47 */
48
49#include "opt_suiddir.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/resourcevar.h>
54#include <sys/kernel.h>
55#include <sys/fcntl.h>
56#include <sys/stat.h>
57#include <sys/bio.h>
58#include <sys/buf.h>
59#include <sys/proc.h>
60#include <sys/mount.h>
61#include <sys/unistd.h>
62#include <sys/time.h>
63#include <sys/vnode.h>
64#include <sys/namei.h>
65#include <sys/lockf.h>
66#include <sys/event.h>
67#include <sys/conf.h>
68#include <sys/file.h>
69
70#include <vm/vm.h>
71#include <vm/vm_extern.h>
72#include <vm/vnode_pager.h>
73
74#include <fs/fifofs/fifo.h>
75
76#include <sys/signalvar.h>
77#include <ufs/ufs/dir.h>
78
79#include <gnu/ext2fs/inode.h>
80#include <gnu/ext2fs/ext2_mount.h>
81#include <gnu/ext2fs/ext2_fs_sb.h>
82#include <gnu/ext2fs/fs.h>
83#include <gnu/ext2fs/ext2_extern.h>
84#include <gnu/ext2fs/ext2_fs.h>
85
86static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
87
88static int ext2_access(struct vop_access_args *);
89static int ext2_advlock(struct vop_advlock_args *);
90static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
91static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
92 struct thread *);
93static int ext2_close(struct vop_close_args *);
94static int ext2_create(struct vop_create_args *);
95static int ext2_fsync(struct vop_fsync_args *);
96static int ext2_getattr(struct vop_getattr_args *);
97static int ext2_kqfilter(struct vop_kqfilter_args *ap);
98static int ext2_link(struct vop_link_args *);
99static int ext2_mkdir(struct vop_mkdir_args *);
100static int ext2_mknod(struct vop_mknod_args *);
101static int ext2_open(struct vop_open_args *);
102static int ext2_pathconf(struct vop_pathconf_args *);
103static int ext2_print(struct vop_print_args *);
104static int ext2_read(struct vop_read_args *);
105static int ext2_readlink(struct vop_readlink_args *);
106static int ext2_remove(struct vop_remove_args *);
107static int ext2_rename(struct vop_rename_args *);
108static int ext2_rmdir(struct vop_rmdir_args *);
109static int ext2_setattr(struct vop_setattr_args *);
110static int ext2_strategy(struct vop_strategy_args *);
111static int ext2_symlink(struct vop_symlink_args *);
112static int ext2_write(struct vop_write_args *);
113static int ext2fifo_close(struct vop_close_args *);
114static int ext2fifo_kqfilter(struct vop_kqfilter_args *);
115static int ext2fifo_read(struct vop_read_args *);
116static int ext2fifo_write(struct vop_write_args *);
117static int ext2spec_close(struct vop_close_args *);
118static int ext2spec_read(struct vop_read_args *);
119static int ext2spec_write(struct vop_write_args *);
120static int filt_ext2read(struct knote *kn, long hint);
121static int filt_ext2write(struct knote *kn, long hint);
122static int filt_ext2vnode(struct knote *kn, long hint);
123static void filt_ext2detach(struct knote *kn);
124
125/* Global vfs data structures for ext2. */
126vop_t **ext2_vnodeop_p;
127static struct vnodeopv_entry_desc ext2_vnodeop_entries[] = {
128 { &vop_default_desc, (vop_t *) vop_defaultop },
129 { &vop_access_desc, (vop_t *) ext2_access },
130 { &vop_advlock_desc, (vop_t *) ext2_advlock },
131 { &vop_bmap_desc, (vop_t *) ext2_bmap },
132 { &vop_cachedlookup_desc, (vop_t *) ext2_lookup },
133 { &vop_close_desc, (vop_t *) ext2_close },
134 { &vop_create_desc, (vop_t *) ext2_create },
135 { &vop_fsync_desc, (vop_t *) ext2_fsync },
136 { &vop_getattr_desc, (vop_t *) ext2_getattr },
137 { &vop_getwritemount_desc, (vop_t *) vop_stdgetwritemount },
138 { &vop_inactive_desc, (vop_t *) ext2_inactive },
139 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
140 { &vop_link_desc, (vop_t *) ext2_link },
141 { &vop_lock_desc, (vop_t *) vop_stdlock },
142 { &vop_lookup_desc, (vop_t *) vfs_cache_lookup },
143 { &vop_mkdir_desc, (vop_t *) ext2_mkdir },
144 { &vop_mknod_desc, (vop_t *) ext2_mknod },
145 { &vop_open_desc, (vop_t *) ext2_open },
146 { &vop_pathconf_desc, (vop_t *) ext2_pathconf },
147 { &vop_poll_desc, (vop_t *) vop_stdpoll },
148 { &vop_kqfilter_desc, (vop_t *) ext2_kqfilter },
149 { &vop_print_desc, (vop_t *) ext2_print },
150 { &vop_read_desc, (vop_t *) ext2_read },
151 { &vop_readdir_desc, (vop_t *) ext2_readdir },
152 { &vop_readlink_desc, (vop_t *) ext2_readlink },
153 { &vop_reallocblks_desc, (vop_t *) ext2_reallocblks },
154 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
155 { &vop_remove_desc, (vop_t *) ext2_remove },
156 { &vop_rename_desc, (vop_t *) ext2_rename },
157 { &vop_rmdir_desc, (vop_t *) ext2_rmdir },
158 { &vop_setattr_desc, (vop_t *) ext2_setattr },
159 { &vop_strategy_desc, (vop_t *) ext2_strategy },
160 { &vop_symlink_desc, (vop_t *) ext2_symlink },
161 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
162 { &vop_write_desc, (vop_t *) ext2_write },
163 { NULL, NULL }
164};
165static struct vnodeopv_desc ext2fs_vnodeop_opv_desc =
166 { &ext2_vnodeop_p, ext2_vnodeop_entries };
167
168vop_t **ext2_specop_p;
169static struct vnodeopv_entry_desc ext2_specop_entries[] = {
170 { &vop_default_desc, (vop_t *) spec_vnoperate },
171 { &vop_access_desc, (vop_t *) ext2_access },
172 { &vop_close_desc, (vop_t *) ext2spec_close },
173 { &vop_fsync_desc, (vop_t *) ext2_fsync },
174 { &vop_getattr_desc, (vop_t *) ext2_getattr },
175 { &vop_inactive_desc, (vop_t *) ext2_inactive },
176 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
177 { &vop_lock_desc, (vop_t *) vop_stdlock },
178 { &vop_print_desc, (vop_t *) ext2_print },
179 { &vop_read_desc, (vop_t *) ext2spec_read },
180 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
181 { &vop_setattr_desc, (vop_t *) ext2_setattr },
182 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
183 { &vop_write_desc, (vop_t *) ext2spec_write },
184 { NULL, NULL }
185};
186static struct vnodeopv_desc ext2fs_specop_opv_desc =
187 { &ext2_specop_p, ext2_specop_entries };
188
189vop_t **ext2_fifoop_p;
190static struct vnodeopv_entry_desc ext2_fifoop_entries[] = {
191 { &vop_default_desc, (vop_t *) fifo_vnoperate },
192 { &vop_access_desc, (vop_t *) ext2_access },
193 { &vop_close_desc, (vop_t *) ext2fifo_close },
194 { &vop_fsync_desc, (vop_t *) ext2_fsync },
195 { &vop_getattr_desc, (vop_t *) ext2_getattr },
196 { &vop_inactive_desc, (vop_t *) ext2_inactive },
197 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
198 { &vop_kqfilter_desc, (vop_t *) ext2fifo_kqfilter },
199 { &vop_lock_desc, (vop_t *) vop_stdlock },
200 { &vop_print_desc, (vop_t *) ext2_print },
201 { &vop_read_desc, (vop_t *) ext2fifo_read },
202 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
203 { &vop_setattr_desc, (vop_t *) ext2_setattr },
204 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
205 { &vop_write_desc, (vop_t *) ext2fifo_write },
206 { NULL, NULL }
207};
208static struct vnodeopv_desc ext2fs_fifoop_opv_desc =
209 { &ext2_fifoop_p, ext2_fifoop_entries };
210
211 VNODEOP_SET(ext2fs_vnodeop_opv_desc);
212 VNODEOP_SET(ext2fs_specop_opv_desc);
213 VNODEOP_SET(ext2fs_fifoop_opv_desc);
214
215#include <gnu/ext2fs/ext2_readwrite.c>
216
217union _qcvt {
218 int64_t qcvt;
219 int32_t val[2];
220};
221#define SETHIGH(q, h) { \
222 union _qcvt tmp; \
223 tmp.qcvt = (q); \
224 tmp.val[_QUAD_HIGHWORD] = (h); \
225 (q) = tmp.qcvt; \
226}
227#define SETLOW(q, l) { \
228 union _qcvt tmp; \
229 tmp.qcvt = (q); \
230 tmp.val[_QUAD_LOWWORD] = (l); \
231 (q) = tmp.qcvt; \
232}
233
234/*
235 * A virgin directory (no blushing please).
236 * Note that the type and namlen fields are reversed relative to ext2.
237 * Also, we don't use `struct odirtemplate', since it would just cause
238 * endianness problems.
239 */
240static struct dirtemplate mastertemplate = {
241 0, 12, 1, EXT2_FT_DIR, ".",
242 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
243};
244static struct dirtemplate omastertemplate = {
245 0, 12, 1, EXT2_FT_UNKNOWN, ".",
246 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
247};
248
249void
250ext2_itimes(vp)
251 struct vnode *vp;
252{
253 struct inode *ip;
254 struct timespec ts;
255
256 ip = VTOI(vp);
257 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
258 return;
259 if ((vp->v_type == VBLK || vp->v_type == VCHR))
260 ip->i_flag |= IN_LAZYMOD;
261 else
262 ip->i_flag |= IN_MODIFIED;
263 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
264 vfs_timestamp(&ts);
265 if (ip->i_flag & IN_ACCESS) {
266 ip->i_atime = ts.tv_sec;
267 ip->i_atimensec = ts.tv_nsec;
268 }
269 if (ip->i_flag & IN_UPDATE) {
270 ip->i_mtime = ts.tv_sec;
271 ip->i_mtimensec = ts.tv_nsec;
272 ip->i_modrev++;
273 }
274 if (ip->i_flag & IN_CHANGE) {
275 ip->i_ctime = ts.tv_sec;
276 ip->i_ctimensec = ts.tv_nsec;
277 }
278 }
279 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
280}
281
282/*
283 * Create a regular file
284 */
285static int
286ext2_create(ap)
287 struct vop_create_args /* {
288 struct vnode *a_dvp;
289 struct vnode **a_vpp;
290 struct componentname *a_cnp;
291 struct vattr *a_vap;
292 } */ *ap;
293{
294 int error;
295
296 error =
297 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
298 ap->a_dvp, ap->a_vpp, ap->a_cnp);
299 if (error)
300 return (error);
301 return (0);
302}
303
304/*
305 * Open called.
306 *
307 * Nothing to do.
308 */
309int
310ext2_open(ap)
311 struct vop_open_args /* {
312 struct vnode *a_vp;
313 int a_mode;
314 struct ucred *a_cred;
315 struct thread *a_td;
316 } */ *ap;
317{
318
319 /*
320 * Files marked append-only must be opened for appending.
321 */
322 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
323 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
324 return (EPERM);
325 return (0);
326}
327
328/*
329 * Close called.
330 *
331 * Update the times on the inode.
332 */
333static int
334ext2_close(ap)
335 struct vop_close_args /* {
336 struct vnode *a_vp;
337 int a_fflag;
338 struct ucred *a_cred;
339 struct thread *a_td;
340 } */ *ap;
341{
342 struct vnode *vp = ap->a_vp;
343 struct mount *mp;
344
345 mtx_lock(&vp->v_interlock);
346 if (vp->v_usecount > 1) {
347 ext2_itimes(vp);
348 mtx_unlock(&vp->v_interlock);
349 } else {
350 mtx_unlock(&vp->v_interlock);
351 /*
352 * If we are closing the last reference to an unlinked
353 * file, then it will be freed by the inactive routine.
354 * Because the freeing causes a the filesystem to be
355 * modified, it must be held up during periods when the
356 * filesystem is suspended.
357 *
358 * XXX - EAGAIN is returned to prevent vn_close from
359 * repeating the vrele operation.
360 */
361 if (vp->v_type == VREG && VTOI(vp)->i_nlink == 0) {
362 (void) vn_start_write(vp, &mp, V_WAIT);
363 vrele(vp);
364 vn_finished_write(mp);
365 return (EAGAIN);
366 }
367 }
368 return (0);
369}
370
371static int
372ext2_access(ap)
373 struct vop_access_args /* {
374 struct vnode *a_vp;
375 int a_mode;
376 struct ucred *a_cred;
377 struct thread *a_td;
378 } */ *ap;
379{
380 struct vnode *vp = ap->a_vp;
381 struct inode *ip = VTOI(vp);
382 mode_t mode = ap->a_mode;
383 int error;
384
385 /*
386 * Disallow write attempts on read-only file systems;
387 * unless the file is a socket, fifo, or a block or
388 * character device resident on the file system.
389 */
390 if (mode & VWRITE) {
391 switch (vp->v_type) {
392 case VDIR:
393 case VLNK:
394 case VREG:
395 if (vp->v_mount->mnt_flag & MNT_RDONLY)
396 return (EROFS);
397 break;
398 default:
399 break;
400 }
401 }
402
403 /* If immutable bit set, nobody gets to write it. */
404 if ((mode & VWRITE) && (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
405 return (EPERM);
406
407 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
408 ap->a_mode, ap->a_cred, NULL);
409 return (error);
410}
411
412static int
413ext2_getattr(ap)
414 struct vop_getattr_args /* {
415 struct vnode *a_vp;
416 struct vattr *a_vap;
417 struct ucred *a_cred;
418 struct thread *a_td;
419 } */ *ap;
420{
421 struct vnode *vp = ap->a_vp;
422 struct inode *ip = VTOI(vp);
423 struct vattr *vap = ap->a_vap;
424
425 ext2_itimes(vp);
426 /*
427 * Copy from inode table
428 */
429 vap->va_fsid = dev2udev(ip->i_dev);
430 vap->va_fileid = ip->i_number;
431 vap->va_mode = ip->i_mode & ~IFMT;
432 vap->va_nlink = ip->i_nlink;
433 vap->va_uid = ip->i_uid;
434 vap->va_gid = ip->i_gid;
435 vap->va_rdev = ip->i_rdev;
436 vap->va_size = ip->i_size;
437 vap->va_atime.tv_sec = ip->i_atime;
438 vap->va_atime.tv_nsec = ip->i_atimensec;
439 vap->va_mtime.tv_sec = ip->i_mtime;
440 vap->va_mtime.tv_nsec = ip->i_mtimensec;
441 vap->va_ctime.tv_sec = ip->i_ctime;
442 vap->va_ctime.tv_nsec = ip->i_ctimensec;
443 vap->va_flags = ip->i_flags;
444 vap->va_gen = ip->i_gen;
445 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
446 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
447 vap->va_type = IFTOVT(ip->i_mode);
448 vap->va_filerev = ip->i_modrev;
449 return (0);
450}
451
452/*
453 * Set attribute vnode op. called from several syscalls
454 */
455int
456ext2_setattr(ap)
457 struct vop_setattr_args /* {
458 struct vnode *a_vp;
459 struct vattr *a_vap;
460 struct ucred *a_cred;
461 struct thread *a_td;
462 } */ *ap;
463{
464 struct vattr *vap = ap->a_vap;
465 struct vnode *vp = ap->a_vp;
466 struct inode *ip = VTOI(vp);
467 struct ucred *cred = ap->a_cred;
468 struct thread *td = ap->a_td;
469 int error;
470
471 /*
472 * Check for unsettable attributes.
473 */
474 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
475 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
476 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
477 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
478 return (EINVAL);
479 }
480 if (vap->va_flags != VNOVAL) {
481 if (vp->v_mount->mnt_flag & MNT_RDONLY)
482 return (EROFS);
483 /*
484 * Callers may only modify the file flags on objects they
485 * have VADMIN rights for.
486 */
487 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
488 return (error);
489 /*
490 * Unprivileged processes and privileged processes in
491 * jail() are not permitted to unset system flags, or
492 * modify flags if any system flags are set.
493 * Privileged non-jail processes may not modify system flags
494 * if securelevel > 0 and any existing system flags are set.
495 */
496 if (!suser_cred(cred, PRISON_ROOT)) {
497 if (ip->i_flags
498 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
499 error = securelevel_gt(cred, 0);
500 if (error)
501 return (error);
502 }
503 ip->i_flags = vap->va_flags;
504 } else {
505 if (ip->i_flags
506 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
507 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
508 return (EPERM);
509 ip->i_flags &= SF_SETTABLE;
510 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
511 }
512 ip->i_flag |= IN_CHANGE;
513 if (vap->va_flags & (IMMUTABLE | APPEND))
514 return (0);
515 }
516 if (ip->i_flags & (IMMUTABLE | APPEND))
517 return (EPERM);
518 /*
519 * Go through the fields and update iff not VNOVAL.
520 */
521 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
522 if (vp->v_mount->mnt_flag & MNT_RDONLY)
523 return (EROFS);
524 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
525 td)) != 0)
526 return (error);
527 }
528 if (vap->va_size != VNOVAL) {
529 /*
530 * Disallow write attempts on read-only file systems;
531 * unless the file is a socket, fifo, or a block or
532 * character device resident on the file system.
533 */
534 switch (vp->v_type) {
535 case VDIR:
536 return (EISDIR);
537 case VLNK:
538 case VREG:
539 if (vp->v_mount->mnt_flag & MNT_RDONLY)
540 return (EROFS);
541 break;
542 default:
543 break;
544 }
545 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
546 return (error);
547 }
548 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
549 if (vp->v_mount->mnt_flag & MNT_RDONLY)
550 return (EROFS);
551 /*
552 * From utimes(2):
553 * If times is NULL, ... The caller must be the owner of
554 * the file, have permission to write the file, or be the
555 * super-user.
556 * If times is non-NULL, ... The caller must be the owner of
557 * the file or be the super-user.
558 */
559 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
560 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
561 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
562 return (error);
563 if (vap->va_atime.tv_sec != VNOVAL)
564 ip->i_flag |= IN_ACCESS;
565 if (vap->va_mtime.tv_sec != VNOVAL)
566 ip->i_flag |= IN_CHANGE | IN_UPDATE;
567 ext2_itimes(vp);
568 if (vap->va_atime.tv_sec != VNOVAL) {
569 ip->i_atime = vap->va_atime.tv_sec;
570 ip->i_atimensec = vap->va_atime.tv_nsec;
571 }
572 if (vap->va_mtime.tv_sec != VNOVAL) {
573 ip->i_mtime = vap->va_mtime.tv_sec;
574 ip->i_mtimensec = vap->va_mtime.tv_nsec;
575 }
576 error = ext2_update(vp, 0);
577 if (error)
578 return (error);
579 }
580 error = 0;
581 if (vap->va_mode != (mode_t)VNOVAL) {
582 if (vp->v_mount->mnt_flag & MNT_RDONLY)
583 return (EROFS);
584 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
585 }
586 VN_KNOTE(vp, NOTE_ATTRIB);
587 return (error);
588}
589
590/*
591 * Change the mode on a file.
592 * Inode must be locked before calling.
593 */
594static int
595ext2_chmod(vp, mode, cred, td)
596 struct vnode *vp;
597 int mode;
598 struct ucred *cred;
599 struct thread *td;
600{
601 struct inode *ip = VTOI(vp);
602 int error;
603
604 /*
605 * To modify the permissions on a file, must possess VADMIN
606 * for that file.
607 */
608 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
609 return (error);
610 /*
611 * Privileged processes may set the sticky bit on non-directories,
612 * as well as set the setgid bit on a file with a group that the
613 * process is not a member of.
614 */
615 if (suser_cred(cred, PRISON_ROOT)) {
616 if (vp->v_type != VDIR && (mode & S_ISTXT))
617 return (EFTYPE);
618 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
619 return (EPERM);
620 }
621 ip->i_mode &= ~ALLPERMS;
622 ip->i_mode |= (mode & ALLPERMS);
623 ip->i_flag |= IN_CHANGE;
624 return (0);
625}
626
627/*
628 * Perform chown operation on inode ip;
629 * inode must be locked prior to call.
630 */
631static int
632ext2_chown(vp, uid, gid, cred, td)
633 struct vnode *vp;
634 uid_t uid;
635 gid_t gid;
636 struct ucred *cred;
637 struct thread *td;
638{
639 struct inode *ip = VTOI(vp);
640 uid_t ouid;
641 gid_t ogid;
642 int error = 0;
643
644 if (uid == (uid_t)VNOVAL)
645 uid = ip->i_uid;
646 if (gid == (gid_t)VNOVAL)
647 gid = ip->i_gid;
648 /*
649 * To modify the ownership of a file, must possess VADMIN
650 * for that file.
651 */
652 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
653 return (error);
654 /*
655 * To change the owner of a file, or change the group of a file
656 * to a group of which we are not a member, the caller must
657 * have privilege.
658 */
659 if ((uid != ip->i_uid ||
660 (gid != ip->i_gid && !groupmember(gid, cred))) &&
661 (error = suser_cred(cred, PRISON_ROOT)))
662 return (error);
663 ogid = ip->i_gid;
664 ouid = ip->i_uid;
665 ip->i_gid = gid;
666 ip->i_uid = uid;
667 ip->i_flag |= IN_CHANGE;
668 if (suser_cred(cred, PRISON_ROOT) && (ouid != uid || ogid != gid))
669 ip->i_mode &= ~(ISUID | ISGID);
670 return (0);
671}
672
673/*
674 * Synch an open file.
675 */
676/* ARGSUSED */
677static int
678ext2_fsync(ap)
679 struct vop_fsync_args /* {
680 struct vnode *a_vp;
681 struct ucred *a_cred;
682 int a_waitfor;
683 struct thread *a_td;
684 } */ *ap;
685{
686 struct vnode *vp = ap->a_vp;
687 struct buf *bp;
688 struct buf *nbp;
689 int s;
690
691 /*
692 * XXX why is all this fs specific?
693 */
694
695 /*
696 * Flush all dirty buffers associated with a vnode.
697 */
698 ext2_discard_prealloc(VTOI(vp));
699
700loop:
701 s = splbio();
702 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
703 nbp = TAILQ_NEXT(bp, b_vnbufs);
704 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
705 continue;
706 if ((bp->b_flags & B_DELWRI) == 0)
707 panic("ext2_fsync: not dirty");
708 bremfree(bp);
709 splx(s);
710 /*
711 * Wait for I/O associated with indirect blocks to complete,
712 * since there is no way to quickly wait for them below.
713 */
714 if (bp->b_vp == vp || ap->a_waitfor == MNT_NOWAIT)
715 (void) bawrite(bp);
716 else
717 (void) bwrite(bp);
718 goto loop;
719 }
720 if (ap->a_waitfor == MNT_WAIT) {
721 while (vp->v_numoutput) {
722 vp->v_flag |= VBWAIT;
723 tsleep(&vp->v_numoutput, PRIBIO + 1, "e2fsyn", 0);
724 }
725#if DIAGNOSTIC
726 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
727 vprint("ext2_fsync: dirty", vp);
728 goto loop;
729 }
730#endif
731 }
732 splx(s);
733 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
734}
735
736/*
737 * Mknod vnode call
738 */
739/* ARGSUSED */
740static int
741ext2_mknod(ap)
742 struct vop_mknod_args /* {
743 struct vnode *a_dvp;
744 struct vnode **a_vpp;
745 struct componentname *a_cnp;
746 struct vattr *a_vap;
747 } */ *ap;
748{
749 struct vattr *vap = ap->a_vap;
750 struct vnode **vpp = ap->a_vpp;
751 struct inode *ip;
752 ino_t ino;
753 int error;
754
755 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
756 ap->a_dvp, vpp, ap->a_cnp);
757 if (error)
758 return (error);
759 ip = VTOI(*vpp);
760 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
761 if (vap->va_rdev != VNOVAL) {
762 /*
763 * Want to be able to use this to make badblock
764 * inodes, so don't truncate the dev number.
765 */
766 ip->i_rdev = vap->va_rdev;
767 }
768 /*
769 * Remove inode, then reload it through VFS_VGET so it is
770 * checked to see if it is an alias of an existing entry in
771 * the inode cache.
772 */
773 vput(*vpp);
774 (*vpp)->v_type = VNON;
775 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
776 vgone(*vpp);
777 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
778 if (error) {
779 *vpp = NULL;
780 return (error);
781 }
782 return (0);
783}
784
785static int
786ext2_remove(ap)
787 struct vop_remove_args /* {
788 struct vnode *a_dvp;
789 struct vnode *a_vp;
790 struct componentname *a_cnp;
791 } */ *ap;
792{
793 struct inode *ip;
794 struct vnode *vp = ap->a_vp;
795 struct vnode *dvp = ap->a_dvp;
796 int error;
797
798 ip = VTOI(vp);
799 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
800 (VTOI(dvp)->i_flags & APPEND)) {
801 error = EPERM;
802 goto out;
803 }
804 error = ext2_dirremove(dvp, ap->a_cnp);
805 if (error == 0) {
806 ip->i_nlink--;
807 ip->i_flag |= IN_CHANGE;
808 }
809out:
810 return (error);
811}
812
813/*
814 * link vnode call
815 */
816static int
817ext2_link(ap)
818 struct vop_link_args /* {
819 struct vnode *a_tdvp;
820 struct vnode *a_vp;
821 struct componentname *a_cnp;
822 } */ *ap;
823{
824 struct vnode *vp = ap->a_vp;
825 struct vnode *tdvp = ap->a_tdvp;
826 struct componentname *cnp = ap->a_cnp;
827 struct thread *td = cnp->cn_thread;
828 struct inode *ip;
829 int error;
830
831#ifdef DIAGNOSTIC
832 if ((cnp->cn_flags & HASBUF) == 0)
833 panic("ext2_link: no name");
834#endif
835 if (tdvp->v_mount != vp->v_mount) {
836 error = EXDEV;
837 goto out2;
838 }
839 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE, td))) {
840 goto out2;
841 }
842 ip = VTOI(vp);
843 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
844 error = EMLINK;
845 goto out1;
846 }
847 if (ip->i_flags & (IMMUTABLE | APPEND)) {
848 error = EPERM;
849 goto out1;
850 }
851 ip->i_nlink++;
852 ip->i_flag |= IN_CHANGE;
853 error = ext2_update(vp, 1);
854 if (!error)
855 error = ext2_direnter(ip, tdvp, cnp);
856 if (error) {
857 ip->i_nlink--;
858 ip->i_flag |= IN_CHANGE;
859 }
860out1:
861 if (tdvp != vp)
862 VOP_UNLOCK(vp, 0, td);
863out2:
864 return (error);
865}
866
867/*
868 * Rename system call.
869 * See comments in sys/ufs/ufs/ufs_vnops.c
870 */
871static int
872ext2_rename(ap)
873 struct vop_rename_args /* {
874 struct vnode *a_fdvp;
875 struct vnode *a_fvp;
876 struct componentname *a_fcnp;
877 struct vnode *a_tdvp;
878 struct vnode *a_tvp;
879 struct componentname *a_tcnp;
880 } */ *ap;
881{
882 struct vnode *tvp = ap->a_tvp;
883 struct vnode *tdvp = ap->a_tdvp;
884 struct vnode *fvp = ap->a_fvp;
885 struct vnode *fdvp = ap->a_fdvp;
886 struct componentname *tcnp = ap->a_tcnp;
887 struct componentname *fcnp = ap->a_fcnp;
888 struct thread *td = fcnp->cn_thread;
889 struct inode *ip, *xp, *dp;
890 struct dirtemplate dirbuf;
891 int doingdirectory = 0, oldparent = 0, newparent = 0;
892 int error = 0;
893 u_char namlen;
894
895#ifdef DIAGNOSTIC
896 if ((tcnp->cn_flags & HASBUF) == 0 ||
897 (fcnp->cn_flags & HASBUF) == 0)
898 panic("ext2_rename: no name");
899#endif
900 /*
901 * Check for cross-device rename.
902 */
903 if ((fvp->v_mount != tdvp->v_mount) ||
904 (tvp && (fvp->v_mount != tvp->v_mount))) {
905 error = EXDEV;
906abortit:
907 if (tdvp == tvp)
908 vrele(tdvp);
909 else
910 vput(tdvp);
911 if (tvp)
912 vput(tvp);
913 vrele(fdvp);
914 vrele(fvp);
915 return (error);
916 }
917
918 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
919 (VTOI(tdvp)->i_flags & APPEND))) {
920 error = EPERM;
921 goto abortit;
922 }
923
924 /*
925 * Check if just deleting a link name or if we've lost a race.
926 * If another process completes the same rename after we've looked
927 * up the source and have blocked looking up the target, then the
928 * source and target inodes may be identical now although the
929 * names were never linked.
930 */
931 if (fvp == tvp) {
932 if (fvp->v_type == VDIR) {
933 /*
934 * Linked directories are impossible, so we must
935 * have lost the race. Pretend that the rename
936 * completed before the lookup.
937 */
938#ifdef UFS_RENAME_DEBUG
939 printf("ext2_rename: fvp == tvp for directories\n");
940#endif
941 error = ENOENT;
942 goto abortit;
943 }
944
945 /* Release destination completely. */
946 vput(tdvp);
947 vput(tvp);
948
949 /*
950 * Delete source. There is another race now that everything
951 * is unlocked, but this doesn't cause any new complications.
952 * Relookup() may find a file that is unrelated to the
953 * original one, or it may fail. Too bad.
954 */
955 vrele(fdvp);
956 vrele(fvp);
957 fcnp->cn_flags &= ~MODMASK;
958 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
959 fcnp->cn_nameiop = DELETE;
960 VREF(fdvp);
961 error = relookup(fdvp, &fvp, fcnp);
962 if (error == 0)
963 vrele(fdvp);
964 if (fvp == NULL) {
965#ifdef UFS_RENAME_DEBUG
966 printf("ext2_rename: from name disappeared\n");
967#endif
968 return (ENOENT);
969 }
970 error = VOP_REMOVE(fdvp, fvp, fcnp);
971 if (fdvp == fvp)
972 vrele(fdvp);
973 else
974 vput(fdvp);
975 if (fvp != NULLVP)
976 vput(fvp);
977 return (error);
978 }
979 if ((error = vn_lock(fvp, LK_EXCLUSIVE, td)) != 0)
980 goto abortit;
981 dp = VTOI(fdvp);
982 ip = VTOI(fvp);
983 if (ip->i_nlink >= LINK_MAX) {
984 VOP_UNLOCK(fvp, 0, td);
985 error = EMLINK;
986 goto abortit;
987 }
988 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
989 || (dp->i_flags & APPEND)) {
990 VOP_UNLOCK(fvp, 0, td);
991 error = EPERM;
992 goto abortit;
993 }
994 if ((ip->i_mode & IFMT) == IFDIR) {
995 /*
996 * Avoid ".", "..", and aliases of "." for obvious reasons.
997 */
998 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
999 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
1000 (ip->i_flag & IN_RENAME)) {
1001 VOP_UNLOCK(fvp, 0, td);
1002 error = EINVAL;
1003 goto abortit;
1004 }
1005 ip->i_flag |= IN_RENAME;
1006 oldparent = dp->i_number;
1007 doingdirectory++;
1008 }
1009 vrele(fdvp);
1010
1011 /*
1012 * When the target exists, both the directory
1013 * and target vnodes are returned locked.
1014 */
1015 dp = VTOI(tdvp);
1016 xp = NULL;
1017 if (tvp)
1018 xp = VTOI(tvp);
1019
1020 /*
1021 * 1) Bump link count while we're moving stuff
1022 * around. If we crash somewhere before
1023 * completing our work, the link count
1024 * may be wrong, but correctable.
1025 */
1026 ip->i_nlink++;
1027 ip->i_flag |= IN_CHANGE;
1028 if ((error = ext2_update(fvp, 1)) != 0) {
1029 VOP_UNLOCK(fvp, 0, td);
1030 goto bad;
1031 }
1032
1033 /*
1034 * If ".." must be changed (ie the directory gets a new
1035 * parent) then the source directory must not be in the
1036 * directory heirarchy above the target, as this would
1037 * orphan everything below the source directory. Also
1038 * the user must have write permission in the source so
1039 * as to be able to change "..". We must repeat the call
1040 * to namei, as the parent directory is unlocked by the
1041 * call to checkpath().
1042 */
1043 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1044 VOP_UNLOCK(fvp, 0, td);
1045 if (oldparent != dp->i_number)
1046 newparent = dp->i_number;
1047 if (doingdirectory && newparent) {
1048 if (error) /* write access check above */
1049 goto bad;
1050 if (xp != NULL)
1051 vput(tvp);
1052 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
1053 if (error)
1054 goto out;
1055 VREF(tdvp);
1056 error = relookup(tdvp, &tvp, tcnp);
1057 if (error)
1058 goto out;
1059 vrele(tdvp);
1060 dp = VTOI(tdvp);
1061 xp = NULL;
1062 if (tvp)
1063 xp = VTOI(tvp);
1064 }
1065 /*
1066 * 2) If target doesn't exist, link the target
1067 * to the source and unlink the source.
1068 * Otherwise, rewrite the target directory
1069 * entry to reference the source inode and
1070 * expunge the original entry's existence.
1071 */
1072 if (xp == NULL) {
1073 if (dp->i_dev != ip->i_dev)
1074 panic("ext2_rename: EXDEV");
1075 /*
1076 * Account for ".." in new directory.
1077 * When source and destination have the same
1078 * parent we don't fool with the link count.
1079 */
1080 if (doingdirectory && newparent) {
1081 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1082 error = EMLINK;
1083 goto bad;
1084 }
1085 dp->i_nlink++;
1086 dp->i_flag |= IN_CHANGE;
1087 error = ext2_update(tdvp, 1);
1088 if (error)
1089 goto bad;
1090 }
1091 error = ext2_direnter(ip, tdvp, tcnp);
1092 if (error) {
1093 if (doingdirectory && newparent) {
1094 dp->i_nlink--;
1095 dp->i_flag |= IN_CHANGE;
1096 (void)ext2_update(tdvp, 1);
1097 }
1098 goto bad;
1099 }
1100 vput(tdvp);
1101 } else {
1102 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1103 panic("ext2_rename: EXDEV");
1104 /*
1105 * Short circuit rename(foo, foo).
1106 */
1107 if (xp->i_number == ip->i_number)
1108 panic("ext2_rename: same file");
1109 /*
1110 * If the parent directory is "sticky", then the user must
1111 * own the parent directory, or the destination of the rename,
1112 * otherwise the destination may not be changed (except by
1113 * root). This implements append-only directories.
1114 */
1115 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1116 tcnp->cn_cred->cr_uid != dp->i_uid &&
1117 xp->i_uid != tcnp->cn_cred->cr_uid) {
1118 error = EPERM;
1119 goto bad;
1120 }
1121 /*
1122 * Target must be empty if a directory and have no links
1123 * to it. Also, ensure source and target are compatible
1124 * (both directories, or both not directories).
1125 */
1126 if ((xp->i_mode&IFMT) == IFDIR) {
1127 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
1128 xp->i_nlink > 2) {
1129 error = ENOTEMPTY;
1130 goto bad;
1131 }
1132 if (!doingdirectory) {
1133 error = ENOTDIR;
1134 goto bad;
1135 }
1136 cache_purge(tdvp);
1137 } else if (doingdirectory) {
1138 error = EISDIR;
1139 goto bad;
1140 }
1141 error = ext2_dirrewrite(dp, ip, tcnp);
1142 if (error)
1143 goto bad;
1144 /*
1145 * If the target directory is in the same
1146 * directory as the source directory,
1147 * decrement the link count on the parent
1148 * of the target directory.
1149 */
1150 if (doingdirectory && !newparent) {
1151 dp->i_nlink--;
1152 dp->i_flag |= IN_CHANGE;
1153 }
1154 vput(tdvp);
1155 /*
1156 * Adjust the link count of the target to
1157 * reflect the dirrewrite above. If this is
1158 * a directory it is empty and there are
1159 * no links to it, so we can squash the inode and
1160 * any space associated with it. We disallowed
1161 * renaming over top of a directory with links to
1162 * it above, as the remaining link would point to
1163 * a directory without "." or ".." entries.
1164 */
1165 xp->i_nlink--;
1166 if (doingdirectory) {
1167 if (--xp->i_nlink != 0)
1168 panic("ext2_rename: linked directory");
1169 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1170 tcnp->cn_cred, tcnp->cn_thread);
1171 }
1172 xp->i_flag |= IN_CHANGE;
1173 vput(tvp);
1174 xp = NULL;
1175 }
1176
1177 /*
1178 * 3) Unlink the source.
1179 */
1180 fcnp->cn_flags &= ~MODMASK;
1181 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1182 VREF(fdvp);
1183 error = relookup(fdvp, &fvp, fcnp);
1184 if (error == 0)
1185 vrele(fdvp);
1186 if (fvp != NULL) {
1187 xp = VTOI(fvp);
1188 dp = VTOI(fdvp);
1189 } else {
1190 /*
1191 * From name has disappeared.
1192 */
1193 if (doingdirectory)
1194 panic("ext2_rename: lost dir entry");
1195 vrele(ap->a_fvp);
1196 return (0);
1197 }
1198 /*
1199 * Ensure that the directory entry still exists and has not
1200 * changed while the new name has been entered. If the source is
1201 * a file then the entry may have been unlinked or renamed. In
1202 * either case there is no further work to be done. If the source
1203 * is a directory then it cannot have been rmdir'ed; its link
1204 * count of three would cause a rmdir to fail with ENOTEMPTY.
1205 * The IN_RENAME flag ensures that it cannot be moved by another
1206 * rename.
1207 */
1208 if (xp != ip) {
1209 if (doingdirectory)
1210 panic("ext2_rename: lost dir entry");
1211 } else {
1212 /*
1213 * If the source is a directory with a
1214 * new parent, the link count of the old
1215 * parent directory must be decremented
1216 * and ".." set to point to the new parent.
1217 */
1218 if (doingdirectory && newparent) {
1219 dp->i_nlink--;
1220 dp->i_flag |= IN_CHANGE;
1221 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1222 sizeof (struct dirtemplate), (off_t)0,
1223 UIO_SYSSPACE, IO_NODELOCKED,
1224 tcnp->cn_cred, (int *)0, (struct thread *)0);
1225 if (error == 0) {
1226 /* Like ufs little-endian: */
1227 namlen = dirbuf.dotdot_type;
1228 if (namlen != 2 ||
1229 dirbuf.dotdot_name[0] != '.' ||
1230 dirbuf.dotdot_name[1] != '.') {
1231 ext2_dirbad(xp, (doff_t)12,
1232 "rename: mangled dir");
1233 } else {
1234 dirbuf.dotdot_ino = newparent;
1235 (void) vn_rdwr(UIO_WRITE, fvp,
1236 (caddr_t)&dirbuf,
1237 sizeof (struct dirtemplate),
1238 (off_t)0, UIO_SYSSPACE,
1239 IO_NODELOCKED|IO_SYNC,
1240 tcnp->cn_cred, (int *)0,
1241 (struct thread *)0);
1242 cache_purge(fdvp);
1243 }
1244 }
1245 }
1246 error = ext2_dirremove(fdvp, fcnp);
1247 if (!error) {
1248 xp->i_nlink--;
1249 xp->i_flag |= IN_CHANGE;
1250 }
1251 xp->i_flag &= ~IN_RENAME;
1252 }
1253 if (dp)
1254 vput(fdvp);
1255 if (xp)
1256 vput(fvp);
1257 vrele(ap->a_fvp);
1258 return (error);
1259
1260bad:
1261 if (xp)
1262 vput(ITOV(xp));
1263 vput(ITOV(dp));
1264out:
1265 if (doingdirectory)
1266 ip->i_flag &= ~IN_RENAME;
1267 if (vn_lock(fvp, LK_EXCLUSIVE, td) == 0) {
1268 ip->i_nlink--;
1269 ip->i_flag |= IN_CHANGE;
1270 ip->i_flag &= ~IN_RENAME;
1271 vput(fvp);
1272 } else
1273 vrele(fvp);
1274 return (error);
1275}
1276
1277/*
1278 * Mkdir system call
1279 */
1280static int
1281ext2_mkdir(ap)
1282 struct vop_mkdir_args /* {
1283 struct vnode *a_dvp;
1284 struct vnode **a_vpp;
1285 struct componentname *a_cnp;
1286 struct vattr *a_vap;
1287 } */ *ap;
1288{
1289 struct vnode *dvp = ap->a_dvp;
1290 struct vattr *vap = ap->a_vap;
1291 struct componentname *cnp = ap->a_cnp;
1292 struct inode *ip, *dp;
1293 struct vnode *tvp;
1294 struct dirtemplate dirtemplate, *dtp;
1295 int error, dmode;
1296
1297#ifdef DIAGNOSTIC
1298 if ((cnp->cn_flags & HASBUF) == 0)
1299 panic("ext2_mkdir: no name");
1300#endif
1301 dp = VTOI(dvp);
1302 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1303 error = EMLINK;
1304 goto out;
1305 }
1306 dmode = vap->va_mode & 0777;
1307 dmode |= IFDIR;
1308 /*
1309 * Must simulate part of ext2_makeinode here to acquire the inode,
1310 * but not have it entered in the parent directory. The entry is
1311 * made later after writing "." and ".." entries.
1312 */
1313 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1314 if (error)
1315 goto out;
1316 ip = VTOI(tvp);
1317 ip->i_gid = dp->i_gid;
1318#ifdef SUIDDIR
1319 {
1320 /*
1321 * if we are hacking owners here, (only do this where told to)
1322 * and we are not giving it TOO root, (would subvert quotas)
1323 * then go ahead and give it to the other user.
1324 * The new directory also inherits the SUID bit.
1325 * If user's UID and dir UID are the same,
1326 * 'give it away' so that the SUID is still forced on.
1327 */
1328 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1329 (dp->i_mode & ISUID) && dp->i_uid) {
1330 dmode |= ISUID;
1331 ip->i_uid = dp->i_uid;
1332 } else {
1333 ip->i_uid = cnp->cn_cred->cr_uid;
1334 }
1335 }
1336#else
1337 ip->i_uid = cnp->cn_cred->cr_uid;
1338#endif
1339 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1340 ip->i_mode = dmode;
1341 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1342 ip->i_nlink = 2;
1343 if (cnp->cn_flags & ISWHITEOUT)
1344 ip->i_flags |= UF_OPAQUE;
1345 error = ext2_update(tvp, 1);
1346
1347 /*
1348 * Bump link count in parent directory
1349 * to reflect work done below. Should
1350 * be done before reference is created
1351 * so reparation is possible if we crash.
1352 */
1353 dp->i_nlink++;
1354 dp->i_flag |= IN_CHANGE;
1355 error = ext2_update(dvp, 1);
1356 if (error)
1357 goto bad;
1358
1359 /* Initialize directory with "." and ".." from static template. */
1360 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
1361 EXT2_FEATURE_INCOMPAT_FILETYPE))
1362 dtp = &mastertemplate;
1363 else
1364 dtp = &omastertemplate;
1365 dirtemplate = *dtp;
1366 dirtemplate.dot_ino = ip->i_number;
1367 dirtemplate.dotdot_ino = dp->i_number;
1368 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1369 * so let's just redefine it - for this function only
1370 */
1371#undef DIRBLKSIZ
1372#define DIRBLKSIZ VTOI(dvp)->i_e2fs->s_blocksize
1373 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1374 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1375 sizeof (dirtemplate), (off_t)0, UIO_SYSSPACE,
1376 IO_NODELOCKED|IO_SYNC, cnp->cn_cred, (int *)0, (struct thread *)0);
1377 if (error) {
1378 dp->i_nlink--;
1379 dp->i_flag |= IN_CHANGE;
1380 goto bad;
1381 }
1382 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1383 /* XXX should grow with balloc() */
1384 panic("ext2_mkdir: blksize");
1385 else {
1386 ip->i_size = DIRBLKSIZ;
1387 ip->i_flag |= IN_CHANGE;
1388 }
1389
1390 /* Directory set up, now install its entry in the parent directory. */
1391 error = ext2_direnter(ip, dvp, cnp);
1392 if (error) {
1393 dp->i_nlink--;
1394 dp->i_flag |= IN_CHANGE;
1395 }
1396bad:
1397 /*
1398 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1399 * for us because we set the link count to 0.
1400 */
1401 if (error) {
1402 ip->i_nlink = 0;
1403 ip->i_flag |= IN_CHANGE;
1404 vput(tvp);
1405 } else
1406 *ap->a_vpp = tvp;
1407out:
1408 return (error);
1409#undef DIRBLKSIZ
1410#define DIRBLKSIZ DEV_BSIZE
1411}
1412
1413/*
1414 * Rmdir system call.
1415 */
1416static int
1417ext2_rmdir(ap)
1418 struct vop_rmdir_args /* {
1419 struct vnode *a_dvp;
1420 struct vnode *a_vp;
1421 struct componentname *a_cnp;
1422 } */ *ap;
1423{
1424 struct vnode *vp = ap->a_vp;
1425 struct vnode *dvp = ap->a_dvp;
1426 struct componentname *cnp = ap->a_cnp;
1427 struct thread *td = cnp->cn_thread;
1428 struct inode *ip, *dp;
1429 int error;
1430
1431 ip = VTOI(vp);
1432 dp = VTOI(dvp);
1433
1434 /*
1435 * Verify the directory is empty (and valid).
1436 * (Rmdir ".." won't be valid since
1437 * ".." will contain a reference to
1438 * the current directory and thus be
1439 * non-empty.)
1440 */
1441 error = 0;
1442 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1443 error = ENOTEMPTY;
1444 goto out;
1445 }
1446 if ((dp->i_flags & APPEND)
1447 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1448 error = EPERM;
1449 goto out;
1450 }
1451 /*
1452 * Delete reference to directory before purging
1453 * inode. If we crash in between, the directory
1454 * will be reattached to lost+found,
1455 */
1456 error = ext2_dirremove(dvp, cnp);
1457 if (error)
1458 goto out;
1459 dp->i_nlink--;
1460 dp->i_flag |= IN_CHANGE;
1461 cache_purge(dvp);
1462 VOP_UNLOCK(dvp, 0, td);
1463 /*
1464 * Truncate inode. The only stuff left
1465 * in the directory is "." and "..". The
1466 * "." reference is inconsequential since
1467 * we're quashing it. The ".." reference
1468 * has already been adjusted above. We've
1469 * removed the "." reference and the reference
1470 * in the parent directory, but there may be
1471 * other hard links so decrement by 2 and
1472 * worry about them later.
1473 */
1474 ip->i_nlink -= 2;
1475 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, td);
1476 cache_purge(ITOV(ip));
1477 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1478out:
1479 return (error);
1480}
1481
1482/*
1483 * symlink -- make a symbolic link
1484 */
1485static int
1486ext2_symlink(ap)
1487 struct vop_symlink_args /* {
1488 struct vnode *a_dvp;
1489 struct vnode **a_vpp;
1490 struct componentname *a_cnp;
1491 struct vattr *a_vap;
1492 char *a_target;
1493 } */ *ap;
1494{
1495 struct vnode *vp, **vpp = ap->a_vpp;
1496 struct inode *ip;
1497 int len, error;
1498
1499 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1500 vpp, ap->a_cnp);
1501 if (error)
1502 return (error);
1503 vp = *vpp;
1504 len = strlen(ap->a_target);
1505 if (len < vp->v_mount->mnt_maxsymlinklen) {
1506 ip = VTOI(vp);
1507 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1508 ip->i_size = len;
1509 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1510 } else
1511 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1512 UIO_SYSSPACE, IO_NODELOCKED, ap->a_cnp->cn_cred, (int *)0,
1513 (struct thread *)0);
1514 if (error)
1515 vput(vp);
1516 return (error);
1517}
1518
1519/*
1520 * Return target name of a symbolic link
1521 */
1522static int
1523ext2_readlink(ap)
1524 struct vop_readlink_args /* {
1525 struct vnode *a_vp;
1526 struct uio *a_uio;
1527 struct ucred *a_cred;
1528 } */ *ap;
1529{
1530 struct vnode *vp = ap->a_vp;
1531 struct inode *ip = VTOI(vp);
1532 int isize;
1533
1534 isize = ip->i_size;
1535 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1536 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1537 return (0);
1538 }
1539 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1540}
1541
1542/*
1543 * Calculate the logical to physical mapping if not done already,
1544 * then call the device strategy routine.
1545 *
1546 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1547 * deadlock on memory. See ext2_bmap() for details.
1548 */
1549int
1550ext2_strategy(ap)
1551 struct vop_strategy_args /* {
1552 struct vnode *a_vp;
1553 struct buf *a_bp;
1554 } */ *ap;
1555{
1556 struct buf *bp = ap->a_bp;
1557 struct vnode *vp = ap->a_vp;
1558 struct inode *ip;
| 47 */
48
49#include "opt_suiddir.h"
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/resourcevar.h>
54#include <sys/kernel.h>
55#include <sys/fcntl.h>
56#include <sys/stat.h>
57#include <sys/bio.h>
58#include <sys/buf.h>
59#include <sys/proc.h>
60#include <sys/mount.h>
61#include <sys/unistd.h>
62#include <sys/time.h>
63#include <sys/vnode.h>
64#include <sys/namei.h>
65#include <sys/lockf.h>
66#include <sys/event.h>
67#include <sys/conf.h>
68#include <sys/file.h>
69
70#include <vm/vm.h>
71#include <vm/vm_extern.h>
72#include <vm/vnode_pager.h>
73
74#include <fs/fifofs/fifo.h>
75
76#include <sys/signalvar.h>
77#include <ufs/ufs/dir.h>
78
79#include <gnu/ext2fs/inode.h>
80#include <gnu/ext2fs/ext2_mount.h>
81#include <gnu/ext2fs/ext2_fs_sb.h>
82#include <gnu/ext2fs/fs.h>
83#include <gnu/ext2fs/ext2_extern.h>
84#include <gnu/ext2fs/ext2_fs.h>
85
86static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *);
87
88static int ext2_access(struct vop_access_args *);
89static int ext2_advlock(struct vop_advlock_args *);
90static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *);
91static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *,
92 struct thread *);
93static int ext2_close(struct vop_close_args *);
94static int ext2_create(struct vop_create_args *);
95static int ext2_fsync(struct vop_fsync_args *);
96static int ext2_getattr(struct vop_getattr_args *);
97static int ext2_kqfilter(struct vop_kqfilter_args *ap);
98static int ext2_link(struct vop_link_args *);
99static int ext2_mkdir(struct vop_mkdir_args *);
100static int ext2_mknod(struct vop_mknod_args *);
101static int ext2_open(struct vop_open_args *);
102static int ext2_pathconf(struct vop_pathconf_args *);
103static int ext2_print(struct vop_print_args *);
104static int ext2_read(struct vop_read_args *);
105static int ext2_readlink(struct vop_readlink_args *);
106static int ext2_remove(struct vop_remove_args *);
107static int ext2_rename(struct vop_rename_args *);
108static int ext2_rmdir(struct vop_rmdir_args *);
109static int ext2_setattr(struct vop_setattr_args *);
110static int ext2_strategy(struct vop_strategy_args *);
111static int ext2_symlink(struct vop_symlink_args *);
112static int ext2_write(struct vop_write_args *);
113static int ext2fifo_close(struct vop_close_args *);
114static int ext2fifo_kqfilter(struct vop_kqfilter_args *);
115static int ext2fifo_read(struct vop_read_args *);
116static int ext2fifo_write(struct vop_write_args *);
117static int ext2spec_close(struct vop_close_args *);
118static int ext2spec_read(struct vop_read_args *);
119static int ext2spec_write(struct vop_write_args *);
120static int filt_ext2read(struct knote *kn, long hint);
121static int filt_ext2write(struct knote *kn, long hint);
122static int filt_ext2vnode(struct knote *kn, long hint);
123static void filt_ext2detach(struct knote *kn);
124
125/* Global vfs data structures for ext2. */
126vop_t **ext2_vnodeop_p;
127static struct vnodeopv_entry_desc ext2_vnodeop_entries[] = {
128 { &vop_default_desc, (vop_t *) vop_defaultop },
129 { &vop_access_desc, (vop_t *) ext2_access },
130 { &vop_advlock_desc, (vop_t *) ext2_advlock },
131 { &vop_bmap_desc, (vop_t *) ext2_bmap },
132 { &vop_cachedlookup_desc, (vop_t *) ext2_lookup },
133 { &vop_close_desc, (vop_t *) ext2_close },
134 { &vop_create_desc, (vop_t *) ext2_create },
135 { &vop_fsync_desc, (vop_t *) ext2_fsync },
136 { &vop_getattr_desc, (vop_t *) ext2_getattr },
137 { &vop_getwritemount_desc, (vop_t *) vop_stdgetwritemount },
138 { &vop_inactive_desc, (vop_t *) ext2_inactive },
139 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
140 { &vop_link_desc, (vop_t *) ext2_link },
141 { &vop_lock_desc, (vop_t *) vop_stdlock },
142 { &vop_lookup_desc, (vop_t *) vfs_cache_lookup },
143 { &vop_mkdir_desc, (vop_t *) ext2_mkdir },
144 { &vop_mknod_desc, (vop_t *) ext2_mknod },
145 { &vop_open_desc, (vop_t *) ext2_open },
146 { &vop_pathconf_desc, (vop_t *) ext2_pathconf },
147 { &vop_poll_desc, (vop_t *) vop_stdpoll },
148 { &vop_kqfilter_desc, (vop_t *) ext2_kqfilter },
149 { &vop_print_desc, (vop_t *) ext2_print },
150 { &vop_read_desc, (vop_t *) ext2_read },
151 { &vop_readdir_desc, (vop_t *) ext2_readdir },
152 { &vop_readlink_desc, (vop_t *) ext2_readlink },
153 { &vop_reallocblks_desc, (vop_t *) ext2_reallocblks },
154 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
155 { &vop_remove_desc, (vop_t *) ext2_remove },
156 { &vop_rename_desc, (vop_t *) ext2_rename },
157 { &vop_rmdir_desc, (vop_t *) ext2_rmdir },
158 { &vop_setattr_desc, (vop_t *) ext2_setattr },
159 { &vop_strategy_desc, (vop_t *) ext2_strategy },
160 { &vop_symlink_desc, (vop_t *) ext2_symlink },
161 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
162 { &vop_write_desc, (vop_t *) ext2_write },
163 { NULL, NULL }
164};
165static struct vnodeopv_desc ext2fs_vnodeop_opv_desc =
166 { &ext2_vnodeop_p, ext2_vnodeop_entries };
167
168vop_t **ext2_specop_p;
169static struct vnodeopv_entry_desc ext2_specop_entries[] = {
170 { &vop_default_desc, (vop_t *) spec_vnoperate },
171 { &vop_access_desc, (vop_t *) ext2_access },
172 { &vop_close_desc, (vop_t *) ext2spec_close },
173 { &vop_fsync_desc, (vop_t *) ext2_fsync },
174 { &vop_getattr_desc, (vop_t *) ext2_getattr },
175 { &vop_inactive_desc, (vop_t *) ext2_inactive },
176 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
177 { &vop_lock_desc, (vop_t *) vop_stdlock },
178 { &vop_print_desc, (vop_t *) ext2_print },
179 { &vop_read_desc, (vop_t *) ext2spec_read },
180 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
181 { &vop_setattr_desc, (vop_t *) ext2_setattr },
182 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
183 { &vop_write_desc, (vop_t *) ext2spec_write },
184 { NULL, NULL }
185};
186static struct vnodeopv_desc ext2fs_specop_opv_desc =
187 { &ext2_specop_p, ext2_specop_entries };
188
189vop_t **ext2_fifoop_p;
190static struct vnodeopv_entry_desc ext2_fifoop_entries[] = {
191 { &vop_default_desc, (vop_t *) fifo_vnoperate },
192 { &vop_access_desc, (vop_t *) ext2_access },
193 { &vop_close_desc, (vop_t *) ext2fifo_close },
194 { &vop_fsync_desc, (vop_t *) ext2_fsync },
195 { &vop_getattr_desc, (vop_t *) ext2_getattr },
196 { &vop_inactive_desc, (vop_t *) ext2_inactive },
197 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
198 { &vop_kqfilter_desc, (vop_t *) ext2fifo_kqfilter },
199 { &vop_lock_desc, (vop_t *) vop_stdlock },
200 { &vop_print_desc, (vop_t *) ext2_print },
201 { &vop_read_desc, (vop_t *) ext2fifo_read },
202 { &vop_reclaim_desc, (vop_t *) ext2_reclaim },
203 { &vop_setattr_desc, (vop_t *) ext2_setattr },
204 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
205 { &vop_write_desc, (vop_t *) ext2fifo_write },
206 { NULL, NULL }
207};
208static struct vnodeopv_desc ext2fs_fifoop_opv_desc =
209 { &ext2_fifoop_p, ext2_fifoop_entries };
210
211 VNODEOP_SET(ext2fs_vnodeop_opv_desc);
212 VNODEOP_SET(ext2fs_specop_opv_desc);
213 VNODEOP_SET(ext2fs_fifoop_opv_desc);
214
215#include <gnu/ext2fs/ext2_readwrite.c>
216
217union _qcvt {
218 int64_t qcvt;
219 int32_t val[2];
220};
221#define SETHIGH(q, h) { \
222 union _qcvt tmp; \
223 tmp.qcvt = (q); \
224 tmp.val[_QUAD_HIGHWORD] = (h); \
225 (q) = tmp.qcvt; \
226}
227#define SETLOW(q, l) { \
228 union _qcvt tmp; \
229 tmp.qcvt = (q); \
230 tmp.val[_QUAD_LOWWORD] = (l); \
231 (q) = tmp.qcvt; \
232}
233
234/*
235 * A virgin directory (no blushing please).
236 * Note that the type and namlen fields are reversed relative to ext2.
237 * Also, we don't use `struct odirtemplate', since it would just cause
238 * endianness problems.
239 */
240static struct dirtemplate mastertemplate = {
241 0, 12, 1, EXT2_FT_DIR, ".",
242 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".."
243};
244static struct dirtemplate omastertemplate = {
245 0, 12, 1, EXT2_FT_UNKNOWN, ".",
246 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".."
247};
248
249void
250ext2_itimes(vp)
251 struct vnode *vp;
252{
253 struct inode *ip;
254 struct timespec ts;
255
256 ip = VTOI(vp);
257 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
258 return;
259 if ((vp->v_type == VBLK || vp->v_type == VCHR))
260 ip->i_flag |= IN_LAZYMOD;
261 else
262 ip->i_flag |= IN_MODIFIED;
263 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
264 vfs_timestamp(&ts);
265 if (ip->i_flag & IN_ACCESS) {
266 ip->i_atime = ts.tv_sec;
267 ip->i_atimensec = ts.tv_nsec;
268 }
269 if (ip->i_flag & IN_UPDATE) {
270 ip->i_mtime = ts.tv_sec;
271 ip->i_mtimensec = ts.tv_nsec;
272 ip->i_modrev++;
273 }
274 if (ip->i_flag & IN_CHANGE) {
275 ip->i_ctime = ts.tv_sec;
276 ip->i_ctimensec = ts.tv_nsec;
277 }
278 }
279 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
280}
281
282/*
283 * Create a regular file
284 */
285static int
286ext2_create(ap)
287 struct vop_create_args /* {
288 struct vnode *a_dvp;
289 struct vnode **a_vpp;
290 struct componentname *a_cnp;
291 struct vattr *a_vap;
292 } */ *ap;
293{
294 int error;
295
296 error =
297 ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
298 ap->a_dvp, ap->a_vpp, ap->a_cnp);
299 if (error)
300 return (error);
301 return (0);
302}
303
304/*
305 * Open called.
306 *
307 * Nothing to do.
308 */
309int
310ext2_open(ap)
311 struct vop_open_args /* {
312 struct vnode *a_vp;
313 int a_mode;
314 struct ucred *a_cred;
315 struct thread *a_td;
316 } */ *ap;
317{
318
319 /*
320 * Files marked append-only must be opened for appending.
321 */
322 if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
323 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
324 return (EPERM);
325 return (0);
326}
327
328/*
329 * Close called.
330 *
331 * Update the times on the inode.
332 */
333static int
334ext2_close(ap)
335 struct vop_close_args /* {
336 struct vnode *a_vp;
337 int a_fflag;
338 struct ucred *a_cred;
339 struct thread *a_td;
340 } */ *ap;
341{
342 struct vnode *vp = ap->a_vp;
343 struct mount *mp;
344
345 mtx_lock(&vp->v_interlock);
346 if (vp->v_usecount > 1) {
347 ext2_itimes(vp);
348 mtx_unlock(&vp->v_interlock);
349 } else {
350 mtx_unlock(&vp->v_interlock);
351 /*
352 * If we are closing the last reference to an unlinked
353 * file, then it will be freed by the inactive routine.
354 * Because the freeing causes a the filesystem to be
355 * modified, it must be held up during periods when the
356 * filesystem is suspended.
357 *
358 * XXX - EAGAIN is returned to prevent vn_close from
359 * repeating the vrele operation.
360 */
361 if (vp->v_type == VREG && VTOI(vp)->i_nlink == 0) {
362 (void) vn_start_write(vp, &mp, V_WAIT);
363 vrele(vp);
364 vn_finished_write(mp);
365 return (EAGAIN);
366 }
367 }
368 return (0);
369}
370
371static int
372ext2_access(ap)
373 struct vop_access_args /* {
374 struct vnode *a_vp;
375 int a_mode;
376 struct ucred *a_cred;
377 struct thread *a_td;
378 } */ *ap;
379{
380 struct vnode *vp = ap->a_vp;
381 struct inode *ip = VTOI(vp);
382 mode_t mode = ap->a_mode;
383 int error;
384
385 /*
386 * Disallow write attempts on read-only file systems;
387 * unless the file is a socket, fifo, or a block or
388 * character device resident on the file system.
389 */
390 if (mode & VWRITE) {
391 switch (vp->v_type) {
392 case VDIR:
393 case VLNK:
394 case VREG:
395 if (vp->v_mount->mnt_flag & MNT_RDONLY)
396 return (EROFS);
397 break;
398 default:
399 break;
400 }
401 }
402
403 /* If immutable bit set, nobody gets to write it. */
404 if ((mode & VWRITE) && (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
405 return (EPERM);
406
407 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
408 ap->a_mode, ap->a_cred, NULL);
409 return (error);
410}
411
412static int
413ext2_getattr(ap)
414 struct vop_getattr_args /* {
415 struct vnode *a_vp;
416 struct vattr *a_vap;
417 struct ucred *a_cred;
418 struct thread *a_td;
419 } */ *ap;
420{
421 struct vnode *vp = ap->a_vp;
422 struct inode *ip = VTOI(vp);
423 struct vattr *vap = ap->a_vap;
424
425 ext2_itimes(vp);
426 /*
427 * Copy from inode table
428 */
429 vap->va_fsid = dev2udev(ip->i_dev);
430 vap->va_fileid = ip->i_number;
431 vap->va_mode = ip->i_mode & ~IFMT;
432 vap->va_nlink = ip->i_nlink;
433 vap->va_uid = ip->i_uid;
434 vap->va_gid = ip->i_gid;
435 vap->va_rdev = ip->i_rdev;
436 vap->va_size = ip->i_size;
437 vap->va_atime.tv_sec = ip->i_atime;
438 vap->va_atime.tv_nsec = ip->i_atimensec;
439 vap->va_mtime.tv_sec = ip->i_mtime;
440 vap->va_mtime.tv_nsec = ip->i_mtimensec;
441 vap->va_ctime.tv_sec = ip->i_ctime;
442 vap->va_ctime.tv_nsec = ip->i_ctimensec;
443 vap->va_flags = ip->i_flags;
444 vap->va_gen = ip->i_gen;
445 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
446 vap->va_bytes = dbtob((u_quad_t)ip->i_blocks);
447 vap->va_type = IFTOVT(ip->i_mode);
448 vap->va_filerev = ip->i_modrev;
449 return (0);
450}
451
452/*
453 * Set attribute vnode op. called from several syscalls
454 */
455int
456ext2_setattr(ap)
457 struct vop_setattr_args /* {
458 struct vnode *a_vp;
459 struct vattr *a_vap;
460 struct ucred *a_cred;
461 struct thread *a_td;
462 } */ *ap;
463{
464 struct vattr *vap = ap->a_vap;
465 struct vnode *vp = ap->a_vp;
466 struct inode *ip = VTOI(vp);
467 struct ucred *cred = ap->a_cred;
468 struct thread *td = ap->a_td;
469 int error;
470
471 /*
472 * Check for unsettable attributes.
473 */
474 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
475 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
476 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
477 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
478 return (EINVAL);
479 }
480 if (vap->va_flags != VNOVAL) {
481 if (vp->v_mount->mnt_flag & MNT_RDONLY)
482 return (EROFS);
483 /*
484 * Callers may only modify the file flags on objects they
485 * have VADMIN rights for.
486 */
487 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
488 return (error);
489 /*
490 * Unprivileged processes and privileged processes in
491 * jail() are not permitted to unset system flags, or
492 * modify flags if any system flags are set.
493 * Privileged non-jail processes may not modify system flags
494 * if securelevel > 0 and any existing system flags are set.
495 */
496 if (!suser_cred(cred, PRISON_ROOT)) {
497 if (ip->i_flags
498 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
499 error = securelevel_gt(cred, 0);
500 if (error)
501 return (error);
502 }
503 ip->i_flags = vap->va_flags;
504 } else {
505 if (ip->i_flags
506 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
507 (vap->va_flags & UF_SETTABLE) != vap->va_flags)
508 return (EPERM);
509 ip->i_flags &= SF_SETTABLE;
510 ip->i_flags |= (vap->va_flags & UF_SETTABLE);
511 }
512 ip->i_flag |= IN_CHANGE;
513 if (vap->va_flags & (IMMUTABLE | APPEND))
514 return (0);
515 }
516 if (ip->i_flags & (IMMUTABLE | APPEND))
517 return (EPERM);
518 /*
519 * Go through the fields and update iff not VNOVAL.
520 */
521 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
522 if (vp->v_mount->mnt_flag & MNT_RDONLY)
523 return (EROFS);
524 if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred,
525 td)) != 0)
526 return (error);
527 }
528 if (vap->va_size != VNOVAL) {
529 /*
530 * Disallow write attempts on read-only file systems;
531 * unless the file is a socket, fifo, or a block or
532 * character device resident on the file system.
533 */
534 switch (vp->v_type) {
535 case VDIR:
536 return (EISDIR);
537 case VLNK:
538 case VREG:
539 if (vp->v_mount->mnt_flag & MNT_RDONLY)
540 return (EROFS);
541 break;
542 default:
543 break;
544 }
545 if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0)
546 return (error);
547 }
548 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
549 if (vp->v_mount->mnt_flag & MNT_RDONLY)
550 return (EROFS);
551 /*
552 * From utimes(2):
553 * If times is NULL, ... The caller must be the owner of
554 * the file, have permission to write the file, or be the
555 * super-user.
556 * If times is non-NULL, ... The caller must be the owner of
557 * the file or be the super-user.
558 */
559 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) &&
560 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
561 (error = VOP_ACCESS(vp, VWRITE, cred, td))))
562 return (error);
563 if (vap->va_atime.tv_sec != VNOVAL)
564 ip->i_flag |= IN_ACCESS;
565 if (vap->va_mtime.tv_sec != VNOVAL)
566 ip->i_flag |= IN_CHANGE | IN_UPDATE;
567 ext2_itimes(vp);
568 if (vap->va_atime.tv_sec != VNOVAL) {
569 ip->i_atime = vap->va_atime.tv_sec;
570 ip->i_atimensec = vap->va_atime.tv_nsec;
571 }
572 if (vap->va_mtime.tv_sec != VNOVAL) {
573 ip->i_mtime = vap->va_mtime.tv_sec;
574 ip->i_mtimensec = vap->va_mtime.tv_nsec;
575 }
576 error = ext2_update(vp, 0);
577 if (error)
578 return (error);
579 }
580 error = 0;
581 if (vap->va_mode != (mode_t)VNOVAL) {
582 if (vp->v_mount->mnt_flag & MNT_RDONLY)
583 return (EROFS);
584 error = ext2_chmod(vp, (int)vap->va_mode, cred, td);
585 }
586 VN_KNOTE(vp, NOTE_ATTRIB);
587 return (error);
588}
589
590/*
591 * Change the mode on a file.
592 * Inode must be locked before calling.
593 */
594static int
595ext2_chmod(vp, mode, cred, td)
596 struct vnode *vp;
597 int mode;
598 struct ucred *cred;
599 struct thread *td;
600{
601 struct inode *ip = VTOI(vp);
602 int error;
603
604 /*
605 * To modify the permissions on a file, must possess VADMIN
606 * for that file.
607 */
608 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
609 return (error);
610 /*
611 * Privileged processes may set the sticky bit on non-directories,
612 * as well as set the setgid bit on a file with a group that the
613 * process is not a member of.
614 */
615 if (suser_cred(cred, PRISON_ROOT)) {
616 if (vp->v_type != VDIR && (mode & S_ISTXT))
617 return (EFTYPE);
618 if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
619 return (EPERM);
620 }
621 ip->i_mode &= ~ALLPERMS;
622 ip->i_mode |= (mode & ALLPERMS);
623 ip->i_flag |= IN_CHANGE;
624 return (0);
625}
626
627/*
628 * Perform chown operation on inode ip;
629 * inode must be locked prior to call.
630 */
631static int
632ext2_chown(vp, uid, gid, cred, td)
633 struct vnode *vp;
634 uid_t uid;
635 gid_t gid;
636 struct ucred *cred;
637 struct thread *td;
638{
639 struct inode *ip = VTOI(vp);
640 uid_t ouid;
641 gid_t ogid;
642 int error = 0;
643
644 if (uid == (uid_t)VNOVAL)
645 uid = ip->i_uid;
646 if (gid == (gid_t)VNOVAL)
647 gid = ip->i_gid;
648 /*
649 * To modify the ownership of a file, must possess VADMIN
650 * for that file.
651 */
652 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
653 return (error);
654 /*
655 * To change the owner of a file, or change the group of a file
656 * to a group of which we are not a member, the caller must
657 * have privilege.
658 */
659 if ((uid != ip->i_uid ||
660 (gid != ip->i_gid && !groupmember(gid, cred))) &&
661 (error = suser_cred(cred, PRISON_ROOT)))
662 return (error);
663 ogid = ip->i_gid;
664 ouid = ip->i_uid;
665 ip->i_gid = gid;
666 ip->i_uid = uid;
667 ip->i_flag |= IN_CHANGE;
668 if (suser_cred(cred, PRISON_ROOT) && (ouid != uid || ogid != gid))
669 ip->i_mode &= ~(ISUID | ISGID);
670 return (0);
671}
672
673/*
674 * Synch an open file.
675 */
676/* ARGSUSED */
677static int
678ext2_fsync(ap)
679 struct vop_fsync_args /* {
680 struct vnode *a_vp;
681 struct ucred *a_cred;
682 int a_waitfor;
683 struct thread *a_td;
684 } */ *ap;
685{
686 struct vnode *vp = ap->a_vp;
687 struct buf *bp;
688 struct buf *nbp;
689 int s;
690
691 /*
692 * XXX why is all this fs specific?
693 */
694
695 /*
696 * Flush all dirty buffers associated with a vnode.
697 */
698 ext2_discard_prealloc(VTOI(vp));
699
700loop:
701 s = splbio();
702 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
703 nbp = TAILQ_NEXT(bp, b_vnbufs);
704 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
705 continue;
706 if ((bp->b_flags & B_DELWRI) == 0)
707 panic("ext2_fsync: not dirty");
708 bremfree(bp);
709 splx(s);
710 /*
711 * Wait for I/O associated with indirect blocks to complete,
712 * since there is no way to quickly wait for them below.
713 */
714 if (bp->b_vp == vp || ap->a_waitfor == MNT_NOWAIT)
715 (void) bawrite(bp);
716 else
717 (void) bwrite(bp);
718 goto loop;
719 }
720 if (ap->a_waitfor == MNT_WAIT) {
721 while (vp->v_numoutput) {
722 vp->v_flag |= VBWAIT;
723 tsleep(&vp->v_numoutput, PRIBIO + 1, "e2fsyn", 0);
724 }
725#if DIAGNOSTIC
726 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
727 vprint("ext2_fsync: dirty", vp);
728 goto loop;
729 }
730#endif
731 }
732 splx(s);
733 return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT));
734}
735
736/*
737 * Mknod vnode call
738 */
739/* ARGSUSED */
740static int
741ext2_mknod(ap)
742 struct vop_mknod_args /* {
743 struct vnode *a_dvp;
744 struct vnode **a_vpp;
745 struct componentname *a_cnp;
746 struct vattr *a_vap;
747 } */ *ap;
748{
749 struct vattr *vap = ap->a_vap;
750 struct vnode **vpp = ap->a_vpp;
751 struct inode *ip;
752 ino_t ino;
753 int error;
754
755 error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
756 ap->a_dvp, vpp, ap->a_cnp);
757 if (error)
758 return (error);
759 ip = VTOI(*vpp);
760 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
761 if (vap->va_rdev != VNOVAL) {
762 /*
763 * Want to be able to use this to make badblock
764 * inodes, so don't truncate the dev number.
765 */
766 ip->i_rdev = vap->va_rdev;
767 }
768 /*
769 * Remove inode, then reload it through VFS_VGET so it is
770 * checked to see if it is an alias of an existing entry in
771 * the inode cache.
772 */
773 vput(*vpp);
774 (*vpp)->v_type = VNON;
775 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
776 vgone(*vpp);
777 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
778 if (error) {
779 *vpp = NULL;
780 return (error);
781 }
782 return (0);
783}
784
785static int
786ext2_remove(ap)
787 struct vop_remove_args /* {
788 struct vnode *a_dvp;
789 struct vnode *a_vp;
790 struct componentname *a_cnp;
791 } */ *ap;
792{
793 struct inode *ip;
794 struct vnode *vp = ap->a_vp;
795 struct vnode *dvp = ap->a_dvp;
796 int error;
797
798 ip = VTOI(vp);
799 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
800 (VTOI(dvp)->i_flags & APPEND)) {
801 error = EPERM;
802 goto out;
803 }
804 error = ext2_dirremove(dvp, ap->a_cnp);
805 if (error == 0) {
806 ip->i_nlink--;
807 ip->i_flag |= IN_CHANGE;
808 }
809out:
810 return (error);
811}
812
813/*
814 * link vnode call
815 */
816static int
817ext2_link(ap)
818 struct vop_link_args /* {
819 struct vnode *a_tdvp;
820 struct vnode *a_vp;
821 struct componentname *a_cnp;
822 } */ *ap;
823{
824 struct vnode *vp = ap->a_vp;
825 struct vnode *tdvp = ap->a_tdvp;
826 struct componentname *cnp = ap->a_cnp;
827 struct thread *td = cnp->cn_thread;
828 struct inode *ip;
829 int error;
830
831#ifdef DIAGNOSTIC
832 if ((cnp->cn_flags & HASBUF) == 0)
833 panic("ext2_link: no name");
834#endif
835 if (tdvp->v_mount != vp->v_mount) {
836 error = EXDEV;
837 goto out2;
838 }
839 if (tdvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE, td))) {
840 goto out2;
841 }
842 ip = VTOI(vp);
843 if ((nlink_t)ip->i_nlink >= LINK_MAX) {
844 error = EMLINK;
845 goto out1;
846 }
847 if (ip->i_flags & (IMMUTABLE | APPEND)) {
848 error = EPERM;
849 goto out1;
850 }
851 ip->i_nlink++;
852 ip->i_flag |= IN_CHANGE;
853 error = ext2_update(vp, 1);
854 if (!error)
855 error = ext2_direnter(ip, tdvp, cnp);
856 if (error) {
857 ip->i_nlink--;
858 ip->i_flag |= IN_CHANGE;
859 }
860out1:
861 if (tdvp != vp)
862 VOP_UNLOCK(vp, 0, td);
863out2:
864 return (error);
865}
866
867/*
868 * Rename system call.
869 * See comments in sys/ufs/ufs/ufs_vnops.c
870 */
871static int
872ext2_rename(ap)
873 struct vop_rename_args /* {
874 struct vnode *a_fdvp;
875 struct vnode *a_fvp;
876 struct componentname *a_fcnp;
877 struct vnode *a_tdvp;
878 struct vnode *a_tvp;
879 struct componentname *a_tcnp;
880 } */ *ap;
881{
882 struct vnode *tvp = ap->a_tvp;
883 struct vnode *tdvp = ap->a_tdvp;
884 struct vnode *fvp = ap->a_fvp;
885 struct vnode *fdvp = ap->a_fdvp;
886 struct componentname *tcnp = ap->a_tcnp;
887 struct componentname *fcnp = ap->a_fcnp;
888 struct thread *td = fcnp->cn_thread;
889 struct inode *ip, *xp, *dp;
890 struct dirtemplate dirbuf;
891 int doingdirectory = 0, oldparent = 0, newparent = 0;
892 int error = 0;
893 u_char namlen;
894
895#ifdef DIAGNOSTIC
896 if ((tcnp->cn_flags & HASBUF) == 0 ||
897 (fcnp->cn_flags & HASBUF) == 0)
898 panic("ext2_rename: no name");
899#endif
900 /*
901 * Check for cross-device rename.
902 */
903 if ((fvp->v_mount != tdvp->v_mount) ||
904 (tvp && (fvp->v_mount != tvp->v_mount))) {
905 error = EXDEV;
906abortit:
907 if (tdvp == tvp)
908 vrele(tdvp);
909 else
910 vput(tdvp);
911 if (tvp)
912 vput(tvp);
913 vrele(fdvp);
914 vrele(fvp);
915 return (error);
916 }
917
918 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
919 (VTOI(tdvp)->i_flags & APPEND))) {
920 error = EPERM;
921 goto abortit;
922 }
923
924 /*
925 * Check if just deleting a link name or if we've lost a race.
926 * If another process completes the same rename after we've looked
927 * up the source and have blocked looking up the target, then the
928 * source and target inodes may be identical now although the
929 * names were never linked.
930 */
931 if (fvp == tvp) {
932 if (fvp->v_type == VDIR) {
933 /*
934 * Linked directories are impossible, so we must
935 * have lost the race. Pretend that the rename
936 * completed before the lookup.
937 */
938#ifdef UFS_RENAME_DEBUG
939 printf("ext2_rename: fvp == tvp for directories\n");
940#endif
941 error = ENOENT;
942 goto abortit;
943 }
944
945 /* Release destination completely. */
946 vput(tdvp);
947 vput(tvp);
948
949 /*
950 * Delete source. There is another race now that everything
951 * is unlocked, but this doesn't cause any new complications.
952 * Relookup() may find a file that is unrelated to the
953 * original one, or it may fail. Too bad.
954 */
955 vrele(fdvp);
956 vrele(fvp);
957 fcnp->cn_flags &= ~MODMASK;
958 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
959 fcnp->cn_nameiop = DELETE;
960 VREF(fdvp);
961 error = relookup(fdvp, &fvp, fcnp);
962 if (error == 0)
963 vrele(fdvp);
964 if (fvp == NULL) {
965#ifdef UFS_RENAME_DEBUG
966 printf("ext2_rename: from name disappeared\n");
967#endif
968 return (ENOENT);
969 }
970 error = VOP_REMOVE(fdvp, fvp, fcnp);
971 if (fdvp == fvp)
972 vrele(fdvp);
973 else
974 vput(fdvp);
975 if (fvp != NULLVP)
976 vput(fvp);
977 return (error);
978 }
979 if ((error = vn_lock(fvp, LK_EXCLUSIVE, td)) != 0)
980 goto abortit;
981 dp = VTOI(fdvp);
982 ip = VTOI(fvp);
983 if (ip->i_nlink >= LINK_MAX) {
984 VOP_UNLOCK(fvp, 0, td);
985 error = EMLINK;
986 goto abortit;
987 }
988 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
989 || (dp->i_flags & APPEND)) {
990 VOP_UNLOCK(fvp, 0, td);
991 error = EPERM;
992 goto abortit;
993 }
994 if ((ip->i_mode & IFMT) == IFDIR) {
995 /*
996 * Avoid ".", "..", and aliases of "." for obvious reasons.
997 */
998 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
999 dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT ||
1000 (ip->i_flag & IN_RENAME)) {
1001 VOP_UNLOCK(fvp, 0, td);
1002 error = EINVAL;
1003 goto abortit;
1004 }
1005 ip->i_flag |= IN_RENAME;
1006 oldparent = dp->i_number;
1007 doingdirectory++;
1008 }
1009 vrele(fdvp);
1010
1011 /*
1012 * When the target exists, both the directory
1013 * and target vnodes are returned locked.
1014 */
1015 dp = VTOI(tdvp);
1016 xp = NULL;
1017 if (tvp)
1018 xp = VTOI(tvp);
1019
1020 /*
1021 * 1) Bump link count while we're moving stuff
1022 * around. If we crash somewhere before
1023 * completing our work, the link count
1024 * may be wrong, but correctable.
1025 */
1026 ip->i_nlink++;
1027 ip->i_flag |= IN_CHANGE;
1028 if ((error = ext2_update(fvp, 1)) != 0) {
1029 VOP_UNLOCK(fvp, 0, td);
1030 goto bad;
1031 }
1032
1033 /*
1034 * If ".." must be changed (ie the directory gets a new
1035 * parent) then the source directory must not be in the
1036 * directory heirarchy above the target, as this would
1037 * orphan everything below the source directory. Also
1038 * the user must have write permission in the source so
1039 * as to be able to change "..". We must repeat the call
1040 * to namei, as the parent directory is unlocked by the
1041 * call to checkpath().
1042 */
1043 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1044 VOP_UNLOCK(fvp, 0, td);
1045 if (oldparent != dp->i_number)
1046 newparent = dp->i_number;
1047 if (doingdirectory && newparent) {
1048 if (error) /* write access check above */
1049 goto bad;
1050 if (xp != NULL)
1051 vput(tvp);
1052 error = ext2_checkpath(ip, dp, tcnp->cn_cred);
1053 if (error)
1054 goto out;
1055 VREF(tdvp);
1056 error = relookup(tdvp, &tvp, tcnp);
1057 if (error)
1058 goto out;
1059 vrele(tdvp);
1060 dp = VTOI(tdvp);
1061 xp = NULL;
1062 if (tvp)
1063 xp = VTOI(tvp);
1064 }
1065 /*
1066 * 2) If target doesn't exist, link the target
1067 * to the source and unlink the source.
1068 * Otherwise, rewrite the target directory
1069 * entry to reference the source inode and
1070 * expunge the original entry's existence.
1071 */
1072 if (xp == NULL) {
1073 if (dp->i_dev != ip->i_dev)
1074 panic("ext2_rename: EXDEV");
1075 /*
1076 * Account for ".." in new directory.
1077 * When source and destination have the same
1078 * parent we don't fool with the link count.
1079 */
1080 if (doingdirectory && newparent) {
1081 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1082 error = EMLINK;
1083 goto bad;
1084 }
1085 dp->i_nlink++;
1086 dp->i_flag |= IN_CHANGE;
1087 error = ext2_update(tdvp, 1);
1088 if (error)
1089 goto bad;
1090 }
1091 error = ext2_direnter(ip, tdvp, tcnp);
1092 if (error) {
1093 if (doingdirectory && newparent) {
1094 dp->i_nlink--;
1095 dp->i_flag |= IN_CHANGE;
1096 (void)ext2_update(tdvp, 1);
1097 }
1098 goto bad;
1099 }
1100 vput(tdvp);
1101 } else {
1102 if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
1103 panic("ext2_rename: EXDEV");
1104 /*
1105 * Short circuit rename(foo, foo).
1106 */
1107 if (xp->i_number == ip->i_number)
1108 panic("ext2_rename: same file");
1109 /*
1110 * If the parent directory is "sticky", then the user must
1111 * own the parent directory, or the destination of the rename,
1112 * otherwise the destination may not be changed (except by
1113 * root). This implements append-only directories.
1114 */
1115 if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
1116 tcnp->cn_cred->cr_uid != dp->i_uid &&
1117 xp->i_uid != tcnp->cn_cred->cr_uid) {
1118 error = EPERM;
1119 goto bad;
1120 }
1121 /*
1122 * Target must be empty if a directory and have no links
1123 * to it. Also, ensure source and target are compatible
1124 * (both directories, or both not directories).
1125 */
1126 if ((xp->i_mode&IFMT) == IFDIR) {
1127 if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) ||
1128 xp->i_nlink > 2) {
1129 error = ENOTEMPTY;
1130 goto bad;
1131 }
1132 if (!doingdirectory) {
1133 error = ENOTDIR;
1134 goto bad;
1135 }
1136 cache_purge(tdvp);
1137 } else if (doingdirectory) {
1138 error = EISDIR;
1139 goto bad;
1140 }
1141 error = ext2_dirrewrite(dp, ip, tcnp);
1142 if (error)
1143 goto bad;
1144 /*
1145 * If the target directory is in the same
1146 * directory as the source directory,
1147 * decrement the link count on the parent
1148 * of the target directory.
1149 */
1150 if (doingdirectory && !newparent) {
1151 dp->i_nlink--;
1152 dp->i_flag |= IN_CHANGE;
1153 }
1154 vput(tdvp);
1155 /*
1156 * Adjust the link count of the target to
1157 * reflect the dirrewrite above. If this is
1158 * a directory it is empty and there are
1159 * no links to it, so we can squash the inode and
1160 * any space associated with it. We disallowed
1161 * renaming over top of a directory with links to
1162 * it above, as the remaining link would point to
1163 * a directory without "." or ".." entries.
1164 */
1165 xp->i_nlink--;
1166 if (doingdirectory) {
1167 if (--xp->i_nlink != 0)
1168 panic("ext2_rename: linked directory");
1169 error = ext2_truncate(tvp, (off_t)0, IO_SYNC,
1170 tcnp->cn_cred, tcnp->cn_thread);
1171 }
1172 xp->i_flag |= IN_CHANGE;
1173 vput(tvp);
1174 xp = NULL;
1175 }
1176
1177 /*
1178 * 3) Unlink the source.
1179 */
1180 fcnp->cn_flags &= ~MODMASK;
1181 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1182 VREF(fdvp);
1183 error = relookup(fdvp, &fvp, fcnp);
1184 if (error == 0)
1185 vrele(fdvp);
1186 if (fvp != NULL) {
1187 xp = VTOI(fvp);
1188 dp = VTOI(fdvp);
1189 } else {
1190 /*
1191 * From name has disappeared.
1192 */
1193 if (doingdirectory)
1194 panic("ext2_rename: lost dir entry");
1195 vrele(ap->a_fvp);
1196 return (0);
1197 }
1198 /*
1199 * Ensure that the directory entry still exists and has not
1200 * changed while the new name has been entered. If the source is
1201 * a file then the entry may have been unlinked or renamed. In
1202 * either case there is no further work to be done. If the source
1203 * is a directory then it cannot have been rmdir'ed; its link
1204 * count of three would cause a rmdir to fail with ENOTEMPTY.
1205 * The IN_RENAME flag ensures that it cannot be moved by another
1206 * rename.
1207 */
1208 if (xp != ip) {
1209 if (doingdirectory)
1210 panic("ext2_rename: lost dir entry");
1211 } else {
1212 /*
1213 * If the source is a directory with a
1214 * new parent, the link count of the old
1215 * parent directory must be decremented
1216 * and ".." set to point to the new parent.
1217 */
1218 if (doingdirectory && newparent) {
1219 dp->i_nlink--;
1220 dp->i_flag |= IN_CHANGE;
1221 error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
1222 sizeof (struct dirtemplate), (off_t)0,
1223 UIO_SYSSPACE, IO_NODELOCKED,
1224 tcnp->cn_cred, (int *)0, (struct thread *)0);
1225 if (error == 0) {
1226 /* Like ufs little-endian: */
1227 namlen = dirbuf.dotdot_type;
1228 if (namlen != 2 ||
1229 dirbuf.dotdot_name[0] != '.' ||
1230 dirbuf.dotdot_name[1] != '.') {
1231 ext2_dirbad(xp, (doff_t)12,
1232 "rename: mangled dir");
1233 } else {
1234 dirbuf.dotdot_ino = newparent;
1235 (void) vn_rdwr(UIO_WRITE, fvp,
1236 (caddr_t)&dirbuf,
1237 sizeof (struct dirtemplate),
1238 (off_t)0, UIO_SYSSPACE,
1239 IO_NODELOCKED|IO_SYNC,
1240 tcnp->cn_cred, (int *)0,
1241 (struct thread *)0);
1242 cache_purge(fdvp);
1243 }
1244 }
1245 }
1246 error = ext2_dirremove(fdvp, fcnp);
1247 if (!error) {
1248 xp->i_nlink--;
1249 xp->i_flag |= IN_CHANGE;
1250 }
1251 xp->i_flag &= ~IN_RENAME;
1252 }
1253 if (dp)
1254 vput(fdvp);
1255 if (xp)
1256 vput(fvp);
1257 vrele(ap->a_fvp);
1258 return (error);
1259
1260bad:
1261 if (xp)
1262 vput(ITOV(xp));
1263 vput(ITOV(dp));
1264out:
1265 if (doingdirectory)
1266 ip->i_flag &= ~IN_RENAME;
1267 if (vn_lock(fvp, LK_EXCLUSIVE, td) == 0) {
1268 ip->i_nlink--;
1269 ip->i_flag |= IN_CHANGE;
1270 ip->i_flag &= ~IN_RENAME;
1271 vput(fvp);
1272 } else
1273 vrele(fvp);
1274 return (error);
1275}
1276
1277/*
1278 * Mkdir system call
1279 */
1280static int
1281ext2_mkdir(ap)
1282 struct vop_mkdir_args /* {
1283 struct vnode *a_dvp;
1284 struct vnode **a_vpp;
1285 struct componentname *a_cnp;
1286 struct vattr *a_vap;
1287 } */ *ap;
1288{
1289 struct vnode *dvp = ap->a_dvp;
1290 struct vattr *vap = ap->a_vap;
1291 struct componentname *cnp = ap->a_cnp;
1292 struct inode *ip, *dp;
1293 struct vnode *tvp;
1294 struct dirtemplate dirtemplate, *dtp;
1295 int error, dmode;
1296
1297#ifdef DIAGNOSTIC
1298 if ((cnp->cn_flags & HASBUF) == 0)
1299 panic("ext2_mkdir: no name");
1300#endif
1301 dp = VTOI(dvp);
1302 if ((nlink_t)dp->i_nlink >= LINK_MAX) {
1303 error = EMLINK;
1304 goto out;
1305 }
1306 dmode = vap->va_mode & 0777;
1307 dmode |= IFDIR;
1308 /*
1309 * Must simulate part of ext2_makeinode here to acquire the inode,
1310 * but not have it entered in the parent directory. The entry is
1311 * made later after writing "." and ".." entries.
1312 */
1313 error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp);
1314 if (error)
1315 goto out;
1316 ip = VTOI(tvp);
1317 ip->i_gid = dp->i_gid;
1318#ifdef SUIDDIR
1319 {
1320 /*
1321 * if we are hacking owners here, (only do this where told to)
1322 * and we are not giving it TOO root, (would subvert quotas)
1323 * then go ahead and give it to the other user.
1324 * The new directory also inherits the SUID bit.
1325 * If user's UID and dir UID are the same,
1326 * 'give it away' so that the SUID is still forced on.
1327 */
1328 if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
1329 (dp->i_mode & ISUID) && dp->i_uid) {
1330 dmode |= ISUID;
1331 ip->i_uid = dp->i_uid;
1332 } else {
1333 ip->i_uid = cnp->cn_cred->cr_uid;
1334 }
1335 }
1336#else
1337 ip->i_uid = cnp->cn_cred->cr_uid;
1338#endif
1339 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
1340 ip->i_mode = dmode;
1341 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
1342 ip->i_nlink = 2;
1343 if (cnp->cn_flags & ISWHITEOUT)
1344 ip->i_flags |= UF_OPAQUE;
1345 error = ext2_update(tvp, 1);
1346
1347 /*
1348 * Bump link count in parent directory
1349 * to reflect work done below. Should
1350 * be done before reference is created
1351 * so reparation is possible if we crash.
1352 */
1353 dp->i_nlink++;
1354 dp->i_flag |= IN_CHANGE;
1355 error = ext2_update(dvp, 1);
1356 if (error)
1357 goto bad;
1358
1359 /* Initialize directory with "." and ".." from static template. */
1360 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs->s_es,
1361 EXT2_FEATURE_INCOMPAT_FILETYPE))
1362 dtp = &mastertemplate;
1363 else
1364 dtp = &omastertemplate;
1365 dirtemplate = *dtp;
1366 dirtemplate.dot_ino = ip->i_number;
1367 dirtemplate.dotdot_ino = dp->i_number;
1368 /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE
1369 * so let's just redefine it - for this function only
1370 */
1371#undef DIRBLKSIZ
1372#define DIRBLKSIZ VTOI(dvp)->i_e2fs->s_blocksize
1373 dirtemplate.dotdot_reclen = DIRBLKSIZ - 12;
1374 error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
1375 sizeof (dirtemplate), (off_t)0, UIO_SYSSPACE,
1376 IO_NODELOCKED|IO_SYNC, cnp->cn_cred, (int *)0, (struct thread *)0);
1377 if (error) {
1378 dp->i_nlink--;
1379 dp->i_flag |= IN_CHANGE;
1380 goto bad;
1381 }
1382 if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
1383 /* XXX should grow with balloc() */
1384 panic("ext2_mkdir: blksize");
1385 else {
1386 ip->i_size = DIRBLKSIZ;
1387 ip->i_flag |= IN_CHANGE;
1388 }
1389
1390 /* Directory set up, now install its entry in the parent directory. */
1391 error = ext2_direnter(ip, dvp, cnp);
1392 if (error) {
1393 dp->i_nlink--;
1394 dp->i_flag |= IN_CHANGE;
1395 }
1396bad:
1397 /*
1398 * No need to do an explicit VOP_TRUNCATE here, vrele will do this
1399 * for us because we set the link count to 0.
1400 */
1401 if (error) {
1402 ip->i_nlink = 0;
1403 ip->i_flag |= IN_CHANGE;
1404 vput(tvp);
1405 } else
1406 *ap->a_vpp = tvp;
1407out:
1408 return (error);
1409#undef DIRBLKSIZ
1410#define DIRBLKSIZ DEV_BSIZE
1411}
1412
1413/*
1414 * Rmdir system call.
1415 */
1416static int
1417ext2_rmdir(ap)
1418 struct vop_rmdir_args /* {
1419 struct vnode *a_dvp;
1420 struct vnode *a_vp;
1421 struct componentname *a_cnp;
1422 } */ *ap;
1423{
1424 struct vnode *vp = ap->a_vp;
1425 struct vnode *dvp = ap->a_dvp;
1426 struct componentname *cnp = ap->a_cnp;
1427 struct thread *td = cnp->cn_thread;
1428 struct inode *ip, *dp;
1429 int error;
1430
1431 ip = VTOI(vp);
1432 dp = VTOI(dvp);
1433
1434 /*
1435 * Verify the directory is empty (and valid).
1436 * (Rmdir ".." won't be valid since
1437 * ".." will contain a reference to
1438 * the current directory and thus be
1439 * non-empty.)
1440 */
1441 error = 0;
1442 if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) {
1443 error = ENOTEMPTY;
1444 goto out;
1445 }
1446 if ((dp->i_flags & APPEND)
1447 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1448 error = EPERM;
1449 goto out;
1450 }
1451 /*
1452 * Delete reference to directory before purging
1453 * inode. If we crash in between, the directory
1454 * will be reattached to lost+found,
1455 */
1456 error = ext2_dirremove(dvp, cnp);
1457 if (error)
1458 goto out;
1459 dp->i_nlink--;
1460 dp->i_flag |= IN_CHANGE;
1461 cache_purge(dvp);
1462 VOP_UNLOCK(dvp, 0, td);
1463 /*
1464 * Truncate inode. The only stuff left
1465 * in the directory is "." and "..". The
1466 * "." reference is inconsequential since
1467 * we're quashing it. The ".." reference
1468 * has already been adjusted above. We've
1469 * removed the "." reference and the reference
1470 * in the parent directory, but there may be
1471 * other hard links so decrement by 2 and
1472 * worry about them later.
1473 */
1474 ip->i_nlink -= 2;
1475 error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, td);
1476 cache_purge(ITOV(ip));
1477 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1478out:
1479 return (error);
1480}
1481
1482/*
1483 * symlink -- make a symbolic link
1484 */
1485static int
1486ext2_symlink(ap)
1487 struct vop_symlink_args /* {
1488 struct vnode *a_dvp;
1489 struct vnode **a_vpp;
1490 struct componentname *a_cnp;
1491 struct vattr *a_vap;
1492 char *a_target;
1493 } */ *ap;
1494{
1495 struct vnode *vp, **vpp = ap->a_vpp;
1496 struct inode *ip;
1497 int len, error;
1498
1499 error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
1500 vpp, ap->a_cnp);
1501 if (error)
1502 return (error);
1503 vp = *vpp;
1504 len = strlen(ap->a_target);
1505 if (len < vp->v_mount->mnt_maxsymlinklen) {
1506 ip = VTOI(vp);
1507 bcopy(ap->a_target, (char *)ip->i_shortlink, len);
1508 ip->i_size = len;
1509 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1510 } else
1511 error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
1512 UIO_SYSSPACE, IO_NODELOCKED, ap->a_cnp->cn_cred, (int *)0,
1513 (struct thread *)0);
1514 if (error)
1515 vput(vp);
1516 return (error);
1517}
1518
1519/*
1520 * Return target name of a symbolic link
1521 */
1522static int
1523ext2_readlink(ap)
1524 struct vop_readlink_args /* {
1525 struct vnode *a_vp;
1526 struct uio *a_uio;
1527 struct ucred *a_cred;
1528 } */ *ap;
1529{
1530 struct vnode *vp = ap->a_vp;
1531 struct inode *ip = VTOI(vp);
1532 int isize;
1533
1534 isize = ip->i_size;
1535 if (isize < vp->v_mount->mnt_maxsymlinklen) {
1536 uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
1537 return (0);
1538 }
1539 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
1540}
1541
1542/*
1543 * Calculate the logical to physical mapping if not done already,
1544 * then call the device strategy routine.
1545 *
1546 * In order to be able to swap to a file, the ext2_bmaparray() operation may not
1547 * deadlock on memory. See ext2_bmap() for details.
1548 */
1549int
1550ext2_strategy(ap)
1551 struct vop_strategy_args /* {
1552 struct vnode *a_vp;
1553 struct buf *a_bp;
1554 } */ *ap;
1555{
1556 struct buf *bp = ap->a_bp;
1557 struct vnode *vp = ap->a_vp;
1558 struct inode *ip;
|