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 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013 The FreeBSD Foundation
12 *
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 *
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
41 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD: releng/9.2/sys/kern/vfs_vnops.c 251762 2013-06-14 21:56:10Z jhb $");
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/kdb.h>
51 #include <sys/stat.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/limits.h>
55 #include <sys/lock.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
59 #include <sys/vnode.h>
60 #include <sys/bio.h>
61 #include <sys/buf.h>
62 #include <sys/filio.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/ttycom.h>
67 #include <sys/conf.h>
68 #include <sys/syslog.h>
69 #include <sys/unistd.h>
70
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_extern.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80
81 static fo_rdwr_t vn_read;
82 static fo_rdwr_t vn_write;
83 static fo_rdwr_t vn_io_fault;
84 static fo_truncate_t vn_truncate;
85 static fo_ioctl_t vn_ioctl;
86 static fo_poll_t vn_poll;
87 static fo_kqfilter_t vn_kqfilter;
88 static fo_stat_t vn_statfile;
89 static fo_close_t vn_closefile;
90
91 struct fileops vnops = {
92 .fo_read = vn_io_fault,
93 .fo_write = vn_io_fault,
94 .fo_truncate = vn_truncate,
95 .fo_ioctl = vn_ioctl,
96 .fo_poll = vn_poll,
97 .fo_kqfilter = vn_kqfilter,
98 .fo_stat = vn_statfile,
99 .fo_close = vn_closefile,
100 .fo_chmod = vn_chmod,
101 .fo_chown = vn_chown,
102 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
103 };
104
105 int
106 vn_open(ndp, flagp, cmode, fp)
107 struct nameidata *ndp;
108 int *flagp, cmode;
109 struct file *fp;
110 {
111 struct thread *td = ndp->ni_cnd.cn_thread;
112
113 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
114 }
115
116 /*
117 * Common code for vnode open operations.
118 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
119 *
120 * Note that this does NOT free nameidata for the successful case,
121 * due to the NDINIT being done elsewhere.
122 */
123 int
124 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
125 struct ucred *cred, struct file *fp)
126 {
127 struct vnode *vp;
128 struct mount *mp;
129 struct thread *td = ndp->ni_cnd.cn_thread;
130 struct vattr vat;
131 struct vattr *vap = &vat;
132 int fmode, error;
133 accmode_t accmode;
134 int vfslocked, mpsafe;
135
136 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
137 restart:
138 vfslocked = 0;
139 fmode = *flagp;
140 if (fmode & O_CREAT) {
141 ndp->ni_cnd.cn_nameiop = CREATE;
142 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
143 MPSAFE;
144 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
145 ndp->ni_cnd.cn_flags |= FOLLOW;
146 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
147 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
148 bwillwrite();
149 if ((error = namei(ndp)) != 0)
150 return (error);
151 vfslocked = NDHASGIANT(ndp);
152 if (!mpsafe)
153 ndp->ni_cnd.cn_flags &= ~MPSAFE;
154 if (ndp->ni_vp == NULL) {
155 VATTR_NULL(vap);
156 vap->va_type = VREG;
157 vap->va_mode = cmode;
158 if (fmode & O_EXCL)
159 vap->va_vaflags |= VA_EXCLUSIVE;
160 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
161 NDFREE(ndp, NDF_ONLY_PNBUF);
162 vput(ndp->ni_dvp);
163 VFS_UNLOCK_GIANT(vfslocked);
164 if ((error = vn_start_write(NULL, &mp,
165 V_XSLEEP | PCATCH)) != 0)
166 return (error);
167 goto restart;
168 }
169 #ifdef MAC
170 error = mac_vnode_check_create(cred, ndp->ni_dvp,
171 &ndp->ni_cnd, vap);
172 if (error == 0)
173 #endif
174 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
175 &ndp->ni_cnd, vap);
176 vput(ndp->ni_dvp);
177 vn_finished_write(mp);
178 if (error) {
179 VFS_UNLOCK_GIANT(vfslocked);
180 NDFREE(ndp, NDF_ONLY_PNBUF);
181 return (error);
182 }
183 fmode &= ~O_TRUNC;
184 vp = ndp->ni_vp;
185 } else {
186 if (ndp->ni_dvp == ndp->ni_vp)
187 vrele(ndp->ni_dvp);
188 else
189 vput(ndp->ni_dvp);
190 ndp->ni_dvp = NULL;
191 vp = ndp->ni_vp;
192 if (fmode & O_EXCL) {
193 error = EEXIST;
194 goto bad;
195 }
196 fmode &= ~O_CREAT;
197 }
198 } else {
199 ndp->ni_cnd.cn_nameiop = LOOKUP;
200 ndp->ni_cnd.cn_flags = ISOPEN |
201 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
202 LOCKLEAF | MPSAFE;
203 if (!(fmode & FWRITE))
204 ndp->ni_cnd.cn_flags |= LOCKSHARED;
205 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
206 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
207 if ((error = namei(ndp)) != 0)
208 return (error);
209 if (!mpsafe)
210 ndp->ni_cnd.cn_flags &= ~MPSAFE;
211 vfslocked = NDHASGIANT(ndp);
212 vp = ndp->ni_vp;
213 }
214 if (vp->v_type == VLNK) {
215 error = EMLINK;
216 goto bad;
217 }
218 if (vp->v_type == VSOCK) {
219 error = EOPNOTSUPP;
220 goto bad;
221 }
222 if (vp->v_type != VDIR && fmode & O_DIRECTORY) {
223 error = ENOTDIR;
224 goto bad;
225 }
226 accmode = 0;
227 if (fmode & (FWRITE | O_TRUNC)) {
228 if (vp->v_type == VDIR) {
229 error = EISDIR;
230 goto bad;
231 }
232 accmode |= VWRITE;
233 }
234 if (fmode & FREAD)
235 accmode |= VREAD;
236 if (fmode & FEXEC)
237 accmode |= VEXEC;
238 if ((fmode & O_APPEND) && (fmode & FWRITE))
239 accmode |= VAPPEND;
240 #ifdef MAC
241 error = mac_vnode_check_open(cred, vp, accmode);
242 if (error)
243 goto bad;
244 #endif
245 if ((fmode & O_CREAT) == 0) {
246 if (accmode & VWRITE) {
247 error = vn_writechk(vp);
248 if (error)
249 goto bad;
250 }
251 if (accmode) {
252 error = VOP_ACCESS(vp, accmode, cred, td);
253 if (error)
254 goto bad;
255 }
256 }
257 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
258 goto bad;
259
260 if (fmode & FWRITE)
261 VOP_ADD_WRITECOUNT(vp, 1);
262 *flagp = fmode;
263 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
264 if (!mpsafe)
265 VFS_UNLOCK_GIANT(vfslocked);
266 return (0);
267 bad:
268 NDFREE(ndp, NDF_ONLY_PNBUF);
269 vput(vp);
270 VFS_UNLOCK_GIANT(vfslocked);
271 *flagp = fmode;
272 ndp->ni_vp = NULL;
273 return (error);
274 }
275
276 /*
277 * Check for write permissions on the specified vnode.
278 * Prototype text segments cannot be written.
279 */
280 int
281 vn_writechk(vp)
282 register struct vnode *vp;
283 {
284
285 ASSERT_VOP_LOCKED(vp, "vn_writechk");
286 /*
287 * If there's shared text associated with
288 * the vnode, try to free it up once. If
289 * we fail, we can't allow writing.
290 */
291 if (VOP_IS_TEXT(vp))
292 return (ETXTBSY);
293
294 return (0);
295 }
296
297 /*
298 * Vnode close call
299 */
300 int
301 vn_close(vp, flags, file_cred, td)
302 register struct vnode *vp;
303 int flags;
304 struct ucred *file_cred;
305 struct thread *td;
306 {
307 struct mount *mp;
308 int error, lock_flags;
309
310 if (!(flags & FWRITE) && vp->v_mount != NULL &&
311 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
312 lock_flags = LK_SHARED;
313 else
314 lock_flags = LK_EXCLUSIVE;
315
316 VFS_ASSERT_GIANT(vp->v_mount);
317
318 vn_start_write(vp, &mp, V_WAIT);
319 vn_lock(vp, lock_flags | LK_RETRY);
320 if (flags & FWRITE) {
321 VNASSERT(vp->v_writecount > 0, vp,
322 ("vn_close: negative writecount"));
323 VOP_ADD_WRITECOUNT(vp, -1);
324 }
325 error = VOP_CLOSE(vp, flags, file_cred, td);
326 vput(vp);
327 vn_finished_write(mp);
328 return (error);
329 }
330
331 /*
332 * Heuristic to detect sequential operation.
333 */
334 static int
335 sequential_heuristic(struct uio *uio, struct file *fp)
336 {
337
338 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
339 return (fp->f_seqcount << IO_SEQSHIFT);
340
341 /*
342 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
343 * that the first I/O is normally considered to be slightly
344 * sequential. Seeking to offset 0 doesn't change sequentiality
345 * unless previous seeks have reduced f_seqcount to 0, in which
346 * case offset 0 is not special.
347 */
348 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
349 uio->uio_offset == fp->f_nextoff) {
350 /*
351 * f_seqcount is in units of fixed-size blocks so that it
352 * depends mainly on the amount of sequential I/O and not
353 * much on the number of sequential I/O's. The fixed size
354 * of 16384 is hard-coded here since it is (not quite) just
355 * a magic size that works well here. This size is more
356 * closely related to the best I/O size for real disks than
357 * to any block size used by software.
358 */
359 fp->f_seqcount += howmany(uio->uio_resid, 16384);
360 if (fp->f_seqcount > IO_SEQMAX)
361 fp->f_seqcount = IO_SEQMAX;
362 return (fp->f_seqcount << IO_SEQSHIFT);
363 }
364
365 /* Not sequential. Quickly draw-down sequentiality. */
366 if (fp->f_seqcount > 1)
367 fp->f_seqcount = 1;
368 else
369 fp->f_seqcount = 0;
370 return (0);
371 }
372
373 /*
374 * Package up an I/O request on a vnode into a uio and do it.
375 */
376 int
377 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
378 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
379 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
380 {
381 struct uio auio;
382 struct iovec aiov;
383 struct mount *mp;
384 struct ucred *cred;
385 void *rl_cookie;
386 int error, lock_flags;
387
388 VFS_ASSERT_GIANT(vp->v_mount);
389
390 auio.uio_iov = &aiov;
391 auio.uio_iovcnt = 1;
392 aiov.iov_base = base;
393 aiov.iov_len = len;
394 auio.uio_resid = len;
395 auio.uio_offset = offset;
396 auio.uio_segflg = segflg;
397 auio.uio_rw = rw;
398 auio.uio_td = td;
399 error = 0;
400
401 if ((ioflg & IO_NODELOCKED) == 0) {
402 if (rw == UIO_READ) {
403 rl_cookie = vn_rangelock_rlock(vp, offset,
404 offset + len);
405 } else {
406 rl_cookie = vn_rangelock_wlock(vp, offset,
407 offset + len);
408 }
409 mp = NULL;
410 if (rw == UIO_WRITE) {
411 if (vp->v_type != VCHR &&
412 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
413 != 0)
414 goto out;
415 if (MNT_SHARED_WRITES(mp) ||
416 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
417 lock_flags = LK_SHARED;
418 else
419 lock_flags = LK_EXCLUSIVE;
420 } else
421 lock_flags = LK_SHARED;
422 vn_lock(vp, lock_flags | LK_RETRY);
423 } else
424 rl_cookie = NULL;
425
426 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
427 #ifdef MAC
428 if ((ioflg & IO_NOMACCHECK) == 0) {
429 if (rw == UIO_READ)
430 error = mac_vnode_check_read(active_cred, file_cred,
431 vp);
432 else
433 error = mac_vnode_check_write(active_cred, file_cred,
434 vp);
435 }
436 #endif
437 if (error == 0) {
438 if (file_cred != NULL)
439 cred = file_cred;
440 else
441 cred = active_cred;
442 if (rw == UIO_READ)
443 error = VOP_READ(vp, &auio, ioflg, cred);
444 else
445 error = VOP_WRITE(vp, &auio, ioflg, cred);
446 }
447 if (aresid)
448 *aresid = auio.uio_resid;
449 else
450 if (auio.uio_resid && error == 0)
451 error = EIO;
452 if ((ioflg & IO_NODELOCKED) == 0) {
453 VOP_UNLOCK(vp, 0);
454 if (mp != NULL)
455 vn_finished_write(mp);
456 }
457 out:
458 if (rl_cookie != NULL)
459 vn_rangelock_unlock(vp, rl_cookie);
460 return (error);
461 }
462
463 /*
464 * Package up an I/O request on a vnode into a uio and do it. The I/O
465 * request is split up into smaller chunks and we try to avoid saturating
466 * the buffer cache while potentially holding a vnode locked, so we
467 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
468 * to give other processes a chance to lock the vnode (either other processes
469 * core'ing the same binary, or unrelated processes scanning the directory).
470 */
471 int
472 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
473 file_cred, aresid, td)
474 enum uio_rw rw;
475 struct vnode *vp;
476 void *base;
477 size_t len;
478 off_t offset;
479 enum uio_seg segflg;
480 int ioflg;
481 struct ucred *active_cred;
482 struct ucred *file_cred;
483 size_t *aresid;
484 struct thread *td;
485 {
486 int error = 0;
487 ssize_t iaresid;
488
489 VFS_ASSERT_GIANT(vp->v_mount);
490
491 do {
492 int chunk;
493
494 /*
495 * Force `offset' to a multiple of MAXBSIZE except possibly
496 * for the first chunk, so that filesystems only need to
497 * write full blocks except possibly for the first and last
498 * chunks.
499 */
500 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
501
502 if (chunk > len)
503 chunk = len;
504 if (rw != UIO_READ && vp->v_type == VREG)
505 bwillwrite();
506 iaresid = 0;
507 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
508 ioflg, active_cred, file_cred, &iaresid, td);
509 len -= chunk; /* aresid calc already includes length */
510 if (error)
511 break;
512 offset += chunk;
513 base = (char *)base + chunk;
514 kern_yield(PRI_USER);
515 } while (len);
516 if (aresid)
517 *aresid = len + iaresid;
518 return (error);
519 }
520
521 off_t
522 foffset_lock(struct file *fp, int flags)
523 {
524 struct mtx *mtxp;
525 off_t res;
526
527 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
528
529 #if OFF_MAX <= LONG_MAX
530 /*
531 * Caller only wants the current f_offset value. Assume that
532 * the long and shorter integer types reads are atomic.
533 */
534 if ((flags & FOF_NOLOCK) != 0)
535 return (fp->f_offset);
536 #endif
537
538 /*
539 * According to McKusick the vn lock was protecting f_offset here.
540 * It is now protected by the FOFFSET_LOCKED flag.
541 */
542 mtxp = mtx_pool_find(mtxpool_sleep, fp);
543 mtx_lock(mtxp);
544 if ((flags & FOF_NOLOCK) == 0) {
545 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
546 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
547 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
548 "vofflock", 0);
549 }
550 fp->f_vnread_flags |= FOFFSET_LOCKED;
551 }
552 res = fp->f_offset;
553 mtx_unlock(mtxp);
554 return (res);
555 }
556
557 void
558 foffset_unlock(struct file *fp, off_t val, int flags)
559 {
560 struct mtx *mtxp;
561
562 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
563
564 #if OFF_MAX <= LONG_MAX
565 if ((flags & FOF_NOLOCK) != 0) {
566 if ((flags & FOF_NOUPDATE) == 0)
567 fp->f_offset = val;
568 if ((flags & FOF_NEXTOFF) != 0)
569 fp->f_nextoff = val;
570 return;
571 }
572 #endif
573
574 mtxp = mtx_pool_find(mtxpool_sleep, fp);
575 mtx_lock(mtxp);
576 if ((flags & FOF_NOUPDATE) == 0)
577 fp->f_offset = val;
578 if ((flags & FOF_NEXTOFF) != 0)
579 fp->f_nextoff = val;
580 if ((flags & FOF_NOLOCK) == 0) {
581 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
582 ("Lost FOFFSET_LOCKED"));
583 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
584 wakeup(&fp->f_vnread_flags);
585 fp->f_vnread_flags = 0;
586 }
587 mtx_unlock(mtxp);
588 }
589
590 void
591 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
592 {
593
594 if ((flags & FOF_OFFSET) == 0)
595 uio->uio_offset = foffset_lock(fp, flags);
596 }
597
598 void
599 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
600 {
601
602 if ((flags & FOF_OFFSET) == 0)
603 foffset_unlock(fp, uio->uio_offset, flags);
604 }
605
606 static int
607 get_advice(struct file *fp, struct uio *uio)
608 {
609 struct mtx *mtxp;
610 int ret;
611
612 ret = POSIX_FADV_NORMAL;
613 if (fp->f_advice == NULL)
614 return (ret);
615
616 mtxp = mtx_pool_find(mtxpool_sleep, fp);
617 mtx_lock(mtxp);
618 if (uio->uio_offset >= fp->f_advice->fa_start &&
619 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
620 ret = fp->f_advice->fa_advice;
621 mtx_unlock(mtxp);
622 return (ret);
623 }
624
625 /*
626 * File table vnode read routine.
627 */
628 static int
629 vn_read(fp, uio, active_cred, flags, td)
630 struct file *fp;
631 struct uio *uio;
632 struct ucred *active_cred;
633 int flags;
634 struct thread *td;
635 {
636 struct vnode *vp;
637 struct mtx *mtxp;
638 int error, ioflag;
639 int advice, vfslocked;
640 off_t offset, start, end;
641
642 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
643 uio->uio_td, td));
644 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
645 vp = fp->f_vnode;
646 ioflag = 0;
647 if (fp->f_flag & FNONBLOCK)
648 ioflag |= IO_NDELAY;
649 if (fp->f_flag & O_DIRECT)
650 ioflag |= IO_DIRECT;
651 advice = get_advice(fp, uio);
652 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
653 vn_lock(vp, LK_SHARED | LK_RETRY);
654
655 switch (advice) {
656 case POSIX_FADV_NORMAL:
657 case POSIX_FADV_SEQUENTIAL:
658 case POSIX_FADV_NOREUSE:
659 ioflag |= sequential_heuristic(uio, fp);
660 break;
661 case POSIX_FADV_RANDOM:
662 /* Disable read-ahead for random I/O. */
663 break;
664 }
665 offset = uio->uio_offset;
666
667 #ifdef MAC
668 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
669 if (error == 0)
670 #endif
671 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
672 fp->f_nextoff = uio->uio_offset;
673 VOP_UNLOCK(vp, 0);
674 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
675 offset != uio->uio_offset) {
676 /*
677 * Use POSIX_FADV_DONTNEED to flush clean pages and
678 * buffers for the backing file after a
679 * POSIX_FADV_NOREUSE read(2). To optimize the common
680 * case of using POSIX_FADV_NOREUSE with sequential
681 * access, track the previous implicit DONTNEED
682 * request and grow this request to include the
683 * current read(2) in addition to the previous
684 * DONTNEED. With purely sequential access this will
685 * cause the DONTNEED requests to continously grow to
686 * cover all of the previously read regions of the
687 * file. This allows filesystem blocks that are
688 * accessed by multiple calls to read(2) to be flushed
689 * once the last read(2) finishes.
690 */
691 start = offset;
692 end = uio->uio_offset - 1;
693 mtxp = mtx_pool_find(mtxpool_sleep, fp);
694 mtx_lock(mtxp);
695 if (fp->f_advice != NULL &&
696 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
697 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
698 start = fp->f_advice->fa_prevstart;
699 else if (fp->f_advice->fa_prevstart != 0 &&
700 fp->f_advice->fa_prevstart == end + 1)
701 end = fp->f_advice->fa_prevend;
702 fp->f_advice->fa_prevstart = start;
703 fp->f_advice->fa_prevend = end;
704 }
705 mtx_unlock(mtxp);
706 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
707 }
708 VFS_UNLOCK_GIANT(vfslocked);
709 return (error);
710 }
711
712 /*
713 * File table vnode write routine.
714 */
715 static int
716 vn_write(fp, uio, active_cred, flags, td)
717 struct file *fp;
718 struct uio *uio;
719 struct ucred *active_cred;
720 int flags;
721 struct thread *td;
722 {
723 struct vnode *vp;
724 struct mount *mp;
725 struct mtx *mtxp;
726 int error, ioflag, lock_flags;
727 int advice, vfslocked;
728 off_t offset, start, end;
729
730 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
731 uio->uio_td, td));
732 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
733 vp = fp->f_vnode;
734 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
735 if (vp->v_type == VREG)
736 bwillwrite();
737 ioflag = IO_UNIT;
738 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
739 ioflag |= IO_APPEND;
740 if (fp->f_flag & FNONBLOCK)
741 ioflag |= IO_NDELAY;
742 if (fp->f_flag & O_DIRECT)
743 ioflag |= IO_DIRECT;
744 if ((fp->f_flag & O_FSYNC) ||
745 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
746 ioflag |= IO_SYNC;
747 mp = NULL;
748 if (vp->v_type != VCHR &&
749 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
750 goto unlock;
751
752 advice = get_advice(fp, uio);
753
754 if ((MNT_SHARED_WRITES(mp) ||
755 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
756 (flags & FOF_OFFSET) != 0) {
757 lock_flags = LK_SHARED;
758 } else {
759 lock_flags = LK_EXCLUSIVE;
760 }
761
762 vn_lock(vp, lock_flags | LK_RETRY);
763 switch (advice) {
764 case POSIX_FADV_NORMAL:
765 case POSIX_FADV_SEQUENTIAL:
766 case POSIX_FADV_NOREUSE:
767 ioflag |= sequential_heuristic(uio, fp);
768 break;
769 case POSIX_FADV_RANDOM:
770 /* XXX: Is this correct? */
771 break;
772 }
773 offset = uio->uio_offset;
774
775 #ifdef MAC
776 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
777 if (error == 0)
778 #endif
779 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
780 fp->f_nextoff = uio->uio_offset;
781 VOP_UNLOCK(vp, 0);
782 if (vp->v_type != VCHR)
783 vn_finished_write(mp);
784 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
785 offset != uio->uio_offset) {
786 /*
787 * Use POSIX_FADV_DONTNEED to flush clean pages and
788 * buffers for the backing file after a
789 * POSIX_FADV_NOREUSE write(2). To optimize the
790 * common case of using POSIX_FADV_NOREUSE with
791 * sequential access, track the previous implicit
792 * DONTNEED request and grow this request to include
793 * the current write(2) in addition to the previous
794 * DONTNEED. With purely sequential access this will
795 * cause the DONTNEED requests to continously grow to
796 * cover all of the previously written regions of the
797 * file.
798 *
799 * Note that the blocks just written are almost
800 * certainly still dirty, so this only works when
801 * VOP_ADVISE() calls from subsequent writes push out
802 * the data written by this write(2) once the backing
803 * buffers are clean. However, as compared to forcing
804 * IO_DIRECT, this gives much saner behavior. Write
805 * clustering is still allowed, and clean pages are
806 * merely moved to the cache page queue rather than
807 * outright thrown away. This means a subsequent
808 * read(2) can still avoid hitting the disk if the
809 * pages have not been reclaimed.
810 *
811 * This does make POSIX_FADV_NOREUSE largely useless
812 * with non-sequential access. However, sequential
813 * access is the more common use case and the flag is
814 * merely advisory.
815 */
816 start = offset;
817 end = uio->uio_offset - 1;
818 mtxp = mtx_pool_find(mtxpool_sleep, fp);
819 mtx_lock(mtxp);
820 if (fp->f_advice != NULL &&
821 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
822 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
823 start = fp->f_advice->fa_prevstart;
824 else if (fp->f_advice->fa_prevstart != 0 &&
825 fp->f_advice->fa_prevstart == end + 1)
826 end = fp->f_advice->fa_prevend;
827 fp->f_advice->fa_prevstart = start;
828 fp->f_advice->fa_prevend = end;
829 }
830 mtx_unlock(mtxp);
831 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
832 }
833
834 unlock:
835 VFS_UNLOCK_GIANT(vfslocked);
836 return (error);
837 }
838
839 static const int io_hold_cnt = 16;
840 static int vn_io_fault_enable = 0;
841 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
842 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
843 static u_long vn_io_faults_cnt;
844 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
845 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
846
847 /*
848 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
849 * prevent the following deadlock:
850 *
851 * Assume that the thread A reads from the vnode vp1 into userspace
852 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
853 * currently not resident, then system ends up with the call chain
854 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
855 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
856 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
857 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
858 * backed by the pages of vnode vp1, and some page in buf2 is not
859 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
860 *
861 * To prevent the lock order reversal and deadlock, vn_io_fault() does
862 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
863 * Instead, it first tries to do the whole range i/o with pagefaults
864 * disabled. If all pages in the i/o buffer are resident and mapped,
865 * VOP will succeed (ignoring the genuine filesystem errors).
866 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
867 * i/o in chunks, with all pages in the chunk prefaulted and held
868 * using vm_fault_quick_hold_pages().
869 *
870 * Filesystems using this deadlock avoidance scheme should use the
871 * array of the held pages from uio, saved in the curthread->td_ma,
872 * instead of doing uiomove(). A helper function
873 * vn_io_fault_uiomove() converts uiomove request into
874 * uiomove_fromphys() over td_ma array.
875 *
876 * Since vnode locks do not cover the whole i/o anymore, rangelocks
877 * make the current i/o request atomic with respect to other i/os and
878 * truncations.
879 */
880 static int
881 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
882 int flags, struct thread *td)
883 {
884 vm_page_t ma[io_hold_cnt + 2];
885 struct uio *uio_clone, short_uio;
886 struct iovec short_iovec[1];
887 fo_rdwr_t *doio;
888 struct vnode *vp;
889 void *rl_cookie;
890 struct mount *mp;
891 vm_page_t *prev_td_ma;
892 int cnt, error, save, saveheld, prev_td_ma_cnt;
893 vm_offset_t addr, end;
894 vm_prot_t prot;
895 size_t len, resid;
896 ssize_t adv;
897
898 if (uio->uio_rw == UIO_READ)
899 doio = vn_read;
900 else
901 doio = vn_write;
902 vp = fp->f_vnode;
903 foffset_lock_uio(fp, uio, flags);
904
905 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
906 ((mp = vp->v_mount) != NULL &&
907 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
908 !vn_io_fault_enable) {
909 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
910 goto out_last;
911 }
912
913 /*
914 * The UFS follows IO_UNIT directive and replays back both
915 * uio_offset and uio_resid if an error is encountered during the
916 * operation. But, since the iovec may be already advanced,
917 * uio is still in an inconsistent state.
918 *
919 * Cache a copy of the original uio, which is advanced to the redo
920 * point using UIO_NOCOPY below.
921 */
922 uio_clone = cloneuio(uio);
923 resid = uio->uio_resid;
924
925 short_uio.uio_segflg = UIO_USERSPACE;
926 short_uio.uio_rw = uio->uio_rw;
927 short_uio.uio_td = uio->uio_td;
928
929 if (uio->uio_rw == UIO_READ) {
930 prot = VM_PROT_WRITE;
931 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
932 uio->uio_offset + uio->uio_resid);
933 } else {
934 prot = VM_PROT_READ;
935 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
936 /* For appenders, punt and lock the whole range. */
937 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
938 else
939 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
940 uio->uio_offset + uio->uio_resid);
941 }
942
943 save = vm_fault_disable_pagefaults();
944 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
945 if (error != EFAULT)
946 goto out;
947
948 atomic_add_long(&vn_io_faults_cnt, 1);
949 uio_clone->uio_segflg = UIO_NOCOPY;
950 uiomove(NULL, resid - uio->uio_resid, uio_clone);
951 uio_clone->uio_segflg = uio->uio_segflg;
952
953 saveheld = curthread_pflags_set(TDP_UIOHELD);
954 prev_td_ma = td->td_ma;
955 prev_td_ma_cnt = td->td_ma_cnt;
956
957 while (uio_clone->uio_resid != 0) {
958 len = uio_clone->uio_iov->iov_len;
959 if (len == 0) {
960 KASSERT(uio_clone->uio_iovcnt >= 1,
961 ("iovcnt underflow"));
962 uio_clone->uio_iov++;
963 uio_clone->uio_iovcnt--;
964 continue;
965 }
966
967 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
968 end = round_page(addr + len);
969 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
970 /*
971 * A perfectly misaligned address and length could cause
972 * both the start and the end of the chunk to use partial
973 * page. +2 accounts for such a situation.
974 */
975 if (cnt > io_hold_cnt + 2) {
976 len = io_hold_cnt * PAGE_SIZE;
977 KASSERT(howmany(round_page(addr + len) -
978 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
979 ("cnt overflow"));
980 }
981 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
982 addr, len, prot, ma, io_hold_cnt + 2);
983 if (cnt == -1) {
984 error = EFAULT;
985 break;
986 }
987 short_uio.uio_iov = &short_iovec[0];
988 short_iovec[0].iov_base = (void *)addr;
989 short_uio.uio_iovcnt = 1;
990 short_uio.uio_resid = short_iovec[0].iov_len = len;
991 short_uio.uio_offset = uio_clone->uio_offset;
992 td->td_ma = ma;
993 td->td_ma_cnt = cnt;
994
995 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
996 td);
997 vm_page_unhold_pages(ma, cnt);
998 adv = len - short_uio.uio_resid;
999
1000 uio_clone->uio_iov->iov_base =
1001 (char *)uio_clone->uio_iov->iov_base + adv;
1002 uio_clone->uio_iov->iov_len -= adv;
1003 uio_clone->uio_resid -= adv;
1004 uio_clone->uio_offset += adv;
1005
1006 uio->uio_resid -= adv;
1007 uio->uio_offset += adv;
1008
1009 if (error != 0 || adv == 0)
1010 break;
1011 }
1012 td->td_ma = prev_td_ma;
1013 td->td_ma_cnt = prev_td_ma_cnt;
1014 curthread_pflags_restore(saveheld);
1015 out:
1016 vm_fault_enable_pagefaults(save);
1017 vn_rangelock_unlock(vp, rl_cookie);
1018 free(uio_clone, M_IOV);
1019 out_last:
1020 foffset_unlock_uio(fp, uio, flags);
1021 return (error);
1022 }
1023
1024 /*
1025 * Helper function to perform the requested uiomove operation using
1026 * the held pages for io->uio_iov[0].iov_base buffer instead of
1027 * copyin/copyout. Access to the pages with uiomove_fromphys()
1028 * instead of iov_base prevents page faults that could occur due to
1029 * pmap_collect() invalidating the mapping created by
1030 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1031 * object cleanup revoking the write access from page mappings.
1032 *
1033 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1034 * instead of plain uiomove().
1035 */
1036 int
1037 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1038 {
1039 struct uio transp_uio;
1040 struct iovec transp_iov[1];
1041 struct thread *td;
1042 size_t adv;
1043 int error, pgadv;
1044
1045 td = curthread;
1046 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1047 uio->uio_segflg != UIO_USERSPACE)
1048 return (uiomove(data, xfersize, uio));
1049
1050 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1051 transp_iov[0].iov_base = data;
1052 transp_uio.uio_iov = &transp_iov[0];
1053 transp_uio.uio_iovcnt = 1;
1054 if (xfersize > uio->uio_resid)
1055 xfersize = uio->uio_resid;
1056 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1057 transp_uio.uio_offset = 0;
1058 transp_uio.uio_segflg = UIO_SYSSPACE;
1059 /*
1060 * Since transp_iov points to data, and td_ma page array
1061 * corresponds to original uio->uio_iov, we need to invert the
1062 * direction of the i/o operation as passed to
1063 * uiomove_fromphys().
1064 */
1065 switch (uio->uio_rw) {
1066 case UIO_WRITE:
1067 transp_uio.uio_rw = UIO_READ;
1068 break;
1069 case UIO_READ:
1070 transp_uio.uio_rw = UIO_WRITE;
1071 break;
1072 }
1073 transp_uio.uio_td = uio->uio_td;
1074 error = uiomove_fromphys(td->td_ma,
1075 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1076 xfersize, &transp_uio);
1077 adv = xfersize - transp_uio.uio_resid;
1078 pgadv =
1079 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1080 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1081 td->td_ma += pgadv;
1082 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1083 pgadv));
1084 td->td_ma_cnt -= pgadv;
1085 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1086 uio->uio_iov->iov_len -= adv;
1087 uio->uio_resid -= adv;
1088 uio->uio_offset += adv;
1089 return (error);
1090 }
1091
1092 int
1093 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1094 struct uio *uio)
1095 {
1096 struct thread *td;
1097 vm_offset_t iov_base;
1098 int cnt, pgadv;
1099
1100 td = curthread;
1101 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1102 uio->uio_segflg != UIO_USERSPACE)
1103 return (uiomove_fromphys(ma, offset, xfersize, uio));
1104
1105 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1106 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1107 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1108 switch (uio->uio_rw) {
1109 case UIO_WRITE:
1110 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1111 offset, cnt);
1112 break;
1113 case UIO_READ:
1114 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1115 cnt);
1116 break;
1117 }
1118 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1119 td->td_ma += pgadv;
1120 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1121 pgadv));
1122 td->td_ma_cnt -= pgadv;
1123 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1124 uio->uio_iov->iov_len -= cnt;
1125 uio->uio_resid -= cnt;
1126 uio->uio_offset += cnt;
1127 return (0);
1128 }
1129
1130
1131 /*
1132 * File table truncate routine.
1133 */
1134 static int
1135 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1136 struct thread *td)
1137 {
1138 struct vattr vattr;
1139 struct mount *mp;
1140 struct vnode *vp;
1141 void *rl_cookie;
1142 int vfslocked;
1143 int error;
1144
1145 vp = fp->f_vnode;
1146
1147 /*
1148 * Lock the whole range for truncation. Otherwise split i/o
1149 * might happen partly before and partly after the truncation.
1150 */
1151 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1152 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1153 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1154 if (error)
1155 goto out1;
1156 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1157 if (vp->v_type == VDIR) {
1158 error = EISDIR;
1159 goto out;
1160 }
1161 #ifdef MAC
1162 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1163 if (error)
1164 goto out;
1165 #endif
1166 error = vn_writechk(vp);
1167 if (error == 0) {
1168 VATTR_NULL(&vattr);
1169 vattr.va_size = length;
1170 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1171 }
1172 out:
1173 VOP_UNLOCK(vp, 0);
1174 vn_finished_write(mp);
1175 out1:
1176 VFS_UNLOCK_GIANT(vfslocked);
1177 vn_rangelock_unlock(vp, rl_cookie);
1178 return (error);
1179 }
1180
1181 /*
1182 * File table vnode stat routine.
1183 */
1184 static int
1185 vn_statfile(fp, sb, active_cred, td)
1186 struct file *fp;
1187 struct stat *sb;
1188 struct ucred *active_cred;
1189 struct thread *td;
1190 {
1191 struct vnode *vp = fp->f_vnode;
1192 int vfslocked;
1193 int error;
1194
1195 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1196 vn_lock(vp, LK_SHARED | LK_RETRY);
1197 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1198 VOP_UNLOCK(vp, 0);
1199 VFS_UNLOCK_GIANT(vfslocked);
1200
1201 return (error);
1202 }
1203
1204 /*
1205 * Stat a vnode; implementation for the stat syscall
1206 */
1207 int
1208 vn_stat(vp, sb, active_cred, file_cred, td)
1209 struct vnode *vp;
1210 register struct stat *sb;
1211 struct ucred *active_cred;
1212 struct ucred *file_cred;
1213 struct thread *td;
1214 {
1215 struct vattr vattr;
1216 register struct vattr *vap;
1217 int error;
1218 u_short mode;
1219
1220 #ifdef MAC
1221 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1222 if (error)
1223 return (error);
1224 #endif
1225
1226 vap = &vattr;
1227
1228 /*
1229 * Initialize defaults for new and unusual fields, so that file
1230 * systems which don't support these fields don't need to know
1231 * about them.
1232 */
1233 vap->va_birthtime.tv_sec = -1;
1234 vap->va_birthtime.tv_nsec = 0;
1235 vap->va_fsid = VNOVAL;
1236 vap->va_rdev = NODEV;
1237
1238 error = VOP_GETATTR(vp, vap, active_cred);
1239 if (error)
1240 return (error);
1241
1242 /*
1243 * Zero the spare stat fields
1244 */
1245 bzero(sb, sizeof *sb);
1246
1247 /*
1248 * Copy from vattr table
1249 */
1250 if (vap->va_fsid != VNOVAL)
1251 sb->st_dev = vap->va_fsid;
1252 else
1253 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1254 sb->st_ino = vap->va_fileid;
1255 mode = vap->va_mode;
1256 switch (vap->va_type) {
1257 case VREG:
1258 mode |= S_IFREG;
1259 break;
1260 case VDIR:
1261 mode |= S_IFDIR;
1262 break;
1263 case VBLK:
1264 mode |= S_IFBLK;
1265 break;
1266 case VCHR:
1267 mode |= S_IFCHR;
1268 break;
1269 case VLNK:
1270 mode |= S_IFLNK;
1271 break;
1272 case VSOCK:
1273 mode |= S_IFSOCK;
1274 break;
1275 case VFIFO:
1276 mode |= S_IFIFO;
1277 break;
1278 default:
1279 return (EBADF);
1280 };
1281 sb->st_mode = mode;
1282 sb->st_nlink = vap->va_nlink;
1283 sb->st_uid = vap->va_uid;
1284 sb->st_gid = vap->va_gid;
1285 sb->st_rdev = vap->va_rdev;
1286 if (vap->va_size > OFF_MAX)
1287 return (EOVERFLOW);
1288 sb->st_size = vap->va_size;
1289 sb->st_atim = vap->va_atime;
1290 sb->st_mtim = vap->va_mtime;
1291 sb->st_ctim = vap->va_ctime;
1292 sb->st_birthtim = vap->va_birthtime;
1293
1294 /*
1295 * According to www.opengroup.org, the meaning of st_blksize is
1296 * "a filesystem-specific preferred I/O block size for this
1297 * object. In some filesystem types, this may vary from file
1298 * to file"
1299 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1300 */
1301
1302 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1303
1304 sb->st_flags = vap->va_flags;
1305 if (priv_check(td, PRIV_VFS_GENERATION))
1306 sb->st_gen = 0;
1307 else
1308 sb->st_gen = vap->va_gen;
1309
1310 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1311 return (0);
1312 }
1313
1314 /*
1315 * File table vnode ioctl routine.
1316 */
1317 static int
1318 vn_ioctl(fp, com, data, active_cred, td)
1319 struct file *fp;
1320 u_long com;
1321 void *data;
1322 struct ucred *active_cred;
1323 struct thread *td;
1324 {
1325 struct vnode *vp = fp->f_vnode;
1326 struct vattr vattr;
1327 int vfslocked;
1328 int error;
1329
1330 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1331 error = ENOTTY;
1332 switch (vp->v_type) {
1333 case VREG:
1334 case VDIR:
1335 if (com == FIONREAD) {
1336 vn_lock(vp, LK_SHARED | LK_RETRY);
1337 error = VOP_GETATTR(vp, &vattr, active_cred);
1338 VOP_UNLOCK(vp, 0);
1339 if (!error)
1340 *(int *)data = vattr.va_size - fp->f_offset;
1341 } else if (com == FIONBIO || com == FIOASYNC) /* XXX */
1342 error = 0;
1343 else
1344 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1345 active_cred, td);
1346 break;
1347
1348 default:
1349 break;
1350 }
1351 VFS_UNLOCK_GIANT(vfslocked);
1352 return (error);
1353 }
1354
1355 /*
1356 * File table vnode poll routine.
1357 */
1358 static int
1359 vn_poll(fp, events, active_cred, td)
1360 struct file *fp;
1361 int events;
1362 struct ucred *active_cred;
1363 struct thread *td;
1364 {
1365 struct vnode *vp;
1366 int vfslocked;
1367 int error;
1368
1369 vp = fp->f_vnode;
1370 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1371 #ifdef MAC
1372 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1373 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1374 VOP_UNLOCK(vp, 0);
1375 if (!error)
1376 #endif
1377
1378 error = VOP_POLL(vp, events, fp->f_cred, td);
1379 VFS_UNLOCK_GIANT(vfslocked);
1380 return (error);
1381 }
1382
1383 /*
1384 * Acquire the requested lock and then check for validity. LK_RETRY
1385 * permits vn_lock to return doomed vnodes.
1386 */
1387 int
1388 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1389 {
1390 int error;
1391
1392 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1393 ("vn_lock called with no locktype."));
1394 do {
1395 #ifdef DEBUG_VFS_LOCKS
1396 KASSERT(vp->v_holdcnt != 0,
1397 ("vn_lock %p: zero hold count", vp));
1398 #endif
1399 error = VOP_LOCK1(vp, flags, file, line);
1400 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1401 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1402 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1403 flags, error));
1404 /*
1405 * Callers specify LK_RETRY if they wish to get dead vnodes.
1406 * If RETRY is not set, we return ENOENT instead.
1407 */
1408 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1409 (flags & LK_RETRY) == 0) {
1410 VOP_UNLOCK(vp, 0);
1411 error = ENOENT;
1412 break;
1413 }
1414 } while (flags & LK_RETRY && error != 0);
1415 return (error);
1416 }
1417
1418 /*
1419 * File table vnode close routine.
1420 */
1421 static int
1422 vn_closefile(fp, td)
1423 struct file *fp;
1424 struct thread *td;
1425 {
1426 struct vnode *vp;
1427 struct flock lf;
1428 int vfslocked;
1429 int error;
1430
1431 vp = fp->f_vnode;
1432
1433 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1434 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1435 lf.l_whence = SEEK_SET;
1436 lf.l_start = 0;
1437 lf.l_len = 0;
1438 lf.l_type = F_UNLCK;
1439 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1440 }
1441
1442 fp->f_ops = &badfileops;
1443
1444 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1445 VFS_UNLOCK_GIANT(vfslocked);
1446 return (error);
1447 }
1448
1449 /*
1450 * Preparing to start a filesystem write operation. If the operation is
1451 * permitted, then we bump the count of operations in progress and
1452 * proceed. If a suspend request is in progress, we wait until the
1453 * suspension is over, and then proceed.
1454 */
1455 static int
1456 vn_start_write_locked(struct mount *mp, int flags)
1457 {
1458 int error;
1459
1460 mtx_assert(MNT_MTX(mp), MA_OWNED);
1461 error = 0;
1462
1463 /*
1464 * Check on status of suspension.
1465 */
1466 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1467 mp->mnt_susp_owner != curthread) {
1468 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1469 if (flags & V_NOWAIT) {
1470 error = EWOULDBLOCK;
1471 goto unlock;
1472 }
1473 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1474 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1475 if (error)
1476 goto unlock;
1477 }
1478 }
1479 if (flags & V_XSLEEP)
1480 goto unlock;
1481 mp->mnt_writeopcount++;
1482 unlock:
1483 if (error != 0 || (flags & V_XSLEEP) != 0)
1484 MNT_REL(mp);
1485 MNT_IUNLOCK(mp);
1486 return (error);
1487 }
1488
1489 int
1490 vn_start_write(vp, mpp, flags)
1491 struct vnode *vp;
1492 struct mount **mpp;
1493 int flags;
1494 {
1495 struct mount *mp;
1496 int error;
1497
1498 error = 0;
1499 /*
1500 * If a vnode is provided, get and return the mount point that
1501 * to which it will write.
1502 */
1503 if (vp != NULL) {
1504 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1505 *mpp = NULL;
1506 if (error != EOPNOTSUPP)
1507 return (error);
1508 return (0);
1509 }
1510 }
1511 if ((mp = *mpp) == NULL)
1512 return (0);
1513
1514 /*
1515 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1516 * a vfs_ref().
1517 * As long as a vnode is not provided we need to acquire a
1518 * refcount for the provided mountpoint too, in order to
1519 * emulate a vfs_ref().
1520 */
1521 MNT_ILOCK(mp);
1522 if (vp == NULL)
1523 MNT_REF(mp);
1524
1525 return (vn_start_write_locked(mp, flags));
1526 }
1527
1528 /*
1529 * Secondary suspension. Used by operations such as vop_inactive
1530 * routines that are needed by the higher level functions. These
1531 * are allowed to proceed until all the higher level functions have
1532 * completed (indicated by mnt_writeopcount dropping to zero). At that
1533 * time, these operations are halted until the suspension is over.
1534 */
1535 int
1536 vn_start_secondary_write(vp, mpp, flags)
1537 struct vnode *vp;
1538 struct mount **mpp;
1539 int flags;
1540 {
1541 struct mount *mp;
1542 int error;
1543
1544 retry:
1545 if (vp != NULL) {
1546 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1547 *mpp = NULL;
1548 if (error != EOPNOTSUPP)
1549 return (error);
1550 return (0);
1551 }
1552 }
1553 /*
1554 * If we are not suspended or have not yet reached suspended
1555 * mode, then let the operation proceed.
1556 */
1557 if ((mp = *mpp) == NULL)
1558 return (0);
1559
1560 /*
1561 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1562 * a vfs_ref().
1563 * As long as a vnode is not provided we need to acquire a
1564 * refcount for the provided mountpoint too, in order to
1565 * emulate a vfs_ref().
1566 */
1567 MNT_ILOCK(mp);
1568 if (vp == NULL)
1569 MNT_REF(mp);
1570 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1571 mp->mnt_secondary_writes++;
1572 mp->mnt_secondary_accwrites++;
1573 MNT_IUNLOCK(mp);
1574 return (0);
1575 }
1576 if (flags & V_NOWAIT) {
1577 MNT_REL(mp);
1578 MNT_IUNLOCK(mp);
1579 return (EWOULDBLOCK);
1580 }
1581 /*
1582 * Wait for the suspension to finish.
1583 */
1584 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1585 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1586 vfs_rel(mp);
1587 if (error == 0)
1588 goto retry;
1589 return (error);
1590 }
1591
1592 /*
1593 * Filesystem write operation has completed. If we are suspending and this
1594 * operation is the last one, notify the suspender that the suspension is
1595 * now in effect.
1596 */
1597 void
1598 vn_finished_write(mp)
1599 struct mount *mp;
1600 {
1601 if (mp == NULL)
1602 return;
1603 MNT_ILOCK(mp);
1604 MNT_REL(mp);
1605 mp->mnt_writeopcount--;
1606 if (mp->mnt_writeopcount < 0)
1607 panic("vn_finished_write: neg cnt");
1608 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1609 mp->mnt_writeopcount <= 0)
1610 wakeup(&mp->mnt_writeopcount);
1611 MNT_IUNLOCK(mp);
1612 }
1613
1614
1615 /*
1616 * Filesystem secondary write operation has completed. If we are
1617 * suspending and this operation is the last one, notify the suspender
1618 * that the suspension is now in effect.
1619 */
1620 void
1621 vn_finished_secondary_write(mp)
1622 struct mount *mp;
1623 {
1624 if (mp == NULL)
1625 return;
1626 MNT_ILOCK(mp);
1627 MNT_REL(mp);
1628 mp->mnt_secondary_writes--;
1629 if (mp->mnt_secondary_writes < 0)
1630 panic("vn_finished_secondary_write: neg cnt");
1631 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1632 mp->mnt_secondary_writes <= 0)
1633 wakeup(&mp->mnt_secondary_writes);
1634 MNT_IUNLOCK(mp);
1635 }
1636
1637
1638
1639 /*
1640 * Request a filesystem to suspend write operations.
1641 */
1642 int
1643 vfs_write_suspend(mp)
1644 struct mount *mp;
1645 {
1646 int error;
1647
1648 MNT_ILOCK(mp);
1649 if (mp->mnt_susp_owner == curthread) {
1650 MNT_IUNLOCK(mp);
1651 return (EALREADY);
1652 }
1653 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1654 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1655 mp->mnt_kern_flag |= MNTK_SUSPEND;
1656 mp->mnt_susp_owner = curthread;
1657 if (mp->mnt_writeopcount > 0)
1658 (void) msleep(&mp->mnt_writeopcount,
1659 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1660 else
1661 MNT_IUNLOCK(mp);
1662 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1663 vfs_write_resume(mp);
1664 return (error);
1665 }
1666
1667 /*
1668 * Request a filesystem to resume write operations.
1669 */
1670 void
1671 vfs_write_resume_flags(struct mount *mp, int flags)
1672 {
1673
1674 MNT_ILOCK(mp);
1675 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1676 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1677 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1678 MNTK_SUSPENDED);
1679 mp->mnt_susp_owner = NULL;
1680 wakeup(&mp->mnt_writeopcount);
1681 wakeup(&mp->mnt_flag);
1682 curthread->td_pflags &= ~TDP_IGNSUSP;
1683 if ((flags & VR_START_WRITE) != 0) {
1684 MNT_REF(mp);
1685 mp->mnt_writeopcount++;
1686 }
1687 MNT_IUNLOCK(mp);
1688 if ((flags & VR_NO_SUSPCLR) == 0)
1689 VFS_SUSP_CLEAN(mp);
1690 } else if ((flags & VR_START_WRITE) != 0) {
1691 MNT_REF(mp);
1692 vn_start_write_locked(mp, 0);
1693 } else {
1694 MNT_IUNLOCK(mp);
1695 }
1696 }
1697
1698 void
1699 vfs_write_resume(struct mount *mp)
1700 {
1701
1702 vfs_write_resume_flags(mp, 0);
1703 }
1704
1705 /*
1706 * Implement kqueues for files by translating it to vnode operation.
1707 */
1708 static int
1709 vn_kqfilter(struct file *fp, struct knote *kn)
1710 {
1711 int vfslocked;
1712 int error;
1713
1714 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1715 error = VOP_KQFILTER(fp->f_vnode, kn);
1716 VFS_UNLOCK_GIANT(vfslocked);
1717
1718 return error;
1719 }
1720
1721 /*
1722 * Simplified in-kernel wrapper calls for extended attribute access.
1723 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1724 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1725 */
1726 int
1727 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1728 const char *attrname, int *buflen, char *buf, struct thread *td)
1729 {
1730 struct uio auio;
1731 struct iovec iov;
1732 int error;
1733
1734 iov.iov_len = *buflen;
1735 iov.iov_base = buf;
1736
1737 auio.uio_iov = &iov;
1738 auio.uio_iovcnt = 1;
1739 auio.uio_rw = UIO_READ;
1740 auio.uio_segflg = UIO_SYSSPACE;
1741 auio.uio_td = td;
1742 auio.uio_offset = 0;
1743 auio.uio_resid = *buflen;
1744
1745 if ((ioflg & IO_NODELOCKED) == 0)
1746 vn_lock(vp, LK_SHARED | LK_RETRY);
1747
1748 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1749
1750 /* authorize attribute retrieval as kernel */
1751 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1752 td);
1753
1754 if ((ioflg & IO_NODELOCKED) == 0)
1755 VOP_UNLOCK(vp, 0);
1756
1757 if (error == 0) {
1758 *buflen = *buflen - auio.uio_resid;
1759 }
1760
1761 return (error);
1762 }
1763
1764 /*
1765 * XXX failure mode if partially written?
1766 */
1767 int
1768 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1769 const char *attrname, int buflen, char *buf, struct thread *td)
1770 {
1771 struct uio auio;
1772 struct iovec iov;
1773 struct mount *mp;
1774 int error;
1775
1776 iov.iov_len = buflen;
1777 iov.iov_base = buf;
1778
1779 auio.uio_iov = &iov;
1780 auio.uio_iovcnt = 1;
1781 auio.uio_rw = UIO_WRITE;
1782 auio.uio_segflg = UIO_SYSSPACE;
1783 auio.uio_td = td;
1784 auio.uio_offset = 0;
1785 auio.uio_resid = buflen;
1786
1787 if ((ioflg & IO_NODELOCKED) == 0) {
1788 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1789 return (error);
1790 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1791 }
1792
1793 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1794
1795 /* authorize attribute setting as kernel */
1796 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1797
1798 if ((ioflg & IO_NODELOCKED) == 0) {
1799 vn_finished_write(mp);
1800 VOP_UNLOCK(vp, 0);
1801 }
1802
1803 return (error);
1804 }
1805
1806 int
1807 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1808 const char *attrname, struct thread *td)
1809 {
1810 struct mount *mp;
1811 int error;
1812
1813 if ((ioflg & IO_NODELOCKED) == 0) {
1814 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1815 return (error);
1816 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1817 }
1818
1819 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1820
1821 /* authorize attribute removal as kernel */
1822 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1823 if (error == EOPNOTSUPP)
1824 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1825 NULL, td);
1826
1827 if ((ioflg & IO_NODELOCKED) == 0) {
1828 vn_finished_write(mp);
1829 VOP_UNLOCK(vp, 0);
1830 }
1831
1832 return (error);
1833 }
1834
1835 int
1836 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1837 {
1838 struct mount *mp;
1839 int ltype, error;
1840
1841 mp = vp->v_mount;
1842 ltype = VOP_ISLOCKED(vp);
1843 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1844 ("vn_vget_ino: vp not locked"));
1845 error = vfs_busy(mp, MBF_NOWAIT);
1846 if (error != 0) {
1847 vfs_ref(mp);
1848 VOP_UNLOCK(vp, 0);
1849 error = vfs_busy(mp, 0);
1850 vn_lock(vp, ltype | LK_RETRY);
1851 vfs_rel(mp);
1852 if (error != 0)
1853 return (ENOENT);
1854 if (vp->v_iflag & VI_DOOMED) {
1855 vfs_unbusy(mp);
1856 return (ENOENT);
1857 }
1858 }
1859 VOP_UNLOCK(vp, 0);
1860 error = VFS_VGET(mp, ino, lkflags, rvp);
1861 vfs_unbusy(mp);
1862 vn_lock(vp, ltype | LK_RETRY);
1863 if (vp->v_iflag & VI_DOOMED) {
1864 if (error == 0)
1865 vput(*rvp);
1866 error = ENOENT;
1867 }
1868 return (error);
1869 }
1870
1871 int
1872 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1873 const struct thread *td)
1874 {
1875
1876 if (vp->v_type != VREG || td == NULL)
1877 return (0);
1878 PROC_LOCK(td->td_proc);
1879 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1880 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1881 kern_psignal(td->td_proc, SIGXFSZ);
1882 PROC_UNLOCK(td->td_proc);
1883 return (EFBIG);
1884 }
1885 PROC_UNLOCK(td->td_proc);
1886 return (0);
1887 }
1888
1889 int
1890 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1891 struct thread *td)
1892 {
1893 struct vnode *vp;
1894 int error, vfslocked;
1895
1896 vp = fp->f_vnode;
1897 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1898 #ifdef AUDIT
1899 vn_lock(vp, LK_SHARED | LK_RETRY);
1900 AUDIT_ARG_VNODE1(vp);
1901 VOP_UNLOCK(vp, 0);
1902 #endif
1903 error = setfmode(td, active_cred, vp, mode);
1904 VFS_UNLOCK_GIANT(vfslocked);
1905 return (error);
1906 }
1907
1908 int
1909 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1910 struct thread *td)
1911 {
1912 struct vnode *vp;
1913 int error, vfslocked;
1914
1915 vp = fp->f_vnode;
1916 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1917 #ifdef AUDIT
1918 vn_lock(vp, LK_SHARED | LK_RETRY);
1919 AUDIT_ARG_VNODE1(vp);
1920 VOP_UNLOCK(vp, 0);
1921 #endif
1922 error = setfown(td, active_cred, vp, uid, gid);
1923 VFS_UNLOCK_GIANT(vfslocked);
1924 return (error);
1925 }
1926
1927 void
1928 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1929 {
1930 vm_object_t object;
1931
1932 if ((object = vp->v_object) == NULL)
1933 return;
1934 VM_OBJECT_LOCK(object);
1935 vm_object_page_remove(object, start, end, 0);
1936 VM_OBJECT_UNLOCK(object);
1937 }
1938
1939 int
1940 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1941 {
1942 struct vattr va;
1943 daddr_t bn, bnp;
1944 uint64_t bsize;
1945 off_t noff;
1946 int error;
1947
1948 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1949 ("Wrong command %lu", cmd));
1950
1951 if (vn_lock(vp, LK_SHARED) != 0)
1952 return (EBADF);
1953 if (vp->v_type != VREG) {
1954 error = ENOTTY;
1955 goto unlock;
1956 }
1957 error = VOP_GETATTR(vp, &va, cred);
1958 if (error != 0)
1959 goto unlock;
1960 noff = *off;
1961 if (noff >= va.va_size) {
1962 error = ENXIO;
1963 goto unlock;
1964 }
1965 bsize = vp->v_mount->mnt_stat.f_iosize;
1966 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1967 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1968 if (error == EOPNOTSUPP) {
1969 error = ENOTTY;
1970 goto unlock;
1971 }
1972 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1973 (bnp != -1 && cmd == FIOSEEKDATA)) {
1974 noff = bn * bsize;
1975 if (noff < *off)
1976 noff = *off;
1977 goto unlock;
1978 }
1979 }
1980 if (noff > va.va_size)
1981 noff = va.va_size;
1982 /* noff == va.va_size. There is an implicit hole at the end of file. */
1983 if (cmd == FIOSEEKDATA)
1984 error = ENXIO;
1985 unlock:
1986 VOP_UNLOCK(vp, 0);
1987 if (error == 0)
1988 *off = noff;
1989 return (error);
1990 }
Cache object: 0d774b00bac233e4cb9a8673c026f44a
|