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, 2014 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("$FreeBSD$");
45
46 #include "opt_hwpmc_hooks.h"
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/disk.h>
51 #include <sys/fail.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/kdb.h>
55 #include <sys/stat.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/mman.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/vnode.h>
65 #include <sys/bio.h>
66 #include <sys/buf.h>
67 #include <sys/filio.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
70 #include <sys/sx.h>
71 #include <sys/sysctl.h>
72 #include <sys/ttycom.h>
73 #include <sys/conf.h>
74 #include <sys/syslog.h>
75 #include <sys/unistd.h>
76 #include <sys/user.h>
77
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80
81 #include <vm/vm.h>
82 #include <vm/vm_extern.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
88
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92
93 static fo_rdwr_t vn_read;
94 static fo_rdwr_t vn_write;
95 static fo_rdwr_t vn_io_fault;
96 static fo_truncate_t vn_truncate;
97 static fo_ioctl_t vn_ioctl;
98 static fo_poll_t vn_poll;
99 static fo_kqfilter_t vn_kqfilter;
100 static fo_stat_t vn_statfile;
101 static fo_close_t vn_closefile;
102 static fo_mmap_t vn_mmap;
103
104 struct fileops vnops = {
105 .fo_read = vn_io_fault,
106 .fo_write = vn_io_fault,
107 .fo_truncate = vn_truncate,
108 .fo_ioctl = vn_ioctl,
109 .fo_poll = vn_poll,
110 .fo_kqfilter = vn_kqfilter,
111 .fo_stat = vn_statfile,
112 .fo_close = vn_closefile,
113 .fo_chmod = vn_chmod,
114 .fo_chown = vn_chown,
115 .fo_sendfile = vn_sendfile,
116 .fo_seek = vn_seek,
117 .fo_fill_kinfo = vn_fill_kinfo,
118 .fo_mmap = vn_mmap,
119 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
120 };
121
122 static const int io_hold_cnt = 16;
123 static int vn_io_fault_enable = 1;
124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
125 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
126 static int vn_io_fault_prefault = 0;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
128 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
129 static u_long vn_io_faults_cnt;
130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
131 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
132
133 /*
134 * Returns true if vn_io_fault mode of handling the i/o request should
135 * be used.
136 */
137 static bool
138 do_vn_io_fault(struct vnode *vp, struct uio *uio)
139 {
140 struct mount *mp;
141
142 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
143 (mp = vp->v_mount) != NULL &&
144 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
145 }
146
147 /*
148 * Structure used to pass arguments to vn_io_fault1(), to do either
149 * file- or vnode-based I/O calls.
150 */
151 struct vn_io_fault_args {
152 enum {
153 VN_IO_FAULT_FOP,
154 VN_IO_FAULT_VOP
155 } kind;
156 struct ucred *cred;
157 int flags;
158 union {
159 struct fop_args_tag {
160 struct file *fp;
161 fo_rdwr_t *doio;
162 } fop_args;
163 struct vop_args_tag {
164 struct vnode *vp;
165 } vop_args;
166 } args;
167 };
168
169 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
170 struct vn_io_fault_args *args, struct thread *td);
171
172 int
173 vn_open(ndp, flagp, cmode, fp)
174 struct nameidata *ndp;
175 int *flagp, cmode;
176 struct file *fp;
177 {
178 struct thread *td = ndp->ni_cnd.cn_thread;
179
180 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
181 }
182
183 /*
184 * Common code for vnode open operations via a name lookup.
185 * Lookup the vnode and invoke VOP_CREATE if needed.
186 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
187 *
188 * Note that this does NOT free nameidata for the successful case,
189 * due to the NDINIT being done elsewhere.
190 */
191 int
192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
193 struct ucred *cred, struct file *fp)
194 {
195 struct vnode *vp;
196 struct mount *mp;
197 struct thread *td = ndp->ni_cnd.cn_thread;
198 struct vattr vat;
199 struct vattr *vap = &vat;
200 int fmode, error;
201
202 restart:
203 fmode = *flagp;
204 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
205 O_EXCL | O_DIRECTORY))
206 return (EINVAL);
207 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
208 ndp->ni_cnd.cn_nameiop = CREATE;
209 /*
210 * Set NOCACHE to avoid flushing the cache when
211 * rolling in many files at once.
212 */
213 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
214 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
215 ndp->ni_cnd.cn_flags |= FOLLOW;
216 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
217 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
218 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
219 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
220 if ((vn_open_flags & VN_OPEN_INVFS) == 0)
221 bwillwrite();
222 if ((error = namei(ndp)) != 0)
223 return (error);
224 if (ndp->ni_vp == NULL) {
225 VATTR_NULL(vap);
226 vap->va_type = VREG;
227 vap->va_mode = cmode;
228 if (fmode & O_EXCL)
229 vap->va_vaflags |= VA_EXCLUSIVE;
230 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
231 NDFREE(ndp, NDF_ONLY_PNBUF);
232 vput(ndp->ni_dvp);
233 if ((error = vn_start_write(NULL, &mp,
234 V_XSLEEP | PCATCH)) != 0)
235 return (error);
236 goto restart;
237 }
238 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
239 ndp->ni_cnd.cn_flags |= MAKEENTRY;
240 #ifdef MAC
241 error = mac_vnode_check_create(cred, ndp->ni_dvp,
242 &ndp->ni_cnd, vap);
243 if (error == 0)
244 #endif
245 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
246 &ndp->ni_cnd, vap);
247 vput(ndp->ni_dvp);
248 vn_finished_write(mp);
249 if (error) {
250 NDFREE(ndp, NDF_ONLY_PNBUF);
251 return (error);
252 }
253 fmode &= ~O_TRUNC;
254 vp = ndp->ni_vp;
255 } else {
256 if (ndp->ni_dvp == ndp->ni_vp)
257 vrele(ndp->ni_dvp);
258 else
259 vput(ndp->ni_dvp);
260 ndp->ni_dvp = NULL;
261 vp = ndp->ni_vp;
262 if (fmode & O_EXCL) {
263 error = EEXIST;
264 goto bad;
265 }
266 if (vp->v_type == VDIR) {
267 error = EISDIR;
268 goto bad;
269 }
270 fmode &= ~O_CREAT;
271 }
272 } else {
273 ndp->ni_cnd.cn_nameiop = LOOKUP;
274 ndp->ni_cnd.cn_flags = ISOPEN |
275 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
276 if (!(fmode & FWRITE))
277 ndp->ni_cnd.cn_flags |= LOCKSHARED;
278 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
279 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
280 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
281 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
282 if ((error = namei(ndp)) != 0)
283 return (error);
284 vp = ndp->ni_vp;
285 }
286 error = vn_open_vnode(vp, fmode, cred, td, fp);
287 if (error)
288 goto bad;
289 *flagp = fmode;
290 return (0);
291 bad:
292 NDFREE(ndp, NDF_ONLY_PNBUF);
293 vput(vp);
294 *flagp = fmode;
295 ndp->ni_vp = NULL;
296 return (error);
297 }
298
299 /*
300 * Common code for vnode open operations once a vnode is located.
301 * Check permissions, and call the VOP_OPEN routine.
302 */
303 int
304 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
305 struct thread *td, struct file *fp)
306 {
307 accmode_t accmode;
308 struct flock lf;
309 int error, lock_flags, type;
310
311 if (vp->v_type == VLNK)
312 return (EMLINK);
313 if (vp->v_type == VSOCK)
314 return (EOPNOTSUPP);
315 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
316 return (ENOTDIR);
317 accmode = 0;
318 if (fmode & (FWRITE | O_TRUNC)) {
319 if (vp->v_type == VDIR)
320 return (EISDIR);
321 accmode |= VWRITE;
322 }
323 if (fmode & FREAD)
324 accmode |= VREAD;
325 if (fmode & FEXEC)
326 accmode |= VEXEC;
327 if ((fmode & O_APPEND) && (fmode & FWRITE))
328 accmode |= VAPPEND;
329 #ifdef MAC
330 if (fmode & O_CREAT)
331 accmode |= VCREAT;
332 if (fmode & O_VERIFY)
333 accmode |= VVERIFY;
334 error = mac_vnode_check_open(cred, vp, accmode);
335 if (error)
336 return (error);
337
338 accmode &= ~(VCREAT | VVERIFY);
339 #endif
340 if ((fmode & O_CREAT) == 0) {
341 if (accmode & VWRITE) {
342 error = vn_writechk(vp);
343 if (error)
344 return (error);
345 }
346 if (accmode) {
347 error = VOP_ACCESS(vp, accmode, cred, td);
348 if (error)
349 return (error);
350 }
351 }
352 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
353 vn_lock(vp, LK_UPGRADE | LK_RETRY);
354 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
355 return (error);
356
357 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) {
358 KASSERT(fp != NULL, ("open with flock requires fp"));
359 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) {
360 error = EOPNOTSUPP;
361 break;
362 }
363 lock_flags = VOP_ISLOCKED(vp);
364 VOP_UNLOCK(vp, 0);
365 lf.l_whence = SEEK_SET;
366 lf.l_start = 0;
367 lf.l_len = 0;
368 if (fmode & O_EXLOCK)
369 lf.l_type = F_WRLCK;
370 else
371 lf.l_type = F_RDLCK;
372 type = F_FLOCK;
373 if ((fmode & FNONBLOCK) == 0)
374 type |= F_WAIT;
375 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
376 if (error == 0)
377 fp->f_flag |= FHASLOCK;
378 vn_lock(vp, lock_flags | LK_RETRY);
379 if (error != 0)
380 break;
381 if ((vp->v_iflag & VI_DOOMED) != 0) {
382 error = ENOENT;
383 break;
384 }
385
386 /*
387 * Another thread might have used this vnode as an
388 * executable while the vnode lock was dropped.
389 * Ensure the vnode is still able to be opened for
390 * writing after the lock has been obtained.
391 */
392 if ((accmode & VWRITE) != 0)
393 error = vn_writechk(vp);
394 break;
395 }
396
397 if (error != 0) {
398 fp->f_flag |= FOPENFAILED;
399 fp->f_vnode = vp;
400 if (fp->f_ops == &badfileops) {
401 fp->f_type = DTYPE_VNODE;
402 fp->f_ops = &vnops;
403 }
404 vref(vp);
405 } else if ((fmode & FWRITE) != 0) {
406 VOP_ADD_WRITECOUNT(vp, 1);
407 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
408 __func__, vp, vp->v_writecount);
409 }
410 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
411 return (error);
412 }
413
414 /*
415 * Check for write permissions on the specified vnode.
416 * Prototype text segments cannot be written.
417 */
418 int
419 vn_writechk(struct vnode *vp)
420 {
421
422 ASSERT_VOP_LOCKED(vp, "vn_writechk");
423 /*
424 * If there's shared text associated with
425 * the vnode, try to free it up once. If
426 * we fail, we can't allow writing.
427 */
428 if (VOP_IS_TEXT(vp))
429 return (ETXTBSY);
430
431 return (0);
432 }
433
434 /*
435 * Vnode close call
436 */
437 static int
438 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
439 struct thread *td, bool keep_ref)
440 {
441 struct mount *mp;
442 int error, lock_flags;
443
444 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
445 MNT_EXTENDED_SHARED(vp->v_mount))
446 lock_flags = LK_SHARED;
447 else
448 lock_flags = LK_EXCLUSIVE;
449
450 vn_start_write(vp, &mp, V_WAIT);
451 vn_lock(vp, lock_flags | LK_RETRY);
452 AUDIT_ARG_VNODE1(vp);
453 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
454 VNASSERT(vp->v_writecount > 0, vp,
455 ("vn_close: negative writecount"));
456 VOP_ADD_WRITECOUNT(vp, -1);
457 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
458 __func__, vp, vp->v_writecount);
459 }
460 error = VOP_CLOSE(vp, flags, file_cred, td);
461 if (keep_ref)
462 VOP_UNLOCK(vp, 0);
463 else
464 vput(vp);
465 vn_finished_write(mp);
466 return (error);
467 }
468
469 int
470 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
471 struct thread *td)
472 {
473
474 return (vn_close1(vp, flags, file_cred, td, false));
475 }
476
477 /*
478 * Heuristic to detect sequential operation.
479 */
480 static int
481 sequential_heuristic(struct uio *uio, struct file *fp)
482 {
483
484 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
485 if (fp->f_flag & FRDAHEAD)
486 return (fp->f_seqcount << IO_SEQSHIFT);
487
488 /*
489 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
490 * that the first I/O is normally considered to be slightly
491 * sequential. Seeking to offset 0 doesn't change sequentiality
492 * unless previous seeks have reduced f_seqcount to 0, in which
493 * case offset 0 is not special.
494 */
495 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
496 uio->uio_offset == fp->f_nextoff) {
497 /*
498 * f_seqcount is in units of fixed-size blocks so that it
499 * depends mainly on the amount of sequential I/O and not
500 * much on the number of sequential I/O's. The fixed size
501 * of 16384 is hard-coded here since it is (not quite) just
502 * a magic size that works well here. This size is more
503 * closely related to the best I/O size for real disks than
504 * to any block size used by software.
505 */
506 fp->f_seqcount += howmany(uio->uio_resid, 16384);
507 if (fp->f_seqcount > IO_SEQMAX)
508 fp->f_seqcount = IO_SEQMAX;
509 return (fp->f_seqcount << IO_SEQSHIFT);
510 }
511
512 /* Not sequential. Quickly draw-down sequentiality. */
513 if (fp->f_seqcount > 1)
514 fp->f_seqcount = 1;
515 else
516 fp->f_seqcount = 0;
517 return (0);
518 }
519
520 /*
521 * Package up an I/O request on a vnode into a uio and do it.
522 */
523 int
524 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
525 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
526 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
527 {
528 struct uio auio;
529 struct iovec aiov;
530 struct mount *mp;
531 struct ucred *cred;
532 void *rl_cookie;
533 struct vn_io_fault_args args;
534 int error, lock_flags;
535
536 if (offset < 0 && vp->v_type != VCHR)
537 return (EINVAL);
538 auio.uio_iov = &aiov;
539 auio.uio_iovcnt = 1;
540 aiov.iov_base = base;
541 aiov.iov_len = len;
542 auio.uio_resid = len;
543 auio.uio_offset = offset;
544 auio.uio_segflg = segflg;
545 auio.uio_rw = rw;
546 auio.uio_td = td;
547 error = 0;
548
549 if ((ioflg & IO_NODELOCKED) == 0) {
550 if ((ioflg & IO_RANGELOCKED) == 0) {
551 if (rw == UIO_READ) {
552 rl_cookie = vn_rangelock_rlock(vp, offset,
553 offset + len);
554 } else {
555 rl_cookie = vn_rangelock_wlock(vp, offset,
556 offset + len);
557 }
558 } else
559 rl_cookie = NULL;
560 mp = NULL;
561 if (rw == UIO_WRITE) {
562 if (vp->v_type != VCHR &&
563 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
564 != 0)
565 goto out;
566 if (MNT_SHARED_WRITES(mp) ||
567 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
568 lock_flags = LK_SHARED;
569 else
570 lock_flags = LK_EXCLUSIVE;
571 } else
572 lock_flags = LK_SHARED;
573 vn_lock(vp, lock_flags | LK_RETRY);
574 } else
575 rl_cookie = NULL;
576
577 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
578 #ifdef MAC
579 if ((ioflg & IO_NOMACCHECK) == 0) {
580 if (rw == UIO_READ)
581 error = mac_vnode_check_read(active_cred, file_cred,
582 vp);
583 else
584 error = mac_vnode_check_write(active_cred, file_cred,
585 vp);
586 }
587 #endif
588 if (error == 0) {
589 if (file_cred != NULL)
590 cred = file_cred;
591 else
592 cred = active_cred;
593 if (do_vn_io_fault(vp, &auio)) {
594 args.kind = VN_IO_FAULT_VOP;
595 args.cred = cred;
596 args.flags = ioflg;
597 args.args.vop_args.vp = vp;
598 error = vn_io_fault1(vp, &auio, &args, td);
599 } else if (rw == UIO_READ) {
600 error = VOP_READ(vp, &auio, ioflg, cred);
601 } else /* if (rw == UIO_WRITE) */ {
602 error = VOP_WRITE(vp, &auio, ioflg, cred);
603 }
604 }
605 if (aresid)
606 *aresid = auio.uio_resid;
607 else
608 if (auio.uio_resid && error == 0)
609 error = EIO;
610 if ((ioflg & IO_NODELOCKED) == 0) {
611 VOP_UNLOCK(vp, 0);
612 if (mp != NULL)
613 vn_finished_write(mp);
614 }
615 out:
616 if (rl_cookie != NULL)
617 vn_rangelock_unlock(vp, rl_cookie);
618 return (error);
619 }
620
621 /*
622 * Package up an I/O request on a vnode into a uio and do it. The I/O
623 * request is split up into smaller chunks and we try to avoid saturating
624 * the buffer cache while potentially holding a vnode locked, so we
625 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
626 * to give other processes a chance to lock the vnode (either other processes
627 * core'ing the same binary, or unrelated processes scanning the directory).
628 */
629 int
630 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
631 file_cred, aresid, td)
632 enum uio_rw rw;
633 struct vnode *vp;
634 void *base;
635 size_t len;
636 off_t offset;
637 enum uio_seg segflg;
638 int ioflg;
639 struct ucred *active_cred;
640 struct ucred *file_cred;
641 size_t *aresid;
642 struct thread *td;
643 {
644 int error = 0;
645 ssize_t iaresid;
646
647 do {
648 int chunk;
649
650 /*
651 * Force `offset' to a multiple of MAXBSIZE except possibly
652 * for the first chunk, so that filesystems only need to
653 * write full blocks except possibly for the first and last
654 * chunks.
655 */
656 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
657
658 if (chunk > len)
659 chunk = len;
660 if (rw != UIO_READ && vp->v_type == VREG)
661 bwillwrite();
662 iaresid = 0;
663 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
664 ioflg, active_cred, file_cred, &iaresid, td);
665 len -= chunk; /* aresid calc already includes length */
666 if (error)
667 break;
668 offset += chunk;
669 base = (char *)base + chunk;
670 kern_yield(PRI_USER);
671 } while (len);
672 if (aresid)
673 *aresid = len + iaresid;
674 return (error);
675 }
676
677 off_t
678 foffset_lock(struct file *fp, int flags)
679 {
680 struct mtx *mtxp;
681 off_t res;
682
683 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
684
685 #if OFF_MAX <= LONG_MAX
686 /*
687 * Caller only wants the current f_offset value. Assume that
688 * the long and shorter integer types reads are atomic.
689 */
690 if ((flags & FOF_NOLOCK) != 0)
691 return (fp->f_offset);
692 #endif
693
694 /*
695 * According to McKusick the vn lock was protecting f_offset here.
696 * It is now protected by the FOFFSET_LOCKED flag.
697 */
698 mtxp = mtx_pool_find(mtxpool_sleep, fp);
699 mtx_lock(mtxp);
700 if ((flags & FOF_NOLOCK) == 0) {
701 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
702 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
703 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
704 "vofflock", 0);
705 }
706 fp->f_vnread_flags |= FOFFSET_LOCKED;
707 }
708 res = fp->f_offset;
709 mtx_unlock(mtxp);
710 return (res);
711 }
712
713 void
714 foffset_unlock(struct file *fp, off_t val, int flags)
715 {
716 struct mtx *mtxp;
717
718 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
719
720 #if OFF_MAX <= LONG_MAX
721 if ((flags & FOF_NOLOCK) != 0) {
722 if ((flags & FOF_NOUPDATE) == 0)
723 fp->f_offset = val;
724 if ((flags & FOF_NEXTOFF) != 0)
725 fp->f_nextoff = val;
726 return;
727 }
728 #endif
729
730 mtxp = mtx_pool_find(mtxpool_sleep, fp);
731 mtx_lock(mtxp);
732 if ((flags & FOF_NOUPDATE) == 0)
733 fp->f_offset = val;
734 if ((flags & FOF_NEXTOFF) != 0)
735 fp->f_nextoff = val;
736 if ((flags & FOF_NOLOCK) == 0) {
737 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
738 ("Lost FOFFSET_LOCKED"));
739 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
740 wakeup(&fp->f_vnread_flags);
741 fp->f_vnread_flags = 0;
742 }
743 mtx_unlock(mtxp);
744 }
745
746 void
747 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
748 {
749
750 if ((flags & FOF_OFFSET) == 0)
751 uio->uio_offset = foffset_lock(fp, flags);
752 }
753
754 void
755 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
756 {
757
758 if ((flags & FOF_OFFSET) == 0)
759 foffset_unlock(fp, uio->uio_offset, flags);
760 }
761
762 static int
763 get_advice(struct file *fp, struct uio *uio)
764 {
765 struct mtx *mtxp;
766 int ret;
767
768 ret = POSIX_FADV_NORMAL;
769 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
770 return (ret);
771
772 mtxp = mtx_pool_find(mtxpool_sleep, fp);
773 mtx_lock(mtxp);
774 if (fp->f_advice != NULL &&
775 uio->uio_offset >= fp->f_advice->fa_start &&
776 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
777 ret = fp->f_advice->fa_advice;
778 mtx_unlock(mtxp);
779 return (ret);
780 }
781
782 /*
783 * File table vnode read routine.
784 */
785 static int
786 vn_read(fp, uio, active_cred, flags, td)
787 struct file *fp;
788 struct uio *uio;
789 struct ucred *active_cred;
790 int flags;
791 struct thread *td;
792 {
793 struct vnode *vp;
794 off_t orig_offset;
795 int error, ioflag;
796 int advice;
797
798 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
799 uio->uio_td, td));
800 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
801 vp = fp->f_vnode;
802 ioflag = 0;
803 if (fp->f_flag & FNONBLOCK)
804 ioflag |= IO_NDELAY;
805 if (fp->f_flag & O_DIRECT)
806 ioflag |= IO_DIRECT;
807 advice = get_advice(fp, uio);
808 vn_lock(vp, LK_SHARED | LK_RETRY);
809
810 switch (advice) {
811 case POSIX_FADV_NORMAL:
812 case POSIX_FADV_SEQUENTIAL:
813 case POSIX_FADV_NOREUSE:
814 ioflag |= sequential_heuristic(uio, fp);
815 break;
816 case POSIX_FADV_RANDOM:
817 /* Disable read-ahead for random I/O. */
818 break;
819 }
820 orig_offset = uio->uio_offset;
821
822 #ifdef MAC
823 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
824 if (error == 0)
825 #endif
826 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
827 fp->f_nextoff = uio->uio_offset;
828 VOP_UNLOCK(vp, 0);
829 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
830 orig_offset != uio->uio_offset)
831 /*
832 * Use POSIX_FADV_DONTNEED to flush pages and buffers
833 * for the backing file after a POSIX_FADV_NOREUSE
834 * read(2).
835 */
836 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
837 POSIX_FADV_DONTNEED);
838 return (error);
839 }
840
841 /*
842 * File table vnode write routine.
843 */
844 static int
845 vn_write(fp, uio, active_cred, flags, td)
846 struct file *fp;
847 struct uio *uio;
848 struct ucred *active_cred;
849 int flags;
850 struct thread *td;
851 {
852 struct vnode *vp;
853 struct mount *mp;
854 off_t orig_offset;
855 int error, ioflag, lock_flags;
856 int advice;
857
858 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
859 uio->uio_td, td));
860 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
861 vp = fp->f_vnode;
862 if (vp->v_type == VREG)
863 bwillwrite();
864 ioflag = IO_UNIT;
865 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
866 ioflag |= IO_APPEND;
867 if (fp->f_flag & FNONBLOCK)
868 ioflag |= IO_NDELAY;
869 if (fp->f_flag & O_DIRECT)
870 ioflag |= IO_DIRECT;
871 if ((fp->f_flag & O_FSYNC) ||
872 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
873 ioflag |= IO_SYNC;
874 mp = NULL;
875 if (vp->v_type != VCHR &&
876 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
877 goto unlock;
878
879 advice = get_advice(fp, uio);
880
881 if (MNT_SHARED_WRITES(mp) ||
882 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
883 lock_flags = LK_SHARED;
884 } else {
885 lock_flags = LK_EXCLUSIVE;
886 }
887
888 vn_lock(vp, lock_flags | LK_RETRY);
889 switch (advice) {
890 case POSIX_FADV_NORMAL:
891 case POSIX_FADV_SEQUENTIAL:
892 case POSIX_FADV_NOREUSE:
893 ioflag |= sequential_heuristic(uio, fp);
894 break;
895 case POSIX_FADV_RANDOM:
896 /* XXX: Is this correct? */
897 break;
898 }
899 orig_offset = uio->uio_offset;
900
901 #ifdef MAC
902 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
903 if (error == 0)
904 #endif
905 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
906 fp->f_nextoff = uio->uio_offset;
907 VOP_UNLOCK(vp, 0);
908 if (vp->v_type != VCHR)
909 vn_finished_write(mp);
910 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
911 orig_offset != uio->uio_offset)
912 /*
913 * Use POSIX_FADV_DONTNEED to flush pages and buffers
914 * for the backing file after a POSIX_FADV_NOREUSE
915 * write(2).
916 */
917 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
918 POSIX_FADV_DONTNEED);
919 unlock:
920 return (error);
921 }
922
923 /*
924 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
925 * prevent the following deadlock:
926 *
927 * Assume that the thread A reads from the vnode vp1 into userspace
928 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
929 * currently not resident, then system ends up with the call chain
930 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
931 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
932 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
933 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
934 * backed by the pages of vnode vp1, and some page in buf2 is not
935 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
936 *
937 * To prevent the lock order reversal and deadlock, vn_io_fault() does
938 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
939 * Instead, it first tries to do the whole range i/o with pagefaults
940 * disabled. If all pages in the i/o buffer are resident and mapped,
941 * VOP will succeed (ignoring the genuine filesystem errors).
942 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
943 * i/o in chunks, with all pages in the chunk prefaulted and held
944 * using vm_fault_quick_hold_pages().
945 *
946 * Filesystems using this deadlock avoidance scheme should use the
947 * array of the held pages from uio, saved in the curthread->td_ma,
948 * instead of doing uiomove(). A helper function
949 * vn_io_fault_uiomove() converts uiomove request into
950 * uiomove_fromphys() over td_ma array.
951 *
952 * Since vnode locks do not cover the whole i/o anymore, rangelocks
953 * make the current i/o request atomic with respect to other i/os and
954 * truncations.
955 */
956
957 /*
958 * Decode vn_io_fault_args and perform the corresponding i/o.
959 */
960 static int
961 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
962 struct thread *td)
963 {
964 int error, save;
965
966 error = 0;
967 save = vm_fault_disable_pagefaults();
968 switch (args->kind) {
969 case VN_IO_FAULT_FOP:
970 error = (args->args.fop_args.doio)(args->args.fop_args.fp,
971 uio, args->cred, args->flags, td);
972 break;
973 case VN_IO_FAULT_VOP:
974 if (uio->uio_rw == UIO_READ) {
975 error = VOP_READ(args->args.vop_args.vp, uio,
976 args->flags, args->cred);
977 } else if (uio->uio_rw == UIO_WRITE) {
978 error = VOP_WRITE(args->args.vop_args.vp, uio,
979 args->flags, args->cred);
980 }
981 break;
982 default:
983 panic("vn_io_fault_doio: unknown kind of io %d %d",
984 args->kind, uio->uio_rw);
985 }
986 vm_fault_enable_pagefaults(save);
987 return (error);
988 }
989
990 static int
991 vn_io_fault_touch(char *base, const struct uio *uio)
992 {
993 int r;
994
995 r = fubyte(base);
996 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
997 return (EFAULT);
998 return (0);
999 }
1000
1001 static int
1002 vn_io_fault_prefault_user(const struct uio *uio)
1003 {
1004 char *base;
1005 const struct iovec *iov;
1006 size_t len;
1007 ssize_t resid;
1008 int error, i;
1009
1010 KASSERT(uio->uio_segflg == UIO_USERSPACE,
1011 ("vn_io_fault_prefault userspace"));
1012
1013 error = i = 0;
1014 iov = uio->uio_iov;
1015 resid = uio->uio_resid;
1016 base = iov->iov_base;
1017 len = iov->iov_len;
1018 while (resid > 0) {
1019 error = vn_io_fault_touch(base, uio);
1020 if (error != 0)
1021 break;
1022 if (len < PAGE_SIZE) {
1023 if (len != 0) {
1024 error = vn_io_fault_touch(base + len - 1, uio);
1025 if (error != 0)
1026 break;
1027 resid -= len;
1028 }
1029 if (++i >= uio->uio_iovcnt)
1030 break;
1031 iov = uio->uio_iov + i;
1032 base = iov->iov_base;
1033 len = iov->iov_len;
1034 } else {
1035 len -= PAGE_SIZE;
1036 base += PAGE_SIZE;
1037 resid -= PAGE_SIZE;
1038 }
1039 }
1040 return (error);
1041 }
1042
1043 /*
1044 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1045 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1046 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1047 * into args and call vn_io_fault1() to handle faults during the user
1048 * mode buffer accesses.
1049 */
1050 static int
1051 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1052 struct thread *td)
1053 {
1054 vm_page_t ma[io_hold_cnt + 2];
1055 struct uio *uio_clone, short_uio;
1056 struct iovec short_iovec[1];
1057 vm_page_t *prev_td_ma;
1058 vm_prot_t prot;
1059 vm_offset_t addr, end;
1060 size_t len, resid;
1061 ssize_t adv;
1062 int error, cnt, saveheld, prev_td_ma_cnt;
1063
1064 if (vn_io_fault_prefault) {
1065 error = vn_io_fault_prefault_user(uio);
1066 if (error != 0)
1067 return (error); /* Or ignore ? */
1068 }
1069
1070 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1071
1072 /*
1073 * The UFS follows IO_UNIT directive and replays back both
1074 * uio_offset and uio_resid if an error is encountered during the
1075 * operation. But, since the iovec may be already advanced,
1076 * uio is still in an inconsistent state.
1077 *
1078 * Cache a copy of the original uio, which is advanced to the redo
1079 * point using UIO_NOCOPY below.
1080 */
1081 uio_clone = cloneuio(uio);
1082 resid = uio->uio_resid;
1083
1084 short_uio.uio_segflg = UIO_USERSPACE;
1085 short_uio.uio_rw = uio->uio_rw;
1086 short_uio.uio_td = uio->uio_td;
1087
1088 error = vn_io_fault_doio(args, uio, td);
1089 if (error != EFAULT)
1090 goto out;
1091
1092 atomic_add_long(&vn_io_faults_cnt, 1);
1093 uio_clone->uio_segflg = UIO_NOCOPY;
1094 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1095 uio_clone->uio_segflg = uio->uio_segflg;
1096
1097 saveheld = curthread_pflags_set(TDP_UIOHELD);
1098 prev_td_ma = td->td_ma;
1099 prev_td_ma_cnt = td->td_ma_cnt;
1100
1101 while (uio_clone->uio_resid != 0) {
1102 len = uio_clone->uio_iov->iov_len;
1103 if (len == 0) {
1104 KASSERT(uio_clone->uio_iovcnt >= 1,
1105 ("iovcnt underflow"));
1106 uio_clone->uio_iov++;
1107 uio_clone->uio_iovcnt--;
1108 continue;
1109 }
1110 if (len > io_hold_cnt * PAGE_SIZE)
1111 len = io_hold_cnt * PAGE_SIZE;
1112 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1113 end = round_page(addr + len);
1114 if (end < addr) {
1115 error = EFAULT;
1116 break;
1117 }
1118 cnt = atop(end - trunc_page(addr));
1119 /*
1120 * A perfectly misaligned address and length could cause
1121 * both the start and the end of the chunk to use partial
1122 * page. +2 accounts for such a situation.
1123 */
1124 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1125 addr, len, prot, ma, io_hold_cnt + 2);
1126 if (cnt == -1) {
1127 error = EFAULT;
1128 break;
1129 }
1130 short_uio.uio_iov = &short_iovec[0];
1131 short_iovec[0].iov_base = (void *)addr;
1132 short_uio.uio_iovcnt = 1;
1133 short_uio.uio_resid = short_iovec[0].iov_len = len;
1134 short_uio.uio_offset = uio_clone->uio_offset;
1135 td->td_ma = ma;
1136 td->td_ma_cnt = cnt;
1137
1138 error = vn_io_fault_doio(args, &short_uio, td);
1139 vm_page_unhold_pages(ma, cnt);
1140 adv = len - short_uio.uio_resid;
1141
1142 uio_clone->uio_iov->iov_base =
1143 (char *)uio_clone->uio_iov->iov_base + adv;
1144 uio_clone->uio_iov->iov_len -= adv;
1145 uio_clone->uio_resid -= adv;
1146 uio_clone->uio_offset += adv;
1147
1148 uio->uio_resid -= adv;
1149 uio->uio_offset += adv;
1150
1151 if (error != 0 || adv == 0)
1152 break;
1153 }
1154 td->td_ma = prev_td_ma;
1155 td->td_ma_cnt = prev_td_ma_cnt;
1156 curthread_pflags_restore(saveheld);
1157 out:
1158 free(uio_clone, M_IOV);
1159 return (error);
1160 }
1161
1162 static int
1163 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1164 int flags, struct thread *td)
1165 {
1166 fo_rdwr_t *doio;
1167 struct vnode *vp;
1168 void *rl_cookie;
1169 struct vn_io_fault_args args;
1170 int error;
1171
1172 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1173 vp = fp->f_vnode;
1174 foffset_lock_uio(fp, uio, flags);
1175 if (do_vn_io_fault(vp, uio)) {
1176 args.kind = VN_IO_FAULT_FOP;
1177 args.args.fop_args.fp = fp;
1178 args.args.fop_args.doio = doio;
1179 args.cred = active_cred;
1180 args.flags = flags | FOF_OFFSET;
1181 if (uio->uio_rw == UIO_READ) {
1182 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1183 uio->uio_offset + uio->uio_resid);
1184 } else if ((fp->f_flag & O_APPEND) != 0 ||
1185 (flags & FOF_OFFSET) == 0) {
1186 /* For appenders, punt and lock the whole range. */
1187 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1188 } else {
1189 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1190 uio->uio_offset + uio->uio_resid);
1191 }
1192 error = vn_io_fault1(vp, uio, &args, td);
1193 vn_rangelock_unlock(vp, rl_cookie);
1194 } else {
1195 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1196 }
1197 foffset_unlock_uio(fp, uio, flags);
1198 return (error);
1199 }
1200
1201 /*
1202 * Helper function to perform the requested uiomove operation using
1203 * the held pages for io->uio_iov[0].iov_base buffer instead of
1204 * copyin/copyout. Access to the pages with uiomove_fromphys()
1205 * instead of iov_base prevents page faults that could occur due to
1206 * pmap_collect() invalidating the mapping created by
1207 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1208 * object cleanup revoking the write access from page mappings.
1209 *
1210 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1211 * instead of plain uiomove().
1212 */
1213 int
1214 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1215 {
1216 struct uio transp_uio;
1217 struct iovec transp_iov[1];
1218 struct thread *td;
1219 size_t adv;
1220 int error, pgadv;
1221
1222 td = curthread;
1223 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1224 uio->uio_segflg != UIO_USERSPACE)
1225 return (uiomove(data, xfersize, uio));
1226
1227 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1228 transp_iov[0].iov_base = data;
1229 transp_uio.uio_iov = &transp_iov[0];
1230 transp_uio.uio_iovcnt = 1;
1231 if (xfersize > uio->uio_resid)
1232 xfersize = uio->uio_resid;
1233 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1234 transp_uio.uio_offset = 0;
1235 transp_uio.uio_segflg = UIO_SYSSPACE;
1236 /*
1237 * Since transp_iov points to data, and td_ma page array
1238 * corresponds to original uio->uio_iov, we need to invert the
1239 * direction of the i/o operation as passed to
1240 * uiomove_fromphys().
1241 */
1242 switch (uio->uio_rw) {
1243 case UIO_WRITE:
1244 transp_uio.uio_rw = UIO_READ;
1245 break;
1246 case UIO_READ:
1247 transp_uio.uio_rw = UIO_WRITE;
1248 break;
1249 }
1250 transp_uio.uio_td = uio->uio_td;
1251 error = uiomove_fromphys(td->td_ma,
1252 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1253 xfersize, &transp_uio);
1254 adv = xfersize - transp_uio.uio_resid;
1255 pgadv =
1256 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1257 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1258 td->td_ma += pgadv;
1259 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1260 pgadv));
1261 td->td_ma_cnt -= pgadv;
1262 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1263 uio->uio_iov->iov_len -= adv;
1264 uio->uio_resid -= adv;
1265 uio->uio_offset += adv;
1266 return (error);
1267 }
1268
1269 int
1270 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1271 struct uio *uio)
1272 {
1273 struct thread *td;
1274 vm_offset_t iov_base;
1275 int cnt, pgadv;
1276
1277 td = curthread;
1278 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1279 uio->uio_segflg != UIO_USERSPACE)
1280 return (uiomove_fromphys(ma, offset, xfersize, uio));
1281
1282 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1283 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1284 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1285 switch (uio->uio_rw) {
1286 case UIO_WRITE:
1287 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1288 offset, cnt);
1289 break;
1290 case UIO_READ:
1291 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1292 cnt);
1293 break;
1294 }
1295 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1296 td->td_ma += pgadv;
1297 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1298 pgadv));
1299 td->td_ma_cnt -= pgadv;
1300 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1301 uio->uio_iov->iov_len -= cnt;
1302 uio->uio_resid -= cnt;
1303 uio->uio_offset += cnt;
1304 return (0);
1305 }
1306
1307
1308 /*
1309 * File table truncate routine.
1310 */
1311 static int
1312 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1313 struct thread *td)
1314 {
1315 struct vattr vattr;
1316 struct mount *mp;
1317 struct vnode *vp;
1318 void *rl_cookie;
1319 int error;
1320
1321 vp = fp->f_vnode;
1322
1323 /*
1324 * Lock the whole range for truncation. Otherwise split i/o
1325 * might happen partly before and partly after the truncation.
1326 */
1327 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1328 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1329 if (error)
1330 goto out1;
1331 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1332 if (vp->v_type == VDIR) {
1333 error = EISDIR;
1334 goto out;
1335 }
1336 #ifdef MAC
1337 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1338 if (error)
1339 goto out;
1340 #endif
1341 error = vn_writechk(vp);
1342 if (error == 0) {
1343 VATTR_NULL(&vattr);
1344 vattr.va_size = length;
1345 if ((fp->f_flag & O_FSYNC) != 0)
1346 vattr.va_vaflags |= VA_SYNC;
1347 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1348 }
1349 out:
1350 VOP_UNLOCK(vp, 0);
1351 vn_finished_write(mp);
1352 out1:
1353 vn_rangelock_unlock(vp, rl_cookie);
1354 return (error);
1355 }
1356
1357 /*
1358 * File table vnode stat routine.
1359 */
1360 static int
1361 vn_statfile(fp, sb, active_cred, td)
1362 struct file *fp;
1363 struct stat *sb;
1364 struct ucred *active_cred;
1365 struct thread *td;
1366 {
1367 struct vnode *vp = fp->f_vnode;
1368 int error;
1369
1370 vn_lock(vp, LK_SHARED | LK_RETRY);
1371 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1372 VOP_UNLOCK(vp, 0);
1373
1374 return (error);
1375 }
1376
1377 /*
1378 * Stat a vnode; implementation for the stat syscall
1379 */
1380 int
1381 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
1382 struct ucred *file_cred, struct thread *td)
1383 {
1384 struct vattr vattr;
1385 struct vattr *vap;
1386 int error;
1387 u_short mode;
1388
1389 AUDIT_ARG_VNODE1(vp);
1390 #ifdef MAC
1391 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1392 if (error)
1393 return (error);
1394 #endif
1395
1396 vap = &vattr;
1397
1398 /*
1399 * Initialize defaults for new and unusual fields, so that file
1400 * systems which don't support these fields don't need to know
1401 * about them.
1402 */
1403 vap->va_birthtime.tv_sec = -1;
1404 vap->va_birthtime.tv_nsec = 0;
1405 vap->va_fsid = VNOVAL;
1406 vap->va_rdev = NODEV;
1407
1408 error = VOP_GETATTR(vp, vap, active_cred);
1409 if (error)
1410 return (error);
1411
1412 /*
1413 * Zero the spare stat fields
1414 */
1415 bzero(sb, sizeof *sb);
1416
1417 /*
1418 * Copy from vattr table
1419 */
1420 if (vap->va_fsid != VNOVAL)
1421 sb->st_dev = vap->va_fsid;
1422 else
1423 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1424 sb->st_ino = vap->va_fileid;
1425 mode = vap->va_mode;
1426 switch (vap->va_type) {
1427 case VREG:
1428 mode |= S_IFREG;
1429 break;
1430 case VDIR:
1431 mode |= S_IFDIR;
1432 break;
1433 case VBLK:
1434 mode |= S_IFBLK;
1435 break;
1436 case VCHR:
1437 mode |= S_IFCHR;
1438 break;
1439 case VLNK:
1440 mode |= S_IFLNK;
1441 break;
1442 case VSOCK:
1443 mode |= S_IFSOCK;
1444 break;
1445 case VFIFO:
1446 mode |= S_IFIFO;
1447 break;
1448 default:
1449 return (EBADF);
1450 }
1451 sb->st_mode = mode;
1452 sb->st_nlink = vap->va_nlink;
1453 sb->st_uid = vap->va_uid;
1454 sb->st_gid = vap->va_gid;
1455 sb->st_rdev = vap->va_rdev;
1456 if (vap->va_size > OFF_MAX)
1457 return (EOVERFLOW);
1458 sb->st_size = vap->va_size;
1459 sb->st_atim = vap->va_atime;
1460 sb->st_mtim = vap->va_mtime;
1461 sb->st_ctim = vap->va_ctime;
1462 sb->st_birthtim = vap->va_birthtime;
1463
1464 /*
1465 * According to www.opengroup.org, the meaning of st_blksize is
1466 * "a filesystem-specific preferred I/O block size for this
1467 * object. In some filesystem types, this may vary from file
1468 * to file"
1469 * Use minimum/default of PAGE_SIZE (e.g. for VCHR).
1470 */
1471
1472 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1473
1474 sb->st_flags = vap->va_flags;
1475 if (priv_check(td, PRIV_VFS_GENERATION))
1476 sb->st_gen = 0;
1477 else
1478 sb->st_gen = vap->va_gen;
1479
1480 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1481 return (0);
1482 }
1483
1484 /*
1485 * File table vnode ioctl routine.
1486 */
1487 static int
1488 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
1489 struct thread *td)
1490 {
1491 struct vattr vattr;
1492 struct vnode *vp;
1493 int error;
1494
1495 vp = fp->f_vnode;
1496 switch (vp->v_type) {
1497 case VDIR:
1498 case VREG:
1499 switch (com) {
1500 case FIONREAD:
1501 vn_lock(vp, LK_SHARED | LK_RETRY);
1502 error = VOP_GETATTR(vp, &vattr, active_cred);
1503 VOP_UNLOCK(vp, 0);
1504 if (error == 0)
1505 *(int *)data = vattr.va_size - fp->f_offset;
1506 return (error);
1507 case FIONBIO:
1508 case FIOASYNC:
1509 return (0);
1510 default:
1511 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1512 active_cred, td));
1513 }
1514 default:
1515 return (ENOTTY);
1516 }
1517 }
1518
1519 /*
1520 * File table vnode poll routine.
1521 */
1522 static int
1523 vn_poll(struct file *fp, int events, struct ucred *active_cred,
1524 struct thread *td)
1525 {
1526 struct vnode *vp;
1527 int error;
1528
1529 vp = fp->f_vnode;
1530 #ifdef MAC
1531 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1532 AUDIT_ARG_VNODE1(vp);
1533 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1534 VOP_UNLOCK(vp, 0);
1535 if (!error)
1536 #endif
1537
1538 error = VOP_POLL(vp, events, fp->f_cred, td);
1539 return (error);
1540 }
1541
1542 /*
1543 * Acquire the requested lock and then check for validity. LK_RETRY
1544 * permits vn_lock to return doomed vnodes.
1545 */
1546 int
1547 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1548 {
1549 int error;
1550
1551 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1552 ("vn_lock: no locktype"));
1553 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1554 retry:
1555 error = VOP_LOCK1(vp, flags, file, line);
1556 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1557 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1558 ("vn_lock: error %d incompatible with flags %#x", error, flags));
1559
1560 if ((flags & LK_RETRY) == 0) {
1561 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1562 VOP_UNLOCK(vp, 0);
1563 error = ENOENT;
1564 }
1565 } else if (error != 0)
1566 goto retry;
1567 return (error);
1568 }
1569
1570 /*
1571 * File table vnode close routine.
1572 */
1573 static int
1574 vn_closefile(struct file *fp, struct thread *td)
1575 {
1576 struct vnode *vp;
1577 struct flock lf;
1578 int error;
1579 bool ref;
1580
1581 vp = fp->f_vnode;
1582 fp->f_ops = &badfileops;
1583 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1584
1585 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1586
1587 if (__predict_false(ref)) {
1588 lf.l_whence = SEEK_SET;
1589 lf.l_start = 0;
1590 lf.l_len = 0;
1591 lf.l_type = F_UNLCK;
1592 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1593 vrele(vp);
1594 }
1595 return (error);
1596 }
1597
1598 static bool
1599 vn_suspendable(struct mount *mp)
1600 {
1601
1602 return (mp->mnt_op->vfs_susp_clean != NULL);
1603 }
1604
1605 /*
1606 * Preparing to start a filesystem write operation. If the operation is
1607 * permitted, then we bump the count of operations in progress and
1608 * proceed. If a suspend request is in progress, we wait until the
1609 * suspension is over, and then proceed.
1610 */
1611 static int
1612 vn_start_write_locked(struct mount *mp, int flags)
1613 {
1614 int error, mflags;
1615
1616 mtx_assert(MNT_MTX(mp), MA_OWNED);
1617 error = 0;
1618
1619 /*
1620 * Check on status of suspension.
1621 */
1622 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1623 mp->mnt_susp_owner != curthread) {
1624 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1625 (flags & PCATCH) : 0) | (PUSER - 1);
1626 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1627 if (flags & V_NOWAIT) {
1628 error = EWOULDBLOCK;
1629 goto unlock;
1630 }
1631 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1632 "suspfs", 0);
1633 if (error)
1634 goto unlock;
1635 }
1636 }
1637 if (flags & V_XSLEEP)
1638 goto unlock;
1639 mp->mnt_writeopcount++;
1640 unlock:
1641 if (error != 0 || (flags & V_XSLEEP) != 0)
1642 MNT_REL(mp);
1643 MNT_IUNLOCK(mp);
1644 return (error);
1645 }
1646
1647 int
1648 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1649 {
1650 struct mount *mp;
1651 int error;
1652
1653 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1654 ("V_MNTREF requires mp"));
1655
1656 error = 0;
1657 /*
1658 * If a vnode is provided, get and return the mount point that
1659 * to which it will write.
1660 */
1661 if (vp != NULL) {
1662 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1663 *mpp = NULL;
1664 if (error != EOPNOTSUPP)
1665 return (error);
1666 return (0);
1667 }
1668 }
1669 if ((mp = *mpp) == NULL)
1670 return (0);
1671
1672 if (!vn_suspendable(mp)) {
1673 if (vp != NULL || (flags & V_MNTREF) != 0)
1674 vfs_rel(mp);
1675 return (0);
1676 }
1677
1678 /*
1679 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1680 * a vfs_ref().
1681 * As long as a vnode is not provided we need to acquire a
1682 * refcount for the provided mountpoint too, in order to
1683 * emulate a vfs_ref().
1684 */
1685 MNT_ILOCK(mp);
1686 if (vp == NULL && (flags & V_MNTREF) == 0)
1687 MNT_REF(mp);
1688
1689 return (vn_start_write_locked(mp, flags));
1690 }
1691
1692 /*
1693 * Secondary suspension. Used by operations such as vop_inactive
1694 * routines that are needed by the higher level functions. These
1695 * are allowed to proceed until all the higher level functions have
1696 * completed (indicated by mnt_writeopcount dropping to zero). At that
1697 * time, these operations are halted until the suspension is over.
1698 */
1699 int
1700 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1701 {
1702 struct mount *mp;
1703 int error;
1704
1705 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1706 ("V_MNTREF requires mp"));
1707
1708 retry:
1709 if (vp != NULL) {
1710 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1711 *mpp = NULL;
1712 if (error != EOPNOTSUPP)
1713 return (error);
1714 return (0);
1715 }
1716 }
1717 /*
1718 * If we are not suspended or have not yet reached suspended
1719 * mode, then let the operation proceed.
1720 */
1721 if ((mp = *mpp) == NULL)
1722 return (0);
1723
1724 if (!vn_suspendable(mp)) {
1725 if (vp != NULL || (flags & V_MNTREF) != 0)
1726 vfs_rel(mp);
1727 return (0);
1728 }
1729
1730 /*
1731 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1732 * a vfs_ref().
1733 * As long as a vnode is not provided we need to acquire a
1734 * refcount for the provided mountpoint too, in order to
1735 * emulate a vfs_ref().
1736 */
1737 MNT_ILOCK(mp);
1738 if (vp == NULL && (flags & V_MNTREF) == 0)
1739 MNT_REF(mp);
1740 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1741 mp->mnt_secondary_writes++;
1742 mp->mnt_secondary_accwrites++;
1743 MNT_IUNLOCK(mp);
1744 return (0);
1745 }
1746 if (flags & V_NOWAIT) {
1747 MNT_REL(mp);
1748 MNT_IUNLOCK(mp);
1749 return (EWOULDBLOCK);
1750 }
1751 /*
1752 * Wait for the suspension to finish.
1753 */
1754 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1755 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1756 "suspfs", 0);
1757 vfs_rel(mp);
1758 if (error == 0)
1759 goto retry;
1760 return (error);
1761 }
1762
1763 /*
1764 * Filesystem write operation has completed. If we are suspending and this
1765 * operation is the last one, notify the suspender that the suspension is
1766 * now in effect.
1767 */
1768 void
1769 vn_finished_write(struct mount *mp)
1770 {
1771 if (mp == NULL || !vn_suspendable(mp))
1772 return;
1773 MNT_ILOCK(mp);
1774 MNT_REL(mp);
1775 mp->mnt_writeopcount--;
1776 if (mp->mnt_writeopcount < 0)
1777 panic("vn_finished_write: neg cnt");
1778 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1779 mp->mnt_writeopcount <= 0)
1780 wakeup(&mp->mnt_writeopcount);
1781 MNT_IUNLOCK(mp);
1782 }
1783
1784
1785 /*
1786 * Filesystem secondary write operation has completed. If we are
1787 * suspending and this operation is the last one, notify the suspender
1788 * that the suspension is now in effect.
1789 */
1790 void
1791 vn_finished_secondary_write(struct mount *mp)
1792 {
1793 if (mp == NULL || !vn_suspendable(mp))
1794 return;
1795 MNT_ILOCK(mp);
1796 MNT_REL(mp);
1797 mp->mnt_secondary_writes--;
1798 if (mp->mnt_secondary_writes < 0)
1799 panic("vn_finished_secondary_write: neg cnt");
1800 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1801 mp->mnt_secondary_writes <= 0)
1802 wakeup(&mp->mnt_secondary_writes);
1803 MNT_IUNLOCK(mp);
1804 }
1805
1806
1807
1808 /*
1809 * Request a filesystem to suspend write operations.
1810 */
1811 int
1812 vfs_write_suspend(struct mount *mp, int flags)
1813 {
1814 int error;
1815
1816 MPASS(vn_suspendable(mp));
1817
1818 MNT_ILOCK(mp);
1819 if (mp->mnt_susp_owner == curthread) {
1820 MNT_IUNLOCK(mp);
1821 return (EALREADY);
1822 }
1823 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1824 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1825
1826 /*
1827 * Unmount holds a write reference on the mount point. If we
1828 * own busy reference and drain for writers, we deadlock with
1829 * the reference draining in the unmount path. Callers of
1830 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1831 * vfs_busy() reference is owned and caller is not in the
1832 * unmount context.
1833 */
1834 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1835 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1836 MNT_IUNLOCK(mp);
1837 return (EBUSY);
1838 }
1839
1840 mp->mnt_kern_flag |= MNTK_SUSPEND;
1841 mp->mnt_susp_owner = curthread;
1842 if (mp->mnt_writeopcount > 0)
1843 (void) msleep(&mp->mnt_writeopcount,
1844 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1845 else
1846 MNT_IUNLOCK(mp);
1847 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1848 vfs_write_resume(mp, 0);
1849 return (error);
1850 }
1851
1852 /*
1853 * Request a filesystem to resume write operations.
1854 */
1855 void
1856 vfs_write_resume(struct mount *mp, int flags)
1857 {
1858
1859 MPASS(vn_suspendable(mp));
1860
1861 MNT_ILOCK(mp);
1862 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1863 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1864 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1865 MNTK_SUSPENDED);
1866 mp->mnt_susp_owner = NULL;
1867 wakeup(&mp->mnt_writeopcount);
1868 wakeup(&mp->mnt_flag);
1869 curthread->td_pflags &= ~TDP_IGNSUSP;
1870 if ((flags & VR_START_WRITE) != 0) {
1871 MNT_REF(mp);
1872 mp->mnt_writeopcount++;
1873 }
1874 MNT_IUNLOCK(mp);
1875 if ((flags & VR_NO_SUSPCLR) == 0)
1876 VFS_SUSP_CLEAN(mp);
1877 } else if ((flags & VR_START_WRITE) != 0) {
1878 MNT_REF(mp);
1879 vn_start_write_locked(mp, 0);
1880 } else {
1881 MNT_IUNLOCK(mp);
1882 }
1883 }
1884
1885 /*
1886 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1887 * methods.
1888 */
1889 int
1890 vfs_write_suspend_umnt(struct mount *mp)
1891 {
1892 int error;
1893
1894 MPASS(vn_suspendable(mp));
1895 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1896 ("vfs_write_suspend_umnt: recursed"));
1897
1898 /* dounmount() already called vn_start_write(). */
1899 for (;;) {
1900 vn_finished_write(mp);
1901 error = vfs_write_suspend(mp, 0);
1902 if (error != 0) {
1903 vn_start_write(NULL, &mp, V_WAIT);
1904 return (error);
1905 }
1906 MNT_ILOCK(mp);
1907 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1908 break;
1909 MNT_IUNLOCK(mp);
1910 vn_start_write(NULL, &mp, V_WAIT);
1911 }
1912 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1913 wakeup(&mp->mnt_flag);
1914 MNT_IUNLOCK(mp);
1915 curthread->td_pflags |= TDP_IGNSUSP;
1916 return (0);
1917 }
1918
1919 /*
1920 * Implement kqueues for files by translating it to vnode operation.
1921 */
1922 static int
1923 vn_kqfilter(struct file *fp, struct knote *kn)
1924 {
1925
1926 return (VOP_KQFILTER(fp->f_vnode, kn));
1927 }
1928
1929 /*
1930 * Simplified in-kernel wrapper calls for extended attribute access.
1931 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1932 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1933 */
1934 int
1935 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1936 const char *attrname, int *buflen, char *buf, struct thread *td)
1937 {
1938 struct uio auio;
1939 struct iovec iov;
1940 int error;
1941
1942 iov.iov_len = *buflen;
1943 iov.iov_base = buf;
1944
1945 auio.uio_iov = &iov;
1946 auio.uio_iovcnt = 1;
1947 auio.uio_rw = UIO_READ;
1948 auio.uio_segflg = UIO_SYSSPACE;
1949 auio.uio_td = td;
1950 auio.uio_offset = 0;
1951 auio.uio_resid = *buflen;
1952
1953 if ((ioflg & IO_NODELOCKED) == 0)
1954 vn_lock(vp, LK_SHARED | LK_RETRY);
1955
1956 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1957
1958 /* authorize attribute retrieval as kernel */
1959 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1960 td);
1961
1962 if ((ioflg & IO_NODELOCKED) == 0)
1963 VOP_UNLOCK(vp, 0);
1964
1965 if (error == 0) {
1966 *buflen = *buflen - auio.uio_resid;
1967 }
1968
1969 return (error);
1970 }
1971
1972 /*
1973 * XXX failure mode if partially written?
1974 */
1975 int
1976 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1977 const char *attrname, int buflen, char *buf, struct thread *td)
1978 {
1979 struct uio auio;
1980 struct iovec iov;
1981 struct mount *mp;
1982 int error;
1983
1984 iov.iov_len = buflen;
1985 iov.iov_base = buf;
1986
1987 auio.uio_iov = &iov;
1988 auio.uio_iovcnt = 1;
1989 auio.uio_rw = UIO_WRITE;
1990 auio.uio_segflg = UIO_SYSSPACE;
1991 auio.uio_td = td;
1992 auio.uio_offset = 0;
1993 auio.uio_resid = buflen;
1994
1995 if ((ioflg & IO_NODELOCKED) == 0) {
1996 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1997 return (error);
1998 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1999 }
2000
2001 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2002
2003 /* authorize attribute setting as kernel */
2004 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2005
2006 if ((ioflg & IO_NODELOCKED) == 0) {
2007 vn_finished_write(mp);
2008 VOP_UNLOCK(vp, 0);
2009 }
2010
2011 return (error);
2012 }
2013
2014 int
2015 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2016 const char *attrname, struct thread *td)
2017 {
2018 struct mount *mp;
2019 int error;
2020
2021 if ((ioflg & IO_NODELOCKED) == 0) {
2022 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2023 return (error);
2024 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2025 }
2026
2027 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2028
2029 /* authorize attribute removal as kernel */
2030 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2031 if (error == EOPNOTSUPP)
2032 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2033 NULL, td);
2034
2035 if ((ioflg & IO_NODELOCKED) == 0) {
2036 vn_finished_write(mp);
2037 VOP_UNLOCK(vp, 0);
2038 }
2039
2040 return (error);
2041 }
2042
2043 static int
2044 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2045 struct vnode **rvp)
2046 {
2047
2048 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2049 }
2050
2051 int
2052 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2053 {
2054
2055 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2056 lkflags, rvp));
2057 }
2058
2059 int
2060 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2061 int lkflags, struct vnode **rvp)
2062 {
2063 struct mount *mp;
2064 int ltype, error;
2065
2066 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2067 mp = vp->v_mount;
2068 ltype = VOP_ISLOCKED(vp);
2069 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2070 ("vn_vget_ino: vp not locked"));
2071 error = vfs_busy(mp, MBF_NOWAIT);
2072 if (error != 0) {
2073 vfs_ref(mp);
2074 VOP_UNLOCK(vp, 0);
2075 error = vfs_busy(mp, 0);
2076 vn_lock(vp, ltype | LK_RETRY);
2077 vfs_rel(mp);
2078 if (error != 0)
2079 return (ENOENT);
2080 if (vp->v_iflag & VI_DOOMED) {
2081 vfs_unbusy(mp);
2082 return (ENOENT);
2083 }
2084 }
2085 VOP_UNLOCK(vp, 0);
2086 error = alloc(mp, alloc_arg, lkflags, rvp);
2087 vfs_unbusy(mp);
2088 if (error != 0 || *rvp != vp)
2089 vn_lock(vp, ltype | LK_RETRY);
2090 if (vp->v_iflag & VI_DOOMED) {
2091 if (error == 0) {
2092 if (*rvp == vp)
2093 vunref(vp);
2094 else
2095 vput(*rvp);
2096 }
2097 error = ENOENT;
2098 }
2099 return (error);
2100 }
2101
2102 int
2103 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2104 struct thread *td)
2105 {
2106
2107 if (vp->v_type != VREG || td == NULL)
2108 return (0);
2109 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2110 lim_cur(td, RLIMIT_FSIZE)) {
2111 PROC_LOCK(td->td_proc);
2112 kern_psignal(td->td_proc, SIGXFSZ);
2113 PROC_UNLOCK(td->td_proc);
2114 return (EFBIG);
2115 }
2116 return (0);
2117 }
2118
2119 int
2120 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2121 struct thread *td)
2122 {
2123 struct vnode *vp;
2124
2125 vp = fp->f_vnode;
2126 #ifdef AUDIT
2127 vn_lock(vp, LK_SHARED | LK_RETRY);
2128 AUDIT_ARG_VNODE1(vp);
2129 VOP_UNLOCK(vp, 0);
2130 #endif
2131 return (setfmode(td, active_cred, vp, mode));
2132 }
2133
2134 int
2135 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2136 struct thread *td)
2137 {
2138 struct vnode *vp;
2139
2140 vp = fp->f_vnode;
2141 #ifdef AUDIT
2142 vn_lock(vp, LK_SHARED | LK_RETRY);
2143 AUDIT_ARG_VNODE1(vp);
2144 VOP_UNLOCK(vp, 0);
2145 #endif
2146 return (setfown(td, active_cred, vp, uid, gid));
2147 }
2148
2149 void
2150 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2151 {
2152 vm_object_t object;
2153
2154 if ((object = vp->v_object) == NULL)
2155 return;
2156 VM_OBJECT_WLOCK(object);
2157 vm_object_page_remove(object, start, end, 0);
2158 VM_OBJECT_WUNLOCK(object);
2159 }
2160
2161 int
2162 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2163 {
2164 struct vattr va;
2165 daddr_t bn, bnp;
2166 uint64_t bsize;
2167 off_t noff;
2168 int error;
2169
2170 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2171 ("Wrong command %lu", cmd));
2172
2173 if (vn_lock(vp, LK_SHARED) != 0)
2174 return (EBADF);
2175 if (vp->v_type != VREG) {
2176 error = ENOTTY;
2177 goto unlock;
2178 }
2179 error = VOP_GETATTR(vp, &va, cred);
2180 if (error != 0)
2181 goto unlock;
2182 noff = *off;
2183 if (noff >= va.va_size) {
2184 error = ENXIO;
2185 goto unlock;
2186 }
2187 bsize = vp->v_mount->mnt_stat.f_iosize;
2188 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize -
2189 noff % bsize) {
2190 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2191 if (error == EOPNOTSUPP) {
2192 error = ENOTTY;
2193 goto unlock;
2194 }
2195 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2196 (bnp != -1 && cmd == FIOSEEKDATA)) {
2197 noff = bn * bsize;
2198 if (noff < *off)
2199 noff = *off;
2200 goto unlock;
2201 }
2202 }
2203 if (noff > va.va_size)
2204 noff = va.va_size;
2205 /* noff == va.va_size. There is an implicit hole at the end of file. */
2206 if (cmd == FIOSEEKDATA)
2207 error = ENXIO;
2208 unlock:
2209 VOP_UNLOCK(vp, 0);
2210 if (error == 0)
2211 *off = noff;
2212 return (error);
2213 }
2214
2215 int
2216 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2217 {
2218 struct ucred *cred;
2219 struct vnode *vp;
2220 struct vattr vattr;
2221 off_t foffset, size;
2222 int error, noneg;
2223
2224 cred = td->td_ucred;
2225 vp = fp->f_vnode;
2226 foffset = foffset_lock(fp, 0);
2227 noneg = (vp->v_type != VCHR);
2228 error = 0;
2229 switch (whence) {
2230 case L_INCR:
2231 if (noneg &&
2232 (foffset < 0 ||
2233 (offset > 0 && foffset > OFF_MAX - offset))) {
2234 error = EOVERFLOW;
2235 break;
2236 }
2237 offset += foffset;
2238 break;
2239 case L_XTND:
2240 vn_lock(vp, LK_SHARED | LK_RETRY);
2241 error = VOP_GETATTR(vp, &vattr, cred);
2242 VOP_UNLOCK(vp, 0);
2243 if (error)
2244 break;
2245
2246 /*
2247 * If the file references a disk device, then fetch
2248 * the media size and use that to determine the ending
2249 * offset.
2250 */
2251 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2252 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2253 vattr.va_size = size;
2254 if (noneg &&
2255 (vattr.va_size > OFF_MAX ||
2256 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2257 error = EOVERFLOW;
2258 break;
2259 }
2260 offset += vattr.va_size;
2261 break;
2262 case L_SET:
2263 break;
2264 case SEEK_DATA:
2265 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2266 break;
2267 case SEEK_HOLE:
2268 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2269 break;
2270 default:
2271 error = EINVAL;
2272 }
2273 if (error == 0 && noneg && offset < 0)
2274 error = EINVAL;
2275 if (error != 0)
2276 goto drop;
2277 VFS_KNOTE_UNLOCKED(vp, 0);
2278 td->td_uretoff.tdu_off = offset;
2279 drop:
2280 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2281 return (error);
2282 }
2283
2284 int
2285 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2286 struct thread *td)
2287 {
2288 int error;
2289
2290 /*
2291 * Grant permission if the caller is the owner of the file, or
2292 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2293 * on the file. If the time pointer is null, then write
2294 * permission on the file is also sufficient.
2295 *
2296 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2297 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2298 * will be allowed to set the times [..] to the current
2299 * server time.
2300 */
2301 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2302 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2303 error = VOP_ACCESS(vp, VWRITE, cred, td);
2304 return (error);
2305 }
2306
2307 int
2308 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2309 {
2310 struct vnode *vp;
2311 int error;
2312
2313 if (fp->f_type == DTYPE_FIFO)
2314 kif->kf_type = KF_TYPE_FIFO;
2315 else
2316 kif->kf_type = KF_TYPE_VNODE;
2317 vp = fp->f_vnode;
2318 vref(vp);
2319 FILEDESC_SUNLOCK(fdp);
2320 error = vn_fill_kinfo_vnode(vp, kif);
2321 vrele(vp);
2322 FILEDESC_SLOCK(fdp);
2323 return (error);
2324 }
2325
2326 static inline void
2327 vn_fill_junk(struct kinfo_file *kif)
2328 {
2329 size_t len, olen;
2330
2331 /*
2332 * Simulate vn_fullpath returning changing values for a given
2333 * vp during e.g. coredump.
2334 */
2335 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2336 olen = strlen(kif->kf_path);
2337 if (len < olen)
2338 strcpy(&kif->kf_path[len - 1], "$");
2339 else
2340 for (; olen < len; olen++)
2341 strcpy(&kif->kf_path[olen], "A");
2342 }
2343
2344 int
2345 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2346 {
2347 struct vattr va;
2348 char *fullpath, *freepath;
2349 int error;
2350
2351 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2352 freepath = NULL;
2353 fullpath = "-";
2354 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2355 if (error == 0) {
2356 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2357 }
2358 if (freepath != NULL)
2359 free(freepath, M_TEMP);
2360
2361 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2362 vn_fill_junk(kif);
2363 );
2364
2365 /*
2366 * Retrieve vnode attributes.
2367 */
2368 va.va_fsid = VNOVAL;
2369 va.va_rdev = NODEV;
2370 vn_lock(vp, LK_SHARED | LK_RETRY);
2371 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2372 VOP_UNLOCK(vp, 0);
2373 if (error != 0)
2374 return (error);
2375 if (va.va_fsid != VNOVAL)
2376 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2377 else
2378 kif->kf_un.kf_file.kf_file_fsid =
2379 vp->v_mount->mnt_stat.f_fsid.val[0];
2380 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2381 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2382 kif->kf_un.kf_file.kf_file_size = va.va_size;
2383 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2384 return (0);
2385 }
2386
2387 int
2388 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2389 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2390 struct thread *td)
2391 {
2392 #ifdef HWPMC_HOOKS
2393 struct pmckern_map_in pkm;
2394 #endif
2395 struct mount *mp;
2396 struct vnode *vp;
2397 vm_object_t object;
2398 vm_prot_t maxprot;
2399 boolean_t writecounted;
2400 int error;
2401
2402 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2403 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2404 /*
2405 * POSIX shared-memory objects are defined to have
2406 * kernel persistence, and are not defined to support
2407 * read(2)/write(2) -- or even open(2). Thus, we can
2408 * use MAP_ASYNC to trade on-disk coherence for speed.
2409 * The shm_open(3) library routine turns on the FPOSIXSHM
2410 * flag to request this behavior.
2411 */
2412 if ((fp->f_flag & FPOSIXSHM) != 0)
2413 flags |= MAP_NOSYNC;
2414 #endif
2415 vp = fp->f_vnode;
2416
2417 /*
2418 * Ensure that file and memory protections are
2419 * compatible. Note that we only worry about
2420 * writability if mapping is shared; in this case,
2421 * current and max prot are dictated by the open file.
2422 * XXX use the vnode instead? Problem is: what
2423 * credentials do we use for determination? What if
2424 * proc does a setuid?
2425 */
2426 mp = vp->v_mount;
2427 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2428 maxprot = VM_PROT_NONE;
2429 if ((prot & VM_PROT_EXECUTE) != 0)
2430 return (EACCES);
2431 } else
2432 maxprot = VM_PROT_EXECUTE;
2433 if ((fp->f_flag & FREAD) != 0)
2434 maxprot |= VM_PROT_READ;
2435 else if ((prot & VM_PROT_READ) != 0)
2436 return (EACCES);
2437
2438 /*
2439 * If we are sharing potential changes via MAP_SHARED and we
2440 * are trying to get write permission although we opened it
2441 * without asking for it, bail out.
2442 */
2443 if ((flags & MAP_SHARED) != 0) {
2444 if ((fp->f_flag & FWRITE) != 0)
2445 maxprot |= VM_PROT_WRITE;
2446 else if ((prot & VM_PROT_WRITE) != 0)
2447 return (EACCES);
2448 } else {
2449 maxprot |= VM_PROT_WRITE;
2450 cap_maxprot |= VM_PROT_WRITE;
2451 }
2452 maxprot &= cap_maxprot;
2453
2454 /*
2455 * For regular files and shared memory, POSIX requires that
2456 * the value of foff be a legitimate offset within the data
2457 * object. In particular, negative offsets are invalid.
2458 * Blocking negative offsets and overflows here avoids
2459 * possible wraparound or user-level access into reserved
2460 * ranges of the data object later. In contrast, POSIX does
2461 * not dictate how offsets are used by device drivers, so in
2462 * the case of a device mapping a negative offset is passed
2463 * on.
2464 */
2465 if (
2466 #ifdef _LP64
2467 size > OFF_MAX ||
2468 #endif
2469 foff < 0 || foff > OFF_MAX - size)
2470 return (EINVAL);
2471
2472 writecounted = FALSE;
2473 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2474 &foff, &object, &writecounted);
2475 if (error != 0)
2476 return (error);
2477 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2478 foff, writecounted, td);
2479 if (error != 0) {
2480 /*
2481 * If this mapping was accounted for in the vnode's
2482 * writecount, then undo that now.
2483 */
2484 if (writecounted)
2485 vnode_pager_release_writecount(object, 0, size);
2486 vm_object_deallocate(object);
2487 }
2488 #ifdef HWPMC_HOOKS
2489 /* Inform hwpmc(4) if an executable is being mapped. */
2490 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2491 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2492 pkm.pm_file = vp;
2493 pkm.pm_address = (uintptr_t) *addr;
2494 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2495 }
2496 }
2497 #endif
2498 return (error);
2499 }
Cache object: 22f38b095491e4d5275b69bc09fe5de6
|