1 /*-
2 * modified for EXT2FS support in Lites 1.1
3 *
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
6 */
7 /*-
8 * SPDX-License-Identifier: BSD-3-Clause
9 *
10 * Copyright (c) 1982, 1986, 1989, 1993
11 * The Regents of the University of California. All rights reserved.
12 * (c) UNIX System Laboratories, Inc.
13 * All or some portions of this file are derived from material licensed
14 * to the University of California by American Telephone and Telegraph
15 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
16 * the permission of UNIX System Laboratories, Inc.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
20 * are met:
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
26 * 3. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
29 *
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * SUCH DAMAGE.
41 *
42 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94
43 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
44 * $FreeBSD$
45 */
46
47 #include "opt_suiddir.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/fcntl.h>
53 #include <sys/filio.h>
54 #include <sys/limits.h>
55 #include <sys/sdt.h>
56 #include <sys/stat.h>
57 #include <sys/bio.h>
58 #include <sys/buf.h>
59 #include <sys/endian.h>
60 #include <sys/priv.h>
61 #include <sys/rwlock.h>
62 #include <sys/mount.h>
63 #include <sys/unistd.h>
64 #include <sys/time.h>
65 #include <sys/vnode.h>
66 #include <sys/namei.h>
67 #include <sys/lockf.h>
68 #include <sys/event.h>
69 #include <sys/conf.h>
70 #include <sys/file.h>
71 #include <sys/extattr.h>
72 #include <sys/vmmeter.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
81
82 #include "opt_directio.h"
83
84 #include <ufs/ufs/dir.h>
85
86 #include <fs/ext2fs/fs.h>
87 #include <fs/ext2fs/inode.h>
88 #include <fs/ext2fs/ext2_acl.h>
89 #include <fs/ext2fs/ext2fs.h>
90 #include <fs/ext2fs/ext2_extern.h>
91 #include <fs/ext2fs/ext2_dinode.h>
92 #include <fs/ext2fs/ext2_dir.h>
93 #include <fs/ext2fs/ext2_mount.h>
94 #include <fs/ext2fs/ext2_extattr.h>
95 #include <fs/ext2fs/ext2_extents.h>
96
97 SDT_PROVIDER_DECLARE(ext2fs);
98 /*
99 * ext2fs trace probe:
100 * arg0: verbosity. Higher numbers give more verbose messages
101 * arg1: Textual message
102 */
103 SDT_PROBE_DEFINE2(ext2fs, , vnops, trace, "int", "char*");
104
105 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
106 static void ext2_itimes_locked(struct vnode *);
107
108 static vop_access_t ext2_access;
109 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
110 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
111 struct thread *);
112 static vop_close_t ext2_close;
113 static vop_create_t ext2_create;
114 static vop_fsync_t ext2_fsync;
115 static vop_getattr_t ext2_getattr;
116 static vop_ioctl_t ext2_ioctl;
117 static vop_link_t ext2_link;
118 static vop_mkdir_t ext2_mkdir;
119 static vop_mknod_t ext2_mknod;
120 static vop_open_t ext2_open;
121 static vop_pathconf_t ext2_pathconf;
122 static vop_print_t ext2_print;
123 static vop_read_t ext2_read;
124 static vop_readlink_t ext2_readlink;
125 static vop_remove_t ext2_remove;
126 static vop_rename_t ext2_rename;
127 static vop_rmdir_t ext2_rmdir;
128 static vop_setattr_t ext2_setattr;
129 static vop_strategy_t ext2_strategy;
130 static vop_symlink_t ext2_symlink;
131 static vop_write_t ext2_write;
132 static vop_deleteextattr_t ext2_deleteextattr;
133 static vop_getextattr_t ext2_getextattr;
134 static vop_listextattr_t ext2_listextattr;
135 static vop_setextattr_t ext2_setextattr;
136 static vop_vptofh_t ext2_vptofh;
137 static vop_close_t ext2fifo_close;
138 static vop_kqfilter_t ext2fifo_kqfilter;
139
140 /* Global vfs data structures for ext2. */
141 struct vop_vector ext2_vnodeops = {
142 .vop_default = &default_vnodeops,
143 .vop_access = ext2_access,
144 .vop_bmap = ext2_bmap,
145 .vop_cachedlookup = ext2_lookup,
146 .vop_close = ext2_close,
147 .vop_create = ext2_create,
148 .vop_fsync = ext2_fsync,
149 .vop_getpages = vnode_pager_local_getpages,
150 .vop_getpages_async = vnode_pager_local_getpages_async,
151 .vop_getattr = ext2_getattr,
152 .vop_inactive = ext2_inactive,
153 .vop_ioctl = ext2_ioctl,
154 .vop_link = ext2_link,
155 .vop_lookup = vfs_cache_lookup,
156 .vop_mkdir = ext2_mkdir,
157 .vop_mknod = ext2_mknod,
158 .vop_open = ext2_open,
159 .vop_pathconf = ext2_pathconf,
160 .vop_poll = vop_stdpoll,
161 .vop_print = ext2_print,
162 .vop_read = ext2_read,
163 .vop_readdir = ext2_readdir,
164 .vop_readlink = ext2_readlink,
165 .vop_reallocblks = ext2_reallocblks,
166 .vop_reclaim = ext2_reclaim,
167 .vop_remove = ext2_remove,
168 .vop_rename = ext2_rename,
169 .vop_rmdir = ext2_rmdir,
170 .vop_setattr = ext2_setattr,
171 .vop_strategy = ext2_strategy,
172 .vop_symlink = ext2_symlink,
173 .vop_write = ext2_write,
174 .vop_deleteextattr = ext2_deleteextattr,
175 .vop_getextattr = ext2_getextattr,
176 .vop_listextattr = ext2_listextattr,
177 .vop_setextattr = ext2_setextattr,
178 #ifdef UFS_ACL
179 .vop_getacl = ext2_getacl,
180 .vop_setacl = ext2_setacl,
181 .vop_aclcheck = ext2_aclcheck,
182 #endif /* UFS_ACL */
183 .vop_vptofh = ext2_vptofh,
184 };
185
186 struct vop_vector ext2_fifoops = {
187 .vop_default = &fifo_specops,
188 .vop_access = ext2_access,
189 .vop_close = ext2fifo_close,
190 .vop_fsync = ext2_fsync,
191 .vop_getattr = ext2_getattr,
192 .vop_inactive = ext2_inactive,
193 .vop_kqfilter = ext2fifo_kqfilter,
194 .vop_pathconf = ext2_pathconf,
195 .vop_print = ext2_print,
196 .vop_read = VOP_PANIC,
197 .vop_reclaim = ext2_reclaim,
198 .vop_setattr = ext2_setattr,
199 .vop_write = VOP_PANIC,
200 .vop_vptofh = ext2_vptofh,
201 };
202
203 /*
204 * A virgin directory (no blushing please).
205 * Note that the type and namlen fields are reversed relative to ext2.
206 * Also, we don't use `struct odirtemplate', since it would just cause
207 * endianness problems.
208 */
209 static struct dirtemplate mastertemplate = {
210 0, 12, 1, EXT2_FT_DIR, ".",
211 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
212 };
213 static struct dirtemplate omastertemplate = {
214 0, 12, 1, EXT2_FT_UNKNOWN, ".",
215 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
216 };
217
218 static void
219 ext2_itimes_locked(struct vnode *vp)
220 {
221 struct inode *ip;
222 struct timespec ts;
223
224 ASSERT_VI_LOCKED(vp, __func__);
225
226 ip = VTOI(vp);
227 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
228 return;
229 if ((vp->v_type == VBLK || vp->v_type == VCHR))
230 ip->i_flag |= IN_LAZYMOD;
231 else
232 ip->i_flag |= IN_MODIFIED;
233 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
234 vfs_timestamp(&ts);
235 if (ip->i_flag & IN_ACCESS) {
236 ip->i_atime = ts.tv_sec;
237 ip->i_atimensec = ts.tv_nsec;
238 }
239 if (ip->i_flag & IN_UPDATE) {
240 ip->i_mtime = ts.tv_sec;
241 ip->i_mtimensec = ts.tv_nsec;
242 ip->i_modrev++;
243 }
244 if (ip->i_flag & IN_CHANGE) {
245 ip->i_ctime = ts.tv_sec;
246 ip->i_ctimensec = ts.tv_nsec;
247 }
248 }
249 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
250 }
251
252 void
253 ext2_itimes(struct vnode *vp)
254 {
255
256 VI_LOCK(vp);
257 ext2_itimes_locked(vp);
258 VI_UNLOCK(vp);
259 }
260
261 /*
262 * Create a regular file
263 */
264 static int
265 ext2_create(struct vop_create_args *ap)
266 {
267 int error;
268
269 error =
270 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
271 ap->a_dvp, ap->a_vpp, ap->a_cnp);
272 if (error != 0)
273 return (error);
274 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
275 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
276 return (0);
277 }
278
279 static int
280 ext2_open(struct vop_open_args *ap)
281 {
282
283 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
284 return (EOPNOTSUPP);
285
286 /*
287 * Files marked append-only must be opened for appending.
288 */
289 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
290 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
291 return (EPERM);
292
293 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
294
295 return (0);
296 }
297
298 /*
299 * Close called.
300 *
301 * Update the times on the inode.
302 */
303 static int
304 ext2_close(struct vop_close_args *ap)
305 {
306 struct vnode *vp = ap->a_vp;
307
308 VI_LOCK(vp);
309 if (vp->v_usecount > 1)
310 ext2_itimes_locked(vp);
311 VI_UNLOCK(vp);
312 return (0);
313 }
314
315 static int
316 ext2_access(struct vop_access_args *ap)
317 {
318 struct vnode *vp = ap->a_vp;
319 struct inode *ip = VTOI(vp);
320 accmode_t accmode = ap->a_accmode;
321 int error;
322
323 if (vp->v_type == VBLK || vp->v_type == VCHR)
324 return (EOPNOTSUPP);
325
326 /*
327 * Disallow write attempts on read-only file systems;
328 * unless the file is a socket, fifo, or a block or
329 * character device resident on the file system.
330 */
331 if (accmode & VWRITE) {
332 switch (vp->v_type) {
333 case VDIR:
334 case VLNK:
335 case VREG:
336 if (vp->v_mount->mnt_flag & MNT_RDONLY)
337 return (EROFS);
338 break;
339 default:
340 break;
341 }
342 }
343
344 /* If immutable bit set, nobody gets to write it. */
345 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
346 return (EPERM);
347
348 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
349 ap->a_accmode, ap->a_cred, NULL);
350 return (error);
351 }
352
353 static int
354 ext2_getattr(struct vop_getattr_args *ap)
355 {
356 struct vnode *vp = ap->a_vp;
357 struct inode *ip = VTOI(vp);
358 struct vattr *vap = ap->a_vap;
359
360 ext2_itimes(vp);
361 /*
362 * Copy from inode table
363 */
364 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
365 vap->va_fileid = ip->i_number;
366 vap->va_mode = ip->i_mode & ~IFMT;
367 vap->va_nlink = ip->i_nlink;
368 vap->va_uid = ip->i_uid;
369 vap->va_gid = ip->i_gid;
370 vap->va_rdev = ip->i_rdev;
371 vap->va_size = ip->i_size;
372 vap->va_atime.tv_sec = ip->i_atime;
373 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
374 vap->va_mtime.tv_sec = ip->i_mtime;
375 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
376 vap->va_ctime.tv_sec = ip->i_ctime;
377 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
378 if E2DI_HAS_XTIME(ip) {
379 vap->va_birthtime.tv_sec = ip->i_birthtime;
380 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
381 }
382 vap->va_flags = ip->i_flags;
383 vap->va_gen = ip->i_gen;
384 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
385 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
386 vap->va_type = IFTOVT(ip->i_mode);
387 vap->va_filerev = ip->i_modrev;
388 return (0);
389 }
390
391 /*
392 * Set attribute vnode op. called from several syscalls
393 */
394 static int
395 ext2_setattr(struct vop_setattr_args *ap)
396 {
397 struct vattr *vap = ap->a_vap;
398 struct vnode *vp = ap->a_vp;
399 struct inode *ip = VTOI(vp);
400 struct ucred *cred = ap->a_cred;
401 struct thread *td = curthread;
402 int error;
403
404 /*
405 * Check for unsettable attributes.
406 */
407 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
408 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
409 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
410 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
411 return (EINVAL);
412 }
413 if (vap->va_flags != VNOVAL) {
414 /* Disallow flags not supported by ext2fs. */
415 if (vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
416 return (EOPNOTSUPP);
417
418 if (vp->v_mount->mnt_flag & MNT_RDONLY)
419 return (EROFS);
420 /*
421 * Callers may only modify the file flags on objects they
422 * have VADMIN rights for.
423 */
424 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
425 return (error);
426 /*
427 * Unprivileged processes and privileged processes in
428 * jail() are not permitted to unset system flags, or
429 * modify flags if any system flags are set.
430 * Privileged non-jail processes may not modify system flags
431 * if securelevel > 0 and any existing system flags are set.
432 */
433 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
434 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) {
435 error = securelevel_gt(cred, 0);
436 if (error)
437 return (error);
438 }
439 } else {
440 if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) ||
441 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
442 return (EPERM);
443 }
444 ip->i_flags = vap->va_flags;
445 ip->i_flag |= IN_CHANGE;
446 if (ip->i_flags & (IMMUTABLE | APPEND))
447 return (0);
448 }
449 if (ip->i_flags & (IMMUTABLE | APPEND))
450 return (EPERM);
451 /*
452 * Go through the fields and update iff not VNOVAL.
453 */
454 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
455 if (vp->v_mount->mnt_flag & MNT_RDONLY)
456 return (EROFS);
457 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
458 td)) != 0)
459 return (error);
460 }
461 if (vap->va_size != VNOVAL) {
462 /*
463 * Disallow write attempts on read-only file systems;
464 * unless the file is a socket, fifo, or a block or
465 * character device resident on the file system.
466 */
467 switch (vp->v_type) {
468 case VDIR:
469 return (EISDIR);
470 case VLNK:
471 case VREG:
472 if (vp->v_mount->mnt_flag & MNT_RDONLY)
473 return (EROFS);
474 break;
475 default:
476 break;
477 }
478 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
479 return (error);
480 }
481 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
482 if (vp->v_mount->mnt_flag & MNT_RDONLY)
483 return (EROFS);
484 /*
485 * From utimes(2):
486 * If times is NULL, ... The caller must be the owner of
487 * the file, have permission to write the file, or be the
488 * super-user.
489 * If times is non-NULL, ... The caller must be the owner of
490 * the file or be the super-user.
491 */
492 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
493 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
494 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
495 return (error);
496 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
497 if (vap->va_atime.tv_sec != VNOVAL) {
498 ip->i_flag &= ~IN_ACCESS;
499 ip->i_atime = vap->va_atime.tv_sec;
500 ip->i_atimensec = vap->va_atime.tv_nsec;
501 }
502 if (vap->va_mtime.tv_sec != VNOVAL) {
503 ip->i_flag &= ~IN_UPDATE;
504 ip->i_mtime = vap->va_mtime.tv_sec;
505 ip->i_mtimensec = vap->va_mtime.tv_nsec;
506 }
507 ip->i_birthtime = vap->va_birthtime.tv_sec;
508 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
509 error = ext2_update(vp, 0);
510 if (error)
511 return (error);
512 }
513 error = 0;
514 if (vap->va_mode != (mode_t)VNOVAL) {
515 if (vp->v_mount->mnt_flag & MNT_RDONLY)
516 return (EROFS);
517 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
518 }
519 return (error);
520 }
521
522 /*
523 * Change the mode on a file.
524 * Inode must be locked before calling.
525 */
526 static int
527 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
528 {
529 struct inode *ip = VTOI(vp);
530 int error;
531
532 /*
533 * To modify the permissions on a file, must possess VADMIN
534 * for that file.
535 */
536 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
537 return (error);
538 /*
539 * Privileged processes may set the sticky bit on non-directories,
540 * as well as set the setgid bit on a file with a group that the
541 * process is not a member of.
542 */
543 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
544 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
545 if (error)
546 return (EFTYPE);
547 }
548 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
549 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
550 if (error)
551 return (error);
552 }
553 ip->i_mode &= ~ALLPERMS;
554 ip->i_mode |= (mode & ALLPERMS);
555 ip->i_flag |= IN_CHANGE;
556 return (0);
557 }
558
559 /*
560 * Perform chown operation on inode ip;
561 * inode must be locked prior to call.
562 */
563 static int
564 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
565 struct thread *td)
566 {
567 struct inode *ip = VTOI(vp);
568 uid_t ouid;
569 gid_t ogid;
570 int error = 0;
571
572 if (uid == (uid_t)VNOVAL)
573 uid = ip->i_uid;
574 if (gid == (gid_t)VNOVAL)
575 gid = ip->i_gid;
576 /*
577 * To modify the ownership of a file, must possess VADMIN
578 * for that file.
579 */
580 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
581 return (error);
582 /*
583 * To change the owner of a file, or change the group of a file
584 * to a group of which we are not a member, the caller must
585 * have privilege.
586 */
587 if (uid != ip->i_uid || (gid != ip->i_gid &&
588 !groupmember(gid, cred))) {
589 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
590 if (error)
591 return (error);
592 }
593 ogid = ip->i_gid;
594 ouid = ip->i_uid;
595 ip->i_gid = gid;
596 ip->i_uid = uid;
597 ip->i_flag |= IN_CHANGE;
598 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
599 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
600 ip->i_mode &= ~(ISUID | ISGID);
601 }
602 return (0);
603 }
604
605 /*
606 * Synch an open file.
607 */
608 /* ARGSUSED */
609 static int
610 ext2_fsync(struct vop_fsync_args *ap)
611 {
612 /*
613 * Flush all dirty buffers associated with a vnode.
614 */
615
616 vop_stdfsync(ap);
617
618 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
619 }
620
621 /*
622 * Mknod vnode call
623 */
624 /* ARGSUSED */
625 static int
626 ext2_mknod(struct vop_mknod_args *ap)
627 {
628 struct vattr *vap = ap->a_vap;
629 struct vnode **vpp = ap->a_vpp;
630 struct inode *ip;
631 ino_t ino;
632 int error;
633
634 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
635 ap->a_dvp, vpp, ap->a_cnp);
636 if (error)
637 return (error);
638 ip = VTOI(*vpp);
639 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
640 if (vap->va_rdev != VNOVAL) {
641 /*
642 * Want to be able to use this to make badblock
643 * inodes, so don't truncate the dev number.
644 */
645 if (!(ip->i_flag & IN_E4EXTENTS))
646 ip->i_rdev = vap->va_rdev;
647 }
648 /*
649 * Remove inode, then reload it through VFS_VGET so it is
650 * checked to see if it is an alias of an existing entry in
651 * the inode cache. XXX I don't believe this is necessary now.
652 */
653 (*vpp)->v_type = VNON;
654 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
655 vgone(*vpp);
656 vput(*vpp);
657 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
658 if (error) {
659 *vpp = NULL;
660 return (error);
661 }
662 return (0);
663 }
664
665 static int
666 ext2_remove(struct vop_remove_args *ap)
667 {
668 struct inode *ip;
669 struct vnode *vp = ap->a_vp;
670 struct vnode *dvp = ap->a_dvp;
671 int error;
672
673 ip = VTOI(vp);
674 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
675 (VTOI(dvp)->i_flags & APPEND)) {
676 error = EPERM;
677 goto out;
678 }
679 error = ext2_dirremove(dvp, ap->a_cnp);
680 if (error == 0) {
681 ip->i_nlink--;
682 ip->i_flag |= IN_CHANGE;
683 }
684 out:
685 return (error);
686 }
687
688 /*
689 * link vnode call
690 */
691 static int
692 ext2_link(struct vop_link_args *ap)
693 {
694 struct vnode *vp = ap->a_vp;
695 struct vnode *tdvp = ap->a_tdvp;
696 struct componentname *cnp = ap->a_cnp;
697 struct inode *ip;
698 int error;
699
700 #ifdef INVARIANTS
701 if ((cnp->cn_flags & HASBUF) == 0)
702 panic("ext2_link: no name");
703 #endif
704 ip = VTOI(vp);
705 if ((nlink_t)ip->i_nlink >= EXT4_LINK_MAX) {
706 error = EMLINK;
707 goto out;
708 }
709 if (ip->i_flags & (IMMUTABLE | APPEND)) {
710 error = EPERM;
711 goto out;
712 }
713 ip->i_nlink++;
714 ip->i_flag |= IN_CHANGE;
715 error = ext2_update(vp, !DOINGASYNC(vp));
716 if (!error)
717 error = ext2_direnter(ip, tdvp, cnp);
718 if (error) {
719 ip->i_nlink--;
720 ip->i_flag |= IN_CHANGE;
721 }
722 out:
723 return (error);
724 }
725
726 static int
727 ext2_inc_nlink(struct inode *ip)
728 {
729
730 ip->i_nlink++;
731
732 if (S_ISDIR(ip->i_mode) &&
733 EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK) &&
734 ip->i_nlink > 1) {
735 if (ip->i_nlink >= EXT4_LINK_MAX || ip->i_nlink == 2)
736 ip->i_nlink = 1;
737 } else if (ip->i_nlink > EXT4_LINK_MAX) {
738 ip->i_nlink--;
739 return (EMLINK);
740 }
741
742 return (0);
743 }
744
745 static void
746 ext2_dec_nlink(struct inode *ip)
747 {
748
749 if (!S_ISDIR(ip->i_mode) || ip->i_nlink > 2)
750 ip->i_nlink--;
751 }
752
753 /*
754 * Rename system call.
755 * rename("foo", "bar");
756 * is essentially
757 * unlink("bar");
758 * link("foo", "bar");
759 * unlink("foo");
760 * but ``atomically''. Can't do full commit without saving state in the
761 * inode on disk which isn't feasible at this time. Best we can do is
762 * always guarantee the target exists.
763 *
764 * Basic algorithm is:
765 *
766 * 1) Bump link count on source while we're linking it to the
767 * target. This also ensure the inode won't be deleted out
768 * from underneath us while we work (it may be truncated by
769 * a concurrent `trunc' or `open' for creation).
770 * 2) Link source to destination. If destination already exists,
771 * delete it first.
772 * 3) Unlink source reference to inode if still around. If a
773 * directory was moved and the parent of the destination
774 * is different from the source, patch the ".." entry in the
775 * directory.
776 */
777 static int
778 ext2_rename(struct vop_rename_args *ap)
779 {
780 struct vnode *tvp = ap->a_tvp;
781 struct vnode *tdvp = ap->a_tdvp;
782 struct vnode *fvp = ap->a_fvp;
783 struct vnode *fdvp = ap->a_fdvp;
784 struct componentname *tcnp = ap->a_tcnp;
785 struct componentname *fcnp = ap->a_fcnp;
786 struct inode *ip, *xp, *dp;
787 struct dirtemplate *dirbuf;
788 int doingdirectory = 0, oldparent = 0, newparent = 0;
789 int error = 0;
790 u_char namlen;
791
792 #ifdef INVARIANTS
793 if ((tcnp->cn_flags & HASBUF) == 0 ||
794 (fcnp->cn_flags & HASBUF) == 0)
795 panic("ext2_rename: no name");
796 #endif
797 /*
798 * Check for cross-device rename.
799 */
800 if ((fvp->v_mount != tdvp->v_mount) ||
801 (tvp && (fvp->v_mount != tvp->v_mount))) {
802 error = EXDEV;
803 abortit:
804 if (tdvp == tvp)
805 vrele(tdvp);
806 else
807 vput(tdvp);
808 if (tvp)
809 vput(tvp);
810 vrele(fdvp);
811 vrele(fvp);
812 return (error);
813 }
814
815 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
816 (VTOI(tdvp)->i_flags & APPEND))) {
817 error = EPERM;
818 goto abortit;
819 }
820
821 /*
822 * Renaming a file to itself has no effect. The upper layers should
823 * not call us in that case. Temporarily just warn if they do.
824 */
825 if (fvp == tvp) {
826 SDT_PROBE2(ext2fs, , vnops, trace, 1,
827 "rename: fvp == tvp (can't happen)");
828 error = 0;
829 goto abortit;
830 }
831
832 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
833 goto abortit;
834 dp = VTOI(fdvp);
835 ip = VTOI(fvp);
836 if (ip->i_nlink >= EXT4_LINK_MAX &&
837 !EXT2_HAS_RO_COMPAT_FEATURE(ip->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
838 VOP_UNLOCK(fvp, 0);
839 error = EMLINK;
840 goto abortit;
841 }
842 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
843 || (dp->i_flags & APPEND)) {
844 VOP_UNLOCK(fvp, 0);
845 error = EPERM;
846 goto abortit;
847 }
848 if ((ip->i_mode & IFMT) == IFDIR) {
849 /*
850 * Avoid ".", "..", and aliases of "." for obvious reasons.
851 */
852 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
853 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
854 (ip->i_flag & IN_RENAME)) {
855 VOP_UNLOCK(fvp, 0);
856 error = EINVAL;
857 goto abortit;
858 }
859 ip->i_flag |= IN_RENAME;
860 oldparent = dp->i_number;
861 doingdirectory++;
862 }
863 vrele(fdvp);
864
865 /*
866 * When the target exists, both the directory
867 * and target vnodes are returned locked.
868 */
869 dp = VTOI(tdvp);
870 xp = NULL;
871 if (tvp)
872 xp = VTOI(tvp);
873
874 /*
875 * 1) Bump link count while we're moving stuff
876 * around. If we crash somewhere before
877 * completing our work, the link count
878 * may be wrong, but correctable.
879 */
880 ext2_inc_nlink(ip);
881 ip->i_flag |= IN_CHANGE;
882 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
883 VOP_UNLOCK(fvp, 0);
884 goto bad;
885 }
886
887 /*
888 * If ".." must be changed (ie the directory gets a new
889 * parent) then the source directory must not be in the
890 * directory hierarchy above the target, as this would
891 * orphan everything below the source directory. Also
892 * the user must have write permission in the source so
893 * as to be able to change "..". We must repeat the call
894 * to namei, as the parent directory is unlocked by the
895 * call to checkpath().
896 */
897 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
898 VOP_UNLOCK(fvp, 0);
899 if (oldparent != dp->i_number)
900 newparent = dp->i_number;
901 if (doingdirectory && newparent) {
902 if (error) /* write access check above */
903 goto bad;
904 if (xp != NULL)
905 vput(tvp);
906 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
907 if (error)
908 goto out;
909 VREF(tdvp);
910 error = relookup(tdvp, &tvp, tcnp);
911 if (error)
912 goto out;
913 vrele(tdvp);
914 dp = VTOI(tdvp);
915 xp = NULL;
916 if (tvp)
917 xp = VTOI(tvp);
918 }
919 /*
920 * 2) If target doesn't exist, link the target
921 * to the source and unlink the source.
922 * Otherwise, rewrite the target directory
923 * entry to reference the source inode and
924 * expunge the original entry's existence.
925 */
926 if (xp == NULL) {
927 if (dp->i_devvp != ip->i_devvp)
928 panic("ext2_rename: EXDEV");
929 /*
930 * Account for ".." in new directory.
931 * When source and destination have the same
932 * parent we don't fool with the link count.
933 */
934 if (doingdirectory && newparent) {
935 error = ext2_inc_nlink(dp);
936 if (error)
937 goto bad;
938
939 dp->i_flag |= IN_CHANGE;
940 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
941 if (error)
942 goto bad;
943 }
944 error = ext2_direnter(ip, tdvp, tcnp);
945 if (error) {
946 if (doingdirectory && newparent) {
947 ext2_dec_nlink(dp);
948 dp->i_flag |= IN_CHANGE;
949 (void)ext2_update(tdvp, 1);
950 }
951 goto bad;
952 }
953 vput(tdvp);
954 } else {
955 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
956 panic("ext2_rename: EXDEV");
957 /*
958 * Short circuit rename(foo, foo).
959 */
960 if (xp->i_number == ip->i_number)
961 panic("ext2_rename: same file");
962 /*
963 * If the parent directory is "sticky", then the user must
964 * own the parent directory, or the destination of the rename,
965 * otherwise the destination may not be changed (except by
966 * root). This implements append-only directories.
967 */
968 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
969 tcnp->cn_cred->cr_uid != dp->i_uid &&
970 xp->i_uid != tcnp->cn_cred->cr_uid) {
971 error = EPERM;
972 goto bad;
973 }
974 /*
975 * Target must be empty if a directory and have no links
976 * to it. Also, ensure source and target are compatible
977 * (both directories, or both not directories).
978 */
979 if ((xp->i_mode & IFMT) == IFDIR) {
980 if (!ext2_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
981 error = ENOTEMPTY;
982 goto bad;
983 }
984 if (!doingdirectory) {
985 error = ENOTDIR;
986 goto bad;
987 }
988 cache_purge(tdvp);
989 } else if (doingdirectory) {
990 error = EISDIR;
991 goto bad;
992 }
993 error = ext2_dirrewrite(dp, ip, tcnp);
994 if (error)
995 goto bad;
996 /*
997 * If the target directory is in the same
998 * directory as the source directory,
999 * decrement the link count on the parent
1000 * of the target directory.
1001 */
1002 if (doingdirectory && !newparent) {
1003 ext2_dec_nlink(dp);
1004 dp->i_flag |= IN_CHANGE;
1005 }
1006 vput(tdvp);
1007 /*
1008 * Adjust the link count of the target to
1009 * reflect the dirrewrite above. If this is
1010 * a directory it is empty and there are
1011 * no links to it, so we can squash the inode and
1012 * any space associated with it. We disallowed
1013 * renaming over top of a directory with links to
1014 * it above, as the remaining link would point to
1015 * a directory without "." or ".." entries.
1016 */
1017 ext2_dec_nlink(xp);
1018 if (doingdirectory) {
1019 if (xp->i_nlink > 2)
1020 panic("ext2_rename: linked directory");
1021 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1022 tcnp->cn_cred, tcnp->cn_thread);
1023 xp->i_nlink = 0;
1024 }
1025 xp->i_flag |= IN_CHANGE;
1026 vput(tvp);
1027 xp = NULL;
1028 }
1029
1030 /*
1031 * 3) Unlink the source.
1032 */
1033 fcnp->cn_flags &= ~MODMASK;
1034 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1035 VREF(fdvp);
1036 error = relookup(fdvp, &fvp, fcnp);
1037 if (error == 0)
1038 vrele(fdvp);
1039 if (fvp != NULL) {
1040 xp = VTOI(fvp);
1041 dp = VTOI(fdvp);
1042 } else {
1043 /*
1044 * From name has disappeared. IN_RENAME is not sufficient
1045 * to protect against directory races due to timing windows,
1046 * so we can't panic here.
1047 */
1048 vrele(ap->a_fvp);
1049 return (0);
1050 }
1051 /*
1052 * Ensure that the directory entry still exists and has not
1053 * changed while the new name has been entered. If the source is
1054 * a file then the entry may have been unlinked or renamed. In
1055 * either case there is no further work to be done. If the source
1056 * is a directory then it cannot have been rmdir'ed; its link
1057 * count of three would cause a rmdir to fail with ENOTEMPTY.
1058 * The IN_RENAME flag ensures that it cannot be moved by another
1059 * rename.
1060 */
1061 if (xp != ip) {
1062 /*
1063 * From name resolves to a different inode. IN_RENAME is
1064 * not sufficient protection against timing window races
1065 * so we can't panic here.
1066 */
1067 } else {
1068 /*
1069 * If the source is a directory with a
1070 * new parent, the link count of the old
1071 * parent directory must be decremented
1072 * and ".." set to point to the new parent.
1073 */
1074 if (doingdirectory && newparent) {
1075 ext2_dec_nlink(dp);
1076 dp->i_flag |= IN_CHANGE;
1077 dirbuf = malloc(dp->i_e2fs->e2fs_bsize, M_TEMP, M_WAITOK | M_ZERO);
1078 error = vn_rdwr(UIO_READ, fvp, (caddr_t)dirbuf,
1079 ip->i_e2fs->e2fs_bsize, (off_t)0,
1080 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1081 tcnp->cn_cred, NOCRED, NULL, NULL);
1082 if (error == 0) {
1083 /* Like ufs little-endian: */
1084 namlen = dirbuf->dotdot_type;
1085 if (namlen != 2 ||
1086 dirbuf->dotdot_name[0] != '.' ||
1087 dirbuf->dotdot_name[1] != '.') {
1088 ext2_dirbad(xp, (doff_t)12,
1089 "rename: mangled dir");
1090 } else {
1091 dirbuf->dotdot_ino = newparent;
1092 /*
1093 * dirblock 0 could be htree root,
1094 * try both csum update functions.
1095 */
1096 ext2_dirent_csum_set(ip,
1097 (struct ext2fs_direct_2 *)dirbuf);
1098 ext2_dx_csum_set(ip,
1099 (struct ext2fs_direct_2 *)dirbuf);
1100 (void)vn_rdwr(UIO_WRITE, fvp,
1101 (caddr_t)dirbuf,
1102 ip->i_e2fs->e2fs_bsize,
1103 (off_t)0, UIO_SYSSPACE,
1104 IO_NODELOCKED | IO_SYNC |
1105 IO_NOMACCHECK, tcnp->cn_cred,
1106 NOCRED, NULL, NULL);
1107 cache_purge(fdvp);
1108 }
1109 }
1110 free(dirbuf, M_TEMP);
1111 }
1112 error = ext2_dirremove(fdvp, fcnp);
1113 if (!error) {
1114 ext2_dec_nlink(xp);
1115 xp->i_flag |= IN_CHANGE;
1116 }
1117 xp->i_flag &= ~IN_RENAME;
1118 }
1119 if (dp)
1120 vput(fdvp);
1121 if (xp)
1122 vput(fvp);
1123 vrele(ap->a_fvp);
1124 return (error);
1125
1126 bad:
1127 if (xp)
1128 vput(ITOV(xp));
1129 vput(ITOV(dp));
1130 out:
1131 if (doingdirectory)
1132 ip->i_flag &= ~IN_RENAME;
1133 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1134 ext2_dec_nlink(ip);
1135 ip->i_flag |= IN_CHANGE;
1136 ip->i_flag &= ~IN_RENAME;
1137 vput(fvp);
1138 } else
1139 vrele(fvp);
1140 return (error);
1141 }
1142
1143 #ifdef UFS_ACL
1144 static int
1145 ext2_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1146 mode_t dmode, struct ucred *cred, struct thread *td)
1147 {
1148 int error;
1149 struct inode *ip = VTOI(tvp);
1150 struct acl *dacl, *acl;
1151
1152 acl = acl_alloc(M_WAITOK);
1153 dacl = acl_alloc(M_WAITOK);
1154
1155 /*
1156 * Retrieve default ACL from parent, if any.
1157 */
1158 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1159 switch (error) {
1160 case 0:
1161 /*
1162 * Retrieved a default ACL, so merge mode and ACL if
1163 * necessary. If the ACL is empty, fall through to
1164 * the "not defined or available" case.
1165 */
1166 if (acl->acl_cnt != 0) {
1167 dmode = acl_posix1e_newfilemode(dmode, acl);
1168 ip->i_mode = dmode;
1169 *dacl = *acl;
1170 ext2_sync_acl_from_inode(ip, acl);
1171 break;
1172 }
1173 /* FALLTHROUGH */
1174
1175 case EOPNOTSUPP:
1176 /*
1177 * Just use the mode as-is.
1178 */
1179 ip->i_mode = dmode;
1180 error = 0;
1181 goto out;
1182
1183 default:
1184 goto out;
1185 }
1186
1187 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1188 if (error == 0)
1189 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1190 switch (error) {
1191 case 0:
1192 break;
1193
1194 case EOPNOTSUPP:
1195 /*
1196 * XXX: This should not happen, as EOPNOTSUPP above
1197 * was supposed to free acl.
1198 */
1199 #ifdef DEBUG
1200 printf("ext2_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1201 #endif /* DEBUG */
1202 break;
1203
1204 default:
1205 goto out;
1206 }
1207
1208 out:
1209 acl_free(acl);
1210 acl_free(dacl);
1211
1212 return (error);
1213 }
1214
1215 static int
1216 ext2_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1217 mode_t mode, struct ucred *cred, struct thread *td)
1218 {
1219 int error;
1220 struct inode *ip = VTOI(tvp);
1221 struct acl *acl;
1222
1223 acl = acl_alloc(M_WAITOK);
1224
1225 /*
1226 * Retrieve default ACL for parent, if any.
1227 */
1228 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1229 switch (error) {
1230 case 0:
1231 /*
1232 * Retrieved a default ACL, so merge mode and ACL if
1233 * necessary.
1234 */
1235 if (acl->acl_cnt != 0) {
1236 /*
1237 * Two possible ways for default ACL to not
1238 * be present. First, the EA can be
1239 * undefined, or second, the default ACL can
1240 * be blank. If it's blank, fall through to
1241 * the it's not defined case.
1242 */
1243 mode = acl_posix1e_newfilemode(mode, acl);
1244 ip->i_mode = mode;
1245 ext2_sync_acl_from_inode(ip, acl);
1246 break;
1247 }
1248 /* FALLTHROUGH */
1249
1250 case EOPNOTSUPP:
1251 /*
1252 * Just use the mode as-is.
1253 */
1254 ip->i_mode = mode;
1255 error = 0;
1256 goto out;
1257
1258 default:
1259 goto out;
1260 }
1261
1262 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1263 switch (error) {
1264 case 0:
1265 break;
1266
1267 case EOPNOTSUPP:
1268 /*
1269 * XXX: This should not happen, as EOPNOTSUPP above was
1270 * supposed to free acl.
1271 */
1272 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1273 "but no VOP_SETACL()\n");
1274 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1275 "but no VOP_SETACL()"); */
1276 break;
1277
1278 default:
1279 goto out;
1280 }
1281
1282 out:
1283 acl_free(acl);
1284
1285 return (error);
1286 }
1287
1288 #endif /* UFS_ACL */
1289
1290 /*
1291 * Mkdir system call
1292 */
1293 static int
1294 ext2_mkdir(struct vop_mkdir_args *ap)
1295 {
1296 struct m_ext2fs *fs;
1297 struct vnode *dvp = ap->a_dvp;
1298 struct vattr *vap = ap->a_vap;
1299 struct componentname *cnp = ap->a_cnp;
1300 struct inode *ip, *dp;
1301 struct vnode *tvp;
1302 struct dirtemplate dirtemplate, *dtp;
1303 char *buf = NULL;
1304 int error, dmode;
1305
1306 #ifdef INVARIANTS
1307 if ((cnp->cn_flags & HASBUF) == 0)
1308 panic("ext2_mkdir: no name");
1309 #endif
1310 dp = VTOI(dvp);
1311 if ((nlink_t)dp->i_nlink >= EXT4_LINK_MAX &&
1312 !EXT2_HAS_RO_COMPAT_FEATURE(dp->i_e2fs, EXT2F_ROCOMPAT_DIR_NLINK)) {
1313 error = EMLINK;
1314 goto out;
1315 }
1316 dmode = vap->va_mode & 0777;
1317 dmode |= IFDIR;
1318 /*
1319 * Must simulate part of ext2_makeinode here to acquire the inode,
1320 * but not have it entered in the parent directory. The entry is
1321 * made later after writing "." and ".." entries.
1322 */
1323 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1324 if (error)
1325 goto out;
1326 ip = VTOI(tvp);
1327 fs = ip->i_e2fs;
1328 ip->i_gid = dp->i_gid;
1329 #ifdef SUIDDIR
1330 {
1331 /*
1332 * if we are hacking owners here, (only do this where told to)
1333 * and we are not giving it TOO root, (would subvert quotas)
1334 * then go ahead and give it to the other user.
1335 * The new directory also inherits the SUID bit.
1336 * If user's UID and dir UID are the same,
1337 * 'give it away' so that the SUID is still forced on.
1338 */
1339 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1340 (dp->i_mode & ISUID) && dp->i_uid) {
1341 dmode |= ISUID;
1342 ip->i_uid = dp->i_uid;
1343 } else {
1344 ip->i_uid = cnp->cn_cred->cr_uid;
1345 }
1346 }
1347 #else
1348 ip->i_uid = cnp->cn_cred->cr_uid;
1349 #endif
1350 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1351 ip->i_mode = dmode;
1352 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1353 ip->i_nlink = 2;
1354 if (cnp->cn_flags & ISWHITEOUT)
1355 ip->i_flags |= UF_OPAQUE;
1356 error = ext2_update(tvp, 1);
1357
1358 /*
1359 * Bump link count in parent directory
1360 * to reflect work done below. Should
1361 * be done before reference is created
1362 * so reparation is possible if we crash.
1363 */
1364 ext2_inc_nlink(dp);
1365 dp->i_flag |= IN_CHANGE;
1366 error = ext2_update(dvp, !DOINGASYNC(dvp));
1367 if (error)
1368 goto bad;
1369
1370 /* Initialize directory with "." and ".." from static template. */
1371 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1372 EXT2F_INCOMPAT_FTYPE))
1373 dtp = &mastertemplate;
1374 else
1375 dtp = &omastertemplate;
1376 dirtemplate = *dtp;
1377 dirtemplate.dot_ino = ip->i_number;
1378 dirtemplate.dotdot_ino = dp->i_number;
1379 /*
1380 * note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE so let's
1381 * just redefine it - for this function only
1382 */
1383 #undef DIRBLKSIZ
1384 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1385 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1386 buf = malloc(DIRBLKSIZ, M_TEMP, M_WAITOK | M_ZERO);
1387 if (EXT2_HAS_RO_COMPAT_FEATURE(fs, EXT2F_ROCOMPAT_METADATA_CKSUM)) {
1388 dirtemplate.dotdot_reclen -= sizeof(struct ext2fs_direct_tail);
1389 ext2_init_dirent_tail(EXT2_DIRENT_TAIL(buf, DIRBLKSIZ));
1390 }
1391 memcpy(buf, &dirtemplate, sizeof(dirtemplate));
1392 ext2_dirent_csum_set(ip, (struct ext2fs_direct_2 *)buf);
1393 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)buf,
1394 DIRBLKSIZ, (off_t)0, UIO_SYSSPACE,
1395 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1396 NULL, NULL);
1397 if (error) {
1398 ext2_dec_nlink(dp);
1399 dp->i_flag |= IN_CHANGE;
1400 goto bad;
1401 }
1402 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1403 /* XXX should grow with balloc() */
1404 panic("ext2_mkdir: blksize");
1405 else {
1406 ip->i_size = DIRBLKSIZ;
1407 ip->i_flag |= IN_CHANGE;
1408 }
1409
1410 #ifdef UFS_ACL
1411 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
1412 error = ext2_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
1413 cnp->cn_cred, cnp->cn_thread);
1414 if (error)
1415 goto bad;
1416 }
1417
1418 #endif /* UFS_ACL */
1419
1420 /* Directory set up, now install its entry in the parent directory. */
1421 error = ext2_direnter(ip, dvp, cnp);
1422 if (error) {
1423 ext2_dec_nlink(dp);
1424 dp->i_flag |= IN_CHANGE;
1425 }
1426 bad:
1427 /*
1428 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1429 * for us because we set the link count to 0.
1430 */
1431 if (error) {
1432 ip->i_nlink = 0;
1433 ip->i_flag |= IN_CHANGE;
1434 vput(tvp);
1435 } else
1436 *ap->a_vpp = tvp;
1437 out:
1438 free(buf, M_TEMP);
1439 return (error);
1440 #undef DIRBLKSIZ
1441 #define DIRBLKSIZ DEV_BSIZE
1442 }
1443
1444 /*
1445 * Rmdir system call.
1446 */
1447 static int
1448 ext2_rmdir(struct vop_rmdir_args *ap)
1449 {
1450 struct vnode *vp = ap->a_vp;
1451 struct vnode *dvp = ap->a_dvp;
1452 struct componentname *cnp = ap->a_cnp;
1453 struct inode *ip, *dp;
1454 int error;
1455
1456 ip = VTOI(vp);
1457 dp = VTOI(dvp);
1458
1459 /*
1460 * Verify the directory is empty (and valid).
1461 * (Rmdir ".." won't be valid since
1462 * ".." will contain a reference to
1463 * the current directory and thus be
1464 * non-empty.)
1465 */
1466 if (!ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1467 error = ENOTEMPTY;
1468 goto out;
1469 }
1470 if ((dp->i_flags & APPEND)
1471 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1472 error = EPERM;
1473 goto out;
1474 }
1475 /*
1476 * Delete reference to directory before purging
1477 * inode. If we crash in between, the directory
1478 * will be reattached to lost+found,
1479 */
1480 error = ext2_dirremove(dvp, cnp);
1481 if (error)
1482 goto out;
1483 ext2_dec_nlink(dp);
1484 dp->i_flag |= IN_CHANGE;
1485 cache_purge(dvp);
1486 VOP_UNLOCK(dvp, 0);
1487 /*
1488 * Truncate inode. The only stuff left
1489 * in the directory is "." and "..".
1490 */
1491 ip->i_nlink = 0;
1492 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1493 cnp->cn_thread);
1494 cache_purge(ITOV(ip));
1495 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1496 VOP_UNLOCK(vp, 0);
1497 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1498 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1499 }
1500 out:
1501 return (error);
1502 }
1503
1504 /*
1505 * symlink -- make a symbolic link
1506 */
1507 static int
1508 ext2_symlink(struct vop_symlink_args *ap)
1509 {
1510 struct vnode *vp, **vpp = ap->a_vpp;
1511 struct inode *ip;
1512 int len, error;
1513
1514 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1515 vpp, ap->a_cnp);
1516 if (error)
1517 return (error);
1518 vp = *vpp;
1519 len = strlen(ap->a_target);
1520 if (len < vp->v_mount->mnt_maxsymlinklen) {
1521 ip = VTOI(vp);
1522 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1523 ip->i_size = len;
1524 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1525 } else
1526 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1527 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1528 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1529 if (error)
1530 vput(vp);
1531 return (error);
1532 }
1533
1534 /*
1535 * Return target name of a symbolic link
1536 */
1537 static int
1538 ext2_readlink(struct vop_readlink_args *ap)
1539 {
1540 struct vnode *vp = ap->a_vp;
1541 struct inode *ip = VTOI(vp);
1542 int isize;
1543
1544 isize = ip->i_size;
1545 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1546 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1547 return (0);
1548 }
1549 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1550 }
1551
1552 /*
1553 * Calculate the logical to physical mapping if not done already,
1554 * then call the device strategy routine.
1555 *
1556 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1557 * deadlock on memory. See ext2_bmap() for details.
1558 */
1559 static int
1560 ext2_strategy(struct vop_strategy_args *ap)
1561 {
1562 struct buf *bp = ap->a_bp;
1563 struct vnode *vp = ap->a_vp;
1564 struct bufobj *bo;
1565 daddr_t blkno;
1566 int error;
1567
1568 if (vp->v_type == VBLK || vp->v_type == VCHR)
1569 panic("ext2_strategy: spec");
1570 if (bp->b_blkno == bp->b_lblkno) {
1571
1572 if (VTOI(ap->a_vp)->i_flag & IN_E4EXTENTS)
1573 error = ext4_bmapext(vp, bp->b_lblkno, &blkno, NULL, NULL);
1574 else
1575 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1576
1577 bp->b_blkno = blkno;
1578 if (error) {
1579 bp->b_error = error;
1580 bp->b_ioflags |= BIO_ERROR;
1581 bufdone(bp);
1582 return (0);
1583 }
1584 if ((long)bp->b_blkno == -1)
1585 vfs_bio_clrbuf(bp);
1586 }
1587 if ((long)bp->b_blkno == -1) {
1588 bufdone(bp);
1589 return (0);
1590 }
1591 bp->b_iooffset = dbtob(bp->b_blkno);
1592 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1593 BO_STRATEGY(bo, bp);
1594 return (0);
1595 }
1596
1597 /*
1598 * Print out the contents of an inode.
1599 */
1600 static int
1601 ext2_print(struct vop_print_args *ap)
1602 {
1603 struct vnode *vp = ap->a_vp;
1604 struct inode *ip = VTOI(vp);
1605
1606 vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number);
1607 if (vp->v_type == VFIFO)
1608 fifo_printinfo(vp);
1609 printf("\n");
1610 return (0);
1611 }
1612
1613 /*
1614 * Close wrapper for fifos.
1615 *
1616 * Update the times on the inode then do device close.
1617 */
1618 static int
1619 ext2fifo_close(struct vop_close_args *ap)
1620 {
1621 struct vnode *vp = ap->a_vp;
1622
1623 VI_LOCK(vp);
1624 if (vp->v_usecount > 1)
1625 ext2_itimes_locked(vp);
1626 VI_UNLOCK(vp);
1627 return (fifo_specops.vop_close(ap));
1628 }
1629
1630 /*
1631 * Kqfilter wrapper for fifos.
1632 *
1633 * Fall through to ext2 kqfilter routines if needed
1634 */
1635 static int
1636 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1637 {
1638 int error;
1639
1640 error = fifo_specops.vop_kqfilter(ap);
1641 if (error)
1642 error = vfs_kqfilter(ap);
1643 return (error);
1644 }
1645
1646 /*
1647 * Return POSIX pathconf information applicable to ext2 filesystems.
1648 */
1649 static int
1650 ext2_pathconf(struct vop_pathconf_args *ap)
1651 {
1652 int error = 0;
1653
1654 switch (ap->a_name) {
1655 case _PC_LINK_MAX:
1656 if (EXT2_HAS_RO_COMPAT_FEATURE(VTOI(ap->a_vp)->i_e2fs,
1657 EXT2F_ROCOMPAT_DIR_NLINK))
1658 *ap->a_retval = INT_MAX;
1659 else
1660 *ap->a_retval = EXT4_LINK_MAX;
1661 break;
1662 case _PC_NAME_MAX:
1663 *ap->a_retval = NAME_MAX;
1664 break;
1665 case _PC_PIPE_BUF:
1666 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
1667 *ap->a_retval = PIPE_BUF;
1668 else
1669 error = EINVAL;
1670 break;
1671 case _PC_CHOWN_RESTRICTED:
1672 *ap->a_retval = 1;
1673 break;
1674 case _PC_NO_TRUNC:
1675 *ap->a_retval = 1;
1676 break;
1677
1678 #ifdef UFS_ACL
1679 case _PC_ACL_EXTENDED:
1680 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1681 *ap->a_retval = 1;
1682 else
1683 *ap->a_retval = 0;
1684 break;
1685 case _PC_ACL_PATH_MAX:
1686 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
1687 *ap->a_retval = ACL_MAX_ENTRIES;
1688 else
1689 *ap->a_retval = 3;
1690 break;
1691 #endif /* UFS_ACL */
1692
1693 case _PC_MIN_HOLE_SIZE:
1694 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1695 break;
1696 case _PC_PRIO_IO:
1697 *ap->a_retval = 0;
1698 break;
1699 case _PC_SYNC_IO:
1700 *ap->a_retval = 0;
1701 break;
1702 case _PC_ALLOC_SIZE_MIN:
1703 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1704 break;
1705 case _PC_FILESIZEBITS:
1706 *ap->a_retval = 64;
1707 break;
1708 case _PC_REC_INCR_XFER_SIZE:
1709 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1710 break;
1711 case _PC_REC_MAX_XFER_SIZE:
1712 *ap->a_retval = -1; /* means ``unlimited'' */
1713 break;
1714 case _PC_REC_MIN_XFER_SIZE:
1715 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1716 break;
1717 case _PC_REC_XFER_ALIGN:
1718 *ap->a_retval = PAGE_SIZE;
1719 break;
1720 case _PC_SYMLINK_MAX:
1721 *ap->a_retval = MAXPATHLEN;
1722 break;
1723
1724 default:
1725 error = vop_stdpathconf(ap);
1726 break;
1727 }
1728 return (error);
1729 }
1730
1731 /*
1732 * Vnode operation to remove a named attribute.
1733 */
1734 static int
1735 ext2_deleteextattr(struct vop_deleteextattr_args *ap)
1736 {
1737 struct inode *ip;
1738 struct m_ext2fs *fs;
1739 int error;
1740
1741 ip = VTOI(ap->a_vp);
1742 fs = ip->i_e2fs;
1743
1744 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1745 return (EOPNOTSUPP);
1746
1747 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1748 return (EOPNOTSUPP);
1749
1750 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1751 ap->a_cred, ap->a_td, VWRITE);
1752 if (error)
1753 return (error);
1754
1755 error = ENOATTR;
1756
1757 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1758 error = ext2_extattr_inode_delete(ip, ap->a_attrnamespace, ap->a_name);
1759 if (error != ENOATTR)
1760 return (error);
1761 }
1762
1763 if (ip->i_facl)
1764 error = ext2_extattr_block_delete(ip, ap->a_attrnamespace, ap->a_name);
1765
1766 return (error);
1767 }
1768
1769 /*
1770 * Vnode operation to retrieve a named extended attribute.
1771 */
1772 static int
1773 ext2_getextattr(struct vop_getextattr_args *ap)
1774 {
1775 struct inode *ip;
1776 struct m_ext2fs *fs;
1777 int error;
1778
1779 ip = VTOI(ap->a_vp);
1780 fs = ip->i_e2fs;
1781
1782 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1783 return (EOPNOTSUPP);
1784
1785 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1786 return (EOPNOTSUPP);
1787
1788 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1789 ap->a_cred, ap->a_td, VREAD);
1790 if (error)
1791 return (error);
1792
1793 if (ap->a_size != NULL)
1794 *ap->a_size = 0;
1795
1796 error = ENOATTR;
1797
1798 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1799 error = ext2_extattr_inode_get(ip, ap->a_attrnamespace,
1800 ap->a_name, ap->a_uio, ap->a_size);
1801 if (error != ENOATTR)
1802 return (error);
1803 }
1804
1805 if (ip->i_facl)
1806 error = ext2_extattr_block_get(ip, ap->a_attrnamespace,
1807 ap->a_name, ap->a_uio, ap->a_size);
1808
1809 return (error);
1810 }
1811
1812 /*
1813 * Vnode operation to retrieve extended attributes on a vnode.
1814 */
1815 static int
1816 ext2_listextattr(struct vop_listextattr_args *ap)
1817 {
1818 struct inode *ip;
1819 struct m_ext2fs *fs;
1820 int error;
1821
1822 ip = VTOI(ap->a_vp);
1823 fs = ip->i_e2fs;
1824
1825 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1826 return (EOPNOTSUPP);
1827
1828 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1829 return (EOPNOTSUPP);
1830
1831 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1832 ap->a_cred, ap->a_td, VREAD);
1833 if (error)
1834 return (error);
1835
1836 if (ap->a_size != NULL)
1837 *ap->a_size = 0;
1838
1839 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1840 error = ext2_extattr_inode_list(ip, ap->a_attrnamespace,
1841 ap->a_uio, ap->a_size);
1842 if (error)
1843 return (error);
1844 }
1845
1846 if (ip->i_facl)
1847 error = ext2_extattr_block_list(ip, ap->a_attrnamespace,
1848 ap->a_uio, ap->a_size);
1849
1850 return (error);
1851 }
1852
1853 /*
1854 * Vnode operation to set a named attribute.
1855 */
1856 static int
1857 ext2_setextattr(struct vop_setextattr_args *ap)
1858 {
1859 struct inode *ip;
1860 struct m_ext2fs *fs;
1861 int error;
1862
1863 ip = VTOI(ap->a_vp);
1864 fs = ip->i_e2fs;
1865
1866 if (!EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_EXT_ATTR))
1867 return (EOPNOTSUPP);
1868
1869 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1870 return (EOPNOTSUPP);
1871
1872 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1873 ap->a_cred, ap->a_td, VWRITE);
1874 if (error)
1875 return (error);
1876
1877 error = ext2_extattr_valid_attrname(ap->a_attrnamespace, ap->a_name);
1878 if (error)
1879 return (error);
1880
1881 if (EXT2_INODE_SIZE(fs) != E2FS_REV0_INODE_SIZE) {
1882 error = ext2_extattr_inode_set(ip, ap->a_attrnamespace,
1883 ap->a_name, ap->a_uio);
1884 if (error != ENOSPC)
1885 return (error);
1886 }
1887
1888 error = ext2_extattr_block_set(ip, ap->a_attrnamespace,
1889 ap->a_name, ap->a_uio);
1890
1891 return (error);
1892 }
1893
1894 /*
1895 * Vnode pointer to File handle
1896 */
1897 /* ARGSUSED */
1898 static int
1899 ext2_vptofh(struct vop_vptofh_args *ap)
1900 {
1901 struct inode *ip;
1902 struct ufid *ufhp;
1903
1904 ip = VTOI(ap->a_vp);
1905 ufhp = (struct ufid *)ap->a_fhp;
1906 ufhp->ufid_len = sizeof(struct ufid);
1907 ufhp->ufid_ino = ip->i_number;
1908 ufhp->ufid_gen = ip->i_gen;
1909 return (0);
1910 }
1911
1912 /*
1913 * Initialize the vnode associated with a new inode, handle aliased
1914 * vnodes.
1915 */
1916 int
1917 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1918 {
1919 struct inode *ip;
1920 struct vnode *vp;
1921
1922 vp = *vpp;
1923 ip = VTOI(vp);
1924 vp->v_type = IFTOVT(ip->i_mode);
1925 /*
1926 * Only unallocated inodes should be of type VNON.
1927 */
1928 if (ip->i_mode != 0 && vp->v_type == VNON)
1929 return (EINVAL);
1930 if (vp->v_type == VFIFO)
1931 vp->v_op = fifoops;
1932
1933 if (ip->i_number == EXT2_ROOTINO)
1934 vp->v_vflag |= VV_ROOT;
1935 ip->i_modrev = init_va_filerev();
1936 *vpp = vp;
1937 return (0);
1938 }
1939
1940 /*
1941 * Allocate a new inode.
1942 */
1943 static int
1944 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1945 struct componentname *cnp)
1946 {
1947 struct inode *ip, *pdir;
1948 struct vnode *tvp;
1949 int error;
1950
1951 pdir = VTOI(dvp);
1952 #ifdef INVARIANTS
1953 if ((cnp->cn_flags & HASBUF) == 0)
1954 panic("ext2_makeinode: no name");
1955 #endif
1956 *vpp = NULL;
1957 if ((mode & IFMT) == 0)
1958 mode |= IFREG;
1959
1960 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1961 if (error) {
1962 return (error);
1963 }
1964 ip = VTOI(tvp);
1965 ip->i_gid = pdir->i_gid;
1966 #ifdef SUIDDIR
1967 {
1968 /*
1969 * if we are
1970 * not the owner of the directory,
1971 * and we are hacking owners here, (only do this where told to)
1972 * and we are not giving it TOO root, (would subvert quotas)
1973 * then go ahead and give it to the other user.
1974 * Note that this drops off the execute bits for security.
1975 */
1976 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1977 (pdir->i_mode & ISUID) &&
1978 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1979 ip->i_uid = pdir->i_uid;
1980 mode &= ~07111;
1981 } else {
1982 ip->i_uid = cnp->cn_cred->cr_uid;
1983 }
1984 }
1985 #else
1986 ip->i_uid = cnp->cn_cred->cr_uid;
1987 #endif
1988 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1989 ip->i_mode = mode;
1990 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1991 ip->i_nlink = 1;
1992 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1993 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1994 ip->i_mode &= ~ISGID;
1995 }
1996
1997 if (cnp->cn_flags & ISWHITEOUT)
1998 ip->i_flags |= UF_OPAQUE;
1999
2000 /*
2001 * Make sure inode goes to disk before directory entry.
2002 */
2003 error = ext2_update(tvp, !DOINGASYNC(tvp));
2004 if (error)
2005 goto bad;
2006
2007 #ifdef UFS_ACL
2008 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2009 error = ext2_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2010 cnp->cn_cred, cnp->cn_thread);
2011 if (error)
2012 goto bad;
2013 }
2014 #endif /* UFS_ACL */
2015
2016 error = ext2_direnter(ip, dvp, cnp);
2017 if (error)
2018 goto bad;
2019
2020 *vpp = tvp;
2021 return (0);
2022
2023 bad:
2024 /*
2025 * Write error occurred trying to update the inode
2026 * or the directory so must deallocate the inode.
2027 */
2028 ip->i_nlink = 0;
2029 ip->i_flag |= IN_CHANGE;
2030 vput(tvp);
2031 return (error);
2032 }
2033
2034 /*
2035 * Vnode op for reading.
2036 */
2037 static int
2038 ext2_read(struct vop_read_args *ap)
2039 {
2040 struct vnode *vp;
2041 struct inode *ip;
2042 struct uio *uio;
2043 struct m_ext2fs *fs;
2044 struct buf *bp;
2045 daddr_t lbn, nextlbn;
2046 off_t bytesinfile;
2047 long size, xfersize, blkoffset;
2048 int error, orig_resid, seqcount;
2049 int ioflag;
2050
2051 vp = ap->a_vp;
2052 uio = ap->a_uio;
2053 ioflag = ap->a_ioflag;
2054
2055 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
2056 ip = VTOI(vp);
2057
2058 #ifdef INVARIANTS
2059 if (uio->uio_rw != UIO_READ)
2060 panic("%s: mode", "ext2_read");
2061
2062 if (vp->v_type == VLNK) {
2063 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
2064 panic("%s: short symlink", "ext2_read");
2065 } else if (vp->v_type != VREG && vp->v_type != VDIR)
2066 panic("%s: type %d", "ext2_read", vp->v_type);
2067 #endif
2068 orig_resid = uio->uio_resid;
2069 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
2070 if (orig_resid == 0)
2071 return (0);
2072 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
2073 fs = ip->i_e2fs;
2074 if (uio->uio_offset < ip->i_size &&
2075 uio->uio_offset >= fs->e2fs_maxfilesize)
2076 return (EOVERFLOW);
2077
2078 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
2079 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
2080 break;
2081 lbn = lblkno(fs, uio->uio_offset);
2082 nextlbn = lbn + 1;
2083 size = blksize(fs, ip, lbn);
2084 blkoffset = blkoff(fs, uio->uio_offset);
2085
2086 xfersize = fs->e2fs_fsize - blkoffset;
2087 if (uio->uio_resid < xfersize)
2088 xfersize = uio->uio_resid;
2089 if (bytesinfile < xfersize)
2090 xfersize = bytesinfile;
2091
2092 if (lblktosize(fs, nextlbn) >= ip->i_size)
2093 error = bread(vp, lbn, size, NOCRED, &bp);
2094 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
2095 error = cluster_read(vp, ip->i_size, lbn, size,
2096 NOCRED, blkoffset + uio->uio_resid, seqcount,
2097 0, &bp);
2098 } else if (seqcount > 1) {
2099 u_int nextsize = blksize(fs, ip, nextlbn);
2100
2101 error = breadn(vp, lbn,
2102 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
2103 } else
2104 error = bread(vp, lbn, size, NOCRED, &bp);
2105 if (error) {
2106 brelse(bp);
2107 bp = NULL;
2108 break;
2109 }
2110
2111 /*
2112 * We should only get non-zero b_resid when an I/O error
2113 * has occurred, which should cause us to break above.
2114 * However, if the short read did not cause an error,
2115 * then we want to ensure that we do not uiomove bad
2116 * or uninitialized data.
2117 */
2118 size -= bp->b_resid;
2119 if (size < xfersize) {
2120 if (size == 0)
2121 break;
2122 xfersize = size;
2123 }
2124 error = uiomove((char *)bp->b_data + blkoffset,
2125 (int)xfersize, uio);
2126 if (error)
2127 break;
2128 vfs_bio_brelse(bp, ioflag);
2129 }
2130
2131 /*
2132 * This can only happen in the case of an error because the loop
2133 * above resets bp to NULL on each iteration and on normal
2134 * completion has not set a new value into it. so it must have come
2135 * from a 'break' statement
2136 */
2137 if (bp != NULL)
2138 vfs_bio_brelse(bp, ioflag);
2139
2140 if ((error == 0 || uio->uio_resid != orig_resid) &&
2141 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
2142 ip->i_flag |= IN_ACCESS;
2143 return (error);
2144 }
2145
2146 static int
2147 ext2_ioctl(struct vop_ioctl_args *ap)
2148 {
2149 struct vnode *vp;
2150 int error;
2151
2152 vp = ap->a_vp;
2153 switch (ap->a_command) {
2154 case FIOSEEKDATA:
2155 if (!(VTOI(vp)->i_flag & IN_E4EXTENTS)) {
2156 error = vn_lock(vp, LK_SHARED);
2157 if (error == 0) {
2158 error = ext2_bmap_seekdata(vp,
2159 (off_t *)ap->a_data);
2160 VOP_UNLOCK(vp, 0);
2161 } else
2162 error = EBADF;
2163 return (error);
2164 }
2165 case FIOSEEKHOLE:
2166 return (vn_bmap_seekhole(vp, ap->a_command,
2167 (off_t *)ap->a_data, ap->a_cred));
2168 default:
2169 return (ENOTTY);
2170 }
2171 }
2172
2173 /*
2174 * Vnode op for writing.
2175 */
2176 static int
2177 ext2_write(struct vop_write_args *ap)
2178 {
2179 struct vnode *vp;
2180 struct uio *uio;
2181 struct inode *ip;
2182 struct m_ext2fs *fs;
2183 struct buf *bp;
2184 daddr_t lbn;
2185 off_t osize;
2186 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
2187
2188 ioflag = ap->a_ioflag;
2189 uio = ap->a_uio;
2190 vp = ap->a_vp;
2191
2192 seqcount = ioflag >> IO_SEQSHIFT;
2193 ip = VTOI(vp);
2194
2195 #ifdef INVARIANTS
2196 if (uio->uio_rw != UIO_WRITE)
2197 panic("%s: mode", "ext2_write");
2198 #endif
2199
2200 switch (vp->v_type) {
2201 case VREG:
2202 if (ioflag & IO_APPEND)
2203 uio->uio_offset = ip->i_size;
2204 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
2205 return (EPERM);
2206 /* FALLTHROUGH */
2207 case VLNK:
2208 break;
2209 case VDIR:
2210 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
2211 if ((ioflag & IO_SYNC) == 0)
2212 panic("ext2_write: nonsync dir write");
2213 break;
2214 default:
2215 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
2216 vp->v_type, (intmax_t)uio->uio_offset,
2217 (intmax_t)uio->uio_resid);
2218 }
2219
2220 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
2221 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
2222 fs = ip->i_e2fs;
2223 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
2224 return (EFBIG);
2225 /*
2226 * Maybe this should be above the vnode op call, but so long as
2227 * file servers have no limits, I don't think it matters.
2228 */
2229 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
2230 return (EFBIG);
2231
2232 resid = uio->uio_resid;
2233 osize = ip->i_size;
2234 if (seqcount > BA_SEQMAX)
2235 flags = BA_SEQMAX << BA_SEQSHIFT;
2236 else
2237 flags = seqcount << BA_SEQSHIFT;
2238 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
2239 flags |= IO_SYNC;
2240
2241 for (error = 0; uio->uio_resid > 0;) {
2242 lbn = lblkno(fs, uio->uio_offset);
2243 blkoffset = blkoff(fs, uio->uio_offset);
2244 xfersize = fs->e2fs_fsize - blkoffset;
2245 if (uio->uio_resid < xfersize)
2246 xfersize = uio->uio_resid;
2247 if (uio->uio_offset + xfersize > ip->i_size)
2248 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
2249
2250 /*
2251 * We must perform a read-before-write if the transfer size
2252 * does not cover the entire buffer.
2253 */
2254 if (fs->e2fs_bsize > xfersize)
2255 flags |= BA_CLRBUF;
2256 else
2257 flags &= ~BA_CLRBUF;
2258 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
2259 ap->a_cred, &bp, flags);
2260 if (error != 0)
2261 break;
2262
2263 if ((ioflag & (IO_SYNC | IO_INVAL)) == (IO_SYNC | IO_INVAL))
2264 bp->b_flags |= B_NOCACHE;
2265 if (uio->uio_offset + xfersize > ip->i_size)
2266 ip->i_size = uio->uio_offset + xfersize;
2267 size = blksize(fs, ip, lbn) - bp->b_resid;
2268 if (size < xfersize)
2269 xfersize = size;
2270
2271 error =
2272 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
2273 /*
2274 * If the buffer is not already filled and we encounter an
2275 * error while trying to fill it, we have to clear out any
2276 * garbage data from the pages instantiated for the buffer.
2277 * If we do not, a failed uiomove() during a write can leave
2278 * the prior contents of the pages exposed to a userland mmap.
2279 *
2280 * Note that we need only clear buffers with a transfer size
2281 * equal to the block size because buffers with a shorter
2282 * transfer size were cleared above by the call to ext2_balloc()
2283 * with the BA_CLRBUF flag set.
2284 *
2285 * If the source region for uiomove identically mmaps the
2286 * buffer, uiomove() performed the NOP copy, and the buffer
2287 * content remains valid because the page fault handler
2288 * validated the pages.
2289 */
2290 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2291 fs->e2fs_bsize == xfersize)
2292 vfs_bio_clrbuf(bp);
2293
2294 vfs_bio_set_flags(bp, ioflag);
2295
2296 /*
2297 * If IO_SYNC each buffer is written synchronously. Otherwise
2298 * if we have a severe page deficiency write the buffer
2299 * asynchronously. Otherwise try to cluster, and if that
2300 * doesn't do it then either do an async write (if O_DIRECT),
2301 * or a delayed write (if not).
2302 */
2303 if (ioflag & IO_SYNC) {
2304 (void)bwrite(bp);
2305 } else if (vm_page_count_severe() ||
2306 buf_dirty_count_severe() ||
2307 (ioflag & IO_ASYNC)) {
2308 bp->b_flags |= B_CLUSTEROK;
2309 bawrite(bp);
2310 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
2311 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2312 bp->b_flags |= B_CLUSTEROK;
2313 cluster_write(vp, bp, ip->i_size, seqcount, 0);
2314 } else {
2315 bawrite(bp);
2316 }
2317 } else if (ioflag & IO_DIRECT) {
2318 bp->b_flags |= B_CLUSTEROK;
2319 bawrite(bp);
2320 } else {
2321 bp->b_flags |= B_CLUSTEROK;
2322 bdwrite(bp);
2323 }
2324 if (error || xfersize == 0)
2325 break;
2326 }
2327 /*
2328 * If we successfully wrote any data, and we are not the superuser
2329 * we clear the setuid and setgid bits as a precaution against
2330 * tampering.
2331 */
2332 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2333 ap->a_cred) {
2334 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2335 ip->i_mode &= ~(ISUID | ISGID);
2336 }
2337 if (error) {
2338 if (ioflag & IO_UNIT) {
2339 (void)ext2_truncate(vp, osize,
2340 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2341 uio->uio_offset -= resid - uio->uio_resid;
2342 uio->uio_resid = resid;
2343 }
2344 }
2345 if (uio->uio_resid != resid) {
2346 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2347 if (ioflag & IO_SYNC)
2348 error = ext2_update(vp, 1);
2349 }
2350 return (error);
2351 }
Cache object: ebb86ef7537c1272790b712ca8622be9
|