FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_vnops.c
1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_mac.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/file.h>
46 #include <sys/kdb.h>
47 #include <sys/stat.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/limits.h>
51 #include <sys/lock.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/bio.h>
57 #include <sys/buf.h>
58 #include <sys/filio.h>
59 #include <sys/sx.h>
60 #include <sys/ttycom.h>
61 #include <sys/conf.h>
62 #include <sys/syslog.h>
63 #include <sys/unistd.h>
64
65 #include <security/mac/mac_framework.h>
66
67 static fo_rdwr_t vn_read;
68 static fo_rdwr_t vn_write;
69 static fo_ioctl_t vn_ioctl;
70 static fo_poll_t vn_poll;
71 static fo_kqfilter_t vn_kqfilter;
72 static fo_stat_t vn_statfile;
73 static fo_close_t vn_closefile;
74
75 struct fileops vnops = {
76 .fo_read = vn_read,
77 .fo_write = vn_write,
78 .fo_ioctl = vn_ioctl,
79 .fo_poll = vn_poll,
80 .fo_kqfilter = vn_kqfilter,
81 .fo_stat = vn_statfile,
82 .fo_close = vn_closefile,
83 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
84 };
85
86 int
87 vn_open(ndp, flagp, cmode, fp)
88 struct nameidata *ndp;
89 int *flagp, cmode;
90 struct file *fp;
91 {
92 struct thread *td = ndp->ni_cnd.cn_thread;
93
94 return (vn_open_cred(ndp, flagp, cmode, td->td_ucred, fp));
95 }
96
97 /*
98 * Common code for vnode open operations.
99 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
100 *
101 * Note that this does NOT free nameidata for the successful case,
102 * due to the NDINIT being done elsewhere.
103 */
104 int
105 vn_open_cred(ndp, flagp, cmode, cred, fp)
106 struct nameidata *ndp;
107 int *flagp, cmode;
108 struct ucred *cred;
109 struct file *fp;
110 {
111 struct vnode *vp;
112 struct mount *mp;
113 struct thread *td = ndp->ni_cnd.cn_thread;
114 struct vattr vat;
115 struct vattr *vap = &vat;
116 int mode, fmode, error;
117 int vfslocked, mpsafe;
118
119 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
120 restart:
121 vfslocked = 0;
122 fmode = *flagp;
123 if (fmode & O_CREAT) {
124 ndp->ni_cnd.cn_nameiop = CREATE;
125 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
126 MPSAFE | AUDITVNODE1;
127 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
128 ndp->ni_cnd.cn_flags |= FOLLOW;
129 bwillwrite();
130 if ((error = namei(ndp)) != 0)
131 return (error);
132 vfslocked = NDHASGIANT(ndp);
133 if (!mpsafe)
134 ndp->ni_cnd.cn_flags &= ~MPSAFE;
135 if (ndp->ni_vp == NULL) {
136 VATTR_NULL(vap);
137 vap->va_type = VREG;
138 vap->va_mode = cmode;
139 if (fmode & O_EXCL)
140 vap->va_vaflags |= VA_EXCLUSIVE;
141 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
142 NDFREE(ndp, NDF_ONLY_PNBUF);
143 vput(ndp->ni_dvp);
144 VFS_UNLOCK_GIANT(vfslocked);
145 if ((error = vn_start_write(NULL, &mp,
146 V_XSLEEP | PCATCH)) != 0)
147 return (error);
148 goto restart;
149 }
150 #ifdef MAC
151 error = mac_check_vnode_create(cred, ndp->ni_dvp,
152 &ndp->ni_cnd, vap);
153 if (error == 0) {
154 #endif
155 VOP_LEASE(ndp->ni_dvp, td, cred, LEASE_WRITE);
156 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
157 &ndp->ni_cnd, vap);
158 #ifdef MAC
159 }
160 #endif
161 vput(ndp->ni_dvp);
162 vn_finished_write(mp);
163 if (error) {
164 VFS_UNLOCK_GIANT(vfslocked);
165 NDFREE(ndp, NDF_ONLY_PNBUF);
166 return (error);
167 }
168 fmode &= ~O_TRUNC;
169 vp = ndp->ni_vp;
170 } else {
171 if (ndp->ni_dvp == ndp->ni_vp)
172 vrele(ndp->ni_dvp);
173 else
174 vput(ndp->ni_dvp);
175 ndp->ni_dvp = NULL;
176 vp = ndp->ni_vp;
177 if (fmode & O_EXCL) {
178 error = EEXIST;
179 goto bad;
180 }
181 fmode &= ~O_CREAT;
182 }
183 } else {
184 ndp->ni_cnd.cn_nameiop = LOOKUP;
185 ndp->ni_cnd.cn_flags = ISOPEN |
186 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
187 LOCKLEAF | MPSAFE | AUDITVNODE1;
188 if ((error = namei(ndp)) != 0)
189 return (error);
190 if (!mpsafe)
191 ndp->ni_cnd.cn_flags &= ~MPSAFE;
192 vfslocked = NDHASGIANT(ndp);
193 vp = ndp->ni_vp;
194 }
195 if (vp->v_type == VLNK) {
196 error = EMLINK;
197 goto bad;
198 }
199 if (vp->v_type == VSOCK) {
200 error = EOPNOTSUPP;
201 goto bad;
202 }
203 mode = 0;
204 if (fmode & (FWRITE | O_TRUNC)) {
205 if (vp->v_type == VDIR) {
206 error = EISDIR;
207 goto bad;
208 }
209 mode |= VWRITE;
210 }
211 if (fmode & FREAD)
212 mode |= VREAD;
213 if (fmode & O_APPEND)
214 mode |= VAPPEND;
215 #ifdef MAC
216 error = mac_check_vnode_open(cred, vp, mode);
217 if (error)
218 goto bad;
219 #endif
220 if ((fmode & O_CREAT) == 0) {
221 if (mode & VWRITE) {
222 error = vn_writechk(vp);
223 if (error)
224 goto bad;
225 }
226 if (mode) {
227 error = VOP_ACCESS(vp, mode, cred, td);
228 if (error)
229 goto bad;
230 }
231 }
232 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
233 goto bad;
234
235 if (fmode & FWRITE)
236 vp->v_writecount++;
237 *flagp = fmode;
238 ASSERT_VOP_ELOCKED(vp, "vn_open_cred");
239 if (!mpsafe)
240 VFS_UNLOCK_GIANT(vfslocked);
241 return (0);
242 bad:
243 NDFREE(ndp, NDF_ONLY_PNBUF);
244 vput(vp);
245 VFS_UNLOCK_GIANT(vfslocked);
246 *flagp = fmode;
247 ndp->ni_vp = NULL;
248 return (error);
249 }
250
251 /*
252 * Check for write permissions on the specified vnode.
253 * Prototype text segments cannot be written.
254 */
255 int
256 vn_writechk(vp)
257 register struct vnode *vp;
258 {
259
260 ASSERT_VOP_LOCKED(vp, "vn_writechk");
261 /*
262 * If there's shared text associated with
263 * the vnode, try to free it up once. If
264 * we fail, we can't allow writing.
265 */
266 if (vp->v_vflag & VV_TEXT)
267 return (ETXTBSY);
268
269 return (0);
270 }
271
272 /*
273 * Vnode close call
274 */
275 int
276 vn_close(vp, flags, file_cred, td)
277 register struct vnode *vp;
278 int flags;
279 struct ucred *file_cred;
280 struct thread *td;
281 {
282 struct mount *mp;
283 int error;
284
285 VFS_ASSERT_GIANT(vp->v_mount);
286
287 vn_start_write(vp, &mp, V_WAIT);
288 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
289 if (flags & FWRITE) {
290 VNASSERT(vp->v_writecount > 0, vp,
291 ("vn_close: negative writecount"));
292 vp->v_writecount--;
293 }
294 error = VOP_CLOSE(vp, flags, file_cred, td);
295 vput(vp);
296 vn_finished_write(mp);
297 return (error);
298 }
299
300 /*
301 * Sequential heuristic - detect sequential operation
302 */
303 static __inline
304 int
305 sequential_heuristic(struct uio *uio, struct file *fp)
306 {
307
308 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
309 uio->uio_offset == fp->f_nextoff) {
310 /*
311 * XXX we assume that the filesystem block size is
312 * the default. Not true, but still gives us a pretty
313 * good indicator of how sequential the read operations
314 * are.
315 */
316 fp->f_seqcount += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
317 if (fp->f_seqcount > IO_SEQMAX)
318 fp->f_seqcount = IO_SEQMAX;
319 return(fp->f_seqcount << IO_SEQSHIFT);
320 }
321
322 /*
323 * Not sequential, quick draw-down of seqcount
324 */
325 if (fp->f_seqcount > 1)
326 fp->f_seqcount = 1;
327 else
328 fp->f_seqcount = 0;
329 return(0);
330 }
331
332 /*
333 * Package up an I/O request on a vnode into a uio and do it.
334 */
335 int
336 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred,
337 aresid, td)
338 enum uio_rw rw;
339 struct vnode *vp;
340 void *base;
341 int len;
342 off_t offset;
343 enum uio_seg segflg;
344 int ioflg;
345 struct ucred *active_cred;
346 struct ucred *file_cred;
347 int *aresid;
348 struct thread *td;
349 {
350 struct uio auio;
351 struct iovec aiov;
352 struct mount *mp;
353 struct ucred *cred;
354 int error;
355
356 VFS_ASSERT_GIANT(vp->v_mount);
357
358 if ((ioflg & IO_NODELOCKED) == 0) {
359 mp = NULL;
360 if (rw == UIO_WRITE) {
361 if (vp->v_type != VCHR &&
362 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
363 != 0)
364 return (error);
365 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
366 } else {
367 /*
368 * XXX This should be LK_SHARED but I don't trust VFS
369 * enough to leave it like that until it has been
370 * reviewed further.
371 */
372 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
373 }
374
375 }
376 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
377 auio.uio_iov = &aiov;
378 auio.uio_iovcnt = 1;
379 aiov.iov_base = base;
380 aiov.iov_len = len;
381 auio.uio_resid = len;
382 auio.uio_offset = offset;
383 auio.uio_segflg = segflg;
384 auio.uio_rw = rw;
385 auio.uio_td = td;
386 error = 0;
387 #ifdef MAC
388 if ((ioflg & IO_NOMACCHECK) == 0) {
389 if (rw == UIO_READ)
390 error = mac_check_vnode_read(active_cred, file_cred,
391 vp);
392 else
393 error = mac_check_vnode_write(active_cred, file_cred,
394 vp);
395 }
396 #endif
397 if (error == 0) {
398 if (file_cred)
399 cred = file_cred;
400 else
401 cred = active_cred;
402 if (rw == UIO_READ)
403 error = VOP_READ(vp, &auio, ioflg, cred);
404 else
405 error = VOP_WRITE(vp, &auio, ioflg, cred);
406 }
407 if (aresid)
408 *aresid = auio.uio_resid;
409 else
410 if (auio.uio_resid && error == 0)
411 error = EIO;
412 if ((ioflg & IO_NODELOCKED) == 0) {
413 if (rw == UIO_WRITE && vp->v_type != VCHR)
414 vn_finished_write(mp);
415 VOP_UNLOCK(vp, 0, td);
416 }
417 return (error);
418 }
419
420 /*
421 * Package up an I/O request on a vnode into a uio and do it. The I/O
422 * request is split up into smaller chunks and we try to avoid saturating
423 * the buffer cache while potentially holding a vnode locked, so we
424 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield()
425 * to give other processes a chance to lock the vnode (either other processes
426 * core'ing the same binary, or unrelated processes scanning the directory).
427 */
428 int
429 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
430 file_cred, aresid, td)
431 enum uio_rw rw;
432 struct vnode *vp;
433 void *base;
434 size_t len;
435 off_t offset;
436 enum uio_seg segflg;
437 int ioflg;
438 struct ucred *active_cred;
439 struct ucred *file_cred;
440 size_t *aresid;
441 struct thread *td;
442 {
443 int error = 0;
444 int iaresid;
445
446 VFS_ASSERT_GIANT(vp->v_mount);
447
448 do {
449 int chunk;
450
451 /*
452 * Force `offset' to a multiple of MAXBSIZE except possibly
453 * for the first chunk, so that filesystems only need to
454 * write full blocks except possibly for the first and last
455 * chunks.
456 */
457 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
458
459 if (chunk > len)
460 chunk = len;
461 if (rw != UIO_READ && vp->v_type == VREG)
462 bwillwrite();
463 iaresid = 0;
464 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
465 ioflg, active_cred, file_cred, &iaresid, td);
466 len -= chunk; /* aresid calc already includes length */
467 if (error)
468 break;
469 offset += chunk;
470 base = (char *)base + chunk;
471 uio_yield();
472 } while (len);
473 if (aresid)
474 *aresid = len + iaresid;
475 return (error);
476 }
477
478 /*
479 * File table vnode read routine.
480 */
481 static int
482 vn_read(fp, uio, active_cred, flags, td)
483 struct file *fp;
484 struct uio *uio;
485 struct ucred *active_cred;
486 struct thread *td;
487 int flags;
488 {
489 struct vnode *vp;
490 int error, ioflag;
491 int vfslocked;
492
493 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
494 uio->uio_td, td));
495 vp = fp->f_vnode;
496 ioflag = 0;
497 if (fp->f_flag & FNONBLOCK)
498 ioflag |= IO_NDELAY;
499 if (fp->f_flag & O_DIRECT)
500 ioflag |= IO_DIRECT;
501 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
502 VOP_LEASE(vp, td, fp->f_cred, LEASE_READ);
503 /*
504 * According to McKusick the vn lock was protecting f_offset here.
505 * It is now protected by the FOFFSET_LOCKED flag.
506 */
507 if ((flags & FOF_OFFSET) == 0) {
508 FILE_LOCK(fp);
509 while(fp->f_vnread_flags & FOFFSET_LOCKED) {
510 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
511 msleep(&fp->f_vnread_flags,fp->f_mtxp,PUSER -1,"vnread offlock",0);
512 }
513 fp->f_vnread_flags |= FOFFSET_LOCKED;
514 FILE_UNLOCK(fp);
515 vn_lock(vp, LK_SHARED | LK_RETRY, td);
516 uio->uio_offset = fp->f_offset;
517 } else
518 vn_lock(vp, LK_SHARED | LK_RETRY, td);
519
520 ioflag |= sequential_heuristic(uio, fp);
521
522 #ifdef MAC
523 error = mac_check_vnode_read(active_cred, fp->f_cred, vp);
524 if (error == 0)
525 #endif
526 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
527 if ((flags & FOF_OFFSET) == 0) {
528 fp->f_offset = uio->uio_offset;
529 FILE_LOCK(fp);
530 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
531 wakeup(&fp->f_vnread_flags);
532 fp->f_vnread_flags = 0;
533 FILE_UNLOCK(fp);
534 }
535 fp->f_nextoff = uio->uio_offset;
536 VOP_UNLOCK(vp, 0, td);
537 VFS_UNLOCK_GIANT(vfslocked);
538 return (error);
539 }
540
541 /*
542 * File table vnode write routine.
543 */
544 static int
545 vn_write(fp, uio, active_cred, flags, td)
546 struct file *fp;
547 struct uio *uio;
548 struct ucred *active_cred;
549 struct thread *td;
550 int flags;
551 {
552 struct vnode *vp;
553 struct mount *mp;
554 int error, ioflag;
555 int vfslocked;
556
557 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
558 uio->uio_td, td));
559 vp = fp->f_vnode;
560 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
561 if (vp->v_type == VREG)
562 bwillwrite();
563 ioflag = IO_UNIT;
564 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
565 ioflag |= IO_APPEND;
566 if (fp->f_flag & FNONBLOCK)
567 ioflag |= IO_NDELAY;
568 if (fp->f_flag & O_DIRECT)
569 ioflag |= IO_DIRECT;
570 if ((fp->f_flag & O_FSYNC) ||
571 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
572 ioflag |= IO_SYNC;
573 mp = NULL;
574 if (vp->v_type != VCHR &&
575 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
576 goto unlock;
577 VOP_LEASE(vp, td, fp->f_cred, LEASE_WRITE);
578 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
579 if ((flags & FOF_OFFSET) == 0)
580 uio->uio_offset = fp->f_offset;
581 ioflag |= sequential_heuristic(uio, fp);
582 #ifdef MAC
583 error = mac_check_vnode_write(active_cred, fp->f_cred, vp);
584 if (error == 0)
585 #endif
586 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
587 if ((flags & FOF_OFFSET) == 0)
588 fp->f_offset = uio->uio_offset;
589 fp->f_nextoff = uio->uio_offset;
590 VOP_UNLOCK(vp, 0, td);
591 if (vp->v_type != VCHR)
592 vn_finished_write(mp);
593 unlock:
594 VFS_UNLOCK_GIANT(vfslocked);
595 return (error);
596 }
597
598 /*
599 * File table vnode stat routine.
600 */
601 static int
602 vn_statfile(fp, sb, active_cred, td)
603 struct file *fp;
604 struct stat *sb;
605 struct ucred *active_cred;
606 struct thread *td;
607 {
608 struct vnode *vp = fp->f_vnode;
609 int vfslocked;
610 int error;
611
612 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
613 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
614 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
615 VOP_UNLOCK(vp, 0, td);
616 VFS_UNLOCK_GIANT(vfslocked);
617
618 return (error);
619 }
620
621 /*
622 * Stat a vnode; implementation for the stat syscall
623 */
624 int
625 vn_stat(vp, sb, active_cred, file_cred, td)
626 struct vnode *vp;
627 register struct stat *sb;
628 struct ucred *active_cred;
629 struct ucred *file_cred;
630 struct thread *td;
631 {
632 struct vattr vattr;
633 register struct vattr *vap;
634 int error;
635 u_short mode;
636
637 #ifdef MAC
638 error = mac_check_vnode_stat(active_cred, file_cred, vp);
639 if (error)
640 return (error);
641 #endif
642
643 vap = &vattr;
644 error = VOP_GETATTR(vp, vap, active_cred, td);
645 if (error)
646 return (error);
647
648 /*
649 * Zero the spare stat fields
650 */
651 bzero(sb, sizeof *sb);
652
653 /*
654 * Copy from vattr table
655 */
656 if (vap->va_fsid != VNOVAL)
657 sb->st_dev = vap->va_fsid;
658 else
659 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
660 sb->st_ino = vap->va_fileid;
661 mode = vap->va_mode;
662 switch (vap->va_type) {
663 case VREG:
664 mode |= S_IFREG;
665 break;
666 case VDIR:
667 mode |= S_IFDIR;
668 break;
669 case VBLK:
670 mode |= S_IFBLK;
671 break;
672 case VCHR:
673 mode |= S_IFCHR;
674 break;
675 case VLNK:
676 mode |= S_IFLNK;
677 break;
678 case VSOCK:
679 mode |= S_IFSOCK;
680 break;
681 case VFIFO:
682 mode |= S_IFIFO;
683 break;
684 default:
685 return (EBADF);
686 };
687 sb->st_mode = mode;
688 sb->st_nlink = vap->va_nlink;
689 sb->st_uid = vap->va_uid;
690 sb->st_gid = vap->va_gid;
691 sb->st_rdev = vap->va_rdev;
692 if (vap->va_size > OFF_MAX)
693 return (EOVERFLOW);
694 sb->st_size = vap->va_size;
695 sb->st_atimespec = vap->va_atime;
696 sb->st_mtimespec = vap->va_mtime;
697 sb->st_ctimespec = vap->va_ctime;
698 sb->st_birthtimespec = vap->va_birthtime;
699
700 /*
701 * According to www.opengroup.org, the meaning of st_blksize is
702 * "a filesystem-specific preferred I/O block size for this
703 * object. In some filesystem types, this may vary from file
704 * to file"
705 * Default to PAGE_SIZE after much discussion.
706 * XXX: min(PAGE_SIZE, vp->v_bufobj.bo_bsize) may be more correct.
707 */
708
709 sb->st_blksize = PAGE_SIZE;
710
711 sb->st_flags = vap->va_flags;
712 if (priv_check(td, PRIV_VFS_GENERATION))
713 sb->st_gen = 0;
714 else
715 sb->st_gen = vap->va_gen;
716
717 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
718 return (0);
719 }
720
721 /*
722 * File table vnode ioctl routine.
723 */
724 static int
725 vn_ioctl(fp, com, data, active_cred, td)
726 struct file *fp;
727 u_long com;
728 void *data;
729 struct ucred *active_cred;
730 struct thread *td;
731 {
732 struct vnode *vp = fp->f_vnode;
733 struct vattr vattr;
734 int vfslocked;
735 int error;
736
737 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
738 error = ENOTTY;
739 switch (vp->v_type) {
740 case VREG:
741 case VDIR:
742 if (com == FIONREAD) {
743 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
744 error = VOP_GETATTR(vp, &vattr, active_cred, td);
745 VOP_UNLOCK(vp, 0, td);
746 if (!error)
747 *(int *)data = vattr.va_size - fp->f_offset;
748 }
749 if (com == FIONBIO || com == FIOASYNC) /* XXX */
750 error = 0;
751 else
752 error = VOP_IOCTL(vp, com, data, fp->f_flag,
753 active_cred, td);
754 break;
755
756 default:
757 break;
758 }
759 VFS_UNLOCK_GIANT(vfslocked);
760 return (error);
761 }
762
763 /*
764 * File table vnode poll routine.
765 */
766 static int
767 vn_poll(fp, events, active_cred, td)
768 struct file *fp;
769 int events;
770 struct ucred *active_cred;
771 struct thread *td;
772 {
773 struct vnode *vp;
774 int vfslocked;
775 int error;
776
777 vp = fp->f_vnode;
778 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
779 #ifdef MAC
780 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
781 error = mac_check_vnode_poll(active_cred, fp->f_cred, vp);
782 VOP_UNLOCK(vp, 0, td);
783 if (!error)
784 #endif
785
786 error = VOP_POLL(vp, events, fp->f_cred, td);
787 VFS_UNLOCK_GIANT(vfslocked);
788 return (error);
789 }
790
791 /*
792 * Check that the vnode is still valid, and if so
793 * acquire requested lock.
794 */
795 int
796 _vn_lock(struct vnode *vp, int flags, struct thread *td, char *file, int line)
797 {
798 int error;
799
800 do {
801 if ((flags & LK_INTERLOCK) == 0)
802 VI_LOCK(vp);
803 #ifdef DEBUG_VFS_LOCKS
804 KASSERT(vp->v_holdcnt != 0,
805 ("vn_lock %p: zero hold count", vp));
806 #endif
807 if ((flags & LK_NOWAIT || (flags & LK_TYPE_MASK) == 0) &&
808 vp->v_iflag & VI_DOOMED) {
809 VI_UNLOCK(vp);
810 return (ENOENT);
811 }
812 /*
813 * Just polling to check validity.
814 */
815 if ((flags & LK_TYPE_MASK) == 0) {
816 VI_UNLOCK(vp);
817 return (0);
818 }
819 /*
820 * lockmgr drops interlock before it will return for
821 * any reason. So force the code above to relock it.
822 */
823 error = VOP_LOCK1(vp, flags | LK_INTERLOCK, td, file, line);
824 flags &= ~LK_INTERLOCK;
825 KASSERT((flags & LK_RETRY) == 0 || error == 0,
826 ("LK_RETRY set with incompatible flags %d\n", flags));
827 /*
828 * Callers specify LK_RETRY if they wish to get dead vnodes.
829 * If RETRY is not set, we return ENOENT instead.
830 */
831 if (error == 0 && vp->v_iflag & VI_DOOMED &&
832 (flags & LK_RETRY) == 0) {
833 VOP_UNLOCK(vp, 0, td);
834 error = ENOENT;
835 break;
836 }
837 } while (flags & LK_RETRY && error != 0);
838 return (error);
839 }
840
841 /*
842 * File table vnode close routine.
843 */
844 static int
845 vn_closefile(fp, td)
846 struct file *fp;
847 struct thread *td;
848 {
849 struct vnode *vp;
850 struct flock lf;
851 int vfslocked;
852 int error;
853
854 vp = fp->f_vnode;
855
856 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
857 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
858 lf.l_whence = SEEK_SET;
859 lf.l_start = 0;
860 lf.l_len = 0;
861 lf.l_type = F_UNLCK;
862 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
863 }
864
865 fp->f_ops = &badfileops;
866
867 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
868 VFS_UNLOCK_GIANT(vfslocked);
869 return (error);
870 }
871
872 /*
873 * Preparing to start a filesystem write operation. If the operation is
874 * permitted, then we bump the count of operations in progress and
875 * proceed. If a suspend request is in progress, we wait until the
876 * suspension is over, and then proceed.
877 */
878 int
879 vn_start_write(vp, mpp, flags)
880 struct vnode *vp;
881 struct mount **mpp;
882 int flags;
883 {
884 struct mount *mp;
885 int error;
886
887 error = 0;
888 /*
889 * If a vnode is provided, get and return the mount point that
890 * to which it will write.
891 */
892 if (vp != NULL) {
893 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
894 *mpp = NULL;
895 if (error != EOPNOTSUPP)
896 return (error);
897 return (0);
898 }
899 }
900 if ((mp = *mpp) == NULL)
901 return (0);
902 MNT_ILOCK(mp);
903 if (vp == NULL)
904 MNT_REF(mp);
905 /*
906 * Check on status of suspension.
907 */
908 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
909 if (flags & V_NOWAIT) {
910 error = EWOULDBLOCK;
911 goto unlock;
912 }
913 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
914 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
915 if (error)
916 goto unlock;
917 }
918 if (flags & V_XSLEEP)
919 goto unlock;
920 mp->mnt_writeopcount++;
921 unlock:
922 MNT_REL(mp);
923 MNT_IUNLOCK(mp);
924 return (error);
925 }
926
927 /*
928 * Secondary suspension. Used by operations such as vop_inactive
929 * routines that are needed by the higher level functions. These
930 * are allowed to proceed until all the higher level functions have
931 * completed (indicated by mnt_writeopcount dropping to zero). At that
932 * time, these operations are halted until the suspension is over.
933 */
934 int
935 vn_write_suspend_wait(vp, mp, flags)
936 struct vnode *vp;
937 struct mount *mp;
938 int flags;
939 {
940 int error;
941
942 if (vp != NULL) {
943 if ((error = VOP_GETWRITEMOUNT(vp, &mp)) != 0) {
944 if (error != EOPNOTSUPP)
945 return (error);
946 return (0);
947 }
948 }
949 /*
950 * If we are not suspended or have not yet reached suspended
951 * mode, then let the operation proceed.
952 */
953 if (mp == NULL)
954 return (0);
955 MNT_ILOCK(mp);
956 if (vp == NULL)
957 MNT_REF(mp);
958 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) == 0) {
959 MNT_REL(mp);
960 MNT_IUNLOCK(mp);
961 return (0);
962 }
963 if (flags & V_NOWAIT) {
964 MNT_REL(mp);
965 MNT_IUNLOCK(mp);
966 return (EWOULDBLOCK);
967 }
968 /*
969 * Wait for the suspension to finish.
970 */
971 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
972 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
973 vfs_rel(mp);
974 return (error);
975 }
976
977 /*
978 * Secondary suspension. Used by operations such as vop_inactive
979 * routines that are needed by the higher level functions. These
980 * are allowed to proceed until all the higher level functions have
981 * completed (indicated by mnt_writeopcount dropping to zero). At that
982 * time, these operations are halted until the suspension is over.
983 */
984 int
985 vn_start_secondary_write(vp, mpp, flags)
986 struct vnode *vp;
987 struct mount **mpp;
988 int flags;
989 {
990 struct mount *mp;
991 int error;
992
993 retry:
994 if (vp != NULL) {
995 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
996 *mpp = NULL;
997 if (error != EOPNOTSUPP)
998 return (error);
999 return (0);
1000 }
1001 }
1002 /*
1003 * If we are not suspended or have not yet reached suspended
1004 * mode, then let the operation proceed.
1005 */
1006 if ((mp = *mpp) == NULL)
1007 return (0);
1008 MNT_ILOCK(mp);
1009 if (vp == NULL)
1010 MNT_REF(mp);
1011 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1012 mp->mnt_secondary_writes++;
1013 mp->mnt_secondary_accwrites++;
1014 MNT_REL(mp);
1015 MNT_IUNLOCK(mp);
1016 return (0);
1017 }
1018 if (flags & V_NOWAIT) {
1019 MNT_REL(mp);
1020 MNT_IUNLOCK(mp);
1021 return (EWOULDBLOCK);
1022 }
1023 /*
1024 * Wait for the suspension to finish.
1025 */
1026 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1027 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1028 vfs_rel(mp);
1029 if (error == 0)
1030 goto retry;
1031 return (error);
1032 }
1033
1034 /*
1035 * Filesystem write operation has completed. If we are suspending and this
1036 * operation is the last one, notify the suspender that the suspension is
1037 * now in effect.
1038 */
1039 void
1040 vn_finished_write(mp)
1041 struct mount *mp;
1042 {
1043 if (mp == NULL)
1044 return;
1045 MNT_ILOCK(mp);
1046 mp->mnt_writeopcount--;
1047 if (mp->mnt_writeopcount < 0)
1048 panic("vn_finished_write: neg cnt");
1049 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1050 mp->mnt_writeopcount <= 0)
1051 wakeup(&mp->mnt_writeopcount);
1052 MNT_IUNLOCK(mp);
1053 }
1054
1055
1056 /*
1057 * Filesystem secondary write operation has completed. If we are
1058 * suspending and this operation is the last one, notify the suspender
1059 * that the suspension is now in effect.
1060 */
1061 void
1062 vn_finished_secondary_write(mp)
1063 struct mount *mp;
1064 {
1065 if (mp == NULL)
1066 return;
1067 MNT_ILOCK(mp);
1068 mp->mnt_secondary_writes--;
1069 if (mp->mnt_secondary_writes < 0)
1070 panic("vn_finished_secondary_write: neg cnt");
1071 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1072 mp->mnt_secondary_writes <= 0)
1073 wakeup(&mp->mnt_secondary_writes);
1074 MNT_IUNLOCK(mp);
1075 }
1076
1077
1078
1079 /*
1080 * Request a filesystem to suspend write operations.
1081 */
1082 int
1083 vfs_write_suspend(mp)
1084 struct mount *mp;
1085 {
1086 struct thread *td = curthread;
1087 int error;
1088
1089 MNT_ILOCK(mp);
1090 if (mp->mnt_kern_flag & MNTK_SUSPEND) {
1091 MNT_IUNLOCK(mp);
1092 return (0);
1093 }
1094 mp->mnt_kern_flag |= MNTK_SUSPEND;
1095 if (mp->mnt_writeopcount > 0)
1096 (void) msleep(&mp->mnt_writeopcount,
1097 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1098 else
1099 MNT_IUNLOCK(mp);
1100 if ((error = VFS_SYNC(mp, MNT_SUSPEND, td)) != 0)
1101 vfs_write_resume(mp);
1102 return (error);
1103 }
1104
1105 /*
1106 * Request a filesystem to resume write operations.
1107 */
1108 void
1109 vfs_write_resume(mp)
1110 struct mount *mp;
1111 {
1112
1113 MNT_ILOCK(mp);
1114 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1115 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1116 MNTK_SUSPENDED);
1117 wakeup(&mp->mnt_writeopcount);
1118 wakeup(&mp->mnt_flag);
1119 }
1120 MNT_IUNLOCK(mp);
1121 }
1122
1123 /*
1124 * Implement kqueues for files by translating it to vnode operation.
1125 */
1126 static int
1127 vn_kqfilter(struct file *fp, struct knote *kn)
1128 {
1129 int vfslocked;
1130 int error;
1131
1132 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1133 error = VOP_KQFILTER(fp->f_vnode, kn);
1134 VFS_UNLOCK_GIANT(vfslocked);
1135
1136 return error;
1137 }
1138
1139 /*
1140 * Simplified in-kernel wrapper calls for extended attribute access.
1141 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1142 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1143 */
1144 int
1145 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1146 const char *attrname, int *buflen, char *buf, struct thread *td)
1147 {
1148 struct uio auio;
1149 struct iovec iov;
1150 int error;
1151
1152 iov.iov_len = *buflen;
1153 iov.iov_base = buf;
1154
1155 auio.uio_iov = &iov;
1156 auio.uio_iovcnt = 1;
1157 auio.uio_rw = UIO_READ;
1158 auio.uio_segflg = UIO_SYSSPACE;
1159 auio.uio_td = td;
1160 auio.uio_offset = 0;
1161 auio.uio_resid = *buflen;
1162
1163 if ((ioflg & IO_NODELOCKED) == 0)
1164 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1165
1166 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1167
1168 /* authorize attribute retrieval as kernel */
1169 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1170 td);
1171
1172 if ((ioflg & IO_NODELOCKED) == 0)
1173 VOP_UNLOCK(vp, 0, td);
1174
1175 if (error == 0) {
1176 *buflen = *buflen - auio.uio_resid;
1177 }
1178
1179 return (error);
1180 }
1181
1182 /*
1183 * XXX failure mode if partially written?
1184 */
1185 int
1186 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1187 const char *attrname, int buflen, char *buf, struct thread *td)
1188 {
1189 struct uio auio;
1190 struct iovec iov;
1191 struct mount *mp;
1192 int error;
1193
1194 iov.iov_len = buflen;
1195 iov.iov_base = buf;
1196
1197 auio.uio_iov = &iov;
1198 auio.uio_iovcnt = 1;
1199 auio.uio_rw = UIO_WRITE;
1200 auio.uio_segflg = UIO_SYSSPACE;
1201 auio.uio_td = td;
1202 auio.uio_offset = 0;
1203 auio.uio_resid = buflen;
1204
1205 if ((ioflg & IO_NODELOCKED) == 0) {
1206 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1207 return (error);
1208 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1209 }
1210
1211 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1212
1213 /* authorize attribute setting as kernel */
1214 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1215
1216 if ((ioflg & IO_NODELOCKED) == 0) {
1217 vn_finished_write(mp);
1218 VOP_UNLOCK(vp, 0, td);
1219 }
1220
1221 return (error);
1222 }
1223
1224 int
1225 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1226 const char *attrname, struct thread *td)
1227 {
1228 struct mount *mp;
1229 int error;
1230
1231 if ((ioflg & IO_NODELOCKED) == 0) {
1232 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1233 return (error);
1234 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1235 }
1236
1237 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1238
1239 /* authorize attribute removal as kernel */
1240 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1241 if (error == EOPNOTSUPP)
1242 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1243 NULL, td);
1244
1245 if ((ioflg & IO_NODELOCKED) == 0) {
1246 vn_finished_write(mp);
1247 VOP_UNLOCK(vp, 0, td);
1248 }
1249
1250 return (error);
1251 }
Cache object: 24173f44d974e892e85af2289010beb9
|