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_SHARED | 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
645 /*
646 * Initialize defaults for new and unusual fields, so that file
647 * systems which don't support these fields don't need to know
648 * about them.
649 */
650 vap->va_birthtime.tv_sec = -1;
651 vap->va_birthtime.tv_nsec = 0;
652 vap->va_fsid = VNOVAL;
653 vap->va_rdev = NODEV;
654
655 error = VOP_GETATTR(vp, vap, active_cred, td);
656 if (error)
657 return (error);
658
659 /*
660 * Zero the spare stat fields
661 */
662 bzero(sb, sizeof *sb);
663
664 /*
665 * Copy from vattr table
666 */
667 if (vap->va_fsid != VNOVAL)
668 sb->st_dev = vap->va_fsid;
669 else
670 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
671 sb->st_ino = vap->va_fileid;
672 mode = vap->va_mode;
673 switch (vap->va_type) {
674 case VREG:
675 mode |= S_IFREG;
676 break;
677 case VDIR:
678 mode |= S_IFDIR;
679 break;
680 case VBLK:
681 mode |= S_IFBLK;
682 break;
683 case VCHR:
684 mode |= S_IFCHR;
685 break;
686 case VLNK:
687 mode |= S_IFLNK;
688 break;
689 case VSOCK:
690 mode |= S_IFSOCK;
691 break;
692 case VFIFO:
693 mode |= S_IFIFO;
694 break;
695 default:
696 return (EBADF);
697 };
698 sb->st_mode = mode;
699 sb->st_nlink = vap->va_nlink;
700 sb->st_uid = vap->va_uid;
701 sb->st_gid = vap->va_gid;
702 sb->st_rdev = vap->va_rdev;
703 if (vap->va_size > OFF_MAX)
704 return (EOVERFLOW);
705 sb->st_size = vap->va_size;
706 sb->st_atimespec = vap->va_atime;
707 sb->st_mtimespec = vap->va_mtime;
708 sb->st_ctimespec = vap->va_ctime;
709 sb->st_birthtimespec = vap->va_birthtime;
710
711 /*
712 * According to www.opengroup.org, the meaning of st_blksize is
713 * "a filesystem-specific preferred I/O block size for this
714 * object. In some filesystem types, this may vary from file
715 * to file"
716 * Default to PAGE_SIZE after much discussion.
717 * XXX: min(PAGE_SIZE, vp->v_bufobj.bo_bsize) may be more correct.
718 */
719
720 sb->st_blksize = PAGE_SIZE;
721
722 sb->st_flags = vap->va_flags;
723 if (priv_check(td, PRIV_VFS_GENERATION))
724 sb->st_gen = 0;
725 else
726 sb->st_gen = vap->va_gen;
727
728 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
729 return (0);
730 }
731
732 /*
733 * File table vnode ioctl routine.
734 */
735 static int
736 vn_ioctl(fp, com, data, active_cred, td)
737 struct file *fp;
738 u_long com;
739 void *data;
740 struct ucred *active_cred;
741 struct thread *td;
742 {
743 struct vnode *vp = fp->f_vnode;
744 struct vattr vattr;
745 int vfslocked;
746 int error;
747
748 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
749 error = ENOTTY;
750 switch (vp->v_type) {
751 case VREG:
752 case VDIR:
753 if (com == FIONREAD) {
754 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
755 error = VOP_GETATTR(vp, &vattr, active_cred, td);
756 VOP_UNLOCK(vp, 0, td);
757 if (!error)
758 *(int *)data = vattr.va_size - fp->f_offset;
759 }
760 if (com == FIONBIO || com == FIOASYNC) /* XXX */
761 error = 0;
762 else
763 error = VOP_IOCTL(vp, com, data, fp->f_flag,
764 active_cred, td);
765 break;
766
767 default:
768 break;
769 }
770 VFS_UNLOCK_GIANT(vfslocked);
771 return (error);
772 }
773
774 /*
775 * File table vnode poll routine.
776 */
777 static int
778 vn_poll(fp, events, active_cred, td)
779 struct file *fp;
780 int events;
781 struct ucred *active_cred;
782 struct thread *td;
783 {
784 struct vnode *vp;
785 int vfslocked;
786 int error;
787
788 vp = fp->f_vnode;
789 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
790 #ifdef MAC
791 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
792 error = mac_check_vnode_poll(active_cred, fp->f_cred, vp);
793 VOP_UNLOCK(vp, 0, td);
794 if (!error)
795 #endif
796
797 error = VOP_POLL(vp, events, fp->f_cred, td);
798 VFS_UNLOCK_GIANT(vfslocked);
799 return (error);
800 }
801
802 /*
803 * Check that the vnode is still valid, and if so
804 * acquire requested lock.
805 */
806 int
807 _vn_lock(struct vnode *vp, int flags, struct thread *td, char *file, int line)
808 {
809 int error;
810
811 do {
812 if ((flags & LK_INTERLOCK) == 0)
813 VI_LOCK(vp);
814 #ifdef DEBUG_VFS_LOCKS
815 KASSERT(vp->v_holdcnt != 0,
816 ("vn_lock %p: zero hold count", vp));
817 #endif
818 if ((flags & LK_NOWAIT || (flags & LK_TYPE_MASK) == 0) &&
819 vp->v_iflag & VI_DOOMED) {
820 VI_UNLOCK(vp);
821 return (ENOENT);
822 }
823 /*
824 * Just polling to check validity.
825 */
826 if ((flags & LK_TYPE_MASK) == 0) {
827 VI_UNLOCK(vp);
828 return (0);
829 }
830 /*
831 * lockmgr drops interlock before it will return for
832 * any reason. So force the code above to relock it.
833 */
834 error = VOP_LOCK1(vp, flags | LK_INTERLOCK, td, file, line);
835 flags &= ~LK_INTERLOCK;
836 KASSERT((flags & LK_RETRY) == 0 || error == 0,
837 ("LK_RETRY set with incompatible flags %d\n", flags));
838 /*
839 * Callers specify LK_RETRY if they wish to get dead vnodes.
840 * If RETRY is not set, we return ENOENT instead.
841 */
842 if (error == 0 && vp->v_iflag & VI_DOOMED &&
843 (flags & LK_RETRY) == 0) {
844 VOP_UNLOCK(vp, 0, td);
845 error = ENOENT;
846 break;
847 }
848 } while (flags & LK_RETRY && error != 0);
849 return (error);
850 }
851
852 /*
853 * File table vnode close routine.
854 */
855 static int
856 vn_closefile(fp, td)
857 struct file *fp;
858 struct thread *td;
859 {
860 struct vnode *vp;
861 struct flock lf;
862 int vfslocked;
863 int error;
864
865 vp = fp->f_vnode;
866
867 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
868 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
869 lf.l_whence = SEEK_SET;
870 lf.l_start = 0;
871 lf.l_len = 0;
872 lf.l_type = F_UNLCK;
873 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
874 }
875
876 fp->f_ops = &badfileops;
877
878 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
879 VFS_UNLOCK_GIANT(vfslocked);
880 return (error);
881 }
882
883 /*
884 * Preparing to start a filesystem write operation. If the operation is
885 * permitted, then we bump the count of operations in progress and
886 * proceed. If a suspend request is in progress, we wait until the
887 * suspension is over, and then proceed.
888 */
889 int
890 vn_start_write(vp, mpp, flags)
891 struct vnode *vp;
892 struct mount **mpp;
893 int flags;
894 {
895 struct mount *mp;
896 int error;
897
898 error = 0;
899 /*
900 * If a vnode is provided, get and return the mount point that
901 * to which it will write.
902 */
903 if (vp != NULL) {
904 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
905 *mpp = NULL;
906 if (error != EOPNOTSUPP)
907 return (error);
908 return (0);
909 }
910 }
911 if ((mp = *mpp) == NULL)
912 return (0);
913 MNT_ILOCK(mp);
914 if (vp == NULL)
915 MNT_REF(mp);
916 /*
917 * Check on status of suspension.
918 */
919 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
920 mp->mnt_susp_owner != curthread) {
921 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
922 if (flags & V_NOWAIT) {
923 error = EWOULDBLOCK;
924 goto unlock;
925 }
926 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
927 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
928 if (error)
929 goto unlock;
930 }
931 }
932 if (flags & V_XSLEEP)
933 goto unlock;
934 mp->mnt_writeopcount++;
935 unlock:
936 MNT_REL(mp);
937 MNT_IUNLOCK(mp);
938 return (error);
939 }
940
941 /*
942 * Secondary suspension. Used by operations such as vop_inactive
943 * routines that are needed by the higher level functions. These
944 * are allowed to proceed until all the higher level functions have
945 * completed (indicated by mnt_writeopcount dropping to zero). At that
946 * time, these operations are halted until the suspension is over.
947 */
948 int
949 vn_start_secondary_write(vp, mpp, flags)
950 struct vnode *vp;
951 struct mount **mpp;
952 int flags;
953 {
954 struct mount *mp;
955 int error;
956
957 retry:
958 if (vp != NULL) {
959 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
960 *mpp = NULL;
961 if (error != EOPNOTSUPP)
962 return (error);
963 return (0);
964 }
965 }
966 /*
967 * If we are not suspended or have not yet reached suspended
968 * mode, then let the operation proceed.
969 */
970 if ((mp = *mpp) == NULL)
971 return (0);
972 MNT_ILOCK(mp);
973 if (vp == NULL)
974 MNT_REF(mp);
975 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
976 mp->mnt_secondary_writes++;
977 mp->mnt_secondary_accwrites++;
978 MNT_REL(mp);
979 MNT_IUNLOCK(mp);
980 return (0);
981 }
982 if (flags & V_NOWAIT) {
983 MNT_REL(mp);
984 MNT_IUNLOCK(mp);
985 return (EWOULDBLOCK);
986 }
987 /*
988 * Wait for the suspension to finish.
989 */
990 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
991 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
992 vfs_rel(mp);
993 if (error == 0)
994 goto retry;
995 return (error);
996 }
997
998 /*
999 * Filesystem write operation has completed. If we are suspending and this
1000 * operation is the last one, notify the suspender that the suspension is
1001 * now in effect.
1002 */
1003 void
1004 vn_finished_write(mp)
1005 struct mount *mp;
1006 {
1007 if (mp == NULL)
1008 return;
1009 MNT_ILOCK(mp);
1010 mp->mnt_writeopcount--;
1011 if (mp->mnt_writeopcount < 0)
1012 panic("vn_finished_write: neg cnt");
1013 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1014 mp->mnt_writeopcount <= 0)
1015 wakeup(&mp->mnt_writeopcount);
1016 MNT_IUNLOCK(mp);
1017 }
1018
1019
1020 /*
1021 * Filesystem secondary write operation has completed. If we are
1022 * suspending and this operation is the last one, notify the suspender
1023 * that the suspension is now in effect.
1024 */
1025 void
1026 vn_finished_secondary_write(mp)
1027 struct mount *mp;
1028 {
1029 if (mp == NULL)
1030 return;
1031 MNT_ILOCK(mp);
1032 mp->mnt_secondary_writes--;
1033 if (mp->mnt_secondary_writes < 0)
1034 panic("vn_finished_secondary_write: neg cnt");
1035 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1036 mp->mnt_secondary_writes <= 0)
1037 wakeup(&mp->mnt_secondary_writes);
1038 MNT_IUNLOCK(mp);
1039 }
1040
1041
1042
1043 /*
1044 * Request a filesystem to suspend write operations.
1045 */
1046 int
1047 vfs_write_suspend(mp)
1048 struct mount *mp;
1049 {
1050 struct thread *td = curthread;
1051 int error;
1052
1053 MNT_ILOCK(mp);
1054 if (mp->mnt_susp_owner == curthread) {
1055 MNT_IUNLOCK(mp);
1056 return (EALREADY);
1057 }
1058 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1059 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1060 mp->mnt_kern_flag |= MNTK_SUSPEND;
1061 mp->mnt_susp_owner = curthread;
1062 if (mp->mnt_writeopcount > 0)
1063 (void) msleep(&mp->mnt_writeopcount,
1064 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1065 else
1066 MNT_IUNLOCK(mp);
1067 if ((error = VFS_SYNC(mp, MNT_SUSPEND, td)) != 0)
1068 vfs_write_resume(mp);
1069 return (error);
1070 }
1071
1072 /*
1073 * Request a filesystem to resume write operations.
1074 */
1075 void
1076 vfs_write_resume(mp)
1077 struct mount *mp;
1078 {
1079
1080 MNT_ILOCK(mp);
1081 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1082 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1083 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1084 MNTK_SUSPENDED);
1085 mp->mnt_susp_owner = NULL;
1086 wakeup(&mp->mnt_writeopcount);
1087 wakeup(&mp->mnt_flag);
1088 curthread->td_pflags &= ~TDP_IGNSUSP;
1089 MNT_IUNLOCK(mp);
1090 VFS_SUSP_CLEAN(mp);
1091 } else
1092 MNT_IUNLOCK(mp);
1093 }
1094
1095 /*
1096 * Implement kqueues for files by translating it to vnode operation.
1097 */
1098 static int
1099 vn_kqfilter(struct file *fp, struct knote *kn)
1100 {
1101 int vfslocked;
1102 int error;
1103
1104 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1105 error = VOP_KQFILTER(fp->f_vnode, kn);
1106 VFS_UNLOCK_GIANT(vfslocked);
1107
1108 return error;
1109 }
1110
1111 /*
1112 * Simplified in-kernel wrapper calls for extended attribute access.
1113 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1114 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1115 */
1116 int
1117 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1118 const char *attrname, int *buflen, char *buf, struct thread *td)
1119 {
1120 struct uio auio;
1121 struct iovec iov;
1122 int error;
1123
1124 iov.iov_len = *buflen;
1125 iov.iov_base = buf;
1126
1127 auio.uio_iov = &iov;
1128 auio.uio_iovcnt = 1;
1129 auio.uio_rw = UIO_READ;
1130 auio.uio_segflg = UIO_SYSSPACE;
1131 auio.uio_td = td;
1132 auio.uio_offset = 0;
1133 auio.uio_resid = *buflen;
1134
1135 if ((ioflg & IO_NODELOCKED) == 0)
1136 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1137
1138 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1139
1140 /* authorize attribute retrieval as kernel */
1141 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1142 td);
1143
1144 if ((ioflg & IO_NODELOCKED) == 0)
1145 VOP_UNLOCK(vp, 0, td);
1146
1147 if (error == 0) {
1148 *buflen = *buflen - auio.uio_resid;
1149 }
1150
1151 return (error);
1152 }
1153
1154 /*
1155 * XXX failure mode if partially written?
1156 */
1157 int
1158 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1159 const char *attrname, int buflen, char *buf, struct thread *td)
1160 {
1161 struct uio auio;
1162 struct iovec iov;
1163 struct mount *mp;
1164 int error;
1165
1166 iov.iov_len = buflen;
1167 iov.iov_base = buf;
1168
1169 auio.uio_iov = &iov;
1170 auio.uio_iovcnt = 1;
1171 auio.uio_rw = UIO_WRITE;
1172 auio.uio_segflg = UIO_SYSSPACE;
1173 auio.uio_td = td;
1174 auio.uio_offset = 0;
1175 auio.uio_resid = buflen;
1176
1177 if ((ioflg & IO_NODELOCKED) == 0) {
1178 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1179 return (error);
1180 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1181 }
1182
1183 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1184
1185 /* authorize attribute setting as kernel */
1186 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1187
1188 if ((ioflg & IO_NODELOCKED) == 0) {
1189 vn_finished_write(mp);
1190 VOP_UNLOCK(vp, 0, td);
1191 }
1192
1193 return (error);
1194 }
1195
1196 int
1197 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1198 const char *attrname, struct thread *td)
1199 {
1200 struct mount *mp;
1201 int error;
1202
1203 if ((ioflg & IO_NODELOCKED) == 0) {
1204 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1205 return (error);
1206 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1207 }
1208
1209 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1210
1211 /* authorize attribute removal as kernel */
1212 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1213 if (error == EOPNOTSUPP)
1214 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1215 NULL, td);
1216
1217 if ((ioflg & IO_NODELOCKED) == 0) {
1218 vn_finished_write(mp);
1219 VOP_UNLOCK(vp, 0, td);
1220 }
1221
1222 return (error);
1223 }
1224
1225 int
1226 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1227 {
1228 struct mount *mp;
1229 int ltype, error;
1230
1231 mp = vp->v_mount;
1232 ltype = VOP_ISLOCKED(vp, curthread);
1233 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1234 ("vn_vget_ino: vp not locked"));
1235 for (;;) {
1236 error = vfs_busy(mp, LK_NOWAIT, NULL, curthread);
1237 if (error == 0)
1238 break;
1239 VOP_UNLOCK(vp, 0, curthread);
1240 pause("vn_vget", 1);
1241 vn_lock(vp, ltype | LK_RETRY, curthread);
1242 if (vp->v_iflag & VI_DOOMED)
1243 return (ENOENT);
1244 }
1245 VOP_UNLOCK(vp, 0, curthread);
1246 error = VFS_VGET(mp, ino, lkflags, rvp);
1247 vfs_unbusy(mp, curthread);
1248 vn_lock(vp, ltype | LK_RETRY, curthread);
1249 if (vp->v_iflag & VI_DOOMED) {
1250 if (error == 0)
1251 vput(*rvp);
1252 error = ENOENT;
1253 }
1254 return (error);
1255 }
Cache object: 40fbb9ca042b32023902fc0d3644be00
|