1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/8.0/sys/ufs/ufs/ufs_vnops.c 197473 2009-09-24 21:32:56Z brooks $");
39
40 #include "opt_quota.h"
41 #include "opt_suiddir.h"
42 #include "opt_ufs.h"
43 #include "opt_ffs.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/namei.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/stat.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/mount.h>
55 #include <sys/priv.h>
56 #include <sys/refcount.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/lockf.h>
61 #include <sys/conf.h>
62 #include <sys/acl.h>
63
64 #include <machine/mutex.h>
65
66 #include <security/mac/mac_framework.h>
67
68 #include <sys/file.h> /* XXX */
69
70 #include <vm/vm.h>
71 #include <vm/vm_extern.h>
72
73 #include <fs/fifofs/fifo.h>
74
75 #include <ufs/ufs/acl.h>
76 #include <ufs/ufs/extattr.h>
77 #include <ufs/ufs/quota.h>
78 #include <ufs/ufs/inode.h>
79 #include <ufs/ufs/dir.h>
80 #include <ufs/ufs/ufsmount.h>
81 #include <ufs/ufs/ufs_extern.h>
82 #ifdef UFS_DIRHASH
83 #include <ufs/ufs/dirhash.h>
84 #endif
85 #ifdef UFS_GJOURNAL
86 #include <ufs/ufs/gjournal.h>
87 #endif
88
89 #include <ufs/ffs/ffs_extern.h>
90
91 static vop_access_t ufs_access;
92 static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *);
93 static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *);
94 static vop_close_t ufs_close;
95 static vop_create_t ufs_create;
96 static vop_getattr_t ufs_getattr;
97 static vop_link_t ufs_link;
98 static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
99 static vop_markatime_t ufs_markatime;
100 static vop_mkdir_t ufs_mkdir;
101 static vop_mknod_t ufs_mknod;
102 static vop_open_t ufs_open;
103 static vop_pathconf_t ufs_pathconf;
104 static vop_print_t ufs_print;
105 static vop_readlink_t ufs_readlink;
106 static vop_remove_t ufs_remove;
107 static vop_rename_t ufs_rename;
108 static vop_rmdir_t ufs_rmdir;
109 static vop_setattr_t ufs_setattr;
110 static vop_strategy_t ufs_strategy;
111 static vop_symlink_t ufs_symlink;
112 static vop_whiteout_t ufs_whiteout;
113 static vop_close_t ufsfifo_close;
114 static vop_kqfilter_t ufsfifo_kqfilter;
115 static vop_pathconf_t ufsfifo_pathconf;
116
117 /*
118 * A virgin directory (no blushing please).
119 */
120 static struct dirtemplate mastertemplate = {
121 0, 12, DT_DIR, 1, ".",
122 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
123 };
124 static struct odirtemplate omastertemplate = {
125 0, 12, 1, ".",
126 0, DIRBLKSIZ - 12, 2, ".."
127 };
128
129 static void
130 ufs_itimes_locked(struct vnode *vp)
131 {
132 struct inode *ip;
133 struct timespec ts;
134
135 ASSERT_VI_LOCKED(vp, __func__);
136
137 ip = VTOI(vp);
138 if (UFS_RDONLY(ip))
139 goto out;
140 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
141 return;
142
143 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
144 ip->i_flag |= IN_LAZYMOD;
145 else if (((vp->v_mount->mnt_kern_flag &
146 (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) ||
147 (ip->i_flag & (IN_CHANGE | IN_UPDATE)))
148 ip->i_flag |= IN_MODIFIED;
149 else if (ip->i_flag & IN_ACCESS)
150 ip->i_flag |= IN_LAZYACCESS;
151 vfs_timestamp(&ts);
152 if (ip->i_flag & IN_ACCESS) {
153 DIP_SET(ip, i_atime, ts.tv_sec);
154 DIP_SET(ip, i_atimensec, ts.tv_nsec);
155 }
156 if (ip->i_flag & IN_UPDATE) {
157 DIP_SET(ip, i_mtime, ts.tv_sec);
158 DIP_SET(ip, i_mtimensec, ts.tv_nsec);
159 }
160 if (ip->i_flag & IN_CHANGE) {
161 DIP_SET(ip, i_ctime, ts.tv_sec);
162 DIP_SET(ip, i_ctimensec, ts.tv_nsec);
163 DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1);
164 }
165
166 out:
167 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
168 }
169
170 void
171 ufs_itimes(struct vnode *vp)
172 {
173
174 VI_LOCK(vp);
175 ufs_itimes_locked(vp);
176 VI_UNLOCK(vp);
177 }
178
179 /*
180 * Create a regular file
181 */
182 static int
183 ufs_create(ap)
184 struct vop_create_args /* {
185 struct vnode *a_dvp;
186 struct vnode **a_vpp;
187 struct componentname *a_cnp;
188 struct vattr *a_vap;
189 } */ *ap;
190 {
191 int error;
192
193 error =
194 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
195 ap->a_dvp, ap->a_vpp, ap->a_cnp);
196 if (error)
197 return (error);
198 return (0);
199 }
200
201 /*
202 * Mknod vnode call
203 */
204 /* ARGSUSED */
205 static int
206 ufs_mknod(ap)
207 struct vop_mknod_args /* {
208 struct vnode *a_dvp;
209 struct vnode **a_vpp;
210 struct componentname *a_cnp;
211 struct vattr *a_vap;
212 } */ *ap;
213 {
214 struct vattr *vap = ap->a_vap;
215 struct vnode **vpp = ap->a_vpp;
216 struct inode *ip;
217 ino_t ino;
218 int error;
219
220 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
221 ap->a_dvp, vpp, ap->a_cnp);
222 if (error)
223 return (error);
224 ip = VTOI(*vpp);
225 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
226 if (vap->va_rdev != VNOVAL) {
227 /*
228 * Want to be able to use this to make badblock
229 * inodes, so don't truncate the dev number.
230 */
231 DIP_SET(ip, i_rdev, vap->va_rdev);
232 }
233 /*
234 * Remove inode, then reload it through VFS_VGET so it is
235 * checked to see if it is an alias of an existing entry in
236 * the inode cache. XXX I don't believe this is necessary now.
237 */
238 (*vpp)->v_type = VNON;
239 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
240 vgone(*vpp);
241 vput(*vpp);
242 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
243 if (error) {
244 *vpp = NULL;
245 return (error);
246 }
247 return (0);
248 }
249
250 /*
251 * Open called.
252 */
253 /* ARGSUSED */
254 static int
255 ufs_open(struct vop_open_args *ap)
256 {
257 struct vnode *vp = ap->a_vp;
258 struct inode *ip;
259
260 if (vp->v_type == VCHR || vp->v_type == VBLK)
261 return (EOPNOTSUPP);
262
263 ip = VTOI(vp);
264 /*
265 * Files marked append-only must be opened for appending.
266 */
267 if ((ip->i_flags & APPEND) &&
268 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
269 return (EPERM);
270 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td);
271 return (0);
272 }
273
274 /*
275 * Close called.
276 *
277 * Update the times on the inode.
278 */
279 /* ARGSUSED */
280 static int
281 ufs_close(ap)
282 struct vop_close_args /* {
283 struct vnode *a_vp;
284 int a_fflag;
285 struct ucred *a_cred;
286 struct thread *a_td;
287 } */ *ap;
288 {
289 struct vnode *vp = ap->a_vp;
290 int usecount;
291
292 VI_LOCK(vp);
293 usecount = vp->v_usecount;
294 if (usecount > 1)
295 ufs_itimes_locked(vp);
296 VI_UNLOCK(vp);
297 return (0);
298 }
299
300 static int
301 ufs_access(ap)
302 struct vop_access_args /* {
303 struct vnode *a_vp;
304 accmode_t a_accmode;
305 struct ucred *a_cred;
306 struct thread *a_td;
307 } */ *ap;
308 {
309 struct vnode *vp = ap->a_vp;
310 struct inode *ip = VTOI(vp);
311 accmode_t accmode = ap->a_accmode;
312 int error;
313 #ifdef QUOTA
314 int relocked;
315 #endif
316 #ifdef UFS_ACL
317 struct acl *acl;
318 #endif
319
320 /*
321 * Disallow write attempts on read-only filesystems;
322 * unless the file is a socket, fifo, or a block or
323 * character device resident on the filesystem.
324 */
325 if (accmode & VWRITE) {
326 switch (vp->v_type) {
327 case VDIR:
328 case VLNK:
329 case VREG:
330 if (vp->v_mount->mnt_flag & MNT_RDONLY)
331 return (EROFS);
332 #ifdef QUOTA
333 /*
334 * Inode is accounted in the quotas only if struct
335 * dquot is attached to it. VOP_ACCESS() is called
336 * from vn_open_cred() and provides a convenient
337 * point to call getinoquota().
338 */
339 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
340
341 /*
342 * Upgrade vnode lock, since getinoquota()
343 * requires exclusive lock to modify inode.
344 */
345 relocked = 1;
346 vhold(vp);
347 vn_lock(vp, LK_UPGRADE | LK_RETRY);
348 VI_LOCK(vp);
349 if (vp->v_iflag & VI_DOOMED) {
350 vdropl(vp);
351 error = ENOENT;
352 goto relock;
353 }
354 vdropl(vp);
355 } else
356 relocked = 0;
357 error = getinoquota(ip);
358 relock:
359 if (relocked)
360 vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
361 if (error != 0)
362 return (error);
363 #endif
364 break;
365 default:
366 break;
367 }
368 }
369
370 /* If immutable bit set, nobody gets to write it. */
371 if ((accmode & VWRITE) && (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
372 return (EPERM);
373
374 #ifdef UFS_ACL
375 if ((vp->v_mount->mnt_flag & MNT_ACLS) != 0) {
376 acl = acl_alloc(M_WAITOK);
377 error = VOP_GETACL(vp, ACL_TYPE_ACCESS, acl, ap->a_cred,
378 ap->a_td);
379 switch (error) {
380 case EOPNOTSUPP:
381 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid,
382 ip->i_gid, ap->a_accmode, ap->a_cred, NULL);
383 break;
384 case 0:
385 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid,
386 ip->i_gid, acl, ap->a_accmode, ap->a_cred, NULL);
387 break;
388 default:
389 printf(
390 "ufs_access(): Error retrieving ACL on object (%d).\n",
391 error);
392 /*
393 * XXX: Fall back until debugged. Should
394 * eventually possibly log an error, and return
395 * EPERM for safety.
396 */
397 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid,
398 ip->i_gid, ap->a_accmode, ap->a_cred, NULL);
399 }
400 acl_free(acl);
401 } else
402 #endif /* !UFS_ACL */
403 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
404 ap->a_accmode, ap->a_cred, NULL);
405 return (error);
406 }
407
408 /* ARGSUSED */
409 static int
410 ufs_getattr(ap)
411 struct vop_getattr_args /* {
412 struct vnode *a_vp;
413 struct vattr *a_vap;
414 struct ucred *a_cred;
415 } */ *ap;
416 {
417 struct vnode *vp = ap->a_vp;
418 struct inode *ip = VTOI(vp);
419 struct vattr *vap = ap->a_vap;
420
421 VI_LOCK(vp);
422 ufs_itimes_locked(vp);
423 if (ip->i_ump->um_fstype == UFS1) {
424 vap->va_atime.tv_sec = ip->i_din1->di_atime;
425 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
426 } else {
427 vap->va_atime.tv_sec = ip->i_din2->di_atime;
428 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
429 }
430 VI_UNLOCK(vp);
431 /*
432 * Copy from inode table
433 */
434 vap->va_fsid = dev2udev(ip->i_dev);
435 vap->va_fileid = ip->i_number;
436 vap->va_mode = ip->i_mode & ~IFMT;
437 vap->va_nlink = ip->i_effnlink;
438 vap->va_uid = ip->i_uid;
439 vap->va_gid = ip->i_gid;
440 if (ip->i_ump->um_fstype == UFS1) {
441 vap->va_rdev = ip->i_din1->di_rdev;
442 vap->va_size = ip->i_din1->di_size;
443 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
444 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
445 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
446 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
447 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks);
448 vap->va_filerev = ip->i_din1->di_modrev;
449 } else {
450 vap->va_rdev = ip->i_din2->di_rdev;
451 vap->va_size = ip->i_din2->di_size;
452 vap->va_mtime.tv_sec = ip->i_din2->di_mtime;
453 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec;
454 vap->va_ctime.tv_sec = ip->i_din2->di_ctime;
455 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec;
456 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime;
457 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec;
458 vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks);
459 vap->va_filerev = ip->i_din2->di_modrev;
460 }
461 vap->va_flags = ip->i_flags;
462 vap->va_gen = ip->i_gen;
463 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
464 vap->va_type = IFTOVT(ip->i_mode);
465 return (0);
466 }
467
468 /*
469 * Set attribute vnode op. called from several syscalls
470 */
471 static int
472 ufs_setattr(ap)
473 struct vop_setattr_args /* {
474 struct vnode *a_vp;
475 struct vattr *a_vap;
476 struct ucred *a_cred;
477 } */ *ap;
478 {
479 struct vattr *vap = ap->a_vap;
480 struct vnode *vp = ap->a_vp;
481 struct inode *ip = VTOI(vp);
482 struct ucred *cred = ap->a_cred;
483 struct thread *td = curthread;
484 int error;
485
486 /*
487 * Check for unsettable attributes.
488 */
489 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
490 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
491 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
492 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
493 return (EINVAL);
494 }
495 if (vap->va_flags != VNOVAL) {
496 if (vp->v_mount->mnt_flag & MNT_RDONLY)
497 return (EROFS);
498 /*
499 * Callers may only modify the file flags on objects they
500 * have VADMIN rights for.
501 */
502 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
503 return (error);
504 /*
505 * Unprivileged processes are not permitted to unset system
506 * flags, or modify flags if any system flags are set.
507 * Privileged non-jail processes may not modify system flags
508 * if securelevel > 0 and any existing system flags are set.
509 * Privileged jail processes behave like privileged non-jail
510 * processes if the security.jail.chflags_allowed sysctl is
511 * is non-zero; otherwise, they behave like unprivileged
512 * processes.
513 */
514 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
515 if (ip->i_flags
516 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
517 error = securelevel_gt(cred, 0);
518 if (error)
519 return (error);
520 }
521 /* Snapshot flag cannot be set or cleared */
522 if (((vap->va_flags & SF_SNAPSHOT) != 0 &&
523 (ip->i_flags & SF_SNAPSHOT) == 0) ||
524 ((vap->va_flags & SF_SNAPSHOT) == 0 &&
525 (ip->i_flags & SF_SNAPSHOT) != 0))
526 return (EPERM);
527 ip->i_flags = vap->va_flags;
528 DIP_SET(ip, i_flags, vap->va_flags);
529 } else {
530 if (ip->i_flags
531 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
532 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
533 return (EPERM);
534 ip->i_flags &= SF_SETTABLE;
535 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
536 DIP_SET(ip, i_flags, ip->i_flags);
537 }
538 ip->i_flag |= IN_CHANGE;
539 if (vap->va_flags & (IMMUTABLE | APPEND))
540 return (0);
541 }
542 if (ip->i_flags & (IMMUTABLE | APPEND))
543 return (EPERM);
544 /*
545 * Go through the fields and update iff not VNOVAL.
546 */
547 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
548 if (vp->v_mount->mnt_flag & MNT_RDONLY)
549 return (EROFS);
550 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
551 td)) != 0)
552 return (error);
553 }
554 if (vap->va_size != VNOVAL) {
555 /*
556 * XXX most of the following special cases should be in
557 * callers instead of in N filesystems. The VDIR check
558 * mostly already is.
559 */
560 switch (vp->v_type) {
561 case VDIR:
562 return (EISDIR);
563 case VLNK:
564 case VREG:
565 /*
566 * Truncation should have an effect in these cases.
567 * Disallow it if the filesystem is read-only or
568 * the file is being snapshotted.
569 */
570 if (vp->v_mount->mnt_flag & MNT_RDONLY)
571 return (EROFS);
572 if ((ip->i_flags & SF_SNAPSHOT) != 0)
573 return (EPERM);
574 break;
575 default:
576 /*
577 * According to POSIX, the result is unspecified
578 * for file types other than regular files,
579 * directories and shared memory objects. We
580 * don't support shared memory objects in the file
581 * system, and have dubious support for truncating
582 * symlinks. Just ignore the request in other cases.
583 */
584 return (0);
585 }
586 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL,
587 cred, td)) != 0)
588 return (error);
589 }
590 if (vap->va_atime.tv_sec != VNOVAL ||
591 vap->va_mtime.tv_sec != VNOVAL ||
592 vap->va_birthtime.tv_sec != VNOVAL) {
593 if (vp->v_mount->mnt_flag & MNT_RDONLY)
594 return (EROFS);
595 if ((ip->i_flags & SF_SNAPSHOT) != 0)
596 return (EPERM);
597 /*
598 * From utimes(2):
599 * If times is NULL, ... The caller must be the owner of
600 * the file, have permission to write the file, or be the
601 * super-user.
602 * If times is non-NULL, ... The caller must be the owner of
603 * the file or be the super-user.
604 *
605 * Possibly for historical reasons, try to use VADMIN in
606 * preference to VWRITE for a NULL timestamp. This means we
607 * will return EACCES in preference to EPERM if neither
608 * check succeeds.
609 */
610 if (vap->va_vaflags & VA_UTIMES_NULL) {
611 error = VOP_ACCESS(vp, VADMIN, cred, td);
612 if (error)
613 error = VOP_ACCESS(vp, VWRITE, cred, td);
614 } else
615 error = VOP_ACCESS(vp, VADMIN, cred, td);
616 if (error)
617 return (error);
618 if (vap->va_atime.tv_sec != VNOVAL)
619 ip->i_flag |= IN_ACCESS;
620 if (vap->va_mtime.tv_sec != VNOVAL)
621 ip->i_flag |= IN_CHANGE | IN_UPDATE;
622 if (vap->va_birthtime.tv_sec != VNOVAL &&
623 ip->i_ump->um_fstype == UFS2)
624 ip->i_flag |= IN_MODIFIED;
625 ufs_itimes(vp);
626 if (vap->va_atime.tv_sec != VNOVAL) {
627 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
628 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
629 }
630 if (vap->va_mtime.tv_sec != VNOVAL) {
631 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
632 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
633 }
634 if (vap->va_birthtime.tv_sec != VNOVAL &&
635 ip->i_ump->um_fstype == UFS2) {
636 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
637 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
638 }
639 error = UFS_UPDATE(vp, 0);
640 if (error)
641 return (error);
642 }
643 error = 0;
644 if (vap->va_mode != (mode_t)VNOVAL) {
645 if (vp->v_mount->mnt_flag & MNT_RDONLY)
646 return (EROFS);
647 if ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode &
648 (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)))
649 return (EPERM);
650 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
651 }
652 return (error);
653 }
654
655 /*
656 * Mark this file's access time for update for vfs_mark_atime(). This
657 * is called from execve() and mmap().
658 */
659 static int
660 ufs_markatime(ap)
661 struct vop_markatime_args /* {
662 struct vnode *a_vp;
663 } */ *ap;
664 {
665 struct vnode *vp = ap->a_vp;
666 struct inode *ip = VTOI(vp);
667
668 VI_LOCK(vp);
669 ip->i_flag |= IN_ACCESS;
670 VI_UNLOCK(vp);
671 return (0);
672 }
673
674 /*
675 * Change the mode on a file.
676 * Inode must be locked before calling.
677 */
678 static int
679 ufs_chmod(vp, mode, cred, td)
680 struct vnode *vp;
681 int mode;
682 struct ucred *cred;
683 struct thread *td;
684 {
685 struct inode *ip = VTOI(vp);
686 int error;
687
688 /*
689 * To modify the permissions on a file, must possess VADMIN
690 * for that file.
691 */
692 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
693 return (error);
694 /*
695 * Privileged processes may set the sticky bit on non-directories,
696 * as well as set the setgid bit on a file with a group that the
697 * process is not a member of. Both of these are allowed in
698 * jail(8).
699 */
700 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
701 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
702 return (EFTYPE);
703 }
704 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
705 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
706 if (error)
707 return (error);
708 }
709 ip->i_mode &= ~ALLPERMS;
710 ip->i_mode |= (mode & ALLPERMS);
711 DIP_SET(ip, i_mode, ip->i_mode);
712 ip->i_flag |= IN_CHANGE;
713 return (0);
714 }
715
716 /*
717 * Perform chown operation on inode ip;
718 * inode must be locked prior to call.
719 */
720 static int
721 ufs_chown(vp, uid, gid, cred, td)
722 struct vnode *vp;
723 uid_t uid;
724 gid_t gid;
725 struct ucred *cred;
726 struct thread *td;
727 {
728 struct inode *ip = VTOI(vp);
729 uid_t ouid;
730 gid_t ogid;
731 int error = 0;
732 #ifdef QUOTA
733 int i;
734 ufs2_daddr_t change;
735 #endif
736
737 if (uid == (uid_t)VNOVAL)
738 uid = ip->i_uid;
739 if (gid == (gid_t)VNOVAL)
740 gid = ip->i_gid;
741 /*
742 * To modify the ownership of a file, must possess VADMIN for that
743 * file.
744 */
745 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
746 return (error);
747 /*
748 * To change the owner of a file, or change the group of a file to a
749 * group of which we are not a member, the caller must have
750 * privilege.
751 */
752 if ((uid != ip->i_uid ||
753 (gid != ip->i_gid && !groupmember(gid, cred))) &&
754 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
755 return (error);
756 ogid = ip->i_gid;
757 ouid = ip->i_uid;
758 #ifdef QUOTA
759 if ((error = getinoquota(ip)) != 0)
760 return (error);
761 if (ouid == uid) {
762 dqrele(vp, ip->i_dquot[USRQUOTA]);
763 ip->i_dquot[USRQUOTA] = NODQUOT;
764 }
765 if (ogid == gid) {
766 dqrele(vp, ip->i_dquot[GRPQUOTA]);
767 ip->i_dquot[GRPQUOTA] = NODQUOT;
768 }
769 change = DIP(ip, i_blocks);
770 (void) chkdq(ip, -change, cred, CHOWN);
771 (void) chkiq(ip, -1, cred, CHOWN);
772 for (i = 0; i < MAXQUOTAS; i++) {
773 dqrele(vp, ip->i_dquot[i]);
774 ip->i_dquot[i] = NODQUOT;
775 }
776 #endif
777 ip->i_gid = gid;
778 DIP_SET(ip, i_gid, gid);
779 ip->i_uid = uid;
780 DIP_SET(ip, i_uid, uid);
781 #ifdef QUOTA
782 if ((error = getinoquota(ip)) == 0) {
783 if (ouid == uid) {
784 dqrele(vp, ip->i_dquot[USRQUOTA]);
785 ip->i_dquot[USRQUOTA] = NODQUOT;
786 }
787 if (ogid == gid) {
788 dqrele(vp, ip->i_dquot[GRPQUOTA]);
789 ip->i_dquot[GRPQUOTA] = NODQUOT;
790 }
791 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
792 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
793 goto good;
794 else
795 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
796 }
797 for (i = 0; i < MAXQUOTAS; i++) {
798 dqrele(vp, ip->i_dquot[i]);
799 ip->i_dquot[i] = NODQUOT;
800 }
801 }
802 ip->i_gid = ogid;
803 DIP_SET(ip, i_gid, ogid);
804 ip->i_uid = ouid;
805 DIP_SET(ip, i_uid, ouid);
806 if (getinoquota(ip) == 0) {
807 if (ouid == uid) {
808 dqrele(vp, ip->i_dquot[USRQUOTA]);
809 ip->i_dquot[USRQUOTA] = NODQUOT;
810 }
811 if (ogid == gid) {
812 dqrele(vp, ip->i_dquot[GRPQUOTA]);
813 ip->i_dquot[GRPQUOTA] = NODQUOT;
814 }
815 (void) chkdq(ip, change, cred, FORCE|CHOWN);
816 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
817 (void) getinoquota(ip);
818 }
819 return (error);
820 good:
821 if (getinoquota(ip))
822 panic("ufs_chown: lost quota");
823 #endif /* QUOTA */
824 ip->i_flag |= IN_CHANGE;
825 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
826 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) {
827 ip->i_mode &= ~(ISUID | ISGID);
828 DIP_SET(ip, i_mode, ip->i_mode);
829 }
830 }
831 return (0);
832 }
833
834 static int
835 ufs_remove(ap)
836 struct vop_remove_args /* {
837 struct vnode *a_dvp;
838 struct vnode *a_vp;
839 struct componentname *a_cnp;
840 } */ *ap;
841 {
842 struct inode *ip;
843 struct vnode *vp = ap->a_vp;
844 struct vnode *dvp = ap->a_dvp;
845 int error;
846 struct thread *td;
847
848 td = curthread;
849 ip = VTOI(vp);
850 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
851 (VTOI(dvp)->i_flags & APPEND)) {
852 error = EPERM;
853 goto out;
854 }
855 #ifdef UFS_GJOURNAL
856 ufs_gjournal_orphan(vp);
857 #endif
858 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
859 if (ip->i_nlink <= 0)
860 vp->v_vflag |= VV_NOSYNC;
861 if ((ip->i_flags & SF_SNAPSHOT) != 0) {
862 /*
863 * Avoid deadlock where another thread is trying to
864 * update the inodeblock for dvp and is waiting on
865 * snaplk. Temporary unlock the vnode lock for the
866 * unlinked file and sync the directory. This should
867 * allow vput() of the directory to not block later on
868 * while holding the snapshot vnode locked, assuming
869 * that the directory hasn't been unlinked too.
870 */
871 VOP_UNLOCK(vp, 0);
872 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
873 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
874 }
875 out:
876 return (error);
877 }
878
879 /*
880 * link vnode call
881 */
882 static int
883 ufs_link(ap)
884 struct vop_link_args /* {
885 struct vnode *a_tdvp;
886 struct vnode *a_vp;
887 struct componentname *a_cnp;
888 } */ *ap;
889 {
890 struct vnode *vp = ap->a_vp;
891 struct vnode *tdvp = ap->a_tdvp;
892 struct componentname *cnp = ap->a_cnp;
893 struct inode *ip;
894 struct direct newdir;
895 int error;
896
897 #ifdef INVARIANTS
898 if ((cnp->cn_flags & HASBUF) == 0)
899 panic("ufs_link: no name");
900 #endif
901 if (tdvp->v_mount != vp->v_mount) {
902 error = EXDEV;
903 goto out;
904 }
905 ip = VTOI(vp);
906 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
907 error = EMLINK;
908 goto out;
909 }
910 if (ip->i_flags & (IMMUTABLE | APPEND)) {
911 error = EPERM;
912 goto out;
913 }
914 ip->i_effnlink++;
915 ip->i_nlink++;
916 DIP_SET(ip, i_nlink, ip->i_nlink);
917 ip->i_flag |= IN_CHANGE;
918 if (DOINGSOFTDEP(vp))
919 softdep_change_linkcnt(ip);
920 error = UFS_UPDATE(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
921 if (!error) {
922 ufs_makedirentry(ip, cnp, &newdir);
923 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
924 }
925
926 if (error) {
927 ip->i_effnlink--;
928 ip->i_nlink--;
929 DIP_SET(ip, i_nlink, ip->i_nlink);
930 ip->i_flag |= IN_CHANGE;
931 if (DOINGSOFTDEP(vp))
932 softdep_change_linkcnt(ip);
933 }
934 out:
935 return (error);
936 }
937
938 /*
939 * whiteout vnode call
940 */
941 static int
942 ufs_whiteout(ap)
943 struct vop_whiteout_args /* {
944 struct vnode *a_dvp;
945 struct componentname *a_cnp;
946 int a_flags;
947 } */ *ap;
948 {
949 struct vnode *dvp = ap->a_dvp;
950 struct componentname *cnp = ap->a_cnp;
951 struct direct newdir;
952 int error = 0;
953
954 switch (ap->a_flags) {
955 case LOOKUP:
956 /* 4.4 format directories support whiteout operations */
957 if (dvp->v_mount->mnt_maxsymlinklen > 0)
958 return (0);
959 return (EOPNOTSUPP);
960
961 case CREATE:
962 /* create a new directory whiteout */
963 #ifdef INVARIANTS
964 if ((cnp->cn_flags & SAVENAME) == 0)
965 panic("ufs_whiteout: missing name");
966 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
967 panic("ufs_whiteout: old format filesystem");
968 #endif
969
970 newdir.d_ino = WINO;
971 newdir.d_namlen = cnp->cn_namelen;
972 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
973 newdir.d_type = DT_WHT;
974 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
975 break;
976
977 case DELETE:
978 /* remove an existing directory whiteout */
979 #ifdef INVARIANTS
980 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
981 panic("ufs_whiteout: old format filesystem");
982 #endif
983
984 cnp->cn_flags &= ~DOWHITEOUT;
985 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
986 break;
987 default:
988 panic("ufs_whiteout: unknown op");
989 }
990 return (error);
991 }
992
993 /*
994 * Rename system call.
995 * rename("foo", "bar");
996 * is essentially
997 * unlink("bar");
998 * link("foo", "bar");
999 * unlink("foo");
1000 * but ``atomically''. Can't do full commit without saving state in the
1001 * inode on disk which isn't feasible at this time. Best we can do is
1002 * always guarantee the target exists.
1003 *
1004 * Basic algorithm is:
1005 *
1006 * 1) Bump link count on source while we're linking it to the
1007 * target. This also ensure the inode won't be deleted out
1008 * from underneath us while we work (it may be truncated by
1009 * a concurrent `trunc' or `open' for creation).
1010 * 2) Link source to destination. If destination already exists,
1011 * delete it first.
1012 * 3) Unlink source reference to inode if still around. If a
1013 * directory was moved and the parent of the destination
1014 * is different from the source, patch the ".." entry in the
1015 * directory.
1016 */
1017 static int
1018 ufs_rename(ap)
1019 struct vop_rename_args /* {
1020 struct vnode *a_fdvp;
1021 struct vnode *a_fvp;
1022 struct componentname *a_fcnp;
1023 struct vnode *a_tdvp;
1024 struct vnode *a_tvp;
1025 struct componentname *a_tcnp;
1026 } */ *ap;
1027 {
1028 struct vnode *tvp = ap->a_tvp;
1029 struct vnode *tdvp = ap->a_tdvp;
1030 struct vnode *fvp = ap->a_fvp;
1031 struct vnode *fdvp = ap->a_fdvp;
1032 struct componentname *tcnp = ap->a_tcnp;
1033 struct componentname *fcnp = ap->a_fcnp;
1034 struct thread *td = fcnp->cn_thread;
1035 struct inode *ip, *xp, *dp;
1036 struct direct newdir;
1037 int doingdirectory = 0, oldparent = 0, newparent = 0;
1038 int error = 0, ioflag;
1039 ino_t fvp_ino;
1040
1041 #ifdef INVARIANTS
1042 if ((tcnp->cn_flags & HASBUF) == 0 ||
1043 (fcnp->cn_flags & HASBUF) == 0)
1044 panic("ufs_rename: no name");
1045 #endif
1046 /*
1047 * Check for cross-device rename.
1048 */
1049 if ((fvp->v_mount != tdvp->v_mount) ||
1050 (tvp && (fvp->v_mount != tvp->v_mount))) {
1051 error = EXDEV;
1052 abortit:
1053 if (tdvp == tvp)
1054 vrele(tdvp);
1055 else
1056 vput(tdvp);
1057 if (tvp)
1058 vput(tvp);
1059 vrele(fdvp);
1060 vrele(fvp);
1061 return (error);
1062 }
1063
1064 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1065 (VTOI(tdvp)->i_flags & APPEND))) {
1066 error = EPERM;
1067 goto abortit;
1068 }
1069
1070 /*
1071 * Renaming a file to itself has no effect. The upper layers should
1072 * not call us in that case. Temporarily just warn if they do.
1073 */
1074 if (fvp == tvp) {
1075 printf("ufs_rename: fvp == tvp (can't happen)\n");
1076 error = 0;
1077 goto abortit;
1078 }
1079
1080 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1081 goto abortit;
1082 dp = VTOI(fdvp);
1083 ip = VTOI(fvp);
1084 if (ip->i_nlink >= LINK_MAX) {
1085 VOP_UNLOCK(fvp, 0);
1086 error = EMLINK;
1087 goto abortit;
1088 }
1089 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1090 || (dp->i_flags & APPEND)) {
1091 VOP_UNLOCK(fvp, 0);
1092 error = EPERM;
1093 goto abortit;
1094 }
1095 if ((ip->i_mode & IFMT) == IFDIR) {
1096 /*
1097 * Avoid ".", "..", and aliases of "." for obvious reasons.
1098 */
1099 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1100 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
1101 (ip->i_flag & IN_RENAME)) {
1102 VOP_UNLOCK(fvp, 0);
1103 error = EINVAL;
1104 goto abortit;
1105 }
1106 ip->i_flag |= IN_RENAME;
1107 oldparent = dp->i_number;
1108 doingdirectory = 1;
1109 }
1110 vrele(fdvp);
1111
1112 /*
1113 * When the target exists, both the directory
1114 * and target vnodes are returned locked.
1115 */
1116 dp = VTOI(tdvp);
1117 xp = NULL;
1118 if (tvp)
1119 xp = VTOI(tvp);
1120
1121 /*
1122 * 1) Bump link count while we're moving stuff
1123 * around. If we crash somewhere before
1124 * completing our work, the link count
1125 * may be wrong, but correctable.
1126 */
1127 ip->i_effnlink++;
1128 ip->i_nlink++;
1129 DIP_SET(ip, i_nlink, ip->i_nlink);
1130 ip->i_flag |= IN_CHANGE;
1131 if (DOINGSOFTDEP(fvp))
1132 softdep_change_linkcnt(ip);
1133 if ((error = UFS_UPDATE(fvp, !(DOINGSOFTDEP(fvp) |
1134 DOINGASYNC(fvp)))) != 0) {
1135 VOP_UNLOCK(fvp, 0);
1136 goto bad;
1137 }
1138
1139 /*
1140 * If ".." must be changed (ie the directory gets a new
1141 * parent) then the source directory must not be in the
1142 * directory hierarchy above the target, as this would
1143 * orphan everything below the source directory. Also
1144 * the user must have write permission in the source so
1145 * as to be able to change "..". We must repeat the call
1146 * to namei, as the parent directory is unlocked by the
1147 * call to checkpath().
1148 */
1149 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1150 fvp_ino = ip->i_number;
1151 VOP_UNLOCK(fvp, 0);
1152 if (oldparent != dp->i_number)
1153 newparent = dp->i_number;
1154 if (doingdirectory && newparent) {
1155 if (error) /* write access check above */
1156 goto bad;
1157 if (xp != NULL)
1158 vput(tvp);
1159 error = ufs_checkpath(fvp_ino, dp, tcnp->cn_cred);
1160 if (error)
1161 goto out;
1162 if ((tcnp->cn_flags & SAVESTART) == 0)
1163 panic("ufs_rename: lost to startdir");
1164 VREF(tdvp);
1165 error = relookup(tdvp, &tvp, tcnp);
1166 if (error)
1167 goto out;
1168 vrele(tdvp);
1169 dp = VTOI(tdvp);
1170 xp = NULL;
1171 if (tvp)
1172 xp = VTOI(tvp);
1173 }
1174 /*
1175 * 2) If target doesn't exist, link the target
1176 * to the source and unlink the source.
1177 * Otherwise, rewrite the target directory
1178 * entry to reference the source inode and
1179 * expunge the original entry's existence.
1180 */
1181 if (xp == NULL) {
1182 if (dp->i_dev != ip->i_dev)
1183 panic("ufs_rename: EXDEV");
1184 /*
1185 * Account for ".." in new directory.
1186 * When source and destination have the same
1187 * parent we don't fool with the link count.
1188 */
1189 if (doingdirectory && newparent) {
1190 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1191 error = EMLINK;
1192 goto bad;
1193 }
1194 dp->i_effnlink++;
1195 dp->i_nlink++;
1196 DIP_SET(dp, i_nlink, dp->i_nlink);
1197 dp->i_flag |= IN_CHANGE;
1198 if (DOINGSOFTDEP(tdvp))
1199 softdep_change_linkcnt(dp);
1200 error = UFS_UPDATE(tdvp, !(DOINGSOFTDEP(tdvp) |
1201 DOINGASYNC(tdvp)));
1202 if (error)
1203 goto bad;
1204 }
1205 ufs_makedirentry(ip, tcnp, &newdir);
1206 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1207 if (error) {
1208 if (doingdirectory && newparent) {
1209 dp->i_effnlink--;
1210 dp->i_nlink--;
1211 DIP_SET(dp, i_nlink, dp->i_nlink);
1212 dp->i_flag |= IN_CHANGE;
1213 if (DOINGSOFTDEP(tdvp))
1214 softdep_change_linkcnt(dp);
1215 (void)UFS_UPDATE(tdvp, 1);
1216 }
1217 goto bad;
1218 }
1219 vput(tdvp);
1220 } else {
1221 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1222 panic("ufs_rename: EXDEV");
1223 /*
1224 * Short circuit rename(foo, foo).
1225 */
1226 if (xp->i_number == ip->i_number)
1227 panic("ufs_rename: same file");
1228 /*
1229 * If the parent directory is "sticky", then the caller
1230 * must possess VADMIN for the parent directory, or the
1231 * destination of the rename. This implements append-only
1232 * directories.
1233 */
1234 if ((dp->i_mode & S_ISTXT) &&
1235 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1236 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1237 error = EPERM;
1238 goto bad;
1239 }
1240 /*
1241 * Target must be empty if a directory and have no links
1242 * to it. Also, ensure source and target are compatible
1243 * (both directories, or both not directories).
1244 */
1245 if ((xp->i_mode&IFMT) == IFDIR) {
1246 if ((xp->i_effnlink > 2) ||
1247 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1248 error = ENOTEMPTY;
1249 goto bad;
1250 }
1251 if (!doingdirectory) {
1252 error = ENOTDIR;
1253 goto bad;
1254 }
1255 cache_purge(tdvp);
1256 } else if (doingdirectory) {
1257 error = EISDIR;
1258 goto bad;
1259 }
1260 error = ufs_dirrewrite(dp, xp, ip->i_number,
1261 IFTODT(ip->i_mode),
1262 (doingdirectory && newparent) ? newparent : doingdirectory);
1263 if (error)
1264 goto bad;
1265 if (doingdirectory) {
1266 if (!newparent) {
1267 dp->i_effnlink--;
1268 if (DOINGSOFTDEP(tdvp))
1269 softdep_change_linkcnt(dp);
1270 }
1271 xp->i_effnlink--;
1272 if (DOINGSOFTDEP(tvp))
1273 softdep_change_linkcnt(xp);
1274 }
1275 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1276 /*
1277 * Truncate inode. The only stuff left in the directory
1278 * is "." and "..". The "." reference is inconsequential
1279 * since we are quashing it. We have removed the "."
1280 * reference and the reference in the parent directory,
1281 * but there may be other hard links. The soft
1282 * dependency code will arrange to do these operations
1283 * after the parent directory entry has been deleted on
1284 * disk, so when running with that code we avoid doing
1285 * them now.
1286 */
1287 if (!newparent) {
1288 dp->i_nlink--;
1289 DIP_SET(dp, i_nlink, dp->i_nlink);
1290 dp->i_flag |= IN_CHANGE;
1291 }
1292 xp->i_nlink--;
1293 DIP_SET(xp, i_nlink, xp->i_nlink);
1294 xp->i_flag |= IN_CHANGE;
1295 ioflag = IO_NORMAL;
1296 if (!DOINGASYNC(tvp))
1297 ioflag |= IO_SYNC;
1298 if ((error = UFS_TRUNCATE(tvp, (off_t)0, ioflag,
1299 tcnp->cn_cred, tcnp->cn_thread)) != 0)
1300 goto bad;
1301 }
1302 vput(tdvp);
1303 vput(tvp);
1304 xp = NULL;
1305 }
1306
1307 /*
1308 * 3) Unlink the source.
1309 */
1310 fcnp->cn_flags &= ~MODMASK;
1311 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1312 if ((fcnp->cn_flags & SAVESTART) == 0)
1313 panic("ufs_rename: lost from startdir");
1314 VREF(fdvp);
1315 error = relookup(fdvp, &fvp, fcnp);
1316 if (error == 0)
1317 vrele(fdvp);
1318 if (fvp != NULL) {
1319 xp = VTOI(fvp);
1320 dp = VTOI(fdvp);
1321 } else {
1322 /*
1323 * From name has disappeared. IN_RENAME is not sufficient
1324 * to protect against directory races due to timing windows,
1325 * so we have to remove the panic. XXX the only real way
1326 * to solve this issue is at a much higher level. By the
1327 * time we hit ufs_rename() it's too late.
1328 */
1329 #if 0
1330 if (doingdirectory)
1331 panic("ufs_rename: lost dir entry");
1332 #endif
1333 vrele(ap->a_fvp);
1334 return (0);
1335 }
1336 /*
1337 * Ensure that the directory entry still exists and has not
1338 * changed while the new name has been entered. If the source is
1339 * a file then the entry may have been unlinked or renamed. In
1340 * either case there is no further work to be done. If the source
1341 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1342 * flag ensures that it cannot be moved by another rename or removed
1343 * by a rmdir.
1344 */
1345 if (xp != ip) {
1346 /*
1347 * From name resolves to a different inode. IN_RENAME is
1348 * not sufficient protection against timing window races
1349 * so we can't panic here. XXX the only real way
1350 * to solve this issue is at a much higher level. By the
1351 * time we hit ufs_rename() it's too late.
1352 */
1353 #if 0
1354 if (doingdirectory)
1355 panic("ufs_rename: lost dir entry");
1356 #endif
1357 } else {
1358 /*
1359 * If the source is a directory with a
1360 * new parent, the link count of the old
1361 * parent directory must be decremented
1362 * and ".." set to point to the new parent.
1363 */
1364 if (doingdirectory && newparent) {
1365 xp->i_offset = mastertemplate.dot_reclen;
1366 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1367 cache_purge(fdvp);
1368 }
1369 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1370 xp->i_flag &= ~IN_RENAME;
1371 }
1372 if (dp)
1373 vput(fdvp);
1374 if (xp)
1375 vput(fvp);
1376 vrele(ap->a_fvp);
1377 return (error);
1378
1379 bad:
1380 if (xp)
1381 vput(ITOV(xp));
1382 vput(ITOV(dp));
1383 out:
1384 if (doingdirectory)
1385 ip->i_flag &= ~IN_RENAME;
1386 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1387 ip->i_effnlink--;
1388 ip->i_nlink--;
1389 DIP_SET(ip, i_nlink, ip->i_nlink);
1390 ip->i_flag |= IN_CHANGE;
1391 ip->i_flag &= ~IN_RENAME;
1392 if (DOINGSOFTDEP(fvp))
1393 softdep_change_linkcnt(ip);
1394 vput(fvp);
1395 } else
1396 vrele(fvp);
1397 return (error);
1398 }
1399
1400 /*
1401 * Mkdir system call
1402 */
1403 static int
1404 ufs_mkdir(ap)
1405 struct vop_mkdir_args /* {
1406 struct vnode *a_dvp;
1407 struct vnode **a_vpp;
1408 struct componentname *a_cnp;
1409 struct vattr *a_vap;
1410 } */ *ap;
1411 {
1412 struct vnode *dvp = ap->a_dvp;
1413 struct vattr *vap = ap->a_vap;
1414 struct componentname *cnp = ap->a_cnp;
1415 struct inode *ip, *dp;
1416 struct vnode *tvp;
1417 struct buf *bp;
1418 struct dirtemplate dirtemplate, *dtp;
1419 struct direct newdir;
1420 #ifdef UFS_ACL
1421 struct acl *acl, *dacl;
1422 #endif
1423 int error, dmode;
1424 long blkoff;
1425
1426 #ifdef INVARIANTS
1427 if ((cnp->cn_flags & HASBUF) == 0)
1428 panic("ufs_mkdir: no name");
1429 #endif
1430 dp = VTOI(dvp);
1431 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1432 error = EMLINK;
1433 goto out;
1434 }
1435 dmode = vap->va_mode & 0777;
1436 dmode |= IFDIR;
1437 /*
1438 * Must simulate part of ufs_makeinode here to acquire the inode,
1439 * but not have it entered in the parent directory. The entry is
1440 * made later after writing "." and ".." entries.
1441 */
1442 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
1443 if (error)
1444 goto out;
1445 ip = VTOI(tvp);
1446 ip->i_gid = dp->i_gid;
1447 DIP_SET(ip, i_gid, dp->i_gid);
1448 #ifdef SUIDDIR
1449 {
1450 #ifdef QUOTA
1451 struct ucred ucred, *ucp;
1452 gid_t ucred_group;
1453 ucp = cnp->cn_cred;
1454 #endif
1455 /*
1456 * If we are hacking owners here, (only do this where told to)
1457 * and we are not giving it TO root, (would subvert quotas)
1458 * then go ahead and give it to the other user.
1459 * The new directory also inherits the SUID bit.
1460 * If user's UID and dir UID are the same,
1461 * 'give it away' so that the SUID is still forced on.
1462 */
1463 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1464 (dp->i_mode & ISUID) && dp->i_uid) {
1465 dmode |= ISUID;
1466 ip->i_uid = dp->i_uid;
1467 DIP_SET(ip, i_uid, dp->i_uid);
1468 #ifdef QUOTA
1469 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1470 /*
1471 * Make sure the correct user gets charged
1472 * for the space.
1473 * Make a dummy credential for the victim.
1474 * XXX This seems to never be accessed out of
1475 * our context so a stack variable is ok.
1476 */
1477 refcount_init(&ucred.cr_ref, 1);
1478 ucred.cr_uid = ip->i_uid;
1479 ucred.cr_ngroups = 1;
1480 ucred.cr_groups = &ucred_group;
1481 ucred.cr_groups[0] = dp->i_gid;
1482 ucp = &ucred;
1483 }
1484 #endif
1485 } else {
1486 ip->i_uid = cnp->cn_cred->cr_uid;
1487 DIP_SET(ip, i_uid, ip->i_uid);
1488 }
1489 #ifdef QUOTA
1490 if ((error = getinoquota(ip)) ||
1491 (error = chkiq(ip, 1, ucp, 0))) {
1492 UFS_VFREE(tvp, ip->i_number, dmode);
1493 vput(tvp);
1494 return (error);
1495 }
1496 #endif
1497 }
1498 #else /* !SUIDDIR */
1499 ip->i_uid = cnp->cn_cred->cr_uid;
1500 DIP_SET(ip, i_uid, ip->i_uid);
1501 #ifdef QUOTA
1502 if ((error = getinoquota(ip)) ||
1503 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
1504 UFS_VFREE(tvp, ip->i_number, dmode);
1505 vput(tvp);
1506 return (error);
1507 }
1508 #endif
1509 #endif /* !SUIDDIR */
1510 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1511 #ifdef UFS_ACL
1512 acl = dacl = NULL;
1513 if ((dvp->v_mount->mnt_flag & MNT_ACLS) != 0) {
1514 acl = acl_alloc(M_WAITOK);
1515 dacl = acl_alloc(M_WAITOK);
1516
1517 /*
1518 * Retrieve default ACL from parent, if any.
1519 */
1520 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cnp->cn_cred,
1521 cnp->cn_thread);
1522 switch (error) {
1523 case 0:
1524 /*
1525 * Retrieved a default ACL, so merge mode and ACL if
1526 * necessary. If the ACL is empty, fall through to
1527 * the "not defined or available" case.
1528 */
1529 if (acl->acl_cnt != 0) {
1530 dmode = acl_posix1e_newfilemode(dmode, acl);
1531 ip->i_mode = dmode;
1532 DIP_SET(ip, i_mode, dmode);
1533 *dacl = *acl;
1534 ufs_sync_acl_from_inode(ip, acl);
1535 break;
1536 }
1537 /* FALLTHROUGH */
1538
1539 case EOPNOTSUPP:
1540 /*
1541 * Just use the mode as-is.
1542 */
1543 ip->i_mode = dmode;
1544 DIP_SET(ip, i_mode, dmode);
1545 acl_free(acl);
1546 acl_free(dacl);
1547 dacl = acl = NULL;
1548 break;
1549
1550 default:
1551 UFS_VFREE(tvp, ip->i_number, dmode);
1552 vput(tvp);
1553 acl_free(acl);
1554 acl_free(dacl);
1555 return (error);
1556 }
1557 } else {
1558 #endif /* !UFS_ACL */
1559 ip->i_mode = dmode;
1560 DIP_SET(ip, i_mode, dmode);
1561 #ifdef UFS_ACL
1562 }
1563 #endif
1564 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1565 ip->i_effnlink = 2;
1566 ip->i_nlink = 2;
1567 DIP_SET(ip, i_nlink, 2);
1568 if (DOINGSOFTDEP(tvp))
1569 softdep_change_linkcnt(ip);
1570 if (cnp->cn_flags & ISWHITEOUT) {
1571 ip->i_flags |= UF_OPAQUE;
1572 DIP_SET(ip, i_flags, ip->i_flags);
1573 }
1574
1575 /*
1576 * Bump link count in parent directory to reflect work done below.
1577 * Should be done before reference is created so cleanup is
1578 * possible if we crash.
1579 */
1580 dp->i_effnlink++;
1581 dp->i_nlink++;
1582 DIP_SET(dp, i_nlink, dp->i_nlink);
1583 dp->i_flag |= IN_CHANGE;
1584 if (DOINGSOFTDEP(dvp))
1585 softdep_change_linkcnt(dp);
1586 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1587 if (error)
1588 goto bad;
1589 #ifdef MAC
1590 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
1591 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
1592 dvp, tvp, cnp);
1593 if (error)
1594 goto bad;
1595 }
1596 #endif
1597 #ifdef UFS_ACL
1598 if (acl != NULL) {
1599 /*
1600 * XXX: If we abort now, will Soft Updates notify the extattr
1601 * code that the EAs for the file need to be released?
1602 */
1603 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cnp->cn_cred,
1604 cnp->cn_thread);
1605 if (error == 0)
1606 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl,
1607 cnp->cn_cred, cnp->cn_thread);
1608 switch (error) {
1609 case 0:
1610 break;
1611
1612 case EOPNOTSUPP:
1613 /*
1614 * XXX: This should not happen, as EOPNOTSUPP above
1615 * was supposed to free acl.
1616 */
1617 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1618 /*
1619 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1620 */
1621 break;
1622
1623 default:
1624 acl_free(acl);
1625 acl_free(dacl);
1626 dacl = acl = NULL;
1627 goto bad;
1628 }
1629 acl_free(acl);
1630 acl_free(dacl);
1631 dacl = acl = NULL;
1632 }
1633 #endif /* !UFS_ACL */
1634
1635 /*
1636 * Initialize directory with "." and ".." from static template.
1637 */
1638 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1639 dtp = &mastertemplate;
1640 else
1641 dtp = (struct dirtemplate *)&omastertemplate;
1642 dirtemplate = *dtp;
1643 dirtemplate.dot_ino = ip->i_number;
1644 dirtemplate.dotdot_ino = dp->i_number;
1645 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
1646 BA_CLRBUF, &bp)) != 0)
1647 goto bad;
1648 ip->i_size = DIRBLKSIZ;
1649 DIP_SET(ip, i_size, DIRBLKSIZ);
1650 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1651 vnode_pager_setsize(tvp, (u_long)ip->i_size);
1652 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1653 if (DOINGSOFTDEP(tvp)) {
1654 /*
1655 * Ensure that the entire newly allocated block is a
1656 * valid directory so that future growth within the
1657 * block does not have to ensure that the block is
1658 * written before the inode.
1659 */
1660 blkoff = DIRBLKSIZ;
1661 while (blkoff < bp->b_bcount) {
1662 ((struct direct *)
1663 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1664 blkoff += DIRBLKSIZ;
1665 }
1666 }
1667 if ((error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) |
1668 DOINGASYNC(tvp)))) != 0) {
1669 (void)bwrite(bp);
1670 goto bad;
1671 }
1672 /*
1673 * Directory set up, now install its entry in the parent directory.
1674 *
1675 * If we are not doing soft dependencies, then we must write out the
1676 * buffer containing the new directory body before entering the new
1677 * name in the parent. If we are doing soft dependencies, then the
1678 * buffer containing the new directory body will be passed to and
1679 * released in the soft dependency code after the code has attached
1680 * an appropriate ordering dependency to the buffer which ensures that
1681 * the buffer is written before the new name is written in the parent.
1682 */
1683 if (DOINGASYNC(dvp))
1684 bdwrite(bp);
1685 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
1686 goto bad;
1687 ufs_makedirentry(ip, cnp, &newdir);
1688 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1689
1690 bad:
1691 if (error == 0) {
1692 *ap->a_vpp = tvp;
1693 } else {
1694 #ifdef UFS_ACL
1695 if (acl != NULL)
1696 acl_free(acl);
1697 if (dacl != NULL)
1698 acl_free(dacl);
1699 #endif
1700 dp->i_effnlink--;
1701 dp->i_nlink--;
1702 DIP_SET(dp, i_nlink, dp->i_nlink);
1703 dp->i_flag |= IN_CHANGE;
1704 if (DOINGSOFTDEP(dvp))
1705 softdep_change_linkcnt(dp);
1706 /*
1707 * No need to do an explicit VOP_TRUNCATE here, vrele will
1708 * do this for us because we set the link count to 0.
1709 */
1710 ip->i_effnlink = 0;
1711 ip->i_nlink = 0;
1712 DIP_SET(ip, i_nlink, 0);
1713 ip->i_flag |= IN_CHANGE;
1714 if (DOINGSOFTDEP(tvp))
1715 softdep_change_linkcnt(ip);
1716 vput(tvp);
1717 }
1718 out:
1719 return (error);
1720 }
1721
1722 /*
1723 * Rmdir system call.
1724 */
1725 static int
1726 ufs_rmdir(ap)
1727 struct vop_rmdir_args /* {
1728 struct vnode *a_dvp;
1729 struct vnode *a_vp;
1730 struct componentname *a_cnp;
1731 } */ *ap;
1732 {
1733 struct vnode *vp = ap->a_vp;
1734 struct vnode *dvp = ap->a_dvp;
1735 struct componentname *cnp = ap->a_cnp;
1736 struct inode *ip, *dp;
1737 int error, ioflag;
1738
1739 ip = VTOI(vp);
1740 dp = VTOI(dvp);
1741
1742 /*
1743 * Do not remove a directory that is in the process of being renamed.
1744 * Verify the directory is empty (and valid). Rmdir ".." will not be
1745 * valid since ".." will contain a reference to the current directory
1746 * and thus be non-empty. Do not allow the removal of mounted on
1747 * directories (this can happen when an NFS exported filesystem
1748 * tries to remove a locally mounted on directory).
1749 */
1750 error = 0;
1751 if ((ip->i_flag & IN_RENAME) || ip->i_effnlink < 2) {
1752 error = EINVAL;
1753 goto out;
1754 }
1755 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1756 error = ENOTEMPTY;
1757 goto out;
1758 }
1759 if ((dp->i_flags & APPEND)
1760 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1761 error = EPERM;
1762 goto out;
1763 }
1764 if (vp->v_mountedhere != 0) {
1765 error = EINVAL;
1766 goto out;
1767 }
1768 #ifdef UFS_GJOURNAL
1769 ufs_gjournal_orphan(vp);
1770 #endif
1771 /*
1772 * Delete reference to directory before purging
1773 * inode. If we crash in between, the directory
1774 * will be reattached to lost+found,
1775 */
1776 dp->i_effnlink--;
1777 ip->i_effnlink--;
1778 if (DOINGSOFTDEP(vp)) {
1779 softdep_change_linkcnt(dp);
1780 softdep_change_linkcnt(ip);
1781 }
1782 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1783 if (error) {
1784 dp->i_effnlink++;
1785 ip->i_effnlink++;
1786 if (DOINGSOFTDEP(vp)) {
1787 softdep_change_linkcnt(dp);
1788 softdep_change_linkcnt(ip);
1789 }
1790 goto out;
1791 }
1792 cache_purge(dvp);
1793 /*
1794 * Truncate inode. The only stuff left in the directory is "." and
1795 * "..". The "." reference is inconsequential since we are quashing
1796 * it. The soft dependency code will arrange to do these operations
1797 * after the parent directory entry has been deleted on disk, so
1798 * when running with that code we avoid doing them now.
1799 */
1800 if (!DOINGSOFTDEP(vp)) {
1801 dp->i_nlink--;
1802 DIP_SET(dp, i_nlink, dp->i_nlink);
1803 dp->i_flag |= IN_CHANGE;
1804 ip->i_nlink--;
1805 DIP_SET(ip, i_nlink, ip->i_nlink);
1806 ip->i_flag |= IN_CHANGE;
1807 ioflag = IO_NORMAL;
1808 if (!DOINGASYNC(vp))
1809 ioflag |= IO_SYNC;
1810 error = UFS_TRUNCATE(vp, (off_t)0, ioflag, cnp->cn_cred,
1811 cnp->cn_thread);
1812 }
1813 cache_purge(vp);
1814 #ifdef UFS_DIRHASH
1815 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1816 if (ip->i_dirhash != NULL)
1817 ufsdirhash_free(ip);
1818 #endif
1819 out:
1820 return (error);
1821 }
1822
1823 /*
1824 * symlink -- make a symbolic link
1825 */
1826 static int
1827 ufs_symlink(ap)
1828 struct vop_symlink_args /* {
1829 struct vnode *a_dvp;
1830 struct vnode **a_vpp;
1831 struct componentname *a_cnp;
1832 struct vattr *a_vap;
1833 char *a_target;
1834 } */ *ap;
1835 {
1836 struct vnode *vp, **vpp = ap->a_vpp;
1837 struct inode *ip;
1838 int len, error;
1839
1840 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1841 vpp, ap->a_cnp);
1842 if (error)
1843 return (error);
1844 vp = *vpp;
1845 len = strlen(ap->a_target);
1846 if (len < vp->v_mount->mnt_maxsymlinklen) {
1847 ip = VTOI(vp);
1848 bcopy(ap->a_target, SHORTLINK(ip), len);
1849 ip->i_size = len;
1850 DIP_SET(ip, i_size, len);
1851 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1852 } else
1853 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1854 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1855 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1856 if (error)
1857 vput(vp);
1858 return (error);
1859 }
1860
1861 /*
1862 * Vnode op for reading directories.
1863 *
1864 * The routine below assumes that the on-disk format of a directory
1865 * is the same as that defined by <sys/dirent.h>. If the on-disk
1866 * format changes, then it will be necessary to do a conversion
1867 * from the on-disk format that read returns to the format defined
1868 * by <sys/dirent.h>.
1869 */
1870 int
1871 ufs_readdir(ap)
1872 struct vop_readdir_args /* {
1873 struct vnode *a_vp;
1874 struct uio *a_uio;
1875 struct ucred *a_cred;
1876 int *a_eofflag;
1877 int *a_ncookies;
1878 u_long **a_cookies;
1879 } */ *ap;
1880 {
1881 struct uio *uio = ap->a_uio;
1882 int error;
1883 size_t count, lost;
1884 off_t off;
1885
1886 if (ap->a_ncookies != NULL)
1887 /*
1888 * Ensure that the block is aligned. The caller can use
1889 * the cookies to determine where in the block to start.
1890 */
1891 uio->uio_offset &= ~(DIRBLKSIZ - 1);
1892 off = uio->uio_offset;
1893 count = uio->uio_resid;
1894 /* Make sure we don't return partial entries. */
1895 if (count <= ((uio->uio_offset + count) & (DIRBLKSIZ -1)))
1896 return (EINVAL);
1897 count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
1898 lost = uio->uio_resid - count;
1899 uio->uio_resid = count;
1900 uio->uio_iov->iov_len = count;
1901 # if (BYTE_ORDER == LITTLE_ENDIAN)
1902 if (ap->a_vp->v_mount->mnt_maxsymlinklen > 0) {
1903 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1904 } else {
1905 struct dirent *dp, *edp;
1906 struct uio auio;
1907 struct iovec aiov;
1908 caddr_t dirbuf;
1909 int readcnt;
1910 u_char tmp;
1911
1912 auio = *uio;
1913 auio.uio_iov = &aiov;
1914 auio.uio_iovcnt = 1;
1915 auio.uio_segflg = UIO_SYSSPACE;
1916 aiov.iov_len = count;
1917 dirbuf = malloc(count, M_TEMP, M_WAITOK);
1918 aiov.iov_base = dirbuf;
1919 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
1920 if (error == 0) {
1921 readcnt = count - auio.uio_resid;
1922 edp = (struct dirent *)&dirbuf[readcnt];
1923 for (dp = (struct dirent *)dirbuf; dp < edp; ) {
1924 tmp = dp->d_namlen;
1925 dp->d_namlen = dp->d_type;
1926 dp->d_type = tmp;
1927 if (dp->d_reclen > 0) {
1928 dp = (struct dirent *)
1929 ((char *)dp + dp->d_reclen);
1930 } else {
1931 error = EIO;
1932 break;
1933 }
1934 }
1935 if (dp >= edp)
1936 error = uiomove(dirbuf, readcnt, uio);
1937 }
1938 free(dirbuf, M_TEMP);
1939 }
1940 # else
1941 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1942 # endif
1943 if (!error && ap->a_ncookies != NULL) {
1944 struct dirent* dpStart;
1945 struct dirent* dpEnd;
1946 struct dirent* dp;
1947 int ncookies;
1948 u_long *cookies;
1949 u_long *cookiep;
1950
1951 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1952 panic("ufs_readdir: unexpected uio from NFS server");
1953 dpStart = (struct dirent *)
1954 ((char *)uio->uio_iov->iov_base - (uio->uio_offset - off));
1955 dpEnd = (struct dirent *) uio->uio_iov->iov_base;
1956 for (dp = dpStart, ncookies = 0;
1957 dp < dpEnd;
1958 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen))
1959 ncookies++;
1960 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1961 M_WAITOK);
1962 for (dp = dpStart, cookiep = cookies;
1963 dp < dpEnd;
1964 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen)) {
1965 off += dp->d_reclen;
1966 *cookiep++ = (u_long) off;
1967 }
1968 *ap->a_ncookies = ncookies;
1969 *ap->a_cookies = cookies;
1970 }
1971 uio->uio_resid += lost;
1972 if (ap->a_eofflag)
1973 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
1974 return (error);
1975 }
1976
1977 /*
1978 * Return target name of a symbolic link
1979 */
1980 static int
1981 ufs_readlink(ap)
1982 struct vop_readlink_args /* {
1983 struct vnode *a_vp;
1984 struct uio *a_uio;
1985 struct ucred *a_cred;
1986 } */ *ap;
1987 {
1988 struct vnode *vp = ap->a_vp;
1989 struct inode *ip = VTOI(vp);
1990 doff_t isize;
1991
1992 isize = ip->i_size;
1993 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1994 DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */
1995 return (uiomove(SHORTLINK(ip), isize, ap->a_uio));
1996 }
1997 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1998 }
1999
2000 /*
2001 * Calculate the logical to physical mapping if not done already,
2002 * then call the device strategy routine.
2003 *
2004 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
2005 * deadlock on memory. See ufs_bmap() for details.
2006 */
2007 static int
2008 ufs_strategy(ap)
2009 struct vop_strategy_args /* {
2010 struct vnode *a_vp;
2011 struct buf *a_bp;
2012 } */ *ap;
2013 {
2014 struct buf *bp = ap->a_bp;
2015 struct vnode *vp = ap->a_vp;
2016 struct bufobj *bo;
2017 struct inode *ip;
2018 ufs2_daddr_t blkno;
2019 int error;
2020
2021 ip = VTOI(vp);
2022 if (bp->b_blkno == bp->b_lblkno) {
2023 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2024 bp->b_blkno = blkno;
2025 if (error) {
2026 bp->b_error = error;
2027 bp->b_ioflags |= BIO_ERROR;
2028 bufdone(bp);
2029 return (0);
2030 }
2031 if ((long)bp->b_blkno == -1)
2032 vfs_bio_clrbuf(bp);
2033 }
2034 if ((long)bp->b_blkno == -1) {
2035 bufdone(bp);
2036 return (0);
2037 }
2038 bp->b_iooffset = dbtob(bp->b_blkno);
2039 bo = ip->i_umbufobj;
2040 BO_STRATEGY(bo, bp);
2041 return (0);
2042 }
2043
2044 /*
2045 * Print out the contents of an inode.
2046 */
2047 static int
2048 ufs_print(ap)
2049 struct vop_print_args /* {
2050 struct vnode *a_vp;
2051 } */ *ap;
2052 {
2053 struct vnode *vp = ap->a_vp;
2054 struct inode *ip = VTOI(vp);
2055
2056 printf("\tino %lu, on dev %s", (u_long)ip->i_number,
2057 devtoname(ip->i_dev));
2058 if (vp->v_type == VFIFO)
2059 fifo_printinfo(vp);
2060 printf("\n");
2061 return (0);
2062 }
2063
2064 /*
2065 * Close wrapper for fifos.
2066 *
2067 * Update the times on the inode then do device close.
2068 */
2069 static int
2070 ufsfifo_close(ap)
2071 struct vop_close_args /* {
2072 struct vnode *a_vp;
2073 int a_fflag;
2074 struct ucred *a_cred;
2075 struct thread *a_td;
2076 } */ *ap;
2077 {
2078 struct vnode *vp = ap->a_vp;
2079 int usecount;
2080
2081 VI_LOCK(vp);
2082 usecount = vp->v_usecount;
2083 if (usecount > 1)
2084 ufs_itimes_locked(vp);
2085 VI_UNLOCK(vp);
2086 return (fifo_specops.vop_close(ap));
2087 }
2088
2089 /*
2090 * Kqfilter wrapper for fifos.
2091 *
2092 * Fall through to ufs kqfilter routines if needed
2093 */
2094 static int
2095 ufsfifo_kqfilter(ap)
2096 struct vop_kqfilter_args *ap;
2097 {
2098 int error;
2099
2100 error = fifo_specops.vop_kqfilter(ap);
2101 if (error)
2102 error = vfs_kqfilter(ap);
2103 return (error);
2104 }
2105
2106 /*
2107 * Return POSIX pathconf information applicable to fifos.
2108 */
2109 static int
2110 ufsfifo_pathconf(ap)
2111 struct vop_pathconf_args /* {
2112 struct vnode *a_vp;
2113 int a_name;
2114 int *a_retval;
2115 } */ *ap;
2116 {
2117
2118 switch (ap->a_name) {
2119 case _PC_ACL_EXTENDED:
2120 case _PC_ACL_PATH_MAX:
2121 case _PC_MAC_PRESENT:
2122 return (ufs_pathconf(ap));
2123 default:
2124 return (fifo_specops.vop_pathconf(ap));
2125 }
2126 /* NOTREACHED */
2127 }
2128
2129 /*
2130 * Return POSIX pathconf information applicable to ufs filesystems.
2131 */
2132 static int
2133 ufs_pathconf(ap)
2134 struct vop_pathconf_args /* {
2135 struct vnode *a_vp;
2136 int a_name;
2137 int *a_retval;
2138 } */ *ap;
2139 {
2140 int error;
2141
2142 error = 0;
2143 switch (ap->a_name) {
2144 case _PC_LINK_MAX:
2145 *ap->a_retval = LINK_MAX;
2146 break;
2147 case _PC_NAME_MAX:
2148 *ap->a_retval = NAME_MAX;
2149 break;
2150 case _PC_PATH_MAX:
2151 *ap->a_retval = PATH_MAX;
2152 break;
2153 case _PC_PIPE_BUF:
2154 *ap->a_retval = PIPE_BUF;
2155 break;
2156 case _PC_CHOWN_RESTRICTED:
2157 *ap->a_retval = 1;
2158 break;
2159 case _PC_NO_TRUNC:
2160 *ap->a_retval = 1;
2161 break;
2162 case _PC_ACL_EXTENDED:
2163 #ifdef UFS_ACL
2164 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2165 *ap->a_retval = 1;
2166 else
2167 *ap->a_retval = 0;
2168 #else
2169 *ap->a_retval = 0;
2170 #endif
2171 break;
2172 case _PC_ACL_PATH_MAX:
2173 #ifdef UFS_ACL
2174 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2175 *ap->a_retval = ACL_MAX_ENTRIES;
2176 else
2177 *ap->a_retval = 3;
2178 #else
2179 *ap->a_retval = 3;
2180 #endif
2181 break;
2182 case _PC_MAC_PRESENT:
2183 #ifdef MAC
2184 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2185 *ap->a_retval = 1;
2186 else
2187 *ap->a_retval = 0;
2188 #else
2189 *ap->a_retval = 0;
2190 #endif
2191 break;
2192 case _PC_ASYNC_IO:
2193 /* _PC_ASYNC_IO should have been handled by upper layers. */
2194 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
2195 error = EINVAL;
2196 break;
2197 case _PC_PRIO_IO:
2198 *ap->a_retval = 0;
2199 break;
2200 case _PC_SYNC_IO:
2201 *ap->a_retval = 0;
2202 break;
2203 case _PC_ALLOC_SIZE_MIN:
2204 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2205 break;
2206 case _PC_FILESIZEBITS:
2207 *ap->a_retval = 64;
2208 break;
2209 case _PC_REC_INCR_XFER_SIZE:
2210 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2211 break;
2212 case _PC_REC_MAX_XFER_SIZE:
2213 *ap->a_retval = -1; /* means ``unlimited'' */
2214 break;
2215 case _PC_REC_MIN_XFER_SIZE:
2216 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2217 break;
2218 case _PC_REC_XFER_ALIGN:
2219 *ap->a_retval = PAGE_SIZE;
2220 break;
2221 case _PC_SYMLINK_MAX:
2222 *ap->a_retval = MAXPATHLEN;
2223 break;
2224
2225 default:
2226 error = EINVAL;
2227 break;
2228 }
2229 return (error);
2230 }
2231
2232 /*
2233 * Initialize the vnode associated with a new inode, handle aliased
2234 * vnodes.
2235 */
2236 int
2237 ufs_vinit(mntp, fifoops, vpp)
2238 struct mount *mntp;
2239 struct vop_vector *fifoops;
2240 struct vnode **vpp;
2241 {
2242 struct inode *ip;
2243 struct vnode *vp;
2244
2245 vp = *vpp;
2246 ip = VTOI(vp);
2247 vp->v_type = IFTOVT(ip->i_mode);
2248 if (vp->v_type == VFIFO)
2249 vp->v_op = fifoops;
2250 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2251 if (ip->i_number == ROOTINO)
2252 vp->v_vflag |= VV_ROOT;
2253 *vpp = vp;
2254 return (0);
2255 }
2256
2257 /*
2258 * Allocate a new inode.
2259 * Vnode dvp must be locked.
2260 */
2261 static int
2262 ufs_makeinode(mode, dvp, vpp, cnp)
2263 int mode;
2264 struct vnode *dvp;
2265 struct vnode **vpp;
2266 struct componentname *cnp;
2267 {
2268 struct inode *ip, *pdir;
2269 struct direct newdir;
2270 struct vnode *tvp;
2271 #ifdef UFS_ACL
2272 struct acl *acl;
2273 #endif
2274 int error;
2275
2276 pdir = VTOI(dvp);
2277 #ifdef INVARIANTS
2278 if ((cnp->cn_flags & HASBUF) == 0)
2279 panic("ufs_makeinode: no name");
2280 #endif
2281 *vpp = NULL;
2282 if ((mode & IFMT) == 0)
2283 mode |= IFREG;
2284
2285 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2286 if (error)
2287 return (error);
2288 ip = VTOI(tvp);
2289 ip->i_gid = pdir->i_gid;
2290 DIP_SET(ip, i_gid, pdir->i_gid);
2291 #ifdef SUIDDIR
2292 {
2293 #ifdef QUOTA
2294 struct ucred ucred, *ucp;
2295 gid_t ucred_group;
2296 ucp = cnp->cn_cred;
2297 #endif
2298 /*
2299 * If we are not the owner of the directory,
2300 * and we are hacking owners here, (only do this where told to)
2301 * and we are not giving it TO root, (would subvert quotas)
2302 * then go ahead and give it to the other user.
2303 * Note that this drops off the execute bits for security.
2304 */
2305 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2306 (pdir->i_mode & ISUID) &&
2307 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2308 ip->i_uid = pdir->i_uid;
2309 DIP_SET(ip, i_uid, ip->i_uid);
2310 mode &= ~07111;
2311 #ifdef QUOTA
2312 /*
2313 * Make sure the correct user gets charged
2314 * for the space.
2315 * Quickly knock up a dummy credential for the victim.
2316 * XXX This seems to never be accessed out of our
2317 * context so a stack variable is ok.
2318 */
2319 refcount_init(&ucred.cr_ref, 1);
2320 ucred.cr_uid = ip->i_uid;
2321 ucred.cr_ngroups = 1;
2322 ucred.cr_groups = &ucred_group;
2323 ucred.cr_groups[0] = pdir->i_gid;
2324 ucp = &ucred;
2325 #endif
2326 } else {
2327 ip->i_uid = cnp->cn_cred->cr_uid;
2328 DIP_SET(ip, i_uid, ip->i_uid);
2329 }
2330
2331 #ifdef QUOTA
2332 if ((error = getinoquota(ip)) ||
2333 (error = chkiq(ip, 1, ucp, 0))) {
2334 UFS_VFREE(tvp, ip->i_number, mode);
2335 vput(tvp);
2336 return (error);
2337 }
2338 #endif
2339 }
2340 #else /* !SUIDDIR */
2341 ip->i_uid = cnp->cn_cred->cr_uid;
2342 DIP_SET(ip, i_uid, ip->i_uid);
2343 #ifdef QUOTA
2344 if ((error = getinoquota(ip)) ||
2345 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2346 UFS_VFREE(tvp, ip->i_number, mode);
2347 vput(tvp);
2348 return (error);
2349 }
2350 #endif
2351 #endif /* !SUIDDIR */
2352 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2353 #ifdef UFS_ACL
2354 acl = NULL;
2355 if ((dvp->v_mount->mnt_flag & MNT_ACLS) != 0) {
2356 acl = acl_alloc(M_WAITOK);
2357
2358 /*
2359 * Retrieve default ACL for parent, if any.
2360 */
2361 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cnp->cn_cred,
2362 cnp->cn_thread);
2363 switch (error) {
2364 case 0:
2365 /*
2366 * Retrieved a default ACL, so merge mode and ACL if
2367 * necessary.
2368 */
2369 if (acl->acl_cnt != 0) {
2370 /*
2371 * Two possible ways for default ACL to not
2372 * be present. First, the EA can be
2373 * undefined, or second, the default ACL can
2374 * be blank. If it's blank, fall through to
2375 * the it's not defined case.
2376 */
2377 mode = acl_posix1e_newfilemode(mode, acl);
2378 ip->i_mode = mode;
2379 DIP_SET(ip, i_mode, mode);
2380 ufs_sync_acl_from_inode(ip, acl);
2381 break;
2382 }
2383 /* FALLTHROUGH */
2384
2385 case EOPNOTSUPP:
2386 /*
2387 * Just use the mode as-is.
2388 */
2389 ip->i_mode = mode;
2390 DIP_SET(ip, i_mode, mode);
2391 acl_free(acl);
2392 acl = NULL;
2393 break;
2394
2395 default:
2396 UFS_VFREE(tvp, ip->i_number, mode);
2397 vput(tvp);
2398 acl_free(acl);
2399 acl = NULL;
2400 return (error);
2401 }
2402 } else {
2403 #endif
2404 ip->i_mode = mode;
2405 DIP_SET(ip, i_mode, mode);
2406 #ifdef UFS_ACL
2407 }
2408 #endif
2409 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2410 ip->i_effnlink = 1;
2411 ip->i_nlink = 1;
2412 DIP_SET(ip, i_nlink, 1);
2413 if (DOINGSOFTDEP(tvp))
2414 softdep_change_linkcnt(ip);
2415 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2416 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2417 ip->i_mode &= ~ISGID;
2418 DIP_SET(ip, i_mode, ip->i_mode);
2419 }
2420
2421 if (cnp->cn_flags & ISWHITEOUT) {
2422 ip->i_flags |= UF_OPAQUE;
2423 DIP_SET(ip, i_flags, ip->i_flags);
2424 }
2425
2426 /*
2427 * Make sure inode goes to disk before directory entry.
2428 */
2429 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2430 if (error)
2431 goto bad;
2432 #ifdef MAC
2433 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2434 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2435 dvp, tvp, cnp);
2436 if (error)
2437 goto bad;
2438 }
2439 #endif
2440 #ifdef UFS_ACL
2441 if (acl != NULL) {
2442 /*
2443 * XXX: If we abort now, will Soft Updates notify the extattr
2444 * code that the EAs for the file need to be released?
2445 */
2446 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cnp->cn_cred,
2447 cnp->cn_thread);
2448 switch (error) {
2449 case 0:
2450 break;
2451
2452 case EOPNOTSUPP:
2453 /*
2454 * XXX: This should not happen, as EOPNOTSUPP above was
2455 * supposed to free acl.
2456 */
2457 printf("ufs_makeinode: VOP_GETACL() but no "
2458 "VOP_SETACL()\n");
2459 /* panic("ufs_makeinode: VOP_GETACL() but no "
2460 "VOP_SETACL()"); */
2461 break;
2462
2463 default:
2464 acl_free(acl);
2465 goto bad;
2466 }
2467 acl_free(acl);
2468 }
2469 #endif /* !UFS_ACL */
2470 ufs_makedirentry(ip, cnp, &newdir);
2471 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2472 if (error)
2473 goto bad;
2474 *vpp = tvp;
2475 return (0);
2476
2477 bad:
2478 /*
2479 * Write error occurred trying to update the inode
2480 * or the directory so must deallocate the inode.
2481 */
2482 ip->i_effnlink = 0;
2483 ip->i_nlink = 0;
2484 DIP_SET(ip, i_nlink, 0);
2485 ip->i_flag |= IN_CHANGE;
2486 if (DOINGSOFTDEP(tvp))
2487 softdep_change_linkcnt(ip);
2488 vput(tvp);
2489 return (error);
2490 }
2491
2492 /* Global vfs data structures for ufs. */
2493 struct vop_vector ufs_vnodeops = {
2494 .vop_default = &default_vnodeops,
2495 .vop_fsync = VOP_PANIC,
2496 .vop_read = VOP_PANIC,
2497 .vop_reallocblks = VOP_PANIC,
2498 .vop_write = VOP_PANIC,
2499 .vop_access = ufs_access,
2500 .vop_bmap = ufs_bmap,
2501 .vop_cachedlookup = ufs_lookup,
2502 .vop_close = ufs_close,
2503 .vop_create = ufs_create,
2504 .vop_getattr = ufs_getattr,
2505 .vop_inactive = ufs_inactive,
2506 .vop_link = ufs_link,
2507 .vop_lookup = vfs_cache_lookup,
2508 .vop_markatime = ufs_markatime,
2509 .vop_mkdir = ufs_mkdir,
2510 .vop_mknod = ufs_mknod,
2511 .vop_open = ufs_open,
2512 .vop_pathconf = ufs_pathconf,
2513 .vop_poll = vop_stdpoll,
2514 .vop_print = ufs_print,
2515 .vop_readdir = ufs_readdir,
2516 .vop_readlink = ufs_readlink,
2517 .vop_reclaim = ufs_reclaim,
2518 .vop_remove = ufs_remove,
2519 .vop_rename = ufs_rename,
2520 .vop_rmdir = ufs_rmdir,
2521 .vop_setattr = ufs_setattr,
2522 #ifdef MAC
2523 .vop_setlabel = vop_stdsetlabel_ea,
2524 #endif
2525 .vop_strategy = ufs_strategy,
2526 .vop_symlink = ufs_symlink,
2527 .vop_whiteout = ufs_whiteout,
2528 #ifdef UFS_EXTATTR
2529 .vop_getextattr = ufs_getextattr,
2530 .vop_deleteextattr = ufs_deleteextattr,
2531 .vop_setextattr = ufs_setextattr,
2532 #endif
2533 #ifdef UFS_ACL
2534 .vop_getacl = ufs_getacl,
2535 .vop_setacl = ufs_setacl,
2536 .vop_aclcheck = ufs_aclcheck,
2537 #endif
2538 };
2539
2540 struct vop_vector ufs_fifoops = {
2541 .vop_default = &fifo_specops,
2542 .vop_fsync = VOP_PANIC,
2543 .vop_access = ufs_access,
2544 .vop_close = ufsfifo_close,
2545 .vop_getattr = ufs_getattr,
2546 .vop_inactive = ufs_inactive,
2547 .vop_kqfilter = ufsfifo_kqfilter,
2548 .vop_markatime = ufs_markatime,
2549 .vop_pathconf = ufsfifo_pathconf,
2550 .vop_print = ufs_print,
2551 .vop_read = VOP_PANIC,
2552 .vop_reclaim = ufs_reclaim,
2553 .vop_setattr = ufs_setattr,
2554 #ifdef MAC
2555 .vop_setlabel = vop_stdsetlabel_ea,
2556 #endif
2557 .vop_write = VOP_PANIC,
2558 #ifdef UFS_EXTATTR
2559 .vop_getextattr = ufs_getextattr,
2560 .vop_deleteextattr = ufs_deleteextattr,
2561 .vop_setextattr = ufs_setextattr,
2562 #endif
2563 #ifdef UFS_ACL
2564 .vop_getacl = ufs_getacl,
2565 .vop_setacl = ufs_setacl,
2566 .vop_aclcheck = ufs_aclcheck,
2567 #endif
2568 };
Cache object: ecca5720941b05a4b679ba8f3521899b
|