1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1994, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
25 * Copyright 2015, Joyent, Inc.
26 */
27
28 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
29 /* All Rights Reserved */
30
31 /*
32 * Portions of this source code were derived from Berkeley 4.3 BSD
33 * under license from the Regents of the University of California.
34 */
35
36
37 #include <sys/param.h>
38 #include <sys/isa_defs.h>
39 #include <sys/types.h>
40 #include <sys/sysmacros.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/fcntl.h>
44 #include <sys/flock.h>
45 #include <sys/vnode.h>
46 #include <sys/file.h>
47 #include <sys/mode.h>
48 #include <sys/proc.h>
49 #include <sys/filio.h>
50 #include <sys/share.h>
51 #include <sys/debug.h>
52 #include <sys/rctl.h>
53 #include <sys/nbmlock.h>
54
55 #include <sys/cmn_err.h>
56
57 /* This is global so that it can be used by brand emulation. */
58 int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
59 static int flock_get_start(vnode_t *, flock64_t *, offset_t, u_offset_t *);
60 static void fd_too_big(proc_t *);
61
62 /*
63 * File control.
64 */
65 int
66 fcntl(int fdes, int cmd, intptr_t arg)
67 {
68 int iarg;
69 int error = 0;
70 int retval;
71 proc_t *p;
72 file_t *fp;
73 vnode_t *vp;
74 u_offset_t offset;
75 u_offset_t start;
76 struct vattr vattr;
77 int in_crit;
78 int flag;
79 struct flock sbf;
80 struct flock64 bf;
81 struct o_flock obf;
82 struct flock64_32 bf64_32;
83 struct fshare fsh;
84 struct shrlock shr;
85 struct shr_locowner shr_own;
86 offset_t maxoffset;
87 model_t datamodel;
88 int fdres;
89
90 #if defined(_ILP32) && !defined(lint) && defined(_SYSCALL32)
91 ASSERT(sizeof (struct flock) == sizeof (struct flock32));
92 ASSERT(sizeof (struct flock64) == sizeof (struct flock64_32));
93 #endif
94 #if defined(_LP64) && !defined(lint) && defined(_SYSCALL32)
95 ASSERT(sizeof (struct flock) == sizeof (struct flock64_64));
96 ASSERT(sizeof (struct flock64) == sizeof (struct flock64_64));
97 #endif
98
99 /*
100 * First, for speed, deal with the subset of cases
101 * that do not require getf() / releasef().
102 */
103 switch (cmd) {
104 case F_GETFD:
105 if ((error = f_getfd_error(fdes, &flag)) == 0)
106 retval = flag;
107 goto out;
108
109 case F_SETFD:
110 error = f_setfd_error(fdes, (int)arg);
111 retval = 0;
112 goto out;
113
114 case F_GETFL:
115 if ((error = f_getfl(fdes, &flag)) == 0) {
116 retval = (flag & (FMASK | FASYNC));
117 if ((flag & (FSEARCH | FEXEC)) == 0)
118 retval += FOPEN;
119 else
120 retval |= (flag & (FSEARCH | FEXEC));
121 }
122 goto out;
123
124 case F_GETXFL:
125 if ((error = f_getfl(fdes, &flag)) == 0) {
126 retval = flag;
127 if ((flag & (FSEARCH | FEXEC)) == 0)
128 retval += FOPEN;
129 }
130 goto out;
131
132 case F_BADFD:
133 if ((error = f_badfd(fdes, &fdres, (int)arg)) == 0)
134 retval = fdres;
135 goto out;
136 }
137
138 /*
139 * Second, for speed, deal with the subset of cases that
140 * require getf() / releasef() but do not require copyin.
141 */
142 if ((fp = getf(fdes)) == NULL) {
143 error = EBADF;
144 goto out;
145 }
146 iarg = (int)arg;
147
148 switch (cmd) {
149 case F_DUPFD:
150 case F_DUPFD_CLOEXEC:
151 p = curproc;
152 if ((uint_t)iarg >= p->p_fno_ctl) {
153 if (iarg >= 0)
154 fd_too_big(p);
155 error = EINVAL;
156 goto done;
157 }
158 /*
159 * We need to increment the f_count reference counter
160 * before allocating a new file descriptor.
161 * Doing it other way round opens a window for race condition
162 * with closeandsetf() on the target file descriptor which can
163 * close the file still referenced by the original
164 * file descriptor.
165 */
166 mutex_enter(&fp->f_tlock);
167 fp->f_count++;
168 mutex_exit(&fp->f_tlock);
169 if ((retval = ufalloc_file(iarg, fp)) == -1) {
170 /*
171 * New file descriptor can't be allocated.
172 * Revert the reference count.
173 */
174 mutex_enter(&fp->f_tlock);
175 fp->f_count--;
176 mutex_exit(&fp->f_tlock);
177 error = EMFILE;
178 } else {
179 if (cmd == F_DUPFD_CLOEXEC) {
180 f_setfd(retval, FD_CLOEXEC);
181 }
182 }
183 goto done;
184
185 case F_DUP2FD_CLOEXEC:
186 if (fdes == iarg) {
187 error = EINVAL;
188 goto done;
189 }
190
191 /*FALLTHROUGH*/
192
193 case F_DUP2FD:
194 p = curproc;
195 if (fdes == iarg) {
196 retval = iarg;
197 } else if ((uint_t)iarg >= p->p_fno_ctl) {
198 if (iarg >= 0)
199 fd_too_big(p);
200 error = EBADF;
201 } else {
202 /*
203 * We can't hold our getf(fdes) across the call to
204 * closeandsetf() because it creates a window for
205 * deadlock: if one thread is doing dup2(a, b) while
206 * another is doing dup2(b, a), each one will block
207 * waiting for the other to call releasef(). The
208 * solution is to increment the file reference count
209 * (which we have to do anyway), then releasef(fdes),
210 * then closeandsetf(). Incrementing f_count ensures
211 * that fp won't disappear after we call releasef().
212 * When closeandsetf() fails, we try avoid calling
213 * closef() because of all the side effects.
214 */
215 mutex_enter(&fp->f_tlock);
216 fp->f_count++;
217 mutex_exit(&fp->f_tlock);
218 releasef(fdes);
219 if ((error = closeandsetf(iarg, fp)) == 0) {
220 if (cmd == F_DUP2FD_CLOEXEC) {
221 f_setfd(iarg, FD_CLOEXEC);
222 }
223 retval = iarg;
224 } else {
225 mutex_enter(&fp->f_tlock);
226 if (fp->f_count > 1) {
227 fp->f_count--;
228 mutex_exit(&fp->f_tlock);
229 } else {
230 mutex_exit(&fp->f_tlock);
231 (void) closef(fp);
232 }
233 }
234 goto out;
235 }
236 goto done;
237
238 case F_SETFL:
239 vp = fp->f_vnode;
240 flag = fp->f_flag;
241 if ((iarg & (FNONBLOCK|FNDELAY)) == (FNONBLOCK|FNDELAY))
242 iarg &= ~FNDELAY;
243 if ((error = VOP_SETFL(vp, flag, iarg, fp->f_cred, NULL)) ==
244 0) {
245 iarg &= FMASK;
246 mutex_enter(&fp->f_tlock);
247 fp->f_flag &= ~FMASK | (FREAD|FWRITE);
248 fp->f_flag |= (iarg - FOPEN) & ~(FREAD|FWRITE);
249 mutex_exit(&fp->f_tlock);
250 }
251 retval = 0;
252 goto done;
253 }
254
255 /*
256 * Finally, deal with the expensive cases.
257 */
258 retval = 0;
259 in_crit = 0;
260 maxoffset = MAXOFF_T;
261 datamodel = DATAMODEL_NATIVE;
262 #if defined(_SYSCALL32_IMPL)
263 if ((datamodel = get_udatamodel()) == DATAMODEL_ILP32)
264 maxoffset = MAXOFF32_T;
265 #endif
266
267 vp = fp->f_vnode;
268 flag = fp->f_flag;
269 offset = fp->f_offset;
270
271 switch (cmd) {
272 /*
273 * The file system and vnode layers understand and implement
274 * locking with flock64 structures. So here once we pass through
275 * the test for compatibility as defined by LFS API, (for F_SETLK,
276 * F_SETLKW, F_GETLK, F_GETLKW, F_OFD_GETLK, F_OFD_SETLK, F_OFD_SETLKW,
277 * F_FREESP) we transform the flock structure to a flock64 structure
278 * and send it to the lower layers. Similarly in case of GETLK and
279 * OFD_GETLK the returned flock64 structure is transformed to a flock
280 * structure if everything fits in nicely, otherwise we return
281 * EOVERFLOW.
282 */
283
284 case F_GETLK:
285 case F_O_GETLK:
286 case F_SETLK:
287 case F_SETLKW:
288 case F_SETLK_NBMAND:
289 case F_OFD_GETLK:
290 case F_OFD_SETLK:
291 case F_OFD_SETLKW:
292 case F_FLOCK:
293 case F_FLOCKW:
294
295 /*
296 * Copy in input fields only.
297 */
298
299 if (cmd == F_O_GETLK) {
300 if (datamodel != DATAMODEL_ILP32) {
301 error = EINVAL;
302 break;
303 }
304
305 if (copyin((void *)arg, &obf, sizeof (obf))) {
306 error = EFAULT;
307 break;
308 }
309 bf.l_type = obf.l_type;
310 bf.l_whence = obf.l_whence;
311 bf.l_start = (off64_t)obf.l_start;
312 bf.l_len = (off64_t)obf.l_len;
313 bf.l_sysid = (int)obf.l_sysid;
314 bf.l_pid = obf.l_pid;
315 } else if (datamodel == DATAMODEL_NATIVE) {
316 if (copyin((void *)arg, &sbf, sizeof (sbf))) {
317 error = EFAULT;
318 break;
319 }
320 /*
321 * XXX In an LP64 kernel with an LP64 application
322 * there's no need to do a structure copy here
323 * struct flock == struct flock64. However,
324 * we did it this way to avoid more conditional
325 * compilation.
326 */
327 bf.l_type = sbf.l_type;
328 bf.l_whence = sbf.l_whence;
329 bf.l_start = (off64_t)sbf.l_start;
330 bf.l_len = (off64_t)sbf.l_len;
331 bf.l_sysid = sbf.l_sysid;
332 bf.l_pid = sbf.l_pid;
333 }
334 #if defined(_SYSCALL32_IMPL)
335 else {
336 struct flock32 sbf32;
337 if (copyin((void *)arg, &sbf32, sizeof (sbf32))) {
338 error = EFAULT;
339 break;
340 }
341 bf.l_type = sbf32.l_type;
342 bf.l_whence = sbf32.l_whence;
343 bf.l_start = (off64_t)sbf32.l_start;
344 bf.l_len = (off64_t)sbf32.l_len;
345 bf.l_sysid = sbf32.l_sysid;
346 bf.l_pid = sbf32.l_pid;
347 }
348 #endif /* _SYSCALL32_IMPL */
349
350 /*
351 * 64-bit support: check for overflow for 32-bit lock ops
352 */
353 if ((error = flock_check(vp, &bf, offset, maxoffset)) != 0)
354 break;
355
356 if (cmd == F_FLOCK || cmd == F_FLOCKW) {
357 /* FLOCK* locking is always over the entire file. */
358 if (bf.l_whence != 0 || bf.l_start != 0 ||
359 bf.l_len != 0) {
360 error = EINVAL;
361 break;
362 }
363 if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) {
364 error = EINVAL;
365 break;
366 }
367 }
368
369 if (cmd == F_OFD_GETLK || cmd == F_OFD_SETLK ||
370 cmd == F_OFD_SETLKW) {
371 /*
372 * TBD OFD-style locking is currently limited to
373 * covering the entire file.
374 */
375 if (bf.l_whence != 0 || bf.l_start != 0 ||
376 bf.l_len != 0) {
377 error = EINVAL;
378 break;
379 }
380 }
381
382 /*
383 * Not all of the filesystems understand F_O_GETLK, and
384 * there's no need for them to know. Map it to F_GETLK.
385 *
386 * The *_frlock functions in the various file systems basically
387 * do some validation and then funnel everything through the
388 * fs_frlock function. For OFD-style locks fs_frlock will do
389 * nothing so that once control returns here we can call the
390 * ofdlock function with the correct fp. For OFD-style locks
391 * the unsupported remote file systems, such as NFS, detect and
392 * reject the OFD-style cmd argument.
393 */
394 if ((error = VOP_FRLOCK(vp, (cmd == F_O_GETLK) ? F_GETLK : cmd,
395 &bf, flag, offset, NULL, fp->f_cred, NULL)) != 0)
396 break;
397
398 if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK ||
399 cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
400 /*
401 * This is an OFD-style lock so we need to handle it
402 * here. Because OFD-style locks are associated with
403 * the file_t we didn't have enough info down the
404 * VOP_FRLOCK path immediately above.
405 */
406 if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0)
407 break;
408 }
409
410 /*
411 * If command is GETLK and no lock is found, only
412 * the type field is changed.
413 */
414 if ((cmd == F_O_GETLK || cmd == F_GETLK ||
415 cmd == F_OFD_GETLK) && bf.l_type == F_UNLCK) {
416 /* l_type always first entry, always a short */
417 if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
418 sizeof (bf.l_type)))
419 error = EFAULT;
420 break;
421 }
422
423 if (cmd == F_O_GETLK) {
424 /*
425 * Return an SVR3 flock structure to the user.
426 */
427 obf.l_type = (int16_t)bf.l_type;
428 obf.l_whence = (int16_t)bf.l_whence;
429 obf.l_start = (int32_t)bf.l_start;
430 obf.l_len = (int32_t)bf.l_len;
431 if (bf.l_sysid > SHRT_MAX || bf.l_pid > SHRT_MAX) {
432 /*
433 * One or both values for the above fields
434 * is too large to store in an SVR3 flock
435 * structure.
436 */
437 error = EOVERFLOW;
438 break;
439 }
440 obf.l_sysid = (int16_t)bf.l_sysid;
441 obf.l_pid = (int16_t)bf.l_pid;
442 if (copyout(&obf, (void *)arg, sizeof (obf)))
443 error = EFAULT;
444 } else if (cmd == F_GETLK || cmd == F_OFD_GETLK) {
445 /*
446 * Copy out SVR4 flock.
447 */
448 int i;
449
450 if (bf.l_start > maxoffset || bf.l_len > maxoffset) {
451 error = EOVERFLOW;
452 break;
453 }
454
455 if (datamodel == DATAMODEL_NATIVE) {
456 for (i = 0; i < 4; i++)
457 sbf.l_pad[i] = 0;
458 /*
459 * XXX In an LP64 kernel with an LP64
460 * application there's no need to do a
461 * structure copy here as currently
462 * struct flock == struct flock64.
463 * We did it this way to avoid more
464 * conditional compilation.
465 */
466 sbf.l_type = bf.l_type;
467 sbf.l_whence = bf.l_whence;
468 sbf.l_start = (off_t)bf.l_start;
469 sbf.l_len = (off_t)bf.l_len;
470 sbf.l_sysid = bf.l_sysid;
471 sbf.l_pid = bf.l_pid;
472 if (copyout(&sbf, (void *)arg, sizeof (sbf)))
473 error = EFAULT;
474 }
475 #if defined(_SYSCALL32_IMPL)
476 else {
477 struct flock32 sbf32;
478 if (bf.l_start > MAXOFF32_T ||
479 bf.l_len > MAXOFF32_T) {
480 error = EOVERFLOW;
481 break;
482 }
483 for (i = 0; i < 4; i++)
484 sbf32.l_pad[i] = 0;
485 sbf32.l_type = (int16_t)bf.l_type;
486 sbf32.l_whence = (int16_t)bf.l_whence;
487 sbf32.l_start = (off32_t)bf.l_start;
488 sbf32.l_len = (off32_t)bf.l_len;
489 sbf32.l_sysid = (int32_t)bf.l_sysid;
490 sbf32.l_pid = (pid32_t)bf.l_pid;
491 if (copyout(&sbf32,
492 (void *)arg, sizeof (sbf32)))
493 error = EFAULT;
494 }
495 #endif
496 }
497 break;
498
499 case F_CHKFL:
500 /*
501 * This is for internal use only, to allow the vnode layer
502 * to validate a flags setting before applying it. User
503 * programs can't issue it.
504 */
505 error = EINVAL;
506 break;
507
508 case F_ALLOCSP:
509 case F_FREESP:
510 case F_ALLOCSP64:
511 case F_FREESP64:
512 /*
513 * Test for not-a-regular-file (and returning EINVAL)
514 * before testing for open-for-writing (and returning EBADF).
515 * This is relied upon by posix_fallocate() in libc.
516 */
517 if (vp->v_type != VREG) {
518 error = EINVAL;
519 break;
520 }
521
522 if ((flag & FWRITE) == 0) {
523 error = EBADF;
524 break;
525 }
526
527 if (datamodel != DATAMODEL_ILP32 &&
528 (cmd == F_ALLOCSP64 || cmd == F_FREESP64)) {
529 error = EINVAL;
530 break;
531 }
532
533 #if defined(_ILP32) || defined(_SYSCALL32_IMPL)
534 if (datamodel == DATAMODEL_ILP32 &&
535 (cmd == F_ALLOCSP || cmd == F_FREESP)) {
536 struct flock32 sbf32;
537 /*
538 * For compatibility we overlay an SVR3 flock on an SVR4
539 * flock. This works because the input field offsets
540 * in "struct flock" were preserved.
541 */
542 if (copyin((void *)arg, &sbf32, sizeof (sbf32))) {
543 error = EFAULT;
544 break;
545 } else {
546 bf.l_type = sbf32.l_type;
547 bf.l_whence = sbf32.l_whence;
548 bf.l_start = (off64_t)sbf32.l_start;
549 bf.l_len = (off64_t)sbf32.l_len;
550 bf.l_sysid = sbf32.l_sysid;
551 bf.l_pid = sbf32.l_pid;
552 }
553 }
554 #endif /* _ILP32 || _SYSCALL32_IMPL */
555
556 #if defined(_LP64)
557 if (datamodel == DATAMODEL_LP64 &&
558 (cmd == F_ALLOCSP || cmd == F_FREESP)) {
559 if (copyin((void *)arg, &bf, sizeof (bf))) {
560 error = EFAULT;
561 break;
562 }
563 }
564 #endif /* defined(_LP64) */
565
566 #if !defined(_LP64) || defined(_SYSCALL32_IMPL)
567 if (datamodel == DATAMODEL_ILP32 &&
568 (cmd == F_ALLOCSP64 || cmd == F_FREESP64)) {
569 if (copyin((void *)arg, &bf64_32, sizeof (bf64_32))) {
570 error = EFAULT;
571 break;
572 } else {
573 /*
574 * Note that the size of flock64 is different in
575 * the ILP32 and LP64 models, due to the l_pad
576 * field. We do not want to assume that the
577 * flock64 structure is laid out the same in
578 * ILP32 and LP64 environments, so we will
579 * copy in the ILP32 version of flock64
580 * explicitly and copy it to the native
581 * flock64 structure.
582 */
583 bf.l_type = (short)bf64_32.l_type;
584 bf.l_whence = (short)bf64_32.l_whence;
585 bf.l_start = bf64_32.l_start;
586 bf.l_len = bf64_32.l_len;
587 bf.l_sysid = (int)bf64_32.l_sysid;
588 bf.l_pid = (pid_t)bf64_32.l_pid;
589 }
590 }
591 #endif /* !defined(_LP64) || defined(_SYSCALL32_IMPL) */
592
593 if (cmd == F_ALLOCSP || cmd == F_FREESP)
594 error = flock_check(vp, &bf, offset, maxoffset);
595 else if (cmd == F_ALLOCSP64 || cmd == F_FREESP64)
596 error = flock_check(vp, &bf, offset, MAXOFFSET_T);
597 if (error)
598 break;
599
600 if (vp->v_type == VREG && bf.l_len == 0 &&
601 bf.l_start > OFFSET_MAX(fp)) {
602 error = EFBIG;
603 break;
604 }
605
606 /*
607 * Make sure that there are no conflicting non-blocking
608 * mandatory locks in the region being manipulated. If
609 * there are such locks then return EACCES.
610 */
611 if ((error = flock_get_start(vp, &bf, offset, &start)) != 0)
612 break;
613
614 if (nbl_need_check(vp)) {
615 u_offset_t begin;
616 ssize_t length;
617
618 nbl_start_crit(vp, RW_READER);
619 in_crit = 1;
620 vattr.va_mask = AT_SIZE;
621 if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
622 != 0)
623 break;
624 begin = start > vattr.va_size ? vattr.va_size : start;
625 length = vattr.va_size > start ? vattr.va_size - start :
626 start - vattr.va_size;
627 if (nbl_conflict(vp, NBL_WRITE, begin, length, 0,
628 NULL)) {
629 error = EACCES;
630 break;
631 }
632 }
633
634 if (cmd == F_ALLOCSP64)
635 cmd = F_ALLOCSP;
636 else if (cmd == F_FREESP64)
637 cmd = F_FREESP;
638
639 error = VOP_SPACE(vp, cmd, &bf, flag, offset, fp->f_cred, NULL);
640
641 break;
642
643 #if !defined(_LP64) || defined(_SYSCALL32_IMPL)
644 case F_GETLK64:
645 case F_SETLK64:
646 case F_SETLKW64:
647 case F_SETLK64_NBMAND:
648 case F_OFD_GETLK64:
649 case F_OFD_SETLK64:
650 case F_OFD_SETLKW64:
651 case F_FLOCK64:
652 case F_FLOCKW64:
653 /*
654 * Large Files: Here we set cmd as *LK and send it to
655 * lower layers. *LK64 is only for the user land.
656 * Most of the comments described above for F_SETLK
657 * applies here too.
658 * Large File support is only needed for ILP32 apps!
659 */
660 if (datamodel != DATAMODEL_ILP32) {
661 error = EINVAL;
662 break;
663 }
664
665 if (cmd == F_GETLK64)
666 cmd = F_GETLK;
667 else if (cmd == F_SETLK64)
668 cmd = F_SETLK;
669 else if (cmd == F_SETLKW64)
670 cmd = F_SETLKW;
671 else if (cmd == F_SETLK64_NBMAND)
672 cmd = F_SETLK_NBMAND;
673 else if (cmd == F_OFD_GETLK64)
674 cmd = F_OFD_GETLK;
675 else if (cmd == F_OFD_SETLK64)
676 cmd = F_OFD_SETLK;
677 else if (cmd == F_OFD_SETLKW64)
678 cmd = F_OFD_SETLKW;
679 else if (cmd == F_FLOCK64)
680 cmd = F_FLOCK;
681 else if (cmd == F_FLOCKW64)
682 cmd = F_FLOCKW;
683
684 /*
685 * Note that the size of flock64 is different in the ILP32
686 * and LP64 models, due to the sucking l_pad field.
687 * We do not want to assume that the flock64 structure is
688 * laid out in the same in ILP32 and LP64 environments, so
689 * we will copy in the ILP32 version of flock64 explicitly
690 * and copy it to the native flock64 structure.
691 */
692
693 if (copyin((void *)arg, &bf64_32, sizeof (bf64_32))) {
694 error = EFAULT;
695 break;
696 }
697
698 bf.l_type = (short)bf64_32.l_type;
699 bf.l_whence = (short)bf64_32.l_whence;
700 bf.l_start = bf64_32.l_start;
701 bf.l_len = bf64_32.l_len;
702 bf.l_sysid = (int)bf64_32.l_sysid;
703 bf.l_pid = (pid_t)bf64_32.l_pid;
704
705 if ((error = flock_check(vp, &bf, offset, MAXOFFSET_T)) != 0)
706 break;
707
708 if (cmd == F_FLOCK || cmd == F_FLOCKW) {
709 /* FLOCK* locking is always over the entire file. */
710 if (bf.l_whence != 0 || bf.l_start != 0 ||
711 bf.l_len != 0) {
712 error = EINVAL;
713 break;
714 }
715 if (bf.l_type < F_RDLCK || bf.l_type > F_UNLCK) {
716 error = EINVAL;
717 break;
718 }
719 }
720
721 if (cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
722 /*
723 * TBD OFD-style locking is currently limited to
724 * covering the entire file.
725 */
726 if (bf.l_whence != 0 || bf.l_start != 0 ||
727 bf.l_len != 0) {
728 error = EINVAL;
729 break;
730 }
731 }
732
733 /*
734 * The *_frlock functions in the various file systems basically
735 * do some validation and then funnel everything through the
736 * fs_frlock function. For OFD-style locks fs_frlock will do
737 * nothing so that once control returns here we can call the
738 * ofdlock function with the correct fp. For OFD-style locks
739 * the unsupported remote file systems, such as NFS, detect and
740 * reject the OFD-style cmd argument.
741 */
742 if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset,
743 NULL, fp->f_cred, NULL)) != 0)
744 break;
745
746 if (cmd == F_FLOCK || cmd == F_FLOCKW || cmd == F_OFD_GETLK ||
747 cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW) {
748 /*
749 * This is an OFD-style lock so we need to handle it
750 * here. Because OFD-style locks are associated with
751 * the file_t we didn't have enough info down the
752 * VOP_FRLOCK path immediately above.
753 */
754 if ((error = ofdlock(fp, cmd, &bf, flag, offset)) != 0)
755 break;
756 }
757
758 if ((cmd == F_GETLK || cmd == F_OFD_GETLK) &&
759 bf.l_type == F_UNLCK) {
760 if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
761 sizeof (bf.l_type)))
762 error = EFAULT;
763 break;
764 }
765
766 if (cmd == F_GETLK || cmd == F_OFD_GETLK) {
767 int i;
768
769 /*
770 * We do not want to assume that the flock64 structure
771 * is laid out in the same in ILP32 and LP64
772 * environments, so we will copy out the ILP32 version
773 * of flock64 explicitly after copying the native
774 * flock64 structure to it.
775 */
776 for (i = 0; i < 4; i++)
777 bf64_32.l_pad[i] = 0;
778 bf64_32.l_type = (int16_t)bf.l_type;
779 bf64_32.l_whence = (int16_t)bf.l_whence;
780 bf64_32.l_start = bf.l_start;
781 bf64_32.l_len = bf.l_len;
782 bf64_32.l_sysid = (int32_t)bf.l_sysid;
783 bf64_32.l_pid = (pid32_t)bf.l_pid;
784 if (copyout(&bf64_32, (void *)arg, sizeof (bf64_32)))
785 error = EFAULT;
786 }
787 break;
788 #endif /* !defined(_LP64) || defined(_SYSCALL32_IMPL) */
789
790 case F_SHARE:
791 case F_SHARE_NBMAND:
792 case F_UNSHARE:
793
794 /*
795 * Copy in input fields only.
796 */
797 if (copyin((void *)arg, &fsh, sizeof (fsh))) {
798 error = EFAULT;
799 break;
800 }
801
802 /*
803 * Local share reservations always have this simple form
804 */
805 shr.s_access = fsh.f_access;
806 shr.s_deny = fsh.f_deny;
807 shr.s_sysid = 0;
808 shr.s_pid = ttoproc(curthread)->p_pid;
809 shr_own.sl_pid = shr.s_pid;
810 shr_own.sl_id = fsh.f_id;
811 shr.s_own_len = sizeof (shr_own);
812 shr.s_owner = (caddr_t)&shr_own;
813 error = VOP_SHRLOCK(vp, cmd, &shr, flag, fp->f_cred, NULL);
814 break;
815
816 default:
817 error = EINVAL;
818 break;
819 }
820
821 if (in_crit)
822 nbl_end_crit(vp);
823
824 done:
825 releasef(fdes);
826 out:
827 if (error)
828 return (set_errno(error));
829 return (retval);
830 }
831
832 int
833 flock_check(vnode_t *vp, flock64_t *flp, offset_t offset, offset_t max)
834 {
835 struct vattr vattr;
836 int error;
837 u_offset_t start, end;
838
839 /*
840 * Determine the starting point of the request
841 */
842 switch (flp->l_whence) {
843 case 0: /* SEEK_SET */
844 start = (u_offset_t)flp->l_start;
845 if (start > max)
846 return (EINVAL);
847 break;
848 case 1: /* SEEK_CUR */
849 if (flp->l_start > (max - offset))
850 return (EOVERFLOW);
851 start = (u_offset_t)(flp->l_start + offset);
852 if (start > max)
853 return (EINVAL);
854 break;
855 case 2: /* SEEK_END */
856 vattr.va_mask = AT_SIZE;
857 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
858 return (error);
859 if (flp->l_start > (max - (offset_t)vattr.va_size))
860 return (EOVERFLOW);
861 start = (u_offset_t)(flp->l_start + (offset_t)vattr.va_size);
862 if (start > max)
863 return (EINVAL);
864 break;
865 default:
866 return (EINVAL);
867 }
868
869 /*
870 * Determine the range covered by the request.
871 */
872 if (flp->l_len == 0)
873 end = MAXEND;
874 else if ((offset_t)flp->l_len > 0) {
875 if (flp->l_len > (max - start + 1))
876 return (EOVERFLOW);
877 end = (u_offset_t)(start + (flp->l_len - 1));
878 ASSERT(end <= max);
879 } else {
880 /*
881 * Negative length; why do we even allow this ?
882 * Because this allows easy specification of
883 * the last n bytes of the file.
884 */
885 end = start;
886 start += (u_offset_t)flp->l_len;
887 (start)++;
888 if (start > max)
889 return (EINVAL);
890 ASSERT(end <= max);
891 }
892 ASSERT(start <= max);
893 if (flp->l_type == F_UNLCK && flp->l_len > 0 &&
894 end == (offset_t)max) {
895 flp->l_len = 0;
896 }
897 if (start > end)
898 return (EINVAL);
899 return (0);
900 }
901
902 static int
903 flock_get_start(vnode_t *vp, flock64_t *flp, offset_t offset, u_offset_t *start)
904 {
905 struct vattr vattr;
906 int error;
907
908 /*
909 * Determine the starting point of the request. Assume that it is
910 * a valid starting point.
911 */
912 switch (flp->l_whence) {
913 case 0: /* SEEK_SET */
914 *start = (u_offset_t)flp->l_start;
915 break;
916 case 1: /* SEEK_CUR */
917 *start = (u_offset_t)(flp->l_start + offset);
918 break;
919 case 2: /* SEEK_END */
920 vattr.va_mask = AT_SIZE;
921 if (error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL))
922 return (error);
923 *start = (u_offset_t)(flp->l_start + (offset_t)vattr.va_size);
924 break;
925 default:
926 return (EINVAL);
927 }
928
929 return (0);
930 }
931
932 /*
933 * Take rctl action when the requested file descriptor is too big.
934 */
935 static void
936 fd_too_big(proc_t *p)
937 {
938 mutex_enter(&p->p_lock);
939 (void) rctl_action(rctlproc_legacy[RLIMIT_NOFILE],
940 p->p_rctls, p, RCA_SAFE);
941 mutex_exit(&p->p_lock);
942 }