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 * Copyright (c) 1982, 1986, 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 * (c) UNIX System Laboratories, Inc.
11 * All or some portions of this file are derived from material licensed
12 * to the University of California by American Telephone and Telegraph
13 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14 * the permission of UNIX System Laboratories, Inc.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 *
40 * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94
41 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
42 * $FreeBSD$
43 */
44
45 #include "opt_suiddir.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
51 #include <sys/filio.h>
52 #include <sys/stat.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/endian.h>
56 #include <sys/priv.h>
57 #include <sys/mount.h>
58 #include <sys/unistd.h>
59 #include <sys/time.h>
60 #include <sys/vnode.h>
61 #include <sys/namei.h>
62 #include <sys/lockf.h>
63 #include <sys/event.h>
64 #include <sys/conf.h>
65 #include <sys/file.h>
66
67 #include <vm/vm.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vnode_pager.h>
72
73 #include "opt_directio.h"
74
75 #include <fs/fifofs/fifo.h>
76
77 #include <ufs/ufs/dir.h>
78
79 #include <fs/ext2fs/fs.h>
80 #include <fs/ext2fs/inode.h>
81 #include <fs/ext2fs/ext2_extern.h>
82 #include <fs/ext2fs/ext2fs.h>
83 #include <fs/ext2fs/ext2_dinode.h>
84 #include <fs/ext2fs/ext2_dir.h>
85 #include <fs/ext2fs/ext2_mount.h>
86
87 static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
88 static void ext2_itimes_locked(struct vnode *);
89 static int ext4_ext_read(struct vop_read_args *);
90 static int ext2_ind_read(struct vop_read_args *);
91
92 static vop_access_t ext2_access;
93 static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
94 static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
95 struct thread *);
96 static vop_close_t ext2_close;
97 static vop_create_t ext2_create;
98 static vop_fsync_t ext2_fsync;
99 static vop_getattr_t ext2_getattr;
100 static vop_ioctl_t ext2_ioctl;
101 static vop_link_t ext2_link;
102 static vop_mkdir_t ext2_mkdir;
103 static vop_mknod_t ext2_mknod;
104 static vop_open_t ext2_open;
105 static vop_pathconf_t ext2_pathconf;
106 static vop_print_t ext2_print;
107 static vop_read_t ext2_read;
108 static vop_readlink_t ext2_readlink;
109 static vop_remove_t ext2_remove;
110 static vop_rename_t ext2_rename;
111 static vop_rmdir_t ext2_rmdir;
112 static vop_setattr_t ext2_setattr;
113 static vop_strategy_t ext2_strategy;
114 static vop_symlink_t ext2_symlink;
115 static vop_write_t ext2_write;
116 static vop_vptofh_t ext2_vptofh;
117 static vop_close_t ext2fifo_close;
118 static vop_kqfilter_t ext2fifo_kqfilter;
119
120 /* Global vfs data structures for ext2. */
121 struct vop_vector ext2_vnodeops = {
122 .vop_default = &default_vnodeops,
123 .vop_access = ext2_access,
124 .vop_bmap = ext2_bmap,
125 .vop_cachedlookup = ext2_lookup,
126 .vop_close = ext2_close,
127 .vop_create = ext2_create,
128 .vop_fsync = ext2_fsync,
129 .vop_getattr = ext2_getattr,
130 .vop_inactive = ext2_inactive,
131 .vop_ioctl = ext2_ioctl,
132 .vop_link = ext2_link,
133 .vop_lookup = vfs_cache_lookup,
134 .vop_mkdir = ext2_mkdir,
135 .vop_mknod = ext2_mknod,
136 .vop_open = ext2_open,
137 .vop_pathconf = ext2_pathconf,
138 .vop_poll = vop_stdpoll,
139 .vop_print = ext2_print,
140 .vop_read = ext2_read,
141 .vop_readdir = ext2_readdir,
142 .vop_readlink = ext2_readlink,
143 .vop_reallocblks = ext2_reallocblks,
144 .vop_reclaim = ext2_reclaim,
145 .vop_remove = ext2_remove,
146 .vop_rename = ext2_rename,
147 .vop_rmdir = ext2_rmdir,
148 .vop_setattr = ext2_setattr,
149 .vop_strategy = ext2_strategy,
150 .vop_symlink = ext2_symlink,
151 .vop_write = ext2_write,
152 .vop_vptofh = ext2_vptofh,
153 };
154
155 struct vop_vector ext2_fifoops = {
156 .vop_default = &fifo_specops,
157 .vop_access = ext2_access,
158 .vop_close = ext2fifo_close,
159 .vop_fsync = ext2_fsync,
160 .vop_getattr = ext2_getattr,
161 .vop_inactive = ext2_inactive,
162 .vop_kqfilter = ext2fifo_kqfilter,
163 .vop_print = ext2_print,
164 .vop_read = VOP_PANIC,
165 .vop_reclaim = ext2_reclaim,
166 .vop_setattr = ext2_setattr,
167 .vop_write = VOP_PANIC,
168 .vop_vptofh = ext2_vptofh,
169 };
170
171 /*
172 * A virgin directory (no blushing please).
173 * Note that the type and namlen fields are reversed relative to ext2.
174 * Also, we don't use `struct odirtemplate', since it would just cause
175 * endianness problems.
176 */
177 static struct dirtemplate mastertemplate = {
178 0, 12, 1, EXT2_FT_DIR, ".",
179 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
180 };
181 static struct dirtemplate omastertemplate = {
182 0, 12, 1, EXT2_FT_UNKNOWN, ".",
183 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
184 };
185
186 static void
187 ext2_itimes_locked(struct vnode *vp)
188 {
189 struct inode *ip;
190 struct timespec ts;
191
192 ASSERT_VI_LOCKED(vp, __func__);
193
194 ip = VTOI(vp);
195 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
196 return;
197 if ((vp->v_type == VBLK || vp->v_type == VCHR))
198 ip->i_flag |= IN_LAZYMOD;
199 else
200 ip->i_flag |= IN_MODIFIED;
201 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
202 vfs_timestamp(&ts);
203 if (ip->i_flag & IN_ACCESS) {
204 ip->i_atime = ts.tv_sec;
205 ip->i_atimensec = ts.tv_nsec;
206 }
207 if (ip->i_flag & IN_UPDATE) {
208 ip->i_mtime = ts.tv_sec;
209 ip->i_mtimensec = ts.tv_nsec;
210 ip->i_modrev++;
211 }
212 if (ip->i_flag & IN_CHANGE) {
213 ip->i_ctime = ts.tv_sec;
214 ip->i_ctimensec = ts.tv_nsec;
215 }
216 }
217 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
218 }
219
220 void
221 ext2_itimes(struct vnode *vp)
222 {
223
224 VI_LOCK(vp);
225 ext2_itimes_locked(vp);
226 VI_UNLOCK(vp);
227 }
228
229 /*
230 * Create a regular file
231 */
232 static int
233 ext2_create(struct vop_create_args *ap)
234 {
235 int error;
236
237 error =
238 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
239 ap->a_dvp, ap->a_vpp, ap->a_cnp);
240 if (error)
241 return (error);
242 return (0);
243 }
244
245 static int
246 ext2_open(struct vop_open_args *ap)
247 {
248
249 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR)
250 return (EOPNOTSUPP);
251
252 /*
253 * Files marked append-only must be opened for appending.
254 */
255 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
256 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
257 return (EPERM);
258
259 vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td);
260
261 return (0);
262 }
263
264 /*
265 * Close called.
266 *
267 * Update the times on the inode.
268 */
269 static int
270 ext2_close(struct vop_close_args *ap)
271 {
272 struct vnode *vp = ap->a_vp;
273
274 VI_LOCK(vp);
275 if (vp->v_usecount > 1)
276 ext2_itimes_locked(vp);
277 VI_UNLOCK(vp);
278 return (0);
279 }
280
281 static int
282 ext2_access(struct vop_access_args *ap)
283 {
284 struct vnode *vp = ap->a_vp;
285 struct inode *ip = VTOI(vp);
286 accmode_t accmode = ap->a_accmode;
287 int error;
288
289 if (vp->v_type == VBLK || vp->v_type == VCHR)
290 return (EOPNOTSUPP);
291
292 /*
293 * Disallow write attempts on read-only file systems;
294 * unless the file is a socket, fifo, or a block or
295 * character device resident on the file system.
296 */
297 if (accmode & VWRITE) {
298 switch (vp->v_type) {
299 case VDIR:
300 case VLNK:
301 case VREG:
302 if (vp->v_mount->mnt_flag & MNT_RDONLY)
303 return (EROFS);
304 break;
305 default:
306 break;
307 }
308 }
309
310 /* If immutable bit set, nobody gets to write it. */
311 if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT)))
312 return (EPERM);
313
314 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
315 ap->a_accmode, ap->a_cred, NULL);
316 return (error);
317 }
318
319 static int
320 ext2_getattr(struct vop_getattr_args *ap)
321 {
322 struct vnode *vp = ap->a_vp;
323 struct inode *ip = VTOI(vp);
324 struct vattr *vap = ap->a_vap;
325
326 ext2_itimes(vp);
327 /*
328 * Copy from inode table
329 */
330 vap->va_fsid = dev2udev(ip->i_devvp->v_rdev);
331 vap->va_fileid = ip->i_number;
332 vap->va_mode = ip->i_mode & ~IFMT;
333 vap->va_nlink = ip->i_nlink;
334 vap->va_uid = ip->i_uid;
335 vap->va_gid = ip->i_gid;
336 vap->va_rdev = ip->i_rdev;
337 vap->va_size = ip->i_size;
338 vap->va_atime.tv_sec = ip->i_atime;
339 vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0;
340 vap->va_mtime.tv_sec = ip->i_mtime;
341 vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0;
342 vap->va_ctime.tv_sec = ip->i_ctime;
343 vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0;
344 if E2DI_HAS_XTIME(ip) {
345 vap->va_birthtime.tv_sec = ip->i_birthtime;
346 vap->va_birthtime.tv_nsec = ip->i_birthnsec;
347 }
348 vap->va_flags = ip->i_flags;
349 vap->va_gen = ip->i_gen;
350 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
351 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
352 vap->va_type = IFTOVT(ip->i_mode);
353 vap->va_filerev = ip->i_modrev;
354 return (0);
355 }
356
357 /*
358 * Set attribute vnode op. called from several syscalls
359 */
360 static int
361 ext2_setattr(struct vop_setattr_args *ap)
362 {
363 struct vattr *vap = ap->a_vap;
364 struct vnode *vp = ap->a_vp;
365 struct inode *ip = VTOI(vp);
366 struct ucred *cred = ap->a_cred;
367 struct thread *td = curthread;
368 int error;
369
370 /*
371 * Check for unsettable attributes.
372 */
373 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
374 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
375 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
376 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
377 return (EINVAL);
378 }
379 if (vap->va_flags != VNOVAL) {
380 /* Disallow flags not supported by ext2fs. */
381 if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP))
382 return (EOPNOTSUPP);
383
384 if (vp->v_mount->mnt_flag & MNT_RDONLY)
385 return (EROFS);
386 /*
387 * Callers may only modify the file flags on objects they
388 * have VADMIN rights for.
389 */
390 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
391 return (error);
392 /*
393 * Unprivileged processes and privileged processes in
394 * jail() are not permitted to unset system flags, or
395 * modify flags if any system flags are set.
396 * Privileged non-jail processes may not modify system flags
397 * if securelevel > 0 and any existing system flags are set.
398 */
399 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
400 if (ip->i_flags &
401 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
402 error = securelevel_gt(cred, 0);
403 if (error)
404 return (error);
405 }
406 ip->i_flags = vap->va_flags;
407 } else {
408 if (ip->i_flags &
409 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
410 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
411 return (EPERM);
412 ip->i_flags &= SF_SETTABLE;
413 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
414 }
415 ip->i_flag |= IN_CHANGE;
416 if (ip->i_flags & (IMMUTABLE | APPEND))
417 return (0);
418 }
419 if (ip->i_flags & (IMMUTABLE | APPEND))
420 return (EPERM);
421 /*
422 * Go through the fields and update iff not VNOVAL.
423 */
424 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
425 if (vp->v_mount->mnt_flag & MNT_RDONLY)
426 return (EROFS);
427 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
428 td)) != 0)
429 return (error);
430 }
431 if (vap->va_size != VNOVAL) {
432 /*
433 * Disallow write attempts on read-only file systems;
434 * unless the file is a socket, fifo, or a block or
435 * character device resident on the file system.
436 */
437 switch (vp->v_type) {
438 case VDIR:
439 return (EISDIR);
440 case VLNK:
441 case VREG:
442 if (vp->v_mount->mnt_flag & MNT_RDONLY)
443 return (EROFS);
444 break;
445 default:
446 break;
447 }
448 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
449 return (error);
450 }
451 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
452 if (vp->v_mount->mnt_flag & MNT_RDONLY)
453 return (EROFS);
454 /*
455 * From utimes(2):
456 * If times is NULL, ... The caller must be the owner of
457 * the file, have permission to write the file, or be the
458 * super-user.
459 * If times is non-NULL, ... The caller must be the owner of
460 * the file or be the super-user.
461 */
462 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
463 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
464 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
465 return (error);
466 if (vap->va_atime.tv_sec != VNOVAL)
467 ip->i_flag |= IN_ACCESS;
468 if (vap->va_mtime.tv_sec != VNOVAL)
469 ip->i_flag |= IN_CHANGE | IN_UPDATE;
470 ext2_itimes(vp);
471 if (vap->va_atime.tv_sec != VNOVAL) {
472 ip->i_atime = vap->va_atime.tv_sec;
473 ip->i_atimensec = vap->va_atime.tv_nsec;
474 }
475 if (vap->va_mtime.tv_sec != VNOVAL) {
476 ip->i_mtime = vap->va_mtime.tv_sec;
477 ip->i_mtimensec = vap->va_mtime.tv_nsec;
478 }
479 ip->i_birthtime = vap->va_birthtime.tv_sec;
480 ip->i_birthnsec = vap->va_birthtime.tv_nsec;
481 error = ext2_update(vp, 0);
482 if (error)
483 return (error);
484 }
485 error = 0;
486 if (vap->va_mode != (mode_t)VNOVAL) {
487 if (vp->v_mount->mnt_flag & MNT_RDONLY)
488 return (EROFS);
489 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
490 }
491 return (error);
492 }
493
494 /*
495 * Change the mode on a file.
496 * Inode must be locked before calling.
497 */
498 static int
499 ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
500 {
501 struct inode *ip = VTOI(vp);
502 int error;
503
504 /*
505 * To modify the permissions on a file, must possess VADMIN
506 * for that file.
507 */
508 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
509 return (error);
510 /*
511 * Privileged processes may set the sticky bit on non-directories,
512 * as well as set the setgid bit on a file with a group that the
513 * process is not a member of.
514 */
515 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
516 error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0);
517 if (error)
518 return (EFTYPE);
519 }
520 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
521 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
522 if (error)
523 return (error);
524 }
525 ip->i_mode &= ~ALLPERMS;
526 ip->i_mode |= (mode & ALLPERMS);
527 ip->i_flag |= IN_CHANGE;
528 return (0);
529 }
530
531 /*
532 * Perform chown operation on inode ip;
533 * inode must be locked prior to call.
534 */
535 static int
536 ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
537 struct thread *td)
538 {
539 struct inode *ip = VTOI(vp);
540 uid_t ouid;
541 gid_t ogid;
542 int error = 0;
543
544 if (uid == (uid_t)VNOVAL)
545 uid = ip->i_uid;
546 if (gid == (gid_t)VNOVAL)
547 gid = ip->i_gid;
548 /*
549 * To modify the ownership of a file, must possess VADMIN
550 * for that file.
551 */
552 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
553 return (error);
554 /*
555 * To change the owner of a file, or change the group of a file
556 * to a group of which we are not a member, the caller must
557 * have privilege.
558 */
559 if (uid != ip->i_uid || (gid != ip->i_gid &&
560 !groupmember(gid, cred))) {
561 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
562 if (error)
563 return (error);
564 }
565 ogid = ip->i_gid;
566 ouid = ip->i_uid;
567 ip->i_gid = gid;
568 ip->i_uid = uid;
569 ip->i_flag |= IN_CHANGE;
570 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
571 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0)
572 ip->i_mode &= ~(ISUID | ISGID);
573 }
574 return (0);
575 }
576
577 /*
578 * Synch an open file.
579 */
580 /* ARGSUSED */
581 static int
582 ext2_fsync(struct vop_fsync_args *ap)
583 {
584 /*
585 * Flush all dirty buffers associated with a vnode.
586 */
587
588 vop_stdfsync(ap);
589
590 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
591 }
592
593 /*
594 * Mknod vnode call
595 */
596 /* ARGSUSED */
597 static int
598 ext2_mknod(struct vop_mknod_args *ap)
599 {
600 struct vattr *vap = ap->a_vap;
601 struct vnode **vpp = ap->a_vpp;
602 struct inode *ip;
603 ino_t ino;
604 int error;
605
606 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
607 ap->a_dvp, vpp, ap->a_cnp);
608 if (error)
609 return (error);
610 ip = VTOI(*vpp);
611 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
612 if (vap->va_rdev != VNOVAL) {
613 /*
614 * Want to be able to use this to make badblock
615 * inodes, so don't truncate the dev number.
616 */
617 ip->i_rdev = vap->va_rdev;
618 }
619 /*
620 * Remove inode, then reload it through VFS_VGET so it is
621 * checked to see if it is an alias of an existing entry in
622 * the inode cache. XXX I don't believe this is necessary now.
623 */
624 (*vpp)->v_type = VNON;
625 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
626 vgone(*vpp);
627 vput(*vpp);
628 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
629 if (error) {
630 *vpp = NULL;
631 return (error);
632 }
633 return (0);
634 }
635
636 static int
637 ext2_remove(struct vop_remove_args *ap)
638 {
639 struct inode *ip;
640 struct vnode *vp = ap->a_vp;
641 struct vnode *dvp = ap->a_dvp;
642 int error;
643
644 ip = VTOI(vp);
645 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
646 (VTOI(dvp)->i_flags & APPEND)) {
647 error = EPERM;
648 goto out;
649 }
650 error = ext2_dirremove(dvp, ap->a_cnp);
651 if (error == 0) {
652 ip->i_nlink--;
653 ip->i_flag |= IN_CHANGE;
654 }
655 out:
656 return (error);
657 }
658
659 /*
660 * link vnode call
661 */
662 static int
663 ext2_link(struct vop_link_args *ap)
664 {
665 struct vnode *vp = ap->a_vp;
666 struct vnode *tdvp = ap->a_tdvp;
667 struct componentname *cnp = ap->a_cnp;
668 struct inode *ip;
669 int error;
670
671 #ifdef INVARIANTS
672 if ((cnp->cn_flags & HASBUF) == 0)
673 panic("ext2_link: no name");
674 #endif
675 if (tdvp->v_mount != vp->v_mount) {
676 error = EXDEV;
677 goto out;
678 }
679 ip = VTOI(vp);
680 if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) {
681 error = EMLINK;
682 goto out;
683 }
684 if (ip->i_flags & (IMMUTABLE | APPEND)) {
685 error = EPERM;
686 goto out;
687 }
688 ip->i_nlink++;
689 ip->i_flag |= IN_CHANGE;
690 error = ext2_update(vp, !DOINGASYNC(vp));
691 if (!error)
692 error = ext2_direnter(ip, tdvp, cnp);
693 if (error) {
694 ip->i_nlink--;
695 ip->i_flag |= IN_CHANGE;
696 }
697 out:
698 return (error);
699 }
700
701 /*
702 * Rename system call.
703 * rename("foo", "bar");
704 * is essentially
705 * unlink("bar");
706 * link("foo", "bar");
707 * unlink("foo");
708 * but ``atomically''. Can't do full commit without saving state in the
709 * inode on disk which isn't feasible at this time. Best we can do is
710 * always guarantee the target exists.
711 *
712 * Basic algorithm is:
713 *
714 * 1) Bump link count on source while we're linking it to the
715 * target. This also ensure the inode won't be deleted out
716 * from underneath us while we work (it may be truncated by
717 * a concurrent `trunc' or `open' for creation).
718 * 2) Link source to destination. If destination already exists,
719 * delete it first.
720 * 3) Unlink source reference to inode if still around. If a
721 * directory was moved and the parent of the destination
722 * is different from the source, patch the ".." entry in the
723 * directory.
724 */
725 static int
726 ext2_rename(struct vop_rename_args *ap)
727 {
728 struct vnode *tvp = ap->a_tvp;
729 struct vnode *tdvp = ap->a_tdvp;
730 struct vnode *fvp = ap->a_fvp;
731 struct vnode *fdvp = ap->a_fdvp;
732 struct componentname *tcnp = ap->a_tcnp;
733 struct componentname *fcnp = ap->a_fcnp;
734 struct inode *ip, *xp, *dp;
735 struct dirtemplate dirbuf;
736 int doingdirectory = 0, oldparent = 0, newparent = 0;
737 int error = 0;
738 u_char namlen;
739
740 #ifdef INVARIANTS
741 if ((tcnp->cn_flags & HASBUF) == 0 ||
742 (fcnp->cn_flags & HASBUF) == 0)
743 panic("ext2_rename: no name");
744 #endif
745 /*
746 * Check for cross-device rename.
747 */
748 if ((fvp->v_mount != tdvp->v_mount) ||
749 (tvp && (fvp->v_mount != tvp->v_mount))) {
750 error = EXDEV;
751 abortit:
752 if (tdvp == tvp)
753 vrele(tdvp);
754 else
755 vput(tdvp);
756 if (tvp)
757 vput(tvp);
758 vrele(fdvp);
759 vrele(fvp);
760 return (error);
761 }
762
763 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
764 (VTOI(tdvp)->i_flags & APPEND))) {
765 error = EPERM;
766 goto abortit;
767 }
768
769 /*
770 * Renaming a file to itself has no effect. The upper layers should
771 * not call us in that case. Temporarily just warn if they do.
772 */
773 if (fvp == tvp) {
774 printf("ext2_rename: fvp == tvp (can't happen)\n");
775 error = 0;
776 goto abortit;
777 }
778
779 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
780 goto abortit;
781 dp = VTOI(fdvp);
782 ip = VTOI(fvp);
783 if (ip->i_nlink >= EXT2_LINK_MAX) {
784 VOP_UNLOCK(fvp, 0);
785 error = EMLINK;
786 goto abortit;
787 }
788 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
789 || (dp->i_flags & APPEND)) {
790 VOP_UNLOCK(fvp, 0);
791 error = EPERM;
792 goto abortit;
793 }
794 if ((ip->i_mode & IFMT) == IFDIR) {
795 /*
796 * Avoid ".", "..", and aliases of "." for obvious reasons.
797 */
798 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
799 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
800 (ip->i_flag & IN_RENAME)) {
801 VOP_UNLOCK(fvp, 0);
802 error = EINVAL;
803 goto abortit;
804 }
805 ip->i_flag |= IN_RENAME;
806 oldparent = dp->i_number;
807 doingdirectory++;
808 }
809 vrele(fdvp);
810
811 /*
812 * When the target exists, both the directory
813 * and target vnodes are returned locked.
814 */
815 dp = VTOI(tdvp);
816 xp = NULL;
817 if (tvp)
818 xp = VTOI(tvp);
819
820 /*
821 * 1) Bump link count while we're moving stuff
822 * around. If we crash somewhere before
823 * completing our work, the link count
824 * may be wrong, but correctable.
825 */
826 ip->i_nlink++;
827 ip->i_flag |= IN_CHANGE;
828 if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) {
829 VOP_UNLOCK(fvp, 0);
830 goto bad;
831 }
832
833 /*
834 * If ".." must be changed (ie the directory gets a new
835 * parent) then the source directory must not be in the
836 * directory hierarchy above the target, as this would
837 * orphan everything below the source directory. Also
838 * the user must have write permission in the source so
839 * as to be able to change "..". We must repeat the call
840 * to namei, as the parent directory is unlocked by the
841 * call to checkpath().
842 */
843 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
844 VOP_UNLOCK(fvp, 0);
845 if (oldparent != dp->i_number)
846 newparent = dp->i_number;
847 if (doingdirectory && newparent) {
848 if (error) /* write access check above */
849 goto bad;
850 if (xp != NULL)
851 vput(tvp);
852 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
853 if (error)
854 goto out;
855 VREF(tdvp);
856 error = relookup(tdvp, &tvp, tcnp);
857 if (error)
858 goto out;
859 vrele(tdvp);
860 dp = VTOI(tdvp);
861 xp = NULL;
862 if (tvp)
863 xp = VTOI(tvp);
864 }
865 /*
866 * 2) If target doesn't exist, link the target
867 * to the source and unlink the source.
868 * Otherwise, rewrite the target directory
869 * entry to reference the source inode and
870 * expunge the original entry's existence.
871 */
872 if (xp == NULL) {
873 if (dp->i_devvp != ip->i_devvp)
874 panic("ext2_rename: EXDEV");
875 /*
876 * Account for ".." in new directory.
877 * When source and destination have the same
878 * parent we don't fool with the link count.
879 */
880 if (doingdirectory && newparent) {
881 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
882 error = EMLINK;
883 goto bad;
884 }
885 dp->i_nlink++;
886 dp->i_flag |= IN_CHANGE;
887 error = ext2_update(tdvp, !DOINGASYNC(tdvp));
888 if (error)
889 goto bad;
890 }
891 error = ext2_direnter(ip, tdvp, tcnp);
892 if (error) {
893 if (doingdirectory && newparent) {
894 dp->i_nlink--;
895 dp->i_flag |= IN_CHANGE;
896 (void)ext2_update(tdvp, 1);
897 }
898 goto bad;
899 }
900 vput(tdvp);
901 } else {
902 if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp)
903 panic("ext2_rename: EXDEV");
904 /*
905 * Short circuit rename(foo, foo).
906 */
907 if (xp->i_number == ip->i_number)
908 panic("ext2_rename: same file");
909 /*
910 * If the parent directory is "sticky", then the user must
911 * own the parent directory, or the destination of the rename,
912 * otherwise the destination may not be changed (except by
913 * root). This implements append-only directories.
914 */
915 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
916 tcnp->cn_cred->cr_uid != dp->i_uid &&
917 xp->i_uid != tcnp->cn_cred->cr_uid) {
918 error = EPERM;
919 goto bad;
920 }
921 /*
922 * Target must be empty if a directory and have no links
923 * to it. Also, ensure source and target are compatible
924 * (both directories, or both not directories).
925 */
926 if ((xp->i_mode&IFMT) == IFDIR) {
927 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
928 xp->i_nlink > 2) {
929 error = ENOTEMPTY;
930 goto bad;
931 }
932 if (!doingdirectory) {
933 error = ENOTDIR;
934 goto bad;
935 }
936 cache_purge(tdvp);
937 } else if (doingdirectory) {
938 error = EISDIR;
939 goto bad;
940 }
941 error = ext2_dirrewrite(dp, ip, tcnp);
942 if (error)
943 goto bad;
944 /*
945 * If the target directory is in the same
946 * directory as the source directory,
947 * decrement the link count on the parent
948 * of the target directory.
949 */
950 if (doingdirectory && !newparent) {
951 dp->i_nlink--;
952 dp->i_flag |= IN_CHANGE;
953 }
954 vput(tdvp);
955 /*
956 * Adjust the link count of the target to
957 * reflect the dirrewrite above. If this is
958 * a directory it is empty and there are
959 * no links to it, so we can squash the inode and
960 * any space associated with it. We disallowed
961 * renaming over top of a directory with links to
962 * it above, as the remaining link would point to
963 * a directory without "." or ".." entries.
964 */
965 xp->i_nlink--;
966 if (doingdirectory) {
967 if (--xp->i_nlink != 0)
968 panic("ext2_rename: linked directory");
969 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
970 tcnp->cn_cred, tcnp->cn_thread);
971 }
972 xp->i_flag |= IN_CHANGE;
973 vput(tvp);
974 xp = NULL;
975 }
976
977 /*
978 * 3) Unlink the source.
979 */
980 fcnp->cn_flags &= ~MODMASK;
981 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
982 VREF(fdvp);
983 error = relookup(fdvp, &fvp, fcnp);
984 if (error == 0)
985 vrele(fdvp);
986 if (fvp != NULL) {
987 xp = VTOI(fvp);
988 dp = VTOI(fdvp);
989 } else {
990 /*
991 * From name has disappeared. IN_RENAME is not sufficient
992 * to protect against directory races due to timing windows,
993 * so we can't panic here.
994 */
995 vrele(ap->a_fvp);
996 return (0);
997 }
998 /*
999 * Ensure that the directory entry still exists and has not
1000 * changed while the new name has been entered. If the source is
1001 * a file then the entry may have been unlinked or renamed. In
1002 * either case there is no further work to be done. If the source
1003 * is a directory then it cannot have been rmdir'ed; its link
1004 * count of three would cause a rmdir to fail with ENOTEMPTY.
1005 * The IN_RENAME flag ensures that it cannot be moved by another
1006 * rename.
1007 */
1008 if (xp != ip) {
1009 /*
1010 * From name resolves to a different inode. IN_RENAME is
1011 * not sufficient protection against timing window races
1012 * so we can't panic here.
1013 */
1014 } else {
1015 /*
1016 * If the source is a directory with a
1017 * new parent, the link count of the old
1018 * parent directory must be decremented
1019 * and ".." set to point to the new parent.
1020 */
1021 if (doingdirectory && newparent) {
1022 dp->i_nlink--;
1023 dp->i_flag |= IN_CHANGE;
1024 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1025 sizeof(struct dirtemplate), (off_t)0,
1026 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1027 tcnp->cn_cred, NOCRED, NULL, NULL);
1028 if (error == 0) {
1029 /* Like ufs little-endian: */
1030 namlen = dirbuf.dotdot_type;
1031 if (namlen != 2 ||
1032 dirbuf.dotdot_name[0] != '.' ||
1033 dirbuf.dotdot_name[1] != '.') {
1034 ext2_dirbad(xp, (doff_t)12,
1035 "rename: mangled dir");
1036 } else {
1037 dirbuf.dotdot_ino = newparent;
1038 (void) vn_rdwr(UIO_WRITE, fvp,
1039 (caddr_t)&dirbuf,
1040 sizeof(struct dirtemplate),
1041 (off_t)0, UIO_SYSSPACE,
1042 IO_NODELOCKED | IO_SYNC |
1043 IO_NOMACCHECK, tcnp->cn_cred,
1044 NOCRED, NULL, NULL);
1045 cache_purge(fdvp);
1046 }
1047 }
1048 }
1049 error = ext2_dirremove(fdvp, fcnp);
1050 if (!error) {
1051 xp->i_nlink--;
1052 xp->i_flag |= IN_CHANGE;
1053 }
1054 xp->i_flag &= ~IN_RENAME;
1055 }
1056 if (dp)
1057 vput(fdvp);
1058 if (xp)
1059 vput(fvp);
1060 vrele(ap->a_fvp);
1061 return (error);
1062
1063 bad:
1064 if (xp)
1065 vput(ITOV(xp));
1066 vput(ITOV(dp));
1067 out:
1068 if (doingdirectory)
1069 ip->i_flag &= ~IN_RENAME;
1070 if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1071 ip->i_nlink--;
1072 ip->i_flag |= IN_CHANGE;
1073 ip->i_flag &= ~IN_RENAME;
1074 vput(fvp);
1075 } else
1076 vrele(fvp);
1077 return (error);
1078 }
1079
1080 /*
1081 * Mkdir system call
1082 */
1083 static int
1084 ext2_mkdir(struct vop_mkdir_args *ap)
1085 {
1086 struct vnode *dvp = ap->a_dvp;
1087 struct vattr *vap = ap->a_vap;
1088 struct componentname *cnp = ap->a_cnp;
1089 struct inode *ip, *dp;
1090 struct vnode *tvp;
1091 struct dirtemplate dirtemplate, *dtp;
1092 int error, dmode;
1093
1094 #ifdef INVARIANTS
1095 if ((cnp->cn_flags & HASBUF) == 0)
1096 panic("ext2_mkdir: no name");
1097 #endif
1098 dp = VTOI(dvp);
1099 if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) {
1100 error = EMLINK;
1101 goto out;
1102 }
1103 dmode = vap->va_mode & 0777;
1104 dmode |= IFDIR;
1105 /*
1106 * Must simulate part of ext2_makeinode here to acquire the inode,
1107 * but not have it entered in the parent directory. The entry is
1108 * made later after writing "." and ".." entries.
1109 */
1110 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1111 if (error)
1112 goto out;
1113 ip = VTOI(tvp);
1114 ip->i_gid = dp->i_gid;
1115 #ifdef SUIDDIR
1116 {
1117 /*
1118 * if we are hacking owners here, (only do this where told to)
1119 * and we are not giving it TOO root, (would subvert quotas)
1120 * then go ahead and give it to the other user.
1121 * The new directory also inherits the SUID bit.
1122 * If user's UID and dir UID are the same,
1123 * 'give it away' so that the SUID is still forced on.
1124 */
1125 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1126 (dp->i_mode & ISUID) && dp->i_uid) {
1127 dmode |= ISUID;
1128 ip->i_uid = dp->i_uid;
1129 } else {
1130 ip->i_uid = cnp->cn_cred->cr_uid;
1131 }
1132 }
1133 #else
1134 ip->i_uid = cnp->cn_cred->cr_uid;
1135 #endif
1136 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1137 ip->i_mode = dmode;
1138 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1139 ip->i_nlink = 2;
1140 if (cnp->cn_flags & ISWHITEOUT)
1141 ip->i_flags |= UF_OPAQUE;
1142 error = ext2_update(tvp, 1);
1143
1144 /*
1145 * Bump link count in parent directory
1146 * to reflect work done below. Should
1147 * be done before reference is created
1148 * so reparation is possible if we crash.
1149 */
1150 dp->i_nlink++;
1151 dp->i_flag |= IN_CHANGE;
1152 error = ext2_update(dvp, !DOINGASYNC(dvp));
1153 if (error)
1154 goto bad;
1155
1156 /* Initialize directory with "." and ".." from static template. */
1157 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1158 EXT2F_INCOMPAT_FTYPE))
1159 dtp = &mastertemplate;
1160 else
1161 dtp = &omastertemplate;
1162 dirtemplate = *dtp;
1163 dirtemplate.dot_ino = ip->i_number;
1164 dirtemplate.dotdot_ino = dp->i_number;
1165 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1166 * so let's just redefine it - for this function only
1167 */
1168 #undef DIRBLKSIZ
1169 #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize
1170 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1171 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1172 sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE,
1173 IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED,
1174 NULL, NULL);
1175 if (error) {
1176 dp->i_nlink--;
1177 dp->i_flag |= IN_CHANGE;
1178 goto bad;
1179 }
1180 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1181 /* XXX should grow with balloc() */
1182 panic("ext2_mkdir: blksize");
1183 else {
1184 ip->i_size = DIRBLKSIZ;
1185 ip->i_flag |= IN_CHANGE;
1186 }
1187
1188 /* Directory set up, now install its entry in the parent directory. */
1189 error = ext2_direnter(ip, dvp, cnp);
1190 if (error) {
1191 dp->i_nlink--;
1192 dp->i_flag |= IN_CHANGE;
1193 }
1194 bad:
1195 /*
1196 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1197 * for us because we set the link count to 0.
1198 */
1199 if (error) {
1200 ip->i_nlink = 0;
1201 ip->i_flag |= IN_CHANGE;
1202 vput(tvp);
1203 } else
1204 *ap->a_vpp = tvp;
1205 out:
1206 return (error);
1207 #undef DIRBLKSIZ
1208 #define DIRBLKSIZ DEV_BSIZE
1209 }
1210
1211 /*
1212 * Rmdir system call.
1213 */
1214 static int
1215 ext2_rmdir(struct vop_rmdir_args *ap)
1216 {
1217 struct vnode *vp = ap->a_vp;
1218 struct vnode *dvp = ap->a_dvp;
1219 struct componentname *cnp = ap->a_cnp;
1220 struct inode *ip, *dp;
1221 int error;
1222
1223 ip = VTOI(vp);
1224 dp = VTOI(dvp);
1225
1226 /*
1227 * Verify the directory is empty (and valid).
1228 * (Rmdir ".." won't be valid since
1229 * ".." will contain a reference to
1230 * the current directory and thus be
1231 * non-empty.)
1232 */
1233 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1234 error = ENOTEMPTY;
1235 goto out;
1236 }
1237 if ((dp->i_flags & APPEND)
1238 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1239 error = EPERM;
1240 goto out;
1241 }
1242 /*
1243 * Delete reference to directory before purging
1244 * inode. If we crash in between, the directory
1245 * will be reattached to lost+found,
1246 */
1247 error = ext2_dirremove(dvp, cnp);
1248 if (error)
1249 goto out;
1250 dp->i_nlink--;
1251 dp->i_flag |= IN_CHANGE;
1252 cache_purge(dvp);
1253 VOP_UNLOCK(dvp, 0);
1254 /*
1255 * Truncate inode. The only stuff left
1256 * in the directory is "." and "..". The
1257 * "." reference is inconsequential since
1258 * we're quashing it. The ".." reference
1259 * has already been adjusted above. We've
1260 * removed the "." reference and the reference
1261 * in the parent directory, but there may be
1262 * other hard links so decrement by 2 and
1263 * worry about them later.
1264 */
1265 ip->i_nlink -= 2;
1266 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
1267 cnp->cn_thread);
1268 cache_purge(ITOV(ip));
1269 if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1270 VOP_UNLOCK(vp, 0);
1271 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1272 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1273 }
1274 out:
1275 return (error);
1276 }
1277
1278 /*
1279 * symlink -- make a symbolic link
1280 */
1281 static int
1282 ext2_symlink(struct vop_symlink_args *ap)
1283 {
1284 struct vnode *vp, **vpp = ap->a_vpp;
1285 struct inode *ip;
1286 int len, error;
1287
1288 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1289 vpp, ap->a_cnp);
1290 if (error)
1291 return (error);
1292 vp = *vpp;
1293 len = strlen(ap->a_target);
1294 if (len < vp->v_mount->mnt_maxsymlinklen) {
1295 ip = VTOI(vp);
1296 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1297 ip->i_size = len;
1298 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1299 } else
1300 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1301 UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
1302 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
1303 if (error)
1304 vput(vp);
1305 return (error);
1306 }
1307
1308 /*
1309 * Return target name of a symbolic link
1310 */
1311 static int
1312 ext2_readlink(struct vop_readlink_args *ap)
1313 {
1314 struct vnode *vp = ap->a_vp;
1315 struct inode *ip = VTOI(vp);
1316 int isize;
1317
1318 isize = ip->i_size;
1319 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1320 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1321 return (0);
1322 }
1323 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1324 }
1325
1326 /*
1327 * Calculate the logical to physical mapping if not done already,
1328 * then call the device strategy routine.
1329 *
1330 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1331 * deadlock on memory. See ext2_bmap() for details.
1332 */
1333 static int
1334 ext2_strategy(struct vop_strategy_args *ap)
1335 {
1336 struct buf *bp = ap->a_bp;
1337 struct vnode *vp = ap->a_vp;
1338 struct bufobj *bo;
1339 daddr_t blkno;
1340 int error;
1341
1342 if (vp->v_type == VBLK || vp->v_type == VCHR)
1343 panic("ext2_strategy: spec");
1344 if (bp->b_blkno == bp->b_lblkno) {
1345 error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL);
1346 bp->b_blkno = blkno;
1347 if (error) {
1348 bp->b_error = error;
1349 bp->b_ioflags |= BIO_ERROR;
1350 bufdone(bp);
1351 return (0);
1352 }
1353 if ((long)bp->b_blkno == -1)
1354 vfs_bio_clrbuf(bp);
1355 }
1356 if ((long)bp->b_blkno == -1) {
1357 bufdone(bp);
1358 return (0);
1359 }
1360 bp->b_iooffset = dbtob(bp->b_blkno);
1361 bo = VFSTOEXT2(vp->v_mount)->um_bo;
1362 BO_STRATEGY(bo, bp);
1363 return (0);
1364 }
1365
1366 /*
1367 * Print out the contents of an inode.
1368 */
1369 static int
1370 ext2_print(struct vop_print_args *ap)
1371 {
1372 struct vnode *vp = ap->a_vp;
1373 struct inode *ip = VTOI(vp);
1374
1375 vn_printf(ip->i_devvp, "\tino %lu", (u_long)ip->i_number);
1376 if (vp->v_type == VFIFO)
1377 fifo_printinfo(vp);
1378 printf("\n");
1379 return (0);
1380 }
1381
1382 /*
1383 * Close wrapper for fifos.
1384 *
1385 * Update the times on the inode then do device close.
1386 */
1387 static int
1388 ext2fifo_close(struct vop_close_args *ap)
1389 {
1390 struct vnode *vp = ap->a_vp;
1391
1392 VI_LOCK(vp);
1393 if (vp->v_usecount > 1)
1394 ext2_itimes_locked(vp);
1395 VI_UNLOCK(vp);
1396 return (fifo_specops.vop_close(ap));
1397 }
1398
1399 /*
1400 * Kqfilter wrapper for fifos.
1401 *
1402 * Fall through to ext2 kqfilter routines if needed
1403 */
1404 static int
1405 ext2fifo_kqfilter(struct vop_kqfilter_args *ap)
1406 {
1407 int error;
1408
1409 error = fifo_specops.vop_kqfilter(ap);
1410 if (error)
1411 error = vfs_kqfilter(ap);
1412 return (error);
1413 }
1414
1415 /*
1416 * Return POSIX pathconf information applicable to ext2 filesystems.
1417 */
1418 static int
1419 ext2_pathconf(struct vop_pathconf_args *ap)
1420 {
1421 int error = 0;
1422
1423 switch (ap->a_name) {
1424 case _PC_LINK_MAX:
1425 *ap->a_retval = EXT2_LINK_MAX;
1426 break;
1427 case _PC_NAME_MAX:
1428 *ap->a_retval = NAME_MAX;
1429 break;
1430 case _PC_PATH_MAX:
1431 *ap->a_retval = PATH_MAX;
1432 break;
1433 case _PC_PIPE_BUF:
1434 *ap->a_retval = PIPE_BUF;
1435 break;
1436 case _PC_CHOWN_RESTRICTED:
1437 *ap->a_retval = 1;
1438 break;
1439 case _PC_NO_TRUNC:
1440 *ap->a_retval = 1;
1441 break;
1442 case _PC_MIN_HOLE_SIZE:
1443 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1444 break;
1445 case _PC_ASYNC_IO:
1446 /* _PC_ASYNC_IO should have been handled by upper layers. */
1447 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
1448 error = EINVAL;
1449 break;
1450 case _PC_PRIO_IO:
1451 *ap->a_retval = 0;
1452 break;
1453 case _PC_SYNC_IO:
1454 *ap->a_retval = 0;
1455 break;
1456 case _PC_ALLOC_SIZE_MIN:
1457 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
1458 break;
1459 case _PC_FILESIZEBITS:
1460 *ap->a_retval = 64;
1461 break;
1462 case _PC_REC_INCR_XFER_SIZE:
1463 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1464 break;
1465 case _PC_REC_MAX_XFER_SIZE:
1466 *ap->a_retval = -1; /* means ``unlimited'' */
1467 break;
1468 case _PC_REC_MIN_XFER_SIZE:
1469 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
1470 break;
1471 case _PC_REC_XFER_ALIGN:
1472 *ap->a_retval = PAGE_SIZE;
1473 break;
1474 case _PC_SYMLINK_MAX:
1475 *ap->a_retval = MAXPATHLEN;
1476 break;
1477
1478 default:
1479 error = EINVAL;
1480 break;
1481 }
1482 return (error);
1483 }
1484
1485 /*
1486 * Vnode pointer to File handle
1487 */
1488 /* ARGSUSED */
1489 static int
1490 ext2_vptofh(struct vop_vptofh_args *ap)
1491 {
1492 struct inode *ip;
1493 struct ufid *ufhp;
1494
1495 ip = VTOI(ap->a_vp);
1496 ufhp = (struct ufid *)ap->a_fhp;
1497 ufhp->ufid_len = sizeof(struct ufid);
1498 ufhp->ufid_ino = ip->i_number;
1499 ufhp->ufid_gen = ip->i_gen;
1500 return (0);
1501 }
1502
1503 /*
1504 * Initialize the vnode associated with a new inode, handle aliased
1505 * vnodes.
1506 */
1507 int
1508 ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
1509 {
1510 struct inode *ip;
1511 struct vnode *vp;
1512
1513 vp = *vpp;
1514 ip = VTOI(vp);
1515 vp->v_type = IFTOVT(ip->i_mode);
1516 if (vp->v_type == VFIFO)
1517 vp->v_op = fifoops;
1518
1519 if (ip->i_number == EXT2_ROOTINO)
1520 vp->v_vflag |= VV_ROOT;
1521 ip->i_modrev = init_va_filerev();
1522 *vpp = vp;
1523 return (0);
1524 }
1525
1526 /*
1527 * Allocate a new inode.
1528 */
1529 static int
1530 ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
1531 struct componentname *cnp)
1532 {
1533 struct inode *ip, *pdir;
1534 struct vnode *tvp;
1535 int error;
1536
1537 pdir = VTOI(dvp);
1538 #ifdef INVARIANTS
1539 if ((cnp->cn_flags & HASBUF) == 0)
1540 panic("ext2_makeinode: no name");
1541 #endif
1542 *vpp = NULL;
1543 if ((mode & IFMT) == 0)
1544 mode |= IFREG;
1545
1546 error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp);
1547 if (error) {
1548 return (error);
1549 }
1550 ip = VTOI(tvp);
1551 ip->i_gid = pdir->i_gid;
1552 #ifdef SUIDDIR
1553 {
1554 /*
1555 * if we are
1556 * not the owner of the directory,
1557 * and we are hacking owners here, (only do this where told to)
1558 * and we are not giving it TOO root, (would subvert quotas)
1559 * then go ahead and give it to the other user.
1560 * Note that this drops off the execute bits for security.
1561 */
1562 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1563 (pdir->i_mode & ISUID) &&
1564 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
1565 ip->i_uid = pdir->i_uid;
1566 mode &= ~07111;
1567 } else {
1568 ip->i_uid = cnp->cn_cred->cr_uid;
1569 }
1570 }
1571 #else
1572 ip->i_uid = cnp->cn_cred->cr_uid;
1573 #endif
1574 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1575 ip->i_mode = mode;
1576 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
1577 ip->i_nlink = 1;
1578 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) {
1579 if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0))
1580 ip->i_mode &= ~ISGID;
1581 }
1582
1583 if (cnp->cn_flags & ISWHITEOUT)
1584 ip->i_flags |= UF_OPAQUE;
1585
1586 /*
1587 * Make sure inode goes to disk before directory entry.
1588 */
1589 error = ext2_update(tvp, !DOINGASYNC(tvp));
1590 if (error)
1591 goto bad;
1592 error = ext2_direnter(ip, dvp, cnp);
1593 if (error)
1594 goto bad;
1595
1596 *vpp = tvp;
1597 return (0);
1598
1599 bad:
1600 /*
1601 * Write error occurred trying to update the inode
1602 * or the directory so must deallocate the inode.
1603 */
1604 ip->i_nlink = 0;
1605 ip->i_flag |= IN_CHANGE;
1606 vput(tvp);
1607 return (error);
1608 }
1609
1610 /*
1611 * Vnode op for reading.
1612 */
1613 static int
1614 ext2_read(struct vop_read_args *ap)
1615 {
1616 struct vnode *vp;
1617 struct inode *ip;
1618 int error;
1619
1620 vp = ap->a_vp;
1621 ip = VTOI(vp);
1622
1623 /*EXT4_EXT_LOCK(ip);*/
1624 if (ip->i_flag & IN_E4EXTENTS)
1625 error = ext4_ext_read(ap);
1626 else
1627 error = ext2_ind_read(ap);
1628 /*EXT4_EXT_UNLOCK(ip);*/
1629 return (error);
1630 }
1631
1632 /*
1633 * Vnode op for reading.
1634 */
1635 static int
1636 ext2_ind_read(struct vop_read_args *ap)
1637 {
1638 struct vnode *vp;
1639 struct inode *ip;
1640 struct uio *uio;
1641 struct m_ext2fs *fs;
1642 struct buf *bp;
1643 daddr_t lbn, nextlbn;
1644 off_t bytesinfile;
1645 long size, xfersize, blkoffset;
1646 int error, orig_resid, seqcount;
1647 int ioflag;
1648
1649 vp = ap->a_vp;
1650 uio = ap->a_uio;
1651 ioflag = ap->a_ioflag;
1652
1653 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
1654 ip = VTOI(vp);
1655
1656 #ifdef INVARIANTS
1657 if (uio->uio_rw != UIO_READ)
1658 panic("%s: mode", "ext2_read");
1659
1660 if (vp->v_type == VLNK) {
1661 if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
1662 panic("%s: short symlink", "ext2_read");
1663 } else if (vp->v_type != VREG && vp->v_type != VDIR)
1664 panic("%s: type %d", "ext2_read", vp->v_type);
1665 #endif
1666 orig_resid = uio->uio_resid;
1667 KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0"));
1668 if (orig_resid == 0)
1669 return (0);
1670 KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0"));
1671 fs = ip->i_e2fs;
1672 if (uio->uio_offset < ip->i_size &&
1673 uio->uio_offset >= fs->e2fs_maxfilesize)
1674 return (EOVERFLOW);
1675
1676 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1677 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1678 break;
1679 lbn = lblkno(fs, uio->uio_offset);
1680 nextlbn = lbn + 1;
1681 size = blksize(fs, ip, lbn);
1682 blkoffset = blkoff(fs, uio->uio_offset);
1683
1684 xfersize = fs->e2fs_fsize - blkoffset;
1685 if (uio->uio_resid < xfersize)
1686 xfersize = uio->uio_resid;
1687 if (bytesinfile < xfersize)
1688 xfersize = bytesinfile;
1689
1690 if (lblktosize(fs, nextlbn) >= ip->i_size)
1691 error = bread(vp, lbn, size, NOCRED, &bp);
1692 else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0)
1693 error = cluster_read(vp, ip->i_size, lbn, size,
1694 NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
1695 else if (seqcount > 1) {
1696 u_int nextsize = blksize(fs, ip, nextlbn);
1697 error = breadn(vp, lbn,
1698 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1699 } else
1700 error = bread(vp, lbn, size, NOCRED, &bp);
1701 if (error) {
1702 brelse(bp);
1703 bp = NULL;
1704 break;
1705 }
1706
1707 /*
1708 * If IO_DIRECT then set B_DIRECT for the buffer. This
1709 * will cause us to attempt to release the buffer later on
1710 * and will cause the buffer cache to attempt to free the
1711 * underlying pages.
1712 */
1713 if (ioflag & IO_DIRECT)
1714 bp->b_flags |= B_DIRECT;
1715
1716 /*
1717 * We should only get non-zero b_resid when an I/O error
1718 * has occurred, which should cause us to break above.
1719 * However, if the short read did not cause an error,
1720 * then we want to ensure that we do not uiomove bad
1721 * or uninitialized data.
1722 */
1723 size -= bp->b_resid;
1724 if (size < xfersize) {
1725 if (size == 0)
1726 break;
1727 xfersize = size;
1728 }
1729 error = uiomove((char *)bp->b_data + blkoffset,
1730 (int)xfersize, uio);
1731 if (error)
1732 break;
1733
1734 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1735 /*
1736 * If it's VMIO or direct I/O, then we don't
1737 * need the buf, mark it available for
1738 * freeing. If it's non-direct VMIO, the VM has
1739 * the data.
1740 */
1741 bp->b_flags |= B_RELBUF;
1742 brelse(bp);
1743 } else {
1744 /*
1745 * Otherwise let whoever
1746 * made the request take care of
1747 * freeing it. We just queue
1748 * it onto another list.
1749 */
1750 bqrelse(bp);
1751 }
1752 }
1753
1754 /*
1755 * This can only happen in the case of an error
1756 * because the loop above resets bp to NULL on each iteration
1757 * and on normal completion has not set a new value into it.
1758 * so it must have come from a 'break' statement
1759 */
1760 if (bp != NULL) {
1761 if (ioflag & (IO_VMIO|IO_DIRECT)) {
1762 bp->b_flags |= B_RELBUF;
1763 brelse(bp);
1764 } else {
1765 bqrelse(bp);
1766 }
1767 }
1768
1769 if ((error == 0 || uio->uio_resid != orig_resid) &&
1770 (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1771 ip->i_flag |= IN_ACCESS;
1772 return (error);
1773 }
1774
1775 static int
1776 ext2_ioctl(struct vop_ioctl_args *ap)
1777 {
1778
1779 switch (ap->a_command) {
1780 case FIOSEEKDATA:
1781 case FIOSEEKHOLE:
1782 return (vn_bmap_seekhole(ap->a_vp, ap->a_command,
1783 (off_t *)ap->a_data, ap->a_cred));
1784 default:
1785 return (ENOTTY);
1786 }
1787 }
1788
1789 /*
1790 * this function handles ext4 extents block mapping
1791 */
1792 static int
1793 ext4_ext_read(struct vop_read_args *ap)
1794 {
1795 struct vnode *vp;
1796 struct inode *ip;
1797 struct uio *uio;
1798 struct m_ext2fs *fs;
1799 struct buf *bp;
1800 struct ext4_extent nex, *ep;
1801 struct ext4_extent_path path;
1802 daddr_t lbn, newblk;
1803 off_t bytesinfile;
1804 int cache_type;
1805 ssize_t orig_resid;
1806 int error;
1807 long size, xfersize, blkoffset;
1808
1809 vp = ap->a_vp;
1810 ip = VTOI(vp);
1811 uio = ap->a_uio;
1812 memset(&path, 0, sizeof(path));
1813
1814 orig_resid = uio->uio_resid;
1815 KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__));
1816 if (orig_resid == 0)
1817 return (0);
1818 KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__));
1819 fs = ip->i_e2fs;
1820 if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize)
1821 return (EOVERFLOW);
1822
1823 while (uio->uio_resid > 0) {
1824 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
1825 break;
1826 lbn = lblkno(fs, uio->uio_offset);
1827 size = blksize(fs, ip, lbn);
1828 blkoffset = blkoff(fs, uio->uio_offset);
1829
1830 xfersize = fs->e2fs_fsize - blkoffset;
1831 xfersize = MIN(xfersize, uio->uio_resid);
1832 xfersize = MIN(xfersize, bytesinfile);
1833
1834 /* get block from ext4 extent cache */
1835 cache_type = ext4_ext_in_cache(ip, lbn, &nex);
1836 switch (cache_type) {
1837 case EXT4_EXT_CACHE_NO:
1838 ext4_ext_find_extent(fs, ip, lbn, &path);
1839 ep = path.ep_ext;
1840 if (ep == NULL)
1841 return (EIO);
1842
1843 ext4_ext_put_cache(ip, ep, EXT4_EXT_CACHE_IN);
1844
1845 newblk = lbn - ep->e_blk + (ep->e_start_lo |
1846 (daddr_t)ep->e_start_hi << 32);
1847
1848 if (path.ep_bp != NULL) {
1849 brelse(path.ep_bp);
1850 path.ep_bp = NULL;
1851 }
1852 break;
1853
1854 case EXT4_EXT_CACHE_GAP:
1855 /* block has not been allocated yet */
1856 return (0);
1857
1858 case EXT4_EXT_CACHE_IN:
1859 newblk = lbn - nex.e_blk + (nex.e_start_lo |
1860 (daddr_t)nex.e_start_hi << 32);
1861 break;
1862
1863 default:
1864 panic("%s: invalid cache type", __func__);
1865 }
1866
1867 error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp);
1868 if (error) {
1869 brelse(bp);
1870 return (error);
1871 }
1872
1873 size -= bp->b_resid;
1874 if (size < xfersize) {
1875 if (size == 0) {
1876 bqrelse(bp);
1877 break;
1878 }
1879 xfersize = size;
1880 }
1881 error = uiomove(bp->b_data + blkoffset, (int)xfersize, uio);
1882 bqrelse(bp);
1883 if (error)
1884 return (error);
1885 }
1886
1887 return (0);
1888 }
1889
1890 /*
1891 * Vnode op for writing.
1892 */
1893 static int
1894 ext2_write(struct vop_write_args *ap)
1895 {
1896 struct vnode *vp;
1897 struct uio *uio;
1898 struct inode *ip;
1899 struct m_ext2fs *fs;
1900 struct buf *bp;
1901 daddr_t lbn;
1902 off_t osize;
1903 int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
1904
1905 ioflag = ap->a_ioflag;
1906 uio = ap->a_uio;
1907 vp = ap->a_vp;
1908
1909 seqcount = ioflag >> IO_SEQSHIFT;
1910 ip = VTOI(vp);
1911
1912 #ifdef INVARIANTS
1913 if (uio->uio_rw != UIO_WRITE)
1914 panic("%s: mode", "ext2_write");
1915 #endif
1916
1917 switch (vp->v_type) {
1918 case VREG:
1919 if (ioflag & IO_APPEND)
1920 uio->uio_offset = ip->i_size;
1921 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
1922 return (EPERM);
1923 /* FALLTHROUGH */
1924 case VLNK:
1925 break;
1926 case VDIR:
1927 /* XXX differs from ffs -- this is called from ext2_mkdir(). */
1928 if ((ioflag & IO_SYNC) == 0)
1929 panic("ext2_write: nonsync dir write");
1930 break;
1931 default:
1932 panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
1933 vp->v_type, (intmax_t)uio->uio_offset,
1934 (intmax_t)uio->uio_resid);
1935 }
1936
1937 KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
1938 KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
1939 fs = ip->i_e2fs;
1940 if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
1941 return (EFBIG);
1942 /*
1943 * Maybe this should be above the vnode op call, but so long as
1944 * file servers have no limits, I don't think it matters.
1945 */
1946 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
1947 return (EFBIG);
1948
1949 resid = uio->uio_resid;
1950 osize = ip->i_size;
1951 if (seqcount > BA_SEQMAX)
1952 flags = BA_SEQMAX << BA_SEQSHIFT;
1953 else
1954 flags = seqcount << BA_SEQSHIFT;
1955 if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1956 flags |= IO_SYNC;
1957
1958 for (error = 0; uio->uio_resid > 0;) {
1959 lbn = lblkno(fs, uio->uio_offset);
1960 blkoffset = blkoff(fs, uio->uio_offset);
1961 xfersize = fs->e2fs_fsize - blkoffset;
1962 if (uio->uio_resid < xfersize)
1963 xfersize = uio->uio_resid;
1964 if (uio->uio_offset + xfersize > ip->i_size)
1965 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
1966
1967 /*
1968 * We must perform a read-before-write if the transfer size
1969 * does not cover the entire buffer.
1970 */
1971 if (fs->e2fs_bsize > xfersize)
1972 flags |= BA_CLRBUF;
1973 else
1974 flags &= ~BA_CLRBUF;
1975 error = ext2_balloc(ip, lbn, blkoffset + xfersize,
1976 ap->a_cred, &bp, flags);
1977 if (error != 0)
1978 break;
1979
1980 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
1981 bp->b_flags |= B_NOCACHE;
1982 if (uio->uio_offset + xfersize > ip->i_size)
1983 ip->i_size = uio->uio_offset + xfersize;
1984 size = blksize(fs, ip, lbn) - bp->b_resid;
1985 if (size < xfersize)
1986 xfersize = size;
1987
1988 error =
1989 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1990 /*
1991 * If the buffer is not already filled and we encounter an
1992 * error while trying to fill it, we have to clear out any
1993 * garbage data from the pages instantiated for the buffer.
1994 * If we do not, a failed uiomove() during a write can leave
1995 * the prior contents of the pages exposed to a userland mmap.
1996 *
1997 * Note that we need only clear buffers with a transfer size
1998 * equal to the block size because buffers with a shorter
1999 * transfer size were cleared above by the call to ext2_balloc()
2000 * with the BA_CLRBUF flag set.
2001 *
2002 * If the source region for uiomove identically mmaps the
2003 * buffer, uiomove() performed the NOP copy, and the buffer
2004 * content remains valid because the page fault handler
2005 * validated the pages.
2006 */
2007 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
2008 fs->e2fs_bsize == xfersize)
2009 vfs_bio_clrbuf(bp);
2010 if (ioflag & (IO_VMIO|IO_DIRECT)) {
2011 bp->b_flags |= B_RELBUF;
2012 }
2013
2014 /*
2015 * If IO_SYNC each buffer is written synchronously. Otherwise
2016 * if we have a severe page deficiency write the buffer
2017 * asynchronously. Otherwise try to cluster, and if that
2018 * doesn't do it then either do an async write (if O_DIRECT),
2019 * or a delayed write (if not).
2020 */
2021 if (ioflag & IO_SYNC) {
2022 (void)bwrite(bp);
2023 } else if (vm_page_count_severe() ||
2024 buf_dirty_count_severe() ||
2025 (ioflag & IO_ASYNC)) {
2026 bp->b_flags |= B_CLUSTEROK;
2027 bawrite(bp);
2028 } else if (xfersize + blkoffset == fs->e2fs_fsize) {
2029 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
2030 bp->b_flags |= B_CLUSTEROK;
2031 cluster_write(vp, bp, ip->i_size, seqcount);
2032 } else {
2033 bawrite(bp);
2034 }
2035 } else if (ioflag & IO_DIRECT) {
2036 bp->b_flags |= B_CLUSTEROK;
2037 bawrite(bp);
2038 } else {
2039 bp->b_flags |= B_CLUSTEROK;
2040 bdwrite(bp);
2041 }
2042 if (error || xfersize == 0)
2043 break;
2044 }
2045 /*
2046 * If we successfully wrote any data, and we are not the superuser
2047 * we clear the setuid and setgid bits as a precaution against
2048 * tampering.
2049 */
2050 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
2051 ap->a_cred) {
2052 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
2053 ip->i_mode &= ~(ISUID | ISGID);
2054 }
2055 if (error) {
2056 if (ioflag & IO_UNIT) {
2057 (void)ext2_truncate(vp, osize,
2058 ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
2059 uio->uio_offset -= resid - uio->uio_resid;
2060 uio->uio_resid = resid;
2061 }
2062 }
2063 if (uio->uio_resid != resid) {
2064 ip->i_flag |= IN_CHANGE | IN_UPDATE;
2065 if (ioflag & IO_SYNC)
2066 error = ext2_update(vp, 1);
2067 }
2068 return (error);
2069 }
Cache object: 53381b4e501e47751d8f350f416b5369
|