bytes = ptoa(node->tn_reg.tn_pages);
530 #ifdef __ILP32__
531 VM_OBJECT_RUNLOCK(obj);
532 #endif
533 } else {
534 vap->va_bytes = node->tn_size;
535 }
536 vap->va_filerev = 0;
537
538 return (0);
539 }
540
541 int
tmpfs_setattr(struct vop_setattr_args * v)542 tmpfs_setattr(struct vop_setattr_args *v)
543 {
544 struct vnode *vp = v->a_vp;
545 struct vattr *vap = v->a_vap;
546 struct ucred *cred = v->a_cred;
547 struct thread *td = curthread;
548
549 int error;
550
551 ASSERT_VOP_IN_SEQC(vp);
552
553 error = 0;
554
555 /* Abort if any unsettable attribute is given. */
556 if (vap->va_type != VNON ||
557 vap->va_nlink != VNOVAL ||
558 vap->va_fsid != VNOVAL ||
559 vap->va_fileid != VNOVAL ||
560 vap->va_blocksize != VNOVAL ||
561 vap->va_gen != VNOVAL ||
562 vap->va_rdev != VNOVAL ||
563 vap->va_bytes != VNOVAL)
564 error = EINVAL;
565
566 if (error == 0 && (vap->va_flags != VNOVAL))
567 error = tmpfs_chflags(vp, vap->va_flags, cred, td);
568
569 if (error == 0 && (vap->va_size != VNOVAL))
570 error = tmpfs_chsize(vp, vap->va_size, cred, td);
571
572 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
573 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
574
575 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
576 error = tmpfs_chmod(vp, vap->va_mode, cred, td);
577
578 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
579 vap->va_atime.tv_nsec != VNOVAL) ||
580 (vap->va_mtime.tv_sec != VNOVAL &&
581 vap->va_mtime.tv_nsec != VNOVAL) ||
582 (vap->va_birthtime.tv_sec != VNOVAL &&
583 vap->va_birthtime.tv_nsec != VNOVAL)))
584 error = tmpfs_chtimes(vp, vap, cred, td);
585
586 /*
587 * Update the node times. We give preference to the error codes
588 * generated by this function rather than the ones that may arise
589 * from tmpfs_update.
590 */
591 tmpfs_update(vp);
592
593 return (error);
594 }
595
596 static int
tmpfs_read(struct vop_read_args * v)597 tmpfs_read(struct vop_read_args *v)
598 {
599 struct vnode *vp;
600 struct uio *uio;
601 struct tmpfs_node *node;
602
603 vp = v->a_vp;
604 if (vp->v_type != VREG)
605 return (EISDIR);
606 uio = v->a_uio;
607 if (uio->uio_offset < 0)
608 return (EINVAL);
609 node = VP_TO_TMPFS_NODE(vp);
610 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node);
611 return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio));
612 }
613
614 static int
tmpfs_read_pgcache(struct vop_read_pgcache_args * v)615 tmpfs_read_pgcache(struct vop_read_pgcache_args *v)
616 {
617 struct vnode *vp;
618 struct tmpfs_node *node;
619 vm_object_t object;
620 off_t size;
621 int error;
622
623 vp = v->a_vp;
624 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp);
625
626 if (v->a_uio->uio_offset < 0)
627 return (EINVAL);
628
629 error = EJUSTRETURN;
630 vfs_smr_enter();
631
632 node = VP_TO_TMPFS_NODE_SMR(vp);
633 if (node == NULL)
634 goto out_smr;
635 MPASS(node->tn_type == VREG);
636 MPASS(node->tn_refcount >= 1);
637 object = node->tn_reg.tn_aobj;
638 if (object == NULL)
639 goto out_smr;
640
641 MPASS(object->type == tmpfs_pager_type);
642 MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) ==
643 OBJ_SWAP);
644 if (!VN_IS_DOOMED(vp)) {
645 /* size cannot become shorter due to rangelock. */
646 size = node->tn_size;
647 tmpfs_set_accessed(node->tn_reg.tn_tmp, node);
648 vfs_smr_exit();
649 error = uiomove_object(object, size, v->a_uio);
650 return (error);
651 }
652 out_smr:
653 vfs_smr_exit();
654 return (error);
655 }
656
657 static int
tmpfs_write(struct vop_write_args * v)658 tmpfs_write(struct vop_write_args *v)
659 {
660 struct vnode *vp;
661 struct uio *uio;
662 struct tmpfs_node *node;
663 off_t oldsize;
664 ssize_t r;
665 int error, ioflag;
666 mode_t newmode;
667
668 vp = v->a_vp;
669 uio = v->a_uio;
670 ioflag = v->a_ioflag;
671 error = 0;
672 node = VP_TO_TMPFS_NODE(vp);
673 oldsize = node->tn_size;
674
675 if (uio->uio_offset < 0 || vp->v_type != VREG)
676 return (EINVAL);
677 if (uio->uio_resid == 0)
678 return (0);
679 if (ioflag & IO_APPEND)
680 uio->uio_offset = node->tn_size;
681 error = vn_rlimit_fsizex(vp, uio, VFS_TO_TMPFS(vp->v_mount)->
682 tm_maxfilesize, &r, uio->uio_td);
683 if (error != 0) {
684 vn_rlimit_fsizex_res(uio, r);
685 return (error);
686 }
687
688 if (uio->uio_offset + uio->uio_resid > node->tn_size) {
689 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
690 FALSE);
691 if (error != 0)
692 goto out;
693 }
694
695 error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio);
696 node->tn_status |= TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED;
697 node->tn_accessed = true;
698 if (node->tn_mode & (S_ISUID | S_ISGID)) {
699 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID)) {
700 newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
701 vn_seqc_write_begin(vp);
702 atomic_store_short(&node->tn_mode, newmode);
703 vn_seqc_write_end(vp);
704 }
705 }
706 if (error != 0)
707 (void)tmpfs_reg_resize(vp, oldsize, TRUE);
708
709 out:
710 MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
711 MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
712
713 vn_rlimit_fsizex_res(uio, r);
714 return (error);
715 }
716
717 static int
tmpfs_deallocate(struct vop_deallocate_args * v)718 tmpfs_deallocate(struct vop_deallocate_args *v)
719 {
720 return (tmpfs_reg_punch_hole(v->a_vp, v->a_offset, v->a_len));
721 }
722
723 static int
tmpfs_fsync(struct vop_fsync_args * v)724 tmpfs_fsync(struct vop_fsync_args *v)
725 {
726 struct vnode *vp = v->a_vp;
727
728 tmpfs_check_mtime(vp);
729 tmpfs_update(vp);
730
731 return (0);
732 }
733
734 static int
tmpfs_remove(struct vop_remove_args * v)735 tmpfs_remove(struct vop_remove_args *v)
736 {
737 struct vnode *dvp = v->a_dvp;
738 struct vnode *vp = v->a_vp;
739
740 int error;
741 struct tmpfs_dirent *de;
742 struct tmpfs_mount *tmp;
743 struct tmpfs_node *dnode;
744 struct tmpfs_node *node;
745
746 if (vp->v_type == VDIR) {
747 error = EISDIR;
748 goto out;
749 }
750
751 dnode = VP_TO_TMPFS_DIR(dvp);
752 node = VP_TO_TMPFS_NODE(vp);
753 tmp = VFS_TO_TMPFS(vp->v_mount);
754 de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
755 MPASS(de != NULL);
756
757 /* Files marked as immutable or append-only cannot be deleted. */
758 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
759 (dnode->tn_flags & APPEND)) {
760 error = EPERM;
761 goto out;
762 }
763
764 /* Remove the entry from the directory; as it is a file, we do not
765 * have to change the number of hard links of the directory. */
766 tmpfs_dir_detach(dvp, de);
767 if (v->a_cnp->cn_flags & DOWHITEOUT)
768 tmpfs_dir_whiteout_add(dvp, v->a_cnp);
769
770 /* Free the directory entry we just deleted. Note that the node
771 * referred by it will not be removed until the vnode is really
772 * reclaimed. */
773 tmpfs_free_dirent(tmp, de);
774
775 node->tn_status |= TMPFS_NODE_CHANGED;
776 node->tn_accessed = true;
777 error = 0;
778
779 out:
780 return (error);
781 }
782
783 static int
tmpfs_link(struct vop_link_args * v)784 tmpfs_link(struct vop_link_args *v)
785 {
786 struct vnode *dvp = v->a_tdvp;
787 struct vnode *vp = v->a_vp;
788 struct componentname *cnp = v->a_cnp;
789
790 int error;
791 struct tmpfs_dirent *de;
792 struct tmpfs_node *node;
793
794 MPASS(dvp != vp); /* XXX When can this be false? */
795 node = VP_TO_TMPFS_NODE(vp);
796
797 /* Ensure that we do not overflow the maximum number of links imposed
798 * by the system. */
799 MPASS(node->tn_links <= TMPFS_LINK_MAX);
800 if (node->tn_links == TMPFS_LINK_MAX) {
801 error = EMLINK;
802 goto out;
803 }
804
805 /* We cannot create links of files marked immutable or append-only. */
806 if (node->tn_flags & (IMMUTABLE | APPEND)) {
807 error = EPERM;
808 goto out;
809 }
810
811 /* Allocate a new directory entry to represent the node. */
812 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
813 cnp->cn_nameptr, cnp->cn_namelen, &de);
814 if (error != 0)
815 goto out;
816
817 /* Insert the new directory entry into the appropriate directory. */
818 if (cnp->cn_flags & ISWHITEOUT)
819 tmpfs_dir_whiteout_remove(dvp, cnp);
820 tmpfs_dir_attach(dvp, de);
821
822 /* vp link count has changed, so update node times. */
823 node->tn_status |= TMPFS_NODE_CHANGED;
824 tmpfs_update(vp);
825
826 error = 0;
827
828 out:
829 return (error);
830 }
831
832 /*
833 * We acquire all but fdvp locks using non-blocking acquisitions. If we
834 * fail to acquire any lock in the path we will drop all held locks,
835 * acquire the new lock in a blocking fashion, and then release it and
836 * restart the rename. This acquire/release step ensures that we do not
837 * spin on a lock waiting for release. On error release all vnode locks
838 * and decrement references the way tmpfs_rename() would do.
839 */
840 static int
tmpfs_rename_relock(struct vnode * fdvp,struct vnode ** fvpp,struct vnode * tdvp,struct vnode ** tvpp,struct componentname * fcnp,struct componentname * tcnp)841 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
842 struct vnode *tdvp, struct vnode **tvpp,
843 struct componentname *fcnp, struct componentname *tcnp)
844 {
845 struct vnode *nvp;
846 struct mount *mp;
847 struct tmpfs_dirent *de;
848 int error, restarts = 0;
849
850 VOP_UNLOCK(tdvp);
851 if (*tvpp != NULL && *tvpp != tdvp)
852 VOP_UNLOCK(*tvpp);
853 mp = fdvp->v_mount;
854
855 relock:
856 restarts += 1;
857 error = vn_lock(fdvp, LK_EXCLUSIVE);
858 if (error)
859 goto releout;
860 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
861 VOP_UNLOCK(fdvp);
862 error = vn_lock(tdvp, LK_EXCLUSIVE);
863 if (error)
864 goto releout;
865 VOP_UNLOCK(tdvp);
866 goto relock;
867 }
868 /*
869 * Re-resolve fvp to be certain it still exists and fetch the
870 * correct vnode.
871 */
872 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
873 if (de == NULL) {
874 VOP_UNLOCK(fdvp);
875 VOP_UNLOCK(tdvp);
876 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
877 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
878 error = EINVAL;
879 else
880 error = ENOENT;
881 goto releout;
882 }
883 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
884 if (error != 0) {
885 VOP_UNLOCK(fdvp);
886 VOP_UNLOCK(tdvp);
887 if (error != EBUSY)
888 goto releout;
889 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
890 if (error != 0)
891 goto releout;
892 VOP_UNLOCK(nvp);
893 /*
894 * Concurrent rename race.
895 */
896 if (nvp == tdvp) {
897 vrele(nvp);
898 error = EINVAL;
899 goto releout;
900 }
901 vrele(*fvpp);
902 *fvpp = nvp;
903 goto relock;
904 }
905 vrele(*fvpp);
906 *fvpp = nvp;
907 VOP_UNLOCK(*fvpp);
908 /*
909 * Re-resolve tvp and acquire the vnode lock if present.
910 */
911 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
912 /*
913 * If tvp disappeared we just carry on.
914 */
915 if (de == NULL && *tvpp != NULL) {
916 vrele(*tvpp);
917 *tvpp = NULL;
918 }
919 /*
920 * Get the tvp ino if the lookup succeeded. We may have to restart
921 * if the non-blocking acquire fails.
922 */
923 if (de != NULL) {
924 nvp = NULL;
925 error = tmpfs_alloc_vp(mp, de->td_node,
926 LK_EXCLUSIVE | LK_NOWAIT, &nvp);
927 if (*tvpp != NULL)
928 vrele(*tvpp);
929 *tvpp = nvp;
930 if (error != 0) {
931 VOP_UNLOCK(fdvp);
932 VOP_UNLOCK(tdvp);
933 if (error != EBUSY)
934 goto releout;
935 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
936 &nvp);
937 if (error != 0)
938 goto releout;
939 VOP_UNLOCK(nvp);
940 /*
941 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
942 */
943 if (nvp == fdvp) {
944 error = ENOTEMPTY;
945 goto releout;
946 }
947 goto relock;
948 }
949 }
950 tmpfs_rename_restarts += restarts;
951
952 return (0);
953
954 releout:
955 vrele(fdvp);
956 vrele(*fvpp);
957 vrele(tdvp);
958 if (*tvpp != NULL)
959 vrele(*tvpp);
960 tmpfs_rename_restarts += restarts;
961
962 return (error);
963 }
964
965 static int
tmpfs_rename(struct vop_rename_args * v)966 tmpfs_rename(struct vop_rename_args *v)
967 {
968 struct vnode *fdvp = v->a_fdvp;
969 struct vnode *fvp = v->a_fvp;
970 struct componentname *fcnp = v->a_fcnp;
971 struct vnode *tdvp = v->a_tdvp;
972 struct vnode *tvp = v->a_tvp;
973 struct componentname *tcnp = v->a_tcnp;
974 char *newname;
975 struct tmpfs_dirent *de;
976 struct tmpfs_mount *tmp;
977 struct tmpfs_node *fdnode;
978 struct tmpfs_node *fnode;
979 struct tmpfs_node *tnode;
980 struct tmpfs_node *tdnode;
981 int error;
982 bool want_seqc_end;
983
984 want_seqc_end = false;
985
986 /*
987 * Disallow cross-device renames.
988 * XXX Why isn't this done by the caller?
989 */
990 if (fvp->v_mount != tdvp->v_mount ||
991 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
992 error = EXDEV;
993 goto out;
994 }
995
996 /* If source and target are the same file, there is nothing to do. */
997 if (fvp == tvp) {
998 error = 0;
999 goto out;
1000 }
1001
1002 /*
1003 * If we need to move the directory between entries, lock the
1004 * source so that we can safely operate on it.
1005 */
1006 if (fdvp != tdvp && fdvp != tvp) {
1007 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1008 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1009 fcnp, tcnp);
1010 if (error != 0)
1011 return (error);
1012 ASSERT_VOP_ELOCKED(fdvp,
1013 "tmpfs_rename: fdvp not locked");
1014 ASSERT_VOP_ELOCKED(tdvp,
1015 "tmpfs_rename: tdvp not locked");
1016 if (tvp != NULL)
1017 ASSERT_VOP_ELOCKED(tvp,
1018 "tmpfs_rename: tvp not locked");
1019 if (fvp == tvp) {
1020 error = 0;
1021 goto out_locked;
1022 }
1023 }
1024 }
1025
1026 /*
1027 * Avoid manipulating '.' and '..' entries.
1028 */
1029 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1030 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) {
1031 error = EINVAL;
1032 goto out_locked;
1033 }
1034
1035 if (tvp != NULL)
1036 vn_seqc_write_begin(tvp);
1037 vn_seqc_write_begin(tdvp);
1038 vn_seqc_write_begin(fvp);
1039 vn_seqc_write_begin(fdvp);
1040 want_seqc_end = true;
1041
1042 tmp = VFS_TO_TMPFS(tdvp->v_mount);
1043 tdnode = VP_TO_TMPFS_DIR(tdvp);
1044 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
1045 fdnode = VP_TO_TMPFS_DIR(fdvp);
1046 fnode = VP_TO_TMPFS_NODE(fvp);
1047 de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
1048
1049 /*
1050 * Entry can disappear before we lock fdvp.
1051 */
1052 if (de == NULL) {
1053 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1054 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1055 error = EINVAL;
1056 else
1057 error = ENOENT;
1058 goto out_locked;
1059 }
1060 MPASS(de->td_node == fnode);
1061
1062 /*
1063 * If re-naming a directory to another preexisting directory
1064 * ensure that the target directory is empty so that its
1065 * removal causes no side effects.
1066 * Kern_rename guarantees the destination to be a directory
1067 * if the source is one.
1068 */
1069 if (tvp != NULL) {
1070 MPASS(tnode != NULL);
1071
1072 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1073 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1074 error = EPERM;
1075 goto out_locked;
1076 }
1077
1078 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1079 if (tnode->tn_size != 0 &&
1080 ((tcnp->cn_flags & IGNOREWHITEOUT) == 0 ||
1081 tnode->tn_size > tnode->tn_dir.tn_wht_size)) {
1082 error = ENOTEMPTY;
1083 goto out_locked;
1084 }
1085 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1086 error = ENOTDIR;
1087 goto out_locked;
1088 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1089 error = EISDIR;
1090 goto out_locked;
1091 } else {
1092 MPASS(fnode->tn_type != VDIR &&
1093 tnode->tn_type != VDIR);
1094 }
1095 }
1096
1097 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1098 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1099 error = EPERM;
1100 goto out_locked;
1101 }
1102
1103 /*
1104 * Ensure that we have enough memory to hold the new name, if it
1105 * has to be changed.
1106 */
1107 if (fcnp->cn_namelen != tcnp->cn_namelen ||
1108 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1109 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1110 } else
1111 newname = NULL;
1112
1113 /*
1114 * If the node is being moved to another directory, we have to do
1115 * the move.
1116 */
1117 if (fdnode != tdnode) {
1118 /*
1119 * In case we are moving a directory, we have to adjust its
1120 * parent to point to the new parent.
1121 */
1122 if (de->td_node->tn_type == VDIR) {
1123 struct tmpfs_node *n;
1124
1125 TMPFS_NODE_LOCK(fnode);
1126 error = tmpfs_access_locked(fvp, fnode, VWRITE,
1127 tcnp->cn_cred);
1128 TMPFS_NODE_UNLOCK(fnode);
1129 if (error) {
1130 if (newname != NULL)
1131 free(newname, M_TMPFSNAME);
1132 goto out_locked;
1133 }
1134
1135 /*
1136 * Ensure the target directory is not a child of the
1137 * directory being moved. Otherwise, we'd end up
1138 * with stale nodes.
1139 */
1140 n = tdnode;
1141 /*
1142 * TMPFS_LOCK guaranties that no nodes are freed while
1143 * traversing the list. Nodes can only be marked as
1144 * removed: tn_parent == NULL.
1145 */
1146 TMPFS_LOCK(tmp);
1147 TMPFS_NODE_LOCK(n);
1148 while (n != n->tn_dir.tn_parent) {
1149 struct tmpfs_node *parent;
1150
1151 if (n == fnode) {
1152 TMPFS_NODE_UNLOCK(n);
1153 TMPFS_UNLOCK(tmp);
1154 error = EINVAL;
1155 if (newname != NULL)
1156 free(newname, M_TMPFSNAME);
1157 goto out_locked;
1158 }
1159 parent = n->tn_dir.tn_parent;
1160 TMPFS_NODE_UNLOCK(n);
1161 if (parent == NULL) {
1162 n = NULL;
1163 break;
1164 }
1165 TMPFS_NODE_LOCK(parent);
1166 if (parent->tn_dir.tn_parent == NULL) {
1167 TMPFS_NODE_UNLOCK(parent);
1168 n = NULL;
1169 break;
1170 }
1171 n = parent;
1172 }
1173 TMPFS_UNLOCK(tmp);
1174 if (n == NULL) {
1175 error = EINVAL;
1176 if (newname != NULL)
1177 free(newname, M_TMPFSNAME);
1178 goto out_locked;
1179 }
1180 TMPFS_NODE_UNLOCK(n);
1181
1182 /* Adjust the parent pointer. */
1183 TMPFS_VALIDATE_DIR(fnode);
1184 TMPFS_NODE_LOCK(de->td_node);
1185 de->td_node->tn_dir.tn_parent = tdnode;
1186 TMPFS_NODE_UNLOCK(de->td_node);
1187
1188 /*
1189 * As a result of changing the target of the '..'
1190 * entry, the link count of the source and target
1191 * directories has to be adjusted.
1192 */
1193 TMPFS_NODE_LOCK(tdnode);
1194 TMPFS_ASSERT_LOCKED(tdnode);
1195 tdnode->tn_links++;
1196 TMPFS_NODE_UNLOCK(tdnode);
1197
1198 TMPFS_NODE_LOCK(fdnode);
1199 TMPFS_ASSERT_LOCKED(fdnode);
1200 fdnode->tn_links--;
1201 TMPFS_NODE_UNLOCK(fdnode);
1202 }
1203 }
1204
1205 /*
1206 * Do the move: just remove the entry from the source directory
1207 * and insert it into the target one.
1208 */
1209 tmpfs_dir_detach(fdvp, de);
1210
1211 if (fcnp->cn_flags & DOWHITEOUT)
1212 tmpfs_dir_whiteout_add(fdvp, fcnp);
1213 if (tcnp->cn_flags & ISWHITEOUT)
1214 tmpfs_dir_whiteout_remove(tdvp, tcnp);
1215
1216 /*
1217 * If the name has changed, we need to make it effective by changing
1218 * it in the directory entry.
1219 */
1220 if (newname != NULL) {
1221 MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1222
1223 free(de->ud.td_name, M_TMPFSNAME);
1224 de->ud.td_name = newname;
1225 tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1226
1227 fnode->tn_status |= TMPFS_NODE_CHANGED;
1228 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1229 }
1230
1231 /*
1232 * If we are overwriting an entry, we have to remove the old one
1233 * from the target directory.
1234 */
1235 if (tvp != NULL) {
1236 struct tmpfs_dirent *tde;
1237
1238 /* Remove the old entry from the target directory. */
1239 tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1240 tmpfs_dir_detach(tdvp, tde);
1241
1242 /*
1243 * If we are overwriting a directory, per the ENOTEMPTY check
1244 * above it must either be empty or contain only whiteout
1245 * entries. In the latter case (which can only happen if
1246 * IGNOREWHITEOUT was passed in tcnp->cn_flags), clear the
1247 * whiteout entries to avoid leaking memory.
1248 */
1249 if (tnode->tn_type == VDIR && tnode->tn_size > 0)
1250 tmpfs_dir_clear_whiteouts(tvp);
1251
1252 /* Update node's ctime because of possible hardlinks. */
1253 tnode->tn_status |= TMPFS_NODE_CHANGED;
1254 tmpfs_update(tvp);
1255
1256 /*
1257 * Free the directory entry we just deleted. Note that the
1258 * node referred by it will not be removed until the vnode is
1259 * really reclaimed.
1260 */
1261 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1262 }
1263
1264 tmpfs_dir_attach(tdvp, de);
1265
1266 if (tmpfs_use_nc(fvp)) {
1267 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp);
1268 }
1269
1270 error = 0;
1271
1272 out_locked:
1273 if (fdvp != tdvp && fdvp != tvp)
1274 VOP_UNLOCK(fdvp);
1275
1276 out:
1277 if (want_seqc_end) {
1278 if (tvp != NULL)
1279 vn_seqc_write_end(tvp);
1280 vn_seqc_write_end(tdvp);
1281 vn_seqc_write_end(fvp);
1282 vn_seqc_write_end(fdvp);
1283 }
1284
1285 /*
1286 * Release target nodes.
1287 * XXX: I don't understand when tdvp can be the same as tvp, but
1288 * other code takes care of this...
1289 */
1290 if (tdvp == tvp)
1291 vrele(tdvp);
1292 else
1293 vput(tdvp);
1294 if (tvp != NULL)
1295 vput(tvp);
1296
1297 /* Release source nodes. */
1298 vrele(fdvp);
1299 vrele(fvp);
1300
1301 return (error);
1302 }
1303
1304 static int
tmpfs_mkdir(struct vop_mkdir_args * v)1305 tmpfs_mkdir(struct vop_mkdir_args *v)
1306 {
1307 struct vnode *dvp = v->a_dvp;
1308 struct vnode **vpp = v->a_vpp;
1309 struct componentname *cnp = v->a_cnp;
1310 struct vattr *vap = v->a_vap;
1311
1312 MPASS(vap->va_type == VDIR);
1313
1314 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL));
1315 }
1316
1317 static int
tmpfs_rmdir(struct vop_rmdir_args * v)1318 tmpfs_rmdir(struct vop_rmdir_args *v)
1319 {
1320 struct vnode *dvp = v->a_dvp;
1321 struct vnode *vp = v->a_vp;
1322 struct componentname *cnp = v->a_cnp;
1323
1324 int error;
1325 struct tmpfs_dirent *de;
1326 struct tmpfs_mount *tmp;
1327 struct tmpfs_node *dnode;
1328 struct tmpfs_node *node;
1329
1330 tmp = VFS_TO_TMPFS(dvp->v_mount);
1331 dnode = VP_TO_TMPFS_DIR(dvp);
1332 node = VP_TO_TMPFS_DIR(vp);
1333
1334 /*
1335 * Directories with more than two non-whiteout entries ('.' and '..')
1336 * cannot be removed.
1337 */
1338 if (node->tn_size != 0 &&
1339 ((cnp->cn_flags & IGNOREWHITEOUT) == 0 ||
1340 node->tn_size > node->tn_dir.tn_wht_size)) {
1341 error = ENOTEMPTY;
1342 goto out;
1343 }
1344
1345 /* Check flags to see if we are allowed to remove the directory. */
1346 if ((dnode->tn_flags & APPEND)
1347 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1348 error = EPERM;
1349 goto out;
1350 }
1351
1352 /* This invariant holds only if we are not trying to remove "..".
1353 * We checked for that above so this is safe now. */
1354 MPASS(node->tn_dir.tn_parent == dnode);
1355
1356 /* Get the directory entry associated with node (vp). This was
1357 * filled by tmpfs_lookup while looking up the entry. */
1358 de = tmpfs_dir_lookup(dnode, node, cnp);
1359 MPASS(TMPFS_DIRENT_MATCHES(de,
1360 cnp->cn_nameptr,
1361 cnp->cn_namelen));
1362
1363 /* Detach the directory entry from the directory (dnode). */
1364 tmpfs_dir_detach(dvp, de);
1365
1366 /*
1367 * If we are removing a directory, per the ENOTEMPTY check above it
1368 * must either be empty or contain only whiteout entries. In the
1369 * latter case (which can only happen if IGNOREWHITEOUT was passed
1370 * in cnp->cn_flags), clear the whiteout entries to avoid leaking
1371 * memory.
1372 */
1373 if (node->tn_size > 0)
1374 tmpfs_dir_clear_whiteouts(vp);
1375
1376 if (cnp->cn_flags & DOWHITEOUT)
1377 tmpfs_dir_whiteout_add(dvp, cnp);
1378
1379 /* No vnode should be allocated for this entry from this point */
1380 TMPFS_NODE_LOCK(node);
1381 node->tn_links--;
1382 node->tn_dir.tn_parent = NULL;
1383 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1384 node->tn_accessed = true;
1385
1386 TMPFS_NODE_UNLOCK(node);
1387
1388 TMPFS_NODE_LOCK(dnode);
1389 dnode->tn_links--;
1390 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1391 dnode->tn_accessed = true;
1392 TMPFS_NODE_UNLOCK(dnode);
1393
1394 if (tmpfs_use_nc(dvp)) {
1395 cache_vop_rmdir(dvp, vp);
1396 }
1397
1398 /* Free the directory entry we just deleted. Note that the node
1399 * referred by it will not be removed until the vnode is really
1400 * reclaimed. */
1401 tmpfs_free_dirent(tmp, de);
1402
1403 /* Release the deleted vnode (will destroy the node, notify
1404 * interested parties and clean it from the cache). */
1405
1406 dnode->tn_status |= TMPFS_NODE_CHANGED;
1407 tmpfs_update(dvp);
1408
1409 error = 0;
1410
1411 out:
1412 return (error);
1413 }
1414
1415 static int
tmpfs_symlink(struct vop_symlink_args * v)1416 tmpfs_symlink(struct vop_symlink_args *v)
1417 {
1418 struct vnode *dvp = v->a_dvp;
1419 struct vnode **vpp = v->a_vpp;
1420 struct componentname *cnp = v->a_cnp;
1421 struct vattr *vap = v->a_vap;
1422 const char *target = v->a_target;
1423
1424 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1425 MPASS(vap->va_type == VLNK);
1426 #else
1427 vap->va_type = VLNK;
1428 #endif
1429
1430 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, target));
1431 }
1432
1433 static int
tmpfs_readdir(struct vop_readdir_args * va)1434 tmpfs_readdir(struct vop_readdir_args *va)
1435 {
1436 struct vnode *vp;
1437 struct uio *uio;
1438 struct tmpfs_mount *tm;
1439 struct tmpfs_node *node;
1440 uint64_t **cookies;
1441 int *eofflag, *ncookies;
1442 ssize_t startresid;
1443 int error, maxcookies;
1444
1445 vp = va->a_vp;
1446 uio = va->a_uio;
1447 eofflag = va->a_eofflag;
1448 cookies = va->a_cookies;
1449 ncookies = va->a_ncookies;
1450
1451 /* This operation only makes sense on directory nodes. */
1452 if (vp->v_type != VDIR)
1453 return (ENOTDIR);
1454
1455 maxcookies = 0;
1456 node = VP_TO_TMPFS_DIR(vp);
1457 tm = VFS_TO_TMPFS(vp->v_mount);
1458
1459 startresid = uio->uio_resid;
1460
1461 /* Allocate cookies for NFS and compat modules. */
1462 if (cookies != NULL && ncookies != NULL) {
1463 maxcookies = howmany(node->tn_size,
1464 sizeof(struct tmpfs_dirent)) + 2;
1465 *cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP,
1466 M_WAITOK);
1467 *ncookies = 0;
1468 }
1469
1470 if (cookies == NULL)
1471 error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL);
1472 else
1473 error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies,
1474 ncookies);
1475
1476 /* Buffer was filled without hitting EOF. */
1477 if (error == EJUSTRETURN)
1478 error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1479
1480 if (error != 0 && cookies != NULL && ncookies != NULL) {
1481 free(*cookies, M_TEMP);
1482 *cookies = NULL;
1483 *ncookies = 0;
1484 }
1485
1486 if (eofflag != NULL)
1487 *eofflag =
1488 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1489
1490 return (error);
1491 }
1492
1493 static int
tmpfs_readlink(struct vop_readlink_args * v)1494 tmpfs_readlink(struct vop_readlink_args *v)
1495 {
1496 struct vnode *vp = v->a_vp;
1497 struct uio *uio = v->a_uio;
1498
1499 int error;
1500 struct tmpfs_node *node;
1501
1502 MPASS(uio->uio_offset == 0);
1503 MPASS(vp->v_type == VLNK);
1504
1505 node = VP_TO_TMPFS_NODE(vp);
1506
1507 error = uiomove(node->tn_link_target, MIN(node->tn_size, uio->uio_resid),
1508 uio);
1509 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node);
1510
1511 return (error);
1512 }
1513
1514 /*
1515 * VOP_FPLOOKUP_SYMLINK routines are subject to special circumstances, see
1516 * the comment above cache_fplookup for details.
1517 *
1518 * Check tmpfs_alloc_node for tmpfs-specific synchronisation notes.
1519 */
1520 static int
tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args * v)1521 tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args *v)
1522 {
1523 struct vnode *vp;
1524 struct tmpfs_node *node;
1525 char *symlink;
1526
1527 vp = v->a_vp;
1528 node = VP_TO_TMPFS_NODE_SMR(vp);
1529 if (__predict_false(node == NULL))
1530 return (EAGAIN);
1531 if (!atomic_load_char(&node->tn_link_smr))
1532 return (EAGAIN);
1533 symlink = atomic_load_ptr(&node->tn_link_target);
1534 if (symlink == NULL)
1535 return (EAGAIN);
1536
1537 return (cache_symlink_resolve(v->a_fpl, symlink, node->tn_size));
1538 }
1539
1540 static int
tmpfs_inactive(struct vop_inactive_args * v)1541 tmpfs_inactive(struct vop_inactive_args *v)
1542 {
1543 struct vnode *vp;
1544 struct tmpfs_node *node;
1545
1546 vp = v->a_vp;
1547 node = VP_TO_TMPFS_NODE(vp);
1548 if (node->tn_links == 0)
1549 vrecycle(vp);
1550 else
1551 tmpfs_check_mtime(vp);
1552 return (0);
1553 }
1554
1555 static int
tmpfs_need_inactive(struct vop_need_inactive_args * ap)1556 tmpfs_need_inactive(struct vop_need_inactive_args *ap)
1557 {
1558 struct vnode *vp;
1559 struct tmpfs_node *node;
1560 struct vm_object *obj;
1561
1562 vp = ap->a_vp;
1563 node = VP_TO_TMPFS_NODE(vp);
1564 if (node->tn_links == 0)
1565 goto need;
1566 if (vp->v_type == VREG) {
1567 obj = vp->v_object;
1568 if (obj->generation != obj->cleangeneration)
1569 goto need;
1570 }
1571 return (0);
1572 need:
1573 return (1);
1574 }
1575
1576 int
tmpfs_reclaim(struct vop_reclaim_args * v)1577 tmpfs_reclaim(struct vop_reclaim_args *v)
1578 {
1579 struct vnode *vp;
1580 struct tmpfs_mount *tmp;
1581 struct tmpfs_node *node;
1582 bool unlock;
1583
1584 vp = v->a_vp;
1585 node = VP_TO_TMPFS_NODE(vp);
1586 tmp = VFS_TO_TMPFS(vp->v_mount);
1587
1588 if (vp->v_type == VREG)
1589 tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1590 vp->v_object = NULL;
1591
1592 TMPFS_LOCK(tmp);
1593 TMPFS_NODE_LOCK(node);
1594 tmpfs_free_vp(vp);
1595
1596 /*
1597 * If the node referenced by this vnode was deleted by the user,
1598 * we must free its associated data structures (now that the vnode
1599 * is being reclaimed).
1600 */
1601 unlock = true;
1602 if (node->tn_links == 0 &&
1603 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1604 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1605 unlock = !tmpfs_free_node_locked(tmp, node, true);
1606 }
1607
1608 if (unlock) {
1609 TMPFS_NODE_UNLOCK(node);
1610 TMPFS_UNLOCK(tmp);
1611 }
1612
1613 MPASS(vp->v_data == NULL);
1614 return (0);
1615 }
1616
1617 int
tmpfs_print(struct vop_print_args * v)1618 tmpfs_print(struct vop_print_args *v)
1619 {
1620 struct vnode *vp = v->a_vp;
1621
1622 struct tmpfs_node *node;
1623
1624 node = VP_TO_TMPFS_NODE(vp);
1625
1626 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n",
1627 node, node->tn_flags, (uintmax_t)node->tn_links);
1628 printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1629 node->tn_mode, node->tn_uid, node->tn_gid,
1630 (intmax_t)node->tn_size, node->tn_status);
1631
1632 if (vp->v_type == VFIFO)
1633 fifo_printinfo(vp);
1634
1635 printf("\n");
1636
1637 return (0);
1638 }
1639
1640 int
tmpfs_pathconf(struct vop_pathconf_args * v)1641 tmpfs_pathconf(struct vop_pathconf_args *v)
1642 {
1643 struct vnode *vp = v->a_vp;
1644 int name = v->a_name;
1645 long *retval = v->a_retval;
1646
1647 int error;
1648
1649 error = 0;
1650
1651 switch (name) {
1652 case _PC_LINK_MAX:
1653 *retval = TMPFS_LINK_MAX;
1654 break;
1655
1656 case _PC_SYMLINK_MAX:
1657 *retval = MAXPATHLEN;
1658 break;
1659
1660 case _PC_NAME_MAX:
1661 *retval = NAME_MAX;
1662 break;
1663
1664 case _PC_PIPE_BUF:
1665 if (vp->v_type == VDIR || vp->v_type == VFIFO)
1666 *retval = PIPE_BUF;
1667 else
1668 error = EINVAL;
1669 break;
1670
1671 case _PC_CHOWN_RESTRICTED:
1672 *retval = 1;
1673 break;
1674
1675 case _PC_NO_TRUNC:
1676 *retval = 1;
1677 break;
1678
1679 case _PC_SYNC_IO:
1680 *retval = 1;
1681 break;
1682
1683 case _PC_FILESIZEBITS:
1684 *retval = 64;
1685 break;
1686
1687 case _PC_MIN_HOLE_SIZE:
1688 *retval = PAGE_SIZE;
1689 break;
1690
1691 case _PC_HAS_HIDDENSYSTEM:
1692 *retval = 1;
1693 break;
1694
1695 default:
1696 error = vop_stdpathconf(v);
1697 }
1698
1699 return (error);
1700 }
1701
1702 static int
tmpfs_vptofh(struct vop_vptofh_args * ap)1703 tmpfs_vptofh(struct vop_vptofh_args *ap)
1704 /*
1705 vop_vptofh {
1706 IN struct vnode *a_vp;
1707 IN struct fid *a_fhp;
1708 };
1709 */
1710 {
1711 struct tmpfs_fid_data *const tfd = (struct tmpfs_fid_data *)ap->a_fhp;
1712 struct tmpfs_node *node;
1713 _Static_assert(sizeof(struct tmpfs_fid_data) <= sizeof(struct fid),
1714 "struct tmpfs_fid_data cannot be larger than struct fid");
1715
1716 node = VP_TO_TMPFS_NODE(ap->a_vp);
1717 tfd->tfd_len = sizeof(*tfd);
1718 tfd->tfd_gen = node->tn_gen;
1719 tfd->tfd_id = node->tn_id;
1720
1721 return (0);
1722 }
1723
1724 static int
tmpfs_whiteout(struct vop_whiteout_args * ap)1725 tmpfs_whiteout(struct vop_whiteout_args *ap)
1726 {
1727 struct vnode *dvp = ap->a_dvp;
1728 struct componentname *cnp = ap->a_cnp;
1729 struct tmpfs_dirent *de;
1730
1731 switch (ap->a_flags) {
1732 case LOOKUP:
1733 return (0);
1734 case CREATE:
1735 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1736 if (de != NULL)
1737 return (de->td_node == NULL ? 0 : EEXIST);
1738 return (tmpfs_dir_whiteout_add(dvp, cnp));
1739 case DELETE:
1740 tmpfs_dir_whiteout_remove(dvp, cnp);
1741 return (0);
1742 default:
1743 panic("tmpfs_whiteout: unknown op");
1744 }
1745 }
1746
1747 static int
tmpfs_vptocnp_dir(struct tmpfs_node * tn,struct tmpfs_node * tnp,struct tmpfs_dirent ** pde)1748 tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp,
1749 struct tmpfs_dirent **pde)
1750 {
1751 struct tmpfs_dir_cursor dc;
1752 struct tmpfs_dirent *de;
1753
1754 for (de = tmpfs_dir_first(tnp, &dc); de != NULL;
1755 de = tmpfs_dir_next(tnp, &dc)) {
1756 if (de->td_node == tn) {
1757 *pde = de;
1758 return (0);
1759 }
1760 }
1761 return (ENOENT);
1762 }
1763
1764 static int
tmpfs_vptocnp_fill(struct vnode * vp,struct tmpfs_node * tn,struct tmpfs_node * tnp,char * buf,size_t * buflen,struct vnode ** dvp)1765 tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn,
1766 struct tmpfs_node *tnp, char *buf, size_t *buflen, struct vnode **dvp)
1767 {
1768 struct tmpfs_dirent *de;
1769 int error, i;
1770
1771 error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED,
1772 dvp);
1773 if (error != 0)
1774 return (error);
1775 error = tmpfs_vptocnp_dir(tn, tnp, &de);
1776 if (error == 0) {
1777 i = *buflen;
1778 i -= de->td_namelen;
1779 if (i < 0) {
1780 error = ENOMEM;
1781 } else {
1782 bcopy(de->ud.td_name, buf + i, de->td_namelen);
1783 *buflen = i;
1784 }
1785 }
1786 if (error == 0) {
1787 if (vp != *dvp)
1788 VOP_UNLOCK(*dvp);
1789 } else {
1790 if (vp != *dvp)
1791 vput(*dvp);
1792 else
1793 vrele(vp);
1794 }
1795 return (error);
1796 }
1797
1798 static int
tmpfs_vptocnp(struct vop_vptocnp_args * ap)1799 tmpfs_vptocnp(struct vop_vptocnp_args *ap)
1800 {
1801 struct vnode *vp, **dvp;
1802 struct tmpfs_node *tn, *tnp, *tnp1;
1803 struct tmpfs_dirent *de;
1804 struct tmpfs_mount *tm;
1805 char *buf;
1806 size_t *buflen;
1807 int error;
1808
1809 vp = ap->a_vp;
1810 dvp = ap->a_vpp;
1811 buf = ap->a_buf;
1812 buflen = ap->a_buflen;
1813
1814 tm = VFS_TO_TMPFS(vp->v_mount);
1815 tn = VP_TO_TMPFS_NODE(vp);
1816 if (tn->tn_type == VDIR) {
1817 tnp = tn->tn_dir.tn_parent;
1818 if (tnp == NULL)
1819 return (ENOENT);
1820 tmpfs_ref_node(tnp);
1821 error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf,
1822 buflen, dvp);
1823 tmpfs_free_node(tm, tnp);
1824 return (error);
1825 }
1826 restart:
1827 TMPFS_LOCK(tm);
1828 restart_locked:
1829 LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) {
1830 if (tnp->tn_type != VDIR)
1831 continue;
1832 TMPFS_NODE_LOCK(tnp);
1833 tmpfs_ref_node(tnp);
1834
1835 /*
1836 * tn_vnode cannot be instantiated while we hold the
1837 * node lock, so the directory cannot be changed while
1838 * we iterate over it. Do this to avoid instantiating
1839 * vnode for directories which cannot point to our
1840 * node.
1841 */
1842 error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp,
1843 &de) : 0;
1844
1845 if (error == 0) {
1846 TMPFS_NODE_UNLOCK(tnp);
1847 TMPFS_UNLOCK(tm);
1848 error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen,
1849 dvp);
1850 if (error == 0) {
1851 tmpfs_free_node(tm, tnp);
1852 return (0);
1853 }
1854 if (VN_IS_DOOMED(vp)) {
1855 tmpfs_free_node(tm, tnp);
1856 return (ENOENT);
1857 }
1858 TMPFS_LOCK(tm);
1859 TMPFS_NODE_LOCK(tnp);
1860 }
1861 if (tmpfs_free_node_locked(tm, tnp, false)) {
1862 goto restart;
1863 } else {
1864 KASSERT(tnp->tn_refcount > 0,
1865 ("node %p refcount zero", tnp));
1866 if (tnp->tn_attached) {
1867 tnp1 = LIST_NEXT(tnp, tn_entries);
1868 TMPFS_NODE_UNLOCK(tnp);
1869 } else {
1870 TMPFS_NODE_UNLOCK(tnp);
1871 goto restart_locked;
1872 }
1873 }
1874 }
1875 TMPFS_UNLOCK(tm);
1876 return (ENOENT);
1877 }
1878
1879 void
tmpfs_extattr_free(struct tmpfs_extattr * ea)1880 tmpfs_extattr_free(struct tmpfs_extattr *ea)
1881 {
1882 free(ea->ea_name, M_TMPFSEA);
1883 free(ea->ea_value, M_TMPFSEA);
1884 free(ea, M_TMPFSEA);
1885 }
1886
1887 static bool
tmpfs_extattr_update_mem(struct tmpfs_mount * tmp,ssize_t size)1888 tmpfs_extattr_update_mem(struct tmpfs_mount *tmp, ssize_t size)
1889 {
1890 TMPFS_LOCK(tmp);
1891 if (size > 0 &&
1892 !tmpfs_pages_check_avail(tmp, howmany(size, PAGE_SIZE))) {
1893 TMPFS_UNLOCK(tmp);
1894 return (false);
1895 }
1896 if (tmp->tm_ea_memory_inuse + size > tmp->tm_ea_memory_max) {
1897 TMPFS_UNLOCK(tmp);
1898 return (false);
1899 }
1900 tmp->tm_ea_memory_inuse += size;
1901 TMPFS_UNLOCK(tmp);
1902 return (true);
1903 }
1904
1905 static int
tmpfs_deleteextattr(struct vop_deleteextattr_args * ap)1906 tmpfs_deleteextattr(struct vop_deleteextattr_args *ap)
1907 {
1908 struct vnode *vp = ap->a_vp;
1909 struct tmpfs_mount *tmp;
1910 struct tmpfs_node *node;
1911 struct tmpfs_extattr *ea;
1912 size_t namelen;
1913 ssize_t diff;
1914 int error;
1915
1916 node = VP_TO_TMPFS_NODE(vp);
1917 tmp = VFS_TO_TMPFS(vp->v_mount);
1918 if (VN_ISDEV(ap->a_vp))
1919 return (EOPNOTSUPP);
1920 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1921 ap->a_cred, ap->a_td, VWRITE);
1922 if (error != 0)
1923 return (error);
1924 if (ap->a_name == NULL || ap->a_name[0] == '\0')
1925 return (EINVAL);
1926 namelen = strlen(ap->a_name);
1927 if (namelen > EXTATTR_MAXNAMELEN)
1928 return (EINVAL);
1929
1930 LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
1931 if (ea->ea_namespace == ap->a_attrnamespace &&
1932 namelen == ea->ea_namelen &&
1933 memcmp(ap->a_name, ea->ea_name, namelen) == 0)
1934 break;
1935 }
1936
1937 if (ea == NULL)
1938 return (ENOATTR);
1939 LIST_REMOVE(ea, ea_extattrs);
1940 diff = -(sizeof(struct tmpfs_extattr) + namelen + ea->ea_size);
1941 tmpfs_extattr_update_mem(tmp, diff);
1942 tmpfs_extattr_free(ea);
1943 return (0);
1944 }
1945
1946 static int
tmpfs_getextattr(struct vop_getextattr_args * ap)1947 tmpfs_getextattr(struct vop_getextattr_args *ap)
1948 {
1949 struct vnode *vp = ap->a_vp;
1950 struct tmpfs_node *node;
1951 struct tmpfs_extattr *ea;
1952 size_t namelen;
1953 int error;
1954
1955 node = VP_TO_TMPFS_NODE(vp);
1956 if (VN_ISDEV(ap->a_vp))
1957 return (EOPNOTSUPP);
1958 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1959 ap->a_cred, ap->a_td, VREAD);
1960 if (error != 0)
1961 return (error);
1962 if (ap->a_name == NULL || ap->a_name[0] == '\0')
1963 return (EINVAL);
1964 namelen = strlen(ap->a_name);
1965 if (namelen > EXTATTR_MAXNAMELEN)
1966 return (EINVAL);
1967
1968 LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
1969 if (ea->ea_namespace == ap->a_attrnamespace &&
1970 namelen == ea->ea_namelen &&
1971 memcmp(ap->a_name, ea->ea_name, namelen) == 0)
1972 break;
1973 }
1974
1975 if (ea == NULL)
1976 return (ENOATTR);
1977 if (ap->a_size != NULL)
1978 *ap->a_size = ea->ea_size;
1979 if (ap->a_uio != NULL && ea->ea_size != 0)
1980 error = uiomove(ea->ea_value, ea->ea_size, ap->a_uio);
1981 return (error);
1982 }
1983
1984 static int
tmpfs_listextattr(struct vop_listextattr_args * ap)1985 tmpfs_listextattr(struct vop_listextattr_args *ap)
1986 {
1987 struct vnode *vp = ap->a_vp;
1988 struct tmpfs_node *node;
1989 struct tmpfs_extattr *ea;
1990 int error;
1991
1992 node = VP_TO_TMPFS_NODE(vp);
1993 if (VN_ISDEV(ap->a_vp))
1994 return (EOPNOTSUPP);
1995 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1996 ap->a_cred, ap->a_td, VREAD);
1997 if (error != 0)
1998 return (error);
1999 if (ap->a_size != NULL)
2000 *ap->a_size = 0;
2001
2002 LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
2003 if (ea->ea_namespace != ap->a_attrnamespace)
2004 continue;
2005 if (ap->a_size != NULL)
2006 *ap->a_size += ea->ea_namelen + 1;
2007 if (ap->a_uio != NULL) {
2008 error = uiomove(&ea->ea_namelen, 1, ap->a_uio);
2009 if (error != 0)
2010 break;
2011 error = uiomove(ea->ea_name, ea->ea_namelen, ap->a_uio);
2012 if (error != 0)
2013 break;
2014 }
2015 }
2016
2017 return (error);
2018 }
2019
2020 static int
tmpfs_setextattr(struct vop_setextattr_args * ap)2021 tmpfs_setextattr(struct vop_setextattr_args *ap)
2022 {
2023 struct vnode *vp = ap->a_vp;
2024 struct tmpfs_mount *tmp;
2025 struct tmpfs_node *node;
2026 struct tmpfs_extattr *ea;
2027 struct tmpfs_extattr *new_ea;
2028 size_t attr_size;
2029 size_t namelen;
2030 ssize_t diff;
2031 int error;
2032
2033 node = VP_TO_TMPFS_NODE(vp);
2034 tmp = VFS_TO_TMPFS(vp->v_mount);
2035 attr_size = ap->a_uio->uio_resid;
2036 diff = 0;
2037 if (VN_ISDEV(ap->a_vp))
2038 return (EOPNOTSUPP);
2039 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
2040 ap->a_cred, ap->a_td, VWRITE);
2041 if (error != 0)
2042 return (error);
2043 if (ap->a_name == NULL || ap->a_name[0] == '\0')
2044 return (EINVAL);
2045 namelen = strlen(ap->a_name);
2046 if (namelen > EXTATTR_MAXNAMELEN)
2047 return (EINVAL);
2048
2049 LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
2050 if (ea->ea_namespace == ap->a_attrnamespace &&
2051 namelen == ea->ea_namelen &&
2052 memcmp(ap->a_name, ea->ea_name, namelen) == 0) {
2053 diff -= sizeof(struct tmpfs_extattr) + ea->ea_namelen +
2054 ea->ea_size;
2055 break;
2056 }
2057 }
2058
2059 diff += sizeof(struct tmpfs_extattr) + namelen + attr_size;
2060 if (!tmpfs_extattr_update_mem(tmp, diff))
2061 return (ENOSPC);
2062 new_ea = malloc(sizeof(struct tmpfs_extattr), M_TMPFSEA, M_WAITOK);
2063 new_ea->ea_namespace = ap->a_attrnamespace;
2064 new_ea->ea_name = malloc(namelen, M_TMPFSEA, M_WAITOK);
2065 new_ea->ea_namelen = namelen;
2066 memcpy(new_ea->ea_name, ap->a_name, namelen);
2067 if (attr_size != 0) {
2068 new_ea->ea_value = malloc(attr_size, M_TMPFSEA, M_WAITOK);
2069 new_ea->ea_size = attr_size;
2070 error = uiomove(new_ea->ea_value, attr_size, ap->a_uio);
2071 } else {
2072 new_ea->ea_value = NULL;
2073 new_ea->ea_size = 0;
2074 }
2075 if (error != 0) {
2076 tmpfs_extattr_update_mem(tmp, -diff);
2077 tmpfs_extattr_free(new_ea);
2078 return (error);
2079 }
2080 if (ea != NULL) {
2081 LIST_REMOVE(ea, ea_extattrs);
2082 tmpfs_extattr_free(ea);
2083 }
2084 LIST_INSERT_HEAD(&node->tn_extattrs, new_ea, ea_extattrs);
2085 return (0);
2086 }
2087
2088 static off_t
tmpfs_seek_data_locked(vm_object_t obj,off_t noff)2089 tmpfs_seek_data_locked(vm_object_t obj, off_t noff)
2090 {
2091 vm_pindex_t p;
2092
2093 p = swap_pager_seek_data(obj, OFF_TO_IDX(noff));
2094 return (p == OFF_TO_IDX(noff) ? noff : IDX_TO_OFF(p));
2095 }
2096
2097 static int
tmpfs_seek_clamp(struct tmpfs_node * tn,off_t * noff,bool seekdata)2098 tmpfs_seek_clamp(struct tmpfs_node *tn, off_t *noff, bool seekdata)
2099 {
2100 if (*noff < tn->tn_size)
2101 return (0);
2102 if (seekdata)
2103 return (ENXIO);
2104 *noff = tn->tn_size;
2105 return (0);
2106 }
2107
2108 static off_t
tmpfs_seek_hole_locked(vm_object_t obj,off_t noff)2109 tmpfs_seek_hole_locked(vm_object_t obj, off_t noff)
2110 {
2111
2112 return (IDX_TO_OFF(swap_pager_seek_hole(obj, OFF_TO_IDX(noff))));
2113 }
2114
2115 static int
tmpfs_seek_datahole(struct vnode * vp,off_t * off,bool seekdata)2116 tmpfs_seek_datahole(struct vnode *vp, off_t *off, bool seekdata)
2117 {
2118 struct tmpfs_node *tn;
2119 vm_object_t obj;
2120 off_t noff;
2121 int error;
2122
2123 if (vp->v_type != VREG)
2124 return (ENOTTY);
2125 tn = VP_TO_TMPFS_NODE(vp);
2126 noff = *off;
2127 if (noff < 0)
2128 return (ENXIO);
2129 error = tmpfs_seek_clamp(tn, &noff, seekdata);
2130 if (error != 0)
2131 return (error);
2132 obj = tn->tn_reg.tn_aobj;
2133
2134 VM_OBJECT_RLOCK(obj);
2135 noff = seekdata ? tmpfs_seek_data_locked(obj, noff) :
2136 tmpfs_seek_hole_locked(obj, noff);
2137 VM_OBJECT_RUNLOCK(obj);
2138
2139 error = tmpfs_seek_clamp(tn, &noff, seekdata);
2140 if (error == 0)
2141 *off = noff;
2142 return (error);
2143 }
2144
2145 static int
tmpfs_ioctl(struct vop_ioctl_args * ap)2146 tmpfs_ioctl(struct vop_ioctl_args *ap)
2147 {
2148 struct vnode *vp = ap->a_vp;
2149 int error = 0;
2150
2151 switch (ap->a_command) {
2152 case FIOSEEKDATA:
2153 case FIOSEEKHOLE:
2154 error = vn_lock(vp, LK_SHARED);
2155 if (error != 0) {
2156 error = EBADF;
2157 break;
2158 }
2159 error = tmpfs_seek_datahole(vp, (off_t *)ap->a_data,
2160 ap->a_command == FIOSEEKDATA);
2161 VOP_UNLOCK(vp);
2162 break;
2163 default:
2164 error = ENOTTY;
2165 break;
2166 }
2167 return (error);
2168 }
2169
2170 /*
2171 * Vnode operations vector used for files stored in a tmpfs file system.
2172 */
2173 struct vop_vector tmpfs_vnodeop_entries = {
2174 .vop_default = &default_vnodeops,
2175 .vop_lookup = vfs_cache_lookup,
2176 .vop_cachedlookup = tmpfs_cached_lookup,
2177 .vop_create = tmpfs_create,
2178 .vop_mknod = tmpfs_mknod,
2179 .vop_open = tmpfs_open,
2180 .vop_close = tmpfs_close,
2181 .vop_fplookup_vexec = tmpfs_fplookup_vexec,
2182 .vop_fplookup_symlink = tmpfs_fplookup_symlink,
2183 .vop_access = tmpfs_access,
2184 .vop_stat = tmpfs_stat,
2185 .vop_getattr = tmpfs_getattr,
2186 .vop_setattr = tmpfs_setattr,
2187 .vop_read = tmpfs_read,
2188 .vop_read_pgcache = tmpfs_read_pgcache,
2189 .vop_write = tmpfs_write,
2190 .vop_deallocate = tmpfs_deallocate,
2191 .vop_fsync = tmpfs_fsync,
2192 .vop_remove = tmpfs_remove,
2193 .vop_link = tmpfs_link,
2194 .vop_rename = tmpfs_rename,
2195 .vop_mkdir = tmpfs_mkdir,
2196 .vop_rmdir = tmpfs_rmdir,
2197 .vop_symlink = tmpfs_symlink,
2198 .vop_readdir = tmpfs_readdir,
2199 .vop_readlink = tmpfs_readlink,
2200 .vop_inactive = tmpfs_inactive,
2201 .vop_need_inactive = tmpfs_need_inactive,
2202 .vop_reclaim = tmpfs_reclaim,
2203 .vop_print = tmpfs_print,
2204 .vop_pathconf = tmpfs_pathconf,
2205 .vop_vptofh = tmpfs_vptofh,
2206 .vop_whiteout = tmpfs_whiteout,
2207 .vop_bmap = VOP_EOPNOTSUPP,
2208 .vop_vptocnp = tmpfs_vptocnp,
2209 .vop_lock1 = vop_lock,
2210 .vop_unlock = vop_unlock,
2211 .vop_islocked = vop_islocked,
2212 .vop_deleteextattr = tmpfs_deleteextattr,
2213 .vop_getextattr = tmpfs_getextattr,
2214 .vop_listextattr = tmpfs_listextattr,
2215 .vop_setextattr = tmpfs_setextattr,
2216 .vop_add_writecount = vop_stdadd_writecount_nomsync,
2217 .vop_ioctl = tmpfs_ioctl,
2218 };
2219 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_entries);
2220
2221 /*
2222 * Same vector for mounts which do not use namecache.
2223 */
2224 struct vop_vector tmpfs_vnodeop_nonc_entries = {
2225 .vop_default = &tmpfs_vnodeop_entries,
2226 .vop_lookup = tmpfs_lookup,
2227 };
2228 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_nonc_entries);
2229