?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks...
[linux-2.6.git] / fs / compat.c
blobfa9d721ecfee57f8f3b4c2f3b37f2b4e52182ffc
1 /*
2 * linux/fs/compat.c
3 *
4 * Kernel compatibililty routines for e.g. 32 bit syscall support
5 * on 64 bit kernels.
6 *
7 * Copyright (C) 2002 Stephen Rothwell, IBM Corporation
8 * Copyright (C) 1997-2000 Jakub Jelinek (jakub@redhat.com)
9 * Copyright (C) 1998 Eddie C. Dost (ecd@skynet.be)
10 * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs
11 * Copyright (C) 2003 Pavel Machek (pavel@ucw.cz)
12 *
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License version 2 as
15 * published by the Free Software Foundation.
16 */
17
18 #include <linux/stddef.h>
19 #include <linux/kernel.h>
20 #include <linux/linkage.h>
21 #include <linux/compat.h>
22 #include <linux/errno.h>
23 #include <linux/time.h>
24 #include <linux/fs.h>
25 #include <linux/fcntl.h>
26 #include <linux/namei.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/vfs.h>
30 #include <linux/ioctl.h>
31 #include <linux/init.h>
32 #include <linux/ncp_mount.h>
33 #include <linux/nfs4_mount.h>
34 #include <linux/syscalls.h>
35 #include <linux/ctype.h>
36 #include <linux/module.h>
37 #include <linux/dirent.h>
38 #include <linux/fsnotify.h>
39 #include <linux/highuid.h>
40 #include <linux/personality.h>
41 #include <linux/rwsem.h>
42 #include <linux/tsacct_kern.h>
43 #include <linux/security.h>
44 #include <linux/highmem.h>
45 #include <linux/signal.h>
46 #include <linux/poll.h>
47 #include <linux/mm.h>
48 #include <linux/eventpoll.h>
49 #include <linux/fs_struct.h>
50 #include <linux/slab.h>
51 #include <linux/pagemap.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/ioctls.h>
56 #include "internal.h"
57
58 int compat_log = 1;
59
60 int compat_printk(const char *fmt, ...)
61 {
62 va_list ap;
63 int ret;
64 if (!compat_log)
65 return 0;
66 va_start(ap, fmt);
67 ret = vprintk(fmt, ap);
68 va_end(ap);
69 return ret;
70 }
71
72 #include "read_write.h"
73
74 /*
75 * Not all architectures have sys_utime, so implement this in terms
76 * of sys_utimes.
77 */
78 asmlinkage long compat_sys_utime(const char __user *filename,
79 struct compat_utimbuf __user *t)
80 {
81 struct timespec tv[2];
82
83 if (t) {
84 if (get_user(tv[0].tv_sec, &t->actime) ||
85 get_user(tv[1].tv_sec, &t->modtime))
86 return -EFAULT;
87 tv[0].tv_nsec = 0;
88 tv[1].tv_nsec = 0;
89 }
90 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
91 }
92
93 asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
94 {
95 struct timespec tv[2];
96
97 if (t) {
98 if (get_compat_timespec(&tv[0], &t[0]) ||
99 get_compat_timespec(&tv[1], &t[1]))
100 return -EFAULT;
101
102 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
103 return 0;
104 }
105 return do_utimes(dfd, filename, t ? tv : NULL, flags);
106 }
107
108 asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
109 {
110 struct timespec tv[2];
111
112 if (t) {
113 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
114 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
115 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
116 get_user(tv[1].tv_nsec, &t[1].tv_usec))
117 return -EFAULT;
118 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
119 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
120 return -EINVAL;
121 tv[0].tv_nsec *= 1000;
122 tv[1].tv_nsec *= 1000;
123 }
124 return do_utimes(dfd, filename, t ? tv : NULL, 0);
125 }
126
127 asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
128 {
129 return compat_sys_futimesat(AT_FDCWD, filename, t);
130 }
131
132 static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
133 {
134 compat_ino_t ino = stat->ino;
135 typeof(ubuf->st_uid) uid = 0;
136 typeof(ubuf->st_gid) gid = 0;
137 int err;
138
139 SET_UID(uid, stat->uid);
140 SET_GID(gid, stat->gid);
141
142 if ((u64) stat->size > MAX_NON_LFS ||
143 !old_valid_dev(stat->dev) ||
144 !old_valid_dev(stat->rdev))
145 return -EOVERFLOW;
146 if (sizeof(ino) < sizeof(stat->ino) && ino != stat->ino)
147 return -EOVERFLOW;
148
149 if (clear_user(ubuf, sizeof(*ubuf)))
150 return -EFAULT;
151
152 err = __put_user(old_encode_dev(stat->dev), &ubuf->st_dev);
153 err |= __put_user(ino, &ubuf->st_ino);
154 err |= __put_user(stat->mode, &ubuf->st_mode);
155 err |= __put_user(stat->nlink, &ubuf->st_nlink);
156 err |= __put_user(uid, &ubuf->st_uid);
157 err |= __put_user(gid, &ubuf->st_gid);
158 err |= __put_user(old_encode_dev(stat->rdev), &ubuf->st_rdev);
159 err |= __put_user(stat->size, &ubuf->st_size);
160 err |= __put_user(stat->atime.tv_sec, &ubuf->st_atime);
161 err |= __put_user(stat->atime.tv_nsec, &ubuf->st_atime_nsec);
162 err |= __put_user(stat->mtime.tv_sec, &ubuf->st_mtime);
163 err |= __put_user(stat->mtime.tv_nsec, &ubuf->st_mtime_nsec);
164 err |= __put_user(stat->ctime.tv_sec, &ubuf->st_ctime);
165 err |= __put_user(stat->ctime.tv_nsec, &ubuf->st_ctime_nsec);
166 err |= __put_user(stat->blksize, &ubuf->st_blksize);
167 err |= __put_user(stat->blocks, &ubuf->st_blocks);
168 return err;
169 }
170
171 asmlinkage long compat_sys_newstat(const char __user * filename,
172 struct compat_stat __user *statbuf)
173 {
174 struct kstat stat;
175 int error;
176
177 error = vfs_stat(filename, &stat);
178 if (error)
179 return error;
180 return cp_compat_stat(&stat, statbuf);
181 }
182
183 asmlinkage long compat_sys_newlstat(const char __user * filename,
184 struct compat_stat __user *statbuf)
185 {
186 struct kstat stat;
187 int error;
188
189 error = vfs_lstat(filename, &stat);
190 if (error)
191 return error;
192 return cp_compat_stat(&stat, statbuf);
193 }
194
195 #ifndef __ARCH_WANT_STAT64
196 asmlinkage long compat_sys_newfstatat(unsigned int dfd,
197 const char __user *filename,
198 struct compat_stat __user *statbuf, int flag)
199 {
200 struct kstat stat;
201 int error;
202
203 error = vfs_fstatat(dfd, filename, &stat, flag);
204 if (error)
205 return error;
206 return cp_compat_stat(&stat, statbuf);
207 }
208 #endif
209
210 asmlinkage long compat_sys_newfstat(unsigned int fd,
211 struct compat_stat __user * statbuf)
212 {
213 struct kstat stat;
214 int error = vfs_fstat(fd, &stat);
215
216 if (!error)
217 error = cp_compat_stat(&stat, statbuf);
218 return error;
219 }
220
221 static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
222 {
223
224 if (sizeof ubuf->f_blocks == 4) {
225 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
226 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
227 return -EOVERFLOW;
228 /* f_files and f_ffree may be -1; it's okay
229 * to stuff that into 32 bits */
230 if (kbuf->f_files != 0xffffffffffffffffULL
231 && (kbuf->f_files & 0xffffffff00000000ULL))
232 return -EOVERFLOW;
233 if (kbuf->f_ffree != 0xffffffffffffffffULL
234 && (kbuf->f_ffree & 0xffffffff00000000ULL))
235 return -EOVERFLOW;
236 }
237 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
238 __put_user(kbuf->f_type, &ubuf->f_type) ||
239 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
240 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
241 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
242 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
243 __put_user(kbuf->f_files, &ubuf->f_files) ||
244 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
245 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
246 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
247 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
248 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
249 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
250 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
251 return -EFAULT;
252 return 0;
253 }
254
255 /*
256 * The following statfs calls are copies of code from fs/statfs.c and
257 * should be checked against those from time to time
258 */
259 asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
260 {
261 struct kstatfs tmp;
262 int error = user_statfs(pathname, &tmp);
263 if (!error)
264 error = put_compat_statfs(buf, &tmp);
265 return error;
266 }
267
268 asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
269 {
270 struct kstatfs tmp;
271 int error = fd_statfs(fd, &tmp);
272 if (!error)
273 error = put_compat_statfs(buf, &tmp);
274 return error;
275 }
276
277 static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
278 {
279 if (sizeof ubuf->f_blocks == 4) {
280 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
281 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
282 return -EOVERFLOW;
283 /* f_files and f_ffree may be -1; it's okay
284 * to stuff that into 32 bits */
285 if (kbuf->f_files != 0xffffffffffffffffULL
286 && (kbuf->f_files & 0xffffffff00000000ULL))
287 return -EOVERFLOW;
288 if (kbuf->f_ffree != 0xffffffffffffffffULL
289 && (kbuf->f_ffree & 0xffffffff00000000ULL))
290 return -EOVERFLOW;
291 }
292 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
293 __put_user(kbuf->f_type, &ubuf->f_type) ||
294 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
295 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
296 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
297 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
298 __put_user(kbuf->f_files, &ubuf->f_files) ||
299 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
300 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
301 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
302 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
303 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
304 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
305 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
306 return -EFAULT;
307 return 0;
308 }
309
310 asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
311 {
312 struct kstatfs tmp;
313 int error;
314
315 if (sz != sizeof(*buf))
316 return -EINVAL;
317
318 error = user_statfs(pathname, &tmp);
319 if (!error)
320 error = put_compat_statfs64(buf, &tmp);
321 return error;
322 }
323
324 asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
325 {
326 struct kstatfs tmp;
327 int error;
328
329 if (sz != sizeof(*buf))
330 return -EINVAL;
331
332 error = fd_statfs(fd, &tmp);
333 if (!error)
334 error = put_compat_statfs64(buf, &tmp);
335 return error;
336 }
337
338 /*
339 * This is a copy of sys_ustat, just dealing with a structure layout.
340 * Given how simple this syscall is that apporach is more maintainable
341 * than the various conversion hacks.
342 */
343 asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
344 {
345 struct compat_ustat tmp;
346 struct kstatfs sbuf;
347 int err = vfs_ustat(new_decode_dev(dev), &sbuf);
348 if (err)
349 return err;
350
351 memset(&tmp, 0, sizeof(struct compat_ustat));
352 tmp.f_tfree = sbuf.f_bfree;
353 tmp.f_tinode = sbuf.f_ffree;
354 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
355 return -EFAULT;
356 return 0;
357 }
358
359 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
360 {
361 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
362 __get_user(kfl->l_type, &ufl->l_type) ||
363 __get_user(kfl->l_whence, &ufl->l_whence) ||
364 __get_user(kfl->l_start, &ufl->l_start) ||
365 __get_user(kfl->l_len, &ufl->l_len) ||
366 __get_user(kfl->l_pid, &ufl->l_pid))
367 return -EFAULT;
368 return 0;
369 }
370
371 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
372 {
373 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
374 __put_user(kfl->l_type, &ufl->l_type) ||
375 __put_user(kfl->l_whence, &ufl->l_whence) ||
376 __put_user(kfl->l_start, &ufl->l_start) ||
377 __put_user(kfl->l_len, &ufl->l_len) ||
378 __put_user(kfl->l_pid, &ufl->l_pid))
379 return -EFAULT;
380 return 0;
381 }
382
383 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
384 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
385 {
386 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
387 __get_user(kfl->l_type, &ufl->l_type) ||
388 __get_user(kfl->l_whence, &ufl->l_whence) ||
389 __get_user(kfl->l_start, &ufl->l_start) ||
390 __get_user(kfl->l_len, &ufl->l_len) ||
391 __get_user(kfl->l_pid, &ufl->l_pid))
392 return -EFAULT;
393 return 0;
394 }
395 #endif
396
397 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
398 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
399 {
400 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
401 __put_user(kfl->l_type, &ufl->l_type) ||
402 __put_user(kfl->l_whence, &ufl->l_whence) ||
403 __put_user(kfl->l_start, &ufl->l_start) ||
404 __put_user(kfl->l_len, &ufl->l_len) ||
405 __put_user(kfl->l_pid, &ufl->l_pid))
406 return -EFAULT;
407 return 0;
408 }
409 #endif
410
411 asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
412 unsigned long arg)
413 {
414 mm_segment_t old_fs;
415 struct flock f;
416 long ret;
417
418 switch (cmd) {
419 case F_GETLK:
420 case F_SETLK:
421 case F_SETLKW:
422 ret = get_compat_flock(&f, compat_ptr(arg));
423 if (ret != 0)
424 break;
425 old_fs = get_fs();
426 set_fs(KERNEL_DS);
427 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
428 set_fs(old_fs);
429 if (cmd == F_GETLK && ret == 0) {
430 /* GETLK was successful and we need to return the data...
431 * but it needs to fit in the compat structure.
432 * l_start shouldn't be too big, unless the original
433 * start + end is greater than COMPAT_OFF_T_MAX, in which
434 * case the app was asking for trouble, so we return
435 * -EOVERFLOW in that case.
436 * l_len could be too big, in which case we just truncate it,
437 * and only allow the app to see that part of the conflicting
438 * lock that might make sense to it anyway
439 */
440
441 if (f.l_start > COMPAT_OFF_T_MAX)
442 ret = -EOVERFLOW;
443 if (f.l_len > COMPAT_OFF_T_MAX)
444 f.l_len = COMPAT_OFF_T_MAX;
445 if (ret == 0)
446 ret = put_compat_flock(&f, compat_ptr(arg));
447 }
448 break;
449
450 case F_GETLK64:
451 case F_SETLK64:
452 case F_SETLKW64:
453 ret = get_compat_flock64(&f, compat_ptr(arg));
454 if (ret != 0)
455 break;
456 old_fs = get_fs();
457 set_fs(KERNEL_DS);
458 ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
459 ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
460 (unsigned long)&f);
461 set_fs(old_fs);
462 if (cmd == F_GETLK64 && ret == 0) {
463 /* need to return lock information - see above for commentary */
464 if (f.l_start > COMPAT_LOFF_T_MAX)
465 ret = -EOVERFLOW;
466 if (f.l_len > COMPAT_LOFF_T_MAX)
467 f.l_len = COMPAT_LOFF_T_MAX;
468 if (ret == 0)
469 ret = put_compat_flock64(&f, compat_ptr(arg));
470 }
471 break;
472
473 default:
474 ret = sys_fcntl(fd, cmd, arg);
475 break;
476 }
477 return ret;
478 }
479
480 asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
481 unsigned long arg)
482 {
483 if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
484 return -EINVAL;
485 return compat_sys_fcntl64(fd, cmd, arg);
486 }
487
488 asmlinkage long
489 compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
490 {
491 long ret;
492 aio_context_t ctx64;
493
494 mm_segment_t oldfs = get_fs();
495 if (unlikely(get_user(ctx64, ctx32p)))
496 return -EFAULT;
497
498 set_fs(KERNEL_DS);
499 /* The __user pointer cast is valid because of the set_fs() */
500 ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
501 set_fs(oldfs);
502 /* truncating is ok because it's a user address */
503 if (!ret)
504 ret = put_user((u32) ctx64, ctx32p);
505 return ret;
506 }
507
508 asmlinkage long
509 compat_sys_io_getevents(aio_context_t ctx_id,
510 unsigned long min_nr,
511 unsigned long nr,
512 struct io_event __user *events,
513 struct compat_timespec __user *timeout)
514 {
515 long ret;
516 struct timespec t;
517 struct timespec __user *ut = NULL;
518
519 ret = -EFAULT;
520 if (unlikely(!access_ok(VERIFY_WRITE, events,
521 nr * sizeof(struct io_event))))
522 goto out;
523 if (timeout) {
524 if (get_compat_timespec(&t, timeout))
525 goto out;
526
527 ut = compat_alloc_user_space(sizeof(*ut));
528 if (copy_to_user(ut, &t, sizeof(t)) )
529 goto out;
530 }
531 ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
532 out:
533 return ret;
534 }
535
536 /* A write operation does a read from user space and vice versa */
537 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
538
539 ssize_t compat_rw_copy_check_uvector(int type,
540 const struct compat_iovec __user *uvector, unsigned long nr_segs,
541 unsigned long fast_segs, struct iovec *fast_pointer,
542 struct iovec **ret_pointer, int check_access)
543 {
544 compat_ssize_t tot_len;
545 struct iovec *iov = *ret_pointer = fast_pointer;
546 ssize_t ret = 0;
547 int seg;
548
549 /*
550 * SuS says "The readv() function *may* fail if the iovcnt argument
551 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
552 * traditionally returned zero for zero segments, so...
553 */
554 if (nr_segs == 0)
555 goto out;
556
557 ret = -EINVAL;
558 if (nr_segs > UIO_MAXIOV || nr_segs < 0)
559 goto out;
560 if (nr_segs > fast_segs) {
561 ret = -ENOMEM;
562 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
563 if (iov == NULL)
564 goto out;
565 }
566 *ret_pointer = iov;
567
568 /*
569 * Single unix specification:
570 * We should -EINVAL if an element length is not >= 0 and fitting an
571 * ssize_t.
572 *
573 * In Linux, the total length is limited to MAX_RW_COUNT, there is
574 * no overflow possibility.
575 */
576 tot_len = 0;
577 ret = -EINVAL;
578 for (seg = 0; seg < nr_segs; seg++) {
579 compat_uptr_t buf;
580 compat_ssize_t len;
581
582 if (__get_user(len, &uvector->iov_len) ||
583 __get_user(buf, &uvector->iov_base)) {
584 ret = -EFAULT;
585 goto out;
586 }
587 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
588 goto out;
589 if (check_access &&
590 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
591 ret = -EFAULT;
592 goto out;
593 }
594 if (len > MAX_RW_COUNT - tot_len)
595 len = MAX_RW_COUNT - tot_len;
596 tot_len += len;
597 iov->iov_base = compat_ptr(buf);
598 iov->iov_len = (compat_size_t) len;
599 uvector++;
600 iov++;
601 }
602 ret = tot_len;
603
604 out:
605 return ret;
606 }
607
608 static inline long
609 copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
610 {
611 compat_uptr_t uptr;
612 int i;
613
614 for (i = 0; i < nr; ++i) {
615 if (get_user(uptr, ptr32 + i))
616 return -EFAULT;
617 if (put_user(compat_ptr(uptr), ptr64 + i))
618 return -EFAULT;
619 }
620 return 0;
621 }
622
623 #define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *))
624
625 asmlinkage long
626 compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
627 {
628 struct iocb __user * __user *iocb64;
629 long ret;
630
631 if (unlikely(nr < 0))
632 return -EINVAL;
633
634 if (nr > MAX_AIO_SUBMITS)
635 nr = MAX_AIO_SUBMITS;
636
637 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
638 ret = copy_iocb(nr, iocb, iocb64);
639 if (!ret)
640 ret = do_io_submit(ctx_id, nr, iocb64, 1);
641 return ret;
642 }
643
644 struct compat_ncp_mount_data {
645 compat_int_t version;
646 compat_uint_t ncp_fd;
647 __compat_uid_t mounted_uid;
648 compat_pid_t wdog_pid;
649 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
650 compat_uint_t time_out;
651 compat_uint_t retry_count;
652 compat_uint_t flags;
653 __compat_uid_t uid;
654 __compat_gid_t gid;
655 compat_mode_t file_mode;
656 compat_mode_t dir_mode;
657 };
658
659 struct compat_ncp_mount_data_v4 {
660 compat_int_t version;
661 compat_ulong_t flags;
662 compat_ulong_t mounted_uid;
663 compat_long_t wdog_pid;
664 compat_uint_t ncp_fd;
665 compat_uint_t time_out;
666 compat_uint_t retry_count;
667 compat_ulong_t uid;
668 compat_ulong_t gid;
669 compat_ulong_t file_mode;
670 compat_ulong_t dir_mode;
671 };
672
673 static void *do_ncp_super_data_conv(void *raw_data)
674 {
675 int version = *(unsigned int *)raw_data;
676
677 if (version == 3) {
678 struct compat_ncp_mount_data *c_n = raw_data;
679 struct ncp_mount_data *n = raw_data;
680
681 n->dir_mode = c_n->dir_mode;
682 n->file_mode = c_n->file_mode;
683 n->gid = c_n->gid;
684 n->uid = c_n->uid;
685 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
686 n->wdog_pid = c_n->wdog_pid;
687 n->mounted_uid = c_n->mounted_uid;
688 } else if (version == 4) {
689 struct compat_ncp_mount_data_v4 *c_n = raw_data;
690 struct ncp_mount_data_v4 *n = raw_data;
691
692 n->dir_mode = c_n->dir_mode;
693 n->file_mode = c_n->file_mode;
694 n->gid = c_n->gid;
695 n->uid = c_n->uid;
696 n->retry_count = c_n->retry_count;
697 n->time_out = c_n->time_out;
698 n->ncp_fd = c_n->ncp_fd;
699 n->wdog_pid = c_n->wdog_pid;
700 n->mounted_uid = c_n->mounted_uid;
701 n->flags = c_n->flags;
702 } else if (version != 5) {
703 return NULL;
704 }
705
706 return raw_data;
707 }
708
709
710 struct compat_nfs_string {
711 compat_uint_t len;
712 compat_uptr_t data;
713 };
714
715 static inline void compat_nfs_string(struct nfs_string *dst,
716 struct compat_nfs_string *src)
717 {
718 dst->data = compat_ptr(src->data);
719 dst->len = src->len;
720 }
721
722 struct compat_nfs4_mount_data_v1 {
723 compat_int_t version;
724 compat_int_t flags;
725 compat_int_t rsize;
726 compat_int_t wsize;
727 compat_int_t timeo;
728 compat_int_t retrans;
729 compat_int_t acregmin;
730 compat_int_t acregmax;
731 compat_int_t acdirmin;
732 compat_int_t acdirmax;
733 struct compat_nfs_string client_addr;
734 struct compat_nfs_string mnt_path;
735 struct compat_nfs_string hostname;
736 compat_uint_t host_addrlen;
737 compat_uptr_t host_addr;
738 compat_int_t proto;
739 compat_int_t auth_flavourlen;
740 compat_uptr_t auth_flavours;
741 };
742
743 static int do_nfs4_super_data_conv(void *raw_data)
744 {
745 int version = *(compat_uint_t *) raw_data;
746
747 if (version == 1) {
748 struct compat_nfs4_mount_data_v1 *raw = raw_data;
749 struct nfs4_mount_data *real = raw_data;
750
751 /* copy the fields backwards */
752 real->auth_flavours = compat_ptr(raw->auth_flavours);
753 real->auth_flavourlen = raw->auth_flavourlen;
754 real->proto = raw->proto;
755 real->host_addr = compat_ptr(raw->host_addr);
756 real->host_addrlen = raw->host_addrlen;
757 compat_nfs_string(&real->hostname, &raw->hostname);
758 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
759 compat_nfs_string(&real->client_addr, &raw->client_addr);
760 real->acdirmax = raw->acdirmax;
761 real->acdirmin = raw->acdirmin;
762 real->acregmax = raw->acregmax;
763 real->acregmin = raw->acregmin;
764 real->retrans = raw->retrans;
765 real->timeo = raw->timeo;
766 real->wsize = raw->wsize;
767 real->rsize = raw->rsize;
768 real->flags = raw->flags;
769 real->version = raw->version;
770 }
771
772 return 0;
773 }
774
775 #define NCPFS_NAME "ncpfs"
776 #define NFS4_NAME "nfs4"
777
778 asmlinkage long compat_sys_mount(const char __user * dev_name,
779 const char __user * dir_name,
780 const char __user * type, unsigned long flags,
781 const void __user * data)
782 {
783 char *kernel_type;
784 unsigned long data_page;
785 char *kernel_dev;
786 char *dir_page;
787 int retval;
788
789 retval = copy_mount_string(type, &kernel_type);
790 if (retval < 0)
791 goto out;
792
793 dir_page = getname(dir_name);
794 retval = PTR_ERR(dir_page);
795 if (IS_ERR(dir_page))
796 goto out1;
797
798 retval = copy_mount_string(dev_name, &kernel_dev);
799 if (retval < 0)
800 goto out2;
801
802 retval = copy_mount_options(data, &data_page);
803 if (retval < 0)
804 goto out3;
805
806 retval = -EINVAL;
807
808 if (kernel_type && data_page) {
809 if (!strcmp(kernel_type, NCPFS_NAME)) {
810 do_ncp_super_data_conv((void *)data_page);
811 } else if (!strcmp(kernel_type, NFS4_NAME)) {
812 if (do_nfs4_super_data_conv((void *) data_page))
813 goto out4;
814 }
815 }
816
817 retval = do_mount(kernel_dev, dir_page, kernel_type,
818 flags, (void*)data_page);
819
820 out4:
821 free_page(data_page);
822 out3:
823 kfree(kernel_dev);
824 out2:
825 putname(dir_page);
826 out1:
827 kfree(kernel_type);
828 out:
829 return retval;
830 }
831
832 struct compat_old_linux_dirent {
833 compat_ulong_t d_ino;
834 compat_ulong_t d_offset;
835 unsigned short d_namlen;
836 char d_name[1];
837 };
838
839 struct compat_readdir_callback {
840 struct compat_old_linux_dirent __user *dirent;
841 int result;
842 };
843
844 static int compat_fillonedir(void *__buf, const char *name, int namlen,
845 loff_t offset, u64 ino, unsigned int d_type)
846 {
847 struct compat_readdir_callback *buf = __buf;
848 struct compat_old_linux_dirent __user *dirent;
849 compat_ulong_t d_ino;
850
851 if (buf->result)
852 return -EINVAL;
853 d_ino = ino;
854 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
855 buf->result = -EOVERFLOW;
856 return -EOVERFLOW;
857 }
858 buf->result++;
859 dirent = buf->dirent;
860 if (!access_ok(VERIFY_WRITE, dirent,
861 (unsigned long)(dirent->d_name + namlen + 1) -
862 (unsigned long)dirent))
863 goto efault;
864 if ( __put_user(d_ino, &dirent->d_ino) ||
865 __put_user(offset, &dirent->d_offset) ||
866 __put_user(namlen, &dirent->d_namlen) ||
867 __copy_to_user(dirent->d_name, name, namlen) ||
868 __put_user(0, dirent->d_name + namlen))
869 goto efault;
870 return 0;
871 efault:
872 buf->result = -EFAULT;
873 return -EFAULT;
874 }
875
876 asmlinkage long compat_sys_old_readdir(unsigned int fd,
877 struct compat_old_linux_dirent __user *dirent, unsigned int count)
878 {
879 int error;
880 struct file *file;
881 struct compat_readdir_callback buf;
882
883 error = -EBADF;
884 file = fget(fd);
885 if (!file)
886 goto out;
887
888 buf.result = 0;
889 buf.dirent = dirent;
890
891 error = vfs_readdir(file, compat_fillonedir, &buf);
892 if (buf.result)
893 error = buf.result;
894
895 fput(file);
896 out:
897 return error;
898 }
899
900 struct compat_linux_dirent {
901 compat_ulong_t d_ino;
902 compat_ulong_t d_off;
903 unsigned short d_reclen;
904 char d_name[1];
905 };
906
907 struct compat_getdents_callback {
908 struct compat_linux_dirent __user *current_dir;
909 struct compat_linux_dirent __user *previous;
910 int count;
911 int error;
912 };
913
914 static int compat_filldir(void *__buf, const char *name, int namlen,
915 loff_t offset, u64 ino, unsigned int d_type)
916 {
917 struct compat_linux_dirent __user * dirent;
918 struct compat_getdents_callback *buf = __buf;
919 compat_ulong_t d_ino;
920 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
921 namlen + 2, sizeof(compat_long_t));
922
923 buf->error = -EINVAL; /* only used if we fail.. */
924 if (reclen > buf->count)
925 return -EINVAL;
926 d_ino = ino;
927 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
928 buf->error = -EOVERFLOW;
929 return -EOVERFLOW;
930 }
931 dirent = buf->previous;
932 if (dirent) {
933 if (__put_user(offset, &dirent->d_off))
934 goto efault;
935 }
936 dirent = buf->current_dir;
937 if (__put_user(d_ino, &dirent->d_ino))
938 goto efault;
939 if (__put_user(reclen, &dirent->d_reclen))
940 goto efault;
941 if (copy_to_user(dirent->d_name, name, namlen))
942 goto efault;
943 if (__put_user(0, dirent->d_name + namlen))
944 goto efault;
945 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
946 goto efault;
947 buf->previous = dirent;
948 dirent = (void __user *)dirent + reclen;
949 buf->current_dir = dirent;
950 buf->count -= reclen;
951 return 0;
952 efault:
953 buf->error = -EFAULT;
954 return -EFAULT;
955 }
956
957 asmlinkage long compat_sys_getdents(unsigned int fd,
958 struct compat_linux_dirent __user *dirent, unsigned int count)
959 {
960 struct file * file;
961 struct compat_linux_dirent __user * lastdirent;
962 struct compat_getdents_callback buf;
963 int error;
964
965 error = -EFAULT;
966 if (!access_ok(VERIFY_WRITE, dirent, count))
967 goto out;
968
969 error = -EBADF;
970 file = fget(fd);
971 if (!file)
972 goto out;
973
974 buf.current_dir = dirent;
975 buf.previous = NULL;
976 buf.count = count;
977 buf.error = 0;
978
979 error = vfs_readdir(file, compat_filldir, &buf);
980 if (error >= 0)
981 error = buf.error;
982 lastdirent = buf.previous;
983 if (lastdirent) {
984 if (put_user(file->f_pos, &lastdirent->d_off))
985 error = -EFAULT;
986 else
987 error = count - buf.count;
988 }
989 fput(file);
990 out:
991 return error;
992 }
993
994 #ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
995
996 struct compat_getdents_callback64 {
997 struct linux_dirent64 __user *current_dir;
998 struct linux_dirent64 __user *previous;
999 int count;
1000 int error;
1001 };
1002
1003 static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
1004 u64 ino, unsigned int d_type)
1005 {
1006 struct linux_dirent64 __user *dirent;
1007 struct compat_getdents_callback64 *buf = __buf;
1008 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1009 sizeof(u64));
1010 u64 off;
1011
1012 buf->error = -EINVAL; /* only used if we fail.. */
1013 if (reclen > buf->count)
1014 return -EINVAL;
1015 dirent = buf->previous;
1016
1017 if (dirent) {
1018 if (__put_user_unaligned(offset, &dirent->d_off))
1019 goto efault;
1020 }
1021 dirent = buf->current_dir;
1022 if (__put_user_unaligned(ino, &dirent->d_ino))
1023 goto efault;
1024 off = 0;
1025 if (__put_user_unaligned(off, &dirent->d_off))
1026 goto efault;
1027 if (__put_user(reclen, &dirent->d_reclen))
1028 goto efault;
1029 if (__put_user(d_type, &dirent->d_type))
1030 goto efault;
1031 if (copy_to_user(dirent->d_name, name, namlen))
1032 goto efault;
1033 if (__put_user(0, dirent->d_name + namlen))
1034 goto efault;
1035 buf->previous = dirent;
1036 dirent = (void __user *)dirent + reclen;
1037 buf->current_dir = dirent;
1038 buf->count -= reclen;
1039 return 0;
1040 efault:
1041 buf->error = -EFAULT;
1042 return -EFAULT;
1043 }
1044
1045 asmlinkage long compat_sys_getdents64(unsigned int fd,
1046 struct linux_dirent64 __user * dirent, unsigned int count)
1047 {
1048 struct file * file;
1049 struct linux_dirent64 __user * lastdirent;
1050 struct compat_getdents_callback64 buf;
1051 int error;
1052
1053 error = -EFAULT;
1054 if (!access_ok(VERIFY_WRITE, dirent, count))
1055 goto out;
1056
1057 error = -EBADF;
1058 file = fget(fd);
1059 if (!file)
1060 goto out;
1061
1062 buf.current_dir = dirent;
1063 buf.previous = NULL;
1064 buf.count = count;
1065 buf.error = 0;
1066
1067 error = vfs_readdir(file, compat_filldir64, &buf);
1068 if (error >= 0)
1069 error = buf.error;
1070 lastdirent = buf.previous;
1071 if (lastdirent) {
1072 typeof(lastdirent->d_off) d_off = file->f_pos;
1073 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1074 error = -EFAULT;
1075 else
1076 error = count - buf.count;
1077 }
1078 fput(file);
1079 out:
1080 return error;
1081 }
1082 #endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1083
1084 static ssize_t compat_do_readv_writev(int type, struct file *file,
1085 const struct compat_iovec __user *uvector,
1086 unsigned long nr_segs, loff_t *pos)
1087 {
1088 compat_ssize_t tot_len;
1089 struct iovec iovstack[UIO_FASTIOV];
1090 struct iovec *iov = iovstack;
1091 ssize_t ret;
1092 io_fn_t fn;
1093 iov_fn_t fnv;
1094
1095 ret = -EINVAL;
1096 if (!file->f_op)
1097 goto out;
1098
1099 ret = -EFAULT;
1100 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1101 goto out;
1102
1103 tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1104 UIO_FASTIOV, iovstack, &iov, 1);
1105 if (tot_len == 0) {
1106 ret = 0;
1107 goto out;
1108 }
1109
1110 ret = rw_verify_area(type, file, pos, tot_len);
1111 if (ret < 0)
1112 goto out;
1113
1114 fnv = NULL;
1115 if (type == READ) {
1116 fn = file->f_op->read;
1117 fnv = file->f_op->aio_read;
1118 } else {
1119 fn = (io_fn_t)file->f_op->write;
1120 fnv = file->f_op->aio_write;
1121 }
1122
1123 if (fnv)
1124 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1125 pos, fnv);
1126 else
1127 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1128
1129 out:
1130 if (iov != iovstack)
1131 kfree(iov);
1132 if ((ret + (type == READ)) > 0) {
1133 if (type == READ)
1134 fsnotify_access(file);
1135 else
1136 fsnotify_modify(file);
1137 }
1138 return ret;
1139 }
1140
1141 static size_t compat_readv(struct file *file,
1142 const struct compat_iovec __user *vec,
1143 unsigned long vlen, loff_t *pos)
1144 {
1145 ssize_t ret = -EBADF;
1146
1147 if (!(file->f_mode & FMODE_READ))
1148 goto out;
1149
1150 ret = -EINVAL;
1151 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1152 goto out;
1153
1154 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1155
1156 out:
1157 if (ret > 0)
1158 add_rchar(current, ret);
1159 inc_syscr(current);
1160 return ret;
1161 }
1162
1163 asmlinkage ssize_t
1164 compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1165 unsigned long vlen)
1166 {
1167 struct file *file;
1168 int fput_needed;
1169 ssize_t ret;
1170
1171 file = fget_light(fd, &fput_needed);
1172 if (!file)
1173 return -EBADF;
1174 ret = compat_readv(file, vec, vlen, &file->f_pos);
1175 fput_light(file, fput_needed);
1176 return ret;
1177 }
1178
1179 asmlinkage ssize_t
1180 compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1181 unsigned long vlen, u32 pos_low, u32 pos_high)
1182 {
1183 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1184 struct file *file;
1185 int fput_needed;
1186 ssize_t ret;
1187
1188 if (pos < 0)
1189 return -EINVAL;
1190 file = fget_light(fd, &fput_needed);
1191 if (!file)
1192 return -EBADF;
1193 ret = -ESPIPE;
1194 if (file->f_mode & FMODE_PREAD)
1195 ret = compat_readv(file, vec, vlen, &pos);
1196 fput_light(file, fput_needed);
1197 return ret;
1198 }
1199
1200 static size_t compat_writev(struct file *file,
1201 const struct compat_iovec __user *vec,
1202 unsigned long vlen, loff_t *pos)
1203 {
1204 ssize_t ret = -EBADF;
1205
1206 if (!(file->f_mode & FMODE_WRITE))
1207 goto out;
1208
1209 ret = -EINVAL;
1210 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1211 goto out;
1212
1213 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1214
1215 out:
1216 if (ret > 0)
1217 add_wchar(current, ret);
1218 inc_syscw(current);
1219 return ret;
1220 }
1221
1222 asmlinkage ssize_t
1223 compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1224 unsigned long vlen)
1225 {
1226 struct file *file;
1227 int fput_needed;
1228 ssize_t ret;
1229
1230 file = fget_light(fd, &fput_needed);
1231 if (!file)
1232 return -EBADF;
1233 ret = compat_writev(file, vec, vlen, &file->f_pos);
1234 fput_light(file, fput_needed);
1235 return ret;
1236 }
1237
1238 asmlinkage ssize_t
1239 compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1240 unsigned long vlen, u32 pos_low, u32 pos_high)
1241 {
1242 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1243 struct file *file;
1244 int fput_needed;
1245 ssize_t ret;
1246
1247 if (pos < 0)
1248 return -EINVAL;
1249 file = fget_light(fd, &fput_needed);
1250 if (!file)
1251 return -EBADF;
1252 ret = -ESPIPE;
1253 if (file->f_mode & FMODE_PWRITE)
1254 ret = compat_writev(file, vec, vlen, &pos);
1255 fput_light(file, fput_needed);
1256 return ret;
1257 }
1258
1259 asmlinkage long
1260 compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1261 unsigned int nr_segs, unsigned int flags)
1262 {
1263 unsigned i;
1264 struct iovec __user *iov;
1265 if (nr_segs > UIO_MAXIOV)
1266 return -EINVAL;
1267 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1268 for (i = 0; i < nr_segs; i++) {
1269 struct compat_iovec v;
1270 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1271 get_user(v.iov_len, &iov32[i].iov_len) ||
1272 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1273 put_user(v.iov_len, &iov[i].iov_len))
1274 return -EFAULT;
1275 }
1276 return sys_vmsplice(fd, iov, nr_segs, flags);
1277 }
1278
1279 /*
1280 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1281 * O_LARGEFILE flag.
1282 */
1283 asmlinkage long
1284 compat_sys_open(const char __user *filename, int flags, umode_t mode)
1285 {
1286 return do_sys_open(AT_FDCWD, filename, flags, mode);
1287 }
1288
1289 /*
1290 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1291 * O_LARGEFILE flag.
1292 */
1293 asmlinkage long
1294 compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, umode_t mode)
1295 {
1296 return do_sys_open(dfd, filename, flags, mode);
1297 }
1298
1299 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1300
1301 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1302 int timeval, int ret)
1303 {
1304 struct timespec ts;
1305
1306 if (!p)
1307 return ret;
1308
1309 if (current->personality & STICKY_TIMEOUTS)
1310 goto sticky;
1311
1312 /* No update for zero timeout */
1313 if (!end_time->tv_sec && !end_time->tv_nsec)
1314 return ret;
1315
1316 ktime_get_ts(&ts);
1317 ts = timespec_sub(*end_time, ts);
1318 if (ts.tv_sec < 0)
1319 ts.tv_sec = ts.tv_nsec = 0;
1320
1321 if (timeval) {
1322 struct compat_timeval rtv;
1323
1324 rtv.tv_sec = ts.tv_sec;
1325 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1326
1327 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1328 return ret;
1329 } else {
1330 struct compat_timespec rts;
1331
1332 rts.tv_sec = ts.tv_sec;
1333 rts.tv_nsec = ts.tv_nsec;
1334
1335 if (!copy_to_user(p, &rts, sizeof(rts)))
1336 return ret;
1337 }
1338 /*
1339 * If an application puts its timeval in read-only memory, we
1340 * don't want the Linux-specific update to the timeval to
1341 * cause a fault after the select has completed
1342 * successfully. However, because we're not updating the
1343 * timeval, we can't restart the system call.
1344 */
1345
1346 sticky:
1347 if (ret == -ERESTARTNOHAND)
1348 ret = -EINTR;
1349 return ret;
1350 }
1351
1352 /*
1353 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1354 * 64-bit unsigned longs.
1355 */
1356 static
1357 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1358 unsigned long *fdset)
1359 {
1360 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1361 if (ufdset) {
1362 unsigned long odd;
1363
1364 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1365 return -EFAULT;
1366
1367 odd = nr & 1UL;
1368 nr &= ~1UL;
1369 while (nr) {
1370 unsigned long h, l;
1371 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1372 return -EFAULT;
1373 ufdset += 2;
1374 *fdset++ = h << 32 | l;
1375 nr -= 2;
1376 }
1377 if (odd && __get_user(*fdset, ufdset))
1378 return -EFAULT;
1379 } else {
1380 /* Tricky, must clear full unsigned long in the
1381 * kernel fdset at the end, this makes sure that
1382 * actually happens.
1383 */
1384 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1385 }
1386 return 0;
1387 }
1388
1389 static
1390 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1391 unsigned long *fdset)
1392 {
1393 unsigned long odd;
1394 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1395
1396 if (!ufdset)
1397 return 0;
1398
1399 odd = nr & 1UL;
1400 nr &= ~1UL;
1401 while (nr) {
1402 unsigned long h, l;
1403 l = *fdset++;
1404 h = l >> 32;
1405 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1406 return -EFAULT;
1407 ufdset += 2;
1408 nr -= 2;
1409 }
1410 if (odd && __put_user(*fdset, ufdset))
1411 return -EFAULT;
1412 return 0;
1413 }
1414
1415
1416 /*
1417 * This is a virtual copy of sys_select from fs/select.c and probably
1418 * should be compared to it from time to time
1419 */
1420
1421 /*
1422 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1423 * like to be certain this leads to no problems. So I return
1424 * EINTR just for safety.
1425 *
1426 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1427 * I'm trying ERESTARTNOHAND which restart only when you want to.
1428 */
1429 int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1430 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1431 struct timespec *end_time)
1432 {
1433 fd_set_bits fds;
1434 void *bits;
1435 int size, max_fds, ret = -EINVAL;
1436 struct fdtable *fdt;
1437 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1438
1439 if (n < 0)
1440 goto out_nofds;
1441
1442 /* max_fds can increase, so grab it once to avoid race */
1443 rcu_read_lock();
1444 fdt = files_fdtable(current->files);
1445 max_fds = fdt->max_fds;
1446 rcu_read_unlock();
1447 if (n > max_fds)
1448 n = max_fds;
1449
1450 /*
1451 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1452 * since we used fdset we need to allocate memory in units of
1453 * long-words.
1454 */
1455 size = FDS_BYTES(n);
1456 bits = stack_fds;
1457 if (size > sizeof(stack_fds) / 6) {
1458 bits = kmalloc(6 * size, GFP_KERNEL);
1459 ret = -ENOMEM;
1460 if (!bits)
1461 goto out_nofds;
1462 }
1463 fds.in = (unsigned long *) bits;
1464 fds.out = (unsigned long *) (bits + size);
1465 fds.ex = (unsigned long *) (bits + 2*size);
1466 fds.res_in = (unsigned long *) (bits + 3*size);
1467 fds.res_out = (unsigned long *) (bits + 4*size);
1468 fds.res_ex = (unsigned long *) (bits + 5*size);
1469
1470 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1471 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1472 (ret = compat_get_fd_set(n, exp, fds.ex)))
1473 goto out;
1474 zero_fd_set(n, fds.res_in);
1475 zero_fd_set(n, fds.res_out);
1476 zero_fd_set(n, fds.res_ex);
1477
1478 ret = do_select(n, &fds, end_time);
1479
1480 if (ret < 0)
1481 goto out;
1482 if (!ret) {
1483 ret = -ERESTARTNOHAND;
1484 if (signal_pending(current))
1485 goto out;
1486 ret = 0;
1487 }
1488
1489 if (compat_set_fd_set(n, inp, fds.res_in) ||
1490 compat_set_fd_set(n, outp, fds.res_out) ||
1491 compat_set_fd_set(n, exp, fds.res_ex))
1492 ret = -EFAULT;
1493 out:
1494 if (bits != stack_fds)
1495 kfree(bits);
1496 out_nofds:
1497 return ret;
1498 }
1499
1500 asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1501 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1502 struct compat_timeval __user *tvp)
1503 {
1504 struct timespec end_time, *to = NULL;
1505 struct compat_timeval tv;
1506 int ret;
1507
1508 if (tvp) {
1509 if (copy_from_user(&tv, tvp, sizeof(tv)))
1510 return -EFAULT;
1511
1512 to = &end_time;
1513 if (poll_select_set_timeout(to,
1514 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1515 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1516 return -EINVAL;
1517 }
1518
1519 ret = compat_core_sys_select(n, inp, outp, exp, to);
1520 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1521
1522 return ret;
1523 }
1524
1525 struct compat_sel_arg_struct {
1526 compat_ulong_t n;
1527 compat_uptr_t inp;
1528 compat_uptr_t outp;
1529 compat_uptr_t exp;
1530 compat_uptr_t tvp;
1531 };
1532
1533 asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1534 {
1535 struct compat_sel_arg_struct a;
1536
1537 if (copy_from_user(&a, arg, sizeof(a)))
1538 return -EFAULT;
1539 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1540 compat_ptr(a.exp), compat_ptr(a.tvp));
1541 }
1542
1543 #ifdef HAVE_SET_RESTORE_SIGMASK
1544 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1545 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1546 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1547 compat_size_t sigsetsize)
1548 {
1549 compat_sigset_t ss32;
1550 sigset_t ksigmask, sigsaved;
1551 struct compat_timespec ts;
1552 struct timespec end_time, *to = NULL;
1553 int ret;
1554
1555 if (tsp) {
1556 if (copy_from_user(&ts, tsp, sizeof(ts)))
1557 return -EFAULT;
1558
1559 to = &end_time;
1560 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1561 return -EINVAL;
1562 }
1563
1564 if (sigmask) {
1565 if (sigsetsize != sizeof(compat_sigset_t))
1566 return -EINVAL;
1567 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1568 return -EFAULT;
1569 sigset_from_compat(&ksigmask, &ss32);
1570
1571 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1572 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1573 }
1574
1575 ret = compat_core_sys_select(n, inp, outp, exp, to);
1576 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1577
1578 if (ret == -ERESTARTNOHAND) {
1579 /*
1580 * Don't restore the signal mask yet. Let do_signal() deliver
1581 * the signal on the way back to userspace, before the signal
1582 * mask is restored.
1583 */
1584 if (sigmask) {
1585 memcpy(&current->saved_sigmask, &sigsaved,
1586 sizeof(sigsaved));
1587 set_restore_sigmask();
1588 }
1589 } else if (sigmask)
1590 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1591
1592 return ret;
1593 }
1594
1595 asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1596 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1597 struct compat_timespec __user *tsp, void __user *sig)
1598 {
1599 compat_size_t sigsetsize = 0;
1600 compat_uptr_t up = 0;
1601
1602 if (sig) {
1603 if (!access_ok(VERIFY_READ, sig,
1604 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1605 __get_user(up, (compat_uptr_t __user *)sig) ||
1606 __get_user(sigsetsize,
1607 (compat_size_t __user *)(sig+sizeof(up))))
1608 return -EFAULT;
1609 }
1610 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1611 sigsetsize);
1612 }
1613
1614 asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1615 unsigned int nfds, struct compat_timespec __user *tsp,
1616 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1617 {
1618 compat_sigset_t ss32;
1619 sigset_t ksigmask, sigsaved;
1620 struct compat_timespec ts;
1621 struct timespec end_time, *to = NULL;
1622 int ret;
1623
1624 if (tsp) {
1625 if (copy_from_user(&ts, tsp, sizeof(ts)))
1626 return -EFAULT;
1627
1628 to = &end_time;
1629 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1630 return -EINVAL;
1631 }
1632
1633 if (sigmask) {
1634 if (sigsetsize != sizeof(compat_sigset_t))
1635 return -EINVAL;
1636 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1637 return -EFAULT;
1638 sigset_from_compat(&ksigmask, &ss32);
1639
1640 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1641 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1642 }
1643
1644 ret = do_sys_poll(ufds, nfds, to);
1645
1646 /* We can restart this syscall, usually */
1647 if (ret == -EINTR) {
1648 /*
1649 * Don't restore the signal mask yet. Let do_signal() deliver
1650 * the signal on the way back to userspace, before the signal
1651 * mask is restored.
1652 */
1653 if (sigmask) {
1654 memcpy(&current->saved_sigmask, &sigsaved,
1655 sizeof(sigsaved));
1656 set_restore_sigmask();
1657 }
1658 ret = -ERESTARTNOHAND;
1659 } else if (sigmask)
1660 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1661
1662 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1663
1664 return ret;
1665 }
1666 #endif /* HAVE_SET_RESTORE_SIGMASK */
1667
1668 #ifdef CONFIG_EPOLL
1669
1670 #ifdef HAVE_SET_RESTORE_SIGMASK
1671 asmlinkage long compat_sys_epoll_pwait(int epfd,
1672 struct compat_epoll_event __user *events,
1673 int maxevents, int timeout,
1674 const compat_sigset_t __user *sigmask,
1675 compat_size_t sigsetsize)
1676 {
1677 long err;
1678 compat_sigset_t csigmask;
1679 sigset_t ksigmask, sigsaved;
1680
1681 /*
1682 * If the caller wants a certain signal mask to be set during the wait,
1683 * we apply it here.
1684 */
1685 if (sigmask) {
1686 if (sigsetsize != sizeof(compat_sigset_t))
1687 return -EINVAL;
1688 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1689 return -EFAULT;
1690 sigset_from_compat(&ksigmask, &csigmask);
1691 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1692 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1693 }
1694
1695 err = sys_epoll_wait(epfd, events, maxevents, timeout);
1696
1697 /*
1698 * If we changed the signal mask, we need to restore the original one.
1699 * In case we've got a signal while waiting, we do not restore the
1700 * signal mask yet, and we allow do_signal() to deliver the signal on
1701 * the way back to userspace, before the signal mask is restored.
1702 */
1703 if (sigmask) {
1704 if (err == -EINTR) {
1705 memcpy(&current->saved_sigmask, &sigsaved,
1706 sizeof(sigsaved));
1707 set_restore_sigmask();
1708 } else
1709 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1710 }
1711
1712 return err;
1713 }
1714 #endif /* HAVE_SET_RESTORE_SIGMASK */
1715
1716 #endif /* CONFIG_EPOLL */
1717
1718 #ifdef CONFIG_SIGNALFD
1719
1720 asmlinkage long compat_sys_signalfd4(int ufd,
1721 const compat_sigset_t __user *sigmask,
1722 compat_size_t sigsetsize, int flags)
1723 {
1724 compat_sigset_t ss32;
1725 sigset_t tmp;
1726 sigset_t __user *ksigmask;
1727
1728 if (sigsetsize != sizeof(compat_sigset_t))
1729 return -EINVAL;
1730 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1731 return -EFAULT;
1732 sigset_from_compat(&tmp, &ss32);
1733 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1734 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1735 return -EFAULT;
1736
1737 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1738 }
1739
1740 asmlinkage long compat_sys_signalfd(int ufd,
1741 const compat_sigset_t __user *sigmask,
1742 compat_size_t sigsetsize)
1743 {
1744 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1745 }
1746 #endif /* CONFIG_SIGNALFD */
1747
1748 #ifdef CONFIG_TIMERFD
1749
1750 asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1751 const struct compat_itimerspec __user *utmr,
1752 struct compat_itimerspec __user *otmr)
1753 {
1754 int error;
1755 struct itimerspec t;
1756 struct itimerspec __user *ut;
1757
1758 if (get_compat_itimerspec(&t, utmr))
1759 return -EFAULT;
1760 ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1761 if (copy_to_user(&ut[0], &t, sizeof(t)))
1762 return -EFAULT;
1763 error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1764 if (!error && otmr)
1765 error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1766 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1767
1768 return error;
1769 }
1770
1771 asmlinkage long compat_sys_timerfd_gettime(int ufd,
1772 struct compat_itimerspec __user *otmr)
1773 {
1774 int error;
1775 struct itimerspec t;
1776 struct itimerspec __user *ut;
1777
1778 ut = compat_alloc_user_space(sizeof(struct itimerspec));
1779 error = sys_timerfd_gettime(ufd, ut);
1780 if (!error)
1781 error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1782 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1783
1784 return error;
1785 }
1786
1787 #endif /* CONFIG_TIMERFD */
1788
1789 #ifdef CONFIG_FHANDLE
1790 /*
1791 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1792 * doesn't set the O_LARGEFILE flag.
1793 */
1794 asmlinkage long
1795 compat_sys_open_by_handle_at(int mountdirfd,
1796 struct file_handle __user *handle, int flags)
1797 {
1798 return do_handle_open(mountdirfd, handle, flags);
1799 }
1800 #endif
Linus' kernel tree