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/7.4/sys/ufs/ufs/ufs_vnops.c 194696 2009-06-23 10:59:59Z kib $");
39
40 #include "opt_mac.h"
41 #include "opt_quota.h"
42 #include "opt_suiddir.h"
43 #include "opt_ufs.h"
44 #include "opt_ffs.h"
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
49 #include <sys/namei.h>
50 #include <sys/kernel.h>
51 #include <sys/fcntl.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/mount.h>
56 #include <sys/priv.h>
57 #include <sys/refcount.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/lockf.h>
62 #include <sys/conf.h>
63 #include <sys/acl.h>
64 #include <sys/jail.h>
65
66 #include <machine/mutex.h>
67
68 #include <security/mac/mac_framework.h>
69
70 #include <sys/file.h> /* XXX */
71
72 #include <vm/vm.h>
73 #include <vm/vm_extern.h>
74
75 #include <fs/fifofs/fifo.h>
76
77 #include <ufs/ufs/acl.h>
78 #include <ufs/ufs/extattr.h>
79 #include <ufs/ufs/quota.h>
80 #include <ufs/ufs/inode.h>
81 #include <ufs/ufs/dir.h>
82 #include <ufs/ufs/ufsmount.h>
83 #include <ufs/ufs/ufs_extern.h>
84 #ifdef UFS_DIRHASH
85 #include <ufs/ufs/dirhash.h>
86 #endif
87 #ifdef UFS_GJOURNAL
88 #include <ufs/ufs/gjournal.h>
89 #endif
90
91 #include <ufs/ffs/ffs_extern.h>
92
93 static vop_access_t ufs_access;
94 static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *);
95 static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *);
96 static vop_close_t ufs_close;
97 static vop_create_t ufs_create;
98 static vop_getattr_t ufs_getattr;
99 static vop_link_t ufs_link;
100 static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
101 static vop_mkdir_t ufs_mkdir;
102 static vop_mknod_t ufs_mknod;
103 static vop_open_t ufs_open;
104 static vop_pathconf_t ufs_pathconf;
105 static vop_print_t ufs_print;
106 static vop_readlink_t ufs_readlink;
107 static vop_remove_t ufs_remove;
108 static vop_rename_t ufs_rename;
109 static vop_rmdir_t ufs_rmdir;
110 static vop_setattr_t ufs_setattr;
111 static vop_strategy_t ufs_strategy;
112 static vop_symlink_t ufs_symlink;
113 static vop_whiteout_t ufs_whiteout;
114 static vop_close_t ufsfifo_close;
115 static vop_kqfilter_t ufsfifo_kqfilter;
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 ip->i_modrev++;
160 }
161 if (ip->i_flag & IN_CHANGE) {
162 DIP_SET(ip, i_ctime, ts.tv_sec);
163 DIP_SET(ip, i_ctimensec, ts.tv_nsec);
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 int a_mode;
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 mode_t mode = ap->a_mode;
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 (mode & 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, ap->a_td) != 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, ap->a_td);
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, ap->a_td);
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 ((mode & 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 = uma_zalloc(acl_zone, 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_mode, 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_mode, 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_mode, ap->a_cred, NULL);
399 }
400 uma_zfree(acl_zone, 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_mode, 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 struct thread *a_td;
416 } */ *ap;
417 {
418 struct vnode *vp = ap->a_vp;
419 struct inode *ip = VTOI(vp);
420 struct vattr *vap = ap->a_vap;
421
422 VI_LOCK(vp);
423 ufs_itimes_locked(vp);
424 if (ip->i_ump->um_fstype == UFS1) {
425 vap->va_atime.tv_sec = ip->i_din1->di_atime;
426 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
427 } else {
428 vap->va_atime.tv_sec = ip->i_din2->di_atime;
429 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
430 }
431 VI_UNLOCK(vp);
432 /*
433 * Copy from inode table
434 */
435 vap->va_fsid = dev2udev(ip->i_dev);
436 vap->va_fileid = ip->i_number;
437 vap->va_mode = ip->i_mode & ~IFMT;
438 vap->va_nlink = ip->i_effnlink;
439 vap->va_uid = ip->i_uid;
440 vap->va_gid = ip->i_gid;
441 if (ip->i_ump->um_fstype == UFS1) {
442 vap->va_rdev = ip->i_din1->di_rdev;
443 vap->va_size = ip->i_din1->di_size;
444 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
445 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
446 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
447 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
448 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks);
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 }
460 vap->va_flags = ip->i_flags;
461 vap->va_gen = ip->i_gen;
462 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
463 vap->va_type = IFTOVT(ip->i_mode);
464 vap->va_filerev = ip->i_modrev;
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 struct thread *a_td;
478 } */ *ap;
479 {
480 struct vattr *vap = ap->a_vap;
481 struct vnode *vp = ap->a_vp;
482 struct inode *ip = VTOI(vp);
483 struct ucred *cred = ap->a_cred;
484 struct thread *td = ap->a_td;
485 int error;
486
487 /*
488 * Check for unsettable attributes.
489 */
490 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
491 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
492 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
493 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
494 return (EINVAL);
495 }
496 /*
497 * Mark for update the file's access time for vfs_mark_atime().
498 * We are doing this here to avoid some of the checks done
499 * below -- this operation is done by request of the kernel and
500 * should bypass some security checks. Things like read-only
501 * checks get handled by other levels (e.g., ffs_update()).
502 */
503 if (vap->va_vaflags & VA_MARK_ATIME) {
504 ip->i_flag |= IN_ACCESS;
505 return (0);
506 }
507 if (vap->va_flags != VNOVAL) {
508 if (vp->v_mount->mnt_flag & MNT_RDONLY)
509 return (EROFS);
510 /*
511 * Callers may only modify the file flags on objects they
512 * have VADMIN rights for.
513 */
514 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
515 return (error);
516 /*
517 * Unprivileged processes are not permitted to unset system
518 * flags, or modify flags if any system flags are set.
519 * Privileged non-jail processes may not modify system flags
520 * if securelevel > 0 and any existing system flags are set.
521 * Privileged jail processes behave like privileged non-jail
522 * processes if the security.jail.chflags_allowed sysctl is
523 * is non-zero; otherwise, they behave like unprivileged
524 * processes.
525 */
526 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
527 if (ip->i_flags
528 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
529 error = securelevel_gt(cred, 0);
530 if (error)
531 return (error);
532 }
533 /* Snapshot flag cannot be set or cleared */
534 if (((vap->va_flags & SF_SNAPSHOT) != 0 &&
535 (ip->i_flags & SF_SNAPSHOT) == 0) ||
536 ((vap->va_flags & SF_SNAPSHOT) == 0 &&
537 (ip->i_flags & SF_SNAPSHOT) != 0))
538 return (EPERM);
539 ip->i_flags = vap->va_flags;
540 DIP_SET(ip, i_flags, vap->va_flags);
541 } else {
542 if (ip->i_flags
543 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
544 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
545 return (EPERM);
546 ip->i_flags &= SF_SETTABLE;
547 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
548 DIP_SET(ip, i_flags, ip->i_flags);
549 }
550 ip->i_flag |= IN_CHANGE;
551 if (vap->va_flags & (IMMUTABLE | APPEND))
552 return (0);
553 }
554 if (ip->i_flags & (IMMUTABLE | APPEND))
555 return (EPERM);
556 /*
557 * Go through the fields and update iff not VNOVAL.
558 */
559 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
560 if (vp->v_mount->mnt_flag & MNT_RDONLY)
561 return (EROFS);
562 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
563 td)) != 0)
564 return (error);
565 }
566 if (vap->va_size != VNOVAL) {
567 /*
568 * XXX most of the following special cases should be in
569 * callers instead of in N filesystems. The VDIR check
570 * mostly already is.
571 */
572 switch (vp->v_type) {
573 case VDIR:
574 return (EISDIR);
575 case VLNK:
576 case VREG:
577 /*
578 * Truncation should have an effect in these cases.
579 * Disallow it if the filesystem is read-only or
580 * the file is being snapshotted.
581 */
582 if (vp->v_mount->mnt_flag & MNT_RDONLY)
583 return (EROFS);
584 if ((ip->i_flags & SF_SNAPSHOT) != 0)
585 return (EPERM);
586 break;
587 default:
588 /*
589 * According to POSIX, the result is unspecified
590 * for file types other than regular files,
591 * directories and shared memory objects. We
592 * don't support shared memory objects in the file
593 * system, and have dubious support for truncating
594 * symlinks. Just ignore the request in other cases.
595 */
596 return (0);
597 }
598 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL,
599 cred, td)) != 0)
600 return (error);
601 }
602 if (vap->va_atime.tv_sec != VNOVAL ||
603 vap->va_mtime.tv_sec != VNOVAL ||
604 vap->va_birthtime.tv_sec != VNOVAL) {
605 if (vp->v_mount->mnt_flag & MNT_RDONLY)
606 return (EROFS);
607 if ((ip->i_flags & SF_SNAPSHOT) != 0)
608 return (EPERM);
609 /*
610 * From utimes(2):
611 * If times is NULL, ... The caller must be the owner of
612 * the file, have permission to write the file, or be the
613 * super-user.
614 * If times is non-NULL, ... The caller must be the owner of
615 * the file or be the super-user.
616 *
617 * Possibly for historical reasons, try to use VADMIN in
618 * preference to VWRITE for a NULL timestamp. This means we
619 * will return EACCES in preference to EPERM if neither
620 * check succeeds.
621 */
622 if (vap->va_vaflags & VA_UTIMES_NULL) {
623 error = VOP_ACCESS(vp, VADMIN, cred, td);
624 if (error)
625 error = VOP_ACCESS(vp, VWRITE, cred, td);
626 } else
627 error = VOP_ACCESS(vp, VADMIN, cred, td);
628 if (error)
629 return (error);
630 if (vap->va_atime.tv_sec != VNOVAL)
631 ip->i_flag |= IN_ACCESS;
632 if (vap->va_mtime.tv_sec != VNOVAL)
633 ip->i_flag |= IN_CHANGE | IN_UPDATE;
634 if (vap->va_birthtime.tv_sec != VNOVAL &&
635 ip->i_ump->um_fstype == UFS2)
636 ip->i_flag |= IN_MODIFIED;
637 ufs_itimes(vp);
638 if (vap->va_atime.tv_sec != VNOVAL) {
639 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
640 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
641 }
642 if (vap->va_mtime.tv_sec != VNOVAL) {
643 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
644 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
645 }
646 if (vap->va_birthtime.tv_sec != VNOVAL &&
647 ip->i_ump->um_fstype == UFS2) {
648 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
649 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
650 }
651 error = UFS_UPDATE(vp, 0);
652 if (error)
653 return (error);
654 }
655 error = 0;
656 if (vap->va_mode != (mode_t)VNOVAL) {
657 if (vp->v_mount->mnt_flag & MNT_RDONLY)
658 return (EROFS);
659 if ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode &
660 (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)))
661 return (EPERM);
662 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
663 }
664 return (error);
665 }
666
667 /*
668 * Change the mode on a file.
669 * Inode must be locked before calling.
670 */
671 static int
672 ufs_chmod(vp, mode, cred, td)
673 struct vnode *vp;
674 int mode;
675 struct ucred *cred;
676 struct thread *td;
677 {
678 struct inode *ip = VTOI(vp);
679 int error;
680
681 /*
682 * To modify the permissions on a file, must possess VADMIN
683 * for that file.
684 */
685 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
686 return (error);
687 /*
688 * Privileged processes may set the sticky bit on non-directories,
689 * as well as set the setgid bit on a file with a group that the
690 * process is not a member of. Both of these are allowed in
691 * jail(8).
692 */
693 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
694 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
695 return (EFTYPE);
696 }
697 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
698 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
699 if (error)
700 return (error);
701 }
702 ip->i_mode &= ~ALLPERMS;
703 ip->i_mode |= (mode & ALLPERMS);
704 DIP_SET(ip, i_mode, ip->i_mode);
705 ip->i_flag |= IN_CHANGE;
706 return (0);
707 }
708
709 /*
710 * Perform chown operation on inode ip;
711 * inode must be locked prior to call.
712 */
713 static int
714 ufs_chown(vp, uid, gid, cred, td)
715 struct vnode *vp;
716 uid_t uid;
717 gid_t gid;
718 struct ucred *cred;
719 struct thread *td;
720 {
721 struct inode *ip = VTOI(vp);
722 uid_t ouid;
723 gid_t ogid;
724 int error = 0;
725 #ifdef QUOTA
726 int i;
727 ufs2_daddr_t change;
728 #endif
729
730 if (uid == (uid_t)VNOVAL)
731 uid = ip->i_uid;
732 if (gid == (gid_t)VNOVAL)
733 gid = ip->i_gid;
734 /*
735 * To modify the ownership of a file, must possess VADMIN for that
736 * file.
737 */
738 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
739 return (error);
740 /*
741 * To change the owner of a file, or change the group of a file to a
742 * group of which we are not a member, the caller must have
743 * privilege.
744 */
745 if ((uid != ip->i_uid ||
746 (gid != ip->i_gid && !groupmember(gid, cred))) &&
747 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
748 return (error);
749 ogid = ip->i_gid;
750 ouid = ip->i_uid;
751 #ifdef QUOTA
752 if ((error = getinoquota(ip)) != 0)
753 return (error);
754 if (ouid == uid) {
755 dqrele(vp, ip->i_dquot[USRQUOTA]);
756 ip->i_dquot[USRQUOTA] = NODQUOT;
757 }
758 if (ogid == gid) {
759 dqrele(vp, ip->i_dquot[GRPQUOTA]);
760 ip->i_dquot[GRPQUOTA] = NODQUOT;
761 }
762 change = DIP(ip, i_blocks);
763 (void) chkdq(ip, -change, cred, CHOWN);
764 (void) chkiq(ip, -1, cred, CHOWN);
765 for (i = 0; i < MAXQUOTAS; i++) {
766 dqrele(vp, ip->i_dquot[i]);
767 ip->i_dquot[i] = NODQUOT;
768 }
769 #endif
770 ip->i_gid = gid;
771 DIP_SET(ip, i_gid, gid);
772 ip->i_uid = uid;
773 DIP_SET(ip, i_uid, uid);
774 #ifdef QUOTA
775 if ((error = getinoquota(ip)) == 0) {
776 if (ouid == uid) {
777 dqrele(vp, ip->i_dquot[USRQUOTA]);
778 ip->i_dquot[USRQUOTA] = NODQUOT;
779 }
780 if (ogid == gid) {
781 dqrele(vp, ip->i_dquot[GRPQUOTA]);
782 ip->i_dquot[GRPQUOTA] = NODQUOT;
783 }
784 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
785 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
786 goto good;
787 else
788 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
789 }
790 for (i = 0; i < MAXQUOTAS; i++) {
791 dqrele(vp, ip->i_dquot[i]);
792 ip->i_dquot[i] = NODQUOT;
793 }
794 }
795 ip->i_gid = ogid;
796 DIP_SET(ip, i_gid, ogid);
797 ip->i_uid = ouid;
798 DIP_SET(ip, i_uid, ouid);
799 if (getinoquota(ip) == 0) {
800 if (ouid == uid) {
801 dqrele(vp, ip->i_dquot[USRQUOTA]);
802 ip->i_dquot[USRQUOTA] = NODQUOT;
803 }
804 if (ogid == gid) {
805 dqrele(vp, ip->i_dquot[GRPQUOTA]);
806 ip->i_dquot[GRPQUOTA] = NODQUOT;
807 }
808 (void) chkdq(ip, change, cred, FORCE|CHOWN);
809 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
810 (void) getinoquota(ip);
811 }
812 return (error);
813 good:
814 if (getinoquota(ip))
815 panic("ufs_chown: lost quota");
816 #endif /* QUOTA */
817 ip->i_flag |= IN_CHANGE;
818 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
819 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) {
820 ip->i_mode &= ~(ISUID | ISGID);
821 DIP_SET(ip, i_mode, ip->i_mode);
822 }
823 }
824 return (0);
825 }
826
827 static int
828 ufs_remove(ap)
829 struct vop_remove_args /* {
830 struct vnode *a_dvp;
831 struct vnode *a_vp;
832 struct componentname *a_cnp;
833 } */ *ap;
834 {
835 struct inode *ip;
836 struct vnode *vp = ap->a_vp;
837 struct vnode *dvp = ap->a_dvp;
838 int error;
839 struct thread *td;
840
841 td = curthread;
842 ip = VTOI(vp);
843 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
844 (VTOI(dvp)->i_flags & APPEND)) {
845 error = EPERM;
846 goto out;
847 }
848 #ifdef UFS_GJOURNAL
849 ufs_gjournal_orphan(vp);
850 #endif
851 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
852 if (ip->i_nlink <= 0)
853 vp->v_vflag |= VV_NOSYNC;
854 if ((ip->i_flags & SF_SNAPSHOT) != 0) {
855 /*
856 * Avoid deadlock where another thread is trying to
857 * update the inodeblock for dvp and is waiting on
858 * snaplk. Temporary unlock the vnode lock for the
859 * unlinked file and sync the directory. This should
860 * allow vput() of the directory to not block later on
861 * while holding the snapshot vnode locked, assuming
862 * that the directory hasn't been unlinked too.
863 */
864 VOP_UNLOCK(vp, 0, td);
865 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
866 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
867 }
868 out:
869 return (error);
870 }
871
872 /*
873 * link vnode call
874 */
875 static int
876 ufs_link(ap)
877 struct vop_link_args /* {
878 struct vnode *a_tdvp;
879 struct vnode *a_vp;
880 struct componentname *a_cnp;
881 } */ *ap;
882 {
883 struct vnode *vp = ap->a_vp;
884 struct vnode *tdvp = ap->a_tdvp;
885 struct componentname *cnp = ap->a_cnp;
886 struct inode *ip;
887 struct direct newdir;
888 int error;
889
890 #ifdef INVARIANTS
891 if ((cnp->cn_flags & HASBUF) == 0)
892 panic("ufs_link: no name");
893 #endif
894 if (tdvp->v_mount != vp->v_mount) {
895 error = EXDEV;
896 goto out;
897 }
898 ip = VTOI(vp);
899 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
900 error = EMLINK;
901 goto out;
902 }
903 if (ip->i_flags & (IMMUTABLE | APPEND)) {
904 error = EPERM;
905 goto out;
906 }
907 ip->i_effnlink++;
908 ip->i_nlink++;
909 DIP_SET(ip, i_nlink, ip->i_nlink);
910 ip->i_flag |= IN_CHANGE;
911 if (DOINGSOFTDEP(vp))
912 softdep_change_linkcnt(ip);
913 error = UFS_UPDATE(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
914 if (!error) {
915 ufs_makedirentry(ip, cnp, &newdir);
916 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
917 }
918
919 if (error) {
920 ip->i_effnlink--;
921 ip->i_nlink--;
922 DIP_SET(ip, i_nlink, ip->i_nlink);
923 ip->i_flag |= IN_CHANGE;
924 if (DOINGSOFTDEP(vp))
925 softdep_change_linkcnt(ip);
926 }
927 out:
928 return (error);
929 }
930
931 /*
932 * whiteout vnode call
933 */
934 static int
935 ufs_whiteout(ap)
936 struct vop_whiteout_args /* {
937 struct vnode *a_dvp;
938 struct componentname *a_cnp;
939 int a_flags;
940 } */ *ap;
941 {
942 struct vnode *dvp = ap->a_dvp;
943 struct componentname *cnp = ap->a_cnp;
944 struct direct newdir;
945 int error = 0;
946
947 switch (ap->a_flags) {
948 case LOOKUP:
949 /* 4.4 format directories support whiteout operations */
950 if (dvp->v_mount->mnt_maxsymlinklen > 0)
951 return (0);
952 return (EOPNOTSUPP);
953
954 case CREATE:
955 /* create a new directory whiteout */
956 #ifdef INVARIANTS
957 if ((cnp->cn_flags & SAVENAME) == 0)
958 panic("ufs_whiteout: missing name");
959 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
960 panic("ufs_whiteout: old format filesystem");
961 #endif
962
963 newdir.d_ino = WINO;
964 newdir.d_namlen = cnp->cn_namelen;
965 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
966 newdir.d_type = DT_WHT;
967 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
968 break;
969
970 case DELETE:
971 /* remove an existing directory whiteout */
972 #ifdef INVARIANTS
973 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
974 panic("ufs_whiteout: old format filesystem");
975 #endif
976
977 cnp->cn_flags &= ~DOWHITEOUT;
978 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
979 break;
980 default:
981 panic("ufs_whiteout: unknown op");
982 }
983 return (error);
984 }
985
986 /*
987 * Rename system call.
988 * rename("foo", "bar");
989 * is essentially
990 * unlink("bar");
991 * link("foo", "bar");
992 * unlink("foo");
993 * but ``atomically''. Can't do full commit without saving state in the
994 * inode on disk which isn't feasible at this time. Best we can do is
995 * always guarantee the target exists.
996 *
997 * Basic algorithm is:
998 *
999 * 1) Bump link count on source while we're linking it to the
1000 * target. This also ensure the inode won't be deleted out
1001 * from underneath us while we work (it may be truncated by
1002 * a concurrent `trunc' or `open' for creation).
1003 * 2) Link source to destination. If destination already exists,
1004 * delete it first.
1005 * 3) Unlink source reference to inode if still around. If a
1006 * directory was moved and the parent of the destination
1007 * is different from the source, patch the ".." entry in the
1008 * directory.
1009 */
1010 static int
1011 ufs_rename(ap)
1012 struct vop_rename_args /* {
1013 struct vnode *a_fdvp;
1014 struct vnode *a_fvp;
1015 struct componentname *a_fcnp;
1016 struct vnode *a_tdvp;
1017 struct vnode *a_tvp;
1018 struct componentname *a_tcnp;
1019 } */ *ap;
1020 {
1021 struct vnode *tvp = ap->a_tvp;
1022 struct vnode *tdvp = ap->a_tdvp;
1023 struct vnode *fvp = ap->a_fvp;
1024 struct vnode *fdvp = ap->a_fdvp;
1025 struct componentname *tcnp = ap->a_tcnp;
1026 struct componentname *fcnp = ap->a_fcnp;
1027 struct thread *td = fcnp->cn_thread;
1028 struct inode *ip, *xp, *dp;
1029 struct direct newdir;
1030 int doingdirectory = 0, oldparent = 0, newparent = 0;
1031 int error = 0, ioflag;
1032 ino_t fvp_ino;
1033
1034 #ifdef INVARIANTS
1035 if ((tcnp->cn_flags & HASBUF) == 0 ||
1036 (fcnp->cn_flags & HASBUF) == 0)
1037 panic("ufs_rename: no name");
1038 #endif
1039 /*
1040 * Check for cross-device rename.
1041 */
1042 if ((fvp->v_mount != tdvp->v_mount) ||
1043 (tvp && (fvp->v_mount != tvp->v_mount))) {
1044 error = EXDEV;
1045 abortit:
1046 if (tdvp == tvp)
1047 vrele(tdvp);
1048 else
1049 vput(tdvp);
1050 if (tvp)
1051 vput(tvp);
1052 vrele(fdvp);
1053 vrele(fvp);
1054 return (error);
1055 }
1056
1057 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1058 (VTOI(tdvp)->i_flags & APPEND))) {
1059 error = EPERM;
1060 goto abortit;
1061 }
1062
1063 /*
1064 * Renaming a file to itself has no effect. The upper layers should
1065 * not call us in that case. Temporarily just warn if they do.
1066 */
1067 if (fvp == tvp) {
1068 printf("ufs_rename: fvp == tvp (can't happen)\n");
1069 error = 0;
1070 goto abortit;
1071 }
1072
1073 if ((error = vn_lock(fvp, LK_EXCLUSIVE, td)) != 0)
1074 goto abortit;
1075 dp = VTOI(fdvp);
1076 ip = VTOI(fvp);
1077 if (ip->i_nlink >= LINK_MAX) {
1078 VOP_UNLOCK(fvp, 0, td);
1079 error = EMLINK;
1080 goto abortit;
1081 }
1082 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1083 || (dp->i_flags & APPEND)) {
1084 VOP_UNLOCK(fvp, 0, td);
1085 error = EPERM;
1086 goto abortit;
1087 }
1088 if ((ip->i_mode & IFMT) == IFDIR) {
1089 /*
1090 * Avoid ".", "..", and aliases of "." for obvious reasons.
1091 */
1092 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1093 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
1094 (ip->i_flag & IN_RENAME)) {
1095 VOP_UNLOCK(fvp, 0, td);
1096 error = EINVAL;
1097 goto abortit;
1098 }
1099 ip->i_flag |= IN_RENAME;
1100 oldparent = dp->i_number;
1101 doingdirectory = 1;
1102 }
1103 vrele(fdvp);
1104
1105 /*
1106 * When the target exists, both the directory
1107 * and target vnodes are returned locked.
1108 */
1109 dp = VTOI(tdvp);
1110 xp = NULL;
1111 if (tvp)
1112 xp = VTOI(tvp);
1113
1114 /*
1115 * 1) Bump link count while we're moving stuff
1116 * around. If we crash somewhere before
1117 * completing our work, the link count
1118 * may be wrong, but correctable.
1119 */
1120 ip->i_effnlink++;
1121 ip->i_nlink++;
1122 DIP_SET(ip, i_nlink, ip->i_nlink);
1123 ip->i_flag |= IN_CHANGE;
1124 if (DOINGSOFTDEP(fvp))
1125 softdep_change_linkcnt(ip);
1126 if ((error = UFS_UPDATE(fvp, !(DOINGSOFTDEP(fvp) |
1127 DOINGASYNC(fvp)))) != 0) {
1128 VOP_UNLOCK(fvp, 0, td);
1129 goto bad;
1130 }
1131
1132 /*
1133 * If ".." must be changed (ie the directory gets a new
1134 * parent) then the source directory must not be in the
1135 * directory hierarchy above the target, as this would
1136 * orphan everything below the source directory. Also
1137 * the user must have write permission in the source so
1138 * as to be able to change "..". We must repeat the call
1139 * to namei, as the parent directory is unlocked by the
1140 * call to checkpath().
1141 */
1142 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1143 fvp_ino = ip->i_number;
1144 VOP_UNLOCK(fvp, 0, td);
1145 if (oldparent != dp->i_number)
1146 newparent = dp->i_number;
1147 if (doingdirectory && newparent) {
1148 if (error) /* write access check above */
1149 goto bad;
1150 if (xp != NULL)
1151 vput(tvp);
1152 error = ufs_checkpath(fvp_ino, dp, tcnp->cn_cred);
1153 if (error)
1154 goto out;
1155 if ((tcnp->cn_flags & SAVESTART) == 0)
1156 panic("ufs_rename: lost to startdir");
1157 VREF(tdvp);
1158 error = relookup(tdvp, &tvp, tcnp);
1159 if (error)
1160 goto out;
1161 vrele(tdvp);
1162 dp = VTOI(tdvp);
1163 xp = NULL;
1164 if (tvp)
1165 xp = VTOI(tvp);
1166 }
1167 /*
1168 * 2) If target doesn't exist, link the target
1169 * to the source and unlink the source.
1170 * Otherwise, rewrite the target directory
1171 * entry to reference the source inode and
1172 * expunge the original entry's existence.
1173 */
1174 if (xp == NULL) {
1175 if (dp->i_dev != ip->i_dev)
1176 panic("ufs_rename: EXDEV");
1177 /*
1178 * Account for ".." in new directory.
1179 * When source and destination have the same
1180 * parent we don't fool with the link count.
1181 */
1182 if (doingdirectory && newparent) {
1183 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1184 error = EMLINK;
1185 goto bad;
1186 }
1187 dp->i_effnlink++;
1188 dp->i_nlink++;
1189 DIP_SET(dp, i_nlink, dp->i_nlink);
1190 dp->i_flag |= IN_CHANGE;
1191 if (DOINGSOFTDEP(tdvp))
1192 softdep_change_linkcnt(dp);
1193 error = UFS_UPDATE(tdvp, !(DOINGSOFTDEP(tdvp) |
1194 DOINGASYNC(tdvp)));
1195 if (error)
1196 goto bad;
1197 }
1198 ufs_makedirentry(ip, tcnp, &newdir);
1199 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1200 if (error) {
1201 if (doingdirectory && newparent) {
1202 dp->i_effnlink--;
1203 dp->i_nlink--;
1204 DIP_SET(dp, i_nlink, dp->i_nlink);
1205 dp->i_flag |= IN_CHANGE;
1206 if (DOINGSOFTDEP(tdvp))
1207 softdep_change_linkcnt(dp);
1208 (void)UFS_UPDATE(tdvp, 1);
1209 }
1210 goto bad;
1211 }
1212 vput(tdvp);
1213 } else {
1214 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1215 panic("ufs_rename: EXDEV");
1216 /*
1217 * Short circuit rename(foo, foo).
1218 */
1219 if (xp->i_number == ip->i_number)
1220 panic("ufs_rename: same file");
1221 /*
1222 * If the parent directory is "sticky", then the caller
1223 * must possess VADMIN for the parent directory, or the
1224 * destination of the rename. This implements append-only
1225 * directories.
1226 */
1227 if ((dp->i_mode & S_ISTXT) &&
1228 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1229 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1230 error = EPERM;
1231 goto bad;
1232 }
1233 /*
1234 * Target must be empty if a directory and have no links
1235 * to it. Also, ensure source and target are compatible
1236 * (both directories, or both not directories).
1237 */
1238 if ((xp->i_mode&IFMT) == IFDIR) {
1239 if ((xp->i_effnlink > 2) ||
1240 !ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
1241 error = ENOTEMPTY;
1242 goto bad;
1243 }
1244 if (!doingdirectory) {
1245 error = ENOTDIR;
1246 goto bad;
1247 }
1248 cache_purge(tdvp);
1249 } else if (doingdirectory) {
1250 error = EISDIR;
1251 goto bad;
1252 }
1253 error = ufs_dirrewrite(dp, xp, ip->i_number,
1254 IFTODT(ip->i_mode),
1255 (doingdirectory && newparent) ? newparent : doingdirectory);
1256 if (error)
1257 goto bad;
1258 if (doingdirectory) {
1259 if (!newparent) {
1260 dp->i_effnlink--;
1261 if (DOINGSOFTDEP(tdvp))
1262 softdep_change_linkcnt(dp);
1263 }
1264 xp->i_effnlink--;
1265 if (DOINGSOFTDEP(tvp))
1266 softdep_change_linkcnt(xp);
1267 }
1268 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1269 /*
1270 * Truncate inode. The only stuff left in the directory
1271 * is "." and "..". The "." reference is inconsequential
1272 * since we are quashing it. We have removed the "."
1273 * reference and the reference in the parent directory,
1274 * but there may be other hard links. The soft
1275 * dependency code will arrange to do these operations
1276 * after the parent directory entry has been deleted on
1277 * disk, so when running with that code we avoid doing
1278 * them now.
1279 */
1280 if (!newparent) {
1281 dp->i_nlink--;
1282 DIP_SET(dp, i_nlink, dp->i_nlink);
1283 dp->i_flag |= IN_CHANGE;
1284 }
1285 xp->i_nlink--;
1286 DIP_SET(xp, i_nlink, xp->i_nlink);
1287 xp->i_flag |= IN_CHANGE;
1288 ioflag = IO_NORMAL;
1289 if (!DOINGASYNC(tvp))
1290 ioflag |= IO_SYNC;
1291 if ((error = UFS_TRUNCATE(tvp, (off_t)0, ioflag,
1292 tcnp->cn_cred, tcnp->cn_thread)) != 0)
1293 goto bad;
1294 }
1295 vput(tdvp);
1296 vput(tvp);
1297 xp = NULL;
1298 }
1299
1300 /*
1301 * 3) Unlink the source.
1302 */
1303 fcnp->cn_flags &= ~MODMASK;
1304 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1305 if ((fcnp->cn_flags & SAVESTART) == 0)
1306 panic("ufs_rename: lost from startdir");
1307 VREF(fdvp);
1308 error = relookup(fdvp, &fvp, fcnp);
1309 if (error == 0)
1310 vrele(fdvp);
1311 if (fvp != NULL) {
1312 xp = VTOI(fvp);
1313 dp = VTOI(fdvp);
1314 } else {
1315 /*
1316 * From name has disappeared. IN_RENAME is not sufficient
1317 * to protect against directory races due to timing windows,
1318 * so we have to remove the panic. XXX the only real way
1319 * to solve this issue is at a much higher level. By the
1320 * time we hit ufs_rename() it's too late.
1321 */
1322 #if 0
1323 if (doingdirectory)
1324 panic("ufs_rename: lost dir entry");
1325 #endif
1326 vrele(ap->a_fvp);
1327 return (0);
1328 }
1329 /*
1330 * Ensure that the directory entry still exists and has not
1331 * changed while the new name has been entered. If the source is
1332 * a file then the entry may have been unlinked or renamed. In
1333 * either case there is no further work to be done. If the source
1334 * is a directory then it cannot have been rmdir'ed; the IN_RENAME
1335 * flag ensures that it cannot be moved by another rename or removed
1336 * by a rmdir.
1337 */
1338 if (xp != ip) {
1339 /*
1340 * From name resolves to a different inode. IN_RENAME is
1341 * not sufficient protection against timing window races
1342 * so we can't panic here. XXX the only real way
1343 * to solve this issue is at a much higher level. By the
1344 * time we hit ufs_rename() it's too late.
1345 */
1346 #if 0
1347 if (doingdirectory)
1348 panic("ufs_rename: lost dir entry");
1349 #endif
1350 } else {
1351 /*
1352 * If the source is a directory with a
1353 * new parent, the link count of the old
1354 * parent directory must be decremented
1355 * and ".." set to point to the new parent.
1356 */
1357 if (doingdirectory && newparent) {
1358 xp->i_offset = mastertemplate.dot_reclen;
1359 ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
1360 cache_purge(fdvp);
1361 }
1362 error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
1363 xp->i_flag &= ~IN_RENAME;
1364 }
1365 if (dp)
1366 vput(fdvp);
1367 if (xp)
1368 vput(fvp);
1369 vrele(ap->a_fvp);
1370 return (error);
1371
1372 bad:
1373 if (xp)
1374 vput(ITOV(xp));
1375 vput(ITOV(dp));
1376 out:
1377 if (doingdirectory)
1378 ip->i_flag &= ~IN_RENAME;
1379 if (vn_lock(fvp, LK_EXCLUSIVE, td) == 0) {
1380 ip->i_effnlink--;
1381 ip->i_nlink--;
1382 DIP_SET(ip, i_nlink, ip->i_nlink);
1383 ip->i_flag |= IN_CHANGE;
1384 ip->i_flag &= ~IN_RENAME;
1385 if (DOINGSOFTDEP(fvp))
1386 softdep_change_linkcnt(ip);
1387 vput(fvp);
1388 } else
1389 vrele(fvp);
1390 return (error);
1391 }
1392
1393 /*
1394 * Mkdir system call
1395 */
1396 static int
1397 ufs_mkdir(ap)
1398 struct vop_mkdir_args /* {
1399 struct vnode *a_dvp;
1400 struct vnode **a_vpp;
1401 struct componentname *a_cnp;
1402 struct vattr *a_vap;
1403 } */ *ap;
1404 {
1405 struct vnode *dvp = ap->a_dvp;
1406 struct vattr *vap = ap->a_vap;
1407 struct componentname *cnp = ap->a_cnp;
1408 struct inode *ip, *dp;
1409 struct vnode *tvp;
1410 struct buf *bp;
1411 struct dirtemplate dirtemplate, *dtp;
1412 struct direct newdir;
1413 #ifdef UFS_ACL
1414 struct acl *acl, *dacl;
1415 #endif
1416 int error, dmode;
1417 long blkoff;
1418
1419 #ifdef INVARIANTS
1420 if ((cnp->cn_flags & HASBUF) == 0)
1421 panic("ufs_mkdir: no name");
1422 #endif
1423 dp = VTOI(dvp);
1424 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1425 error = EMLINK;
1426 goto out;
1427 }
1428 dmode = vap->va_mode & 0777;
1429 dmode |= IFDIR;
1430 /*
1431 * Must simulate part of ufs_makeinode here to acquire the inode,
1432 * but not have it entered in the parent directory. The entry is
1433 * made later after writing "." and ".." entries.
1434 */
1435 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
1436 if (error)
1437 goto out;
1438 ip = VTOI(tvp);
1439 ip->i_gid = dp->i_gid;
1440 DIP_SET(ip, i_gid, dp->i_gid);
1441 #ifdef SUIDDIR
1442 {
1443 #ifdef QUOTA
1444 struct ucred ucred, *ucp;
1445 ucp = cnp->cn_cred;
1446 #endif
1447 /*
1448 * If we are hacking owners here, (only do this where told to)
1449 * and we are not giving it TO root, (would subvert quotas)
1450 * then go ahead and give it to the other user.
1451 * The new directory also inherits the SUID bit.
1452 * If user's UID and dir UID are the same,
1453 * 'give it away' so that the SUID is still forced on.
1454 */
1455 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1456 (dp->i_mode & ISUID) && dp->i_uid) {
1457 dmode |= ISUID;
1458 ip->i_uid = dp->i_uid;
1459 DIP_SET(ip, i_uid, dp->i_uid);
1460 #ifdef QUOTA
1461 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1462 /*
1463 * Make sure the correct user gets charged
1464 * for the space.
1465 * Make a dummy credential for the victim.
1466 * XXX This seems to never be accessed out of
1467 * our context so a stack variable is ok.
1468 */
1469 refcount_init(&ucred.cr_ref, 1);
1470 ucred.cr_uid = ip->i_uid;
1471 ucred.cr_ngroups = 1;
1472 ucred.cr_groups[0] = dp->i_gid;
1473 ucp = &ucred;
1474 }
1475 #endif
1476 } else {
1477 ip->i_uid = cnp->cn_cred->cr_uid;
1478 DIP_SET(ip, i_uid, ip->i_uid);
1479 }
1480 #ifdef QUOTA
1481 if ((error = getinoquota(ip)) ||
1482 (error = chkiq(ip, 1, ucp, 0))) {
1483 UFS_VFREE(tvp, ip->i_number, dmode);
1484 vput(tvp);
1485 return (error);
1486 }
1487 #endif
1488 }
1489 #else /* !SUIDDIR */
1490 ip->i_uid = cnp->cn_cred->cr_uid;
1491 DIP_SET(ip, i_uid, ip->i_uid);
1492 #ifdef QUOTA
1493 if ((error = getinoquota(ip)) ||
1494 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
1495 UFS_VFREE(tvp, ip->i_number, dmode);
1496 vput(tvp);
1497 return (error);
1498 }
1499 #endif
1500 #endif /* !SUIDDIR */
1501 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1502 #ifdef UFS_ACL
1503 acl = dacl = NULL;
1504 if ((dvp->v_mount->mnt_flag & MNT_ACLS) != 0) {
1505 acl = uma_zalloc(acl_zone, M_WAITOK);
1506 dacl = uma_zalloc(acl_zone, M_WAITOK);
1507
1508 /*
1509 * Retrieve default ACL from parent, if any.
1510 */
1511 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cnp->cn_cred,
1512 cnp->cn_thread);
1513 switch (error) {
1514 case 0:
1515 /*
1516 * Retrieved a default ACL, so merge mode and ACL if
1517 * necessary. If the ACL is empty, fall through to
1518 * the "not defined or available" case.
1519 */
1520 if (acl->acl_cnt != 0) {
1521 dmode = acl_posix1e_newfilemode(dmode, acl);
1522 ip->i_mode = dmode;
1523 DIP_SET(ip, i_mode, dmode);
1524 *dacl = *acl;
1525 ufs_sync_acl_from_inode(ip, acl);
1526 break;
1527 }
1528 /* FALLTHROUGH */
1529
1530 case EOPNOTSUPP:
1531 /*
1532 * Just use the mode as-is.
1533 */
1534 ip->i_mode = dmode;
1535 DIP_SET(ip, i_mode, dmode);
1536 uma_zfree(acl_zone, acl);
1537 uma_zfree(acl_zone, dacl);
1538 dacl = acl = NULL;
1539 break;
1540
1541 default:
1542 UFS_VFREE(tvp, ip->i_number, dmode);
1543 vput(tvp);
1544 uma_zfree(acl_zone, acl);
1545 uma_zfree(acl_zone, dacl);
1546 return (error);
1547 }
1548 } else {
1549 #endif /* !UFS_ACL */
1550 ip->i_mode = dmode;
1551 DIP_SET(ip, i_mode, dmode);
1552 #ifdef UFS_ACL
1553 }
1554 #endif
1555 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1556 ip->i_effnlink = 2;
1557 ip->i_nlink = 2;
1558 DIP_SET(ip, i_nlink, 2);
1559 if (DOINGSOFTDEP(tvp))
1560 softdep_change_linkcnt(ip);
1561 if (cnp->cn_flags & ISWHITEOUT) {
1562 ip->i_flags |= UF_OPAQUE;
1563 DIP_SET(ip, i_flags, ip->i_flags);
1564 }
1565
1566 /*
1567 * Bump link count in parent directory to reflect work done below.
1568 * Should be done before reference is created so cleanup is
1569 * possible if we crash.
1570 */
1571 dp->i_effnlink++;
1572 dp->i_nlink++;
1573 DIP_SET(dp, i_nlink, dp->i_nlink);
1574 dp->i_flag |= IN_CHANGE;
1575 if (DOINGSOFTDEP(dvp))
1576 softdep_change_linkcnt(dp);
1577 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1578 if (error)
1579 goto bad;
1580 #ifdef MAC
1581 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
1582 error = mac_create_vnode_extattr(cnp->cn_cred, dvp->v_mount,
1583 dvp, tvp, cnp);
1584 if (error)
1585 goto bad;
1586 }
1587 #endif
1588 #ifdef UFS_ACL
1589 if (acl != NULL) {
1590 /*
1591 * XXX: If we abort now, will Soft Updates notify the extattr
1592 * code that the EAs for the file need to be released?
1593 */
1594 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cnp->cn_cred,
1595 cnp->cn_thread);
1596 if (error == 0)
1597 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl,
1598 cnp->cn_cred, cnp->cn_thread);
1599 switch (error) {
1600 case 0:
1601 break;
1602
1603 case EOPNOTSUPP:
1604 /*
1605 * XXX: This should not happen, as EOPNOTSUPP above
1606 * was supposed to free acl.
1607 */
1608 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1609 /*
1610 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1611 */
1612 break;
1613
1614 default:
1615 uma_zfree(acl_zone, acl);
1616 uma_zfree(acl_zone, dacl);
1617 dacl = acl = NULL;
1618 goto bad;
1619 }
1620 uma_zfree(acl_zone, acl);
1621 uma_zfree(acl_zone, dacl);
1622 dacl = acl = NULL;
1623 }
1624 #endif /* !UFS_ACL */
1625
1626 /*
1627 * Initialize directory with "." and ".." from static template.
1628 */
1629 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1630 dtp = &mastertemplate;
1631 else
1632 dtp = (struct dirtemplate *)&omastertemplate;
1633 dirtemplate = *dtp;
1634 dirtemplate.dot_ino = ip->i_number;
1635 dirtemplate.dotdot_ino = dp->i_number;
1636 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
1637 BA_CLRBUF, &bp)) != 0)
1638 goto bad;
1639 ip->i_size = DIRBLKSIZ;
1640 DIP_SET(ip, i_size, DIRBLKSIZ);
1641 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1642 vnode_pager_setsize(tvp, (u_long)ip->i_size);
1643 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1644 if (DOINGSOFTDEP(tvp)) {
1645 /*
1646 * Ensure that the entire newly allocated block is a
1647 * valid directory so that future growth within the
1648 * block does not have to ensure that the block is
1649 * written before the inode.
1650 */
1651 blkoff = DIRBLKSIZ;
1652 while (blkoff < bp->b_bcount) {
1653 ((struct direct *)
1654 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1655 blkoff += DIRBLKSIZ;
1656 }
1657 }
1658 if ((error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) |
1659 DOINGASYNC(tvp)))) != 0) {
1660 (void)bwrite(bp);
1661 goto bad;
1662 }
1663 /*
1664 * Directory set up, now install its entry in the parent directory.
1665 *
1666 * If we are not doing soft dependencies, then we must write out the
1667 * buffer containing the new directory body before entering the new
1668 * name in the parent. If we are doing soft dependencies, then the
1669 * buffer containing the new directory body will be passed to and
1670 * released in the soft dependency code after the code has attached
1671 * an appropriate ordering dependency to the buffer which ensures that
1672 * the buffer is written before the new name is written in the parent.
1673 */
1674 if (DOINGASYNC(dvp))
1675 bdwrite(bp);
1676 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
1677 goto bad;
1678 ufs_makedirentry(ip, cnp, &newdir);
1679 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
1680
1681 bad:
1682 if (error == 0) {
1683 *ap->a_vpp = tvp;
1684 } else {
1685 #ifdef UFS_ACL
1686 if (acl != NULL)
1687 uma_zfree(acl_zone, acl);
1688 if (dacl != NULL)
1689 uma_zfree(acl_zone, dacl);
1690 #endif
1691 dp->i_effnlink--;
1692 dp->i_nlink--;
1693 DIP_SET(dp, i_nlink, dp->i_nlink);
1694 dp->i_flag |= IN_CHANGE;
1695 if (DOINGSOFTDEP(dvp))
1696 softdep_change_linkcnt(dp);
1697 /*
1698 * No need to do an explicit VOP_TRUNCATE here, vrele will
1699 * do this for us because we set the link count to 0.
1700 */
1701 ip->i_effnlink = 0;
1702 ip->i_nlink = 0;
1703 DIP_SET(ip, i_nlink, 0);
1704 ip->i_flag |= IN_CHANGE;
1705 if (DOINGSOFTDEP(tvp))
1706 softdep_change_linkcnt(ip);
1707 vput(tvp);
1708 }
1709 out:
1710 return (error);
1711 }
1712
1713 /*
1714 * Rmdir system call.
1715 */
1716 static int
1717 ufs_rmdir(ap)
1718 struct vop_rmdir_args /* {
1719 struct vnode *a_dvp;
1720 struct vnode *a_vp;
1721 struct componentname *a_cnp;
1722 } */ *ap;
1723 {
1724 struct vnode *vp = ap->a_vp;
1725 struct vnode *dvp = ap->a_dvp;
1726 struct componentname *cnp = ap->a_cnp;
1727 struct inode *ip, *dp;
1728 int error, ioflag;
1729
1730 ip = VTOI(vp);
1731 dp = VTOI(dvp);
1732
1733 /*
1734 * Do not remove a directory that is in the process of being renamed.
1735 * Verify the directory is empty (and valid). Rmdir ".." will not be
1736 * valid since ".." will contain a reference to the current directory
1737 * and thus be non-empty. Do not allow the removal of mounted on
1738 * directories (this can happen when an NFS exported filesystem
1739 * tries to remove a locally mounted on directory).
1740 */
1741 error = 0;
1742 if ((ip->i_flag & IN_RENAME) || ip->i_effnlink < 2) {
1743 error = EINVAL;
1744 goto out;
1745 }
1746 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1747 error = ENOTEMPTY;
1748 goto out;
1749 }
1750 if ((dp->i_flags & APPEND)
1751 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1752 error = EPERM;
1753 goto out;
1754 }
1755 if (vp->v_mountedhere != 0) {
1756 error = EINVAL;
1757 goto out;
1758 }
1759 #ifdef UFS_GJOURNAL
1760 ufs_gjournal_orphan(vp);
1761 #endif
1762 /*
1763 * Delete reference to directory before purging
1764 * inode. If we crash in between, the directory
1765 * will be reattached to lost+found,
1766 */
1767 dp->i_effnlink--;
1768 ip->i_effnlink--;
1769 if (DOINGSOFTDEP(vp)) {
1770 softdep_change_linkcnt(dp);
1771 softdep_change_linkcnt(ip);
1772 }
1773 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
1774 if (error) {
1775 dp->i_effnlink++;
1776 ip->i_effnlink++;
1777 if (DOINGSOFTDEP(vp)) {
1778 softdep_change_linkcnt(dp);
1779 softdep_change_linkcnt(ip);
1780 }
1781 goto out;
1782 }
1783 cache_purge(dvp);
1784 /*
1785 * Truncate inode. The only stuff left in the directory is "." and
1786 * "..". The "." reference is inconsequential since we are quashing
1787 * it. The soft dependency code will arrange to do these operations
1788 * after the parent directory entry has been deleted on disk, so
1789 * when running with that code we avoid doing them now.
1790 */
1791 if (!DOINGSOFTDEP(vp)) {
1792 dp->i_nlink--;
1793 DIP_SET(dp, i_nlink, dp->i_nlink);
1794 dp->i_flag |= IN_CHANGE;
1795 ip->i_nlink--;
1796 DIP_SET(ip, i_nlink, ip->i_nlink);
1797 ip->i_flag |= IN_CHANGE;
1798 ioflag = IO_NORMAL;
1799 if (!DOINGASYNC(vp))
1800 ioflag |= IO_SYNC;
1801 error = UFS_TRUNCATE(vp, (off_t)0, ioflag, cnp->cn_cred,
1802 cnp->cn_thread);
1803 }
1804 cache_purge(vp);
1805 #ifdef UFS_DIRHASH
1806 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
1807 if (ip->i_dirhash != NULL)
1808 ufsdirhash_free(ip);
1809 #endif
1810 out:
1811 return (error);
1812 }
1813
1814 /*
1815 * symlink -- make a symbolic link
1816 */
1817 static int
1818 ufs_symlink(ap)
1819 struct vop_symlink_args /* {
1820 struct vnode *a_dvp;
1821 struct vnode **a_vpp;
1822 struct componentname *a_cnp;
1823 struct vattr *a_vap;
1824 char *a_target;
1825 } */ *ap;
1826 {
1827 struct vnode *vp, **vpp = ap->a_vpp;
1828 struct inode *ip;
1829 int len, error;
1830
1831 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1832 vpp, ap->a_cnp);
1833 if (error)
1834 return (error);
1835 vp = *vpp;
1836 len = strlen(ap->a_target);
1837 if (len < vp->v_mount->mnt_maxsymlinklen) {
1838 ip = VTOI(vp);
1839 bcopy(ap->a_target, SHORTLINK(ip), len);
1840 ip->i_size = len;
1841 DIP_SET(ip, i_size, len);
1842 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1843 } else
1844 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1845 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1846 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1847 if (error)
1848 vput(vp);
1849 return (error);
1850 }
1851
1852 /*
1853 * Vnode op for reading directories.
1854 *
1855 * The routine below assumes that the on-disk format of a directory
1856 * is the same as that defined by <sys/dirent.h>. If the on-disk
1857 * format changes, then it will be necessary to do a conversion
1858 * from the on-disk format that read returns to the format defined
1859 * by <sys/dirent.h>.
1860 */
1861 int
1862 ufs_readdir(ap)
1863 struct vop_readdir_args /* {
1864 struct vnode *a_vp;
1865 struct uio *a_uio;
1866 struct ucred *a_cred;
1867 int *a_eofflag;
1868 int *a_ncookies;
1869 u_long **a_cookies;
1870 } */ *ap;
1871 {
1872 struct uio *uio = ap->a_uio;
1873 int error;
1874 size_t count, lost;
1875 off_t off;
1876
1877 if (ap->a_ncookies != NULL)
1878 /*
1879 * Ensure that the block is aligned. The caller can use
1880 * the cookies to determine where in the block to start.
1881 */
1882 uio->uio_offset &= ~(DIRBLKSIZ - 1);
1883 off = uio->uio_offset;
1884 count = uio->uio_resid;
1885 /* Make sure we don't return partial entries. */
1886 if (count <= ((uio->uio_offset + count) & (DIRBLKSIZ -1)))
1887 return (EINVAL);
1888 count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
1889 lost = uio->uio_resid - count;
1890 uio->uio_resid = count;
1891 uio->uio_iov->iov_len = count;
1892 # if (BYTE_ORDER == LITTLE_ENDIAN)
1893 if (ap->a_vp->v_mount->mnt_maxsymlinklen > 0) {
1894 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1895 } else {
1896 struct dirent *dp, *edp;
1897 struct uio auio;
1898 struct iovec aiov;
1899 caddr_t dirbuf;
1900 int readcnt;
1901 u_char tmp;
1902
1903 auio = *uio;
1904 auio.uio_iov = &aiov;
1905 auio.uio_iovcnt = 1;
1906 auio.uio_segflg = UIO_SYSSPACE;
1907 aiov.iov_len = count;
1908 MALLOC(dirbuf, caddr_t, count, M_TEMP, M_WAITOK);
1909 aiov.iov_base = dirbuf;
1910 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
1911 if (error == 0) {
1912 readcnt = count - auio.uio_resid;
1913 edp = (struct dirent *)&dirbuf[readcnt];
1914 for (dp = (struct dirent *)dirbuf; dp < edp; ) {
1915 tmp = dp->d_namlen;
1916 dp->d_namlen = dp->d_type;
1917 dp->d_type = tmp;
1918 if (dp->d_reclen > 0) {
1919 dp = (struct dirent *)
1920 ((char *)dp + dp->d_reclen);
1921 } else {
1922 error = EIO;
1923 break;
1924 }
1925 }
1926 if (dp >= edp)
1927 error = uiomove(dirbuf, readcnt, uio);
1928 }
1929 FREE(dirbuf, M_TEMP);
1930 }
1931 # else
1932 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
1933 # endif
1934 if (!error && ap->a_ncookies != NULL) {
1935 struct dirent* dpStart;
1936 struct dirent* dpEnd;
1937 struct dirent* dp;
1938 int ncookies;
1939 u_long *cookies;
1940 u_long *cookiep;
1941
1942 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1943 panic("ufs_readdir: unexpected uio from NFS server");
1944 dpStart = (struct dirent *)
1945 ((char *)uio->uio_iov->iov_base - (uio->uio_offset - off));
1946 dpEnd = (struct dirent *) uio->uio_iov->iov_base;
1947 for (dp = dpStart, ncookies = 0;
1948 dp < dpEnd;
1949 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen))
1950 ncookies++;
1951 MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
1952 M_WAITOK);
1953 for (dp = dpStart, cookiep = cookies;
1954 dp < dpEnd;
1955 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen)) {
1956 off += dp->d_reclen;
1957 *cookiep++ = (u_long) off;
1958 }
1959 *ap->a_ncookies = ncookies;
1960 *ap->a_cookies = cookies;
1961 }
1962 uio->uio_resid += lost;
1963 if (ap->a_eofflag)
1964 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
1965 return (error);
1966 }
1967
1968 /*
1969 * Return target name of a symbolic link
1970 */
1971 static int
1972 ufs_readlink(ap)
1973 struct vop_readlink_args /* {
1974 struct vnode *a_vp;
1975 struct uio *a_uio;
1976 struct ucred *a_cred;
1977 } */ *ap;
1978 {
1979 struct vnode *vp = ap->a_vp;
1980 struct inode *ip = VTOI(vp);
1981 doff_t isize;
1982
1983 isize = ip->i_size;
1984 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
1985 DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */
1986 return (uiomove(SHORTLINK(ip), isize, ap->a_uio));
1987 }
1988 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1989 }
1990
1991 /*
1992 * Calculate the logical to physical mapping if not done already,
1993 * then call the device strategy routine.
1994 *
1995 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
1996 * deadlock on memory. See ufs_bmap() for details.
1997 */
1998 static int
1999 ufs_strategy(ap)
2000 struct vop_strategy_args /* {
2001 struct vnode *a_vp;
2002 struct buf *a_bp;
2003 } */ *ap;
2004 {
2005 struct buf *bp = ap->a_bp;
2006 struct vnode *vp = ap->a_vp;
2007 struct bufobj *bo;
2008 struct inode *ip;
2009 ufs2_daddr_t blkno;
2010 int error;
2011
2012 ip = VTOI(vp);
2013 if (bp->b_blkno == bp->b_lblkno) {
2014 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2015 bp->b_blkno = blkno;
2016 if (error) {
2017 bp->b_error = error;
2018 bp->b_ioflags |= BIO_ERROR;
2019 bufdone(bp);
2020 return (0);
2021 }
2022 if ((long)bp->b_blkno == -1)
2023 vfs_bio_clrbuf(bp);
2024 }
2025 if ((long)bp->b_blkno == -1) {
2026 bufdone(bp);
2027 return (0);
2028 }
2029 bp->b_iooffset = dbtob(bp->b_blkno);
2030 bo = ip->i_umbufobj;
2031 BO_STRATEGY(bo, bp);
2032 return (0);
2033 }
2034
2035 /*
2036 * Print out the contents of an inode.
2037 */
2038 static int
2039 ufs_print(ap)
2040 struct vop_print_args /* {
2041 struct vnode *a_vp;
2042 } */ *ap;
2043 {
2044 struct vnode *vp = ap->a_vp;
2045 struct inode *ip = VTOI(vp);
2046
2047 printf("\tino %lu, on dev %s", (u_long)ip->i_number,
2048 devtoname(ip->i_dev));
2049 if (vp->v_type == VFIFO)
2050 fifo_printinfo(vp);
2051 printf("\n");
2052 return (0);
2053 }
2054
2055 /*
2056 * Close wrapper for fifos.
2057 *
2058 * Update the times on the inode then do device close.
2059 */
2060 static int
2061 ufsfifo_close(ap)
2062 struct vop_close_args /* {
2063 struct vnode *a_vp;
2064 int a_fflag;
2065 struct ucred *a_cred;
2066 struct thread *a_td;
2067 } */ *ap;
2068 {
2069 struct vnode *vp = ap->a_vp;
2070 int usecount;
2071
2072 VI_LOCK(vp);
2073 usecount = vp->v_usecount;
2074 if (usecount > 1)
2075 ufs_itimes_locked(vp);
2076 VI_UNLOCK(vp);
2077 return (fifo_specops.vop_close(ap));
2078 }
2079
2080 /*
2081 * Kqfilter wrapper for fifos.
2082 *
2083 * Fall through to ufs kqfilter routines if needed
2084 */
2085 static int
2086 ufsfifo_kqfilter(ap)
2087 struct vop_kqfilter_args *ap;
2088 {
2089 int error;
2090
2091 error = fifo_specops.vop_kqfilter(ap);
2092 if (error)
2093 error = vfs_kqfilter(ap);
2094 return (error);
2095 }
2096
2097 /*
2098 * Return POSIX pathconf information applicable to ufs filesystems.
2099 */
2100 static int
2101 ufs_pathconf(ap)
2102 struct vop_pathconf_args /* {
2103 struct vnode *a_vp;
2104 int a_name;
2105 int *a_retval;
2106 } */ *ap;
2107 {
2108 int error;
2109
2110 error = 0;
2111 switch (ap->a_name) {
2112 case _PC_LINK_MAX:
2113 *ap->a_retval = LINK_MAX;
2114 break;
2115 case _PC_NAME_MAX:
2116 *ap->a_retval = NAME_MAX;
2117 break;
2118 case _PC_PATH_MAX:
2119 *ap->a_retval = PATH_MAX;
2120 break;
2121 case _PC_PIPE_BUF:
2122 *ap->a_retval = PIPE_BUF;
2123 break;
2124 case _PC_CHOWN_RESTRICTED:
2125 *ap->a_retval = 1;
2126 break;
2127 case _PC_NO_TRUNC:
2128 *ap->a_retval = 1;
2129 break;
2130 case _PC_ACL_EXTENDED:
2131 #ifdef UFS_ACL
2132 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2133 *ap->a_retval = 1;
2134 else
2135 *ap->a_retval = 0;
2136 #else
2137 *ap->a_retval = 0;
2138 #endif
2139 break;
2140 case _PC_ACL_PATH_MAX:
2141 #ifdef UFS_ACL
2142 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2143 *ap->a_retval = ACL_MAX_ENTRIES;
2144 else
2145 *ap->a_retval = 3;
2146 #else
2147 *ap->a_retval = 3;
2148 #endif
2149 break;
2150 case _PC_MAC_PRESENT:
2151 #ifdef MAC
2152 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2153 *ap->a_retval = 1;
2154 else
2155 *ap->a_retval = 0;
2156 #else
2157 *ap->a_retval = 0;
2158 #endif
2159 break;
2160 case _PC_ASYNC_IO:
2161 /* _PC_ASYNC_IO should have been handled by upper layers. */
2162 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
2163 error = EINVAL;
2164 break;
2165 case _PC_PRIO_IO:
2166 *ap->a_retval = 0;
2167 break;
2168 case _PC_SYNC_IO:
2169 *ap->a_retval = 0;
2170 break;
2171 case _PC_ALLOC_SIZE_MIN:
2172 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2173 break;
2174 case _PC_FILESIZEBITS:
2175 *ap->a_retval = 64;
2176 break;
2177 case _PC_REC_INCR_XFER_SIZE:
2178 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2179 break;
2180 case _PC_REC_MAX_XFER_SIZE:
2181 *ap->a_retval = -1; /* means ``unlimited'' */
2182 break;
2183 case _PC_REC_MIN_XFER_SIZE:
2184 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2185 break;
2186 case _PC_REC_XFER_ALIGN:
2187 *ap->a_retval = PAGE_SIZE;
2188 break;
2189 case _PC_SYMLINK_MAX:
2190 *ap->a_retval = MAXPATHLEN;
2191 break;
2192
2193 default:
2194 error = EINVAL;
2195 break;
2196 }
2197 return (error);
2198 }
2199
2200 /*
2201 * Initialize the vnode associated with a new inode, handle aliased
2202 * vnodes.
2203 */
2204 int
2205 ufs_vinit(mntp, fifoops, vpp)
2206 struct mount *mntp;
2207 struct vop_vector *fifoops;
2208 struct vnode **vpp;
2209 {
2210 struct inode *ip;
2211 struct vnode *vp;
2212
2213 vp = *vpp;
2214 ip = VTOI(vp);
2215 vp->v_type = IFTOVT(ip->i_mode);
2216 if (vp->v_type == VFIFO)
2217 vp->v_op = fifoops;
2218 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2219 if (ip->i_number == ROOTINO)
2220 vp->v_vflag |= VV_ROOT;
2221 ip->i_modrev = init_va_filerev();
2222 *vpp = vp;
2223 return (0);
2224 }
2225
2226 /*
2227 * Allocate a new inode.
2228 * Vnode dvp must be locked.
2229 */
2230 static int
2231 ufs_makeinode(mode, dvp, vpp, cnp)
2232 int mode;
2233 struct vnode *dvp;
2234 struct vnode **vpp;
2235 struct componentname *cnp;
2236 {
2237 struct inode *ip, *pdir;
2238 struct direct newdir;
2239 struct vnode *tvp;
2240 #ifdef UFS_ACL
2241 struct acl *acl;
2242 #endif
2243 int error;
2244
2245 pdir = VTOI(dvp);
2246 #ifdef INVARIANTS
2247 if ((cnp->cn_flags & HASBUF) == 0)
2248 panic("ufs_makeinode: no name");
2249 #endif
2250 *vpp = NULL;
2251 if ((mode & IFMT) == 0)
2252 mode |= IFREG;
2253
2254 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2255 if (error)
2256 return (error);
2257 ip = VTOI(tvp);
2258 ip->i_gid = pdir->i_gid;
2259 DIP_SET(ip, i_gid, pdir->i_gid);
2260 #ifdef SUIDDIR
2261 {
2262 #ifdef QUOTA
2263 struct ucred ucred, *ucp;
2264 ucp = cnp->cn_cred;
2265 #endif
2266 /*
2267 * If we are not the owner of the directory,
2268 * and we are hacking owners here, (only do this where told to)
2269 * and we are not giving it TO root, (would subvert quotas)
2270 * then go ahead and give it to the other user.
2271 * Note that this drops off the execute bits for security.
2272 */
2273 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2274 (pdir->i_mode & ISUID) &&
2275 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2276 ip->i_uid = pdir->i_uid;
2277 DIP_SET(ip, i_uid, ip->i_uid);
2278 mode &= ~07111;
2279 #ifdef QUOTA
2280 /*
2281 * Make sure the correct user gets charged
2282 * for the space.
2283 * Quickly knock up a dummy credential for the victim.
2284 * XXX This seems to never be accessed out of our
2285 * context so a stack variable is ok.
2286 */
2287 refcount_init(&ucred.cr_ref, 1);
2288 ucred.cr_uid = ip->i_uid;
2289 ucred.cr_ngroups = 1;
2290 ucred.cr_groups[0] = pdir->i_gid;
2291 ucp = &ucred;
2292 #endif
2293 } else {
2294 ip->i_uid = cnp->cn_cred->cr_uid;
2295 DIP_SET(ip, i_uid, ip->i_uid);
2296 }
2297
2298 #ifdef QUOTA
2299 if ((error = getinoquota(ip)) ||
2300 (error = chkiq(ip, 1, ucp, 0))) {
2301 UFS_VFREE(tvp, ip->i_number, mode);
2302 vput(tvp);
2303 return (error);
2304 }
2305 #endif
2306 }
2307 #else /* !SUIDDIR */
2308 ip->i_uid = cnp->cn_cred->cr_uid;
2309 DIP_SET(ip, i_uid, ip->i_uid);
2310 #ifdef QUOTA
2311 if ((error = getinoquota(ip)) ||
2312 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2313 UFS_VFREE(tvp, ip->i_number, mode);
2314 vput(tvp);
2315 return (error);
2316 }
2317 #endif
2318 #endif /* !SUIDDIR */
2319 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2320 #ifdef UFS_ACL
2321 acl = NULL;
2322 if ((dvp->v_mount->mnt_flag & MNT_ACLS) != 0) {
2323 acl = uma_zalloc(acl_zone, M_WAITOK);
2324
2325 /*
2326 * Retrieve default ACL for parent, if any.
2327 */
2328 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cnp->cn_cred,
2329 cnp->cn_thread);
2330 switch (error) {
2331 case 0:
2332 /*
2333 * Retrieved a default ACL, so merge mode and ACL if
2334 * necessary.
2335 */
2336 if (acl->acl_cnt != 0) {
2337 /*
2338 * Two possible ways for default ACL to not
2339 * be present. First, the EA can be
2340 * undefined, or second, the default ACL can
2341 * be blank. If it's blank, fall through to
2342 * the it's not defined case.
2343 */
2344 mode = acl_posix1e_newfilemode(mode, acl);
2345 ip->i_mode = mode;
2346 DIP_SET(ip, i_mode, mode);
2347 ufs_sync_acl_from_inode(ip, acl);
2348 break;
2349 }
2350 /* FALLTHROUGH */
2351
2352 case EOPNOTSUPP:
2353 /*
2354 * Just use the mode as-is.
2355 */
2356 ip->i_mode = mode;
2357 DIP_SET(ip, i_mode, mode);
2358 uma_zfree(acl_zone, acl);
2359 acl = NULL;
2360 break;
2361
2362 default:
2363 UFS_VFREE(tvp, ip->i_number, mode);
2364 vput(tvp);
2365 uma_zfree(acl_zone, acl);
2366 acl = NULL;
2367 return (error);
2368 }
2369 } else {
2370 #endif
2371 ip->i_mode = mode;
2372 DIP_SET(ip, i_mode, mode);
2373 #ifdef UFS_ACL
2374 }
2375 #endif
2376 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2377 ip->i_effnlink = 1;
2378 ip->i_nlink = 1;
2379 DIP_SET(ip, i_nlink, 1);
2380 if (DOINGSOFTDEP(tvp))
2381 softdep_change_linkcnt(ip);
2382 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2383 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2384 ip->i_mode &= ~ISGID;
2385 DIP_SET(ip, i_mode, ip->i_mode);
2386 }
2387
2388 if (cnp->cn_flags & ISWHITEOUT) {
2389 ip->i_flags |= UF_OPAQUE;
2390 DIP_SET(ip, i_flags, ip->i_flags);
2391 }
2392
2393 /*
2394 * Make sure inode goes to disk before directory entry.
2395 */
2396 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2397 if (error)
2398 goto bad;
2399 #ifdef MAC
2400 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2401 error = mac_create_vnode_extattr(cnp->cn_cred, dvp->v_mount,
2402 dvp, tvp, cnp);
2403 if (error)
2404 goto bad;
2405 }
2406 #endif
2407 #ifdef UFS_ACL
2408 if (acl != NULL) {
2409 /*
2410 * XXX: If we abort now, will Soft Updates notify the extattr
2411 * code that the EAs for the file need to be released?
2412 */
2413 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cnp->cn_cred,
2414 cnp->cn_thread);
2415 switch (error) {
2416 case 0:
2417 break;
2418
2419 case EOPNOTSUPP:
2420 /*
2421 * XXX: This should not happen, as EOPNOTSUPP above was
2422 * supposed to free acl.
2423 */
2424 printf("ufs_makeinode: VOP_GETACL() but no "
2425 "VOP_SETACL()\n");
2426 /* panic("ufs_makeinode: VOP_GETACL() but no "
2427 "VOP_SETACL()"); */
2428 break;
2429
2430 default:
2431 uma_zfree(acl_zone, acl);
2432 goto bad;
2433 }
2434 uma_zfree(acl_zone, acl);
2435 }
2436 #endif /* !UFS_ACL */
2437 ufs_makedirentry(ip, cnp, &newdir);
2438 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2439 if (error)
2440 goto bad;
2441 *vpp = tvp;
2442 return (0);
2443
2444 bad:
2445 /*
2446 * Write error occurred trying to update the inode
2447 * or the directory so must deallocate the inode.
2448 */
2449 ip->i_effnlink = 0;
2450 ip->i_nlink = 0;
2451 DIP_SET(ip, i_nlink, 0);
2452 ip->i_flag |= IN_CHANGE;
2453 if (DOINGSOFTDEP(tvp))
2454 softdep_change_linkcnt(ip);
2455 vput(tvp);
2456 return (error);
2457 }
2458
2459 /* Global vfs data structures for ufs. */
2460 struct vop_vector ufs_vnodeops = {
2461 .vop_default = &default_vnodeops,
2462 .vop_fsync = VOP_PANIC,
2463 .vop_read = VOP_PANIC,
2464 .vop_reallocblks = VOP_PANIC,
2465 .vop_write = VOP_PANIC,
2466 .vop_access = ufs_access,
2467 .vop_bmap = ufs_bmap,
2468 .vop_cachedlookup = ufs_lookup,
2469 .vop_close = ufs_close,
2470 .vop_create = ufs_create,
2471 .vop_getattr = ufs_getattr,
2472 .vop_inactive = ufs_inactive,
2473 .vop_link = ufs_link,
2474 .vop_lookup = vfs_cache_lookup,
2475 .vop_mkdir = ufs_mkdir,
2476 .vop_mknod = ufs_mknod,
2477 .vop_open = ufs_open,
2478 .vop_pathconf = ufs_pathconf,
2479 .vop_poll = vop_stdpoll,
2480 .vop_print = ufs_print,
2481 .vop_readdir = ufs_readdir,
2482 .vop_readlink = ufs_readlink,
2483 .vop_reclaim = ufs_reclaim,
2484 .vop_remove = ufs_remove,
2485 .vop_rename = ufs_rename,
2486 .vop_rmdir = ufs_rmdir,
2487 .vop_setattr = ufs_setattr,
2488 #ifdef MAC
2489 .vop_setlabel = vop_stdsetlabel_ea,
2490 #endif
2491 .vop_strategy = ufs_strategy,
2492 .vop_symlink = ufs_symlink,
2493 .vop_whiteout = ufs_whiteout,
2494 #ifdef UFS_EXTATTR
2495 .vop_getextattr = ufs_getextattr,
2496 .vop_deleteextattr = ufs_deleteextattr,
2497 .vop_setextattr = ufs_setextattr,
2498 #endif
2499 #ifdef UFS_ACL
2500 .vop_getacl = ufs_getacl,
2501 .vop_setacl = ufs_setacl,
2502 .vop_aclcheck = ufs_aclcheck,
2503 #endif
2504 };
2505
2506 struct vop_vector ufs_fifoops = {
2507 .vop_default = &fifo_specops,
2508 .vop_fsync = VOP_PANIC,
2509 .vop_access = ufs_access,
2510 .vop_close = ufsfifo_close,
2511 .vop_getattr = ufs_getattr,
2512 .vop_inactive = ufs_inactive,
2513 .vop_kqfilter = ufsfifo_kqfilter,
2514 .vop_print = ufs_print,
2515 .vop_read = VOP_PANIC,
2516 .vop_reclaim = ufs_reclaim,
2517 .vop_setattr = ufs_setattr,
2518 #ifdef MAC
2519 .vop_setlabel = vop_stdsetlabel_ea,
2520 #endif
2521 .vop_write = VOP_PANIC,
2522 #ifdef UFS_EXTATTR
2523 .vop_getextattr = ufs_getextattr,
2524 .vop_deleteextattr = ufs_deleteextattr,
2525 .vop_setextattr = ufs_setextattr,
2526 #endif
2527 #ifdef UFS_ACL
2528 .vop_getacl = ufs_getacl,
2529 .vop_setacl = ufs_setacl,
2530 .vop_aclcheck = ufs_aclcheck,
2531 #endif
2532 };
Cache object: 864187a94a5ad1d5b348e33071e5dc4e
|