1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #include <sys/cdefs.h>
38 #include "opt_quota.h"
39 #include "opt_suiddir.h"
40 #include "opt_ufs.h"
41 #include "opt_ffs.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/malloc.h>
46 #include <sys/namei.h>
47 #include <sys/kernel.h>
48 #include <sys/fcntl.h>
49 #include <sys/filio.h>
50 #include <sys/stat.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/mount.h>
54 #include <sys/priv.h>
55 #include <sys/refcount.h>
56 #include <sys/unistd.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/lockf.h>
60 #include <sys/conf.h>
61 #include <sys/acl.h>
62 #include <sys/smr.h>
63
64 #include <security/audit/audit.h>
65 #include <security/mac/mac_framework.h>
66
67 #include <sys/file.h> /* XXX */
68
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71
72 #include <ufs/ufs/acl.h>
73 #include <ufs/ufs/extattr.h>
74 #include <ufs/ufs/quota.h>
75 #include <ufs/ufs/inode.h>
76 #include <ufs/ufs/dir.h>
77 #include <ufs/ufs/ufsmount.h>
78 #include <ufs/ufs/ufs_extern.h>
79 #ifdef UFS_DIRHASH
80 #include <ufs/ufs/dirhash.h>
81 #endif
82 #ifdef UFS_GJOURNAL
83 #include <ufs/ufs/gjournal.h>
84 FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS");
85 #endif
86
87 #ifdef QUOTA
88 FEATURE(ufs_quota, "UFS disk quotas support");
89 FEATURE(ufs_quota64, "64bit UFS disk quotas support");
90 #endif
91
92 #ifdef SUIDDIR
93 FEATURE(suiddir,
94 "Give all new files in directory the same ownership as the directory");
95 #endif
96
97 VFS_SMR_DECLARE;
98
99 #include <ufs/ffs/ffs_extern.h>
100
101 static vop_accessx_t ufs_accessx;
102 vop_fplookup_vexec_t ufs_fplookup_vexec;
103 static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *);
104 static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
105 struct thread *);
106 static vop_close_t ufs_close;
107 static vop_create_t ufs_create;
108 static vop_stat_t ufs_stat;
109 static vop_getattr_t ufs_getattr;
110 static vop_ioctl_t ufs_ioctl;
111 static vop_link_t ufs_link;
112 static int ufs_makeinode(int mode, struct vnode *, struct vnode **,
113 struct componentname *, const char *);
114 static vop_mmapped_t ufs_mmapped;
115 static vop_mkdir_t ufs_mkdir;
116 static vop_mknod_t ufs_mknod;
117 static vop_open_t ufs_open;
118 static vop_pathconf_t ufs_pathconf;
119 static vop_print_t ufs_print;
120 static vop_readlink_t ufs_readlink;
121 static vop_remove_t ufs_remove;
122 static vop_rename_t ufs_rename;
123 static vop_rmdir_t ufs_rmdir;
124 static vop_setattr_t ufs_setattr;
125 static vop_strategy_t ufs_strategy;
126 static vop_symlink_t ufs_symlink;
127 static vop_whiteout_t ufs_whiteout;
128 static vop_close_t ufsfifo_close;
129
130 SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
131 "UFS filesystem");
132
133 /*
134 * A virgin directory (no blushing please).
135 */
136 static struct dirtemplate mastertemplate = {
137 0, 12, DT_DIR, 1, ".",
138 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
139 };
140 static struct odirtemplate omastertemplate = {
141 0, 12, 1, ".",
142 0, DIRBLKSIZ - 12, 2, ".."
143 };
144
145 static void
ufs_itimes_locked(struct vnode * vp)146 ufs_itimes_locked(struct vnode *vp)
147 {
148 struct inode *ip;
149 struct timespec ts;
150
151 ASSERT_VI_LOCKED(vp, __func__);
152
153 ip = VTOI(vp);
154 if (UFS_RDONLY(ip))
155 goto out;
156 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
157 return;
158
159 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
160 UFS_INODE_SET_FLAG(ip, IN_LAZYMOD);
161 else if (((vp->v_mount->mnt_kern_flag &
162 (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) ||
163 (ip->i_flag & (IN_CHANGE | IN_UPDATE)))
164 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
165 else if (ip->i_flag & IN_ACCESS)
166 UFS_INODE_SET_FLAG(ip, IN_LAZYACCESS);
167 vfs_timestamp(&ts);
168 if (ip->i_flag & IN_ACCESS) {
169 DIP_SET(ip, i_atime, ts.tv_sec);
170 DIP_SET(ip, i_atimensec, ts.tv_nsec);
171 }
172 if (ip->i_flag & IN_UPDATE) {
173 DIP_SET(ip, i_mtime, ts.tv_sec);
174 DIP_SET(ip, i_mtimensec, ts.tv_nsec);
175 }
176 if (ip->i_flag & IN_CHANGE) {
177 DIP_SET(ip, i_ctime, ts.tv_sec);
178 DIP_SET(ip, i_ctimensec, ts.tv_nsec);
179 DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1);
180 }
181
182 out:
183 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
184 }
185
186 void
ufs_itimes(struct vnode * vp)187 ufs_itimes(struct vnode *vp)
188 {
189 struct inode *ip;
190
191 ip = VTOI(vp);
192 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
193 return;
194
195 VI_LOCK(vp);
196 ufs_itimes_locked(vp);
197 VI_UNLOCK(vp);
198 }
199
200 static int
ufs_sync_nlink1(struct mount * mp)201 ufs_sync_nlink1(struct mount *mp)
202 {
203 int error;
204
205 error = vfs_busy(mp, 0);
206 if (error == 0) {
207 VFS_SYNC(mp, MNT_WAIT);
208 vfs_unbusy(mp);
209 error = ERELOOKUP;
210 }
211 vfs_rel(mp);
212 return (error);
213 }
214
215 static int
ufs_sync_nlink(struct vnode * vp,struct vnode * vp1)216 ufs_sync_nlink(struct vnode *vp, struct vnode *vp1)
217 {
218 struct inode *ip;
219 struct mount *mp;
220 int error;
221
222 ip = VTOI(vp);
223 if (ip->i_nlink < UFS_LINK_MAX)
224 return (0);
225 if (!DOINGSOFTDEP(vp) || ip->i_effnlink >= UFS_LINK_MAX)
226 return (EMLINK);
227
228 mp = vp->v_mount;
229 vfs_ref(mp);
230 VOP_UNLOCK(vp);
231 if (vp1 != NULL)
232 VOP_UNLOCK(vp1);
233 error = ufs_sync_nlink1(mp);
234 vn_lock_pair(vp, false, LK_EXCLUSIVE, vp1, false, LK_EXCLUSIVE);
235 return (error);
236 }
237
238 /*
239 * Create a regular file
240 */
241 static int
ufs_create(struct vop_create_args * ap)242 ufs_create(
243 struct vop_create_args /* {
244 struct vnode *a_dvp;
245 struct vnode **a_vpp;
246 struct componentname *a_cnp;
247 struct vattr *a_vap;
248 } */ *ap)
249 {
250 int error;
251
252 error =
253 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
254 ap->a_dvp, ap->a_vpp, ap->a_cnp, "ufs_create");
255 if (error != 0)
256 return (error);
257 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
258 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
259 return (0);
260 }
261
262 /*
263 * Mknod vnode call
264 */
265 /* ARGSUSED */
266 static int
ufs_mknod(struct vop_mknod_args * ap)267 ufs_mknod(
268 struct vop_mknod_args /* {
269 struct vnode *a_dvp;
270 struct vnode **a_vpp;
271 struct componentname *a_cnp;
272 struct vattr *a_vap;
273 } */ *ap)
274 {
275 struct vattr *vap = ap->a_vap;
276 struct vnode **vpp = ap->a_vpp;
277 struct inode *ip;
278 ino_t ino;
279 int error;
280
281 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
282 ap->a_dvp, vpp, ap->a_cnp, "ufs_mknod");
283 if (error)
284 return (error);
285 ip = VTOI(*vpp);
286 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
287 if (vap->va_rdev != VNOVAL) {
288 /*
289 * Want to be able to use this to make badblock
290 * inodes, so don't truncate the dev number.
291 */
292 DIP_SET(ip, i_rdev, vap->va_rdev);
293 }
294 /*
295 * Remove inode, then reload it through VFS_VGET(). This is
296 * needed to do further inode initialization, for instance
297 * fifo, which was too early for VFS_VGET() done as part of
298 * UFS_VALLOC().
299 */
300 (*vpp)->v_type = VNON;
301 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
302 vgone(*vpp);
303 vput(*vpp);
304 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
305 if (error) {
306 *vpp = NULL;
307 return (error);
308 }
309 return (0);
310 }
311
312 /*
313 * Open called.
314 */
315 /* ARGSUSED */
316 static int
ufs_open(struct vop_open_args * ap)317 ufs_open(struct vop_open_args *ap)
318 {
319 struct vnode *vp = ap->a_vp;
320 struct inode *ip;
321
322 if (vp->v_type == VCHR || vp->v_type == VBLK)
323 return (EOPNOTSUPP);
324
325 ip = VTOI(vp);
326 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td);
327 if (vp->v_type == VREG && (vn_irflag_read(vp) & VIRF_PGREAD) == 0 &&
328 ip->i_ump->um_bsize >= PAGE_SIZE) {
329 vn_irflag_set_cond(vp, VIRF_PGREAD);
330 }
331
332 /*
333 * Files marked append-only must be opened for appending.
334 */
335 if ((ip->i_flags & APPEND) &&
336 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
337 return (EPERM);
338
339 return (0);
340 }
341
342 /*
343 * Close called.
344 *
345 * Update the times on the inode.
346 */
347 /* ARGSUSED */
348 static int
ufs_close(struct vop_close_args * ap)349 ufs_close(
350 struct vop_close_args /* {
351 struct vnode *a_vp;
352 int a_fflag;
353 struct ucred *a_cred;
354 struct thread *a_td;
355 } */ *ap)
356 {
357 struct vnode *vp = ap->a_vp;
358
359 ufs_itimes(vp);
360 return (0);
361 }
362
363 static int
ufs_accessx(struct vop_accessx_args * ap)364 ufs_accessx(
365 struct vop_accessx_args /* {
366 struct vnode *a_vp;
367 accmode_t a_accmode;
368 struct ucred *a_cred;
369 struct thread *a_td;
370 } */ *ap)
371 {
372 struct vnode *vp = ap->a_vp;
373 struct inode *ip = VTOI(vp);
374 accmode_t accmode = ap->a_accmode;
375 int error;
376 #ifdef UFS_ACL
377 struct acl *acl;
378 acl_type_t type;
379 #endif
380
381 /*
382 * Disallow write attempts on read-only filesystems;
383 * unless the file is a socket, fifo, or a block or
384 * character device resident on the filesystem.
385 */
386 if (accmode & VMODIFY_PERMS) {
387 switch (vp->v_type) {
388 case VDIR:
389 case VLNK:
390 case VREG:
391 if (vp->v_mount->mnt_flag & MNT_RDONLY)
392 return (EROFS);
393 #ifdef QUOTA
394 /*
395 * Inode is accounted in the quotas only if struct
396 * dquot is attached to it. VOP_ACCESS() is called
397 * from vn_open_cred() and provides a convenient
398 * point to call getinoquota(). The lock mode is
399 * exclusive when the file is opening for write.
400 */
401 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
402 error = getinoquota(ip);
403 if (error != 0)
404 return (error);
405 }
406 #endif
407 break;
408 default:
409 break;
410 }
411 }
412
413 /*
414 * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS"
415 * permits the owner of the file to remove the IMMUTABLE flag.
416 */
417 if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) &&
418 (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
419 return (EPERM);
420
421 #ifdef UFS_ACL
422 if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) {
423 if (vp->v_mount->mnt_flag & MNT_NFS4ACLS)
424 type = ACL_TYPE_NFS4;
425 else
426 type = ACL_TYPE_ACCESS;
427
428 acl = acl_alloc(M_WAITOK);
429 if (type == ACL_TYPE_NFS4)
430 error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td);
431 else
432 error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td);
433 switch (error) {
434 case 0:
435 if (type == ACL_TYPE_NFS4) {
436 error = vaccess_acl_nfs4(vp->v_type, ip->i_uid,
437 ip->i_gid, acl, accmode, ap->a_cred);
438 } else {
439 error = vfs_unixify_accmode(&accmode);
440 if (error == 0)
441 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid,
442 ip->i_gid, acl, accmode, ap->a_cred);
443 }
444 break;
445 default:
446 if (error != EOPNOTSUPP)
447 printf(
448 "ufs_accessx(): Error retrieving ACL on object (%d).\n",
449 error);
450 /*
451 * XXX: Fall back until debugged. Should
452 * eventually possibly log an error, and return
453 * EPERM for safety.
454 */
455 error = vfs_unixify_accmode(&accmode);
456 if (error == 0)
457 error = vaccess(vp->v_type, ip->i_mode,
458 ip->i_uid, ip->i_gid, accmode, ap->a_cred);
459 }
460 acl_free(acl);
461
462 return (error);
463 }
464 #endif /* !UFS_ACL */
465 error = vfs_unixify_accmode(&accmode);
466 if (error == 0)
467 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
468 accmode, ap->a_cred);
469 return (error);
470 }
471
472 /*
473 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
474 * the comment above cache_fplookup for details.
475 */
476 int
ufs_fplookup_vexec(struct vop_fplookup_vexec_args * ap)477 ufs_fplookup_vexec(
478 struct vop_fplookup_vexec_args /* {
479 struct vnode *a_vp;
480 struct ucred *a_cred;
481 struct thread *a_td;
482 } */ *ap)
483 {
484 struct vnode *vp;
485 struct inode *ip;
486 struct ucred *cred;
487 mode_t all_x, mode;
488
489 vp = ap->a_vp;
490 ip = VTOI_SMR(vp);
491 if (__predict_false(ip == NULL))
492 return (EAGAIN);
493
494 /*
495 * XXX ACL race
496 *
497 * ACLs are not supported and UFS clears/sets this flag on mount and
498 * remount. However, we may still be racing with seeing them and there
499 * is no provision to make sure they were accounted for. This matches
500 * the behavior of the locked case, since the lookup there is also
501 * racy: mount takes no measures to block anyone from progressing.
502 */
503 all_x = S_IXUSR | S_IXGRP | S_IXOTH;
504 mode = atomic_load_short(&ip->i_mode);
505 if (__predict_true((mode & all_x) == all_x))
506 return (0);
507
508 cred = ap->a_cred;
509 return (vaccess_vexec_smr(mode, ip->i_uid, ip->i_gid, cred));
510 }
511
512 /* ARGSUSED */
513 static int
ufs_stat(struct vop_stat_args * ap)514 ufs_stat(struct vop_stat_args *ap)
515 {
516 struct vnode *vp = ap->a_vp;
517 struct inode *ip = VTOI(vp);
518 struct stat *sb = ap->a_sb;
519 int error;
520
521 error = vop_stat_helper_pre(ap);
522 if (__predict_false(error))
523 return (error);
524
525 VI_LOCK(vp);
526 ufs_itimes_locked(vp);
527 if (I_IS_UFS1(ip)) {
528 sb->st_atim.tv_sec = ip->i_din1->di_atime;
529 sb->st_atim.tv_nsec = ip->i_din1->di_atimensec;
530 } else {
531 sb->st_atim.tv_sec = ip->i_din2->di_atime;
532 sb->st_atim.tv_nsec = ip->i_din2->di_atimensec;
533 }
534 VI_UNLOCK(vp);
535
536 sb->st_dev = dev2udev(ITOUMP(ip)->um_dev);
537 sb->st_ino = ip->i_number;
538 sb->st_mode = (ip->i_mode & ~IFMT) | VTTOIF(vp->v_type);
539 sb->st_nlink = ip->i_effnlink;
540 sb->st_uid = ip->i_uid;
541 sb->st_gid = ip->i_gid;
542 if (I_IS_UFS1(ip)) {
543 sb->st_rdev = ip->i_din1->di_rdev;
544 sb->st_size = ip->i_din1->di_size;
545 sb->st_mtim.tv_sec = ip->i_din1->di_mtime;
546 sb->st_mtim.tv_nsec = ip->i_din1->di_mtimensec;
547 sb->st_ctim.tv_sec = ip->i_din1->di_ctime;
548 sb->st_ctim.tv_nsec = ip->i_din1->di_ctimensec;
549 sb->st_birthtim.tv_sec = -1;
550 sb->st_birthtim.tv_nsec = 0;
551 sb->st_blocks = dbtob((uint64_t)ip->i_din1->di_blocks) / S_BLKSIZE;
552 } else {
553 sb->st_rdev = ip->i_din2->di_rdev;
554 sb->st_size = ip->i_din2->di_size;
555 sb->st_mtim.tv_sec = ip->i_din2->di_mtime;
556 sb->st_mtim.tv_nsec = ip->i_din2->di_mtimensec;
557 sb->st_ctim.tv_sec = ip->i_din2->di_ctime;
558 sb->st_ctim.tv_nsec = ip->i_din2->di_ctimensec;
559 sb->st_birthtim.tv_sec = ip->i_din2->di_birthtime;
560 sb->st_birthtim.tv_nsec = ip->i_din2->di_birthnsec;
561 sb->st_blocks = dbtob((uint64_t)ip->i_din2->di_blocks) / S_BLKSIZE;
562 }
563
564 sb->st_blksize = max(PAGE_SIZE, vp->v_mount->mnt_stat.f_iosize);
565 sb->st_flags = ip->i_flags;
566 sb->st_gen = ip->i_gen;
567
568 return (vop_stat_helper_post(ap, error));
569 }
570
571 /* ARGSUSED */
572 static int
ufs_getattr(struct vop_getattr_args * ap)573 ufs_getattr(
574 struct vop_getattr_args /* {
575 struct vnode *a_vp;
576 struct vattr *a_vap;
577 struct ucred *a_cred;
578 } */ *ap)
579 {
580 struct vnode *vp = ap->a_vp;
581 struct inode *ip = VTOI(vp);
582 struct vattr *vap = ap->a_vap;
583
584 VI_LOCK(vp);
585 ufs_itimes_locked(vp);
586 if (I_IS_UFS1(ip)) {
587 vap->va_atime.tv_sec = ip->i_din1->di_atime;
588 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
589 } else {
590 vap->va_atime.tv_sec = ip->i_din2->di_atime;
591 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
592 }
593 VI_UNLOCK(vp);
594 /*
595 * Copy from inode table
596 */
597 vap->va_fsid = dev2udev(ITOUMP(ip)->um_dev);
598 vap->va_fileid = ip->i_number;
599 vap->va_mode = ip->i_mode & ~IFMT;
600 vap->va_nlink = ip->i_effnlink;
601 vap->va_uid = ip->i_uid;
602 vap->va_gid = ip->i_gid;
603 if (I_IS_UFS1(ip)) {
604 vap->va_rdev = ip->i_din1->di_rdev;
605 vap->va_size = ip->i_din1->di_size;
606 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
607 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
608 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
609 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
610 vap->va_bytes = dbtob((uint64_t)ip->i_din1->di_blocks);
611 vap->va_filerev = ip->i_din1->di_modrev;
612 } else {
613 vap->va_rdev = ip->i_din2->di_rdev;
614 vap->va_size = ip->i_din2->di_size;
615 vap->va_mtime.tv_sec = ip->i_din2->di_mtime;
616 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec;
617 vap->va_ctime.tv_sec = ip->i_din2->di_ctime;
618 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec;
619 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime;
620 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec;
621 vap->va_bytes = dbtob((uint64_t)ip->i_din2->di_blocks);
622 vap->va_filerev = ip->i_din2->di_modrev;
623 }
624 vap->va_flags = ip->i_flags;
625 vap->va_gen = ip->i_gen;
626 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
627 vap->va_type = IFTOVT(ip->i_mode);
628 return (0);
629 }
630
631 /*
632 * Set attribute vnode op. called from several syscalls
633 */
634 static int
ufs_setattr(struct vop_setattr_args * ap)635 ufs_setattr(
636 struct vop_setattr_args /* {
637 struct vnode *a_vp;
638 struct vattr *a_vap;
639 struct ucred *a_cred;
640 } */ *ap)
641 {
642 struct vattr *vap = ap->a_vap;
643 struct vnode *vp = ap->a_vp;
644 struct inode *ip = VTOI(vp);
645 struct ucred *cred = ap->a_cred;
646 struct thread *td = curthread;
647 int error;
648
649 /*
650 * Check for unsettable attributes.
651 */
652 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
653 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
654 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
655 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
656 return (EINVAL);
657 }
658 if (vap->va_flags != VNOVAL) {
659 if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE |
660 SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE |
661 UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK |
662 UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
663 UF_SPARSE | UF_SYSTEM)) != 0)
664 return (EOPNOTSUPP);
665 if (vp->v_mount->mnt_flag & MNT_RDONLY)
666 return (EROFS);
667 /*
668 * Callers may only modify the file flags on objects they
669 * have VADMIN rights for.
670 */
671 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
672 return (error);
673 /*
674 * Unprivileged processes are not permitted to unset system
675 * flags, or modify flags if any system flags are set.
676 * Privileged non-jail processes may not modify system flags
677 * if securelevel > 0 and any existing system flags are set.
678 * Privileged jail processes behave like privileged non-jail
679 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
680 * otherwise, they behave like unprivileged processes.
681 */
682 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
683 if (ip->i_flags &
684 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
685 error = securelevel_gt(cred, 0);
686 if (error)
687 return (error);
688 }
689 /* The snapshot flag cannot be toggled. */
690 if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT)
691 return (EPERM);
692 } else {
693 if (ip->i_flags &
694 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
695 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
696 return (EPERM);
697 }
698 ip->i_flags = vap->va_flags;
699 DIP_SET(ip, i_flags, vap->va_flags);
700 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
701 error = UFS_UPDATE(vp, 0);
702 if (ip->i_flags & (IMMUTABLE | APPEND))
703 return (error);
704 }
705 /*
706 * If immutable or append, no one can change any of its attributes
707 * except the ones already handled (in some cases, file flags
708 * including the immutability flags themselves for the superuser).
709 */
710 if (ip->i_flags & (IMMUTABLE | APPEND))
711 return (EPERM);
712 /*
713 * Go through the fields and update iff not VNOVAL.
714 */
715 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
716 if (vp->v_mount->mnt_flag & MNT_RDONLY)
717 return (EROFS);
718 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
719 td)) != 0)
720 return (error);
721 }
722 if (vap->va_size != VNOVAL) {
723 /*
724 * XXX most of the following special cases should be in
725 * callers instead of in N filesystems. The VDIR check
726 * mostly already is.
727 */
728 switch (vp->v_type) {
729 case VDIR:
730 return (EISDIR);
731 case VLNK:
732 case VREG:
733 /*
734 * Truncation should have an effect in these cases.
735 * Disallow it if the filesystem is read-only or
736 * the file is being snapshotted.
737 */
738 if (vp->v_mount->mnt_flag & MNT_RDONLY)
739 return (EROFS);
740 if (IS_SNAPSHOT(ip))
741 return (EPERM);
742 break;
743 default:
744 /*
745 * According to POSIX, the result is unspecified
746 * for file types other than regular files,
747 * directories and shared memory objects. We
748 * don't support shared memory objects in the file
749 * system, and have dubious support for truncating
750 * symlinks. Just ignore the request in other cases.
751 */
752 return (0);
753 }
754 error = vn_rlimit_trunc(vap->va_size, td);
755 if (error != 0)
756 return (error);
757 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL |
758 ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0),
759 cred)) != 0)
760 return (error);
761 }
762 if (vap->va_atime.tv_sec != VNOVAL ||
763 vap->va_mtime.tv_sec != VNOVAL ||
764 vap->va_birthtime.tv_sec != VNOVAL) {
765 if (vp->v_mount->mnt_flag & MNT_RDONLY)
766 return (EROFS);
767 if (IS_SNAPSHOT(ip))
768 return (EPERM);
769 error = vn_utimes_perm(vp, vap, cred, td);
770 if (error != 0)
771 return (error);
772 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED);
773 if (vap->va_atime.tv_sec != VNOVAL) {
774 ip->i_flag &= ~IN_ACCESS;
775 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
776 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
777 }
778 if (vap->va_mtime.tv_sec != VNOVAL) {
779 ip->i_flag &= ~IN_UPDATE;
780 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
781 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
782 }
783 if (vap->va_birthtime.tv_sec != VNOVAL && I_IS_UFS2(ip)) {
784 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
785 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
786 }
787 error = UFS_UPDATE(vp, 0);
788 if (error)
789 return (error);
790 }
791 error = 0;
792 if (vap->va_mode != (mode_t)VNOVAL) {
793 if (vp->v_mount->mnt_flag & MNT_RDONLY)
794 return (EROFS);
795 if (IS_SNAPSHOT(ip) && (vap->va_mode & (S_IXUSR | S_IWUSR |
796 S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)) != 0)
797 return (EPERM);
798 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
799 }
800 return (error);
801 }
802
803 #ifdef UFS_ACL
804 static int
ufs_update_nfs4_acl_after_mode_change(struct vnode * vp,int mode,int file_owner_id,struct ucred * cred,struct thread * td)805 ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode,
806 int file_owner_id, struct ucred *cred, struct thread *td)
807 {
808 int error;
809 struct acl *aclp;
810
811 aclp = acl_alloc(M_WAITOK);
812 error = ufs_getacl_nfs4_internal(vp, aclp, td);
813 /*
814 * We don't have to handle EOPNOTSUPP here, as the filesystem claims
815 * it supports ACLs.
816 */
817 if (error)
818 goto out;
819
820 acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id);
821 error = ufs_setacl_nfs4_internal(vp, aclp, td);
822
823 out:
824 acl_free(aclp);
825 return (error);
826 }
827 #endif /* UFS_ACL */
828
829 static int
ufs_mmapped(struct vop_mmapped_args * ap)830 ufs_mmapped(
831 struct vop_mmapped_args /* {
832 struct vnode *a_vp;
833 } */ *ap)
834 {
835 struct vnode *vp;
836 struct inode *ip;
837 struct mount *mp;
838
839 vp = ap->a_vp;
840 ip = VTOI(vp);
841 mp = vp->v_mount;
842
843 if ((mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
844 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
845 /*
846 * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there.
847 */
848 return (0);
849 }
850
851 /*
852 * Change the mode on a file.
853 * Inode must be locked before calling.
854 */
855 static int
ufs_chmod(struct vnode * vp,int mode,struct ucred * cred,struct thread * td)856 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
857 {
858 struct inode *ip = VTOI(vp);
859 int newmode, error;
860
861 /*
862 * To modify the permissions on a file, must possess VADMIN
863 * for that file.
864 */
865 if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td)))
866 return (error);
867 /*
868 * Privileged processes may set the sticky bit on non-directories,
869 * as well as set the setgid bit on a file with a group that the
870 * process is not a member of. Both of these are allowed in
871 * jail(8).
872 */
873 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
874 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
875 return (EFTYPE);
876 }
877 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
878 error = priv_check_cred(cred, PRIV_VFS_SETGID);
879 if (error)
880 return (error);
881 }
882
883 /*
884 * Deny setting setuid if we are not the file owner.
885 */
886 if ((mode & ISUID) && ip->i_uid != cred->cr_uid) {
887 error = priv_check_cred(cred, PRIV_VFS_ADMIN);
888 if (error)
889 return (error);
890 }
891
892 newmode = ip->i_mode & ~ALLPERMS;
893 newmode |= (mode & ALLPERMS);
894 UFS_INODE_SET_MODE(ip, newmode);
895 DIP_SET(ip, i_mode, ip->i_mode);
896 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
897 #ifdef UFS_ACL
898 if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0)
899 error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td);
900 #endif
901 if (error == 0 && (ip->i_flag & IN_CHANGE) != 0)
902 error = UFS_UPDATE(vp, 0);
903
904 return (error);
905 }
906
907 /*
908 * Perform chown operation on inode ip;
909 * inode must be locked prior to call.
910 */
911 static int
ufs_chown(struct vnode * vp,uid_t uid,gid_t gid,struct ucred * cred,struct thread * td)912 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
913 struct thread *td)
914 {
915 struct inode *ip = VTOI(vp);
916 uid_t ouid;
917 gid_t ogid;
918 int error = 0;
919 #ifdef QUOTA
920 int i;
921 ufs2_daddr_t change;
922 #endif
923
924 if (uid == (uid_t)VNOVAL)
925 uid = ip->i_uid;
926 if (gid == (gid_t)VNOVAL)
927 gid = ip->i_gid;
928 /*
929 * To modify the ownership of a file, must possess VADMIN for that
930 * file.
931 */
932 if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td)))
933 return (error);
934 /*
935 * To change the owner of a file, or change the group of a file to a
936 * group of which we are not a member, the caller must have
937 * privilege.
938 */
939 if (((uid != ip->i_uid && uid != cred->cr_uid) ||
940 (gid != ip->i_gid && !groupmember(gid, cred))) &&
941 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
942 return (error);
943 ogid = ip->i_gid;
944 ouid = ip->i_uid;
945 #ifdef QUOTA
946 if ((error = getinoquota(ip)) != 0)
947 return (error);
948 if (ouid == uid) {
949 dqrele(vp, ip->i_dquot[USRQUOTA]);
950 ip->i_dquot[USRQUOTA] = NODQUOT;
951 }
952 if (ogid == gid) {
953 dqrele(vp, ip->i_dquot[GRPQUOTA]);
954 ip->i_dquot[GRPQUOTA] = NODQUOT;
955 }
956 change = DIP(ip, i_blocks);
957 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
958 (void) chkiq(ip, -1, cred, CHOWN|FORCE);
959 for (i = 0; i < MAXQUOTAS; i++) {
960 dqrele(vp, ip->i_dquot[i]);
961 ip->i_dquot[i] = NODQUOT;
962 }
963 #endif
964 ip->i_gid = gid;
965 DIP_SET(ip, i_gid, gid);
966 ip->i_uid = uid;
967 DIP_SET(ip, i_uid, uid);
968 #ifdef QUOTA
969 if ((error = getinoquota(ip)) == 0) {
970 if (ouid == uid) {
971 dqrele(vp, ip->i_dquot[USRQUOTA]);
972 ip->i_dquot[USRQUOTA] = NODQUOT;
973 }
974 if (ogid == gid) {
975 dqrele(vp, ip->i_dquot[GRPQUOTA]);
976 ip->i_dquot[GRPQUOTA] = NODQUOT;
977 }
978 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
979 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
980 goto good;
981 else
982 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
983 }
984 for (i = 0; i < MAXQUOTAS; i++) {
985 dqrele(vp, ip->i_dquot[i]);
986 ip->i_dquot[i] = NODQUOT;
987 }
988 }
989 ip->i_gid = ogid;
990 DIP_SET(ip, i_gid, ogid);
991 ip->i_uid = ouid;
992 DIP_SET(ip, i_uid, ouid);
993 if (getinoquota(ip) == 0) {
994 if (ouid == uid) {
995 dqrele(vp, ip->i_dquot[USRQUOTA]);
996 ip->i_dquot[USRQUOTA] = NODQUOT;
997 }
998 if (ogid == gid) {
999 dqrele(vp, ip->i_dquot[GRPQUOTA]);
1000 ip->i_dquot[GRPQUOTA] = NODQUOT;
1001 }
1002 (void) chkdq(ip, change, cred, FORCE|CHOWN);
1003 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
1004 (void) getinoquota(ip);
1005 }
1006 return (error);
1007 good:
1008 if (getinoquota(ip))
1009 panic("ufs_chown: lost quota");
1010 #endif /* QUOTA */
1011 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1012 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
1013 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1014 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1015 DIP_SET(ip, i_mode, ip->i_mode);
1016 }
1017 }
1018 error = UFS_UPDATE(vp, 0);
1019 return (error);
1020 }
1021
1022 static int
ufs_remove(struct vop_remove_args * ap)1023 ufs_remove(
1024 struct vop_remove_args /* {
1025 struct vnode *a_dvp;
1026 struct vnode *a_vp;
1027 struct componentname *a_cnp;
1028 } */ *ap)
1029 {
1030 struct inode *ip;
1031 struct vnode *vp = ap->a_vp;
1032 struct vnode *dvp = ap->a_dvp;
1033 int error;
1034 struct thread *td;
1035
1036 td = curthread;
1037 ip = VTOI(vp);
1038 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1039 (VTOI(dvp)->i_flags & APPEND))
1040 return (EPERM);
1041 if (DOINGSUJ(dvp)) {
1042 error = softdep_prelink(dvp, vp, ap->a_cnp);
1043 if (error != 0) {
1044 MPASS(error == ERELOOKUP);
1045 return (error);
1046 }
1047 }
1048
1049 #ifdef UFS_GJOURNAL
1050 ufs_gjournal_orphan(vp);
1051 #endif
1052 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
1053 if (ip->i_nlink <= 0)
1054 vp->v_vflag |= VV_NOSYNC;
1055 if (IS_SNAPSHOT(ip)) {
1056 /*
1057 * Avoid deadlock where another thread is trying to
1058 * update the inodeblock for dvp and is waiting on
1059 * snaplk. Temporary unlock the vnode lock for the
1060 * unlinked file and sync the directory. This should
1061 * allow vput() of the directory to not block later on
1062 * while holding the snapshot vnode locked, assuming
1063 * that the directory hasn't been unlinked too.
1064 */
1065 VOP_UNLOCK(vp);
1066 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
1067 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1068 }
1069 return (error);
1070 }
1071
1072 static void
print_bad_link_count(const char * funcname,struct vnode * dvp)1073 print_bad_link_count(const char *funcname, struct vnode *dvp)
1074 {
1075 struct inode *dip;
1076
1077 dip = VTOI(dvp);
1078 uprintf("%s: Bad link count %d on parent inode %jd in file system %s\n",
1079 funcname, dip->i_effnlink, (intmax_t)dip->i_number,
1080 dvp->v_mount->mnt_stat.f_mntonname);
1081 }
1082
1083 /*
1084 * link vnode call
1085 */
1086 static int
ufs_link(struct vop_link_args * ap)1087 ufs_link(
1088 struct vop_link_args /* {
1089 struct vnode *a_tdvp;
1090 struct vnode *a_vp;
1091 struct componentname *a_cnp;
1092 } */ *ap)
1093 {
1094 struct vnode *vp = ap->a_vp;
1095 struct vnode *tdvp = ap->a_tdvp;
1096 struct componentname *cnp = ap->a_cnp;
1097 struct inode *ip;
1098 struct direct newdir;
1099 int error;
1100
1101 if (DOINGSUJ(tdvp)) {
1102 error = softdep_prelink(tdvp, vp, cnp);
1103 if (error != 0) {
1104 MPASS(error == ERELOOKUP);
1105 return (error);
1106 }
1107 }
1108
1109 if (VTOI(tdvp)->i_effnlink < 2) {
1110 print_bad_link_count("ufs_link", tdvp);
1111 error = EINVAL;
1112 goto out;
1113 }
1114 error = ufs_sync_nlink(vp, tdvp);
1115 if (error != 0)
1116 goto out;
1117 ip = VTOI(vp);
1118
1119 /*
1120 * The file may have been removed after namei dropped the original
1121 * lock.
1122 */
1123 if (ip->i_effnlink == 0) {
1124 error = ENOENT;
1125 goto out;
1126 }
1127 if (ip->i_flags & (IMMUTABLE | APPEND)) {
1128 error = EPERM;
1129 goto out;
1130 }
1131
1132 ip->i_effnlink++;
1133 ip->i_nlink++;
1134 DIP_SET_NLINK(ip, ip->i_nlink);
1135 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1136 if (DOINGSOFTDEP(vp))
1137 softdep_setup_link(VTOI(tdvp), ip);
1138 error = UFS_UPDATE(vp, !DOINGSOFTDEP(vp) && !DOINGASYNC(vp));
1139 if (!error) {
1140 ufs_makedirentry(ip, cnp, &newdir);
1141 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
1142 }
1143
1144 if (error) {
1145 ip->i_effnlink--;
1146 ip->i_nlink--;
1147 DIP_SET_NLINK(ip, ip->i_nlink);
1148 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1149 if (DOINGSOFTDEP(vp))
1150 softdep_revert_link(VTOI(tdvp), ip);
1151 }
1152 out:
1153 return (error);
1154 }
1155
1156 /*
1157 * whiteout vnode call
1158 */
1159 static int
ufs_whiteout(struct vop_whiteout_args * ap)1160 ufs_whiteout(
1161 struct vop_whiteout_args /* {
1162 struct vnode *a_dvp;
1163 struct componentname *a_cnp;
1164 int a_flags;
1165 } */ *ap)
1166 {
1167 struct vnode *dvp = ap->a_dvp;
1168 struct componentname *cnp = ap->a_cnp;
1169 struct direct newdir;
1170 int error = 0;
1171
1172 if (DOINGSUJ(dvp) && (ap->a_flags == CREATE ||
1173 ap->a_flags == DELETE)) {
1174 error = softdep_prelink(dvp, NULL, cnp);
1175 if (error != 0) {
1176 MPASS(error == ERELOOKUP);
1177 return (error);
1178 }
1179 }
1180
1181 switch (ap->a_flags) {
1182 case LOOKUP:
1183 /* 4.4 format directories support whiteout operations */
1184 if (!OFSFMT(dvp))
1185 return (0);
1186 return (EOPNOTSUPP);
1187
1188 case CREATE:
1189 /* create a new directory whiteout */
1190 #ifdef INVARIANTS
1191 if (OFSFMT(dvp))
1192 panic("ufs_whiteout: old format filesystem");
1193 #endif
1194
1195 newdir.d_ino = UFS_WINO;
1196 newdir.d_namlen = cnp->cn_namelen;
1197 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
1198 newdir.d_type = DT_WHT;
1199 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
1200 break;
1201
1202 case DELETE:
1203 /* remove an existing directory whiteout */
1204 #ifdef INVARIANTS
1205 if (OFSFMT(dvp))
1206 panic("ufs_whiteout: old format filesystem");
1207 #endif
1208
1209 cnp->cn_flags &= ~DOWHITEOUT;
1210 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
1211 break;
1212 default:
1213 panic("ufs_whiteout: unknown op");
1214 }
1215 return (error);
1216 }
1217
1218 static volatile int rename_restarts;
1219 SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD,
1220 __DEVOLATILE(int *, &rename_restarts), 0,
1221 "Times rename had to restart due to lock contention");
1222
1223 /*
1224 * Rename system call.
1225 * rename("foo", "bar");
1226 * is essentially
1227 * unlink("bar");
1228 * link("foo", "bar");
1229 * unlink("foo");
1230 * but ``atomically''. Can't do full commit without saving state in the
1231 * inode on disk which isn't feasible at this time. Best we can do is
1232 * always guarantee the target exists.
1233 *
1234 * Basic algorithm is:
1235 *
1236 * 1) Bump link count on source while we're linking it to the
1237 * target. This also ensure the inode won't be deleted out
1238 * from underneath us while we work (it may be truncated by
1239 * a concurrent `trunc' or `open' for creation).
1240 * 2) Link source to destination. If destination already exists,
1241 * delete it first.
1242 * 3) Unlink source reference to inode if still around. If a
1243 * directory was moved and the parent of the destination
1244 * is different from the source, patch the ".." entry in the
1245 * directory.
1246 */
1247 static int
ufs_rename(struct vop_rename_args * ap)1248 ufs_rename(
1249 struct vop_rename_args /* {
1250 struct vnode *a_fdvp;
1251 struct vnode *a_fvp;
1252 struct componentname *a_fcnp;
1253 struct vnode *a_tdvp;
1254 struct vnode *a_tvp;
1255 struct componentname *a_tcnp;
1256 } */ *ap)
1257 {
1258 struct vnode *tvp = ap->a_tvp;
1259 struct vnode *tdvp = ap->a_tdvp;
1260 struct vnode *fvp = ap->a_fvp;
1261 struct vnode *fdvp = ap->a_fdvp;
1262 struct vnode *nvp;
1263 struct componentname *tcnp = ap->a_tcnp;
1264 struct componentname *fcnp = ap->a_fcnp;
1265 struct thread *td = curthread;
1266 struct inode *fip, *tip, *tdp, *fdp;
1267 struct direct newdir;
1268 off_t endoff;
1269 int doingdirectory, newparent;
1270 int error = 0;
1271 struct mount *mp;
1272 ino_t ino;
1273 seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s;
1274 bool checkpath_locked, want_seqc_end;
1275
1276 checkpath_locked = want_seqc_end = false;
1277
1278 endoff = 0;
1279 mp = tdvp->v_mount;
1280 VOP_UNLOCK(tdvp);
1281 if (tvp && tvp != tdvp)
1282 VOP_UNLOCK(tvp);
1283 /*
1284 * Check for cross-device rename.
1285 */
1286 if ((fvp->v_mount != tdvp->v_mount) ||
1287 (tvp && (fvp->v_mount != tvp->v_mount))) {
1288 error = EXDEV;
1289 mp = NULL;
1290 goto releout;
1291 }
1292
1293 fdvp_s = fvp_s = tdvp_s = tvp_s = SEQC_MOD;
1294 relock:
1295 /*
1296 * We need to acquire 2 to 4 locks depending on whether tvp is NULL
1297 * and fdvp and tdvp are the same directory. Subsequently we need
1298 * to double-check all paths and in the directory rename case we
1299 * need to verify that we are not creating a directory loop. To
1300 * handle this we acquire all but fdvp using non-blocking
1301 * acquisitions. If we fail to acquire any lock in the path we will
1302 * drop all held locks, acquire the new lock in a blocking fashion,
1303 * and then release it and restart the rename. This acquire/release
1304 * step ensures that we do not spin on a lock waiting for release.
1305 */
1306 error = vn_lock(fdvp, LK_EXCLUSIVE);
1307 if (error)
1308 goto releout;
1309 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1310 VOP_UNLOCK(fdvp);
1311 error = vn_lock(tdvp, LK_EXCLUSIVE);
1312 if (error)
1313 goto releout;
1314 VOP_UNLOCK(tdvp);
1315 atomic_add_int(&rename_restarts, 1);
1316 goto relock;
1317 }
1318 /*
1319 * Re-resolve fvp to be certain it still exists and fetch the
1320 * correct vnode.
1321 */
1322 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1323 if (error) {
1324 VOP_UNLOCK(fdvp);
1325 VOP_UNLOCK(tdvp);
1326 goto releout;
1327 }
1328 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1329 if (error) {
1330 VOP_UNLOCK(fdvp);
1331 VOP_UNLOCK(tdvp);
1332 if (error != EBUSY)
1333 goto releout;
1334 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1335 if (error != 0)
1336 goto releout;
1337 VOP_UNLOCK(nvp);
1338 vrele(fvp);
1339 fvp = nvp;
1340 atomic_add_int(&rename_restarts, 1);
1341 goto relock;
1342 }
1343 vrele(fvp);
1344 fvp = nvp;
1345 /*
1346 * Re-resolve tvp and acquire the vnode lock if present.
1347 */
1348 error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino);
1349 if (error != 0 && error != EJUSTRETURN) {
1350 VOP_UNLOCK(fdvp);
1351 VOP_UNLOCK(tdvp);
1352 VOP_UNLOCK(fvp);
1353 goto releout;
1354 }
1355 /*
1356 * If tvp disappeared we just carry on.
1357 */
1358 if (error == EJUSTRETURN && tvp != NULL) {
1359 vrele(tvp);
1360 tvp = NULL;
1361 }
1362 /*
1363 * Get the tvp ino if the lookup succeeded. We may have to restart
1364 * if the non-blocking acquire fails.
1365 */
1366 if (error == 0) {
1367 nvp = NULL;
1368 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1369 if (tvp)
1370 vrele(tvp);
1371 tvp = nvp;
1372 if (error) {
1373 VOP_UNLOCK(fdvp);
1374 VOP_UNLOCK(tdvp);
1375 VOP_UNLOCK(fvp);
1376 if (error != EBUSY)
1377 goto releout;
1378 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1379 if (error != 0)
1380 goto releout;
1381 vput(nvp);
1382 atomic_add_int(&rename_restarts, 1);
1383 goto relock;
1384 }
1385 }
1386
1387 if (DOINGSUJ(fdvp) &&
1388 (seqc_in_modify(fdvp_s) || !vn_seqc_consistent(fdvp, fdvp_s) ||
1389 seqc_in_modify(fvp_s) || !vn_seqc_consistent(fvp, fvp_s) ||
1390 seqc_in_modify(tdvp_s) || !vn_seqc_consistent(tdvp, tdvp_s) ||
1391 (tvp != NULL && (seqc_in_modify(tvp_s) ||
1392 !vn_seqc_consistent(tvp, tvp_s))))) {
1393 error = softdep_prerename(fdvp, fvp, tdvp, tvp);
1394 if (error != 0)
1395 goto releout;
1396 }
1397
1398 fdp = VTOI(fdvp);
1399 fip = VTOI(fvp);
1400 tdp = VTOI(tdvp);
1401 tip = NULL;
1402 if (tvp)
1403 tip = VTOI(tvp);
1404 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1405 (VTOI(tdvp)->i_flags & APPEND))) {
1406 error = EPERM;
1407 goto unlockout;
1408 }
1409 /*
1410 * Renaming a file to itself has no effect. The upper layers should
1411 * not call us in that case. However, things could change after
1412 * we drop the locks above.
1413 */
1414 if (fvp == tvp) {
1415 error = 0;
1416 goto unlockout;
1417 }
1418 doingdirectory = 0;
1419 newparent = 0;
1420 ino = fip->i_number;
1421 if (fip->i_nlink >= UFS_LINK_MAX) {
1422 if (!DOINGSOFTDEP(fvp) || fip->i_effnlink >= UFS_LINK_MAX) {
1423 error = EMLINK;
1424 goto unlockout;
1425 }
1426 vfs_ref(mp);
1427 MPASS(!want_seqc_end);
1428 if (checkpath_locked) {
1429 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1430 checkpath_locked = false;
1431 }
1432 VOP_UNLOCK(fdvp);
1433 VOP_UNLOCK(fvp);
1434 vref(tdvp);
1435 if (tvp != NULL)
1436 vref(tvp);
1437 VOP_VPUT_PAIR(tdvp, &tvp, true);
1438 error = ufs_sync_nlink1(mp);
1439 vrele(fdvp);
1440 vrele(fvp);
1441 vrele(tdvp);
1442 if (tvp != NULL)
1443 vrele(tvp);
1444 return (error);
1445 }
1446 if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1447 || (fdp->i_flags & APPEND)) {
1448 error = EPERM;
1449 goto unlockout;
1450 }
1451 if ((fip->i_mode & IFMT) == IFDIR) {
1452 /*
1453 * Avoid ".", "..", and aliases of "." for obvious reasons.
1454 */
1455 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1456 fdp == fip ||
1457 (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
1458 error = EINVAL;
1459 goto unlockout;
1460 }
1461 if (fdp->i_number != tdp->i_number)
1462 newparent = tdp->i_number;
1463 doingdirectory = 1;
1464 }
1465 if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) ||
1466 (tvp != NULL && tvp->v_type == VDIR &&
1467 tvp->v_mountedhere != NULL)) {
1468 error = EXDEV;
1469 goto unlockout;
1470 }
1471
1472 /*
1473 * If ".." must be changed (ie the directory gets a new
1474 * parent) then the source directory must not be in the
1475 * directory hierarchy above the target, as this would
1476 * orphan everything below the source directory. Also
1477 * the user must have write permission in the source so
1478 * as to be able to change "..".
1479 */
1480 if (doingdirectory && newparent) {
1481 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
1482 if (error)
1483 goto unlockout;
1484
1485 sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock);
1486 checkpath_locked = true;
1487 error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred,
1488 &ino);
1489 /*
1490 * We encountered a lock that we have to wait for. Unlock
1491 * everything else and VGET before restarting.
1492 */
1493 if (ino) {
1494 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1495 checkpath_locked = false;
1496 VOP_UNLOCK(fdvp);
1497 VOP_UNLOCK(fvp);
1498 VOP_UNLOCK(tdvp);
1499 if (tvp)
1500 VOP_UNLOCK(tvp);
1501 error = VFS_VGET(mp, ino, LK_SHARED, &nvp);
1502 if (error == 0)
1503 vput(nvp);
1504 atomic_add_int(&rename_restarts, 1);
1505 goto relock;
1506 }
1507 if (error)
1508 goto unlockout;
1509 }
1510 if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 ||
1511 tdp->i_effnlink == 0)
1512 panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp);
1513
1514 if (tvp != NULL)
1515 vn_seqc_write_begin(tvp);
1516 vn_seqc_write_begin(tdvp);
1517 vn_seqc_write_begin(fvp);
1518 vn_seqc_write_begin(fdvp);
1519 want_seqc_end = true;
1520
1521 /*
1522 * 1) Bump link count while we're moving stuff
1523 * around. If we crash somewhere before
1524 * completing our work, the link count
1525 * may be wrong, but correctable.
1526 */
1527 fip->i_effnlink++;
1528 fip->i_nlink++;
1529 DIP_SET_NLINK(fip, fip->i_nlink);
1530 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1531 if (DOINGSOFTDEP(fvp))
1532 softdep_setup_link(tdp, fip);
1533 error = UFS_UPDATE(fvp, !DOINGSOFTDEP(fvp) && !DOINGASYNC(fvp));
1534 if (error)
1535 goto bad;
1536
1537 /*
1538 * 2) If target doesn't exist, link the target
1539 * to the source and unlink the source.
1540 * Otherwise, rewrite the target directory
1541 * entry to reference the source inode and
1542 * expunge the original entry's existence.
1543 */
1544 if (tip == NULL) {
1545 if (ITODEV(tdp) != ITODEV(fip))
1546 panic("ufs_rename: EXDEV");
1547 if (doingdirectory && newparent) {
1548 /*
1549 * Account for ".." in new directory.
1550 * When source and destination have the same
1551 * parent we don't adjust the link count. The
1552 * actual link modification is completed when
1553 * .. is rewritten below.
1554 */
1555 if (tdp->i_nlink >= UFS_LINK_MAX) {
1556 fip->i_effnlink--;
1557 fip->i_nlink--;
1558 DIP_SET_NLINK(fip, fip->i_nlink);
1559 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1560 if (DOINGSOFTDEP(fvp))
1561 softdep_revert_link(tdp, fip);
1562 if (!DOINGSOFTDEP(tdvp) ||
1563 tdp->i_effnlink >= UFS_LINK_MAX) {
1564 error = EMLINK;
1565 goto unlockout;
1566 }
1567 MPASS(want_seqc_end);
1568 if (tvp != NULL)
1569 vn_seqc_write_end(tvp);
1570 vn_seqc_write_end(tdvp);
1571 vn_seqc_write_end(fvp);
1572 vn_seqc_write_end(fdvp);
1573 want_seqc_end = false;
1574 vfs_ref(mp);
1575 MPASS(checkpath_locked);
1576 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1577 checkpath_locked = false;
1578 VOP_UNLOCK(fdvp);
1579 VOP_UNLOCK(fvp);
1580 vref(tdvp);
1581 if (tvp != NULL)
1582 vref(tvp);
1583 VOP_VPUT_PAIR(tdvp, &tvp, true);
1584 error = ufs_sync_nlink1(mp);
1585 vrele(fdvp);
1586 vrele(fvp);
1587 vrele(tdvp);
1588 if (tvp != NULL)
1589 vrele(tvp);
1590 return (error);
1591 }
1592 }
1593 ufs_makedirentry(fip, tcnp, &newdir);
1594 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1595 if (error)
1596 goto bad;
1597 /* Setup tdvp for directory compaction if needed. */
1598 if (I_COUNT(tdp) != 0 && I_ENDOFF(tdp) != 0 &&
1599 I_ENDOFF(tdp) < tdp->i_size)
1600 endoff = I_ENDOFF(tdp);
1601 } else {
1602 if (ITODEV(tip) != ITODEV(tdp) || ITODEV(tip) != ITODEV(fip))
1603 panic("ufs_rename: EXDEV");
1604 /*
1605 * Short circuit rename(foo, foo).
1606 */
1607 if (tip->i_number == fip->i_number)
1608 panic("ufs_rename: same file");
1609 /*
1610 * If the parent directory is "sticky", then the caller
1611 * must possess VADMIN for the parent directory, or the
1612 * destination of the rename. This implements append-only
1613 * directories.
1614 */
1615 if ((tdp->i_mode & S_ISTXT) &&
1616 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1617 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1618 error = EPERM;
1619 goto bad;
1620 }
1621 /*
1622 * Target must be empty if a directory and have no links
1623 * to it. Also, ensure source and target are compatible
1624 * (both directories, or both not directories).
1625 */
1626 if ((tip->i_mode & IFMT) == IFDIR) {
1627 if ((tip->i_effnlink > 2) ||
1628 !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) {
1629 error = ENOTEMPTY;
1630 goto bad;
1631 }
1632 if (!doingdirectory) {
1633 error = ENOTDIR;
1634 goto bad;
1635 }
1636 cache_purge(tdvp);
1637 } else if (doingdirectory) {
1638 error = EISDIR;
1639 goto bad;
1640 }
1641 if (doingdirectory) {
1642 if (!newparent) {
1643 tdp->i_effnlink--;
1644 if (DOINGSOFTDEP(tdvp))
1645 softdep_change_linkcnt(tdp);
1646 }
1647 tip->i_effnlink--;
1648 if (DOINGSOFTDEP(tvp))
1649 softdep_change_linkcnt(tip);
1650 }
1651 error = ufs_dirrewrite(tdp, tip, fip->i_number,
1652 IFTODT(fip->i_mode),
1653 (doingdirectory && newparent) ? newparent : doingdirectory);
1654 if (error) {
1655 if (doingdirectory) {
1656 if (!newparent) {
1657 tdp->i_effnlink++;
1658 if (DOINGSOFTDEP(tdvp))
1659 softdep_change_linkcnt(tdp);
1660 }
1661 tip->i_effnlink++;
1662 if (DOINGSOFTDEP(tvp))
1663 softdep_change_linkcnt(tip);
1664 }
1665 goto bad;
1666 }
1667 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1668 /*
1669 * The only stuff left in the directory is "."
1670 * and "..". The "." reference is inconsequential
1671 * since we are quashing it. We have removed the "."
1672 * reference and the reference in the parent directory,
1673 * but there may be other hard links. The soft
1674 * dependency code will arrange to do these operations
1675 * after the parent directory entry has been deleted on
1676 * disk, so when running with that code we avoid doing
1677 * them now.
1678 */
1679 if (!newparent) {
1680 tdp->i_nlink--;
1681 DIP_SET_NLINK(tdp, tdp->i_nlink);
1682 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1683 }
1684 tip->i_nlink--;
1685 DIP_SET_NLINK(tip, tip->i_nlink);
1686 UFS_INODE_SET_FLAG(tip, IN_CHANGE);
1687 }
1688 }
1689
1690 /*
1691 * 3) Unlink the source. We have to resolve the path again to
1692 * fixup the directory offset and count for ufs_dirremove.
1693 */
1694 if (fdvp == tdvp) {
1695 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1696 if (error)
1697 panic("ufs_rename: from entry went away!");
1698 if (ino != fip->i_number)
1699 panic("ufs_rename: ino mismatch %ju != %ju\n",
1700 (uintmax_t)ino, (uintmax_t)fip->i_number);
1701 }
1702 /*
1703 * If the source is a directory with a
1704 * new parent, the link count of the old
1705 * parent directory must be decremented
1706 * and ".." set to point to the new parent.
1707 */
1708 if (doingdirectory && newparent) {
1709 /*
1710 * Set the directory depth based on its new parent.
1711 */
1712 DIP_SET(fip, i_dirdepth, DIP(tdp, i_dirdepth) + 1);
1713 /*
1714 * If tip exists we simply use its link, otherwise we must
1715 * add a new one.
1716 */
1717 if (tip == NULL) {
1718 tdp->i_effnlink++;
1719 tdp->i_nlink++;
1720 DIP_SET_NLINK(tdp, tdp->i_nlink);
1721 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1722 if (DOINGSOFTDEP(tdvp))
1723 softdep_setup_dotdot_link(tdp, fip);
1724 error = UFS_UPDATE(tdvp, !DOINGSOFTDEP(tdvp) &&
1725 !DOINGASYNC(tdvp));
1726 /* Don't go to bad here as the new link exists. */
1727 if (error)
1728 goto unlockout;
1729 } else if (DOINGSUJ(tdvp))
1730 /* Journal must account for each new link. */
1731 softdep_setup_dotdot_link(tdp, fip);
1732 SET_I_OFFSET(fip, mastertemplate.dot_reclen);
1733 if (ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0) != 0)
1734 ufs_dirbad(fip, mastertemplate.dot_reclen,
1735 "rename: missing .. entry");
1736 cache_purge(fdvp);
1737 }
1738 error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0);
1739 /*
1740 * The kern_renameat() looks up the fvp using the DELETE flag, which
1741 * causes the removal of the name cache entry for fvp.
1742 * As the relookup of the fvp is done in two steps:
1743 * ufs_lookup_ino() and then VFS_VGET(), another thread might do a
1744 * normal lookup of the from name just before the VFS_VGET() call,
1745 * causing the cache entry to be re-instantiated.
1746 *
1747 * The same issue also applies to tvp if it exists as
1748 * otherwise we may have a stale name cache entry for the new
1749 * name that references the old i-node if it has other links
1750 * or open file descriptors.
1751 */
1752 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp);
1753
1754 unlockout:
1755 if (want_seqc_end) {
1756 if (tvp != NULL)
1757 vn_seqc_write_end(tvp);
1758 vn_seqc_write_end(tdvp);
1759 vn_seqc_write_end(fvp);
1760 vn_seqc_write_end(fdvp);
1761 }
1762
1763 if (checkpath_locked)
1764 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1765
1766 vput(fdvp);
1767 vput(fvp);
1768
1769 /*
1770 * If compaction or fsync was requested do it in
1771 * ffs_vput_pair() now that other locks are no longer needed.
1772 */
1773 if (error == 0 && endoff != 0) {
1774 UFS_INODE_SET_FLAG(tdp, IN_ENDOFF);
1775 SET_I_ENDOFF(tdp, endoff);
1776 }
1777 VOP_VPUT_PAIR(tdvp, &tvp, true);
1778 return (error);
1779
1780 bad:
1781 fip->i_effnlink--;
1782 fip->i_nlink--;
1783 DIP_SET_NLINK(fip, fip->i_nlink);
1784 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1785 if (DOINGSOFTDEP(fvp))
1786 softdep_revert_link(tdp, fip);
1787 goto unlockout;
1788
1789 releout:
1790 if (want_seqc_end) {
1791 if (tvp != NULL)
1792 vn_seqc_write_end(tvp);
1793 vn_seqc_write_end(tdvp);
1794 vn_seqc_write_end(fvp);
1795 vn_seqc_write_end(fdvp);
1796 }
1797
1798 vrele(fdvp);
1799 vrele(fvp);
1800 vrele(tdvp);
1801 if (tvp)
1802 vrele(tvp);
1803
1804 return (error);
1805 }
1806
1807 #ifdef UFS_ACL
1808 static int
ufs_do_posix1e_acl_inheritance_dir(struct vnode * dvp,struct vnode * tvp,mode_t dmode,struct ucred * cred,struct thread * td)1809 ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1810 mode_t dmode, struct ucred *cred, struct thread *td)
1811 {
1812 int error;
1813 struct inode *ip = VTOI(tvp);
1814 struct acl *dacl, *acl;
1815
1816 acl = acl_alloc(M_WAITOK);
1817 dacl = acl_alloc(M_WAITOK);
1818
1819 /*
1820 * Retrieve default ACL from parent, if any.
1821 */
1822 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1823 switch (error) {
1824 case 0:
1825 /*
1826 * Retrieved a default ACL, so merge mode and ACL if
1827 * necessary. If the ACL is empty, fall through to
1828 * the "not defined or available" case.
1829 */
1830 if (acl->acl_cnt != 0) {
1831 dmode = acl_posix1e_newfilemode(dmode, acl);
1832 UFS_INODE_SET_MODE(ip, dmode);
1833 DIP_SET(ip, i_mode, dmode);
1834 *dacl = *acl;
1835 ufs_sync_acl_from_inode(ip, acl);
1836 break;
1837 }
1838 /* FALLTHROUGH */
1839
1840 case EOPNOTSUPP:
1841 /*
1842 * Just use the mode as-is.
1843 */
1844 UFS_INODE_SET_MODE(ip, dmode);
1845 DIP_SET(ip, i_mode, dmode);
1846 error = 0;
1847 goto out;
1848
1849 default:
1850 goto out;
1851 }
1852
1853 /*
1854 * XXX: If we abort now, will Soft Updates notify the extattr
1855 * code that the EAs for the file need to be released?
1856 */
1857 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1858 if (error == 0)
1859 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1860 switch (error) {
1861 case 0:
1862 break;
1863
1864 case EOPNOTSUPP:
1865 /*
1866 * XXX: This should not happen, as EOPNOTSUPP above
1867 * was supposed to free acl.
1868 */
1869 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1870 /*
1871 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1872 */
1873 break;
1874
1875 default:
1876 goto out;
1877 }
1878
1879 out:
1880 acl_free(acl);
1881 acl_free(dacl);
1882
1883 return (error);
1884 }
1885
1886 static int
ufs_do_posix1e_acl_inheritance_file(struct vnode * dvp,struct vnode * tvp,mode_t mode,struct ucred * cred,struct thread * td)1887 ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1888 mode_t mode, struct ucred *cred, struct thread *td)
1889 {
1890 int error;
1891 struct inode *ip = VTOI(tvp);
1892 struct acl *acl;
1893
1894 acl = acl_alloc(M_WAITOK);
1895
1896 /*
1897 * Retrieve default ACL for parent, if any.
1898 */
1899 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1900 switch (error) {
1901 case 0:
1902 /*
1903 * Retrieved a default ACL, so merge mode and ACL if
1904 * necessary.
1905 */
1906 if (acl->acl_cnt != 0) {
1907 /*
1908 * Two possible ways for default ACL to not
1909 * be present. First, the EA can be
1910 * undefined, or second, the default ACL can
1911 * be blank. If it's blank, fall through to
1912 * the it's not defined case.
1913 */
1914 mode = acl_posix1e_newfilemode(mode, acl);
1915 UFS_INODE_SET_MODE(ip, mode);
1916 DIP_SET(ip, i_mode, mode);
1917 ufs_sync_acl_from_inode(ip, acl);
1918 break;
1919 }
1920 /* FALLTHROUGH */
1921
1922 case EOPNOTSUPP:
1923 /*
1924 * Just use the mode as-is.
1925 */
1926 UFS_INODE_SET_MODE(ip, mode);
1927 DIP_SET(ip, i_mode, mode);
1928 error = 0;
1929 goto out;
1930
1931 default:
1932 goto out;
1933 }
1934
1935 /*
1936 * XXX: If we abort now, will Soft Updates notify the extattr
1937 * code that the EAs for the file need to be released?
1938 */
1939 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1940 switch (error) {
1941 case 0:
1942 break;
1943
1944 case EOPNOTSUPP:
1945 /*
1946 * XXX: This should not happen, as EOPNOTSUPP above was
1947 * supposed to free acl.
1948 */
1949 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1950 "but no VOP_SETACL()\n");
1951 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1952 "but no VOP_SETACL()"); */
1953 break;
1954
1955 default:
1956 goto out;
1957 }
1958
1959 out:
1960 acl_free(acl);
1961
1962 return (error);
1963 }
1964
1965 static int
ufs_do_nfs4_acl_inheritance(struct vnode * dvp,struct vnode * tvp,mode_t child_mode,struct ucred * cred,struct thread * td)1966 ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp,
1967 mode_t child_mode, struct ucred *cred, struct thread *td)
1968 {
1969 int error;
1970 struct acl *parent_aclp, *child_aclp;
1971
1972 parent_aclp = acl_alloc(M_WAITOK);
1973 child_aclp = acl_alloc(M_WAITOK | M_ZERO);
1974
1975 error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td);
1976 if (error)
1977 goto out;
1978 acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp,
1979 child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR);
1980 error = ufs_setacl_nfs4_internal(tvp, child_aclp, td);
1981 if (error)
1982 goto out;
1983 out:
1984 acl_free(parent_aclp);
1985 acl_free(child_aclp);
1986
1987 return (error);
1988 }
1989 #endif
1990
1991 /*
1992 * Mkdir system call
1993 */
1994 static int
ufs_mkdir(struct vop_mkdir_args * ap)1995 ufs_mkdir(
1996 struct vop_mkdir_args /* {
1997 struct vnode *a_dvp;
1998 struct vnode **a_vpp;
1999 struct componentname *a_cnp;
2000 struct vattr *a_vap;
2001 } */ *ap)
2002 {
2003 struct vnode *dvp = ap->a_dvp;
2004 struct vattr *vap = ap->a_vap;
2005 struct componentname *cnp = ap->a_cnp;
2006 struct inode *ip, *dp;
2007 struct vnode *tvp;
2008 struct buf *bp;
2009 struct dirtemplate dirtemplate, *dtp;
2010 struct direct newdir;
2011 int error, dmode;
2012 long blkoff;
2013
2014 dp = VTOI(dvp);
2015 error = ufs_sync_nlink(dvp, NULL);
2016 if (error != 0)
2017 goto out;
2018 dmode = vap->va_mode & 0777;
2019 dmode |= IFDIR;
2020
2021 /*
2022 * Must simulate part of ufs_makeinode here to acquire the inode,
2023 * but not have it entered in the parent directory. The entry is
2024 * made later after writing "." and ".." entries.
2025 */
2026 if (dp->i_effnlink < 2) {
2027 print_bad_link_count("ufs_mkdir", dvp);
2028 error = EINVAL;
2029 goto out;
2030 }
2031
2032 if (DOINGSUJ(dvp)) {
2033 error = softdep_prelink(dvp, NULL, cnp);
2034 if (error != 0) {
2035 MPASS(error == ERELOOKUP);
2036 return (error);
2037 }
2038 }
2039
2040 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
2041 if (error)
2042 goto out;
2043 vn_seqc_write_begin(tvp);
2044 ip = VTOI(tvp);
2045 ip->i_gid = dp->i_gid;
2046 DIP_SET(ip, i_gid, dp->i_gid);
2047 #ifdef SUIDDIR
2048 {
2049 #ifdef QUOTA
2050 struct ucred ucred, *ucp;
2051 gid_t ucred_group;
2052 ucp = cnp->cn_cred;
2053 #endif
2054 /*
2055 * If we are hacking owners here, (only do this where told to)
2056 * and we are not giving it TO root, (would subvert quotas)
2057 * then go ahead and give it to the other user.
2058 * The new directory also inherits the SUID bit.
2059 * If user's UID and dir UID are the same,
2060 * 'give it away' so that the SUID is still forced on.
2061 */
2062 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2063 (dp->i_mode & ISUID) && dp->i_uid) {
2064 dmode |= ISUID;
2065 ip->i_uid = dp->i_uid;
2066 DIP_SET(ip, i_uid, dp->i_uid);
2067 #ifdef QUOTA
2068 if (dp->i_uid != cnp->cn_cred->cr_uid) {
2069 /*
2070 * Make sure the correct user gets charged
2071 * for the space.
2072 * Make a dummy credential for the victim.
2073 * XXX This seems to never be accessed out of
2074 * our context so a stack variable is ok.
2075 */
2076 ucred.cr_ref = 1;
2077 ucred.cr_uid = ip->i_uid;
2078 ucred.cr_ngroups = 1;
2079 ucred.cr_groups = &ucred_group;
2080 ucred.cr_groups[0] = dp->i_gid;
2081 ucp = &ucred;
2082 }
2083 #endif
2084 } else {
2085 ip->i_uid = cnp->cn_cred->cr_uid;
2086 DIP_SET(ip, i_uid, ip->i_uid);
2087 }
2088 #ifdef QUOTA
2089 if ((error = getinoquota(ip)) ||
2090 (error = chkiq(ip, 1, ucp, 0))) {
2091 if (DOINGSOFTDEP(tvp))
2092 softdep_revert_link(dp, ip);
2093 UFS_VFREE(tvp, ip->i_number, dmode);
2094 vn_seqc_write_end(tvp);
2095 vgone(tvp);
2096 vput(tvp);
2097 return (error);
2098 }
2099 #endif
2100 }
2101 #else /* !SUIDDIR */
2102 ip->i_uid = cnp->cn_cred->cr_uid;
2103 DIP_SET(ip, i_uid, ip->i_uid);
2104 #ifdef QUOTA
2105 if ((error = getinoquota(ip)) ||
2106 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2107 if (DOINGSOFTDEP(tvp))
2108 softdep_revert_link(dp, ip);
2109 UFS_VFREE(tvp, ip->i_number, dmode);
2110 vn_seqc_write_end(tvp);
2111 vgone(tvp);
2112 vput(tvp);
2113 return (error);
2114 }
2115 #endif
2116 #endif /* !SUIDDIR */
2117 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2118 UFS_INODE_SET_MODE(ip, dmode);
2119 DIP_SET(ip, i_mode, dmode);
2120 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
2121 ip->i_effnlink = 2;
2122 ip->i_nlink = 2;
2123 DIP_SET_NLINK(ip, 2);
2124 DIP_SET(ip, i_dirdepth, DIP(dp,i_dirdepth) + 1);
2125
2126 if (cnp->cn_flags & ISWHITEOUT) {
2127 ip->i_flags |= UF_OPAQUE;
2128 DIP_SET(ip, i_flags, ip->i_flags);
2129 }
2130
2131 /*
2132 * Bump link count in parent directory to reflect work done below.
2133 * Should be done before reference is created so cleanup is
2134 * possible if we crash.
2135 */
2136 dp->i_effnlink++;
2137 dp->i_nlink++;
2138 DIP_SET_NLINK(dp, dp->i_nlink);
2139 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2140 if (DOINGSOFTDEP(dvp))
2141 softdep_setup_mkdir(dp, ip);
2142 error = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp));
2143 if (error)
2144 goto bad;
2145 #ifdef MAC
2146 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2147 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2148 dvp, tvp, cnp);
2149 if (error)
2150 goto bad;
2151 }
2152 #endif
2153 #ifdef UFS_ACL
2154 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2155 error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
2156 cnp->cn_cred, curthread);
2157 if (error)
2158 goto bad;
2159 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2160 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode,
2161 cnp->cn_cred, curthread);
2162 if (error)
2163 goto bad;
2164 }
2165 #endif /* !UFS_ACL */
2166
2167 /*
2168 * Initialize directory with "." and ".." from static template.
2169 */
2170 if (!OFSFMT(dvp))
2171 dtp = &mastertemplate;
2172 else
2173 dtp = (struct dirtemplate *)&omastertemplate;
2174 dirtemplate = *dtp;
2175 dirtemplate.dot_ino = ip->i_number;
2176 dirtemplate.dotdot_ino = dp->i_number;
2177 vnode_pager_setsize(tvp, DIRBLKSIZ);
2178 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
2179 BA_CLRBUF, &bp)) != 0)
2180 goto bad;
2181 ip->i_size = DIRBLKSIZ;
2182 DIP_SET(ip, i_size, DIRBLKSIZ);
2183 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2184 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
2185 if (DOINGSOFTDEP(tvp)) {
2186 /*
2187 * Ensure that the entire newly allocated block is a
2188 * valid directory so that future growth within the
2189 * block does not have to ensure that the block is
2190 * written before the inode.
2191 */
2192 blkoff = DIRBLKSIZ;
2193 while (blkoff < bp->b_bcount) {
2194 ((struct direct *)
2195 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
2196 blkoff += DIRBLKSIZ;
2197 }
2198 }
2199 if ((error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) &&
2200 !DOINGASYNC(tvp))) != 0) {
2201 (void)bwrite(bp);
2202 goto bad;
2203 }
2204 /*
2205 * Directory set up, now install its entry in the parent directory.
2206 *
2207 * If we are not doing soft dependencies, then we must write out the
2208 * buffer containing the new directory body before entering the new
2209 * name in the parent. If we are doing soft dependencies, then the
2210 * buffer containing the new directory body will be passed to and
2211 * released in the soft dependency code after the code has attached
2212 * an appropriate ordering dependency to the buffer which ensures that
2213 * the buffer is written before the new name is written in the parent.
2214 */
2215 if (DOINGASYNC(dvp))
2216 bdwrite(bp);
2217 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
2218 goto bad;
2219 ufs_makedirentry(ip, cnp, &newdir);
2220 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
2221
2222 bad:
2223 if (error == 0) {
2224 *ap->a_vpp = tvp;
2225 vn_seqc_write_end(tvp);
2226 } else {
2227 dp->i_effnlink--;
2228 dp->i_nlink--;
2229 DIP_SET_NLINK(dp, dp->i_nlink);
2230 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2231 /*
2232 * No need to do an explicit VOP_TRUNCATE here, vrele will
2233 * do this for us because we set the link count to 0.
2234 */
2235 ip->i_effnlink = 0;
2236 ip->i_nlink = 0;
2237 DIP_SET_NLINK(ip, 0);
2238 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2239 if (DOINGSOFTDEP(tvp))
2240 softdep_revert_mkdir(dp, ip);
2241 vn_seqc_write_end(tvp);
2242 vgone(tvp);
2243 vput(tvp);
2244 }
2245 out:
2246 return (error);
2247 }
2248
2249 /*
2250 * Rmdir system call.
2251 */
2252 static int
ufs_rmdir(struct vop_rmdir_args * ap)2253 ufs_rmdir(
2254 struct vop_rmdir_args /* {
2255 struct vnode *a_dvp;
2256 struct vnode *a_vp;
2257 struct componentname *a_cnp;
2258 } */ *ap)
2259 {
2260 struct vnode *vp = ap->a_vp;
2261 struct vnode *dvp = ap->a_dvp;
2262 struct componentname *cnp = ap->a_cnp;
2263 struct inode *ip, *dp;
2264 int error;
2265
2266 ip = VTOI(vp);
2267 dp = VTOI(dvp);
2268
2269 /*
2270 * Do not remove a directory that is in the process of being renamed.
2271 * Verify the directory is empty (and valid). Rmdir ".." will not be
2272 * valid since ".." will contain a reference to the current directory
2273 * and thus be non-empty. Do not allow the removal of mounted on
2274 * directories (this can happen when an NFS exported filesystem
2275 * tries to remove a locally mounted on directory).
2276 */
2277 error = 0;
2278 if (dp->i_effnlink <= 2) {
2279 if (dp->i_effnlink == 2)
2280 print_bad_link_count("ufs_rmdir", dvp);
2281 error = EINVAL;
2282 goto out;
2283 }
2284 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
2285 error = ENOTEMPTY;
2286 goto out;
2287 }
2288 if ((dp->i_flags & APPEND)
2289 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
2290 error = EPERM;
2291 goto out;
2292 }
2293 if (vp->v_mountedhere != 0) {
2294 error = EINVAL;
2295 goto out;
2296 }
2297 if (DOINGSUJ(dvp)) {
2298 error = softdep_prelink(dvp, vp, cnp);
2299 if (error != 0) {
2300 MPASS(error == ERELOOKUP);
2301 return (error);
2302 }
2303 }
2304
2305 #ifdef UFS_GJOURNAL
2306 ufs_gjournal_orphan(vp);
2307 #endif
2308 /*
2309 * Delete reference to directory before purging
2310 * inode. If we crash in between, the directory
2311 * will be reattached to lost+found,
2312 */
2313 dp->i_effnlink--;
2314 ip->i_effnlink--;
2315 if (DOINGSOFTDEP(vp))
2316 softdep_setup_rmdir(dp, ip);
2317 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
2318 if (error) {
2319 dp->i_effnlink++;
2320 ip->i_effnlink++;
2321 if (DOINGSOFTDEP(vp))
2322 softdep_revert_rmdir(dp, ip);
2323 goto out;
2324 }
2325 /*
2326 * The only stuff left in the directory is "." and "..". The "."
2327 * reference is inconsequential since we are quashing it. The soft
2328 * dependency code will arrange to do these operations after
2329 * the parent directory entry has been deleted on disk, so
2330 * when running with that code we avoid doing them now.
2331 */
2332 if (!DOINGSOFTDEP(vp)) {
2333 dp->i_nlink--;
2334 DIP_SET_NLINK(dp, dp->i_nlink);
2335 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2336 error = UFS_UPDATE(dvp, 0);
2337 ip->i_nlink--;
2338 DIP_SET_NLINK(ip, ip->i_nlink);
2339 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2340 }
2341 cache_vop_rmdir(dvp, vp);
2342 #ifdef UFS_DIRHASH
2343 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
2344 if (ip->i_dirhash != NULL)
2345 ufsdirhash_free(ip);
2346 #endif
2347 out:
2348 return (error);
2349 }
2350
2351 /*
2352 * symlink -- make a symbolic link
2353 */
2354 static int
ufs_symlink(struct vop_symlink_args * ap)2355 ufs_symlink(
2356 struct vop_symlink_args /* {
2357 struct vnode *a_dvp;
2358 struct vnode **a_vpp;
2359 struct componentname *a_cnp;
2360 struct vattr *a_vap;
2361 const char *a_target;
2362 } */ *ap)
2363 {
2364 struct vnode *vp, **vpp = ap->a_vpp;
2365 struct inode *ip;
2366 int len, error;
2367
2368 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
2369 vpp, ap->a_cnp, "ufs_symlink");
2370 if (error)
2371 return (error);
2372 vp = *vpp;
2373 len = strlen(ap->a_target);
2374 if (len < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) {
2375 ip = VTOI(vp);
2376 bcopy(ap->a_target, DIP(ip, i_shortlink), len);
2377 ip->i_size = len;
2378 DIP_SET(ip, i_size, len);
2379 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2380 error = UFS_UPDATE(vp, 0);
2381 } else
2382 error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
2383 len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
2384 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
2385 if (error)
2386 vput(vp);
2387 return (error);
2388 }
2389
2390 /*
2391 * Vnode op for reading directories.
2392 */
2393 int
ufs_readdir(struct vop_readdir_args * ap)2394 ufs_readdir(
2395 struct vop_readdir_args /* {
2396 struct vnode *a_vp;
2397 struct uio *a_uio;
2398 struct ucred *a_cred;
2399 int *a_eofflag;
2400 int *a_ncookies;
2401 uint64_t **a_cookies;
2402 } */ *ap)
2403 {
2404 struct vnode *vp = ap->a_vp;
2405 struct uio *uio = ap->a_uio;
2406 struct buf *bp;
2407 struct inode *ip;
2408 struct direct *dp, *edp;
2409 uint64_t *cookies;
2410 struct dirent dstdp;
2411 off_t offset, startoffset;
2412 size_t readcnt, skipcnt;
2413 ssize_t startresid;
2414 uint64_t ncookies;
2415 int error;
2416
2417 if (uio->uio_offset < 0)
2418 return (EINVAL);
2419 ip = VTOI(vp);
2420 if (ip->i_effnlink == 0) {
2421 *ap->a_eofflag = 1;
2422 return (0);
2423 }
2424 if (ap->a_ncookies != NULL) {
2425 if (uio->uio_resid < 0)
2426 ncookies = 0;
2427 else
2428 ncookies = uio->uio_resid;
2429 if (uio->uio_offset >= ip->i_size)
2430 ncookies = 0;
2431 else if (ip->i_size - uio->uio_offset < ncookies)
2432 ncookies = ip->i_size - uio->uio_offset;
2433 ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1;
2434 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
2435 *ap->a_ncookies = ncookies;
2436 *ap->a_cookies = cookies;
2437 } else {
2438 ncookies = 0;
2439 cookies = NULL;
2440 }
2441 offset = startoffset = uio->uio_offset;
2442 startresid = uio->uio_resid;
2443 error = 0;
2444 while (error == 0 && uio->uio_resid > 0 &&
2445 uio->uio_offset < ip->i_size) {
2446 error = UFS_BLKATOFF(vp, uio->uio_offset, NULL, &bp);
2447 if (error)
2448 break;
2449 if (bp->b_offset + bp->b_bcount > ip->i_size)
2450 readcnt = ip->i_size - bp->b_offset;
2451 else
2452 readcnt = bp->b_bcount;
2453 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
2454 ~(size_t)(DIRBLKSIZ - 1);
2455 offset = bp->b_offset + skipcnt;
2456 dp = (struct direct *)&bp->b_data[skipcnt];
2457 edp = (struct direct *)&bp->b_data[readcnt];
2458 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
2459 if (dp->d_reclen <= offsetof(struct direct, d_name) ||
2460 (caddr_t)dp + dp->d_reclen > (caddr_t)edp) {
2461 error = EIO;
2462 break;
2463 }
2464 #if BYTE_ORDER == LITTLE_ENDIAN
2465 /* Old filesystem format. */
2466 if (OFSFMT(vp)) {
2467 dstdp.d_namlen = dp->d_type;
2468 dstdp.d_type = dp->d_namlen;
2469 } else
2470 #endif
2471 {
2472 dstdp.d_namlen = dp->d_namlen;
2473 dstdp.d_type = dp->d_type;
2474 }
2475 if (offsetof(struct direct, d_name) + dstdp.d_namlen >
2476 dp->d_reclen) {
2477 error = EIO;
2478 break;
2479 }
2480 if (offset < startoffset || dp->d_ino == 0)
2481 goto nextentry;
2482 dstdp.d_fileno = dp->d_ino;
2483 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
2484 bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen);
2485 /* NOTE: d_off is the offset of the *next* entry. */
2486 dstdp.d_off = offset + dp->d_reclen;
2487 dirent_terminate(&dstdp);
2488 if (dstdp.d_reclen > uio->uio_resid) {
2489 if (uio->uio_resid == startresid)
2490 error = EINVAL;
2491 else
2492 error = EJUSTRETURN;
2493 break;
2494 }
2495 /* Advance dp. */
2496 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
2497 if (error)
2498 break;
2499 if (cookies != NULL) {
2500 KASSERT(ncookies > 0,
2501 ("ufs_readdir: cookies buffer too small"));
2502 *cookies = offset + dp->d_reclen;
2503 cookies++;
2504 ncookies--;
2505 }
2506 nextentry:
2507 offset += dp->d_reclen;
2508 dp = (struct direct *)((caddr_t)dp + dp->d_reclen);
2509 }
2510 bqrelse(bp);
2511 uio->uio_offset = offset;
2512 }
2513 /* We need to correct uio_offset. */
2514 uio->uio_offset = offset;
2515 if (error == EJUSTRETURN)
2516 error = 0;
2517 if (ap->a_ncookies != NULL) {
2518 if (error == 0) {
2519 *ap->a_ncookies -= ncookies;
2520 } else {
2521 free(*ap->a_cookies, M_TEMP);
2522 *ap->a_ncookies = 0;
2523 *ap->a_cookies = NULL;
2524 }
2525 }
2526 if (error == 0 && ap->a_eofflag)
2527 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
2528 return (error);
2529 }
2530
2531 /*
2532 * Return target name of a symbolic link
2533 */
2534 static int
ufs_readlink(struct vop_readlink_args * ap)2535 ufs_readlink(
2536 struct vop_readlink_args /* {
2537 struct vnode *a_vp;
2538 struct uio *a_uio;
2539 struct ucred *a_cred;
2540 } */ *ap)
2541 {
2542 struct vnode *vp = ap->a_vp;
2543 struct inode *ip = VTOI(vp);
2544 doff_t isize;
2545
2546 isize = ip->i_size;
2547 if (isize < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
2548 return (uiomove(DIP(ip, i_shortlink), isize, ap->a_uio));
2549 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
2550 }
2551
2552 /*
2553 * Calculate the logical to physical mapping if not done already,
2554 * then call the device strategy routine.
2555 *
2556 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
2557 * deadlock on memory. See ufs_bmap() for details.
2558 */
2559 static int
ufs_strategy(struct vop_strategy_args * ap)2560 ufs_strategy(
2561 struct vop_strategy_args /* {
2562 struct vnode *a_vp;
2563 struct buf *a_bp;
2564 } */ *ap)
2565 {
2566 struct buf *bp = ap->a_bp;
2567 struct vnode *vp = ap->a_vp;
2568 ufs2_daddr_t blkno;
2569 int error;
2570
2571 if (bp->b_blkno == bp->b_lblkno) {
2572 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2573 bp->b_blkno = blkno;
2574 if (error) {
2575 bp->b_error = error;
2576 bp->b_ioflags |= BIO_ERROR;
2577 bufdone(bp);
2578 return (0);
2579 }
2580 if ((long)bp->b_blkno == -1)
2581 vfs_bio_clrbuf(bp);
2582 }
2583 if ((long)bp->b_blkno == -1) {
2584 bufdone(bp);
2585 return (0);
2586 }
2587 bp->b_iooffset = dbtob(bp->b_blkno);
2588 BO_STRATEGY(VFSTOUFS(vp->v_mount)->um_bo, bp);
2589 return (0);
2590 }
2591
2592 /*
2593 * Print out the contents of an inode.
2594 */
2595 static int
ufs_print(struct vop_print_args * ap)2596 ufs_print(
2597 struct vop_print_args /* {
2598 struct vnode *a_vp;
2599 } */ *ap)
2600 {
2601 struct vnode *vp = ap->a_vp;
2602 struct inode *ip = VTOI(vp);
2603
2604 printf("\tnlink=%d, effnlink=%d, size=%jd", ip->i_nlink,
2605 ip->i_effnlink, (intmax_t)ip->i_size);
2606 if (I_IS_UFS2(ip))
2607 printf(", extsize %d", ip->i_din2->di_extsize);
2608 printf("\n\tgeneration=%jx, uid=%d, gid=%d, flags=0x%b\n",
2609 (uintmax_t)ip->i_gen, ip->i_uid, ip->i_gid,
2610 (uint32_t)ip->i_flags, PRINT_INODE_FLAGS);
2611 printf("\tino %ju, on dev %s", (intmax_t)ip->i_number,
2612 devtoname(ITODEV(ip)));
2613 if (vp->v_type == VFIFO)
2614 fifo_printinfo(vp);
2615 printf("\n");
2616 return (0);
2617 }
2618
2619 /*
2620 * Close wrapper for fifos.
2621 *
2622 * Update the times on the inode then do device close.
2623 */
2624 static int
ufsfifo_close(struct vop_close_args * ap)2625 ufsfifo_close(
2626 struct vop_close_args /* {
2627 struct vnode *a_vp;
2628 int a_fflag;
2629 struct ucred *a_cred;
2630 struct thread *a_td;
2631 } */ *ap)
2632 {
2633
2634 ufs_close(ap);
2635 return (fifo_specops.vop_close(ap));
2636 }
2637
2638 /*
2639 * Return POSIX pathconf information applicable to ufs filesystems.
2640 */
2641 static int
ufs_pathconf(struct vop_pathconf_args * ap)2642 ufs_pathconf(
2643 struct vop_pathconf_args /* {
2644 struct vnode *a_vp;
2645 int a_name;
2646 int *a_retval;
2647 } */ *ap)
2648 {
2649 int error;
2650
2651 error = 0;
2652 switch (ap->a_name) {
2653 case _PC_LINK_MAX:
2654 *ap->a_retval = UFS_LINK_MAX;
2655 break;
2656 case _PC_NAME_MAX:
2657 *ap->a_retval = UFS_MAXNAMLEN;
2658 break;
2659 case _PC_PIPE_BUF:
2660 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
2661 *ap->a_retval = PIPE_BUF;
2662 else
2663 error = EINVAL;
2664 break;
2665 case _PC_CHOWN_RESTRICTED:
2666 *ap->a_retval = 1;
2667 break;
2668 case _PC_NO_TRUNC:
2669 *ap->a_retval = 1;
2670 break;
2671 #ifdef UFS_ACL
2672 case _PC_ACL_EXTENDED:
2673 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2674 *ap->a_retval = 1;
2675 else
2676 *ap->a_retval = 0;
2677 break;
2678 case _PC_ACL_NFS4:
2679 if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS)
2680 *ap->a_retval = 1;
2681 else
2682 *ap->a_retval = 0;
2683 break;
2684 #endif
2685 case _PC_ACL_PATH_MAX:
2686 #ifdef UFS_ACL
2687 if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS))
2688 *ap->a_retval = ACL_MAX_ENTRIES;
2689 else
2690 *ap->a_retval = 3;
2691 #else
2692 *ap->a_retval = 3;
2693 #endif
2694 break;
2695 #ifdef MAC
2696 case _PC_MAC_PRESENT:
2697 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2698 *ap->a_retval = 1;
2699 else
2700 *ap->a_retval = 0;
2701 break;
2702 #endif
2703 case _PC_MIN_HOLE_SIZE:
2704 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2705 break;
2706 case _PC_PRIO_IO:
2707 *ap->a_retval = 0;
2708 break;
2709 case _PC_SYNC_IO:
2710 *ap->a_retval = 0;
2711 break;
2712 case _PC_ALLOC_SIZE_MIN:
2713 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2714 break;
2715 case _PC_FILESIZEBITS:
2716 *ap->a_retval = 64;
2717 break;
2718 case _PC_REC_INCR_XFER_SIZE:
2719 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2720 break;
2721 case _PC_REC_MAX_XFER_SIZE:
2722 *ap->a_retval = -1; /* means ``unlimited'' */
2723 break;
2724 case _PC_REC_MIN_XFER_SIZE:
2725 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2726 break;
2727 case _PC_REC_XFER_ALIGN:
2728 *ap->a_retval = PAGE_SIZE;
2729 break;
2730 case _PC_SYMLINK_MAX:
2731 *ap->a_retval = MAXPATHLEN;
2732 break;
2733
2734 default:
2735 error = vop_stdpathconf(ap);
2736 break;
2737 }
2738 return (error);
2739 }
2740
2741 /*
2742 * Initialize the vnode associated with a new inode, handle aliased
2743 * vnodes.
2744 */
2745 int
ufs_vinit(struct mount * mntp,struct vop_vector * fifoops,struct vnode ** vpp)2746 ufs_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
2747 {
2748 struct inode *ip;
2749 struct vnode *vp;
2750
2751 vp = *vpp;
2752 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2753 ip = VTOI(vp);
2754 vp->v_type = IFTOVT(ip->i_mode);
2755 /*
2756 * Only unallocated inodes should be of type VNON.
2757 */
2758 if (ip->i_mode != 0 && vp->v_type == VNON)
2759 return (EINVAL);
2760 if (vp->v_type == VFIFO)
2761 vp->v_op = fifoops;
2762 if (ip->i_number == UFS_ROOTINO)
2763 vp->v_vflag |= VV_ROOT;
2764 *vpp = vp;
2765 return (0);
2766 }
2767
2768 /*
2769 * Allocate a new inode.
2770 * Vnode dvp must be locked.
2771 */
2772 static int
ufs_makeinode(int mode,struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,const char * callfunc)2773 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2774 struct componentname *cnp, const char *callfunc)
2775 {
2776 struct inode *ip, *pdir;
2777 struct direct newdir;
2778 struct vnode *tvp;
2779 int error;
2780
2781 pdir = VTOI(dvp);
2782 *vpp = NULL;
2783 if ((mode & IFMT) == 0)
2784 mode |= IFREG;
2785
2786 if (pdir->i_effnlink < 2) {
2787 print_bad_link_count(callfunc, dvp);
2788 return (EINVAL);
2789 }
2790 if (DOINGSUJ(dvp)) {
2791 error = softdep_prelink(dvp, NULL, cnp);
2792 if (error != 0) {
2793 MPASS(error == ERELOOKUP);
2794 return (error);
2795 }
2796 }
2797 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2798 if (error)
2799 return (error);
2800 ip = VTOI(tvp);
2801 ip->i_gid = pdir->i_gid;
2802 DIP_SET(ip, i_gid, pdir->i_gid);
2803 #ifdef SUIDDIR
2804 {
2805 #ifdef QUOTA
2806 struct ucred ucred, *ucp;
2807 gid_t ucred_group;
2808 ucp = cnp->cn_cred;
2809 #endif
2810 /*
2811 * If we are not the owner of the directory,
2812 * and we are hacking owners here, (only do this where told to)
2813 * and we are not giving it TO root, (would subvert quotas)
2814 * then go ahead and give it to the other user.
2815 * Note that this drops off the execute bits for security.
2816 */
2817 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2818 (pdir->i_mode & ISUID) &&
2819 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2820 ip->i_uid = pdir->i_uid;
2821 DIP_SET(ip, i_uid, ip->i_uid);
2822 mode &= ~07111;
2823 #ifdef QUOTA
2824 /*
2825 * Make sure the correct user gets charged
2826 * for the space.
2827 * Quickly knock up a dummy credential for the victim.
2828 * XXX This seems to never be accessed out of our
2829 * context so a stack variable is ok.
2830 */
2831 ucred.cr_ref = 1;
2832 ucred.cr_uid = ip->i_uid;
2833 ucred.cr_ngroups = 1;
2834 ucred.cr_groups = &ucred_group;
2835 ucred.cr_groups[0] = pdir->i_gid;
2836 ucp = &ucred;
2837 #endif
2838 } else {
2839 ip->i_uid = cnp->cn_cred->cr_uid;
2840 DIP_SET(ip, i_uid, ip->i_uid);
2841 }
2842
2843 #ifdef QUOTA
2844 if ((error = getinoquota(ip)) ||
2845 (error = chkiq(ip, 1, ucp, 0))) {
2846 if (DOINGSOFTDEP(tvp))
2847 softdep_revert_link(pdir, ip);
2848 UFS_VFREE(tvp, ip->i_number, mode);
2849 vgone(tvp);
2850 vput(tvp);
2851 return (error);
2852 }
2853 #endif
2854 }
2855 #else /* !SUIDDIR */
2856 ip->i_uid = cnp->cn_cred->cr_uid;
2857 DIP_SET(ip, i_uid, ip->i_uid);
2858 #ifdef QUOTA
2859 if ((error = getinoquota(ip)) ||
2860 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2861 if (DOINGSOFTDEP(tvp))
2862 softdep_revert_link(pdir, ip);
2863 UFS_VFREE(tvp, ip->i_number, mode);
2864 vgone(tvp);
2865 vput(tvp);
2866 return (error);
2867 }
2868 #endif
2869 #endif /* !SUIDDIR */
2870 vn_seqc_write_begin(tvp); /* Mostly to cover asserts */
2871 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2872 UFS_INODE_SET_MODE(ip, mode);
2873 DIP_SET(ip, i_mode, mode);
2874 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2875 ip->i_effnlink = 1;
2876 ip->i_nlink = 1;
2877 DIP_SET_NLINK(ip, 1);
2878 if (DOINGSOFTDEP(tvp))
2879 softdep_setup_create(VTOI(dvp), ip);
2880 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2881 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID)) {
2882 UFS_INODE_SET_MODE(ip, ip->i_mode & ~ISGID);
2883 DIP_SET(ip, i_mode, ip->i_mode);
2884 }
2885
2886 if (cnp->cn_flags & ISWHITEOUT) {
2887 ip->i_flags |= UF_OPAQUE;
2888 DIP_SET(ip, i_flags, ip->i_flags);
2889 }
2890
2891 /*
2892 * Make sure inode goes to disk before directory entry.
2893 */
2894 error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp));
2895 if (error)
2896 goto bad;
2897 #ifdef MAC
2898 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2899 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2900 dvp, tvp, cnp);
2901 if (error)
2902 goto bad;
2903 }
2904 #endif
2905 #ifdef UFS_ACL
2906 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2907 error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2908 cnp->cn_cred, curthread);
2909 if (error)
2910 goto bad;
2911 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2912 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode,
2913 cnp->cn_cred, curthread);
2914 if (error)
2915 goto bad;
2916 }
2917 #endif /* !UFS_ACL */
2918 ufs_makedirentry(ip, cnp, &newdir);
2919 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2920 if (error)
2921 goto bad;
2922 vn_seqc_write_end(tvp);
2923 *vpp = tvp;
2924 return (0);
2925
2926 bad:
2927 /*
2928 * Write error occurred trying to update the inode
2929 * or the directory so must deallocate the inode.
2930 */
2931 ip->i_effnlink = 0;
2932 ip->i_nlink = 0;
2933 DIP_SET_NLINK(ip, 0);
2934 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2935 if (DOINGSOFTDEP(tvp))
2936 softdep_revert_create(VTOI(dvp), ip);
2937 vn_seqc_write_end(tvp);
2938 vgone(tvp);
2939 vput(tvp);
2940 return (error);
2941 }
2942
2943 static int
ufs_ioctl(struct vop_ioctl_args * ap)2944 ufs_ioctl(struct vop_ioctl_args *ap)
2945 {
2946 struct vnode *vp;
2947 int error;
2948
2949 vp = ap->a_vp;
2950 switch (ap->a_command) {
2951 case FIOSEEKDATA:
2952 error = vn_lock(vp, LK_EXCLUSIVE);
2953 if (error == 0) {
2954 error = ufs_bmap_seekdata(vp, (off_t *)ap->a_data);
2955 VOP_UNLOCK(vp);
2956 } else
2957 error = EBADF;
2958 return (error);
2959 case FIOSEEKHOLE:
2960 return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data,
2961 ap->a_cred));
2962 default:
2963 return (ENOTTY);
2964 }
2965 }
2966
2967 static int
ufs_read_pgcache(struct vop_read_pgcache_args * ap)2968 ufs_read_pgcache(struct vop_read_pgcache_args *ap)
2969 {
2970 struct uio *uio;
2971 struct vnode *vp;
2972
2973 uio = ap->a_uio;
2974 vp = ap->a_vp;
2975 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp);
2976
2977 if (uio->uio_resid > ptoa(io_hold_cnt) || uio->uio_offset < 0 ||
2978 (ap->a_ioflag & IO_DIRECT) != 0)
2979 return (EJUSTRETURN);
2980 return (vn_read_from_obj(vp, uio));
2981 }
2982
2983 /* Global vfs data structures for ufs. */
2984 struct vop_vector ufs_vnodeops = {
2985 .vop_default = &default_vnodeops,
2986 .vop_fsync = VOP_PANIC,
2987 .vop_read = VOP_PANIC,
2988 .vop_reallocblks = VOP_PANIC,
2989 .vop_write = VOP_PANIC,
2990 .vop_accessx = ufs_accessx,
2991 .vop_bmap = ufs_bmap,
2992 .vop_fplookup_vexec = ufs_fplookup_vexec,
2993 .vop_fplookup_symlink = VOP_EAGAIN,
2994 .vop_cachedlookup = ufs_lookup,
2995 .vop_close = ufs_close,
2996 .vop_create = ufs_create,
2997 .vop_stat = ufs_stat,
2998 .vop_getattr = ufs_getattr,
2999 .vop_inactive = ufs_inactive,
3000 .vop_ioctl = ufs_ioctl,
3001 .vop_link = ufs_link,
3002 .vop_lookup = vfs_cache_lookup,
3003 .vop_mmapped = ufs_mmapped,
3004 .vop_mkdir = ufs_mkdir,
3005 .vop_mknod = ufs_mknod,
3006 .vop_need_inactive = ufs_need_inactive,
3007 .vop_open = ufs_open,
3008 .vop_pathconf = ufs_pathconf,
3009 .vop_poll = vop_stdpoll,
3010 .vop_print = ufs_print,
3011 .vop_read_pgcache = ufs_read_pgcache,
3012 .vop_readdir = ufs_readdir,
3013 .vop_readlink = ufs_readlink,
3014 .vop_reclaim = ufs_reclaim,
3015 .vop_remove = ufs_remove,
3016 .vop_rename = ufs_rename,
3017 .vop_rmdir = ufs_rmdir,
3018 .vop_setattr = ufs_setattr,
3019 #ifdef MAC
3020 .vop_setlabel = vop_stdsetlabel_ea,
3021 #endif
3022 .vop_strategy = ufs_strategy,
3023 .vop_symlink = ufs_symlink,
3024 .vop_whiteout = ufs_whiteout,
3025 #ifdef UFS_EXTATTR
3026 .vop_getextattr = ufs_getextattr,
3027 .vop_deleteextattr = ufs_deleteextattr,
3028 .vop_setextattr = ufs_setextattr,
3029 #endif
3030 #ifdef UFS_ACL
3031 .vop_getacl = ufs_getacl,
3032 .vop_setacl = ufs_setacl,
3033 .vop_aclcheck = ufs_aclcheck,
3034 #endif
3035 };
3036 VFS_VOP_VECTOR_REGISTER(ufs_vnodeops);
3037
3038 struct vop_vector ufs_fifoops = {
3039 .vop_default = &fifo_specops,
3040 .vop_fsync = VOP_PANIC,
3041 .vop_accessx = ufs_accessx,
3042 .vop_close = ufsfifo_close,
3043 .vop_getattr = ufs_getattr,
3044 .vop_inactive = ufs_inactive,
3045 .vop_pathconf = ufs_pathconf,
3046 .vop_print = ufs_print,
3047 .vop_read = VOP_PANIC,
3048 .vop_reclaim = ufs_reclaim,
3049 .vop_setattr = ufs_setattr,
3050 #ifdef MAC
3051 .vop_setlabel = vop_stdsetlabel_ea,
3052 #endif
3053 .vop_write = VOP_PANIC,
3054 #ifdef UFS_EXTATTR
3055 .vop_getextattr = ufs_getextattr,
3056 .vop_deleteextattr = ufs_deleteextattr,
3057 .vop_setextattr = ufs_setextattr,
3058 #endif
3059 #ifdef UFS_ACL
3060 .vop_getacl = ufs_getacl,
3061 .vop_setacl = ufs_setacl,
3062 .vop_aclcheck = ufs_aclcheck,
3063 #endif
3064 .vop_fplookup_vexec = VOP_EAGAIN,
3065 .vop_fplookup_symlink = VOP_EAGAIN,
3066 };
3067 VFS_VOP_VECTOR_REGISTER(ufs_fifoops);
3068