?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nfsd4: move principal name into svc_cred
[linux-2.6/libata-dev.git] / fs / compat.c
blobf2944ace7a7b4a06a29880ed8c55cdd2cccb1ada
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, stat->uid);
148 SET_GID(tmp.st_gid, 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, int check_access)
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 (check_access &&
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 char *dir_page;
780 int retval;
781
782 retval = copy_mount_string(type, &kernel_type);
783 if (retval < 0)
784 goto out;
785
786 dir_page = getname(dir_name);
787 retval = PTR_ERR(dir_page);
788 if (IS_ERR(dir_page))
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_page, 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_page);
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 file *file;
874 struct compat_readdir_callback buf;
875
876 error = -EBADF;
877 file = fget(fd);
878 if (!file)
879 goto out;
880
881 buf.result = 0;
882 buf.dirent = dirent;
883
884 error = vfs_readdir(file, compat_fillonedir, &buf);
885 if (buf.result)
886 error = buf.result;
887
888 fput(file);
889 out:
890 return error;
891 }
892
893 struct compat_linux_dirent {
894 compat_ulong_t d_ino;
895 compat_ulong_t d_off;
896 unsigned short d_reclen;
897 char d_name[1];
898 };
899
900 struct compat_getdents_callback {
901 struct compat_linux_dirent __user *current_dir;
902 struct compat_linux_dirent __user *previous;
903 int count;
904 int error;
905 };
906
907 static int compat_filldir(void *__buf, const char *name, int namlen,
908 loff_t offset, u64 ino, unsigned int d_type)
909 {
910 struct compat_linux_dirent __user * dirent;
911 struct compat_getdents_callback *buf = __buf;
912 compat_ulong_t d_ino;
913 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
914 namlen + 2, sizeof(compat_long_t));
915
916 buf->error = -EINVAL; /* only used if we fail.. */
917 if (reclen > buf->count)
918 return -EINVAL;
919 d_ino = ino;
920 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
921 buf->error = -EOVERFLOW;
922 return -EOVERFLOW;
923 }
924 dirent = buf->previous;
925 if (dirent) {
926 if (__put_user(offset, &dirent->d_off))
927 goto efault;
928 }
929 dirent = buf->current_dir;
930 if (__put_user(d_ino, &dirent->d_ino))
931 goto efault;
932 if (__put_user(reclen, &dirent->d_reclen))
933 goto efault;
934 if (copy_to_user(dirent->d_name, name, namlen))
935 goto efault;
936 if (__put_user(0, dirent->d_name + namlen))
937 goto efault;
938 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
939 goto efault;
940 buf->previous = dirent;
941 dirent = (void __user *)dirent + reclen;
942 buf->current_dir = dirent;
943 buf->count -= reclen;
944 return 0;
945 efault:
946 buf->error = -EFAULT;
947 return -EFAULT;
948 }
949
950 asmlinkage long compat_sys_getdents(unsigned int fd,
951 struct compat_linux_dirent __user *dirent, unsigned int count)
952 {
953 struct file * file;
954 struct compat_linux_dirent __user * lastdirent;
955 struct compat_getdents_callback buf;
956 int error;
957
958 error = -EFAULT;
959 if (!access_ok(VERIFY_WRITE, dirent, count))
960 goto out;
961
962 error = -EBADF;
963 file = fget(fd);
964 if (!file)
965 goto out;
966
967 buf.current_dir = dirent;
968 buf.previous = NULL;
969 buf.count = count;
970 buf.error = 0;
971
972 error = vfs_readdir(file, compat_filldir, &buf);
973 if (error >= 0)
974 error = buf.error;
975 lastdirent = buf.previous;
976 if (lastdirent) {
977 if (put_user(file->f_pos, &lastdirent->d_off))
978 error = -EFAULT;
979 else
980 error = count - buf.count;
981 }
982 fput(file);
983 out:
984 return error;
985 }
986
987 #ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
988
989 struct compat_getdents_callback64 {
990 struct linux_dirent64 __user *current_dir;
991 struct linux_dirent64 __user *previous;
992 int count;
993 int error;
994 };
995
996 static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
997 u64 ino, unsigned int d_type)
998 {
999 struct linux_dirent64 __user *dirent;
1000 struct compat_getdents_callback64 *buf = __buf;
1001 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1002 sizeof(u64));
1003 u64 off;
1004
1005 buf->error = -EINVAL; /* only used if we fail.. */
1006 if (reclen > buf->count)
1007 return -EINVAL;
1008 dirent = buf->previous;
1009
1010 if (dirent) {
1011 if (__put_user_unaligned(offset, &dirent->d_off))
1012 goto efault;
1013 }
1014 dirent = buf->current_dir;
1015 if (__put_user_unaligned(ino, &dirent->d_ino))
1016 goto efault;
1017 off = 0;
1018 if (__put_user_unaligned(off, &dirent->d_off))
1019 goto efault;
1020 if (__put_user(reclen, &dirent->d_reclen))
1021 goto efault;
1022 if (__put_user(d_type, &dirent->d_type))
1023 goto efault;
1024 if (copy_to_user(dirent->d_name, name, namlen))
1025 goto efault;
1026 if (__put_user(0, dirent->d_name + namlen))
1027 goto efault;
1028 buf->previous = dirent;
1029 dirent = (void __user *)dirent + reclen;
1030 buf->current_dir = dirent;
1031 buf->count -= reclen;
1032 return 0;
1033 efault:
1034 buf->error = -EFAULT;
1035 return -EFAULT;
1036 }
1037
1038 asmlinkage long compat_sys_getdents64(unsigned int fd,
1039 struct linux_dirent64 __user * dirent, unsigned int count)
1040 {
1041 struct file * file;
1042 struct linux_dirent64 __user * lastdirent;
1043 struct compat_getdents_callback64 buf;
1044 int error;
1045
1046 error = -EFAULT;
1047 if (!access_ok(VERIFY_WRITE, dirent, count))
1048 goto out;
1049
1050 error = -EBADF;
1051 file = fget(fd);
1052 if (!file)
1053 goto out;
1054
1055 buf.current_dir = dirent;
1056 buf.previous = NULL;
1057 buf.count = count;
1058 buf.error = 0;
1059
1060 error = vfs_readdir(file, compat_filldir64, &buf);
1061 if (error >= 0)
1062 error = buf.error;
1063 lastdirent = buf.previous;
1064 if (lastdirent) {
1065 typeof(lastdirent->d_off) d_off = file->f_pos;
1066 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1067 error = -EFAULT;
1068 else
1069 error = count - buf.count;
1070 }
1071 fput(file);
1072 out:
1073 return error;
1074 }
1075 #endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1076
1077 static ssize_t compat_do_readv_writev(int type, struct file *file,
1078 const struct compat_iovec __user *uvector,
1079 unsigned long nr_segs, loff_t *pos)
1080 {
1081 compat_ssize_t tot_len;
1082 struct iovec iovstack[UIO_FASTIOV];
1083 struct iovec *iov = iovstack;
1084 ssize_t ret;
1085 io_fn_t fn;
1086 iov_fn_t fnv;
1087
1088 ret = -EINVAL;
1089 if (!file->f_op)
1090 goto out;
1091
1092 ret = -EFAULT;
1093 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1094 goto out;
1095
1096 tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1097 UIO_FASTIOV, iovstack, &iov, 1);
1098 if (tot_len == 0) {
1099 ret = 0;
1100 goto out;
1101 }
1102
1103 ret = rw_verify_area(type, file, pos, tot_len);
1104 if (ret < 0)
1105 goto out;
1106
1107 fnv = NULL;
1108 if (type == READ) {
1109 fn = file->f_op->read;
1110 fnv = file->f_op->aio_read;
1111 } else {
1112 fn = (io_fn_t)file->f_op->write;
1113 fnv = file->f_op->aio_write;
1114 }
1115
1116 if (fnv)
1117 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1118 pos, fnv);
1119 else
1120 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1121
1122 out:
1123 if (iov != iovstack)
1124 kfree(iov);
1125 if ((ret + (type == READ)) > 0) {
1126 if (type == READ)
1127 fsnotify_access(file);
1128 else
1129 fsnotify_modify(file);
1130 }
1131 return ret;
1132 }
1133
1134 static size_t compat_readv(struct file *file,
1135 const struct compat_iovec __user *vec,
1136 unsigned long vlen, loff_t *pos)
1137 {
1138 ssize_t ret = -EBADF;
1139
1140 if (!(file->f_mode & FMODE_READ))
1141 goto out;
1142
1143 ret = -EINVAL;
1144 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1145 goto out;
1146
1147 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1148
1149 out:
1150 if (ret > 0)
1151 add_rchar(current, ret);
1152 inc_syscr(current);
1153 return ret;
1154 }
1155
1156 asmlinkage ssize_t
1157 compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1158 unsigned long vlen)
1159 {
1160 struct file *file;
1161 int fput_needed;
1162 ssize_t ret;
1163
1164 file = fget_light(fd, &fput_needed);
1165 if (!file)
1166 return -EBADF;
1167 ret = compat_readv(file, vec, vlen, &file->f_pos);
1168 fput_light(file, fput_needed);
1169 return ret;
1170 }
1171
1172 asmlinkage ssize_t
1173 compat_sys_preadv64(unsigned long fd, const struct compat_iovec __user *vec,
1174 unsigned long vlen, loff_t pos)
1175 {
1176 struct file *file;
1177 int fput_needed;
1178 ssize_t ret;
1179
1180 if (pos < 0)
1181 return -EINVAL;
1182 file = fget_light(fd, &fput_needed);
1183 if (!file)
1184 return -EBADF;
1185 ret = -ESPIPE;
1186 if (file->f_mode & FMODE_PREAD)
1187 ret = compat_readv(file, vec, vlen, &pos);
1188 fput_light(file, fput_needed);
1189 return ret;
1190 }
1191
1192 asmlinkage ssize_t
1193 compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1194 unsigned long vlen, u32 pos_low, u32 pos_high)
1195 {
1196 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1197 return compat_sys_preadv64(fd, vec, vlen, pos);
1198 }
1199
1200 static size_t compat_writev(struct file *file,
1201 const struct compat_iovec __user *vec,
1202 unsigned long vlen, loff_t *pos)
1203 {
1204 ssize_t ret = -EBADF;
1205
1206 if (!(file->f_mode & FMODE_WRITE))
1207 goto out;
1208
1209 ret = -EINVAL;
1210 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1211 goto out;
1212
1213 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1214
1215 out:
1216 if (ret > 0)
1217 add_wchar(current, ret);
1218 inc_syscw(current);
1219 return ret;
1220 }
1221
1222 asmlinkage ssize_t
1223 compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1224 unsigned long vlen)
1225 {
1226 struct file *file;
1227 int fput_needed;
1228 ssize_t ret;
1229
1230 file = fget_light(fd, &fput_needed);
1231 if (!file)
1232 return -EBADF;
1233 ret = compat_writev(file, vec, vlen, &file->f_pos);
1234 fput_light(file, fput_needed);
1235 return ret;
1236 }
1237
1238 asmlinkage ssize_t
1239 compat_sys_pwritev64(unsigned long fd, const struct compat_iovec __user *vec,
1240 unsigned long vlen, loff_t pos)
1241 {
1242 struct file *file;
1243 int fput_needed;
1244 ssize_t ret;
1245
1246 if (pos < 0)
1247 return -EINVAL;
1248 file = fget_light(fd, &fput_needed);
1249 if (!file)
1250 return -EBADF;
1251 ret = -ESPIPE;
1252 if (file->f_mode & FMODE_PWRITE)
1253 ret = compat_writev(file, vec, vlen, &pos);
1254 fput_light(file, fput_needed);
1255 return ret;
1256 }
1257
1258 asmlinkage ssize_t
1259 compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1260 unsigned long vlen, u32 pos_low, u32 pos_high)
1261 {
1262 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1263 return compat_sys_pwritev64(fd, vec, vlen, pos);
1264 }
1265
1266 asmlinkage long
1267 compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1268 unsigned int nr_segs, unsigned int flags)
1269 {
1270 unsigned i;
1271 struct iovec __user *iov;
1272 if (nr_segs > UIO_MAXIOV)
1273 return -EINVAL;
1274 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1275 for (i = 0; i < nr_segs; i++) {
1276 struct compat_iovec v;
1277 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1278 get_user(v.iov_len, &iov32[i].iov_len) ||
1279 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1280 put_user(v.iov_len, &iov[i].iov_len))
1281 return -EFAULT;
1282 }
1283 return sys_vmsplice(fd, iov, nr_segs, flags);
1284 }
1285
1286 /*
1287 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1288 * O_LARGEFILE flag.
1289 */
1290 asmlinkage long
1291 compat_sys_open(const char __user *filename, int flags, umode_t mode)
1292 {
1293 return do_sys_open(AT_FDCWD, filename, flags, mode);
1294 }
1295
1296 /*
1297 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1298 * O_LARGEFILE flag.
1299 */
1300 asmlinkage long
1301 compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, umode_t mode)
1302 {
1303 return do_sys_open(dfd, filename, flags, mode);
1304 }
1305
1306 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1307
1308 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1309 int timeval, int ret)
1310 {
1311 struct timespec ts;
1312
1313 if (!p)
1314 return ret;
1315
1316 if (current->personality & STICKY_TIMEOUTS)
1317 goto sticky;
1318
1319 /* No update for zero timeout */
1320 if (!end_time->tv_sec && !end_time->tv_nsec)
1321 return ret;
1322
1323 ktime_get_ts(&ts);
1324 ts = timespec_sub(*end_time, ts);
1325 if (ts.tv_sec < 0)
1326 ts.tv_sec = ts.tv_nsec = 0;
1327
1328 if (timeval) {
1329 struct compat_timeval rtv;
1330
1331 rtv.tv_sec = ts.tv_sec;
1332 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1333
1334 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1335 return ret;
1336 } else {
1337 struct compat_timespec rts;
1338
1339 rts.tv_sec = ts.tv_sec;
1340 rts.tv_nsec = ts.tv_nsec;
1341
1342 if (!copy_to_user(p, &rts, sizeof(rts)))
1343 return ret;
1344 }
1345 /*
1346 * If an application puts its timeval in read-only memory, we
1347 * don't want the Linux-specific update to the timeval to
1348 * cause a fault after the select has completed
1349 * successfully. However, because we're not updating the
1350 * timeval, we can't restart the system call.
1351 */
1352
1353 sticky:
1354 if (ret == -ERESTARTNOHAND)
1355 ret = -EINTR;
1356 return ret;
1357 }
1358
1359 /*
1360 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1361 * 64-bit unsigned longs.
1362 */
1363 static
1364 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1365 unsigned long *fdset)
1366 {
1367 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1368 if (ufdset) {
1369 unsigned long odd;
1370
1371 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1372 return -EFAULT;
1373
1374 odd = nr & 1UL;
1375 nr &= ~1UL;
1376 while (nr) {
1377 unsigned long h, l;
1378 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1379 return -EFAULT;
1380 ufdset += 2;
1381 *fdset++ = h << 32 | l;
1382 nr -= 2;
1383 }
1384 if (odd && __get_user(*fdset, ufdset))
1385 return -EFAULT;
1386 } else {
1387 /* Tricky, must clear full unsigned long in the
1388 * kernel fdset at the end, this makes sure that
1389 * actually happens.
1390 */
1391 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1392 }
1393 return 0;
1394 }
1395
1396 static
1397 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1398 unsigned long *fdset)
1399 {
1400 unsigned long odd;
1401 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1402
1403 if (!ufdset)
1404 return 0;
1405
1406 odd = nr & 1UL;
1407 nr &= ~1UL;
1408 while (nr) {
1409 unsigned long h, l;
1410 l = *fdset++;
1411 h = l >> 32;
1412 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1413 return -EFAULT;
1414 ufdset += 2;
1415 nr -= 2;
1416 }
1417 if (odd && __put_user(*fdset, ufdset))
1418 return -EFAULT;
1419 return 0;
1420 }
1421
1422
1423 /*
1424 * This is a virtual copy of sys_select from fs/select.c and probably
1425 * should be compared to it from time to time
1426 */
1427
1428 /*
1429 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1430 * like to be certain this leads to no problems. So I return
1431 * EINTR just for safety.
1432 *
1433 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1434 * I'm trying ERESTARTNOHAND which restart only when you want to.
1435 */
1436 int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1437 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1438 struct timespec *end_time)
1439 {
1440 fd_set_bits fds;
1441 void *bits;
1442 int size, max_fds, ret = -EINVAL;
1443 struct fdtable *fdt;
1444 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1445
1446 if (n < 0)
1447 goto out_nofds;
1448
1449 /* max_fds can increase, so grab it once to avoid race */
1450 rcu_read_lock();
1451 fdt = files_fdtable(current->files);
1452 max_fds = fdt->max_fds;
1453 rcu_read_unlock();
1454 if (n > max_fds)
1455 n = max_fds;
1456
1457 /*
1458 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1459 * since we used fdset we need to allocate memory in units of
1460 * long-words.
1461 */
1462 size = FDS_BYTES(n);
1463 bits = stack_fds;
1464 if (size > sizeof(stack_fds) / 6) {
1465 bits = kmalloc(6 * size, GFP_KERNEL);
1466 ret = -ENOMEM;
1467 if (!bits)
1468 goto out_nofds;
1469 }
1470 fds.in = (unsigned long *) bits;
1471 fds.out = (unsigned long *) (bits + size);
1472 fds.ex = (unsigned long *) (bits + 2*size);
1473 fds.res_in = (unsigned long *) (bits + 3*size);
1474 fds.res_out = (unsigned long *) (bits + 4*size);
1475 fds.res_ex = (unsigned long *) (bits + 5*size);
1476
1477 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1478 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1479 (ret = compat_get_fd_set(n, exp, fds.ex)))
1480 goto out;
1481 zero_fd_set(n, fds.res_in);
1482 zero_fd_set(n, fds.res_out);
1483 zero_fd_set(n, fds.res_ex);
1484
1485 ret = do_select(n, &fds, end_time);
1486
1487 if (ret < 0)
1488 goto out;
1489 if (!ret) {
1490 ret = -ERESTARTNOHAND;
1491 if (signal_pending(current))
1492 goto out;
1493 ret = 0;
1494 }
1495
1496 if (compat_set_fd_set(n, inp, fds.res_in) ||
1497 compat_set_fd_set(n, outp, fds.res_out) ||
1498 compat_set_fd_set(n, exp, fds.res_ex))
1499 ret = -EFAULT;
1500 out:
1501 if (bits != stack_fds)
1502 kfree(bits);
1503 out_nofds:
1504 return ret;
1505 }
1506
1507 asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1508 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1509 struct compat_timeval __user *tvp)
1510 {
1511 struct timespec end_time, *to = NULL;
1512 struct compat_timeval tv;
1513 int ret;
1514
1515 if (tvp) {
1516 if (copy_from_user(&tv, tvp, sizeof(tv)))
1517 return -EFAULT;
1518
1519 to = &end_time;
1520 if (poll_select_set_timeout(to,
1521 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1522 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1523 return -EINVAL;
1524 }
1525
1526 ret = compat_core_sys_select(n, inp, outp, exp, to);
1527 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1528
1529 return ret;
1530 }
1531
1532 struct compat_sel_arg_struct {
1533 compat_ulong_t n;
1534 compat_uptr_t inp;
1535 compat_uptr_t outp;
1536 compat_uptr_t exp;
1537 compat_uptr_t tvp;
1538 };
1539
1540 asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1541 {
1542 struct compat_sel_arg_struct a;
1543
1544 if (copy_from_user(&a, arg, sizeof(a)))
1545 return -EFAULT;
1546 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1547 compat_ptr(a.exp), compat_ptr(a.tvp));
1548 }
1549
1550 #ifdef HAVE_SET_RESTORE_SIGMASK
1551 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1552 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1553 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1554 compat_size_t sigsetsize)
1555 {
1556 compat_sigset_t ss32;
1557 sigset_t ksigmask, sigsaved;
1558 struct compat_timespec ts;
1559 struct timespec end_time, *to = NULL;
1560 int ret;
1561
1562 if (tsp) {
1563 if (copy_from_user(&ts, tsp, sizeof(ts)))
1564 return -EFAULT;
1565
1566 to = &end_time;
1567 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1568 return -EINVAL;
1569 }
1570
1571 if (sigmask) {
1572 if (sigsetsize != sizeof(compat_sigset_t))
1573 return -EINVAL;
1574 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1575 return -EFAULT;
1576 sigset_from_compat(&ksigmask, &ss32);
1577
1578 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1579 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1580 }
1581
1582 ret = compat_core_sys_select(n, inp, outp, exp, to);
1583 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1584
1585 if (ret == -ERESTARTNOHAND) {
1586 /*
1587 * Don't restore the signal mask yet. Let do_signal() deliver
1588 * the signal on the way back to userspace, before the signal
1589 * mask is restored.
1590 */
1591 if (sigmask) {
1592 memcpy(&current->saved_sigmask, &sigsaved,
1593 sizeof(sigsaved));
1594 set_restore_sigmask();
1595 }
1596 } else if (sigmask)
1597 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1598
1599 return ret;
1600 }
1601
1602 asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1603 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1604 struct compat_timespec __user *tsp, void __user *sig)
1605 {
1606 compat_size_t sigsetsize = 0;
1607 compat_uptr_t up = 0;
1608
1609 if (sig) {
1610 if (!access_ok(VERIFY_READ, sig,
1611 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1612 __get_user(up, (compat_uptr_t __user *)sig) ||
1613 __get_user(sigsetsize,
1614 (compat_size_t __user *)(sig+sizeof(up))))
1615 return -EFAULT;
1616 }
1617 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1618 sigsetsize);
1619 }
1620
1621 asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1622 unsigned int nfds, struct compat_timespec __user *tsp,
1623 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1624 {
1625 compat_sigset_t ss32;
1626 sigset_t ksigmask, sigsaved;
1627 struct compat_timespec ts;
1628 struct timespec end_time, *to = NULL;
1629 int ret;
1630
1631 if (tsp) {
1632 if (copy_from_user(&ts, tsp, sizeof(ts)))
1633 return -EFAULT;
1634
1635 to = &end_time;
1636 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1637 return -EINVAL;
1638 }
1639
1640 if (sigmask) {
1641 if (sigsetsize != sizeof(compat_sigset_t))
1642 return -EINVAL;
1643 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1644 return -EFAULT;
1645 sigset_from_compat(&ksigmask, &ss32);
1646
1647 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1648 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1649 }
1650
1651 ret = do_sys_poll(ufds, nfds, to);
1652
1653 /* We can restart this syscall, usually */
1654 if (ret == -EINTR) {
1655 /*
1656 * Don't restore the signal mask yet. Let do_signal() deliver
1657 * the signal on the way back to userspace, before the signal
1658 * mask is restored.
1659 */
1660 if (sigmask) {
1661 memcpy(&current->saved_sigmask, &sigsaved,
1662 sizeof(sigsaved));
1663 set_restore_sigmask();
1664 }
1665 ret = -ERESTARTNOHAND;
1666 } else if (sigmask)
1667 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1668
1669 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1670
1671 return ret;
1672 }
1673 #endif /* HAVE_SET_RESTORE_SIGMASK */
1674
1675 #ifdef CONFIG_EPOLL
1676
1677 #ifdef HAVE_SET_RESTORE_SIGMASK
1678 asmlinkage long compat_sys_epoll_pwait(int epfd,
1679 struct compat_epoll_event __user *events,
1680 int maxevents, int timeout,
1681 const compat_sigset_t __user *sigmask,
1682 compat_size_t sigsetsize)
1683 {
1684 long err;
1685 compat_sigset_t csigmask;
1686 sigset_t ksigmask, sigsaved;
1687
1688 /*
1689 * If the caller wants a certain signal mask to be set during the wait,
1690 * we apply it here.
1691 */
1692 if (sigmask) {
1693 if (sigsetsize != sizeof(compat_sigset_t))
1694 return -EINVAL;
1695 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1696 return -EFAULT;
1697 sigset_from_compat(&ksigmask, &csigmask);
1698 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1699 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1700 }
1701
1702 err = sys_epoll_wait(epfd, events, maxevents, timeout);
1703
1704 /*
1705 * If we changed the signal mask, we need to restore the original one.
1706 * In case we've got a signal while waiting, we do not restore the
1707 * signal mask yet, and we allow do_signal() to deliver the signal on
1708 * the way back to userspace, before the signal mask is restored.
1709 */
1710 if (sigmask) {
1711 if (err == -EINTR) {
1712 memcpy(&current->saved_sigmask, &sigsaved,
1713 sizeof(sigsaved));
1714 set_restore_sigmask();
1715 } else
1716 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1717 }
1718
1719 return err;
1720 }
1721 #endif /* HAVE_SET_RESTORE_SIGMASK */
1722
1723 #endif /* CONFIG_EPOLL */
1724
1725 #ifdef CONFIG_SIGNALFD
1726
1727 asmlinkage long compat_sys_signalfd4(int ufd,
1728 const compat_sigset_t __user *sigmask,
1729 compat_size_t sigsetsize, int flags)
1730 {
1731 compat_sigset_t ss32;
1732 sigset_t tmp;
1733 sigset_t __user *ksigmask;
1734
1735 if (sigsetsize != sizeof(compat_sigset_t))
1736 return -EINVAL;
1737 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1738 return -EFAULT;
1739 sigset_from_compat(&tmp, &ss32);
1740 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1741 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1742 return -EFAULT;
1743
1744 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1745 }
1746
1747 asmlinkage long compat_sys_signalfd(int ufd,
1748 const compat_sigset_t __user *sigmask,
1749 compat_size_t sigsetsize)
1750 {
1751 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1752 }
1753 #endif /* CONFIG_SIGNALFD */
1754
1755 #ifdef CONFIG_TIMERFD
1756
1757 asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1758 const struct compat_itimerspec __user *utmr,
1759 struct compat_itimerspec __user *otmr)
1760 {
1761 int error;
1762 struct itimerspec t;
1763 struct itimerspec __user *ut;
1764
1765 if (get_compat_itimerspec(&t, utmr))
1766 return -EFAULT;
1767 ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1768 if (copy_to_user(&ut[0], &t, sizeof(t)))
1769 return -EFAULT;
1770 error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1771 if (!error && otmr)
1772 error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1773 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1774
1775 return error;
1776 }
1777
1778 asmlinkage long compat_sys_timerfd_gettime(int ufd,
1779 struct compat_itimerspec __user *otmr)
1780 {
1781 int error;
1782 struct itimerspec t;
1783 struct itimerspec __user *ut;
1784
1785 ut = compat_alloc_user_space(sizeof(struct itimerspec));
1786 error = sys_timerfd_gettime(ufd, ut);
1787 if (!error)
1788 error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1789 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1790
1791 return error;
1792 }
1793
1794 #endif /* CONFIG_TIMERFD */
1795
1796 #ifdef CONFIG_FHANDLE
1797 /*
1798 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1799 * doesn't set the O_LARGEFILE flag.
1800 */
1801 asmlinkage long
1802 compat_sys_open_by_handle_at(int mountdirfd,
1803 struct file_handle __user *handle, int flags)
1804 {
1805 return do_handle_open(mountdirfd, handle, flags);
1806 }
1807 #endif
Linux 2.6 libata-dev (mirror)