?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add linux-next specific files for 20110831
[linux-2.6/next.git] / fs / compat.c
blobc76adb46b0ce2e5584c8f598f3e1b2a30039f15c
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/nfsd/syscall.h>
41 #include <linux/personality.h>
42 #include <linux/rwsem.h>
43 #include <linux/tsacct_kern.h>
44 #include <linux/security.h>
45 #include <linux/highmem.h>
46 #include <linux/signal.h>
47 #include <linux/poll.h>
48 #include <linux/mm.h>
49 #include <linux/eventpoll.h>
50 #include <linux/fs_struct.h>
51 #include <linux/slab.h>
52 #include <linux/pagemap.h>
53
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
56 #include <asm/ioctls.h>
57 #include "internal.h"
58
59 int compat_log = 1;
60
61 int compat_printk(const char *fmt, ...)
62 {
63 va_list ap;
64 int ret;
65 if (!compat_log)
66 return 0;
67 va_start(ap, fmt);
68 ret = vprintk(fmt, ap);
69 va_end(ap);
70 return ret;
71 }
72
73 #include "read_write.h"
74
75 /*
76 * Not all architectures have sys_utime, so implement this in terms
77 * of sys_utimes.
78 */
79 asmlinkage long compat_sys_utime(const char __user *filename,
80 struct compat_utimbuf __user *t)
81 {
82 struct timespec tv[2];
83
84 if (t) {
85 if (get_user(tv[0].tv_sec, &t->actime) ||
86 get_user(tv[1].tv_sec, &t->modtime))
87 return -EFAULT;
88 tv[0].tv_nsec = 0;
89 tv[1].tv_nsec = 0;
90 }
91 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
92 }
93
94 asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
95 {
96 struct timespec tv[2];
97
98 if (t) {
99 if (get_compat_timespec(&tv[0], &t[0]) ||
100 get_compat_timespec(&tv[1], &t[1]))
101 return -EFAULT;
102
103 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
104 return 0;
105 }
106 return do_utimes(dfd, filename, t ? tv : NULL, flags);
107 }
108
109 asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
110 {
111 struct timespec tv[2];
112
113 if (t) {
114 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
115 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
116 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
117 get_user(tv[1].tv_nsec, &t[1].tv_usec))
118 return -EFAULT;
119 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
120 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
121 return -EINVAL;
122 tv[0].tv_nsec *= 1000;
123 tv[1].tv_nsec *= 1000;
124 }
125 return do_utimes(dfd, filename, t ? tv : NULL, 0);
126 }
127
128 asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
129 {
130 return compat_sys_futimesat(AT_FDCWD, filename, t);
131 }
132
133 static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
134 {
135 compat_ino_t ino = stat->ino;
136 typeof(ubuf->st_uid) uid = 0;
137 typeof(ubuf->st_gid) gid = 0;
138 int err;
139
140 SET_UID(uid, stat->uid);
141 SET_GID(gid, stat->gid);
142
143 if ((u64) stat->size > MAX_NON_LFS ||
144 !old_valid_dev(stat->dev) ||
145 !old_valid_dev(stat->rdev))
146 return -EOVERFLOW;
147 if (sizeof(ino) < sizeof(stat->ino) && ino != stat->ino)
148 return -EOVERFLOW;
149
150 if (clear_user(ubuf, sizeof(*ubuf)))
151 return -EFAULT;
152
153 err = __put_user(old_encode_dev(stat->dev), &ubuf->st_dev);
154 err |= __put_user(ino, &ubuf->st_ino);
155 err |= __put_user(stat->mode, &ubuf->st_mode);
156 err |= __put_user(stat->nlink, &ubuf->st_nlink);
157 err |= __put_user(uid, &ubuf->st_uid);
158 err |= __put_user(gid, &ubuf->st_gid);
159 err |= __put_user(old_encode_dev(stat->rdev), &ubuf->st_rdev);
160 err |= __put_user(stat->size, &ubuf->st_size);
161 err |= __put_user(stat->atime.tv_sec, &ubuf->st_atime);
162 err |= __put_user(stat->atime.tv_nsec, &ubuf->st_atime_nsec);
163 err |= __put_user(stat->mtime.tv_sec, &ubuf->st_mtime);
164 err |= __put_user(stat->mtime.tv_nsec, &ubuf->st_mtime_nsec);
165 err |= __put_user(stat->ctime.tv_sec, &ubuf->st_ctime);
166 err |= __put_user(stat->ctime.tv_nsec, &ubuf->st_ctime_nsec);
167 err |= __put_user(stat->blksize, &ubuf->st_blksize);
168 err |= __put_user(stat->blocks, &ubuf->st_blocks);
169 return err;
170 }
171
172 asmlinkage long compat_sys_newstat(const char __user * filename,
173 struct compat_stat __user *statbuf)
174 {
175 struct kstat stat;
176 int error;
177
178 error = vfs_stat(filename, &stat);
179 if (error)
180 return error;
181 return cp_compat_stat(&stat, statbuf);
182 }
183
184 asmlinkage long compat_sys_newlstat(const char __user * filename,
185 struct compat_stat __user *statbuf)
186 {
187 struct kstat stat;
188 int error;
189
190 error = vfs_lstat(filename, &stat);
191 if (error)
192 return error;
193 return cp_compat_stat(&stat, statbuf);
194 }
195
196 #ifndef __ARCH_WANT_STAT64
197 asmlinkage long compat_sys_newfstatat(unsigned int dfd,
198 const char __user *filename,
199 struct compat_stat __user *statbuf, int flag)
200 {
201 struct kstat stat;
202 int error;
203
204 error = vfs_fstatat(dfd, filename, &stat, flag);
205 if (error)
206 return error;
207 return cp_compat_stat(&stat, statbuf);
208 }
209 #endif
210
211 asmlinkage long compat_sys_newfstat(unsigned int fd,
212 struct compat_stat __user * statbuf)
213 {
214 struct kstat stat;
215 int error = vfs_fstat(fd, &stat);
216
217 if (!error)
218 error = cp_compat_stat(&stat, statbuf);
219 return error;
220 }
221
222 static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
223 {
224
225 if (sizeof ubuf->f_blocks == 4) {
226 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
227 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
228 return -EOVERFLOW;
229 /* f_files and f_ffree may be -1; it's okay
230 * to stuff that into 32 bits */
231 if (kbuf->f_files != 0xffffffffffffffffULL
232 && (kbuf->f_files & 0xffffffff00000000ULL))
233 return -EOVERFLOW;
234 if (kbuf->f_ffree != 0xffffffffffffffffULL
235 && (kbuf->f_ffree & 0xffffffff00000000ULL))
236 return -EOVERFLOW;
237 }
238 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
239 __put_user(kbuf->f_type, &ubuf->f_type) ||
240 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
241 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
242 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
243 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
244 __put_user(kbuf->f_files, &ubuf->f_files) ||
245 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
246 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
247 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
248 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
249 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
250 __put_user(0, &ubuf->f_spare[0]) ||
251 __put_user(0, &ubuf->f_spare[1]) ||
252 __put_user(0, &ubuf->f_spare[2]) ||
253 __put_user(0, &ubuf->f_spare[3]) ||
254 __put_user(0, &ubuf->f_spare[4]))
255 return -EFAULT;
256 return 0;
257 }
258
259 /*
260 * The following statfs calls are copies of code from fs/statfs.c and
261 * should be checked against those from time to time
262 */
263 asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
264 {
265 struct kstatfs tmp;
266 int error = user_statfs(pathname, &tmp);
267 if (!error)
268 error = put_compat_statfs(buf, &tmp);
269 return error;
270 }
271
272 asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
273 {
274 struct kstatfs tmp;
275 int error = fd_statfs(fd, &tmp);
276 if (!error)
277 error = put_compat_statfs(buf, &tmp);
278 return error;
279 }
280
281 static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
282 {
283 if (sizeof ubuf->f_blocks == 4) {
284 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
285 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
286 return -EOVERFLOW;
287 /* f_files and f_ffree may be -1; it's okay
288 * to stuff that into 32 bits */
289 if (kbuf->f_files != 0xffffffffffffffffULL
290 && (kbuf->f_files & 0xffffffff00000000ULL))
291 return -EOVERFLOW;
292 if (kbuf->f_ffree != 0xffffffffffffffffULL
293 && (kbuf->f_ffree & 0xffffffff00000000ULL))
294 return -EOVERFLOW;
295 }
296 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
297 __put_user(kbuf->f_type, &ubuf->f_type) ||
298 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
299 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
300 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
301 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
302 __put_user(kbuf->f_files, &ubuf->f_files) ||
303 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
304 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
305 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
306 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
307 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
308 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
309 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
310 return -EFAULT;
311 return 0;
312 }
313
314 asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
315 {
316 struct kstatfs tmp;
317 int error;
318
319 if (sz != sizeof(*buf))
320 return -EINVAL;
321
322 error = user_statfs(pathname, &tmp);
323 if (!error)
324 error = put_compat_statfs64(buf, &tmp);
325 return error;
326 }
327
328 asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
329 {
330 struct kstatfs tmp;
331 int error;
332
333 if (sz != sizeof(*buf))
334 return -EINVAL;
335
336 error = fd_statfs(fd, &tmp);
337 if (!error)
338 error = put_compat_statfs64(buf, &tmp);
339 return error;
340 }
341
342 /*
343 * This is a copy of sys_ustat, just dealing with a structure layout.
344 * Given how simple this syscall is that apporach is more maintainable
345 * than the various conversion hacks.
346 */
347 asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
348 {
349 struct super_block *sb;
350 struct compat_ustat tmp;
351 struct kstatfs sbuf;
352 int err;
353
354 sb = user_get_super(new_decode_dev(dev));
355 if (!sb)
356 return -EINVAL;
357 err = statfs_by_dentry(sb->s_root, &sbuf);
358 drop_super(sb);
359 if (err)
360 return err;
361
362 memset(&tmp, 0, sizeof(struct compat_ustat));
363 tmp.f_tfree = sbuf.f_bfree;
364 tmp.f_tinode = sbuf.f_ffree;
365 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
366 return -EFAULT;
367 return 0;
368 }
369
370 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
371 {
372 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
373 __get_user(kfl->l_type, &ufl->l_type) ||
374 __get_user(kfl->l_whence, &ufl->l_whence) ||
375 __get_user(kfl->l_start, &ufl->l_start) ||
376 __get_user(kfl->l_len, &ufl->l_len) ||
377 __get_user(kfl->l_pid, &ufl->l_pid))
378 return -EFAULT;
379 return 0;
380 }
381
382 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
383 {
384 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
385 __put_user(kfl->l_type, &ufl->l_type) ||
386 __put_user(kfl->l_whence, &ufl->l_whence) ||
387 __put_user(kfl->l_start, &ufl->l_start) ||
388 __put_user(kfl->l_len, &ufl->l_len) ||
389 __put_user(kfl->l_pid, &ufl->l_pid))
390 return -EFAULT;
391 return 0;
392 }
393
394 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
395 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
396 {
397 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
398 __get_user(kfl->l_type, &ufl->l_type) ||
399 __get_user(kfl->l_whence, &ufl->l_whence) ||
400 __get_user(kfl->l_start, &ufl->l_start) ||
401 __get_user(kfl->l_len, &ufl->l_len) ||
402 __get_user(kfl->l_pid, &ufl->l_pid))
403 return -EFAULT;
404 return 0;
405 }
406 #endif
407
408 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
409 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
410 {
411 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
412 __put_user(kfl->l_type, &ufl->l_type) ||
413 __put_user(kfl->l_whence, &ufl->l_whence) ||
414 __put_user(kfl->l_start, &ufl->l_start) ||
415 __put_user(kfl->l_len, &ufl->l_len) ||
416 __put_user(kfl->l_pid, &ufl->l_pid))
417 return -EFAULT;
418 return 0;
419 }
420 #endif
421
422 asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
423 unsigned long arg)
424 {
425 mm_segment_t old_fs;
426 struct flock f;
427 long ret;
428
429 switch (cmd) {
430 case F_GETLK:
431 case F_SETLK:
432 case F_SETLKW:
433 ret = get_compat_flock(&f, compat_ptr(arg));
434 if (ret != 0)
435 break;
436 old_fs = get_fs();
437 set_fs(KERNEL_DS);
438 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
439 set_fs(old_fs);
440 if (cmd == F_GETLK && ret == 0) {
441 /* GETLK was successful and we need to return the data...
442 * but it needs to fit in the compat structure.
443 * l_start shouldn't be too big, unless the original
444 * start + end is greater than COMPAT_OFF_T_MAX, in which
445 * case the app was asking for trouble, so we return
446 * -EOVERFLOW in that case.
447 * l_len could be too big, in which case we just truncate it,
448 * and only allow the app to see that part of the conflicting
449 * lock that might make sense to it anyway
450 */
451
452 if (f.l_start > COMPAT_OFF_T_MAX)
453 ret = -EOVERFLOW;
454 if (f.l_len > COMPAT_OFF_T_MAX)
455 f.l_len = COMPAT_OFF_T_MAX;
456 if (ret == 0)
457 ret = put_compat_flock(&f, compat_ptr(arg));
458 }
459 break;
460
461 case F_GETLK64:
462 case F_SETLK64:
463 case F_SETLKW64:
464 ret = get_compat_flock64(&f, compat_ptr(arg));
465 if (ret != 0)
466 break;
467 old_fs = get_fs();
468 set_fs(KERNEL_DS);
469 ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
470 ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
471 (unsigned long)&f);
472 set_fs(old_fs);
473 if (cmd == F_GETLK64 && ret == 0) {
474 /* need to return lock information - see above for commentary */
475 if (f.l_start > COMPAT_LOFF_T_MAX)
476 ret = -EOVERFLOW;
477 if (f.l_len > COMPAT_LOFF_T_MAX)
478 f.l_len = COMPAT_LOFF_T_MAX;
479 if (ret == 0)
480 ret = put_compat_flock64(&f, compat_ptr(arg));
481 }
482 break;
483
484 default:
485 ret = sys_fcntl(fd, cmd, arg);
486 break;
487 }
488 return ret;
489 }
490
491 asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
492 unsigned long arg)
493 {
494 if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
495 return -EINVAL;
496 return compat_sys_fcntl64(fd, cmd, arg);
497 }
498
499 asmlinkage long
500 compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
501 {
502 long ret;
503 aio_context_t ctx64;
504
505 mm_segment_t oldfs = get_fs();
506 if (unlikely(get_user(ctx64, ctx32p)))
507 return -EFAULT;
508
509 set_fs(KERNEL_DS);
510 /* The __user pointer cast is valid because of the set_fs() */
511 ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
512 set_fs(oldfs);
513 /* truncating is ok because it's a user address */
514 if (!ret)
515 ret = put_user((u32) ctx64, ctx32p);
516 return ret;
517 }
518
519 asmlinkage long
520 compat_sys_io_getevents(aio_context_t ctx_id,
521 unsigned long min_nr,
522 unsigned long nr,
523 struct io_event __user *events,
524 struct compat_timespec __user *timeout)
525 {
526 long ret;
527 struct timespec t;
528 struct timespec __user *ut = NULL;
529
530 ret = -EFAULT;
531 if (unlikely(!access_ok(VERIFY_WRITE, events,
532 nr * sizeof(struct io_event))))
533 goto out;
534 if (timeout) {
535 if (get_compat_timespec(&t, timeout))
536 goto out;
537
538 ut = compat_alloc_user_space(sizeof(*ut));
539 if (copy_to_user(ut, &t, sizeof(t)) )
540 goto out;
541 }
542 ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
543 out:
544 return ret;
545 }
546
547 /* A write operation does a read from user space and vice versa */
548 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
549
550 ssize_t compat_rw_copy_check_uvector(int type,
551 const struct compat_iovec __user *uvector, unsigned long nr_segs,
552 unsigned long fast_segs, struct iovec *fast_pointer,
553 struct iovec **ret_pointer, int check_access)
554 {
555 compat_ssize_t tot_len;
556 struct iovec *iov = *ret_pointer = fast_pointer;
557 ssize_t ret = 0;
558 int seg;
559
560 /*
561 * SuS says "The readv() function *may* fail if the iovcnt argument
562 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
563 * traditionally returned zero for zero segments, so...
564 */
565 if (nr_segs == 0)
566 goto out;
567
568 ret = -EINVAL;
569 if (nr_segs > UIO_MAXIOV || nr_segs < 0)
570 goto out;
571 if (nr_segs > fast_segs) {
572 ret = -ENOMEM;
573 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
574 if (iov == NULL)
575 goto out;
576 }
577 *ret_pointer = iov;
578
579 /*
580 * Single unix specification:
581 * We should -EINVAL if an element length is not >= 0 and fitting an
582 * ssize_t.
583 *
584 * In Linux, the total length is limited to MAX_RW_COUNT, there is
585 * no overflow possibility.
586 */
587 tot_len = 0;
588 ret = -EINVAL;
589 for (seg = 0; seg < nr_segs; seg++) {
590 compat_uptr_t buf;
591 compat_ssize_t len;
592
593 if (__get_user(len, &uvector->iov_len) ||
594 __get_user(buf, &uvector->iov_base)) {
595 ret = -EFAULT;
596 goto out;
597 }
598 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
599 goto out;
600 if (check_access &&
601 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
602 ret = -EFAULT;
603 goto out;
604 }
605 if (len > MAX_RW_COUNT - tot_len)
606 len = MAX_RW_COUNT - tot_len;
607 tot_len += len;
608 iov->iov_base = compat_ptr(buf);
609 iov->iov_len = (compat_size_t) len;
610 uvector++;
611 iov++;
612 }
613 ret = tot_len;
614
615 out:
616 return ret;
617 }
618
619 static inline long
620 copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
621 {
622 compat_uptr_t uptr;
623 int i;
624
625 for (i = 0; i < nr; ++i) {
626 if (get_user(uptr, ptr32 + i))
627 return -EFAULT;
628 if (put_user(compat_ptr(uptr), ptr64 + i))
629 return -EFAULT;
630 }
631 return 0;
632 }
633
634 #define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *))
635
636 asmlinkage long
637 compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
638 {
639 struct iocb __user * __user *iocb64;
640 long ret;
641
642 if (unlikely(nr < 0))
643 return -EINVAL;
644
645 if (nr > MAX_AIO_SUBMITS)
646 nr = MAX_AIO_SUBMITS;
647
648 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
649 ret = copy_iocb(nr, iocb, iocb64);
650 if (!ret)
651 ret = do_io_submit(ctx_id, nr, iocb64, 1);
652 return ret;
653 }
654
655 struct compat_ncp_mount_data {
656 compat_int_t version;
657 compat_uint_t ncp_fd;
658 __compat_uid_t mounted_uid;
659 compat_pid_t wdog_pid;
660 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
661 compat_uint_t time_out;
662 compat_uint_t retry_count;
663 compat_uint_t flags;
664 __compat_uid_t uid;
665 __compat_gid_t gid;
666 compat_mode_t file_mode;
667 compat_mode_t dir_mode;
668 };
669
670 struct compat_ncp_mount_data_v4 {
671 compat_int_t version;
672 compat_ulong_t flags;
673 compat_ulong_t mounted_uid;
674 compat_long_t wdog_pid;
675 compat_uint_t ncp_fd;
676 compat_uint_t time_out;
677 compat_uint_t retry_count;
678 compat_ulong_t uid;
679 compat_ulong_t gid;
680 compat_ulong_t file_mode;
681 compat_ulong_t dir_mode;
682 };
683
684 static void *do_ncp_super_data_conv(void *raw_data)
685 {
686 int version = *(unsigned int *)raw_data;
687
688 if (version == 3) {
689 struct compat_ncp_mount_data *c_n = raw_data;
690 struct ncp_mount_data *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 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
697 n->wdog_pid = c_n->wdog_pid;
698 n->mounted_uid = c_n->mounted_uid;
699 } else if (version == 4) {
700 struct compat_ncp_mount_data_v4 *c_n = raw_data;
701 struct ncp_mount_data_v4 *n = raw_data;
702
703 n->dir_mode = c_n->dir_mode;
704 n->file_mode = c_n->file_mode;
705 n->gid = c_n->gid;
706 n->uid = c_n->uid;
707 n->retry_count = c_n->retry_count;
708 n->time_out = c_n->time_out;
709 n->ncp_fd = c_n->ncp_fd;
710 n->wdog_pid = c_n->wdog_pid;
711 n->mounted_uid = c_n->mounted_uid;
712 n->flags = c_n->flags;
713 } else if (version != 5) {
714 return NULL;
715 }
716
717 return raw_data;
718 }
719
720
721 struct compat_nfs_string {
722 compat_uint_t len;
723 compat_uptr_t data;
724 };
725
726 static inline void compat_nfs_string(struct nfs_string *dst,
727 struct compat_nfs_string *src)
728 {
729 dst->data = compat_ptr(src->data);
730 dst->len = src->len;
731 }
732
733 struct compat_nfs4_mount_data_v1 {
734 compat_int_t version;
735 compat_int_t flags;
736 compat_int_t rsize;
737 compat_int_t wsize;
738 compat_int_t timeo;
739 compat_int_t retrans;
740 compat_int_t acregmin;
741 compat_int_t acregmax;
742 compat_int_t acdirmin;
743 compat_int_t acdirmax;
744 struct compat_nfs_string client_addr;
745 struct compat_nfs_string mnt_path;
746 struct compat_nfs_string hostname;
747 compat_uint_t host_addrlen;
748 compat_uptr_t host_addr;
749 compat_int_t proto;
750 compat_int_t auth_flavourlen;
751 compat_uptr_t auth_flavours;
752 };
753
754 static int do_nfs4_super_data_conv(void *raw_data)
755 {
756 int version = *(compat_uint_t *) raw_data;
757
758 if (version == 1) {
759 struct compat_nfs4_mount_data_v1 *raw = raw_data;
760 struct nfs4_mount_data *real = raw_data;
761
762 /* copy the fields backwards */
763 real->auth_flavours = compat_ptr(raw->auth_flavours);
764 real->auth_flavourlen = raw->auth_flavourlen;
765 real->proto = raw->proto;
766 real->host_addr = compat_ptr(raw->host_addr);
767 real->host_addrlen = raw->host_addrlen;
768 compat_nfs_string(&real->hostname, &raw->hostname);
769 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
770 compat_nfs_string(&real->client_addr, &raw->client_addr);
771 real->acdirmax = raw->acdirmax;
772 real->acdirmin = raw->acdirmin;
773 real->acregmax = raw->acregmax;
774 real->acregmin = raw->acregmin;
775 real->retrans = raw->retrans;
776 real->timeo = raw->timeo;
777 real->wsize = raw->wsize;
778 real->rsize = raw->rsize;
779 real->flags = raw->flags;
780 real->version = raw->version;
781 }
782
783 return 0;
784 }
785
786 #define NCPFS_NAME "ncpfs"
787 #define NFS4_NAME "nfs4"
788
789 asmlinkage long compat_sys_mount(const char __user * dev_name,
790 const char __user * dir_name,
791 const char __user * type, unsigned long flags,
792 const void __user * data)
793 {
794 char *kernel_type;
795 unsigned long data_page;
796 char *kernel_dev;
797 char *dir_page;
798 int retval;
799
800 retval = copy_mount_string(type, &kernel_type);
801 if (retval < 0)
802 goto out;
803
804 dir_page = getname(dir_name);
805 retval = PTR_ERR(dir_page);
806 if (IS_ERR(dir_page))
807 goto out1;
808
809 retval = copy_mount_string(dev_name, &kernel_dev);
810 if (retval < 0)
811 goto out2;
812
813 retval = copy_mount_options(data, &data_page);
814 if (retval < 0)
815 goto out3;
816
817 retval = -EINVAL;
818
819 if (kernel_type && data_page) {
820 if (!strcmp(kernel_type, NCPFS_NAME)) {
821 do_ncp_super_data_conv((void *)data_page);
822 } else if (!strcmp(kernel_type, NFS4_NAME)) {
823 if (do_nfs4_super_data_conv((void *) data_page))
824 goto out4;
825 }
826 }
827
828 retval = do_mount(kernel_dev, dir_page, kernel_type,
829 flags, (void*)data_page);
830
831 out4:
832 free_page(data_page);
833 out3:
834 kfree(kernel_dev);
835 out2:
836 putname(dir_page);
837 out1:
838 kfree(kernel_type);
839 out:
840 return retval;
841 }
842
843 struct compat_old_linux_dirent {
844 compat_ulong_t d_ino;
845 compat_ulong_t d_offset;
846 unsigned short d_namlen;
847 char d_name[1];
848 };
849
850 struct compat_readdir_callback {
851 struct compat_old_linux_dirent __user *dirent;
852 int result;
853 };
854
855 static int compat_fillonedir(void *__buf, const char *name, int namlen,
856 loff_t offset, u64 ino, unsigned int d_type)
857 {
858 struct compat_readdir_callback *buf = __buf;
859 struct compat_old_linux_dirent __user *dirent;
860 compat_ulong_t d_ino;
861
862 if (buf->result)
863 return -EINVAL;
864 d_ino = ino;
865 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
866 buf->result = -EOVERFLOW;
867 return -EOVERFLOW;
868 }
869 buf->result++;
870 dirent = buf->dirent;
871 if (!access_ok(VERIFY_WRITE, dirent,
872 (unsigned long)(dirent->d_name + namlen + 1) -
873 (unsigned long)dirent))
874 goto efault;
875 if ( __put_user(d_ino, &dirent->d_ino) ||
876 __put_user(offset, &dirent->d_offset) ||
877 __put_user(namlen, &dirent->d_namlen) ||
878 __copy_to_user(dirent->d_name, name, namlen) ||
879 __put_user(0, dirent->d_name + namlen))
880 goto efault;
881 return 0;
882 efault:
883 buf->result = -EFAULT;
884 return -EFAULT;
885 }
886
887 asmlinkage long compat_sys_old_readdir(unsigned int fd,
888 struct compat_old_linux_dirent __user *dirent, unsigned int count)
889 {
890 int error;
891 struct file *file;
892 struct compat_readdir_callback buf;
893
894 error = -EBADF;
895 file = fget(fd);
896 if (!file)
897 goto out;
898
899 buf.result = 0;
900 buf.dirent = dirent;
901
902 error = vfs_readdir(file, compat_fillonedir, &buf);
903 if (buf.result)
904 error = buf.result;
905
906 fput(file);
907 out:
908 return error;
909 }
910
911 struct compat_linux_dirent {
912 compat_ulong_t d_ino;
913 compat_ulong_t d_off;
914 unsigned short d_reclen;
915 char d_name[1];
916 };
917
918 struct compat_getdents_callback {
919 struct compat_linux_dirent __user *current_dir;
920 struct compat_linux_dirent __user *previous;
921 int count;
922 int error;
923 };
924
925 static int compat_filldir(void *__buf, const char *name, int namlen,
926 loff_t offset, u64 ino, unsigned int d_type)
927 {
928 struct compat_linux_dirent __user * dirent;
929 struct compat_getdents_callback *buf = __buf;
930 compat_ulong_t d_ino;
931 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
932 namlen + 2, sizeof(compat_long_t));
933
934 buf->error = -EINVAL; /* only used if we fail.. */
935 if (reclen > buf->count)
936 return -EINVAL;
937 d_ino = ino;
938 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
939 buf->error = -EOVERFLOW;
940 return -EOVERFLOW;
941 }
942 dirent = buf->previous;
943 if (dirent) {
944 if (__put_user(offset, &dirent->d_off))
945 goto efault;
946 }
947 dirent = buf->current_dir;
948 if (__put_user(d_ino, &dirent->d_ino))
949 goto efault;
950 if (__put_user(reclen, &dirent->d_reclen))
951 goto efault;
952 if (copy_to_user(dirent->d_name, name, namlen))
953 goto efault;
954 if (__put_user(0, dirent->d_name + namlen))
955 goto efault;
956 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
957 goto efault;
958 buf->previous = dirent;
959 dirent = (void __user *)dirent + reclen;
960 buf->current_dir = dirent;
961 buf->count -= reclen;
962 return 0;
963 efault:
964 buf->error = -EFAULT;
965 return -EFAULT;
966 }
967
968 asmlinkage long compat_sys_getdents(unsigned int fd,
969 struct compat_linux_dirent __user *dirent, unsigned int count)
970 {
971 struct file * file;
972 struct compat_linux_dirent __user * lastdirent;
973 struct compat_getdents_callback buf;
974 int error;
975
976 error = -EFAULT;
977 if (!access_ok(VERIFY_WRITE, dirent, count))
978 goto out;
979
980 error = -EBADF;
981 file = fget(fd);
982 if (!file)
983 goto out;
984
985 buf.current_dir = dirent;
986 buf.previous = NULL;
987 buf.count = count;
988 buf.error = 0;
989
990 error = vfs_readdir(file, compat_filldir, &buf);
991 if (error >= 0)
992 error = buf.error;
993 lastdirent = buf.previous;
994 if (lastdirent) {
995 if (put_user(file->f_pos, &lastdirent->d_off))
996 error = -EFAULT;
997 else
998 error = count - buf.count;
999 }
1000 fput(file);
1001 out:
1002 return error;
1003 }
1004
1005 #ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
1006
1007 struct compat_getdents_callback64 {
1008 struct linux_dirent64 __user *current_dir;
1009 struct linux_dirent64 __user *previous;
1010 int count;
1011 int error;
1012 };
1013
1014 static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
1015 u64 ino, unsigned int d_type)
1016 {
1017 struct linux_dirent64 __user *dirent;
1018 struct compat_getdents_callback64 *buf = __buf;
1019 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1020 sizeof(u64));
1021 u64 off;
1022
1023 buf->error = -EINVAL; /* only used if we fail.. */
1024 if (reclen > buf->count)
1025 return -EINVAL;
1026 dirent = buf->previous;
1027
1028 if (dirent) {
1029 if (__put_user_unaligned(offset, &dirent->d_off))
1030 goto efault;
1031 }
1032 dirent = buf->current_dir;
1033 if (__put_user_unaligned(ino, &dirent->d_ino))
1034 goto efault;
1035 off = 0;
1036 if (__put_user_unaligned(off, &dirent->d_off))
1037 goto efault;
1038 if (__put_user(reclen, &dirent->d_reclen))
1039 goto efault;
1040 if (__put_user(d_type, &dirent->d_type))
1041 goto efault;
1042 if (copy_to_user(dirent->d_name, name, namlen))
1043 goto efault;
1044 if (__put_user(0, dirent->d_name + namlen))
1045 goto efault;
1046 buf->previous = dirent;
1047 dirent = (void __user *)dirent + reclen;
1048 buf->current_dir = dirent;
1049 buf->count -= reclen;
1050 return 0;
1051 efault:
1052 buf->error = -EFAULT;
1053 return -EFAULT;
1054 }
1055
1056 asmlinkage long compat_sys_getdents64(unsigned int fd,
1057 struct linux_dirent64 __user * dirent, unsigned int count)
1058 {
1059 struct file * file;
1060 struct linux_dirent64 __user * lastdirent;
1061 struct compat_getdents_callback64 buf;
1062 int error;
1063
1064 error = -EFAULT;
1065 if (!access_ok(VERIFY_WRITE, dirent, count))
1066 goto out;
1067
1068 error = -EBADF;
1069 file = fget(fd);
1070 if (!file)
1071 goto out;
1072
1073 buf.current_dir = dirent;
1074 buf.previous = NULL;
1075 buf.count = count;
1076 buf.error = 0;
1077
1078 error = vfs_readdir(file, compat_filldir64, &buf);
1079 if (error >= 0)
1080 error = buf.error;
1081 lastdirent = buf.previous;
1082 if (lastdirent) {
1083 typeof(lastdirent->d_off) d_off = file->f_pos;
1084 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1085 error = -EFAULT;
1086 else
1087 error = count - buf.count;
1088 }
1089 fput(file);
1090 out:
1091 return error;
1092 }
1093 #endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1094
1095 static ssize_t compat_do_readv_writev(int type, struct file *file,
1096 const struct compat_iovec __user *uvector,
1097 unsigned long nr_segs, loff_t *pos)
1098 {
1099 compat_ssize_t tot_len;
1100 struct iovec iovstack[UIO_FASTIOV];
1101 struct iovec *iov = iovstack;
1102 ssize_t ret;
1103 io_fn_t fn;
1104 iov_fn_t fnv;
1105
1106 ret = -EINVAL;
1107 if (!file->f_op)
1108 goto out;
1109
1110 ret = -EFAULT;
1111 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1112 goto out;
1113
1114 tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1115 UIO_FASTIOV, iovstack, &iov, 1);
1116 if (tot_len == 0) {
1117 ret = 0;
1118 goto out;
1119 }
1120
1121 ret = rw_verify_area(type, file, pos, tot_len);
1122 if (ret < 0)
1123 goto out;
1124
1125 fnv = NULL;
1126 if (type == READ) {
1127 fn = file->f_op->read;
1128 fnv = file->f_op->aio_read;
1129 } else {
1130 fn = (io_fn_t)file->f_op->write;
1131 fnv = file->f_op->aio_write;
1132 }
1133
1134 if (fnv)
1135 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1136 pos, fnv);
1137 else
1138 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1139
1140 out:
1141 if (iov != iovstack)
1142 kfree(iov);
1143 if ((ret + (type == READ)) > 0) {
1144 if (type == READ)
1145 fsnotify_access(file);
1146 else
1147 fsnotify_modify(file);
1148 }
1149 return ret;
1150 }
1151
1152 static size_t compat_readv(struct file *file,
1153 const struct compat_iovec __user *vec,
1154 unsigned long vlen, loff_t *pos)
1155 {
1156 ssize_t ret = -EBADF;
1157
1158 if (!(file->f_mode & FMODE_READ))
1159 goto out;
1160
1161 ret = -EINVAL;
1162 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1163 goto out;
1164
1165 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1166
1167 out:
1168 if (ret > 0)
1169 add_rchar(current, ret);
1170 inc_syscr(current);
1171 return ret;
1172 }
1173
1174 asmlinkage ssize_t
1175 compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1176 unsigned long vlen)
1177 {
1178 struct file *file;
1179 int fput_needed;
1180 ssize_t ret;
1181
1182 file = fget_light(fd, &fput_needed);
1183 if (!file)
1184 return -EBADF;
1185 ret = compat_readv(file, vec, vlen, &file->f_pos);
1186 fput_light(file, fput_needed);
1187 return ret;
1188 }
1189
1190 asmlinkage ssize_t
1191 compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1192 unsigned long vlen, u32 pos_low, u32 pos_high)
1193 {
1194 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1195 struct file *file;
1196 int fput_needed;
1197 ssize_t ret;
1198
1199 if (pos < 0)
1200 return -EINVAL;
1201 file = fget_light(fd, &fput_needed);
1202 if (!file)
1203 return -EBADF;
1204 ret = -ESPIPE;
1205 if (file->f_mode & FMODE_PREAD)
1206 ret = compat_readv(file, vec, vlen, &pos);
1207 fput_light(file, fput_needed);
1208 return ret;
1209 }
1210
1211 static size_t compat_writev(struct file *file,
1212 const struct compat_iovec __user *vec,
1213 unsigned long vlen, loff_t *pos)
1214 {
1215 ssize_t ret = -EBADF;
1216
1217 if (!(file->f_mode & FMODE_WRITE))
1218 goto out;
1219
1220 ret = -EINVAL;
1221 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1222 goto out;
1223
1224 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1225
1226 out:
1227 if (ret > 0)
1228 add_wchar(current, ret);
1229 inc_syscw(current);
1230 return ret;
1231 }
1232
1233 asmlinkage ssize_t
1234 compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1235 unsigned long vlen)
1236 {
1237 struct file *file;
1238 int fput_needed;
1239 ssize_t ret;
1240
1241 file = fget_light(fd, &fput_needed);
1242 if (!file)
1243 return -EBADF;
1244 ret = compat_writev(file, vec, vlen, &file->f_pos);
1245 fput_light(file, fput_needed);
1246 return ret;
1247 }
1248
1249 asmlinkage ssize_t
1250 compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1251 unsigned long vlen, u32 pos_low, u32 pos_high)
1252 {
1253 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1254 struct file *file;
1255 int fput_needed;
1256 ssize_t ret;
1257
1258 if (pos < 0)
1259 return -EINVAL;
1260 file = fget_light(fd, &fput_needed);
1261 if (!file)
1262 return -EBADF;
1263 ret = -ESPIPE;
1264 if (file->f_mode & FMODE_PWRITE)
1265 ret = compat_writev(file, vec, vlen, &pos);
1266 fput_light(file, fput_needed);
1267 return ret;
1268 }
1269
1270 asmlinkage long
1271 compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1272 unsigned int nr_segs, unsigned int flags)
1273 {
1274 unsigned i;
1275 struct iovec __user *iov;
1276 if (nr_segs > UIO_MAXIOV)
1277 return -EINVAL;
1278 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1279 for (i = 0; i < nr_segs; i++) {
1280 struct compat_iovec v;
1281 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1282 get_user(v.iov_len, &iov32[i].iov_len) ||
1283 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1284 put_user(v.iov_len, &iov[i].iov_len))
1285 return -EFAULT;
1286 }
1287 return sys_vmsplice(fd, iov, nr_segs, flags);
1288 }
1289
1290 /*
1291 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1292 * O_LARGEFILE flag.
1293 */
1294 asmlinkage long
1295 compat_sys_open(const char __user *filename, int flags, int mode)
1296 {
1297 return do_sys_open(AT_FDCWD, filename, flags, mode);
1298 }
1299
1300 /*
1301 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1302 * O_LARGEFILE flag.
1303 */
1304 asmlinkage long
1305 compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, int mode)
1306 {
1307 return do_sys_open(dfd, filename, flags, mode);
1308 }
1309
1310 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1311
1312 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1313 int timeval, int ret)
1314 {
1315 struct timespec ts;
1316
1317 if (!p)
1318 return ret;
1319
1320 if (current->personality & STICKY_TIMEOUTS)
1321 goto sticky;
1322
1323 /* No update for zero timeout */
1324 if (!end_time->tv_sec && !end_time->tv_nsec)
1325 return ret;
1326
1327 ktime_get_ts(&ts);
1328 ts = timespec_sub(*end_time, ts);
1329 if (ts.tv_sec < 0)
1330 ts.tv_sec = ts.tv_nsec = 0;
1331
1332 if (timeval) {
1333 struct compat_timeval rtv;
1334
1335 rtv.tv_sec = ts.tv_sec;
1336 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1337
1338 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1339 return ret;
1340 } else {
1341 struct compat_timespec rts;
1342
1343 rts.tv_sec = ts.tv_sec;
1344 rts.tv_nsec = ts.tv_nsec;
1345
1346 if (!copy_to_user(p, &rts, sizeof(rts)))
1347 return ret;
1348 }
1349 /*
1350 * If an application puts its timeval in read-only memory, we
1351 * don't want the Linux-specific update to the timeval to
1352 * cause a fault after the select has completed
1353 * successfully. However, because we're not updating the
1354 * timeval, we can't restart the system call.
1355 */
1356
1357 sticky:
1358 if (ret == -ERESTARTNOHAND)
1359 ret = -EINTR;
1360 return ret;
1361 }
1362
1363 /*
1364 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1365 * 64-bit unsigned longs.
1366 */
1367 static
1368 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1369 unsigned long *fdset)
1370 {
1371 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1372 if (ufdset) {
1373 unsigned long odd;
1374
1375 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1376 return -EFAULT;
1377
1378 odd = nr & 1UL;
1379 nr &= ~1UL;
1380 while (nr) {
1381 unsigned long h, l;
1382 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1383 return -EFAULT;
1384 ufdset += 2;
1385 *fdset++ = h << 32 | l;
1386 nr -= 2;
1387 }
1388 if (odd && __get_user(*fdset, ufdset))
1389 return -EFAULT;
1390 } else {
1391 /* Tricky, must clear full unsigned long in the
1392 * kernel fdset at the end, this makes sure that
1393 * actually happens.
1394 */
1395 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1396 }
1397 return 0;
1398 }
1399
1400 static
1401 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1402 unsigned long *fdset)
1403 {
1404 unsigned long odd;
1405 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1406
1407 if (!ufdset)
1408 return 0;
1409
1410 odd = nr & 1UL;
1411 nr &= ~1UL;
1412 while (nr) {
1413 unsigned long h, l;
1414 l = *fdset++;
1415 h = l >> 32;
1416 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1417 return -EFAULT;
1418 ufdset += 2;
1419 nr -= 2;
1420 }
1421 if (odd && __put_user(*fdset, ufdset))
1422 return -EFAULT;
1423 return 0;
1424 }
1425
1426
1427 /*
1428 * This is a virtual copy of sys_select from fs/select.c and probably
1429 * should be compared to it from time to time
1430 */
1431
1432 /*
1433 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1434 * like to be certain this leads to no problems. So I return
1435 * EINTR just for safety.
1436 *
1437 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1438 * I'm trying ERESTARTNOHAND which restart only when you want to.
1439 */
1440 int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1441 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1442 struct timespec *end_time)
1443 {
1444 fd_set_bits fds;
1445 void *bits;
1446 int size, max_fds, ret = -EINVAL;
1447 struct fdtable *fdt;
1448 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1449
1450 if (n < 0)
1451 goto out_nofds;
1452
1453 /* max_fds can increase, so grab it once to avoid race */
1454 rcu_read_lock();
1455 fdt = files_fdtable(current->files);
1456 max_fds = fdt->max_fds;
1457 rcu_read_unlock();
1458 if (n > max_fds)
1459 n = max_fds;
1460
1461 /*
1462 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1463 * since we used fdset we need to allocate memory in units of
1464 * long-words.
1465 */
1466 size = FDS_BYTES(n);
1467 bits = stack_fds;
1468 if (size > sizeof(stack_fds) / 6) {
1469 bits = kmalloc(6 * size, GFP_KERNEL);
1470 ret = -ENOMEM;
1471 if (!bits)
1472 goto out_nofds;
1473 }
1474 fds.in = (unsigned long *) bits;
1475 fds.out = (unsigned long *) (bits + size);
1476 fds.ex = (unsigned long *) (bits + 2*size);
1477 fds.res_in = (unsigned long *) (bits + 3*size);
1478 fds.res_out = (unsigned long *) (bits + 4*size);
1479 fds.res_ex = (unsigned long *) (bits + 5*size);
1480
1481 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1482 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1483 (ret = compat_get_fd_set(n, exp, fds.ex)))
1484 goto out;
1485 zero_fd_set(n, fds.res_in);
1486 zero_fd_set(n, fds.res_out);
1487 zero_fd_set(n, fds.res_ex);
1488
1489 ret = do_select(n, &fds, end_time);
1490
1491 if (ret < 0)
1492 goto out;
1493 if (!ret) {
1494 ret = -ERESTARTNOHAND;
1495 if (signal_pending(current))
1496 goto out;
1497 ret = 0;
1498 }
1499
1500 if (compat_set_fd_set(n, inp, fds.res_in) ||
1501 compat_set_fd_set(n, outp, fds.res_out) ||
1502 compat_set_fd_set(n, exp, fds.res_ex))
1503 ret = -EFAULT;
1504 out:
1505 if (bits != stack_fds)
1506 kfree(bits);
1507 out_nofds:
1508 return ret;
1509 }
1510
1511 asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1512 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1513 struct compat_timeval __user *tvp)
1514 {
1515 struct timespec end_time, *to = NULL;
1516 struct compat_timeval tv;
1517 int ret;
1518
1519 if (tvp) {
1520 if (copy_from_user(&tv, tvp, sizeof(tv)))
1521 return -EFAULT;
1522
1523 to = &end_time;
1524 if (poll_select_set_timeout(to,
1525 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1526 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1527 return -EINVAL;
1528 }
1529
1530 ret = compat_core_sys_select(n, inp, outp, exp, to);
1531 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1532
1533 return ret;
1534 }
1535
1536 struct compat_sel_arg_struct {
1537 compat_ulong_t n;
1538 compat_uptr_t inp;
1539 compat_uptr_t outp;
1540 compat_uptr_t exp;
1541 compat_uptr_t tvp;
1542 };
1543
1544 asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1545 {
1546 struct compat_sel_arg_struct a;
1547
1548 if (copy_from_user(&a, arg, sizeof(a)))
1549 return -EFAULT;
1550 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1551 compat_ptr(a.exp), compat_ptr(a.tvp));
1552 }
1553
1554 #ifdef HAVE_SET_RESTORE_SIGMASK
1555 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1556 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1557 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1558 compat_size_t sigsetsize)
1559 {
1560 compat_sigset_t ss32;
1561 sigset_t ksigmask, sigsaved;
1562 struct compat_timespec ts;
1563 struct timespec end_time, *to = NULL;
1564 int ret;
1565
1566 if (tsp) {
1567 if (copy_from_user(&ts, tsp, sizeof(ts)))
1568 return -EFAULT;
1569
1570 to = &end_time;
1571 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1572 return -EINVAL;
1573 }
1574
1575 if (sigmask) {
1576 if (sigsetsize != sizeof(compat_sigset_t))
1577 return -EINVAL;
1578 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1579 return -EFAULT;
1580 sigset_from_compat(&ksigmask, &ss32);
1581
1582 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1583 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1584 }
1585
1586 ret = compat_core_sys_select(n, inp, outp, exp, to);
1587 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1588
1589 if (ret == -ERESTARTNOHAND) {
1590 /*
1591 * Don't restore the signal mask yet. Let do_signal() deliver
1592 * the signal on the way back to userspace, before the signal
1593 * mask is restored.
1594 */
1595 if (sigmask) {
1596 memcpy(&current->saved_sigmask, &sigsaved,
1597 sizeof(sigsaved));
1598 set_restore_sigmask();
1599 }
1600 } else if (sigmask)
1601 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1602
1603 return ret;
1604 }
1605
1606 asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1607 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1608 struct compat_timespec __user *tsp, void __user *sig)
1609 {
1610 compat_size_t sigsetsize = 0;
1611 compat_uptr_t up = 0;
1612
1613 if (sig) {
1614 if (!access_ok(VERIFY_READ, sig,
1615 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1616 __get_user(up, (compat_uptr_t __user *)sig) ||
1617 __get_user(sigsetsize,
1618 (compat_size_t __user *)(sig+sizeof(up))))
1619 return -EFAULT;
1620 }
1621 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1622 sigsetsize);
1623 }
1624
1625 asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1626 unsigned int nfds, struct compat_timespec __user *tsp,
1627 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1628 {
1629 compat_sigset_t ss32;
1630 sigset_t ksigmask, sigsaved;
1631 struct compat_timespec ts;
1632 struct timespec end_time, *to = NULL;
1633 int ret;
1634
1635 if (tsp) {
1636 if (copy_from_user(&ts, tsp, sizeof(ts)))
1637 return -EFAULT;
1638
1639 to = &end_time;
1640 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1641 return -EINVAL;
1642 }
1643
1644 if (sigmask) {
1645 if (sigsetsize != sizeof(compat_sigset_t))
1646 return -EINVAL;
1647 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1648 return -EFAULT;
1649 sigset_from_compat(&ksigmask, &ss32);
1650
1651 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1652 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1653 }
1654
1655 ret = do_sys_poll(ufds, nfds, to);
1656
1657 /* We can restart this syscall, usually */
1658 if (ret == -EINTR) {
1659 /*
1660 * Don't restore the signal mask yet. Let do_signal() deliver
1661 * the signal on the way back to userspace, before the signal
1662 * mask is restored.
1663 */
1664 if (sigmask) {
1665 memcpy(&current->saved_sigmask, &sigsaved,
1666 sizeof(sigsaved));
1667 set_restore_sigmask();
1668 }
1669 ret = -ERESTARTNOHAND;
1670 } else if (sigmask)
1671 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1672
1673 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1674
1675 return ret;
1676 }
1677 #endif /* HAVE_SET_RESTORE_SIGMASK */
1678
1679 #ifdef CONFIG_EPOLL
1680
1681 #ifdef HAVE_SET_RESTORE_SIGMASK
1682 asmlinkage long compat_sys_epoll_pwait(int epfd,
1683 struct compat_epoll_event __user *events,
1684 int maxevents, int timeout,
1685 const compat_sigset_t __user *sigmask,
1686 compat_size_t sigsetsize)
1687 {
1688 long err;
1689 compat_sigset_t csigmask;
1690 sigset_t ksigmask, sigsaved;
1691
1692 /*
1693 * If the caller wants a certain signal mask to be set during the wait,
1694 * we apply it here.
1695 */
1696 if (sigmask) {
1697 if (sigsetsize != sizeof(compat_sigset_t))
1698 return -EINVAL;
1699 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1700 return -EFAULT;
1701 sigset_from_compat(&ksigmask, &csigmask);
1702 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1703 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1704 }
1705
1706 err = sys_epoll_wait(epfd, events, maxevents, timeout);
1707
1708 /*
1709 * If we changed the signal mask, we need to restore the original one.
1710 * In case we've got a signal while waiting, we do not restore the
1711 * signal mask yet, and we allow do_signal() to deliver the signal on
1712 * the way back to userspace, before the signal mask is restored.
1713 */
1714 if (sigmask) {
1715 if (err == -EINTR) {
1716 memcpy(&current->saved_sigmask, &sigsaved,
1717 sizeof(sigsaved));
1718 set_restore_sigmask();
1719 } else
1720 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1721 }
1722
1723 return err;
1724 }
1725 #endif /* HAVE_SET_RESTORE_SIGMASK */
1726
1727 #endif /* CONFIG_EPOLL */
1728
1729 #ifdef CONFIG_SIGNALFD
1730
1731 asmlinkage long compat_sys_signalfd4(int ufd,
1732 const compat_sigset_t __user *sigmask,
1733 compat_size_t sigsetsize, int flags)
1734 {
1735 compat_sigset_t ss32;
1736 sigset_t tmp;
1737 sigset_t __user *ksigmask;
1738
1739 if (sigsetsize != sizeof(compat_sigset_t))
1740 return -EINVAL;
1741 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1742 return -EFAULT;
1743 sigset_from_compat(&tmp, &ss32);
1744 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1745 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1746 return -EFAULT;
1747
1748 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1749 }
1750
1751 asmlinkage long compat_sys_signalfd(int ufd,
1752 const compat_sigset_t __user *sigmask,
1753 compat_size_t sigsetsize)
1754 {
1755 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1756 }
1757 #endif /* CONFIG_SIGNALFD */
1758
1759 #ifdef CONFIG_TIMERFD
1760
1761 asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1762 const struct compat_itimerspec __user *utmr,
1763 struct compat_itimerspec __user *otmr)
1764 {
1765 int error;
1766 struct itimerspec t;
1767 struct itimerspec __user *ut;
1768
1769 if (get_compat_itimerspec(&t, utmr))
1770 return -EFAULT;
1771 ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1772 if (copy_to_user(&ut[0], &t, sizeof(t)))
1773 return -EFAULT;
1774 error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1775 if (!error && otmr)
1776 error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1777 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1778
1779 return error;
1780 }
1781
1782 asmlinkage long compat_sys_timerfd_gettime(int ufd,
1783 struct compat_itimerspec __user *otmr)
1784 {
1785 int error;
1786 struct itimerspec t;
1787 struct itimerspec __user *ut;
1788
1789 ut = compat_alloc_user_space(sizeof(struct itimerspec));
1790 error = sys_timerfd_gettime(ufd, ut);
1791 if (!error)
1792 error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1793 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1794
1795 return error;
1796 }
1797
1798 #endif /* CONFIG_TIMERFD */
1799
1800 #ifdef CONFIG_FHANDLE
1801 /*
1802 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1803 * doesn't set the O_LARGEFILE flag.
1804 */
1805 asmlinkage long
1806 compat_sys_open_by_handle_at(int mountdirfd,
1807 struct file_handle __user *handle, int flags)
1808 {
1809 return do_handle_open(mountdirfd, handle, flags);
1810 }
1811 #endif
linux-next