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/9.0/sys/ufs/ufs/ufs_vnops.c 227402 2011-11-09 18:51:41Z pho $");
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 <security/mac/mac_framework.h>
65
66 #include <sys/file.h> /* XXX */
67
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70
71 #include <fs/fifofs/fifo.h>
72
73 #include <ufs/ufs/acl.h>
74 #include <ufs/ufs/extattr.h>
75 #include <ufs/ufs/quota.h>
76 #include <ufs/ufs/inode.h>
77 #include <ufs/ufs/dir.h>
78 #include <ufs/ufs/ufsmount.h>
79 #include <ufs/ufs/ufs_extern.h>
80 #ifdef UFS_DIRHASH
81 #include <ufs/ufs/dirhash.h>
82 #endif
83 #ifdef UFS_GJOURNAL
84 #include <ufs/ufs/gjournal.h>
85 FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS");
86 #endif
87
88 #ifdef QUOTA
89 FEATURE(ufs_quota, "UFS disk quotas support");
90 FEATURE(ufs_quota64, "64bit UFS disk quotas support");
91 #endif
92
93 #ifdef SUIDDIR
94 FEATURE(suiddir,
95 "Give all new files in directory the same ownership as the directory");
96 #endif
97
98
99 #include <ufs/ffs/ffs_extern.h>
100
101 static vop_accessx_t ufs_accessx;
102 static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *);
103 static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *);
104 static vop_close_t ufs_close;
105 static vop_create_t ufs_create;
106 static vop_getattr_t ufs_getattr;
107 static vop_link_t ufs_link;
108 static int ufs_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
109 static vop_markatime_t ufs_markatime;
110 static vop_mkdir_t ufs_mkdir;
111 static vop_mknod_t ufs_mknod;
112 static vop_open_t ufs_open;
113 static vop_pathconf_t ufs_pathconf;
114 static vop_print_t ufs_print;
115 static vop_readlink_t ufs_readlink;
116 static vop_remove_t ufs_remove;
117 static vop_rename_t ufs_rename;
118 static vop_rmdir_t ufs_rmdir;
119 static vop_setattr_t ufs_setattr;
120 static vop_strategy_t ufs_strategy;
121 static vop_symlink_t ufs_symlink;
122 static vop_whiteout_t ufs_whiteout;
123 static vop_close_t ufsfifo_close;
124 static vop_kqfilter_t ufsfifo_kqfilter;
125 static vop_pathconf_t ufsfifo_pathconf;
126
127 SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD, 0, "UFS filesystem");
128
129 /*
130 * A virgin directory (no blushing please).
131 */
132 static struct dirtemplate mastertemplate = {
133 0, 12, DT_DIR, 1, ".",
134 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
135 };
136 static struct odirtemplate omastertemplate = {
137 0, 12, 1, ".",
138 0, DIRBLKSIZ - 12, 2, ".."
139 };
140
141 static void
142 ufs_itimes_locked(struct vnode *vp)
143 {
144 struct inode *ip;
145 struct timespec ts;
146
147 ASSERT_VI_LOCKED(vp, __func__);
148
149 ip = VTOI(vp);
150 if (UFS_RDONLY(ip))
151 goto out;
152 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
153 return;
154
155 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
156 ip->i_flag |= IN_LAZYMOD;
157 else if (((vp->v_mount->mnt_kern_flag &
158 (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) ||
159 (ip->i_flag & (IN_CHANGE | IN_UPDATE)))
160 ip->i_flag |= IN_MODIFIED;
161 else if (ip->i_flag & IN_ACCESS)
162 ip->i_flag |= IN_LAZYACCESS;
163 vfs_timestamp(&ts);
164 if (ip->i_flag & IN_ACCESS) {
165 DIP_SET(ip, i_atime, ts.tv_sec);
166 DIP_SET(ip, i_atimensec, ts.tv_nsec);
167 }
168 if (ip->i_flag & IN_UPDATE) {
169 DIP_SET(ip, i_mtime, ts.tv_sec);
170 DIP_SET(ip, i_mtimensec, ts.tv_nsec);
171 }
172 if (ip->i_flag & IN_CHANGE) {
173 DIP_SET(ip, i_ctime, ts.tv_sec);
174 DIP_SET(ip, i_ctimensec, ts.tv_nsec);
175 DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1);
176 }
177
178 out:
179 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
180 }
181
182 void
183 ufs_itimes(struct vnode *vp)
184 {
185
186 VI_LOCK(vp);
187 ufs_itimes_locked(vp);
188 VI_UNLOCK(vp);
189 }
190
191 /*
192 * Create a regular file
193 */
194 static int
195 ufs_create(ap)
196 struct vop_create_args /* {
197 struct vnode *a_dvp;
198 struct vnode **a_vpp;
199 struct componentname *a_cnp;
200 struct vattr *a_vap;
201 } */ *ap;
202 {
203 int error;
204
205 error =
206 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
207 ap->a_dvp, ap->a_vpp, ap->a_cnp);
208 if (error)
209 return (error);
210 return (0);
211 }
212
213 /*
214 * Mknod vnode call
215 */
216 /* ARGSUSED */
217 static int
218 ufs_mknod(ap)
219 struct vop_mknod_args /* {
220 struct vnode *a_dvp;
221 struct vnode **a_vpp;
222 struct componentname *a_cnp;
223 struct vattr *a_vap;
224 } */ *ap;
225 {
226 struct vattr *vap = ap->a_vap;
227 struct vnode **vpp = ap->a_vpp;
228 struct inode *ip;
229 ino_t ino;
230 int error;
231
232 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
233 ap->a_dvp, vpp, ap->a_cnp);
234 if (error)
235 return (error);
236 ip = VTOI(*vpp);
237 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
238 if (vap->va_rdev != VNOVAL) {
239 /*
240 * Want to be able to use this to make badblock
241 * inodes, so don't truncate the dev number.
242 */
243 DIP_SET(ip, i_rdev, vap->va_rdev);
244 }
245 /*
246 * Remove inode, then reload it through VFS_VGET so it is
247 * checked to see if it is an alias of an existing entry in
248 * the inode cache. XXX I don't believe this is necessary now.
249 */
250 (*vpp)->v_type = VNON;
251 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
252 vgone(*vpp);
253 vput(*vpp);
254 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
255 if (error) {
256 *vpp = NULL;
257 return (error);
258 }
259 return (0);
260 }
261
262 /*
263 * Open called.
264 */
265 /* ARGSUSED */
266 static int
267 ufs_open(struct vop_open_args *ap)
268 {
269 struct vnode *vp = ap->a_vp;
270 struct inode *ip;
271
272 if (vp->v_type == VCHR || vp->v_type == VBLK)
273 return (EOPNOTSUPP);
274
275 ip = VTOI(vp);
276 /*
277 * Files marked append-only must be opened for appending.
278 */
279 if ((ip->i_flags & APPEND) &&
280 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
281 return (EPERM);
282 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td);
283 return (0);
284 }
285
286 /*
287 * Close called.
288 *
289 * Update the times on the inode.
290 */
291 /* ARGSUSED */
292 static int
293 ufs_close(ap)
294 struct vop_close_args /* {
295 struct vnode *a_vp;
296 int a_fflag;
297 struct ucred *a_cred;
298 struct thread *a_td;
299 } */ *ap;
300 {
301 struct vnode *vp = ap->a_vp;
302 int usecount;
303
304 VI_LOCK(vp);
305 usecount = vp->v_usecount;
306 if (usecount > 1)
307 ufs_itimes_locked(vp);
308 VI_UNLOCK(vp);
309 return (0);
310 }
311
312 static int
313 ufs_accessx(ap)
314 struct vop_accessx_args /* {
315 struct vnode *a_vp;
316 accmode_t a_accmode;
317 struct ucred *a_cred;
318 struct thread *a_td;
319 } */ *ap;
320 {
321 struct vnode *vp = ap->a_vp;
322 struct inode *ip = VTOI(vp);
323 accmode_t accmode = ap->a_accmode;
324 int error;
325 #ifdef QUOTA
326 int relocked;
327 #endif
328 #ifdef UFS_ACL
329 struct acl *acl;
330 acl_type_t type;
331 #endif
332
333 /*
334 * Disallow write attempts on read-only filesystems;
335 * unless the file is a socket, fifo, or a block or
336 * character device resident on the filesystem.
337 */
338 if (accmode & VMODIFY_PERMS) {
339 switch (vp->v_type) {
340 case VDIR:
341 case VLNK:
342 case VREG:
343 if (vp->v_mount->mnt_flag & MNT_RDONLY)
344 return (EROFS);
345 #ifdef QUOTA
346 /*
347 * Inode is accounted in the quotas only if struct
348 * dquot is attached to it. VOP_ACCESS() is called
349 * from vn_open_cred() and provides a convenient
350 * point to call getinoquota().
351 */
352 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
353
354 /*
355 * Upgrade vnode lock, since getinoquota()
356 * requires exclusive lock to modify inode.
357 */
358 relocked = 1;
359 vhold(vp);
360 vn_lock(vp, LK_UPGRADE | LK_RETRY);
361 VI_LOCK(vp);
362 if (vp->v_iflag & VI_DOOMED) {
363 vdropl(vp);
364 error = ENOENT;
365 goto relock;
366 }
367 vdropl(vp);
368 } else
369 relocked = 0;
370 error = getinoquota(ip);
371 relock:
372 if (relocked)
373 vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
374 if (error != 0)
375 return (error);
376 #endif
377 break;
378 default:
379 break;
380 }
381 }
382
383 /*
384 * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS"
385 * is here, because without it, * it would be impossible for the owner
386 * to remove the IMMUTABLE flag.
387 */
388 if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) &&
389 (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
390 return (EPERM);
391
392 #ifdef UFS_ACL
393 if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) {
394 if (vp->v_mount->mnt_flag & MNT_NFS4ACLS)
395 type = ACL_TYPE_NFS4;
396 else
397 type = ACL_TYPE_ACCESS;
398
399 acl = acl_alloc(M_WAITOK);
400 if (type == ACL_TYPE_NFS4)
401 error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td);
402 else
403 error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td);
404 switch (error) {
405 case 0:
406 if (type == ACL_TYPE_NFS4) {
407 error = vaccess_acl_nfs4(vp->v_type, ip->i_uid,
408 ip->i_gid, acl, accmode, ap->a_cred, NULL);
409 } else {
410 error = vfs_unixify_accmode(&accmode);
411 if (error == 0)
412 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid,
413 ip->i_gid, acl, accmode, ap->a_cred, NULL);
414 }
415 break;
416 default:
417 if (error != EOPNOTSUPP)
418 printf(
419 "ufs_accessx(): Error retrieving ACL on object (%d).\n",
420 error);
421 /*
422 * XXX: Fall back until debugged. Should
423 * eventually possibly log an error, and return
424 * EPERM for safety.
425 */
426 error = vfs_unixify_accmode(&accmode);
427 if (error == 0)
428 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid,
429 ip->i_gid, accmode, ap->a_cred, NULL);
430 }
431 acl_free(acl);
432
433 return (error);
434 }
435 #endif /* !UFS_ACL */
436 error = vfs_unixify_accmode(&accmode);
437 if (error == 0)
438 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
439 accmode, ap->a_cred, NULL);
440 return (error);
441 }
442
443 /* ARGSUSED */
444 static int
445 ufs_getattr(ap)
446 struct vop_getattr_args /* {
447 struct vnode *a_vp;
448 struct vattr *a_vap;
449 struct ucred *a_cred;
450 } */ *ap;
451 {
452 struct vnode *vp = ap->a_vp;
453 struct inode *ip = VTOI(vp);
454 struct vattr *vap = ap->a_vap;
455
456 VI_LOCK(vp);
457 ufs_itimes_locked(vp);
458 if (ip->i_ump->um_fstype == UFS1) {
459 vap->va_atime.tv_sec = ip->i_din1->di_atime;
460 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
461 } else {
462 vap->va_atime.tv_sec = ip->i_din2->di_atime;
463 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
464 }
465 VI_UNLOCK(vp);
466 /*
467 * Copy from inode table
468 */
469 vap->va_fsid = dev2udev(ip->i_dev);
470 vap->va_fileid = ip->i_number;
471 vap->va_mode = ip->i_mode & ~IFMT;
472 vap->va_nlink = ip->i_effnlink;
473 vap->va_uid = ip->i_uid;
474 vap->va_gid = ip->i_gid;
475 if (ip->i_ump->um_fstype == UFS1) {
476 vap->va_rdev = ip->i_din1->di_rdev;
477 vap->va_size = ip->i_din1->di_size;
478 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
479 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
480 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
481 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
482 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks);
483 vap->va_filerev = ip->i_din1->di_modrev;
484 } else {
485 vap->va_rdev = ip->i_din2->di_rdev;
486 vap->va_size = ip->i_din2->di_size;
487 vap->va_mtime.tv_sec = ip->i_din2->di_mtime;
488 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec;
489 vap->va_ctime.tv_sec = ip->i_din2->di_ctime;
490 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec;
491 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime;
492 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec;
493 vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks);
494 vap->va_filerev = ip->i_din2->di_modrev;
495 }
496 vap->va_flags = ip->i_flags;
497 vap->va_gen = ip->i_gen;
498 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
499 vap->va_type = IFTOVT(ip->i_mode);
500 return (0);
501 }
502
503 /*
504 * Set attribute vnode op. called from several syscalls
505 */
506 static int
507 ufs_setattr(ap)
508 struct vop_setattr_args /* {
509 struct vnode *a_vp;
510 struct vattr *a_vap;
511 struct ucred *a_cred;
512 } */ *ap;
513 {
514 struct vattr *vap = ap->a_vap;
515 struct vnode *vp = ap->a_vp;
516 struct inode *ip = VTOI(vp);
517 struct ucred *cred = ap->a_cred;
518 struct thread *td = curthread;
519 int error;
520
521 /*
522 * Check for unsettable attributes.
523 */
524 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
525 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
526 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
527 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
528 return (EINVAL);
529 }
530 if (vap->va_flags != VNOVAL) {
531 if (vp->v_mount->mnt_flag & MNT_RDONLY)
532 return (EROFS);
533 /*
534 * Callers may only modify the file flags on objects they
535 * have VADMIN rights for.
536 */
537 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
538 return (error);
539 /*
540 * Unprivileged processes are not permitted to unset system
541 * flags, or modify flags if any system flags are set.
542 * Privileged non-jail processes may not modify system flags
543 * if securelevel > 0 and any existing system flags are set.
544 * Privileged jail processes behave like privileged non-jail
545 * processes if the security.jail.chflags_allowed sysctl is
546 * is non-zero; otherwise, they behave like unprivileged
547 * processes.
548 */
549 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
550 if (ip->i_flags
551 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
552 error = securelevel_gt(cred, 0);
553 if (error)
554 return (error);
555 }
556 /* Snapshot flag cannot be set or cleared */
557 if (((vap->va_flags & SF_SNAPSHOT) != 0 &&
558 (ip->i_flags & SF_SNAPSHOT) == 0) ||
559 ((vap->va_flags & SF_SNAPSHOT) == 0 &&
560 (ip->i_flags & SF_SNAPSHOT) != 0))
561 return (EPERM);
562 ip->i_flags = vap->va_flags;
563 DIP_SET(ip, i_flags, vap->va_flags);
564 } else {
565 if (ip->i_flags
566 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
567 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
568 return (EPERM);
569 ip->i_flags &= SF_SETTABLE;
570 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
571 DIP_SET(ip, i_flags, ip->i_flags);
572 }
573 ip->i_flag |= IN_CHANGE;
574 if (vap->va_flags & (IMMUTABLE | APPEND))
575 return (0);
576 }
577 if (ip->i_flags & (IMMUTABLE | APPEND))
578 return (EPERM);
579 /*
580 * Go through the fields and update iff not VNOVAL.
581 */
582 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
583 if (vp->v_mount->mnt_flag & MNT_RDONLY)
584 return (EROFS);
585 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
586 td)) != 0)
587 return (error);
588 }
589 if (vap->va_size != VNOVAL) {
590 /*
591 * XXX most of the following special cases should be in
592 * callers instead of in N filesystems. The VDIR check
593 * mostly already is.
594 */
595 switch (vp->v_type) {
596 case VDIR:
597 return (EISDIR);
598 case VLNK:
599 case VREG:
600 /*
601 * Truncation should have an effect in these cases.
602 * Disallow it if the filesystem is read-only or
603 * the file is being snapshotted.
604 */
605 if (vp->v_mount->mnt_flag & MNT_RDONLY)
606 return (EROFS);
607 if ((ip->i_flags & SF_SNAPSHOT) != 0)
608 return (EPERM);
609 break;
610 default:
611 /*
612 * According to POSIX, the result is unspecified
613 * for file types other than regular files,
614 * directories and shared memory objects. We
615 * don't support shared memory objects in the file
616 * system, and have dubious support for truncating
617 * symlinks. Just ignore the request in other cases.
618 */
619 return (0);
620 }
621 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL,
622 cred, td)) != 0)
623 return (error);
624 }
625 if (vap->va_atime.tv_sec != VNOVAL ||
626 vap->va_mtime.tv_sec != VNOVAL ||
627 vap->va_birthtime.tv_sec != VNOVAL) {
628 if (vp->v_mount->mnt_flag & MNT_RDONLY)
629 return (EROFS);
630 if ((ip->i_flags & SF_SNAPSHOT) != 0)
631 return (EPERM);
632 /*
633 * From utimes(2):
634 * If times is NULL, ... The caller must be the owner of
635 * the file, have permission to write the file, or be the
636 * super-user.
637 * If times is non-NULL, ... The caller must be the owner of
638 * the file or be the super-user.
639 *
640 * Possibly for historical reasons, try to use VADMIN in
641 * preference to VWRITE for a NULL timestamp. This means we
642 * will return EACCES in preference to EPERM if neither
643 * check succeeds.
644 */
645 if (vap->va_vaflags & VA_UTIMES_NULL) {
646 /*
647 * NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes
648 *
649 * "A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
650 * will be allowed to set the times [..] to the current
651 * server time."
652 *
653 * XXX: Calling it four times seems a little excessive.
654 */
655 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
656 if (error)
657 error = VOP_ACCESS(vp, VWRITE, cred, td);
658 } else
659 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
660 if (error)
661 return (error);
662 if (vap->va_atime.tv_sec != VNOVAL)
663 ip->i_flag |= IN_ACCESS;
664 if (vap->va_mtime.tv_sec != VNOVAL)
665 ip->i_flag |= IN_CHANGE | IN_UPDATE;
666 if (vap->va_birthtime.tv_sec != VNOVAL &&
667 ip->i_ump->um_fstype == UFS2)
668 ip->i_flag |= IN_MODIFIED;
669 ufs_itimes(vp);
670 if (vap->va_atime.tv_sec != VNOVAL) {
671 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
672 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
673 }
674 if (vap->va_mtime.tv_sec != VNOVAL) {
675 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
676 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
677 }
678 if (vap->va_birthtime.tv_sec != VNOVAL &&
679 ip->i_ump->um_fstype == UFS2) {
680 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
681 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
682 }
683 error = UFS_UPDATE(vp, 0);
684 if (error)
685 return (error);
686 }
687 error = 0;
688 if (vap->va_mode != (mode_t)VNOVAL) {
689 if (vp->v_mount->mnt_flag & MNT_RDONLY)
690 return (EROFS);
691 if ((ip->i_flags & SF_SNAPSHOT) != 0 && (vap->va_mode &
692 (S_IXUSR | S_IWUSR | S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)))
693 return (EPERM);
694 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
695 }
696 return (error);
697 }
698
699 #ifdef UFS_ACL
700 static int
701 ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode,
702 int file_owner_id, struct ucred *cred, struct thread *td)
703 {
704 int error;
705 struct acl *aclp;
706
707 aclp = acl_alloc(M_WAITOK);
708 error = ufs_getacl_nfs4_internal(vp, aclp, td);
709 /*
710 * We don't have to handle EOPNOTSUPP here, as the filesystem claims
711 * it supports ACLs.
712 */
713 if (error)
714 goto out;
715
716 acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id);
717 error = ufs_setacl_nfs4_internal(vp, aclp, td);
718
719 out:
720 acl_free(aclp);
721 return (error);
722 }
723 #endif /* UFS_ACL */
724
725 /*
726 * Mark this file's access time for update for vfs_mark_atime(). This
727 * is called from execve() and mmap().
728 */
729 static int
730 ufs_markatime(ap)
731 struct vop_markatime_args /* {
732 struct vnode *a_vp;
733 } */ *ap;
734 {
735 struct vnode *vp = ap->a_vp;
736 struct inode *ip = VTOI(vp);
737
738 VI_LOCK(vp);
739 ip->i_flag |= IN_ACCESS;
740 VI_UNLOCK(vp);
741 return (0);
742 }
743
744 /*
745 * Change the mode on a file.
746 * Inode must be locked before calling.
747 */
748 static int
749 ufs_chmod(vp, mode, cred, td)
750 struct vnode *vp;
751 int mode;
752 struct ucred *cred;
753 struct thread *td;
754 {
755 struct inode *ip = VTOI(vp);
756 int error;
757
758 /*
759 * To modify the permissions on a file, must possess VADMIN
760 * for that file.
761 */
762 if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td)))
763 return (error);
764 /*
765 * Privileged processes may set the sticky bit on non-directories,
766 * as well as set the setgid bit on a file with a group that the
767 * process is not a member of. Both of these are allowed in
768 * jail(8).
769 */
770 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
771 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
772 return (EFTYPE);
773 }
774 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
775 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
776 if (error)
777 return (error);
778 }
779
780 /*
781 * Deny setting setuid if we are not the file owner.
782 */
783 if ((mode & ISUID) && ip->i_uid != cred->cr_uid) {
784 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
785 if (error)
786 return (error);
787 }
788
789 ip->i_mode &= ~ALLPERMS;
790 ip->i_mode |= (mode & ALLPERMS);
791 DIP_SET(ip, i_mode, ip->i_mode);
792 ip->i_flag |= IN_CHANGE;
793 #ifdef UFS_ACL
794 if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0)
795 error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td);
796 #endif
797 return (error);
798 }
799
800 /*
801 * Perform chown operation on inode ip;
802 * inode must be locked prior to call.
803 */
804 static int
805 ufs_chown(vp, uid, gid, cred, td)
806 struct vnode *vp;
807 uid_t uid;
808 gid_t gid;
809 struct ucred *cred;
810 struct thread *td;
811 {
812 struct inode *ip = VTOI(vp);
813 uid_t ouid;
814 gid_t ogid;
815 int error = 0;
816 #ifdef QUOTA
817 int i;
818 ufs2_daddr_t change;
819 #endif
820
821 if (uid == (uid_t)VNOVAL)
822 uid = ip->i_uid;
823 if (gid == (gid_t)VNOVAL)
824 gid = ip->i_gid;
825 /*
826 * To modify the ownership of a file, must possess VADMIN for that
827 * file.
828 */
829 if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td)))
830 return (error);
831 /*
832 * To change the owner of a file, or change the group of a file to a
833 * group of which we are not a member, the caller must have
834 * privilege.
835 */
836 if (((uid != ip->i_uid && uid != cred->cr_uid) ||
837 (gid != ip->i_gid && !groupmember(gid, cred))) &&
838 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
839 return (error);
840 ogid = ip->i_gid;
841 ouid = ip->i_uid;
842 #ifdef QUOTA
843 if ((error = getinoquota(ip)) != 0)
844 return (error);
845 if (ouid == uid) {
846 dqrele(vp, ip->i_dquot[USRQUOTA]);
847 ip->i_dquot[USRQUOTA] = NODQUOT;
848 }
849 if (ogid == gid) {
850 dqrele(vp, ip->i_dquot[GRPQUOTA]);
851 ip->i_dquot[GRPQUOTA] = NODQUOT;
852 }
853 change = DIP(ip, i_blocks);
854 (void) chkdq(ip, -change, cred, CHOWN);
855 (void) chkiq(ip, -1, cred, CHOWN);
856 for (i = 0; i < MAXQUOTAS; i++) {
857 dqrele(vp, ip->i_dquot[i]);
858 ip->i_dquot[i] = NODQUOT;
859 }
860 #endif
861 ip->i_gid = gid;
862 DIP_SET(ip, i_gid, gid);
863 ip->i_uid = uid;
864 DIP_SET(ip, i_uid, uid);
865 #ifdef QUOTA
866 if ((error = getinoquota(ip)) == 0) {
867 if (ouid == uid) {
868 dqrele(vp, ip->i_dquot[USRQUOTA]);
869 ip->i_dquot[USRQUOTA] = NODQUOT;
870 }
871 if (ogid == gid) {
872 dqrele(vp, ip->i_dquot[GRPQUOTA]);
873 ip->i_dquot[GRPQUOTA] = NODQUOT;
874 }
875 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
876 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
877 goto good;
878 else
879 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
880 }
881 for (i = 0; i < MAXQUOTAS; i++) {
882 dqrele(vp, ip->i_dquot[i]);
883 ip->i_dquot[i] = NODQUOT;
884 }
885 }
886 ip->i_gid = ogid;
887 DIP_SET(ip, i_gid, ogid);
888 ip->i_uid = ouid;
889 DIP_SET(ip, i_uid, ouid);
890 if (getinoquota(ip) == 0) {
891 if (ouid == uid) {
892 dqrele(vp, ip->i_dquot[USRQUOTA]);
893 ip->i_dquot[USRQUOTA] = NODQUOT;
894 }
895 if (ogid == gid) {
896 dqrele(vp, ip->i_dquot[GRPQUOTA]);
897 ip->i_dquot[GRPQUOTA] = NODQUOT;
898 }
899 (void) chkdq(ip, change, cred, FORCE|CHOWN);
900 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
901 (void) getinoquota(ip);
902 }
903 return (error);
904 good:
905 if (getinoquota(ip))
906 panic("ufs_chown: lost quota");
907 #endif /* QUOTA */
908 ip->i_flag |= IN_CHANGE;
909 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
910 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) {
911 ip->i_mode &= ~(ISUID | ISGID);
912 DIP_SET(ip, i_mode, ip->i_mode);
913 }
914 }
915 return (0);
916 }
917
918 static int
919 ufs_remove(ap)
920 struct vop_remove_args /* {
921 struct vnode *a_dvp;
922 struct vnode *a_vp;
923 struct componentname *a_cnp;
924 } */ *ap;
925 {
926 struct inode *ip;
927 struct vnode *vp = ap->a_vp;
928 struct vnode *dvp = ap->a_dvp;
929 int error;
930 struct thread *td;
931
932 td = curthread;
933 ip = VTOI(vp);
934 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
935 (VTOI(dvp)->i_flags & APPEND)) {
936 error = EPERM;
937 goto out;
938 }
939 #ifdef UFS_GJOURNAL
940 ufs_gjournal_orphan(vp);
941 #endif
942 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
943 if (ip->i_nlink <= 0)
944 vp->v_vflag |= VV_NOSYNC;
945 if ((ip->i_flags & SF_SNAPSHOT) != 0) {
946 /*
947 * Avoid deadlock where another thread is trying to
948 * update the inodeblock for dvp and is waiting on
949 * snaplk. Temporary unlock the vnode lock for the
950 * unlinked file and sync the directory. This should
951 * allow vput() of the directory to not block later on
952 * while holding the snapshot vnode locked, assuming
953 * that the directory hasn't been unlinked too.
954 */
955 VOP_UNLOCK(vp, 0);
956 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
957 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
958 }
959 out:
960 return (error);
961 }
962
963 /*
964 * link vnode call
965 */
966 static int
967 ufs_link(ap)
968 struct vop_link_args /* {
969 struct vnode *a_tdvp;
970 struct vnode *a_vp;
971 struct componentname *a_cnp;
972 } */ *ap;
973 {
974 struct vnode *vp = ap->a_vp;
975 struct vnode *tdvp = ap->a_tdvp;
976 struct componentname *cnp = ap->a_cnp;
977 struct inode *ip;
978 struct direct newdir;
979 int error;
980
981 #ifdef INVARIANTS
982 if ((cnp->cn_flags & HASBUF) == 0)
983 panic("ufs_link: no name");
984 #endif
985 if (tdvp->v_mount != vp->v_mount) {
986 error = EXDEV;
987 goto out;
988 }
989 if (VTOI(tdvp)->i_effnlink < 2)
990 panic("ufs_link: Bad link count %d on parent",
991 VTOI(tdvp)->i_effnlink);
992 ip = VTOI(vp);
993 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
994 error = EMLINK;
995 goto out;
996 }
997 if (ip->i_flags & (IMMUTABLE | APPEND)) {
998 error = EPERM;
999 goto out;
1000 }
1001 ip->i_effnlink++;
1002 ip->i_nlink++;
1003 DIP_SET(ip, i_nlink, ip->i_nlink);
1004 ip->i_flag |= IN_CHANGE;
1005 if (DOINGSOFTDEP(vp))
1006 softdep_setup_link(VTOI(tdvp), ip);
1007 error = UFS_UPDATE(vp, !(DOINGSOFTDEP(vp) | DOINGASYNC(vp)));
1008 if (!error) {
1009 ufs_makedirentry(ip, cnp, &newdir);
1010 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL, 0);
1011 }
1012
1013 if (error) {
1014 ip->i_effnlink--;
1015 ip->i_nlink--;
1016 DIP_SET(ip, i_nlink, ip->i_nlink);
1017 ip->i_flag |= IN_CHANGE;
1018 if (DOINGSOFTDEP(vp))
1019 softdep_revert_link(VTOI(tdvp), ip);
1020 }
1021 out:
1022 return (error);
1023 }
1024
1025 /*
1026 * whiteout vnode call
1027 */
1028 static int
1029 ufs_whiteout(ap)
1030 struct vop_whiteout_args /* {
1031 struct vnode *a_dvp;
1032 struct componentname *a_cnp;
1033 int a_flags;
1034 } */ *ap;
1035 {
1036 struct vnode *dvp = ap->a_dvp;
1037 struct componentname *cnp = ap->a_cnp;
1038 struct direct newdir;
1039 int error = 0;
1040
1041 switch (ap->a_flags) {
1042 case LOOKUP:
1043 /* 4.4 format directories support whiteout operations */
1044 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1045 return (0);
1046 return (EOPNOTSUPP);
1047
1048 case CREATE:
1049 /* create a new directory whiteout */
1050 #ifdef INVARIANTS
1051 if ((cnp->cn_flags & SAVENAME) == 0)
1052 panic("ufs_whiteout: missing name");
1053 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
1054 panic("ufs_whiteout: old format filesystem");
1055 #endif
1056
1057 newdir.d_ino = WINO;
1058 newdir.d_namlen = cnp->cn_namelen;
1059 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
1060 newdir.d_type = DT_WHT;
1061 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL, 0);
1062 break;
1063
1064 case DELETE:
1065 /* remove an existing directory whiteout */
1066 #ifdef INVARIANTS
1067 if (dvp->v_mount->mnt_maxsymlinklen <= 0)
1068 panic("ufs_whiteout: old format filesystem");
1069 #endif
1070
1071 cnp->cn_flags &= ~DOWHITEOUT;
1072 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
1073 break;
1074 default:
1075 panic("ufs_whiteout: unknown op");
1076 }
1077 return (error);
1078 }
1079
1080 static volatile int rename_restarts;
1081 SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD,
1082 __DEVOLATILE(int *, &rename_restarts), 0,
1083 "Times rename had to restart due to lock contention");
1084
1085 /*
1086 * Rename system call.
1087 * rename("foo", "bar");
1088 * is essentially
1089 * unlink("bar");
1090 * link("foo", "bar");
1091 * unlink("foo");
1092 * but ``atomically''. Can't do full commit without saving state in the
1093 * inode on disk which isn't feasible at this time. Best we can do is
1094 * always guarantee the target exists.
1095 *
1096 * Basic algorithm is:
1097 *
1098 * 1) Bump link count on source while we're linking it to the
1099 * target. This also ensure the inode won't be deleted out
1100 * from underneath us while we work (it may be truncated by
1101 * a concurrent `trunc' or `open' for creation).
1102 * 2) Link source to destination. If destination already exists,
1103 * delete it first.
1104 * 3) Unlink source reference to inode if still around. If a
1105 * directory was moved and the parent of the destination
1106 * is different from the source, patch the ".." entry in the
1107 * directory.
1108 */
1109 static int
1110 ufs_rename(ap)
1111 struct vop_rename_args /* {
1112 struct vnode *a_fdvp;
1113 struct vnode *a_fvp;
1114 struct componentname *a_fcnp;
1115 struct vnode *a_tdvp;
1116 struct vnode *a_tvp;
1117 struct componentname *a_tcnp;
1118 } */ *ap;
1119 {
1120 struct vnode *tvp = ap->a_tvp;
1121 struct vnode *tdvp = ap->a_tdvp;
1122 struct vnode *fvp = ap->a_fvp;
1123 struct vnode *fdvp = ap->a_fdvp;
1124 struct vnode *nvp;
1125 struct componentname *tcnp = ap->a_tcnp;
1126 struct componentname *fcnp = ap->a_fcnp;
1127 struct thread *td = fcnp->cn_thread;
1128 struct inode *fip, *tip, *tdp, *fdp;
1129 struct direct newdir;
1130 off_t endoff;
1131 int doingdirectory, newparent;
1132 int error = 0;
1133 struct mount *mp;
1134 ino_t ino;
1135
1136 #ifdef INVARIANTS
1137 if ((tcnp->cn_flags & HASBUF) == 0 ||
1138 (fcnp->cn_flags & HASBUF) == 0)
1139 panic("ufs_rename: no name");
1140 #endif
1141 endoff = 0;
1142 mp = tdvp->v_mount;
1143 VOP_UNLOCK(tdvp, 0);
1144 if (tvp && tvp != tdvp)
1145 VOP_UNLOCK(tvp, 0);
1146 /*
1147 * Check for cross-device rename.
1148 */
1149 if ((fvp->v_mount != tdvp->v_mount) ||
1150 (tvp && (fvp->v_mount != tvp->v_mount))) {
1151 error = EXDEV;
1152 mp = NULL;
1153 goto releout;
1154 }
1155 error = vfs_busy(mp, 0);
1156 if (error) {
1157 mp = NULL;
1158 goto releout;
1159 }
1160 relock:
1161 /*
1162 * We need to acquire 2 to 4 locks depending on whether tvp is NULL
1163 * and fdvp and tdvp are the same directory. Subsequently we need
1164 * to double-check all paths and in the directory rename case we
1165 * need to verify that we are not creating a directory loop. To
1166 * handle this we acquire all but fdvp using non-blocking
1167 * acquisitions. If we fail to acquire any lock in the path we will
1168 * drop all held locks, acquire the new lock in a blocking fashion,
1169 * and then release it and restart the rename. This acquire/release
1170 * step ensures that we do not spin on a lock waiting for release.
1171 */
1172 error = vn_lock(fdvp, LK_EXCLUSIVE);
1173 if (error)
1174 goto releout;
1175 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1176 VOP_UNLOCK(fdvp, 0);
1177 error = vn_lock(tdvp, LK_EXCLUSIVE);
1178 if (error)
1179 goto releout;
1180 VOP_UNLOCK(tdvp, 0);
1181 atomic_add_int(&rename_restarts, 1);
1182 goto relock;
1183 }
1184 /*
1185 * Re-resolve fvp to be certain it still exists and fetch the
1186 * correct vnode.
1187 */
1188 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1189 if (error) {
1190 VOP_UNLOCK(fdvp, 0);
1191 VOP_UNLOCK(tdvp, 0);
1192 goto releout;
1193 }
1194 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1195 if (error) {
1196 VOP_UNLOCK(fdvp, 0);
1197 VOP_UNLOCK(tdvp, 0);
1198 if (error != EBUSY)
1199 goto releout;
1200 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1201 if (error != 0)
1202 goto releout;
1203 VOP_UNLOCK(nvp, 0);
1204 vrele(fvp);
1205 fvp = nvp;
1206 atomic_add_int(&rename_restarts, 1);
1207 goto relock;
1208 }
1209 vrele(fvp);
1210 fvp = nvp;
1211 /*
1212 * Re-resolve tvp and acquire the vnode lock if present.
1213 */
1214 error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino);
1215 if (error != 0 && error != EJUSTRETURN) {
1216 VOP_UNLOCK(fdvp, 0);
1217 VOP_UNLOCK(tdvp, 0);
1218 VOP_UNLOCK(fvp, 0);
1219 goto releout;
1220 }
1221 /*
1222 * If tvp disappeared we just carry on.
1223 */
1224 if (error == EJUSTRETURN && tvp != NULL) {
1225 vrele(tvp);
1226 tvp = NULL;
1227 }
1228 /*
1229 * Get the tvp ino if the lookup succeeded. We may have to restart
1230 * if the non-blocking acquire fails.
1231 */
1232 if (error == 0) {
1233 nvp = NULL;
1234 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1235 if (tvp)
1236 vrele(tvp);
1237 tvp = nvp;
1238 if (error) {
1239 VOP_UNLOCK(fdvp, 0);
1240 VOP_UNLOCK(tdvp, 0);
1241 VOP_UNLOCK(fvp, 0);
1242 if (error != EBUSY)
1243 goto releout;
1244 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1245 if (error != 0)
1246 goto releout;
1247 VOP_UNLOCK(nvp, 0);
1248 atomic_add_int(&rename_restarts, 1);
1249 goto relock;
1250 }
1251 }
1252 fdp = VTOI(fdvp);
1253 fip = VTOI(fvp);
1254 tdp = VTOI(tdvp);
1255 tip = NULL;
1256 if (tvp)
1257 tip = VTOI(tvp);
1258 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1259 (VTOI(tdvp)->i_flags & APPEND))) {
1260 error = EPERM;
1261 goto unlockout;
1262 }
1263 /*
1264 * Renaming a file to itself has no effect. The upper layers should
1265 * not call us in that case. However, things could change after
1266 * we drop the locks above.
1267 */
1268 if (fvp == tvp) {
1269 error = 0;
1270 goto unlockout;
1271 }
1272 doingdirectory = 0;
1273 newparent = 0;
1274 ino = fip->i_number;
1275 if (fip->i_nlink >= LINK_MAX) {
1276 error = EMLINK;
1277 goto unlockout;
1278 }
1279 if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1280 || (fdp->i_flags & APPEND)) {
1281 error = EPERM;
1282 goto unlockout;
1283 }
1284 if ((fip->i_mode & IFMT) == IFDIR) {
1285 /*
1286 * Avoid ".", "..", and aliases of "." for obvious reasons.
1287 */
1288 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1289 fdp == fip ||
1290 (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
1291 error = EINVAL;
1292 goto unlockout;
1293 }
1294 if (fdp->i_number != tdp->i_number)
1295 newparent = tdp->i_number;
1296 doingdirectory = 1;
1297 }
1298 if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) ||
1299 (tvp != NULL && tvp->v_type == VDIR &&
1300 tvp->v_mountedhere != NULL)) {
1301 error = EXDEV;
1302 goto unlockout;
1303 }
1304
1305 /*
1306 * If ".." must be changed (ie the directory gets a new
1307 * parent) then the source directory must not be in the
1308 * directory hierarchy above the target, as this would
1309 * orphan everything below the source directory. Also
1310 * the user must have write permission in the source so
1311 * as to be able to change "..".
1312 */
1313 if (doingdirectory && newparent) {
1314 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1315 if (error)
1316 goto unlockout;
1317 error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred,
1318 &ino);
1319 /*
1320 * We encountered a lock that we have to wait for. Unlock
1321 * everything else and VGET before restarting.
1322 */
1323 if (ino) {
1324 VOP_UNLOCK(fdvp, 0);
1325 VOP_UNLOCK(fvp, 0);
1326 VOP_UNLOCK(tdvp, 0);
1327 if (tvp)
1328 VOP_UNLOCK(tvp, 0);
1329 error = VFS_VGET(mp, ino, LK_SHARED, &nvp);
1330 if (error == 0)
1331 vput(nvp);
1332 atomic_add_int(&rename_restarts, 1);
1333 goto relock;
1334 }
1335 if (error)
1336 goto unlockout;
1337 if ((tcnp->cn_flags & SAVESTART) == 0)
1338 panic("ufs_rename: lost to startdir");
1339 }
1340 if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 ||
1341 tdp->i_effnlink == 0)
1342 panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp);
1343
1344 /*
1345 * 1) Bump link count while we're moving stuff
1346 * around. If we crash somewhere before
1347 * completing our work, the link count
1348 * may be wrong, but correctable.
1349 */
1350 fip->i_effnlink++;
1351 fip->i_nlink++;
1352 DIP_SET(fip, i_nlink, fip->i_nlink);
1353 fip->i_flag |= IN_CHANGE;
1354 if (DOINGSOFTDEP(fvp))
1355 softdep_setup_link(tdp, fip);
1356 error = UFS_UPDATE(fvp, !(DOINGSOFTDEP(fvp) | DOINGASYNC(fvp)));
1357 if (error)
1358 goto bad;
1359
1360 /*
1361 * 2) If target doesn't exist, link the target
1362 * to the source and unlink the source.
1363 * Otherwise, rewrite the target directory
1364 * entry to reference the source inode and
1365 * expunge the original entry's existence.
1366 */
1367 if (tip == NULL) {
1368 if (tdp->i_dev != fip->i_dev)
1369 panic("ufs_rename: EXDEV");
1370 if (doingdirectory && newparent) {
1371 /*
1372 * Account for ".." in new directory.
1373 * When source and destination have the same
1374 * parent we don't adjust the link count. The
1375 * actual link modification is completed when
1376 * .. is rewritten below.
1377 */
1378 if ((nlink_t)tdp->i_nlink >= LINK_MAX) {
1379 error = EMLINK;
1380 goto bad;
1381 }
1382 }
1383 ufs_makedirentry(fip, tcnp, &newdir);
1384 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL, 1);
1385 if (error)
1386 goto bad;
1387 /* Setup tdvp for directory compaction if needed. */
1388 if (tdp->i_count && tdp->i_endoff &&
1389 tdp->i_endoff < tdp->i_size)
1390 endoff = tdp->i_endoff;
1391 } else {
1392 if (tip->i_dev != tdp->i_dev || tip->i_dev != fip->i_dev)
1393 panic("ufs_rename: EXDEV");
1394 /*
1395 * Short circuit rename(foo, foo).
1396 */
1397 if (tip->i_number == fip->i_number)
1398 panic("ufs_rename: same file");
1399 /*
1400 * If the parent directory is "sticky", then the caller
1401 * must possess VADMIN for the parent directory, or the
1402 * destination of the rename. This implements append-only
1403 * directories.
1404 */
1405 if ((tdp->i_mode & S_ISTXT) &&
1406 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1407 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1408 error = EPERM;
1409 goto bad;
1410 }
1411 /*
1412 * Target must be empty if a directory and have no links
1413 * to it. Also, ensure source and target are compatible
1414 * (both directories, or both not directories).
1415 */
1416 if ((tip->i_mode & IFMT) == IFDIR) {
1417 if ((tip->i_effnlink > 2) ||
1418 !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) {
1419 error = ENOTEMPTY;
1420 goto bad;
1421 }
1422 if (!doingdirectory) {
1423 error = ENOTDIR;
1424 goto bad;
1425 }
1426 cache_purge(tdvp);
1427 } else if (doingdirectory) {
1428 error = EISDIR;
1429 goto bad;
1430 }
1431 if (doingdirectory) {
1432 if (!newparent) {
1433 tdp->i_effnlink--;
1434 if (DOINGSOFTDEP(tdvp))
1435 softdep_change_linkcnt(tdp);
1436 }
1437 tip->i_effnlink--;
1438 if (DOINGSOFTDEP(tvp))
1439 softdep_change_linkcnt(tip);
1440 }
1441 error = ufs_dirrewrite(tdp, tip, fip->i_number,
1442 IFTODT(fip->i_mode),
1443 (doingdirectory && newparent) ? newparent : doingdirectory);
1444 if (error) {
1445 if (doingdirectory) {
1446 if (!newparent) {
1447 tdp->i_effnlink++;
1448 if (DOINGSOFTDEP(tdvp))
1449 softdep_change_linkcnt(tdp);
1450 }
1451 tip->i_effnlink++;
1452 if (DOINGSOFTDEP(tvp))
1453 softdep_change_linkcnt(tip);
1454 }
1455 }
1456 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1457 /*
1458 * The only stuff left in the directory is "."
1459 * and "..". The "." reference is inconsequential
1460 * since we are quashing it. We have removed the "."
1461 * reference and the reference in the parent directory,
1462 * but there may be other hard links. The soft
1463 * dependency code will arrange to do these operations
1464 * after the parent directory entry has been deleted on
1465 * disk, so when running with that code we avoid doing
1466 * them now.
1467 */
1468 if (!newparent) {
1469 tdp->i_nlink--;
1470 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1471 tdp->i_flag |= IN_CHANGE;
1472 }
1473 tip->i_nlink--;
1474 DIP_SET(tip, i_nlink, tip->i_nlink);
1475 tip->i_flag |= IN_CHANGE;
1476 }
1477 }
1478
1479 /*
1480 * 3) Unlink the source. We have to resolve the path again to
1481 * fixup the directory offset and count for ufs_dirremove.
1482 */
1483 if (fdvp == tdvp) {
1484 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1485 if (error)
1486 panic("ufs_rename: from entry went away!");
1487 if (ino != fip->i_number)
1488 panic("ufs_rename: ino mismatch %d != %d\n", ino,
1489 fip->i_number);
1490 }
1491 /*
1492 * If the source is a directory with a
1493 * new parent, the link count of the old
1494 * parent directory must be decremented
1495 * and ".." set to point to the new parent.
1496 */
1497 if (doingdirectory && newparent) {
1498 /*
1499 * If tip exists we simply use its link, otherwise we must
1500 * add a new one.
1501 */
1502 if (tip == NULL) {
1503 tdp->i_effnlink++;
1504 tdp->i_nlink++;
1505 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1506 tdp->i_flag |= IN_CHANGE;
1507 if (DOINGSOFTDEP(tdvp))
1508 softdep_setup_dotdot_link(tdp, fip);
1509 error = UFS_UPDATE(tdvp, !(DOINGSOFTDEP(tdvp) |
1510 DOINGASYNC(tdvp)));
1511 /* Don't go to bad here as the new link exists. */
1512 if (error)
1513 goto unlockout;
1514 } else if (DOINGSUJ(tdvp))
1515 /* Journal must account for each new link. */
1516 softdep_setup_dotdot_link(tdp, fip);
1517 fip->i_offset = mastertemplate.dot_reclen;
1518 ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0);
1519 cache_purge(fdvp);
1520 }
1521 error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0);
1522 /*
1523 * The kern_renameat() looks up the fvp using the DELETE flag, which
1524 * causes the removal of the name cache entry for fvp.
1525 * As the relookup of the fvp is done in two steps:
1526 * ufs_lookup_ino() and then VFS_VGET(), another thread might do a
1527 * normal lookup of the from name just before the VFS_VGET() call,
1528 * causing the cache entry to be re-instantiated.
1529 */
1530 cache_purge(fvp);
1531
1532 unlockout:
1533 vput(fdvp);
1534 vput(fvp);
1535 if (tvp)
1536 vput(tvp);
1537 /*
1538 * If compaction or fsync was requested do it now that other locks
1539 * are no longer needed.
1540 */
1541 if (error == 0 && endoff != 0) {
1542 #ifdef UFS_DIRHASH
1543 if (tdp->i_dirhash != NULL)
1544 ufsdirhash_dirtrunc(tdp, endoff);
1545 #endif
1546 UFS_TRUNCATE(tdvp, endoff, IO_NORMAL | IO_SYNC, tcnp->cn_cred,
1547 td);
1548 }
1549 if (error == 0 && tdp->i_flag & IN_NEEDSYNC)
1550 error = VOP_FSYNC(tdvp, MNT_WAIT, td);
1551 vput(tdvp);
1552 if (mp)
1553 vfs_unbusy(mp);
1554 return (error);
1555
1556 bad:
1557 fip->i_effnlink--;
1558 fip->i_nlink--;
1559 DIP_SET(fip, i_nlink, fip->i_nlink);
1560 fip->i_flag |= IN_CHANGE;
1561 if (DOINGSOFTDEP(fvp))
1562 softdep_revert_link(tdp, fip);
1563 goto unlockout;
1564
1565 releout:
1566 vrele(fdvp);
1567 vrele(fvp);
1568 vrele(tdvp);
1569 if (tvp)
1570 vrele(tvp);
1571 if (mp)
1572 vfs_unbusy(mp);
1573
1574 return (error);
1575 }
1576
1577 #ifdef UFS_ACL
1578 static int
1579 ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1580 mode_t dmode, struct ucred *cred, struct thread *td)
1581 {
1582 int error;
1583 struct inode *ip = VTOI(tvp);
1584 struct acl *dacl, *acl;
1585
1586 acl = acl_alloc(M_WAITOK);
1587 dacl = acl_alloc(M_WAITOK);
1588
1589 /*
1590 * Retrieve default ACL from parent, if any.
1591 */
1592 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1593 switch (error) {
1594 case 0:
1595 /*
1596 * Retrieved a default ACL, so merge mode and ACL if
1597 * necessary. If the ACL is empty, fall through to
1598 * the "not defined or available" case.
1599 */
1600 if (acl->acl_cnt != 0) {
1601 dmode = acl_posix1e_newfilemode(dmode, acl);
1602 ip->i_mode = dmode;
1603 DIP_SET(ip, i_mode, dmode);
1604 *dacl = *acl;
1605 ufs_sync_acl_from_inode(ip, acl);
1606 break;
1607 }
1608 /* FALLTHROUGH */
1609
1610 case EOPNOTSUPP:
1611 /*
1612 * Just use the mode as-is.
1613 */
1614 ip->i_mode = dmode;
1615 DIP_SET(ip, i_mode, dmode);
1616 error = 0;
1617 goto out;
1618
1619 default:
1620 goto out;
1621 }
1622
1623 /*
1624 * XXX: If we abort now, will Soft Updates notify the extattr
1625 * code that the EAs for the file need to be released?
1626 */
1627 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1628 if (error == 0)
1629 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1630 switch (error) {
1631 case 0:
1632 break;
1633
1634 case EOPNOTSUPP:
1635 /*
1636 * XXX: This should not happen, as EOPNOTSUPP above
1637 * was supposed to free acl.
1638 */
1639 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1640 /*
1641 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1642 */
1643 break;
1644
1645 default:
1646 goto out;
1647 }
1648
1649 out:
1650 acl_free(acl);
1651 acl_free(dacl);
1652
1653 return (error);
1654 }
1655
1656 static int
1657 ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1658 mode_t mode, struct ucred *cred, struct thread *td)
1659 {
1660 int error;
1661 struct inode *ip = VTOI(tvp);
1662 struct acl *acl;
1663
1664 acl = acl_alloc(M_WAITOK);
1665
1666 /*
1667 * Retrieve default ACL for parent, if any.
1668 */
1669 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1670 switch (error) {
1671 case 0:
1672 /*
1673 * Retrieved a default ACL, so merge mode and ACL if
1674 * necessary.
1675 */
1676 if (acl->acl_cnt != 0) {
1677 /*
1678 * Two possible ways for default ACL to not
1679 * be present. First, the EA can be
1680 * undefined, or second, the default ACL can
1681 * be blank. If it's blank, fall through to
1682 * the it's not defined case.
1683 */
1684 mode = acl_posix1e_newfilemode(mode, acl);
1685 ip->i_mode = mode;
1686 DIP_SET(ip, i_mode, mode);
1687 ufs_sync_acl_from_inode(ip, acl);
1688 break;
1689 }
1690 /* FALLTHROUGH */
1691
1692 case EOPNOTSUPP:
1693 /*
1694 * Just use the mode as-is.
1695 */
1696 ip->i_mode = mode;
1697 DIP_SET(ip, i_mode, mode);
1698 error = 0;
1699 goto out;
1700
1701 default:
1702 goto out;
1703 }
1704
1705 /*
1706 * XXX: If we abort now, will Soft Updates notify the extattr
1707 * code that the EAs for the file need to be released?
1708 */
1709 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1710 switch (error) {
1711 case 0:
1712 break;
1713
1714 case EOPNOTSUPP:
1715 /*
1716 * XXX: This should not happen, as EOPNOTSUPP above was
1717 * supposed to free acl.
1718 */
1719 printf("ufs_makeinode: VOP_GETACL() but no "
1720 "VOP_SETACL()\n");
1721 /* panic("ufs_makeinode: VOP_GETACL() but no "
1722 "VOP_SETACL()"); */
1723 break;
1724
1725 default:
1726 goto out;
1727 }
1728
1729 out:
1730 acl_free(acl);
1731
1732 return (error);
1733 }
1734
1735 static int
1736 ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp,
1737 mode_t child_mode, struct ucred *cred, struct thread *td)
1738 {
1739 int error;
1740 struct acl *parent_aclp, *child_aclp;
1741
1742 parent_aclp = acl_alloc(M_WAITOK);
1743 child_aclp = acl_alloc(M_WAITOK | M_ZERO);
1744
1745 error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td);
1746 if (error)
1747 goto out;
1748 acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp,
1749 child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR);
1750 error = ufs_setacl_nfs4_internal(tvp, child_aclp, td);
1751 if (error)
1752 goto out;
1753 out:
1754 acl_free(parent_aclp);
1755 acl_free(child_aclp);
1756
1757 return (error);
1758 }
1759 #endif
1760
1761 /*
1762 * Mkdir system call
1763 */
1764 static int
1765 ufs_mkdir(ap)
1766 struct vop_mkdir_args /* {
1767 struct vnode *a_dvp;
1768 struct vnode **a_vpp;
1769 struct componentname *a_cnp;
1770 struct vattr *a_vap;
1771 } */ *ap;
1772 {
1773 struct vnode *dvp = ap->a_dvp;
1774 struct vattr *vap = ap->a_vap;
1775 struct componentname *cnp = ap->a_cnp;
1776 struct inode *ip, *dp;
1777 struct vnode *tvp;
1778 struct buf *bp;
1779 struct dirtemplate dirtemplate, *dtp;
1780 struct direct newdir;
1781 int error, dmode;
1782 long blkoff;
1783
1784 #ifdef INVARIANTS
1785 if ((cnp->cn_flags & HASBUF) == 0)
1786 panic("ufs_mkdir: no name");
1787 #endif
1788 dp = VTOI(dvp);
1789 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1790 error = EMLINK;
1791 goto out;
1792 }
1793 dmode = vap->va_mode & 0777;
1794 dmode |= IFDIR;
1795 /*
1796 * Must simulate part of ufs_makeinode here to acquire the inode,
1797 * but not have it entered in the parent directory. The entry is
1798 * made later after writing "." and ".." entries.
1799 */
1800 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
1801 if (error)
1802 goto out;
1803 ip = VTOI(tvp);
1804 ip->i_gid = dp->i_gid;
1805 DIP_SET(ip, i_gid, dp->i_gid);
1806 #ifdef SUIDDIR
1807 {
1808 #ifdef QUOTA
1809 struct ucred ucred, *ucp;
1810 gid_t ucred_group;
1811 ucp = cnp->cn_cred;
1812 #endif
1813 /*
1814 * If we are hacking owners here, (only do this where told to)
1815 * and we are not giving it TO root, (would subvert quotas)
1816 * then go ahead and give it to the other user.
1817 * The new directory also inherits the SUID bit.
1818 * If user's UID and dir UID are the same,
1819 * 'give it away' so that the SUID is still forced on.
1820 */
1821 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1822 (dp->i_mode & ISUID) && dp->i_uid) {
1823 dmode |= ISUID;
1824 ip->i_uid = dp->i_uid;
1825 DIP_SET(ip, i_uid, dp->i_uid);
1826 #ifdef QUOTA
1827 if (dp->i_uid != cnp->cn_cred->cr_uid) {
1828 /*
1829 * Make sure the correct user gets charged
1830 * for the space.
1831 * Make a dummy credential for the victim.
1832 * XXX This seems to never be accessed out of
1833 * our context so a stack variable is ok.
1834 */
1835 refcount_init(&ucred.cr_ref, 1);
1836 ucred.cr_uid = ip->i_uid;
1837 ucred.cr_ngroups = 1;
1838 ucred.cr_groups = &ucred_group;
1839 ucred.cr_groups[0] = dp->i_gid;
1840 ucp = &ucred;
1841 }
1842 #endif
1843 } else {
1844 ip->i_uid = cnp->cn_cred->cr_uid;
1845 DIP_SET(ip, i_uid, ip->i_uid);
1846 }
1847 #ifdef QUOTA
1848 if ((error = getinoquota(ip)) ||
1849 (error = chkiq(ip, 1, ucp, 0))) {
1850 if (DOINGSOFTDEP(tvp))
1851 softdep_revert_link(dp, ip);
1852 UFS_VFREE(tvp, ip->i_number, dmode);
1853 vput(tvp);
1854 return (error);
1855 }
1856 #endif
1857 }
1858 #else /* !SUIDDIR */
1859 ip->i_uid = cnp->cn_cred->cr_uid;
1860 DIP_SET(ip, i_uid, ip->i_uid);
1861 #ifdef QUOTA
1862 if ((error = getinoquota(ip)) ||
1863 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
1864 if (DOINGSOFTDEP(tvp))
1865 softdep_revert_link(dp, ip);
1866 UFS_VFREE(tvp, ip->i_number, dmode);
1867 vput(tvp);
1868 return (error);
1869 }
1870 #endif
1871 #endif /* !SUIDDIR */
1872 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1873 ip->i_mode = dmode;
1874 DIP_SET(ip, i_mode, dmode);
1875 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1876 ip->i_effnlink = 2;
1877 ip->i_nlink = 2;
1878 DIP_SET(ip, i_nlink, 2);
1879
1880 if (cnp->cn_flags & ISWHITEOUT) {
1881 ip->i_flags |= UF_OPAQUE;
1882 DIP_SET(ip, i_flags, ip->i_flags);
1883 }
1884
1885 /*
1886 * Bump link count in parent directory to reflect work done below.
1887 * Should be done before reference is created so cleanup is
1888 * possible if we crash.
1889 */
1890 dp->i_effnlink++;
1891 dp->i_nlink++;
1892 DIP_SET(dp, i_nlink, dp->i_nlink);
1893 dp->i_flag |= IN_CHANGE;
1894 if (DOINGSOFTDEP(dvp))
1895 softdep_setup_mkdir(dp, ip);
1896 error = UFS_UPDATE(dvp, !(DOINGSOFTDEP(dvp) | DOINGASYNC(dvp)));
1897 if (error)
1898 goto bad;
1899 #ifdef MAC
1900 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
1901 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
1902 dvp, tvp, cnp);
1903 if (error)
1904 goto bad;
1905 }
1906 #endif
1907 #ifdef UFS_ACL
1908 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1909 error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1910 cnp->cn_cred, cnp->cn_thread);
1911 if (error)
1912 goto bad;
1913 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
1914 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode,
1915 cnp->cn_cred, cnp->cn_thread);
1916 if (error)
1917 goto bad;
1918 }
1919 #endif /* !UFS_ACL */
1920
1921 /*
1922 * Initialize directory with "." and ".." from static template.
1923 */
1924 if (dvp->v_mount->mnt_maxsymlinklen > 0)
1925 dtp = &mastertemplate;
1926 else
1927 dtp = (struct dirtemplate *)&omastertemplate;
1928 dirtemplate = *dtp;
1929 dirtemplate.dot_ino = ip->i_number;
1930 dirtemplate.dotdot_ino = dp->i_number;
1931 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
1932 BA_CLRBUF, &bp)) != 0)
1933 goto bad;
1934 ip->i_size = DIRBLKSIZ;
1935 DIP_SET(ip, i_size, DIRBLKSIZ);
1936 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1937 vnode_pager_setsize(tvp, (u_long)ip->i_size);
1938 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
1939 if (DOINGSOFTDEP(tvp)) {
1940 /*
1941 * Ensure that the entire newly allocated block is a
1942 * valid directory so that future growth within the
1943 * block does not have to ensure that the block is
1944 * written before the inode.
1945 */
1946 blkoff = DIRBLKSIZ;
1947 while (blkoff < bp->b_bcount) {
1948 ((struct direct *)
1949 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
1950 blkoff += DIRBLKSIZ;
1951 }
1952 }
1953 if ((error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) |
1954 DOINGASYNC(tvp)))) != 0) {
1955 (void)bwrite(bp);
1956 goto bad;
1957 }
1958 /*
1959 * Directory set up, now install its entry in the parent directory.
1960 *
1961 * If we are not doing soft dependencies, then we must write out the
1962 * buffer containing the new directory body before entering the new
1963 * name in the parent. If we are doing soft dependencies, then the
1964 * buffer containing the new directory body will be passed to and
1965 * released in the soft dependency code after the code has attached
1966 * an appropriate ordering dependency to the buffer which ensures that
1967 * the buffer is written before the new name is written in the parent.
1968 */
1969 if (DOINGASYNC(dvp))
1970 bdwrite(bp);
1971 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
1972 goto bad;
1973 ufs_makedirentry(ip, cnp, &newdir);
1974 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp, 0);
1975
1976 bad:
1977 if (error == 0) {
1978 *ap->a_vpp = tvp;
1979 } else {
1980 dp->i_effnlink--;
1981 dp->i_nlink--;
1982 DIP_SET(dp, i_nlink, dp->i_nlink);
1983 dp->i_flag |= IN_CHANGE;
1984 /*
1985 * No need to do an explicit VOP_TRUNCATE here, vrele will
1986 * do this for us because we set the link count to 0.
1987 */
1988 ip->i_effnlink = 0;
1989 ip->i_nlink = 0;
1990 DIP_SET(ip, i_nlink, 0);
1991 ip->i_flag |= IN_CHANGE;
1992 if (DOINGSOFTDEP(tvp))
1993 softdep_revert_mkdir(dp, ip);
1994
1995 vput(tvp);
1996 }
1997 out:
1998 return (error);
1999 }
2000
2001 /*
2002 * Rmdir system call.
2003 */
2004 static int
2005 ufs_rmdir(ap)
2006 struct vop_rmdir_args /* {
2007 struct vnode *a_dvp;
2008 struct vnode *a_vp;
2009 struct componentname *a_cnp;
2010 } */ *ap;
2011 {
2012 struct vnode *vp = ap->a_vp;
2013 struct vnode *dvp = ap->a_dvp;
2014 struct componentname *cnp = ap->a_cnp;
2015 struct inode *ip, *dp;
2016 int error;
2017
2018 ip = VTOI(vp);
2019 dp = VTOI(dvp);
2020
2021 /*
2022 * Do not remove a directory that is in the process of being renamed.
2023 * Verify the directory is empty (and valid). Rmdir ".." will not be
2024 * valid since ".." will contain a reference to the current directory
2025 * and thus be non-empty. Do not allow the removal of mounted on
2026 * directories (this can happen when an NFS exported filesystem
2027 * tries to remove a locally mounted on directory).
2028 */
2029 error = 0;
2030 if (ip->i_effnlink < 2) {
2031 error = EINVAL;
2032 goto out;
2033 }
2034 if (dp->i_effnlink < 3)
2035 panic("ufs_dirrem: Bad link count %d on parent",
2036 dp->i_effnlink);
2037 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
2038 error = ENOTEMPTY;
2039 goto out;
2040 }
2041 if ((dp->i_flags & APPEND)
2042 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
2043 error = EPERM;
2044 goto out;
2045 }
2046 if (vp->v_mountedhere != 0) {
2047 error = EINVAL;
2048 goto out;
2049 }
2050 #ifdef UFS_GJOURNAL
2051 ufs_gjournal_orphan(vp);
2052 #endif
2053 /*
2054 * Delete reference to directory before purging
2055 * inode. If we crash in between, the directory
2056 * will be reattached to lost+found,
2057 */
2058 dp->i_effnlink--;
2059 ip->i_effnlink--;
2060 if (DOINGSOFTDEP(vp))
2061 softdep_setup_rmdir(dp, ip);
2062 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
2063 if (error) {
2064 dp->i_effnlink++;
2065 ip->i_effnlink++;
2066 if (DOINGSOFTDEP(vp))
2067 softdep_revert_rmdir(dp, ip);
2068 goto out;
2069 }
2070 cache_purge(dvp);
2071 /*
2072 * The only stuff left in the directory is "." and "..". The "."
2073 * reference is inconsequential since we are quashing it. The soft
2074 * dependency code will arrange to do these operations after
2075 * the parent directory entry has been deleted on disk, so
2076 * when running with that code we avoid doing them now.
2077 */
2078 if (!DOINGSOFTDEP(vp)) {
2079 dp->i_nlink--;
2080 DIP_SET(dp, i_nlink, dp->i_nlink);
2081 dp->i_flag |= IN_CHANGE;
2082 ip->i_nlink--;
2083 DIP_SET(ip, i_nlink, ip->i_nlink);
2084 ip->i_flag |= IN_CHANGE;
2085 }
2086 cache_purge(vp);
2087 #ifdef UFS_DIRHASH
2088 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
2089 if (ip->i_dirhash != NULL)
2090 ufsdirhash_free(ip);
2091 #endif
2092 out:
2093 return (error);
2094 }
2095
2096 /*
2097 * symlink -- make a symbolic link
2098 */
2099 static int
2100 ufs_symlink(ap)
2101 struct vop_symlink_args /* {
2102 struct vnode *a_dvp;
2103 struct vnode **a_vpp;
2104 struct componentname *a_cnp;
2105 struct vattr *a_vap;
2106 char *a_target;
2107 } */ *ap;
2108 {
2109 struct vnode *vp, **vpp = ap->a_vpp;
2110 struct inode *ip;
2111 int len, error;
2112
2113 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
2114 vpp, ap->a_cnp);
2115 if (error)
2116 return (error);
2117 vp = *vpp;
2118 len = strlen(ap->a_target);
2119 if (len < vp->v_mount->mnt_maxsymlinklen) {
2120 ip = VTOI(vp);
2121 bcopy(ap->a_target, SHORTLINK(ip), len);
2122 ip->i_size = len;
2123 DIP_SET(ip, i_size, len);
2124 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2125 } else
2126 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
2127 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
2128 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
2129 if (error)
2130 vput(vp);
2131 return (error);
2132 }
2133
2134 /*
2135 * Vnode op for reading directories.
2136 *
2137 * The routine below assumes that the on-disk format of a directory
2138 * is the same as that defined by <sys/dirent.h>. If the on-disk
2139 * format changes, then it will be necessary to do a conversion
2140 * from the on-disk format that read returns to the format defined
2141 * by <sys/dirent.h>.
2142 */
2143 int
2144 ufs_readdir(ap)
2145 struct vop_readdir_args /* {
2146 struct vnode *a_vp;
2147 struct uio *a_uio;
2148 struct ucred *a_cred;
2149 int *a_eofflag;
2150 int *a_ncookies;
2151 u_long **a_cookies;
2152 } */ *ap;
2153 {
2154 struct uio *uio = ap->a_uio;
2155 struct inode *ip;
2156 int error;
2157 size_t count, lost;
2158 off_t off;
2159
2160 if (ap->a_ncookies != NULL)
2161 /*
2162 * Ensure that the block is aligned. The caller can use
2163 * the cookies to determine where in the block to start.
2164 */
2165 uio->uio_offset &= ~(DIRBLKSIZ - 1);
2166 ip = VTOI(ap->a_vp);
2167 if (ip->i_effnlink == 0)
2168 return (0);
2169 off = uio->uio_offset;
2170 count = uio->uio_resid;
2171 /* Make sure we don't return partial entries. */
2172 if (count <= ((uio->uio_offset + count) & (DIRBLKSIZ -1)))
2173 return (EINVAL);
2174 count -= (uio->uio_offset + count) & (DIRBLKSIZ -1);
2175 lost = uio->uio_resid - count;
2176 uio->uio_resid = count;
2177 uio->uio_iov->iov_len = count;
2178 # if (BYTE_ORDER == LITTLE_ENDIAN)
2179 if (ap->a_vp->v_mount->mnt_maxsymlinklen > 0) {
2180 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
2181 } else {
2182 struct dirent *dp, *edp;
2183 struct uio auio;
2184 struct iovec aiov;
2185 caddr_t dirbuf;
2186 int readcnt;
2187 u_char tmp;
2188
2189 auio = *uio;
2190 auio.uio_iov = &aiov;
2191 auio.uio_iovcnt = 1;
2192 auio.uio_segflg = UIO_SYSSPACE;
2193 aiov.iov_len = count;
2194 dirbuf = malloc(count, M_TEMP, M_WAITOK);
2195 aiov.iov_base = dirbuf;
2196 error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
2197 if (error == 0) {
2198 readcnt = count - auio.uio_resid;
2199 edp = (struct dirent *)&dirbuf[readcnt];
2200 for (dp = (struct dirent *)dirbuf; dp < edp; ) {
2201 tmp = dp->d_namlen;
2202 dp->d_namlen = dp->d_type;
2203 dp->d_type = tmp;
2204 if (dp->d_reclen > 0) {
2205 dp = (struct dirent *)
2206 ((char *)dp + dp->d_reclen);
2207 } else {
2208 error = EIO;
2209 break;
2210 }
2211 }
2212 if (dp >= edp)
2213 error = uiomove(dirbuf, readcnt, uio);
2214 }
2215 free(dirbuf, M_TEMP);
2216 }
2217 # else
2218 error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
2219 # endif
2220 if (!error && ap->a_ncookies != NULL) {
2221 struct dirent* dpStart;
2222 struct dirent* dpEnd;
2223 struct dirent* dp;
2224 int ncookies;
2225 u_long *cookies;
2226 u_long *cookiep;
2227
2228 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2229 panic("ufs_readdir: unexpected uio from NFS server");
2230 dpStart = (struct dirent *)
2231 ((char *)uio->uio_iov->iov_base - (uio->uio_offset - off));
2232 dpEnd = (struct dirent *) uio->uio_iov->iov_base;
2233 for (dp = dpStart, ncookies = 0;
2234 dp < dpEnd;
2235 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen))
2236 ncookies++;
2237 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
2238 M_WAITOK);
2239 for (dp = dpStart, cookiep = cookies;
2240 dp < dpEnd;
2241 dp = (struct dirent *)((caddr_t) dp + dp->d_reclen)) {
2242 off += dp->d_reclen;
2243 *cookiep++ = (u_long) off;
2244 }
2245 *ap->a_ncookies = ncookies;
2246 *ap->a_cookies = cookies;
2247 }
2248 uio->uio_resid += lost;
2249 if (ap->a_eofflag)
2250 *ap->a_eofflag = VTOI(ap->a_vp)->i_size <= uio->uio_offset;
2251 return (error);
2252 }
2253
2254 /*
2255 * Return target name of a symbolic link
2256 */
2257 static int
2258 ufs_readlink(ap)
2259 struct vop_readlink_args /* {
2260 struct vnode *a_vp;
2261 struct uio *a_uio;
2262 struct ucred *a_cred;
2263 } */ *ap;
2264 {
2265 struct vnode *vp = ap->a_vp;
2266 struct inode *ip = VTOI(vp);
2267 doff_t isize;
2268
2269 isize = ip->i_size;
2270 if ((isize < vp->v_mount->mnt_maxsymlinklen) ||
2271 DIP(ip, i_blocks) == 0) { /* XXX - for old fastlink support */
2272 return (uiomove(SHORTLINK(ip), isize, ap->a_uio));
2273 }
2274 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
2275 }
2276
2277 /*
2278 * Calculate the logical to physical mapping if not done already,
2279 * then call the device strategy routine.
2280 *
2281 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
2282 * deadlock on memory. See ufs_bmap() for details.
2283 */
2284 static int
2285 ufs_strategy(ap)
2286 struct vop_strategy_args /* {
2287 struct vnode *a_vp;
2288 struct buf *a_bp;
2289 } */ *ap;
2290 {
2291 struct buf *bp = ap->a_bp;
2292 struct vnode *vp = ap->a_vp;
2293 struct bufobj *bo;
2294 struct inode *ip;
2295 ufs2_daddr_t blkno;
2296 int error;
2297
2298 ip = VTOI(vp);
2299 if (bp->b_blkno == bp->b_lblkno) {
2300 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2301 bp->b_blkno = blkno;
2302 if (error) {
2303 bp->b_error = error;
2304 bp->b_ioflags |= BIO_ERROR;
2305 bufdone(bp);
2306 return (0);
2307 }
2308 if ((long)bp->b_blkno == -1)
2309 vfs_bio_clrbuf(bp);
2310 }
2311 if ((long)bp->b_blkno == -1) {
2312 bufdone(bp);
2313 return (0);
2314 }
2315 bp->b_iooffset = dbtob(bp->b_blkno);
2316 bo = ip->i_umbufobj;
2317 BO_STRATEGY(bo, bp);
2318 return (0);
2319 }
2320
2321 /*
2322 * Print out the contents of an inode.
2323 */
2324 static int
2325 ufs_print(ap)
2326 struct vop_print_args /* {
2327 struct vnode *a_vp;
2328 } */ *ap;
2329 {
2330 struct vnode *vp = ap->a_vp;
2331 struct inode *ip = VTOI(vp);
2332
2333 printf("\tino %lu, on dev %s", (u_long)ip->i_number,
2334 devtoname(ip->i_dev));
2335 if (vp->v_type == VFIFO)
2336 fifo_printinfo(vp);
2337 printf("\n");
2338 return (0);
2339 }
2340
2341 /*
2342 * Close wrapper for fifos.
2343 *
2344 * Update the times on the inode then do device close.
2345 */
2346 static int
2347 ufsfifo_close(ap)
2348 struct vop_close_args /* {
2349 struct vnode *a_vp;
2350 int a_fflag;
2351 struct ucred *a_cred;
2352 struct thread *a_td;
2353 } */ *ap;
2354 {
2355 struct vnode *vp = ap->a_vp;
2356 int usecount;
2357
2358 VI_LOCK(vp);
2359 usecount = vp->v_usecount;
2360 if (usecount > 1)
2361 ufs_itimes_locked(vp);
2362 VI_UNLOCK(vp);
2363 return (fifo_specops.vop_close(ap));
2364 }
2365
2366 /*
2367 * Kqfilter wrapper for fifos.
2368 *
2369 * Fall through to ufs kqfilter routines if needed
2370 */
2371 static int
2372 ufsfifo_kqfilter(ap)
2373 struct vop_kqfilter_args *ap;
2374 {
2375 int error;
2376
2377 error = fifo_specops.vop_kqfilter(ap);
2378 if (error)
2379 error = vfs_kqfilter(ap);
2380 return (error);
2381 }
2382
2383 /*
2384 * Return POSIX pathconf information applicable to fifos.
2385 */
2386 static int
2387 ufsfifo_pathconf(ap)
2388 struct vop_pathconf_args /* {
2389 struct vnode *a_vp;
2390 int a_name;
2391 int *a_retval;
2392 } */ *ap;
2393 {
2394
2395 switch (ap->a_name) {
2396 case _PC_ACL_EXTENDED:
2397 case _PC_ACL_NFS4:
2398 case _PC_ACL_PATH_MAX:
2399 case _PC_MAC_PRESENT:
2400 return (ufs_pathconf(ap));
2401 default:
2402 return (fifo_specops.vop_pathconf(ap));
2403 }
2404 /* NOTREACHED */
2405 }
2406
2407 /*
2408 * Return POSIX pathconf information applicable to ufs filesystems.
2409 */
2410 static int
2411 ufs_pathconf(ap)
2412 struct vop_pathconf_args /* {
2413 struct vnode *a_vp;
2414 int a_name;
2415 int *a_retval;
2416 } */ *ap;
2417 {
2418 int error;
2419
2420 error = 0;
2421 switch (ap->a_name) {
2422 case _PC_LINK_MAX:
2423 *ap->a_retval = LINK_MAX;
2424 break;
2425 case _PC_NAME_MAX:
2426 *ap->a_retval = NAME_MAX;
2427 break;
2428 case _PC_PATH_MAX:
2429 *ap->a_retval = PATH_MAX;
2430 break;
2431 case _PC_PIPE_BUF:
2432 *ap->a_retval = PIPE_BUF;
2433 break;
2434 case _PC_CHOWN_RESTRICTED:
2435 *ap->a_retval = 1;
2436 break;
2437 case _PC_NO_TRUNC:
2438 *ap->a_retval = 1;
2439 break;
2440 case _PC_ACL_EXTENDED:
2441 #ifdef UFS_ACL
2442 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2443 *ap->a_retval = 1;
2444 else
2445 *ap->a_retval = 0;
2446 #else
2447 *ap->a_retval = 0;
2448 #endif
2449 break;
2450
2451 case _PC_ACL_NFS4:
2452 #ifdef UFS_ACL
2453 if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS)
2454 *ap->a_retval = 1;
2455 else
2456 *ap->a_retval = 0;
2457 #else
2458 *ap->a_retval = 0;
2459 #endif
2460 break;
2461
2462 case _PC_ACL_PATH_MAX:
2463 #ifdef UFS_ACL
2464 if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS))
2465 *ap->a_retval = ACL_MAX_ENTRIES;
2466 else
2467 *ap->a_retval = 3;
2468 #else
2469 *ap->a_retval = 3;
2470 #endif
2471 break;
2472 case _PC_MAC_PRESENT:
2473 #ifdef MAC
2474 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2475 *ap->a_retval = 1;
2476 else
2477 *ap->a_retval = 0;
2478 #else
2479 *ap->a_retval = 0;
2480 #endif
2481 break;
2482 case _PC_ASYNC_IO:
2483 /* _PC_ASYNC_IO should have been handled by upper layers. */
2484 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
2485 error = EINVAL;
2486 break;
2487 case _PC_PRIO_IO:
2488 *ap->a_retval = 0;
2489 break;
2490 case _PC_SYNC_IO:
2491 *ap->a_retval = 0;
2492 break;
2493 case _PC_ALLOC_SIZE_MIN:
2494 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2495 break;
2496 case _PC_FILESIZEBITS:
2497 *ap->a_retval = 64;
2498 break;
2499 case _PC_REC_INCR_XFER_SIZE:
2500 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2501 break;
2502 case _PC_REC_MAX_XFER_SIZE:
2503 *ap->a_retval = -1; /* means ``unlimited'' */
2504 break;
2505 case _PC_REC_MIN_XFER_SIZE:
2506 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2507 break;
2508 case _PC_REC_XFER_ALIGN:
2509 *ap->a_retval = PAGE_SIZE;
2510 break;
2511 case _PC_SYMLINK_MAX:
2512 *ap->a_retval = MAXPATHLEN;
2513 break;
2514
2515 default:
2516 error = EINVAL;
2517 break;
2518 }
2519 return (error);
2520 }
2521
2522 /*
2523 * Initialize the vnode associated with a new inode, handle aliased
2524 * vnodes.
2525 */
2526 int
2527 ufs_vinit(mntp, fifoops, vpp)
2528 struct mount *mntp;
2529 struct vop_vector *fifoops;
2530 struct vnode **vpp;
2531 {
2532 struct inode *ip;
2533 struct vnode *vp;
2534
2535 vp = *vpp;
2536 ip = VTOI(vp);
2537 vp->v_type = IFTOVT(ip->i_mode);
2538 if (vp->v_type == VFIFO)
2539 vp->v_op = fifoops;
2540 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2541 if (ip->i_number == ROOTINO)
2542 vp->v_vflag |= VV_ROOT;
2543 *vpp = vp;
2544 return (0);
2545 }
2546
2547 /*
2548 * Allocate a new inode.
2549 * Vnode dvp must be locked.
2550 */
2551 static int
2552 ufs_makeinode(mode, dvp, vpp, cnp)
2553 int mode;
2554 struct vnode *dvp;
2555 struct vnode **vpp;
2556 struct componentname *cnp;
2557 {
2558 struct inode *ip, *pdir;
2559 struct direct newdir;
2560 struct vnode *tvp;
2561 int error;
2562
2563 pdir = VTOI(dvp);
2564 #ifdef INVARIANTS
2565 if ((cnp->cn_flags & HASBUF) == 0)
2566 panic("ufs_makeinode: no name");
2567 #endif
2568 *vpp = NULL;
2569 if ((mode & IFMT) == 0)
2570 mode |= IFREG;
2571
2572 if (VTOI(dvp)->i_effnlink < 2)
2573 panic("ufs_makeinode: Bad link count %d on parent",
2574 VTOI(dvp)->i_effnlink);
2575 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2576 if (error)
2577 return (error);
2578 ip = VTOI(tvp);
2579 ip->i_gid = pdir->i_gid;
2580 DIP_SET(ip, i_gid, pdir->i_gid);
2581 #ifdef SUIDDIR
2582 {
2583 #ifdef QUOTA
2584 struct ucred ucred, *ucp;
2585 gid_t ucred_group;
2586 ucp = cnp->cn_cred;
2587 #endif
2588 /*
2589 * If we are not the owner of the directory,
2590 * and we are hacking owners here, (only do this where told to)
2591 * and we are not giving it TO root, (would subvert quotas)
2592 * then go ahead and give it to the other user.
2593 * Note that this drops off the execute bits for security.
2594 */
2595 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2596 (pdir->i_mode & ISUID) &&
2597 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2598 ip->i_uid = pdir->i_uid;
2599 DIP_SET(ip, i_uid, ip->i_uid);
2600 mode &= ~07111;
2601 #ifdef QUOTA
2602 /*
2603 * Make sure the correct user gets charged
2604 * for the space.
2605 * Quickly knock up a dummy credential for the victim.
2606 * XXX This seems to never be accessed out of our
2607 * context so a stack variable is ok.
2608 */
2609 refcount_init(&ucred.cr_ref, 1);
2610 ucred.cr_uid = ip->i_uid;
2611 ucred.cr_ngroups = 1;
2612 ucred.cr_groups = &ucred_group;
2613 ucred.cr_groups[0] = pdir->i_gid;
2614 ucp = &ucred;
2615 #endif
2616 } else {
2617 ip->i_uid = cnp->cn_cred->cr_uid;
2618 DIP_SET(ip, i_uid, ip->i_uid);
2619 }
2620
2621 #ifdef QUOTA
2622 if ((error = getinoquota(ip)) ||
2623 (error = chkiq(ip, 1, ucp, 0))) {
2624 if (DOINGSOFTDEP(tvp))
2625 softdep_revert_link(pdir, ip);
2626 UFS_VFREE(tvp, ip->i_number, mode);
2627 vput(tvp);
2628 return (error);
2629 }
2630 #endif
2631 }
2632 #else /* !SUIDDIR */
2633 ip->i_uid = cnp->cn_cred->cr_uid;
2634 DIP_SET(ip, i_uid, ip->i_uid);
2635 #ifdef QUOTA
2636 if ((error = getinoquota(ip)) ||
2637 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2638 if (DOINGSOFTDEP(tvp))
2639 softdep_revert_link(pdir, ip);
2640 UFS_VFREE(tvp, ip->i_number, mode);
2641 vput(tvp);
2642 return (error);
2643 }
2644 #endif
2645 #endif /* !SUIDDIR */
2646 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
2647 ip->i_mode = mode;
2648 DIP_SET(ip, i_mode, mode);
2649 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2650 ip->i_effnlink = 1;
2651 ip->i_nlink = 1;
2652 DIP_SET(ip, i_nlink, 1);
2653 if (DOINGSOFTDEP(tvp))
2654 softdep_setup_create(VTOI(dvp), ip);
2655 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2656 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID, 0)) {
2657 ip->i_mode &= ~ISGID;
2658 DIP_SET(ip, i_mode, ip->i_mode);
2659 }
2660
2661 if (cnp->cn_flags & ISWHITEOUT) {
2662 ip->i_flags |= UF_OPAQUE;
2663 DIP_SET(ip, i_flags, ip->i_flags);
2664 }
2665
2666 /*
2667 * Make sure inode goes to disk before directory entry.
2668 */
2669 error = UFS_UPDATE(tvp, !(DOINGSOFTDEP(tvp) | DOINGASYNC(tvp)));
2670 if (error)
2671 goto bad;
2672 #ifdef MAC
2673 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2674 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2675 dvp, tvp, cnp);
2676 if (error)
2677 goto bad;
2678 }
2679 #endif
2680 #ifdef UFS_ACL
2681 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2682 error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2683 cnp->cn_cred, cnp->cn_thread);
2684 if (error)
2685 goto bad;
2686 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2687 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode,
2688 cnp->cn_cred, cnp->cn_thread);
2689 if (error)
2690 goto bad;
2691 }
2692 #endif /* !UFS_ACL */
2693 ufs_makedirentry(ip, cnp, &newdir);
2694 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL, 0);
2695 if (error)
2696 goto bad;
2697 *vpp = tvp;
2698 return (0);
2699
2700 bad:
2701 /*
2702 * Write error occurred trying to update the inode
2703 * or the directory so must deallocate the inode.
2704 */
2705 ip->i_effnlink = 0;
2706 ip->i_nlink = 0;
2707 DIP_SET(ip, i_nlink, 0);
2708 ip->i_flag |= IN_CHANGE;
2709 if (DOINGSOFTDEP(tvp))
2710 softdep_revert_create(VTOI(dvp), ip);
2711 vput(tvp);
2712 return (error);
2713 }
2714
2715 /* Global vfs data structures for ufs. */
2716 struct vop_vector ufs_vnodeops = {
2717 .vop_default = &default_vnodeops,
2718 .vop_fsync = VOP_PANIC,
2719 .vop_read = VOP_PANIC,
2720 .vop_reallocblks = VOP_PANIC,
2721 .vop_write = VOP_PANIC,
2722 .vop_accessx = ufs_accessx,
2723 .vop_bmap = ufs_bmap,
2724 .vop_cachedlookup = ufs_lookup,
2725 .vop_close = ufs_close,
2726 .vop_create = ufs_create,
2727 .vop_getattr = ufs_getattr,
2728 .vop_inactive = ufs_inactive,
2729 .vop_link = ufs_link,
2730 .vop_lookup = vfs_cache_lookup,
2731 .vop_markatime = ufs_markatime,
2732 .vop_mkdir = ufs_mkdir,
2733 .vop_mknod = ufs_mknod,
2734 .vop_open = ufs_open,
2735 .vop_pathconf = ufs_pathconf,
2736 .vop_poll = vop_stdpoll,
2737 .vop_print = ufs_print,
2738 .vop_readdir = ufs_readdir,
2739 .vop_readlink = ufs_readlink,
2740 .vop_reclaim = ufs_reclaim,
2741 .vop_remove = ufs_remove,
2742 .vop_rename = ufs_rename,
2743 .vop_rmdir = ufs_rmdir,
2744 .vop_setattr = ufs_setattr,
2745 #ifdef MAC
2746 .vop_setlabel = vop_stdsetlabel_ea,
2747 #endif
2748 .vop_strategy = ufs_strategy,
2749 .vop_symlink = ufs_symlink,
2750 .vop_whiteout = ufs_whiteout,
2751 #ifdef UFS_EXTATTR
2752 .vop_getextattr = ufs_getextattr,
2753 .vop_deleteextattr = ufs_deleteextattr,
2754 .vop_setextattr = ufs_setextattr,
2755 #endif
2756 #ifdef UFS_ACL
2757 .vop_getacl = ufs_getacl,
2758 .vop_setacl = ufs_setacl,
2759 .vop_aclcheck = ufs_aclcheck,
2760 #endif
2761 };
2762
2763 struct vop_vector ufs_fifoops = {
2764 .vop_default = &fifo_specops,
2765 .vop_fsync = VOP_PANIC,
2766 .vop_accessx = ufs_accessx,
2767 .vop_close = ufsfifo_close,
2768 .vop_getattr = ufs_getattr,
2769 .vop_inactive = ufs_inactive,
2770 .vop_kqfilter = ufsfifo_kqfilter,
2771 .vop_markatime = ufs_markatime,
2772 .vop_pathconf = ufsfifo_pathconf,
2773 .vop_print = ufs_print,
2774 .vop_read = VOP_PANIC,
2775 .vop_reclaim = ufs_reclaim,
2776 .vop_setattr = ufs_setattr,
2777 #ifdef MAC
2778 .vop_setlabel = vop_stdsetlabel_ea,
2779 #endif
2780 .vop_write = VOP_PANIC,
2781 #ifdef UFS_EXTATTR
2782 .vop_getextattr = ufs_getextattr,
2783 .vop_deleteextattr = ufs_deleteextattr,
2784 .vop_setextattr = ufs_setextattr,
2785 #endif
2786 #ifdef UFS_ACL
2787 .vop_getacl = ufs_getacl,
2788 .vop_setacl = ufs_setacl,
2789 .vop_aclcheck = ufs_aclcheck,
2790 #endif
2791 };
Cache object: 96d93f265d1ff12d98d8d3a20b097858
|