?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv6: add support of equal cost multipath (ECMP)
[linux-2.6.git] / fs / compat.c
blob015e1e1f87c649ff48df282c6c04780569c3192d
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/dirent.h>
37 #include <linux/fsnotify.h>
38 #include <linux/highuid.h>
39 #include <linux/personality.h>
40 #include <linux/rwsem.h>
41 #include <linux/tsacct_kern.h>
42 #include <linux/security.h>
43 #include <linux/highmem.h>
44 #include <linux/signal.h>
45 #include <linux/poll.h>
46 #include <linux/mm.h>
47 #include <linux/eventpoll.h>
48 #include <linux/fs_struct.h>
49 #include <linux/slab.h>
50 #include <linux/pagemap.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/mmu_context.h>
54 #include <asm/ioctls.h>
55 #include "internal.h"
56
57 int compat_log = 1;
58
59 int compat_printk(const char *fmt, ...)
60 {
61 va_list ap;
62 int ret;
63 if (!compat_log)
64 return 0;
65 va_start(ap, fmt);
66 ret = vprintk(fmt, ap);
67 va_end(ap);
68 return ret;
69 }
70
71 #include "read_write.h"
72
73 /*
74 * Not all architectures have sys_utime, so implement this in terms
75 * of sys_utimes.
76 */
77 asmlinkage long compat_sys_utime(const char __user *filename,
78 struct compat_utimbuf __user *t)
79 {
80 struct timespec tv[2];
81
82 if (t) {
83 if (get_user(tv[0].tv_sec, &t->actime) ||
84 get_user(tv[1].tv_sec, &t->modtime))
85 return -EFAULT;
86 tv[0].tv_nsec = 0;
87 tv[1].tv_nsec = 0;
88 }
89 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
90 }
91
92 asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
93 {
94 struct timespec tv[2];
95
96 if (t) {
97 if (get_compat_timespec(&tv[0], &t[0]) ||
98 get_compat_timespec(&tv[1], &t[1]))
99 return -EFAULT;
100
101 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
102 return 0;
103 }
104 return do_utimes(dfd, filename, t ? tv : NULL, flags);
105 }
106
107 asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
108 {
109 struct timespec tv[2];
110
111 if (t) {
112 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
113 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
114 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
115 get_user(tv[1].tv_nsec, &t[1].tv_usec))
116 return -EFAULT;
117 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
118 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
119 return -EINVAL;
120 tv[0].tv_nsec *= 1000;
121 tv[1].tv_nsec *= 1000;
122 }
123 return do_utimes(dfd, filename, t ? tv : NULL, 0);
124 }
125
126 asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
127 {
128 return compat_sys_futimesat(AT_FDCWD, filename, t);
129 }
130
131 static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
132 {
133 struct compat_stat tmp;
134
135 if (!old_valid_dev(stat->dev) || !old_valid_dev(stat->rdev))
136 return -EOVERFLOW;
137
138 memset(&tmp, 0, sizeof(tmp));
139 tmp.st_dev = old_encode_dev(stat->dev);
140 tmp.st_ino = stat->ino;
141 if (sizeof(tmp.st_ino) < sizeof(stat->ino) && tmp.st_ino != stat->ino)
142 return -EOVERFLOW;
143 tmp.st_mode = stat->mode;
144 tmp.st_nlink = stat->nlink;
145 if (tmp.st_nlink != stat->nlink)
146 return -EOVERFLOW;
147 SET_UID(tmp.st_uid, from_kuid_munged(current_user_ns(), stat->uid));
148 SET_GID(tmp.st_gid, from_kgid_munged(current_user_ns(), stat->gid));
149 tmp.st_rdev = old_encode_dev(stat->rdev);
150 if ((u64) stat->size > MAX_NON_LFS)
151 return -EOVERFLOW;
152 tmp.st_size = stat->size;
153 tmp.st_atime = stat->atime.tv_sec;
154 tmp.st_atime_nsec = stat->atime.tv_nsec;
155 tmp.st_mtime = stat->mtime.tv_sec;
156 tmp.st_mtime_nsec = stat->mtime.tv_nsec;
157 tmp.st_ctime = stat->ctime.tv_sec;
158 tmp.st_ctime_nsec = stat->ctime.tv_nsec;
159 tmp.st_blocks = stat->blocks;
160 tmp.st_blksize = stat->blksize;
161 return copy_to_user(ubuf, &tmp, sizeof(tmp)) ? -EFAULT : 0;
162 }
163
164 asmlinkage long compat_sys_newstat(const char __user * filename,
165 struct compat_stat __user *statbuf)
166 {
167 struct kstat stat;
168 int error;
169
170 error = vfs_stat(filename, &stat);
171 if (error)
172 return error;
173 return cp_compat_stat(&stat, statbuf);
174 }
175
176 asmlinkage long compat_sys_newlstat(const char __user * filename,
177 struct compat_stat __user *statbuf)
178 {
179 struct kstat stat;
180 int error;
181
182 error = vfs_lstat(filename, &stat);
183 if (error)
184 return error;
185 return cp_compat_stat(&stat, statbuf);
186 }
187
188 #ifndef __ARCH_WANT_STAT64
189 asmlinkage long compat_sys_newfstatat(unsigned int dfd,
190 const char __user *filename,
191 struct compat_stat __user *statbuf, int flag)
192 {
193 struct kstat stat;
194 int error;
195
196 error = vfs_fstatat(dfd, filename, &stat, flag);
197 if (error)
198 return error;
199 return cp_compat_stat(&stat, statbuf);
200 }
201 #endif
202
203 asmlinkage long compat_sys_newfstat(unsigned int fd,
204 struct compat_stat __user * statbuf)
205 {
206 struct kstat stat;
207 int error = vfs_fstat(fd, &stat);
208
209 if (!error)
210 error = cp_compat_stat(&stat, statbuf);
211 return error;
212 }
213
214 static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
215 {
216
217 if (sizeof ubuf->f_blocks == 4) {
218 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
219 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
220 return -EOVERFLOW;
221 /* f_files and f_ffree may be -1; it's okay
222 * to stuff that into 32 bits */
223 if (kbuf->f_files != 0xffffffffffffffffULL
224 && (kbuf->f_files & 0xffffffff00000000ULL))
225 return -EOVERFLOW;
226 if (kbuf->f_ffree != 0xffffffffffffffffULL
227 && (kbuf->f_ffree & 0xffffffff00000000ULL))
228 return -EOVERFLOW;
229 }
230 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
231 __put_user(kbuf->f_type, &ubuf->f_type) ||
232 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
233 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
234 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
235 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
236 __put_user(kbuf->f_files, &ubuf->f_files) ||
237 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
238 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
239 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
240 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
241 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
242 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
243 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
244 return -EFAULT;
245 return 0;
246 }
247
248 /*
249 * The following statfs calls are copies of code from fs/statfs.c and
250 * should be checked against those from time to time
251 */
252 asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
253 {
254 struct kstatfs tmp;
255 int error = user_statfs(pathname, &tmp);
256 if (!error)
257 error = put_compat_statfs(buf, &tmp);
258 return error;
259 }
260
261 asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
262 {
263 struct kstatfs tmp;
264 int error = fd_statfs(fd, &tmp);
265 if (!error)
266 error = put_compat_statfs(buf, &tmp);
267 return error;
268 }
269
270 static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
271 {
272 if (sizeof ubuf->f_blocks == 4) {
273 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
274 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
275 return -EOVERFLOW;
276 /* f_files and f_ffree may be -1; it's okay
277 * to stuff that into 32 bits */
278 if (kbuf->f_files != 0xffffffffffffffffULL
279 && (kbuf->f_files & 0xffffffff00000000ULL))
280 return -EOVERFLOW;
281 if (kbuf->f_ffree != 0xffffffffffffffffULL
282 && (kbuf->f_ffree & 0xffffffff00000000ULL))
283 return -EOVERFLOW;
284 }
285 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
286 __put_user(kbuf->f_type, &ubuf->f_type) ||
287 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
288 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
289 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
290 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
291 __put_user(kbuf->f_files, &ubuf->f_files) ||
292 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
293 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
294 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
295 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
296 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
297 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
298 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
299 return -EFAULT;
300 return 0;
301 }
302
303 asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
304 {
305 struct kstatfs tmp;
306 int error;
307
308 if (sz != sizeof(*buf))
309 return -EINVAL;
310
311 error = user_statfs(pathname, &tmp);
312 if (!error)
313 error = put_compat_statfs64(buf, &tmp);
314 return error;
315 }
316
317 asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
318 {
319 struct kstatfs tmp;
320 int error;
321
322 if (sz != sizeof(*buf))
323 return -EINVAL;
324
325 error = fd_statfs(fd, &tmp);
326 if (!error)
327 error = put_compat_statfs64(buf, &tmp);
328 return error;
329 }
330
331 /*
332 * This is a copy of sys_ustat, just dealing with a structure layout.
333 * Given how simple this syscall is that apporach is more maintainable
334 * than the various conversion hacks.
335 */
336 asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
337 {
338 struct compat_ustat tmp;
339 struct kstatfs sbuf;
340 int err = vfs_ustat(new_decode_dev(dev), &sbuf);
341 if (err)
342 return err;
343
344 memset(&tmp, 0, sizeof(struct compat_ustat));
345 tmp.f_tfree = sbuf.f_bfree;
346 tmp.f_tinode = sbuf.f_ffree;
347 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
348 return -EFAULT;
349 return 0;
350 }
351
352 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
353 {
354 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
355 __get_user(kfl->l_type, &ufl->l_type) ||
356 __get_user(kfl->l_whence, &ufl->l_whence) ||
357 __get_user(kfl->l_start, &ufl->l_start) ||
358 __get_user(kfl->l_len, &ufl->l_len) ||
359 __get_user(kfl->l_pid, &ufl->l_pid))
360 return -EFAULT;
361 return 0;
362 }
363
364 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
365 {
366 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
367 __put_user(kfl->l_type, &ufl->l_type) ||
368 __put_user(kfl->l_whence, &ufl->l_whence) ||
369 __put_user(kfl->l_start, &ufl->l_start) ||
370 __put_user(kfl->l_len, &ufl->l_len) ||
371 __put_user(kfl->l_pid, &ufl->l_pid))
372 return -EFAULT;
373 return 0;
374 }
375
376 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
377 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
378 {
379 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
380 __get_user(kfl->l_type, &ufl->l_type) ||
381 __get_user(kfl->l_whence, &ufl->l_whence) ||
382 __get_user(kfl->l_start, &ufl->l_start) ||
383 __get_user(kfl->l_len, &ufl->l_len) ||
384 __get_user(kfl->l_pid, &ufl->l_pid))
385 return -EFAULT;
386 return 0;
387 }
388 #endif
389
390 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
391 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
392 {
393 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
394 __put_user(kfl->l_type, &ufl->l_type) ||
395 __put_user(kfl->l_whence, &ufl->l_whence) ||
396 __put_user(kfl->l_start, &ufl->l_start) ||
397 __put_user(kfl->l_len, &ufl->l_len) ||
398 __put_user(kfl->l_pid, &ufl->l_pid))
399 return -EFAULT;
400 return 0;
401 }
402 #endif
403
404 asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
405 unsigned long arg)
406 {
407 mm_segment_t old_fs;
408 struct flock f;
409 long ret;
410
411 switch (cmd) {
412 case F_GETLK:
413 case F_SETLK:
414 case F_SETLKW:
415 ret = get_compat_flock(&f, compat_ptr(arg));
416 if (ret != 0)
417 break;
418 old_fs = get_fs();
419 set_fs(KERNEL_DS);
420 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
421 set_fs(old_fs);
422 if (cmd == F_GETLK && ret == 0) {
423 /* GETLK was successful and we need to return the data...
424 * but it needs to fit in the compat structure.
425 * l_start shouldn't be too big, unless the original
426 * start + end is greater than COMPAT_OFF_T_MAX, in which
427 * case the app was asking for trouble, so we return
428 * -EOVERFLOW in that case.
429 * l_len could be too big, in which case we just truncate it,
430 * and only allow the app to see that part of the conflicting
431 * lock that might make sense to it anyway
432 */
433
434 if (f.l_start > COMPAT_OFF_T_MAX)
435 ret = -EOVERFLOW;
436 if (f.l_len > COMPAT_OFF_T_MAX)
437 f.l_len = COMPAT_OFF_T_MAX;
438 if (ret == 0)
439 ret = put_compat_flock(&f, compat_ptr(arg));
440 }
441 break;
442
443 case F_GETLK64:
444 case F_SETLK64:
445 case F_SETLKW64:
446 ret = get_compat_flock64(&f, compat_ptr(arg));
447 if (ret != 0)
448 break;
449 old_fs = get_fs();
450 set_fs(KERNEL_DS);
451 ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
452 ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
453 (unsigned long)&f);
454 set_fs(old_fs);
455 if (cmd == F_GETLK64 && ret == 0) {
456 /* need to return lock information - see above for commentary */
457 if (f.l_start > COMPAT_LOFF_T_MAX)
458 ret = -EOVERFLOW;
459 if (f.l_len > COMPAT_LOFF_T_MAX)
460 f.l_len = COMPAT_LOFF_T_MAX;
461 if (ret == 0)
462 ret = put_compat_flock64(&f, compat_ptr(arg));
463 }
464 break;
465
466 default:
467 ret = sys_fcntl(fd, cmd, arg);
468 break;
469 }
470 return ret;
471 }
472
473 asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
474 unsigned long arg)
475 {
476 if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
477 return -EINVAL;
478 return compat_sys_fcntl64(fd, cmd, arg);
479 }
480
481 asmlinkage long
482 compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
483 {
484 long ret;
485 aio_context_t ctx64;
486
487 mm_segment_t oldfs = get_fs();
488 if (unlikely(get_user(ctx64, ctx32p)))
489 return -EFAULT;
490
491 set_fs(KERNEL_DS);
492 /* The __user pointer cast is valid because of the set_fs() */
493 ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
494 set_fs(oldfs);
495 /* truncating is ok because it's a user address */
496 if (!ret)
497 ret = put_user((u32) ctx64, ctx32p);
498 return ret;
499 }
500
501 asmlinkage long
502 compat_sys_io_getevents(aio_context_t ctx_id,
503 unsigned long min_nr,
504 unsigned long nr,
505 struct io_event __user *events,
506 struct compat_timespec __user *timeout)
507 {
508 long ret;
509 struct timespec t;
510 struct timespec __user *ut = NULL;
511
512 ret = -EFAULT;
513 if (unlikely(!access_ok(VERIFY_WRITE, events,
514 nr * sizeof(struct io_event))))
515 goto out;
516 if (timeout) {
517 if (get_compat_timespec(&t, timeout))
518 goto out;
519
520 ut = compat_alloc_user_space(sizeof(*ut));
521 if (copy_to_user(ut, &t, sizeof(t)) )
522 goto out;
523 }
524 ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
525 out:
526 return ret;
527 }
528
529 /* A write operation does a read from user space and vice versa */
530 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
531
532 ssize_t compat_rw_copy_check_uvector(int type,
533 const struct compat_iovec __user *uvector, unsigned long nr_segs,
534 unsigned long fast_segs, struct iovec *fast_pointer,
535 struct iovec **ret_pointer)
536 {
537 compat_ssize_t tot_len;
538 struct iovec *iov = *ret_pointer = fast_pointer;
539 ssize_t ret = 0;
540 int seg;
541
542 /*
543 * SuS says "The readv() function *may* fail if the iovcnt argument
544 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
545 * traditionally returned zero for zero segments, so...
546 */
547 if (nr_segs == 0)
548 goto out;
549
550 ret = -EINVAL;
551 if (nr_segs > UIO_MAXIOV || nr_segs < 0)
552 goto out;
553 if (nr_segs > fast_segs) {
554 ret = -ENOMEM;
555 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
556 if (iov == NULL)
557 goto out;
558 }
559 *ret_pointer = iov;
560
561 /*
562 * Single unix specification:
563 * We should -EINVAL if an element length is not >= 0 and fitting an
564 * ssize_t.
565 *
566 * In Linux, the total length is limited to MAX_RW_COUNT, there is
567 * no overflow possibility.
568 */
569 tot_len = 0;
570 ret = -EINVAL;
571 for (seg = 0; seg < nr_segs; seg++) {
572 compat_uptr_t buf;
573 compat_ssize_t len;
574
575 if (__get_user(len, &uvector->iov_len) ||
576 __get_user(buf, &uvector->iov_base)) {
577 ret = -EFAULT;
578 goto out;
579 }
580 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
581 goto out;
582 if (type >= 0 &&
583 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
584 ret = -EFAULT;
585 goto out;
586 }
587 if (len > MAX_RW_COUNT - tot_len)
588 len = MAX_RW_COUNT - tot_len;
589 tot_len += len;
590 iov->iov_base = compat_ptr(buf);
591 iov->iov_len = (compat_size_t) len;
592 uvector++;
593 iov++;
594 }
595 ret = tot_len;
596
597 out:
598 return ret;
599 }
600
601 static inline long
602 copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
603 {
604 compat_uptr_t uptr;
605 int i;
606
607 for (i = 0; i < nr; ++i) {
608 if (get_user(uptr, ptr32 + i))
609 return -EFAULT;
610 if (put_user(compat_ptr(uptr), ptr64 + i))
611 return -EFAULT;
612 }
613 return 0;
614 }
615
616 #define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *))
617
618 asmlinkage long
619 compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
620 {
621 struct iocb __user * __user *iocb64;
622 long ret;
623
624 if (unlikely(nr < 0))
625 return -EINVAL;
626
627 if (nr > MAX_AIO_SUBMITS)
628 nr = MAX_AIO_SUBMITS;
629
630 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
631 ret = copy_iocb(nr, iocb, iocb64);
632 if (!ret)
633 ret = do_io_submit(ctx_id, nr, iocb64, 1);
634 return ret;
635 }
636
637 struct compat_ncp_mount_data {
638 compat_int_t version;
639 compat_uint_t ncp_fd;
640 __compat_uid_t mounted_uid;
641 compat_pid_t wdog_pid;
642 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
643 compat_uint_t time_out;
644 compat_uint_t retry_count;
645 compat_uint_t flags;
646 __compat_uid_t uid;
647 __compat_gid_t gid;
648 compat_mode_t file_mode;
649 compat_mode_t dir_mode;
650 };
651
652 struct compat_ncp_mount_data_v4 {
653 compat_int_t version;
654 compat_ulong_t flags;
655 compat_ulong_t mounted_uid;
656 compat_long_t wdog_pid;
657 compat_uint_t ncp_fd;
658 compat_uint_t time_out;
659 compat_uint_t retry_count;
660 compat_ulong_t uid;
661 compat_ulong_t gid;
662 compat_ulong_t file_mode;
663 compat_ulong_t dir_mode;
664 };
665
666 static void *do_ncp_super_data_conv(void *raw_data)
667 {
668 int version = *(unsigned int *)raw_data;
669
670 if (version == 3) {
671 struct compat_ncp_mount_data *c_n = raw_data;
672 struct ncp_mount_data *n = raw_data;
673
674 n->dir_mode = c_n->dir_mode;
675 n->file_mode = c_n->file_mode;
676 n->gid = c_n->gid;
677 n->uid = c_n->uid;
678 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
679 n->wdog_pid = c_n->wdog_pid;
680 n->mounted_uid = c_n->mounted_uid;
681 } else if (version == 4) {
682 struct compat_ncp_mount_data_v4 *c_n = raw_data;
683 struct ncp_mount_data_v4 *n = raw_data;
684
685 n->dir_mode = c_n->dir_mode;
686 n->file_mode = c_n->file_mode;
687 n->gid = c_n->gid;
688 n->uid = c_n->uid;
689 n->retry_count = c_n->retry_count;
690 n->time_out = c_n->time_out;
691 n->ncp_fd = c_n->ncp_fd;
692 n->wdog_pid = c_n->wdog_pid;
693 n->mounted_uid = c_n->mounted_uid;
694 n->flags = c_n->flags;
695 } else if (version != 5) {
696 return NULL;
697 }
698
699 return raw_data;
700 }
701
702
703 struct compat_nfs_string {
704 compat_uint_t len;
705 compat_uptr_t data;
706 };
707
708 static inline void compat_nfs_string(struct nfs_string *dst,
709 struct compat_nfs_string *src)
710 {
711 dst->data = compat_ptr(src->data);
712 dst->len = src->len;
713 }
714
715 struct compat_nfs4_mount_data_v1 {
716 compat_int_t version;
717 compat_int_t flags;
718 compat_int_t rsize;
719 compat_int_t wsize;
720 compat_int_t timeo;
721 compat_int_t retrans;
722 compat_int_t acregmin;
723 compat_int_t acregmax;
724 compat_int_t acdirmin;
725 compat_int_t acdirmax;
726 struct compat_nfs_string client_addr;
727 struct compat_nfs_string mnt_path;
728 struct compat_nfs_string hostname;
729 compat_uint_t host_addrlen;
730 compat_uptr_t host_addr;
731 compat_int_t proto;
732 compat_int_t auth_flavourlen;
733 compat_uptr_t auth_flavours;
734 };
735
736 static int do_nfs4_super_data_conv(void *raw_data)
737 {
738 int version = *(compat_uint_t *) raw_data;
739
740 if (version == 1) {
741 struct compat_nfs4_mount_data_v1 *raw = raw_data;
742 struct nfs4_mount_data *real = raw_data;
743
744 /* copy the fields backwards */
745 real->auth_flavours = compat_ptr(raw->auth_flavours);
746 real->auth_flavourlen = raw->auth_flavourlen;
747 real->proto = raw->proto;
748 real->host_addr = compat_ptr(raw->host_addr);
749 real->host_addrlen = raw->host_addrlen;
750 compat_nfs_string(&real->hostname, &raw->hostname);
751 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
752 compat_nfs_string(&real->client_addr, &raw->client_addr);
753 real->acdirmax = raw->acdirmax;
754 real->acdirmin = raw->acdirmin;
755 real->acregmax = raw->acregmax;
756 real->acregmin = raw->acregmin;
757 real->retrans = raw->retrans;
758 real->timeo = raw->timeo;
759 real->wsize = raw->wsize;
760 real->rsize = raw->rsize;
761 real->flags = raw->flags;
762 real->version = raw->version;
763 }
764
765 return 0;
766 }
767
768 #define NCPFS_NAME "ncpfs"
769 #define NFS4_NAME "nfs4"
770
771 asmlinkage long compat_sys_mount(const char __user * dev_name,
772 const char __user * dir_name,
773 const char __user * type, unsigned long flags,
774 const void __user * data)
775 {
776 char *kernel_type;
777 unsigned long data_page;
778 char *kernel_dev;
779 struct filename *dir;
780 int retval;
781
782 retval = copy_mount_string(type, &kernel_type);
783 if (retval < 0)
784 goto out;
785
786 dir = getname(dir_name);
787 retval = PTR_ERR(dir);
788 if (IS_ERR(dir))
789 goto out1;
790
791 retval = copy_mount_string(dev_name, &kernel_dev);
792 if (retval < 0)
793 goto out2;
794
795 retval = copy_mount_options(data, &data_page);
796 if (retval < 0)
797 goto out3;
798
799 retval = -EINVAL;
800
801 if (kernel_type && data_page) {
802 if (!strcmp(kernel_type, NCPFS_NAME)) {
803 do_ncp_super_data_conv((void *)data_page);
804 } else if (!strcmp(kernel_type, NFS4_NAME)) {
805 if (do_nfs4_super_data_conv((void *) data_page))
806 goto out4;
807 }
808 }
809
810 retval = do_mount(kernel_dev, dir->name, kernel_type,
811 flags, (void*)data_page);
812
813 out4:
814 free_page(data_page);
815 out3:
816 kfree(kernel_dev);
817 out2:
818 putname(dir);
819 out1:
820 kfree(kernel_type);
821 out:
822 return retval;
823 }
824
825 struct compat_old_linux_dirent {
826 compat_ulong_t d_ino;
827 compat_ulong_t d_offset;
828 unsigned short d_namlen;
829 char d_name[1];
830 };
831
832 struct compat_readdir_callback {
833 struct compat_old_linux_dirent __user *dirent;
834 int result;
835 };
836
837 static int compat_fillonedir(void *__buf, const char *name, int namlen,
838 loff_t offset, u64 ino, unsigned int d_type)
839 {
840 struct compat_readdir_callback *buf = __buf;
841 struct compat_old_linux_dirent __user *dirent;
842 compat_ulong_t d_ino;
843
844 if (buf->result)
845 return -EINVAL;
846 d_ino = ino;
847 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
848 buf->result = -EOVERFLOW;
849 return -EOVERFLOW;
850 }
851 buf->result++;
852 dirent = buf->dirent;
853 if (!access_ok(VERIFY_WRITE, dirent,
854 (unsigned long)(dirent->d_name + namlen + 1) -
855 (unsigned long)dirent))
856 goto efault;
857 if ( __put_user(d_ino, &dirent->d_ino) ||
858 __put_user(offset, &dirent->d_offset) ||
859 __put_user(namlen, &dirent->d_namlen) ||
860 __copy_to_user(dirent->d_name, name, namlen) ||
861 __put_user(0, dirent->d_name + namlen))
862 goto efault;
863 return 0;
864 efault:
865 buf->result = -EFAULT;
866 return -EFAULT;
867 }
868
869 asmlinkage long compat_sys_old_readdir(unsigned int fd,
870 struct compat_old_linux_dirent __user *dirent, unsigned int count)
871 {
872 int error;
873 struct fd f = fdget(fd);
874 struct compat_readdir_callback buf;
875
876 if (!f.file)
877 return -EBADF;
878
879 buf.result = 0;
880 buf.dirent = dirent;
881
882 error = vfs_readdir(f.file, compat_fillonedir, &buf);
883 if (buf.result)
884 error = buf.result;
885
886 fdput(f);
887 return error;
888 }
889
890 struct compat_linux_dirent {
891 compat_ulong_t d_ino;
892 compat_ulong_t d_off;
893 unsigned short d_reclen;
894 char d_name[1];
895 };
896
897 struct compat_getdents_callback {
898 struct compat_linux_dirent __user *current_dir;
899 struct compat_linux_dirent __user *previous;
900 int count;
901 int error;
902 };
903
904 static int compat_filldir(void *__buf, const char *name, int namlen,
905 loff_t offset, u64 ino, unsigned int d_type)
906 {
907 struct compat_linux_dirent __user * dirent;
908 struct compat_getdents_callback *buf = __buf;
909 compat_ulong_t d_ino;
910 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
911 namlen + 2, sizeof(compat_long_t));
912
913 buf->error = -EINVAL; /* only used if we fail.. */
914 if (reclen > buf->count)
915 return -EINVAL;
916 d_ino = ino;
917 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
918 buf->error = -EOVERFLOW;
919 return -EOVERFLOW;
920 }
921 dirent = buf->previous;
922 if (dirent) {
923 if (__put_user(offset, &dirent->d_off))
924 goto efault;
925 }
926 dirent = buf->current_dir;
927 if (__put_user(d_ino, &dirent->d_ino))
928 goto efault;
929 if (__put_user(reclen, &dirent->d_reclen))
930 goto efault;
931 if (copy_to_user(dirent->d_name, name, namlen))
932 goto efault;
933 if (__put_user(0, dirent->d_name + namlen))
934 goto efault;
935 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
936 goto efault;
937 buf->previous = dirent;
938 dirent = (void __user *)dirent + reclen;
939 buf->current_dir = dirent;
940 buf->count -= reclen;
941 return 0;
942 efault:
943 buf->error = -EFAULT;
944 return -EFAULT;
945 }
946
947 asmlinkage long compat_sys_getdents(unsigned int fd,
948 struct compat_linux_dirent __user *dirent, unsigned int count)
949 {
950 struct fd f;
951 struct compat_linux_dirent __user * lastdirent;
952 struct compat_getdents_callback buf;
953 int error;
954
955 if (!access_ok(VERIFY_WRITE, dirent, count))
956 return -EFAULT;
957
958 f = fdget(fd);
959 if (!f.file)
960 return -EBADF;
961
962 buf.current_dir = dirent;
963 buf.previous = NULL;
964 buf.count = count;
965 buf.error = 0;
966
967 error = vfs_readdir(f.file, compat_filldir, &buf);
968 if (error >= 0)
969 error = buf.error;
970 lastdirent = buf.previous;
971 if (lastdirent) {
972 if (put_user(f.file->f_pos, &lastdirent->d_off))
973 error = -EFAULT;
974 else
975 error = count - buf.count;
976 }
977 fdput(f);
978 return error;
979 }
980
981 #ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
982
983 struct compat_getdents_callback64 {
984 struct linux_dirent64 __user *current_dir;
985 struct linux_dirent64 __user *previous;
986 int count;
987 int error;
988 };
989
990 static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
991 u64 ino, unsigned int d_type)
992 {
993 struct linux_dirent64 __user *dirent;
994 struct compat_getdents_callback64 *buf = __buf;
995 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
996 sizeof(u64));
997 u64 off;
998
999 buf->error = -EINVAL; /* only used if we fail.. */
1000 if (reclen > buf->count)
1001 return -EINVAL;
1002 dirent = buf->previous;
1003
1004 if (dirent) {
1005 if (__put_user_unaligned(offset, &dirent->d_off))
1006 goto efault;
1007 }
1008 dirent = buf->current_dir;
1009 if (__put_user_unaligned(ino, &dirent->d_ino))
1010 goto efault;
1011 off = 0;
1012 if (__put_user_unaligned(off, &dirent->d_off))
1013 goto efault;
1014 if (__put_user(reclen, &dirent->d_reclen))
1015 goto efault;
1016 if (__put_user(d_type, &dirent->d_type))
1017 goto efault;
1018 if (copy_to_user(dirent->d_name, name, namlen))
1019 goto efault;
1020 if (__put_user(0, dirent->d_name + namlen))
1021 goto efault;
1022 buf->previous = dirent;
1023 dirent = (void __user *)dirent + reclen;
1024 buf->current_dir = dirent;
1025 buf->count -= reclen;
1026 return 0;
1027 efault:
1028 buf->error = -EFAULT;
1029 return -EFAULT;
1030 }
1031
1032 asmlinkage long compat_sys_getdents64(unsigned int fd,
1033 struct linux_dirent64 __user * dirent, unsigned int count)
1034 {
1035 struct fd f;
1036 struct linux_dirent64 __user * lastdirent;
1037 struct compat_getdents_callback64 buf;
1038 int error;
1039
1040 if (!access_ok(VERIFY_WRITE, dirent, count))
1041 return -EFAULT;
1042
1043 f = fdget(fd);
1044 if (!f.file)
1045 return -EBADF;
1046
1047 buf.current_dir = dirent;
1048 buf.previous = NULL;
1049 buf.count = count;
1050 buf.error = 0;
1051
1052 error = vfs_readdir(f.file, compat_filldir64, &buf);
1053 if (error >= 0)
1054 error = buf.error;
1055 lastdirent = buf.previous;
1056 if (lastdirent) {
1057 typeof(lastdirent->d_off) d_off = f.file->f_pos;
1058 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1059 error = -EFAULT;
1060 else
1061 error = count - buf.count;
1062 }
1063 fdput(f);
1064 return error;
1065 }
1066 #endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1067
1068 static ssize_t compat_do_readv_writev(int type, struct file *file,
1069 const struct compat_iovec __user *uvector,
1070 unsigned long nr_segs, loff_t *pos)
1071 {
1072 compat_ssize_t tot_len;
1073 struct iovec iovstack[UIO_FASTIOV];
1074 struct iovec *iov = iovstack;
1075 ssize_t ret;
1076 io_fn_t fn;
1077 iov_fn_t fnv;
1078
1079 ret = -EINVAL;
1080 if (!file->f_op)
1081 goto out;
1082
1083 ret = -EFAULT;
1084 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1085 goto out;
1086
1087 tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1088 UIO_FASTIOV, iovstack, &iov);
1089 if (tot_len == 0) {
1090 ret = 0;
1091 goto out;
1092 }
1093
1094 ret = rw_verify_area(type, file, pos, tot_len);
1095 if (ret < 0)
1096 goto out;
1097
1098 fnv = NULL;
1099 if (type == READ) {
1100 fn = file->f_op->read;
1101 fnv = file->f_op->aio_read;
1102 } else {
1103 fn = (io_fn_t)file->f_op->write;
1104 fnv = file->f_op->aio_write;
1105 }
1106
1107 if (fnv)
1108 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1109 pos, fnv);
1110 else
1111 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1112
1113 out:
1114 if (iov != iovstack)
1115 kfree(iov);
1116 if ((ret + (type == READ)) > 0) {
1117 if (type == READ)
1118 fsnotify_access(file);
1119 else
1120 fsnotify_modify(file);
1121 }
1122 return ret;
1123 }
1124
1125 static size_t compat_readv(struct file *file,
1126 const struct compat_iovec __user *vec,
1127 unsigned long vlen, loff_t *pos)
1128 {
1129 ssize_t ret = -EBADF;
1130
1131 if (!(file->f_mode & FMODE_READ))
1132 goto out;
1133
1134 ret = -EINVAL;
1135 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1136 goto out;
1137
1138 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1139
1140 out:
1141 if (ret > 0)
1142 add_rchar(current, ret);
1143 inc_syscr(current);
1144 return ret;
1145 }
1146
1147 asmlinkage ssize_t
1148 compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1149 unsigned long vlen)
1150 {
1151 struct fd f = fdget(fd);
1152 ssize_t ret;
1153 loff_t pos;
1154
1155 if (!f.file)
1156 return -EBADF;
1157 pos = f.file->f_pos;
1158 ret = compat_readv(f.file, vec, vlen, &pos);
1159 f.file->f_pos = pos;
1160 fdput(f);
1161 return ret;
1162 }
1163
1164 asmlinkage ssize_t
1165 compat_sys_preadv64(unsigned long fd, const struct compat_iovec __user *vec,
1166 unsigned long vlen, loff_t pos)
1167 {
1168 struct fd f;
1169 ssize_t ret;
1170
1171 if (pos < 0)
1172 return -EINVAL;
1173 f = fdget(fd);
1174 if (!f.file)
1175 return -EBADF;
1176 ret = -ESPIPE;
1177 if (f.file->f_mode & FMODE_PREAD)
1178 ret = compat_readv(f.file, vec, vlen, &pos);
1179 fdput(f);
1180 return ret;
1181 }
1182
1183 asmlinkage ssize_t
1184 compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1185 unsigned long vlen, u32 pos_low, u32 pos_high)
1186 {
1187 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1188 return compat_sys_preadv64(fd, vec, vlen, pos);
1189 }
1190
1191 static size_t compat_writev(struct file *file,
1192 const struct compat_iovec __user *vec,
1193 unsigned long vlen, loff_t *pos)
1194 {
1195 ssize_t ret = -EBADF;
1196
1197 if (!(file->f_mode & FMODE_WRITE))
1198 goto out;
1199
1200 ret = -EINVAL;
1201 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1202 goto out;
1203
1204 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1205
1206 out:
1207 if (ret > 0)
1208 add_wchar(current, ret);
1209 inc_syscw(current);
1210 return ret;
1211 }
1212
1213 asmlinkage ssize_t
1214 compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1215 unsigned long vlen)
1216 {
1217 struct fd f = fdget(fd);
1218 ssize_t ret;
1219 loff_t pos;
1220
1221 if (!f.file)
1222 return -EBADF;
1223 pos = f.file->f_pos;
1224 ret = compat_writev(f.file, vec, vlen, &pos);
1225 f.file->f_pos = pos;
1226 fdput(f);
1227 return ret;
1228 }
1229
1230 asmlinkage ssize_t
1231 compat_sys_pwritev64(unsigned long fd, const struct compat_iovec __user *vec,
1232 unsigned long vlen, loff_t pos)
1233 {
1234 struct fd f;
1235 ssize_t ret;
1236
1237 if (pos < 0)
1238 return -EINVAL;
1239 f = fdget(fd);
1240 if (!f.file)
1241 return -EBADF;
1242 ret = -ESPIPE;
1243 if (f.file->f_mode & FMODE_PWRITE)
1244 ret = compat_writev(f.file, vec, vlen, &pos);
1245 fdput(f);
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 return compat_sys_pwritev64(fd, vec, vlen, pos);
1255 }
1256
1257 asmlinkage long
1258 compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1259 unsigned int nr_segs, unsigned int flags)
1260 {
1261 unsigned i;
1262 struct iovec __user *iov;
1263 if (nr_segs > UIO_MAXIOV)
1264 return -EINVAL;
1265 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1266 for (i = 0; i < nr_segs; i++) {
1267 struct compat_iovec v;
1268 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1269 get_user(v.iov_len, &iov32[i].iov_len) ||
1270 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1271 put_user(v.iov_len, &iov[i].iov_len))
1272 return -EFAULT;
1273 }
1274 return sys_vmsplice(fd, iov, nr_segs, flags);
1275 }
1276
1277 /*
1278 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1279 * O_LARGEFILE flag.
1280 */
1281 asmlinkage long
1282 compat_sys_open(const char __user *filename, int flags, umode_t mode)
1283 {
1284 return do_sys_open(AT_FDCWD, filename, flags, mode);
1285 }
1286
1287 /*
1288 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1289 * O_LARGEFILE flag.
1290 */
1291 asmlinkage long
1292 compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, umode_t mode)
1293 {
1294 return do_sys_open(dfd, filename, flags, mode);
1295 }
1296
1297 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1298
1299 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1300 int timeval, int ret)
1301 {
1302 struct timespec ts;
1303
1304 if (!p)
1305 return ret;
1306
1307 if (current->personality & STICKY_TIMEOUTS)
1308 goto sticky;
1309
1310 /* No update for zero timeout */
1311 if (!end_time->tv_sec && !end_time->tv_nsec)
1312 return ret;
1313
1314 ktime_get_ts(&ts);
1315 ts = timespec_sub(*end_time, ts);
1316 if (ts.tv_sec < 0)
1317 ts.tv_sec = ts.tv_nsec = 0;
1318
1319 if (timeval) {
1320 struct compat_timeval rtv;
1321
1322 rtv.tv_sec = ts.tv_sec;
1323 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1324
1325 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1326 return ret;
1327 } else {
1328 struct compat_timespec rts;
1329
1330 rts.tv_sec = ts.tv_sec;
1331 rts.tv_nsec = ts.tv_nsec;
1332
1333 if (!copy_to_user(p, &rts, sizeof(rts)))
1334 return ret;
1335 }
1336 /*
1337 * If an application puts its timeval in read-only memory, we
1338 * don't want the Linux-specific update to the timeval to
1339 * cause a fault after the select has completed
1340 * successfully. However, because we're not updating the
1341 * timeval, we can't restart the system call.
1342 */
1343
1344 sticky:
1345 if (ret == -ERESTARTNOHAND)
1346 ret = -EINTR;
1347 return ret;
1348 }
1349
1350 /*
1351 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1352 * 64-bit unsigned longs.
1353 */
1354 static
1355 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1356 unsigned long *fdset)
1357 {
1358 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1359 if (ufdset) {
1360 unsigned long odd;
1361
1362 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1363 return -EFAULT;
1364
1365 odd = nr & 1UL;
1366 nr &= ~1UL;
1367 while (nr) {
1368 unsigned long h, l;
1369 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1370 return -EFAULT;
1371 ufdset += 2;
1372 *fdset++ = h << 32 | l;
1373 nr -= 2;
1374 }
1375 if (odd && __get_user(*fdset, ufdset))
1376 return -EFAULT;
1377 } else {
1378 /* Tricky, must clear full unsigned long in the
1379 * kernel fdset at the end, this makes sure that
1380 * actually happens.
1381 */
1382 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1383 }
1384 return 0;
1385 }
1386
1387 static
1388 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1389 unsigned long *fdset)
1390 {
1391 unsigned long odd;
1392 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1393
1394 if (!ufdset)
1395 return 0;
1396
1397 odd = nr & 1UL;
1398 nr &= ~1UL;
1399 while (nr) {
1400 unsigned long h, l;
1401 l = *fdset++;
1402 h = l >> 32;
1403 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1404 return -EFAULT;
1405 ufdset += 2;
1406 nr -= 2;
1407 }
1408 if (odd && __put_user(*fdset, ufdset))
1409 return -EFAULT;
1410 return 0;
1411 }
1412
1413
1414 /*
1415 * This is a virtual copy of sys_select from fs/select.c and probably
1416 * should be compared to it from time to time
1417 */
1418
1419 /*
1420 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1421 * like to be certain this leads to no problems. So I return
1422 * EINTR just for safety.
1423 *
1424 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1425 * I'm trying ERESTARTNOHAND which restart only when you want to.
1426 */
1427 int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1428 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1429 struct timespec *end_time)
1430 {
1431 fd_set_bits fds;
1432 void *bits;
1433 int size, max_fds, ret = -EINVAL;
1434 struct fdtable *fdt;
1435 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1436
1437 if (n < 0)
1438 goto out_nofds;
1439
1440 /* max_fds can increase, so grab it once to avoid race */
1441 rcu_read_lock();
1442 fdt = files_fdtable(current->files);
1443 max_fds = fdt->max_fds;
1444 rcu_read_unlock();
1445 if (n > max_fds)
1446 n = max_fds;
1447
1448 /*
1449 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1450 * since we used fdset we need to allocate memory in units of
1451 * long-words.
1452 */
1453 size = FDS_BYTES(n);
1454 bits = stack_fds;
1455 if (size > sizeof(stack_fds) / 6) {
1456 bits = kmalloc(6 * size, GFP_KERNEL);
1457 ret = -ENOMEM;
1458 if (!bits)
1459 goto out_nofds;
1460 }
1461 fds.in = (unsigned long *) bits;
1462 fds.out = (unsigned long *) (bits + size);
1463 fds.ex = (unsigned long *) (bits + 2*size);
1464 fds.res_in = (unsigned long *) (bits + 3*size);
1465 fds.res_out = (unsigned long *) (bits + 4*size);
1466 fds.res_ex = (unsigned long *) (bits + 5*size);
1467
1468 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1469 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1470 (ret = compat_get_fd_set(n, exp, fds.ex)))
1471 goto out;
1472 zero_fd_set(n, fds.res_in);
1473 zero_fd_set(n, fds.res_out);
1474 zero_fd_set(n, fds.res_ex);
1475
1476 ret = do_select(n, &fds, end_time);
1477
1478 if (ret < 0)
1479 goto out;
1480 if (!ret) {
1481 ret = -ERESTARTNOHAND;
1482 if (signal_pending(current))
1483 goto out;
1484 ret = 0;
1485 }
1486
1487 if (compat_set_fd_set(n, inp, fds.res_in) ||
1488 compat_set_fd_set(n, outp, fds.res_out) ||
1489 compat_set_fd_set(n, exp, fds.res_ex))
1490 ret = -EFAULT;
1491 out:
1492 if (bits != stack_fds)
1493 kfree(bits);
1494 out_nofds:
1495 return ret;
1496 }
1497
1498 asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1499 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1500 struct compat_timeval __user *tvp)
1501 {
1502 struct timespec end_time, *to = NULL;
1503 struct compat_timeval tv;
1504 int ret;
1505
1506 if (tvp) {
1507 if (copy_from_user(&tv, tvp, sizeof(tv)))
1508 return -EFAULT;
1509
1510 to = &end_time;
1511 if (poll_select_set_timeout(to,
1512 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1513 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1514 return -EINVAL;
1515 }
1516
1517 ret = compat_core_sys_select(n, inp, outp, exp, to);
1518 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1519
1520 return ret;
1521 }
1522
1523 struct compat_sel_arg_struct {
1524 compat_ulong_t n;
1525 compat_uptr_t inp;
1526 compat_uptr_t outp;
1527 compat_uptr_t exp;
1528 compat_uptr_t tvp;
1529 };
1530
1531 asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1532 {
1533 struct compat_sel_arg_struct a;
1534
1535 if (copy_from_user(&a, arg, sizeof(a)))
1536 return -EFAULT;
1537 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1538 compat_ptr(a.exp), compat_ptr(a.tvp));
1539 }
1540
1541 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1542 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1543 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1544 compat_size_t sigsetsize)
1545 {
1546 compat_sigset_t ss32;
1547 sigset_t ksigmask, sigsaved;
1548 struct compat_timespec ts;
1549 struct timespec end_time, *to = NULL;
1550 int ret;
1551
1552 if (tsp) {
1553 if (copy_from_user(&ts, tsp, sizeof(ts)))
1554 return -EFAULT;
1555
1556 to = &end_time;
1557 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1558 return -EINVAL;
1559 }
1560
1561 if (sigmask) {
1562 if (sigsetsize != sizeof(compat_sigset_t))
1563 return -EINVAL;
1564 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1565 return -EFAULT;
1566 sigset_from_compat(&ksigmask, &ss32);
1567
1568 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1569 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1570 }
1571
1572 ret = compat_core_sys_select(n, inp, outp, exp, to);
1573 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1574
1575 if (ret == -ERESTARTNOHAND) {
1576 /*
1577 * Don't restore the signal mask yet. Let do_signal() deliver
1578 * the signal on the way back to userspace, before the signal
1579 * mask is restored.
1580 */
1581 if (sigmask) {
1582 memcpy(&current->saved_sigmask, &sigsaved,
1583 sizeof(sigsaved));
1584 set_restore_sigmask();
1585 }
1586 } else if (sigmask)
1587 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1588
1589 return ret;
1590 }
1591
1592 asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1593 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1594 struct compat_timespec __user *tsp, void __user *sig)
1595 {
1596 compat_size_t sigsetsize = 0;
1597 compat_uptr_t up = 0;
1598
1599 if (sig) {
1600 if (!access_ok(VERIFY_READ, sig,
1601 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1602 __get_user(up, (compat_uptr_t __user *)sig) ||
1603 __get_user(sigsetsize,
1604 (compat_size_t __user *)(sig+sizeof(up))))
1605 return -EFAULT;
1606 }
1607 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1608 sigsetsize);
1609 }
1610
1611 asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1612 unsigned int nfds, struct compat_timespec __user *tsp,
1613 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1614 {
1615 compat_sigset_t ss32;
1616 sigset_t ksigmask, sigsaved;
1617 struct compat_timespec ts;
1618 struct timespec end_time, *to = NULL;
1619 int ret;
1620
1621 if (tsp) {
1622 if (copy_from_user(&ts, tsp, sizeof(ts)))
1623 return -EFAULT;
1624
1625 to = &end_time;
1626 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1627 return -EINVAL;
1628 }
1629
1630 if (sigmask) {
1631 if (sigsetsize != sizeof(compat_sigset_t))
1632 return -EINVAL;
1633 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1634 return -EFAULT;
1635 sigset_from_compat(&ksigmask, &ss32);
1636
1637 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1638 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1639 }
1640
1641 ret = do_sys_poll(ufds, nfds, to);
1642
1643 /* We can restart this syscall, usually */
1644 if (ret == -EINTR) {
1645 /*
1646 * Don't restore the signal mask yet. Let do_signal() deliver
1647 * the signal on the way back to userspace, before the signal
1648 * mask is restored.
1649 */
1650 if (sigmask) {
1651 memcpy(&current->saved_sigmask, &sigsaved,
1652 sizeof(sigsaved));
1653 set_restore_sigmask();
1654 }
1655 ret = -ERESTARTNOHAND;
1656 } else if (sigmask)
1657 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1658
1659 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1660
1661 return ret;
1662 }
1663
1664 #ifdef CONFIG_EPOLL
1665
1666 asmlinkage long compat_sys_epoll_pwait(int epfd,
1667 struct compat_epoll_event __user *events,
1668 int maxevents, int timeout,
1669 const compat_sigset_t __user *sigmask,
1670 compat_size_t sigsetsize)
1671 {
1672 long err;
1673 compat_sigset_t csigmask;
1674 sigset_t ksigmask, sigsaved;
1675
1676 /*
1677 * If the caller wants a certain signal mask to be set during the wait,
1678 * we apply it here.
1679 */
1680 if (sigmask) {
1681 if (sigsetsize != sizeof(compat_sigset_t))
1682 return -EINVAL;
1683 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1684 return -EFAULT;
1685 sigset_from_compat(&ksigmask, &csigmask);
1686 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1687 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1688 }
1689
1690 err = sys_epoll_wait(epfd, events, maxevents, timeout);
1691
1692 /*
1693 * If we changed the signal mask, we need to restore the original one.
1694 * In case we've got a signal while waiting, we do not restore the
1695 * signal mask yet, and we allow do_signal() to deliver the signal on
1696 * the way back to userspace, before the signal mask is restored.
1697 */
1698 if (sigmask) {
1699 if (err == -EINTR) {
1700 memcpy(&current->saved_sigmask, &sigsaved,
1701 sizeof(sigsaved));
1702 set_restore_sigmask();
1703 } else
1704 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1705 }
1706
1707 return err;
1708 }
1709
1710 #endif /* CONFIG_EPOLL */
1711
1712 #ifdef CONFIG_SIGNALFD
1713
1714 asmlinkage long compat_sys_signalfd4(int ufd,
1715 const compat_sigset_t __user *sigmask,
1716 compat_size_t sigsetsize, int flags)
1717 {
1718 compat_sigset_t ss32;
1719 sigset_t tmp;
1720 sigset_t __user *ksigmask;
1721
1722 if (sigsetsize != sizeof(compat_sigset_t))
1723 return -EINVAL;
1724 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1725 return -EFAULT;
1726 sigset_from_compat(&tmp, &ss32);
1727 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1728 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1729 return -EFAULT;
1730
1731 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1732 }
1733
1734 asmlinkage long compat_sys_signalfd(int ufd,
1735 const compat_sigset_t __user *sigmask,
1736 compat_size_t sigsetsize)
1737 {
1738 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1739 }
1740 #endif /* CONFIG_SIGNALFD */
1741
1742 #ifdef CONFIG_TIMERFD
1743
1744 asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1745 const struct compat_itimerspec __user *utmr,
1746 struct compat_itimerspec __user *otmr)
1747 {
1748 int error;
1749 struct itimerspec t;
1750 struct itimerspec __user *ut;
1751
1752 if (get_compat_itimerspec(&t, utmr))
1753 return -EFAULT;
1754 ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1755 if (copy_to_user(&ut[0], &t, sizeof(t)))
1756 return -EFAULT;
1757 error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1758 if (!error && otmr)
1759 error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1760 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1761
1762 return error;
1763 }
1764
1765 asmlinkage long compat_sys_timerfd_gettime(int ufd,
1766 struct compat_itimerspec __user *otmr)
1767 {
1768 int error;
1769 struct itimerspec t;
1770 struct itimerspec __user *ut;
1771
1772 ut = compat_alloc_user_space(sizeof(struct itimerspec));
1773 error = sys_timerfd_gettime(ufd, ut);
1774 if (!error)
1775 error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1776 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1777
1778 return error;
1779 }
1780
1781 #endif /* CONFIG_TIMERFD */
1782
1783 #ifdef CONFIG_FHANDLE
1784 /*
1785 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1786 * doesn't set the O_LARGEFILE flag.
1787 */
1788 asmlinkage long
1789 compat_sys_open_by_handle_at(int mountdirfd,
1790 struct file_handle __user *handle, int flags)
1791 {
1792 return do_handle_open(mountdirfd, handle, flags);
1793 }
1794 #endif
1795
1796 #ifdef __ARCH_WANT_COMPAT_SYS_SENDFILE
1797 asmlinkage long compat_sys_sendfile(int out_fd, int in_fd,
1798 compat_off_t __user *offset, compat_size_t count)
1799 {
1800 loff_t pos;
1801 off_t off;
1802 ssize_t ret;
1803
1804 if (offset) {
1805 if (unlikely(get_user(off, offset)))
1806 return -EFAULT;
1807 pos = off;
1808 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1809 if (unlikely(put_user(pos, offset)))
1810 return -EFAULT;
1811 return ret;
1812 }
1813
1814 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1815 }
1816 #endif /* __ARCH_WANT_COMPAT_SYS_SENDFILE */
Linus' kernel tree