?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nfs41: implement DESTROY_CLIENTID operation
[linux-btrfs-devel.git] / fs / compat.c
blobf2b36d472c73a524758027c4329ccfa8d7e31e1b
1 /*
2 * linux/fs/compat.c
3 *
4 * Kernel compatibililty routines for e.g. 32 bit syscall support
5 * on 64 bit kernels.
6 *
7 * Copyright (C) 2002 Stephen Rothwell, IBM Corporation
8 * Copyright (C) 1997-2000 Jakub Jelinek (jakub@redhat.com)
9 * Copyright (C) 1998 Eddie C. Dost (ecd@skynet.be)
10 * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs
11 * Copyright (C) 2003 Pavel Machek (pavel@ucw.cz)
12 *
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License version 2 as
15 * published by the Free Software Foundation.
16 */
17
18 #include <linux/stddef.h>
19 #include <linux/kernel.h>
20 #include <linux/linkage.h>
21 #include <linux/compat.h>
22 #include <linux/errno.h>
23 #include <linux/time.h>
24 #include <linux/fs.h>
25 #include <linux/fcntl.h>
26 #include <linux/namei.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/vfs.h>
30 #include <linux/ioctl.h>
31 #include <linux/init.h>
32 #include <linux/ncp_mount.h>
33 #include <linux/nfs4_mount.h>
34 #include <linux/syscalls.h>
35 #include <linux/ctype.h>
36 #include <linux/module.h>
37 #include <linux/dirent.h>
38 #include <linux/fsnotify.h>
39 #include <linux/highuid.h>
40 #include <linux/personality.h>
41 #include <linux/rwsem.h>
42 #include <linux/tsacct_kern.h>
43 #include <linux/security.h>
44 #include <linux/highmem.h>
45 #include <linux/signal.h>
46 #include <linux/poll.h>
47 #include <linux/mm.h>
48 #include <linux/eventpoll.h>
49 #include <linux/fs_struct.h>
50 #include <linux/slab.h>
51 #include <linux/pagemap.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/ioctls.h>
56 #include "internal.h"
57
58 int compat_log = 1;
59
60 int compat_printk(const char *fmt, ...)
61 {
62 va_list ap;
63 int ret;
64 if (!compat_log)
65 return 0;
66 va_start(ap, fmt);
67 ret = vprintk(fmt, ap);
68 va_end(ap);
69 return ret;
70 }
71
72 #include "read_write.h"
73
74 /*
75 * Not all architectures have sys_utime, so implement this in terms
76 * of sys_utimes.
77 */
78 asmlinkage long compat_sys_utime(const char __user *filename,
79 struct compat_utimbuf __user *t)
80 {
81 struct timespec tv[2];
82
83 if (t) {
84 if (get_user(tv[0].tv_sec, &t->actime) ||
85 get_user(tv[1].tv_sec, &t->modtime))
86 return -EFAULT;
87 tv[0].tv_nsec = 0;
88 tv[1].tv_nsec = 0;
89 }
90 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
91 }
92
93 asmlinkage long compat_sys_utimensat(unsigned int dfd, const char __user *filename, struct compat_timespec __user *t, int flags)
94 {
95 struct timespec tv[2];
96
97 if (t) {
98 if (get_compat_timespec(&tv[0], &t[0]) ||
99 get_compat_timespec(&tv[1], &t[1]))
100 return -EFAULT;
101
102 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
103 return 0;
104 }
105 return do_utimes(dfd, filename, t ? tv : NULL, flags);
106 }
107
108 asmlinkage long compat_sys_futimesat(unsigned int dfd, const char __user *filename, struct compat_timeval __user *t)
109 {
110 struct timespec tv[2];
111
112 if (t) {
113 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
114 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
115 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
116 get_user(tv[1].tv_nsec, &t[1].tv_usec))
117 return -EFAULT;
118 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
119 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
120 return -EINVAL;
121 tv[0].tv_nsec *= 1000;
122 tv[1].tv_nsec *= 1000;
123 }
124 return do_utimes(dfd, filename, t ? tv : NULL, 0);
125 }
126
127 asmlinkage long compat_sys_utimes(const char __user *filename, struct compat_timeval __user *t)
128 {
129 return compat_sys_futimesat(AT_FDCWD, filename, t);
130 }
131
132 static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
133 {
134 compat_ino_t ino = stat->ino;
135 typeof(ubuf->st_uid) uid = 0;
136 typeof(ubuf->st_gid) gid = 0;
137 int err;
138
139 SET_UID(uid, stat->uid);
140 SET_GID(gid, stat->gid);
141
142 if ((u64) stat->size > MAX_NON_LFS ||
143 !old_valid_dev(stat->dev) ||
144 !old_valid_dev(stat->rdev))
145 return -EOVERFLOW;
146 if (sizeof(ino) < sizeof(stat->ino) && ino != stat->ino)
147 return -EOVERFLOW;
148
149 if (clear_user(ubuf, sizeof(*ubuf)))
150 return -EFAULT;
151
152 err = __put_user(old_encode_dev(stat->dev), &ubuf->st_dev);
153 err |= __put_user(ino, &ubuf->st_ino);
154 err |= __put_user(stat->mode, &ubuf->st_mode);
155 err |= __put_user(stat->nlink, &ubuf->st_nlink);
156 err |= __put_user(uid, &ubuf->st_uid);
157 err |= __put_user(gid, &ubuf->st_gid);
158 err |= __put_user(old_encode_dev(stat->rdev), &ubuf->st_rdev);
159 err |= __put_user(stat->size, &ubuf->st_size);
160 err |= __put_user(stat->atime.tv_sec, &ubuf->st_atime);
161 err |= __put_user(stat->atime.tv_nsec, &ubuf->st_atime_nsec);
162 err |= __put_user(stat->mtime.tv_sec, &ubuf->st_mtime);
163 err |= __put_user(stat->mtime.tv_nsec, &ubuf->st_mtime_nsec);
164 err |= __put_user(stat->ctime.tv_sec, &ubuf->st_ctime);
165 err |= __put_user(stat->ctime.tv_nsec, &ubuf->st_ctime_nsec);
166 err |= __put_user(stat->blksize, &ubuf->st_blksize);
167 err |= __put_user(stat->blocks, &ubuf->st_blocks);
168 return err;
169 }
170
171 asmlinkage long compat_sys_newstat(const char __user * filename,
172 struct compat_stat __user *statbuf)
173 {
174 struct kstat stat;
175 int error;
176
177 error = vfs_stat(filename, &stat);
178 if (error)
179 return error;
180 return cp_compat_stat(&stat, statbuf);
181 }
182
183 asmlinkage long compat_sys_newlstat(const char __user * filename,
184 struct compat_stat __user *statbuf)
185 {
186 struct kstat stat;
187 int error;
188
189 error = vfs_lstat(filename, &stat);
190 if (error)
191 return error;
192 return cp_compat_stat(&stat, statbuf);
193 }
194
195 #ifndef __ARCH_WANT_STAT64
196 asmlinkage long compat_sys_newfstatat(unsigned int dfd,
197 const char __user *filename,
198 struct compat_stat __user *statbuf, int flag)
199 {
200 struct kstat stat;
201 int error;
202
203 error = vfs_fstatat(dfd, filename, &stat, flag);
204 if (error)
205 return error;
206 return cp_compat_stat(&stat, statbuf);
207 }
208 #endif
209
210 asmlinkage long compat_sys_newfstat(unsigned int fd,
211 struct compat_stat __user * statbuf)
212 {
213 struct kstat stat;
214 int error = vfs_fstat(fd, &stat);
215
216 if (!error)
217 error = cp_compat_stat(&stat, statbuf);
218 return error;
219 }
220
221 static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
222 {
223
224 if (sizeof ubuf->f_blocks == 4) {
225 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
226 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
227 return -EOVERFLOW;
228 /* f_files and f_ffree may be -1; it's okay
229 * to stuff that into 32 bits */
230 if (kbuf->f_files != 0xffffffffffffffffULL
231 && (kbuf->f_files & 0xffffffff00000000ULL))
232 return -EOVERFLOW;
233 if (kbuf->f_ffree != 0xffffffffffffffffULL
234 && (kbuf->f_ffree & 0xffffffff00000000ULL))
235 return -EOVERFLOW;
236 }
237 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
238 __put_user(kbuf->f_type, &ubuf->f_type) ||
239 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
240 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
241 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
242 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
243 __put_user(kbuf->f_files, &ubuf->f_files) ||
244 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
245 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
246 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
247 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
248 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
249 __put_user(0, &ubuf->f_spare[0]) ||
250 __put_user(0, &ubuf->f_spare[1]) ||
251 __put_user(0, &ubuf->f_spare[2]) ||
252 __put_user(0, &ubuf->f_spare[3]) ||
253 __put_user(0, &ubuf->f_spare[4]))
254 return -EFAULT;
255 return 0;
256 }
257
258 /*
259 * The following statfs calls are copies of code from fs/statfs.c and
260 * should be checked against those from time to time
261 */
262 asmlinkage long compat_sys_statfs(const char __user *pathname, struct compat_statfs __user *buf)
263 {
264 struct kstatfs tmp;
265 int error = user_statfs(pathname, &tmp);
266 if (!error)
267 error = put_compat_statfs(buf, &tmp);
268 return error;
269 }
270
271 asmlinkage long compat_sys_fstatfs(unsigned int fd, struct compat_statfs __user *buf)
272 {
273 struct kstatfs tmp;
274 int error = fd_statfs(fd, &tmp);
275 if (!error)
276 error = put_compat_statfs(buf, &tmp);
277 return error;
278 }
279
280 static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
281 {
282 if (sizeof ubuf->f_blocks == 4) {
283 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
284 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
285 return -EOVERFLOW;
286 /* f_files and f_ffree may be -1; it's okay
287 * to stuff that into 32 bits */
288 if (kbuf->f_files != 0xffffffffffffffffULL
289 && (kbuf->f_files & 0xffffffff00000000ULL))
290 return -EOVERFLOW;
291 if (kbuf->f_ffree != 0xffffffffffffffffULL
292 && (kbuf->f_ffree & 0xffffffff00000000ULL))
293 return -EOVERFLOW;
294 }
295 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
296 __put_user(kbuf->f_type, &ubuf->f_type) ||
297 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
298 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
299 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
300 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
301 __put_user(kbuf->f_files, &ubuf->f_files) ||
302 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
303 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
304 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
305 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
306 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
307 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
308 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
309 return -EFAULT;
310 return 0;
311 }
312
313 asmlinkage long compat_sys_statfs64(const char __user *pathname, compat_size_t sz, struct compat_statfs64 __user *buf)
314 {
315 struct kstatfs tmp;
316 int error;
317
318 if (sz != sizeof(*buf))
319 return -EINVAL;
320
321 error = user_statfs(pathname, &tmp);
322 if (!error)
323 error = put_compat_statfs64(buf, &tmp);
324 return error;
325 }
326
327 asmlinkage long compat_sys_fstatfs64(unsigned int fd, compat_size_t sz, struct compat_statfs64 __user *buf)
328 {
329 struct kstatfs tmp;
330 int error;
331
332 if (sz != sizeof(*buf))
333 return -EINVAL;
334
335 error = fd_statfs(fd, &tmp);
336 if (!error)
337 error = put_compat_statfs64(buf, &tmp);
338 return error;
339 }
340
341 /*
342 * This is a copy of sys_ustat, just dealing with a structure layout.
343 * Given how simple this syscall is that apporach is more maintainable
344 * than the various conversion hacks.
345 */
346 asmlinkage long compat_sys_ustat(unsigned dev, struct compat_ustat __user *u)
347 {
348 struct super_block *sb;
349 struct compat_ustat tmp;
350 struct kstatfs sbuf;
351 int err;
352
353 sb = user_get_super(new_decode_dev(dev));
354 if (!sb)
355 return -EINVAL;
356 err = statfs_by_dentry(sb->s_root, &sbuf);
357 drop_super(sb);
358 if (err)
359 return err;
360
361 memset(&tmp, 0, sizeof(struct compat_ustat));
362 tmp.f_tfree = sbuf.f_bfree;
363 tmp.f_tinode = sbuf.f_ffree;
364 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
365 return -EFAULT;
366 return 0;
367 }
368
369 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
370 {
371 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
372 __get_user(kfl->l_type, &ufl->l_type) ||
373 __get_user(kfl->l_whence, &ufl->l_whence) ||
374 __get_user(kfl->l_start, &ufl->l_start) ||
375 __get_user(kfl->l_len, &ufl->l_len) ||
376 __get_user(kfl->l_pid, &ufl->l_pid))
377 return -EFAULT;
378 return 0;
379 }
380
381 static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
382 {
383 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
384 __put_user(kfl->l_type, &ufl->l_type) ||
385 __put_user(kfl->l_whence, &ufl->l_whence) ||
386 __put_user(kfl->l_start, &ufl->l_start) ||
387 __put_user(kfl->l_len, &ufl->l_len) ||
388 __put_user(kfl->l_pid, &ufl->l_pid))
389 return -EFAULT;
390 return 0;
391 }
392
393 #ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
394 static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
395 {
396 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
397 __get_user(kfl->l_type, &ufl->l_type) ||
398 __get_user(kfl->l_whence, &ufl->l_whence) ||
399 __get_user(kfl->l_start, &ufl->l_start) ||
400 __get_user(kfl->l_len, &ufl->l_len) ||
401 __get_user(kfl->l_pid, &ufl->l_pid))
402 return -EFAULT;
403 return 0;
404 }
405 #endif
406
407 #ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
408 static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
409 {
410 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
411 __put_user(kfl->l_type, &ufl->l_type) ||
412 __put_user(kfl->l_whence, &ufl->l_whence) ||
413 __put_user(kfl->l_start, &ufl->l_start) ||
414 __put_user(kfl->l_len, &ufl->l_len) ||
415 __put_user(kfl->l_pid, &ufl->l_pid))
416 return -EFAULT;
417 return 0;
418 }
419 #endif
420
421 asmlinkage long compat_sys_fcntl64(unsigned int fd, unsigned int cmd,
422 unsigned long arg)
423 {
424 mm_segment_t old_fs;
425 struct flock f;
426 long ret;
427
428 switch (cmd) {
429 case F_GETLK:
430 case F_SETLK:
431 case F_SETLKW:
432 ret = get_compat_flock(&f, compat_ptr(arg));
433 if (ret != 0)
434 break;
435 old_fs = get_fs();
436 set_fs(KERNEL_DS);
437 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
438 set_fs(old_fs);
439 if (cmd == F_GETLK && ret == 0) {
440 /* GETLK was successful and we need to return the data...
441 * but it needs to fit in the compat structure.
442 * l_start shouldn't be too big, unless the original
443 * start + end is greater than COMPAT_OFF_T_MAX, in which
444 * case the app was asking for trouble, so we return
445 * -EOVERFLOW in that case.
446 * l_len could be too big, in which case we just truncate it,
447 * and only allow the app to see that part of the conflicting
448 * lock that might make sense to it anyway
449 */
450
451 if (f.l_start > COMPAT_OFF_T_MAX)
452 ret = -EOVERFLOW;
453 if (f.l_len > COMPAT_OFF_T_MAX)
454 f.l_len = COMPAT_OFF_T_MAX;
455 if (ret == 0)
456 ret = put_compat_flock(&f, compat_ptr(arg));
457 }
458 break;
459
460 case F_GETLK64:
461 case F_SETLK64:
462 case F_SETLKW64:
463 ret = get_compat_flock64(&f, compat_ptr(arg));
464 if (ret != 0)
465 break;
466 old_fs = get_fs();
467 set_fs(KERNEL_DS);
468 ret = sys_fcntl(fd, (cmd == F_GETLK64) ? F_GETLK :
469 ((cmd == F_SETLK64) ? F_SETLK : F_SETLKW),
470 (unsigned long)&f);
471 set_fs(old_fs);
472 if (cmd == F_GETLK64 && ret == 0) {
473 /* need to return lock information - see above for commentary */
474 if (f.l_start > COMPAT_LOFF_T_MAX)
475 ret = -EOVERFLOW;
476 if (f.l_len > COMPAT_LOFF_T_MAX)
477 f.l_len = COMPAT_LOFF_T_MAX;
478 if (ret == 0)
479 ret = put_compat_flock64(&f, compat_ptr(arg));
480 }
481 break;
482
483 default:
484 ret = sys_fcntl(fd, cmd, arg);
485 break;
486 }
487 return ret;
488 }
489
490 asmlinkage long compat_sys_fcntl(unsigned int fd, unsigned int cmd,
491 unsigned long arg)
492 {
493 if ((cmd == F_GETLK64) || (cmd == F_SETLK64) || (cmd == F_SETLKW64))
494 return -EINVAL;
495 return compat_sys_fcntl64(fd, cmd, arg);
496 }
497
498 asmlinkage long
499 compat_sys_io_setup(unsigned nr_reqs, u32 __user *ctx32p)
500 {
501 long ret;
502 aio_context_t ctx64;
503
504 mm_segment_t oldfs = get_fs();
505 if (unlikely(get_user(ctx64, ctx32p)))
506 return -EFAULT;
507
508 set_fs(KERNEL_DS);
509 /* The __user pointer cast is valid because of the set_fs() */
510 ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
511 set_fs(oldfs);
512 /* truncating is ok because it's a user address */
513 if (!ret)
514 ret = put_user((u32) ctx64, ctx32p);
515 return ret;
516 }
517
518 asmlinkage long
519 compat_sys_io_getevents(aio_context_t ctx_id,
520 unsigned long min_nr,
521 unsigned long nr,
522 struct io_event __user *events,
523 struct compat_timespec __user *timeout)
524 {
525 long ret;
526 struct timespec t;
527 struct timespec __user *ut = NULL;
528
529 ret = -EFAULT;
530 if (unlikely(!access_ok(VERIFY_WRITE, events,
531 nr * sizeof(struct io_event))))
532 goto out;
533 if (timeout) {
534 if (get_compat_timespec(&t, timeout))
535 goto out;
536
537 ut = compat_alloc_user_space(sizeof(*ut));
538 if (copy_to_user(ut, &t, sizeof(t)) )
539 goto out;
540 }
541 ret = sys_io_getevents(ctx_id, min_nr, nr, events, ut);
542 out:
543 return ret;
544 }
545
546 /* A write operation does a read from user space and vice versa */
547 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
548
549 ssize_t compat_rw_copy_check_uvector(int type,
550 const struct compat_iovec __user *uvector, unsigned long nr_segs,
551 unsigned long fast_segs, struct iovec *fast_pointer,
552 struct iovec **ret_pointer)
553 {
554 compat_ssize_t tot_len;
555 struct iovec *iov = *ret_pointer = fast_pointer;
556 ssize_t ret = 0;
557 int seg;
558
559 /*
560 * SuS says "The readv() function *may* fail if the iovcnt argument
561 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
562 * traditionally returned zero for zero segments, so...
563 */
564 if (nr_segs == 0)
565 goto out;
566
567 ret = -EINVAL;
568 if (nr_segs > UIO_MAXIOV || nr_segs < 0)
569 goto out;
570 if (nr_segs > fast_segs) {
571 ret = -ENOMEM;
572 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
573 if (iov == NULL)
574 goto out;
575 }
576 *ret_pointer = iov;
577
578 /*
579 * Single unix specification:
580 * We should -EINVAL if an element length is not >= 0 and fitting an
581 * ssize_t.
582 *
583 * In Linux, the total length is limited to MAX_RW_COUNT, there is
584 * no overflow possibility.
585 */
586 tot_len = 0;
587 ret = -EINVAL;
588 for (seg = 0; seg < nr_segs; seg++) {
589 compat_uptr_t buf;
590 compat_ssize_t len;
591
592 if (__get_user(len, &uvector->iov_len) ||
593 __get_user(buf, &uvector->iov_base)) {
594 ret = -EFAULT;
595 goto out;
596 }
597 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
598 goto out;
599 if (!access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
600 ret = -EFAULT;
601 goto out;
602 }
603 if (len > MAX_RW_COUNT - tot_len)
604 len = MAX_RW_COUNT - tot_len;
605 tot_len += len;
606 iov->iov_base = compat_ptr(buf);
607 iov->iov_len = (compat_size_t) len;
608 uvector++;
609 iov++;
610 }
611 ret = tot_len;
612
613 out:
614 return ret;
615 }
616
617 static inline long
618 copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
619 {
620 compat_uptr_t uptr;
621 int i;
622
623 for (i = 0; i < nr; ++i) {
624 if (get_user(uptr, ptr32 + i))
625 return -EFAULT;
626 if (put_user(compat_ptr(uptr), ptr64 + i))
627 return -EFAULT;
628 }
629 return 0;
630 }
631
632 #define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *))
633
634 asmlinkage long
635 compat_sys_io_submit(aio_context_t ctx_id, int nr, u32 __user *iocb)
636 {
637 struct iocb __user * __user *iocb64;
638 long ret;
639
640 if (unlikely(nr < 0))
641 return -EINVAL;
642
643 if (nr > MAX_AIO_SUBMITS)
644 nr = MAX_AIO_SUBMITS;
645
646 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
647 ret = copy_iocb(nr, iocb, iocb64);
648 if (!ret)
649 ret = do_io_submit(ctx_id, nr, iocb64, 1);
650 return ret;
651 }
652
653 struct compat_ncp_mount_data {
654 compat_int_t version;
655 compat_uint_t ncp_fd;
656 __compat_uid_t mounted_uid;
657 compat_pid_t wdog_pid;
658 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
659 compat_uint_t time_out;
660 compat_uint_t retry_count;
661 compat_uint_t flags;
662 __compat_uid_t uid;
663 __compat_gid_t gid;
664 compat_mode_t file_mode;
665 compat_mode_t dir_mode;
666 };
667
668 struct compat_ncp_mount_data_v4 {
669 compat_int_t version;
670 compat_ulong_t flags;
671 compat_ulong_t mounted_uid;
672 compat_long_t wdog_pid;
673 compat_uint_t ncp_fd;
674 compat_uint_t time_out;
675 compat_uint_t retry_count;
676 compat_ulong_t uid;
677 compat_ulong_t gid;
678 compat_ulong_t file_mode;
679 compat_ulong_t dir_mode;
680 };
681
682 static void *do_ncp_super_data_conv(void *raw_data)
683 {
684 int version = *(unsigned int *)raw_data;
685
686 if (version == 3) {
687 struct compat_ncp_mount_data *c_n = raw_data;
688 struct ncp_mount_data *n = raw_data;
689
690 n->dir_mode = c_n->dir_mode;
691 n->file_mode = c_n->file_mode;
692 n->gid = c_n->gid;
693 n->uid = c_n->uid;
694 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
695 n->wdog_pid = c_n->wdog_pid;
696 n->mounted_uid = c_n->mounted_uid;
697 } else if (version == 4) {
698 struct compat_ncp_mount_data_v4 *c_n = raw_data;
699 struct ncp_mount_data_v4 *n = raw_data;
700
701 n->dir_mode = c_n->dir_mode;
702 n->file_mode = c_n->file_mode;
703 n->gid = c_n->gid;
704 n->uid = c_n->uid;
705 n->retry_count = c_n->retry_count;
706 n->time_out = c_n->time_out;
707 n->ncp_fd = c_n->ncp_fd;
708 n->wdog_pid = c_n->wdog_pid;
709 n->mounted_uid = c_n->mounted_uid;
710 n->flags = c_n->flags;
711 } else if (version != 5) {
712 return NULL;
713 }
714
715 return raw_data;
716 }
717
718
719 struct compat_nfs_string {
720 compat_uint_t len;
721 compat_uptr_t data;
722 };
723
724 static inline void compat_nfs_string(struct nfs_string *dst,
725 struct compat_nfs_string *src)
726 {
727 dst->data = compat_ptr(src->data);
728 dst->len = src->len;
729 }
730
731 struct compat_nfs4_mount_data_v1 {
732 compat_int_t version;
733 compat_int_t flags;
734 compat_int_t rsize;
735 compat_int_t wsize;
736 compat_int_t timeo;
737 compat_int_t retrans;
738 compat_int_t acregmin;
739 compat_int_t acregmax;
740 compat_int_t acdirmin;
741 compat_int_t acdirmax;
742 struct compat_nfs_string client_addr;
743 struct compat_nfs_string mnt_path;
744 struct compat_nfs_string hostname;
745 compat_uint_t host_addrlen;
746 compat_uptr_t host_addr;
747 compat_int_t proto;
748 compat_int_t auth_flavourlen;
749 compat_uptr_t auth_flavours;
750 };
751
752 static int do_nfs4_super_data_conv(void *raw_data)
753 {
754 int version = *(compat_uint_t *) raw_data;
755
756 if (version == 1) {
757 struct compat_nfs4_mount_data_v1 *raw = raw_data;
758 struct nfs4_mount_data *real = raw_data;
759
760 /* copy the fields backwards */
761 real->auth_flavours = compat_ptr(raw->auth_flavours);
762 real->auth_flavourlen = raw->auth_flavourlen;
763 real->proto = raw->proto;
764 real->host_addr = compat_ptr(raw->host_addr);
765 real->host_addrlen = raw->host_addrlen;
766 compat_nfs_string(&real->hostname, &raw->hostname);
767 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
768 compat_nfs_string(&real->client_addr, &raw->client_addr);
769 real->acdirmax = raw->acdirmax;
770 real->acdirmin = raw->acdirmin;
771 real->acregmax = raw->acregmax;
772 real->acregmin = raw->acregmin;
773 real->retrans = raw->retrans;
774 real->timeo = raw->timeo;
775 real->wsize = raw->wsize;
776 real->rsize = raw->rsize;
777 real->flags = raw->flags;
778 real->version = raw->version;
779 }
780
781 return 0;
782 }
783
784 #define NCPFS_NAME "ncpfs"
785 #define NFS4_NAME "nfs4"
786
787 asmlinkage long compat_sys_mount(const char __user * dev_name,
788 const char __user * dir_name,
789 const char __user * type, unsigned long flags,
790 const void __user * data)
791 {
792 char *kernel_type;
793 unsigned long data_page;
794 char *kernel_dev;
795 char *dir_page;
796 int retval;
797
798 retval = copy_mount_string(type, &kernel_type);
799 if (retval < 0)
800 goto out;
801
802 dir_page = getname(dir_name);
803 retval = PTR_ERR(dir_page);
804 if (IS_ERR(dir_page))
805 goto out1;
806
807 retval = copy_mount_string(dev_name, &kernel_dev);
808 if (retval < 0)
809 goto out2;
810
811 retval = copy_mount_options(data, &data_page);
812 if (retval < 0)
813 goto out3;
814
815 retval = -EINVAL;
816
817 if (kernel_type && data_page) {
818 if (!strcmp(kernel_type, NCPFS_NAME)) {
819 do_ncp_super_data_conv((void *)data_page);
820 } else if (!strcmp(kernel_type, NFS4_NAME)) {
821 if (do_nfs4_super_data_conv((void *) data_page))
822 goto out4;
823 }
824 }
825
826 retval = do_mount(kernel_dev, dir_page, kernel_type,
827 flags, (void*)data_page);
828
829 out4:
830 free_page(data_page);
831 out3:
832 kfree(kernel_dev);
833 out2:
834 putname(dir_page);
835 out1:
836 kfree(kernel_type);
837 out:
838 return retval;
839 }
840
841 struct compat_old_linux_dirent {
842 compat_ulong_t d_ino;
843 compat_ulong_t d_offset;
844 unsigned short d_namlen;
845 char d_name[1];
846 };
847
848 struct compat_readdir_callback {
849 struct compat_old_linux_dirent __user *dirent;
850 int result;
851 };
852
853 static int compat_fillonedir(void *__buf, const char *name, int namlen,
854 loff_t offset, u64 ino, unsigned int d_type)
855 {
856 struct compat_readdir_callback *buf = __buf;
857 struct compat_old_linux_dirent __user *dirent;
858 compat_ulong_t d_ino;
859
860 if (buf->result)
861 return -EINVAL;
862 d_ino = ino;
863 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
864 buf->result = -EOVERFLOW;
865 return -EOVERFLOW;
866 }
867 buf->result++;
868 dirent = buf->dirent;
869 if (!access_ok(VERIFY_WRITE, dirent,
870 (unsigned long)(dirent->d_name + namlen + 1) -
871 (unsigned long)dirent))
872 goto efault;
873 if ( __put_user(d_ino, &dirent->d_ino) ||
874 __put_user(offset, &dirent->d_offset) ||
875 __put_user(namlen, &dirent->d_namlen) ||
876 __copy_to_user(dirent->d_name, name, namlen) ||
877 __put_user(0, dirent->d_name + namlen))
878 goto efault;
879 return 0;
880 efault:
881 buf->result = -EFAULT;
882 return -EFAULT;
883 }
884
885 asmlinkage long compat_sys_old_readdir(unsigned int fd,
886 struct compat_old_linux_dirent __user *dirent, unsigned int count)
887 {
888 int error;
889 struct file *file;
890 struct compat_readdir_callback buf;
891
892 error = -EBADF;
893 file = fget(fd);
894 if (!file)
895 goto out;
896
897 buf.result = 0;
898 buf.dirent = dirent;
899
900 error = vfs_readdir(file, compat_fillonedir, &buf);
901 if (buf.result)
902 error = buf.result;
903
904 fput(file);
905 out:
906 return error;
907 }
908
909 struct compat_linux_dirent {
910 compat_ulong_t d_ino;
911 compat_ulong_t d_off;
912 unsigned short d_reclen;
913 char d_name[1];
914 };
915
916 struct compat_getdents_callback {
917 struct compat_linux_dirent __user *current_dir;
918 struct compat_linux_dirent __user *previous;
919 int count;
920 int error;
921 };
922
923 static int compat_filldir(void *__buf, const char *name, int namlen,
924 loff_t offset, u64 ino, unsigned int d_type)
925 {
926 struct compat_linux_dirent __user * dirent;
927 struct compat_getdents_callback *buf = __buf;
928 compat_ulong_t d_ino;
929 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
930 namlen + 2, sizeof(compat_long_t));
931
932 buf->error = -EINVAL; /* only used if we fail.. */
933 if (reclen > buf->count)
934 return -EINVAL;
935 d_ino = ino;
936 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
937 buf->error = -EOVERFLOW;
938 return -EOVERFLOW;
939 }
940 dirent = buf->previous;
941 if (dirent) {
942 if (__put_user(offset, &dirent->d_off))
943 goto efault;
944 }
945 dirent = buf->current_dir;
946 if (__put_user(d_ino, &dirent->d_ino))
947 goto efault;
948 if (__put_user(reclen, &dirent->d_reclen))
949 goto efault;
950 if (copy_to_user(dirent->d_name, name, namlen))
951 goto efault;
952 if (__put_user(0, dirent->d_name + namlen))
953 goto efault;
954 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
955 goto efault;
956 buf->previous = dirent;
957 dirent = (void __user *)dirent + reclen;
958 buf->current_dir = dirent;
959 buf->count -= reclen;
960 return 0;
961 efault:
962 buf->error = -EFAULT;
963 return -EFAULT;
964 }
965
966 asmlinkage long compat_sys_getdents(unsigned int fd,
967 struct compat_linux_dirent __user *dirent, unsigned int count)
968 {
969 struct file * file;
970 struct compat_linux_dirent __user * lastdirent;
971 struct compat_getdents_callback buf;
972 int error;
973
974 error = -EFAULT;
975 if (!access_ok(VERIFY_WRITE, dirent, count))
976 goto out;
977
978 error = -EBADF;
979 file = fget(fd);
980 if (!file)
981 goto out;
982
983 buf.current_dir = dirent;
984 buf.previous = NULL;
985 buf.count = count;
986 buf.error = 0;
987
988 error = vfs_readdir(file, compat_filldir, &buf);
989 if (error >= 0)
990 error = buf.error;
991 lastdirent = buf.previous;
992 if (lastdirent) {
993 if (put_user(file->f_pos, &lastdirent->d_off))
994 error = -EFAULT;
995 else
996 error = count - buf.count;
997 }
998 fput(file);
999 out:
1000 return error;
1001 }
1002
1003 #ifndef __ARCH_OMIT_COMPAT_SYS_GETDENTS64
1004
1005 struct compat_getdents_callback64 {
1006 struct linux_dirent64 __user *current_dir;
1007 struct linux_dirent64 __user *previous;
1008 int count;
1009 int error;
1010 };
1011
1012 static int compat_filldir64(void * __buf, const char * name, int namlen, loff_t offset,
1013 u64 ino, unsigned int d_type)
1014 {
1015 struct linux_dirent64 __user *dirent;
1016 struct compat_getdents_callback64 *buf = __buf;
1017 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1018 sizeof(u64));
1019 u64 off;
1020
1021 buf->error = -EINVAL; /* only used if we fail.. */
1022 if (reclen > buf->count)
1023 return -EINVAL;
1024 dirent = buf->previous;
1025
1026 if (dirent) {
1027 if (__put_user_unaligned(offset, &dirent->d_off))
1028 goto efault;
1029 }
1030 dirent = buf->current_dir;
1031 if (__put_user_unaligned(ino, &dirent->d_ino))
1032 goto efault;
1033 off = 0;
1034 if (__put_user_unaligned(off, &dirent->d_off))
1035 goto efault;
1036 if (__put_user(reclen, &dirent->d_reclen))
1037 goto efault;
1038 if (__put_user(d_type, &dirent->d_type))
1039 goto efault;
1040 if (copy_to_user(dirent->d_name, name, namlen))
1041 goto efault;
1042 if (__put_user(0, dirent->d_name + namlen))
1043 goto efault;
1044 buf->previous = dirent;
1045 dirent = (void __user *)dirent + reclen;
1046 buf->current_dir = dirent;
1047 buf->count -= reclen;
1048 return 0;
1049 efault:
1050 buf->error = -EFAULT;
1051 return -EFAULT;
1052 }
1053
1054 asmlinkage long compat_sys_getdents64(unsigned int fd,
1055 struct linux_dirent64 __user * dirent, unsigned int count)
1056 {
1057 struct file * file;
1058 struct linux_dirent64 __user * lastdirent;
1059 struct compat_getdents_callback64 buf;
1060 int error;
1061
1062 error = -EFAULT;
1063 if (!access_ok(VERIFY_WRITE, dirent, count))
1064 goto out;
1065
1066 error = -EBADF;
1067 file = fget(fd);
1068 if (!file)
1069 goto out;
1070
1071 buf.current_dir = dirent;
1072 buf.previous = NULL;
1073 buf.count = count;
1074 buf.error = 0;
1075
1076 error = vfs_readdir(file, compat_filldir64, &buf);
1077 if (error >= 0)
1078 error = buf.error;
1079 lastdirent = buf.previous;
1080 if (lastdirent) {
1081 typeof(lastdirent->d_off) d_off = file->f_pos;
1082 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1083 error = -EFAULT;
1084 else
1085 error = count - buf.count;
1086 }
1087 fput(file);
1088 out:
1089 return error;
1090 }
1091 #endif /* ! __ARCH_OMIT_COMPAT_SYS_GETDENTS64 */
1092
1093 static ssize_t compat_do_readv_writev(int type, struct file *file,
1094 const struct compat_iovec __user *uvector,
1095 unsigned long nr_segs, loff_t *pos)
1096 {
1097 compat_ssize_t tot_len;
1098 struct iovec iovstack[UIO_FASTIOV];
1099 struct iovec *iov = iovstack;
1100 ssize_t ret;
1101 io_fn_t fn;
1102 iov_fn_t fnv;
1103
1104 ret = -EINVAL;
1105 if (!file->f_op)
1106 goto out;
1107
1108 ret = -EFAULT;
1109 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
1110 goto out;
1111
1112 tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1113 UIO_FASTIOV, iovstack, &iov);
1114 if (tot_len == 0) {
1115 ret = 0;
1116 goto out;
1117 }
1118
1119 ret = rw_verify_area(type, file, pos, tot_len);
1120 if (ret < 0)
1121 goto out;
1122
1123 fnv = NULL;
1124 if (type == READ) {
1125 fn = file->f_op->read;
1126 fnv = file->f_op->aio_read;
1127 } else {
1128 fn = (io_fn_t)file->f_op->write;
1129 fnv = file->f_op->aio_write;
1130 }
1131
1132 if (fnv)
1133 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1134 pos, fnv);
1135 else
1136 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1137
1138 out:
1139 if (iov != iovstack)
1140 kfree(iov);
1141 if ((ret + (type == READ)) > 0) {
1142 if (type == READ)
1143 fsnotify_access(file);
1144 else
1145 fsnotify_modify(file);
1146 }
1147 return ret;
1148 }
1149
1150 static size_t compat_readv(struct file *file,
1151 const struct compat_iovec __user *vec,
1152 unsigned long vlen, loff_t *pos)
1153 {
1154 ssize_t ret = -EBADF;
1155
1156 if (!(file->f_mode & FMODE_READ))
1157 goto out;
1158
1159 ret = -EINVAL;
1160 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
1161 goto out;
1162
1163 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1164
1165 out:
1166 if (ret > 0)
1167 add_rchar(current, ret);
1168 inc_syscr(current);
1169 return ret;
1170 }
1171
1172 asmlinkage ssize_t
1173 compat_sys_readv(unsigned long fd, const struct compat_iovec __user *vec,
1174 unsigned long vlen)
1175 {
1176 struct file *file;
1177 int fput_needed;
1178 ssize_t ret;
1179
1180 file = fget_light(fd, &fput_needed);
1181 if (!file)
1182 return -EBADF;
1183 ret = compat_readv(file, vec, vlen, &file->f_pos);
1184 fput_light(file, fput_needed);
1185 return ret;
1186 }
1187
1188 asmlinkage ssize_t
1189 compat_sys_preadv(unsigned long fd, const struct compat_iovec __user *vec,
1190 unsigned long vlen, u32 pos_low, u32 pos_high)
1191 {
1192 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1193 struct file *file;
1194 int fput_needed;
1195 ssize_t ret;
1196
1197 if (pos < 0)
1198 return -EINVAL;
1199 file = fget_light(fd, &fput_needed);
1200 if (!file)
1201 return -EBADF;
1202 ret = -ESPIPE;
1203 if (file->f_mode & FMODE_PREAD)
1204 ret = compat_readv(file, vec, vlen, &pos);
1205 fput_light(file, fput_needed);
1206 return ret;
1207 }
1208
1209 static size_t compat_writev(struct file *file,
1210 const struct compat_iovec __user *vec,
1211 unsigned long vlen, loff_t *pos)
1212 {
1213 ssize_t ret = -EBADF;
1214
1215 if (!(file->f_mode & FMODE_WRITE))
1216 goto out;
1217
1218 ret = -EINVAL;
1219 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
1220 goto out;
1221
1222 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1223
1224 out:
1225 if (ret > 0)
1226 add_wchar(current, ret);
1227 inc_syscw(current);
1228 return ret;
1229 }
1230
1231 asmlinkage ssize_t
1232 compat_sys_writev(unsigned long fd, const struct compat_iovec __user *vec,
1233 unsigned long vlen)
1234 {
1235 struct file *file;
1236 int fput_needed;
1237 ssize_t ret;
1238
1239 file = fget_light(fd, &fput_needed);
1240 if (!file)
1241 return -EBADF;
1242 ret = compat_writev(file, vec, vlen, &file->f_pos);
1243 fput_light(file, fput_needed);
1244 return ret;
1245 }
1246
1247 asmlinkage ssize_t
1248 compat_sys_pwritev(unsigned long fd, const struct compat_iovec __user *vec,
1249 unsigned long vlen, u32 pos_low, u32 pos_high)
1250 {
1251 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1252 struct file *file;
1253 int fput_needed;
1254 ssize_t ret;
1255
1256 if (pos < 0)
1257 return -EINVAL;
1258 file = fget_light(fd, &fput_needed);
1259 if (!file)
1260 return -EBADF;
1261 ret = -ESPIPE;
1262 if (file->f_mode & FMODE_PWRITE)
1263 ret = compat_writev(file, vec, vlen, &pos);
1264 fput_light(file, fput_needed);
1265 return ret;
1266 }
1267
1268 asmlinkage long
1269 compat_sys_vmsplice(int fd, const struct compat_iovec __user *iov32,
1270 unsigned int nr_segs, unsigned int flags)
1271 {
1272 unsigned i;
1273 struct iovec __user *iov;
1274 if (nr_segs > UIO_MAXIOV)
1275 return -EINVAL;
1276 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1277 for (i = 0; i < nr_segs; i++) {
1278 struct compat_iovec v;
1279 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1280 get_user(v.iov_len, &iov32[i].iov_len) ||
1281 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1282 put_user(v.iov_len, &iov[i].iov_len))
1283 return -EFAULT;
1284 }
1285 return sys_vmsplice(fd, iov, nr_segs, flags);
1286 }
1287
1288 /*
1289 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1290 * O_LARGEFILE flag.
1291 */
1292 asmlinkage long
1293 compat_sys_open(const char __user *filename, int flags, int mode)
1294 {
1295 return do_sys_open(AT_FDCWD, filename, flags, mode);
1296 }
1297
1298 /*
1299 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1300 * O_LARGEFILE flag.
1301 */
1302 asmlinkage long
1303 compat_sys_openat(unsigned int dfd, const char __user *filename, int flags, int mode)
1304 {
1305 return do_sys_open(dfd, filename, flags, mode);
1306 }
1307
1308 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1309
1310 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1311 int timeval, int ret)
1312 {
1313 struct timespec ts;
1314
1315 if (!p)
1316 return ret;
1317
1318 if (current->personality & STICKY_TIMEOUTS)
1319 goto sticky;
1320
1321 /* No update for zero timeout */
1322 if (!end_time->tv_sec && !end_time->tv_nsec)
1323 return ret;
1324
1325 ktime_get_ts(&ts);
1326 ts = timespec_sub(*end_time, ts);
1327 if (ts.tv_sec < 0)
1328 ts.tv_sec = ts.tv_nsec = 0;
1329
1330 if (timeval) {
1331 struct compat_timeval rtv;
1332
1333 rtv.tv_sec = ts.tv_sec;
1334 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1335
1336 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1337 return ret;
1338 } else {
1339 struct compat_timespec rts;
1340
1341 rts.tv_sec = ts.tv_sec;
1342 rts.tv_nsec = ts.tv_nsec;
1343
1344 if (!copy_to_user(p, &rts, sizeof(rts)))
1345 return ret;
1346 }
1347 /*
1348 * If an application puts its timeval in read-only memory, we
1349 * don't want the Linux-specific update to the timeval to
1350 * cause a fault after the select has completed
1351 * successfully. However, because we're not updating the
1352 * timeval, we can't restart the system call.
1353 */
1354
1355 sticky:
1356 if (ret == -ERESTARTNOHAND)
1357 ret = -EINTR;
1358 return ret;
1359 }
1360
1361 /*
1362 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1363 * 64-bit unsigned longs.
1364 */
1365 static
1366 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1367 unsigned long *fdset)
1368 {
1369 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1370 if (ufdset) {
1371 unsigned long odd;
1372
1373 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1374 return -EFAULT;
1375
1376 odd = nr & 1UL;
1377 nr &= ~1UL;
1378 while (nr) {
1379 unsigned long h, l;
1380 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1381 return -EFAULT;
1382 ufdset += 2;
1383 *fdset++ = h << 32 | l;
1384 nr -= 2;
1385 }
1386 if (odd && __get_user(*fdset, ufdset))
1387 return -EFAULT;
1388 } else {
1389 /* Tricky, must clear full unsigned long in the
1390 * kernel fdset at the end, this makes sure that
1391 * actually happens.
1392 */
1393 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1394 }
1395 return 0;
1396 }
1397
1398 static
1399 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1400 unsigned long *fdset)
1401 {
1402 unsigned long odd;
1403 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1404
1405 if (!ufdset)
1406 return 0;
1407
1408 odd = nr & 1UL;
1409 nr &= ~1UL;
1410 while (nr) {
1411 unsigned long h, l;
1412 l = *fdset++;
1413 h = l >> 32;
1414 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1415 return -EFAULT;
1416 ufdset += 2;
1417 nr -= 2;
1418 }
1419 if (odd && __put_user(*fdset, ufdset))
1420 return -EFAULT;
1421 return 0;
1422 }
1423
1424
1425 /*
1426 * This is a virtual copy of sys_select from fs/select.c and probably
1427 * should be compared to it from time to time
1428 */
1429
1430 /*
1431 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1432 * like to be certain this leads to no problems. So I return
1433 * EINTR just for safety.
1434 *
1435 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1436 * I'm trying ERESTARTNOHAND which restart only when you want to.
1437 */
1438 int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1439 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1440 struct timespec *end_time)
1441 {
1442 fd_set_bits fds;
1443 void *bits;
1444 int size, max_fds, ret = -EINVAL;
1445 struct fdtable *fdt;
1446 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1447
1448 if (n < 0)
1449 goto out_nofds;
1450
1451 /* max_fds can increase, so grab it once to avoid race */
1452 rcu_read_lock();
1453 fdt = files_fdtable(current->files);
1454 max_fds = fdt->max_fds;
1455 rcu_read_unlock();
1456 if (n > max_fds)
1457 n = max_fds;
1458
1459 /*
1460 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1461 * since we used fdset we need to allocate memory in units of
1462 * long-words.
1463 */
1464 size = FDS_BYTES(n);
1465 bits = stack_fds;
1466 if (size > sizeof(stack_fds) / 6) {
1467 bits = kmalloc(6 * size, GFP_KERNEL);
1468 ret = -ENOMEM;
1469 if (!bits)
1470 goto out_nofds;
1471 }
1472 fds.in = (unsigned long *) bits;
1473 fds.out = (unsigned long *) (bits + size);
1474 fds.ex = (unsigned long *) (bits + 2*size);
1475 fds.res_in = (unsigned long *) (bits + 3*size);
1476 fds.res_out = (unsigned long *) (bits + 4*size);
1477 fds.res_ex = (unsigned long *) (bits + 5*size);
1478
1479 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1480 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1481 (ret = compat_get_fd_set(n, exp, fds.ex)))
1482 goto out;
1483 zero_fd_set(n, fds.res_in);
1484 zero_fd_set(n, fds.res_out);
1485 zero_fd_set(n, fds.res_ex);
1486
1487 ret = do_select(n, &fds, end_time);
1488
1489 if (ret < 0)
1490 goto out;
1491 if (!ret) {
1492 ret = -ERESTARTNOHAND;
1493 if (signal_pending(current))
1494 goto out;
1495 ret = 0;
1496 }
1497
1498 if (compat_set_fd_set(n, inp, fds.res_in) ||
1499 compat_set_fd_set(n, outp, fds.res_out) ||
1500 compat_set_fd_set(n, exp, fds.res_ex))
1501 ret = -EFAULT;
1502 out:
1503 if (bits != stack_fds)
1504 kfree(bits);
1505 out_nofds:
1506 return ret;
1507 }
1508
1509 asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1510 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1511 struct compat_timeval __user *tvp)
1512 {
1513 struct timespec end_time, *to = NULL;
1514 struct compat_timeval tv;
1515 int ret;
1516
1517 if (tvp) {
1518 if (copy_from_user(&tv, tvp, sizeof(tv)))
1519 return -EFAULT;
1520
1521 to = &end_time;
1522 if (poll_select_set_timeout(to,
1523 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1524 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1525 return -EINVAL;
1526 }
1527
1528 ret = compat_core_sys_select(n, inp, outp, exp, to);
1529 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1530
1531 return ret;
1532 }
1533
1534 struct compat_sel_arg_struct {
1535 compat_ulong_t n;
1536 compat_uptr_t inp;
1537 compat_uptr_t outp;
1538 compat_uptr_t exp;
1539 compat_uptr_t tvp;
1540 };
1541
1542 asmlinkage long compat_sys_old_select(struct compat_sel_arg_struct __user *arg)
1543 {
1544 struct compat_sel_arg_struct a;
1545
1546 if (copy_from_user(&a, arg, sizeof(a)))
1547 return -EFAULT;
1548 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1549 compat_ptr(a.exp), compat_ptr(a.tvp));
1550 }
1551
1552 #ifdef HAVE_SET_RESTORE_SIGMASK
1553 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1554 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1555 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1556 compat_size_t sigsetsize)
1557 {
1558 compat_sigset_t ss32;
1559 sigset_t ksigmask, sigsaved;
1560 struct compat_timespec ts;
1561 struct timespec end_time, *to = NULL;
1562 int ret;
1563
1564 if (tsp) {
1565 if (copy_from_user(&ts, tsp, sizeof(ts)))
1566 return -EFAULT;
1567
1568 to = &end_time;
1569 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1570 return -EINVAL;
1571 }
1572
1573 if (sigmask) {
1574 if (sigsetsize != sizeof(compat_sigset_t))
1575 return -EINVAL;
1576 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1577 return -EFAULT;
1578 sigset_from_compat(&ksigmask, &ss32);
1579
1580 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1581 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1582 }
1583
1584 ret = compat_core_sys_select(n, inp, outp, exp, to);
1585 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1586
1587 if (ret == -ERESTARTNOHAND) {
1588 /*
1589 * Don't restore the signal mask yet. Let do_signal() deliver
1590 * the signal on the way back to userspace, before the signal
1591 * mask is restored.
1592 */
1593 if (sigmask) {
1594 memcpy(&current->saved_sigmask, &sigsaved,
1595 sizeof(sigsaved));
1596 set_restore_sigmask();
1597 }
1598 } else if (sigmask)
1599 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1600
1601 return ret;
1602 }
1603
1604 asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1605 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1606 struct compat_timespec __user *tsp, void __user *sig)
1607 {
1608 compat_size_t sigsetsize = 0;
1609 compat_uptr_t up = 0;
1610
1611 if (sig) {
1612 if (!access_ok(VERIFY_READ, sig,
1613 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1614 __get_user(up, (compat_uptr_t __user *)sig) ||
1615 __get_user(sigsetsize,
1616 (compat_size_t __user *)(sig+sizeof(up))))
1617 return -EFAULT;
1618 }
1619 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1620 sigsetsize);
1621 }
1622
1623 asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1624 unsigned int nfds, struct compat_timespec __user *tsp,
1625 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1626 {
1627 compat_sigset_t ss32;
1628 sigset_t ksigmask, sigsaved;
1629 struct compat_timespec ts;
1630 struct timespec end_time, *to = NULL;
1631 int ret;
1632
1633 if (tsp) {
1634 if (copy_from_user(&ts, tsp, sizeof(ts)))
1635 return -EFAULT;
1636
1637 to = &end_time;
1638 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1639 return -EINVAL;
1640 }
1641
1642 if (sigmask) {
1643 if (sigsetsize != sizeof(compat_sigset_t))
1644 return -EINVAL;
1645 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1646 return -EFAULT;
1647 sigset_from_compat(&ksigmask, &ss32);
1648
1649 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1650 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1651 }
1652
1653 ret = do_sys_poll(ufds, nfds, to);
1654
1655 /* We can restart this syscall, usually */
1656 if (ret == -EINTR) {
1657 /*
1658 * Don't restore the signal mask yet. Let do_signal() deliver
1659 * the signal on the way back to userspace, before the signal
1660 * mask is restored.
1661 */
1662 if (sigmask) {
1663 memcpy(&current->saved_sigmask, &sigsaved,
1664 sizeof(sigsaved));
1665 set_restore_sigmask();
1666 }
1667 ret = -ERESTARTNOHAND;
1668 } else if (sigmask)
1669 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1670
1671 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1672
1673 return ret;
1674 }
1675 #endif /* HAVE_SET_RESTORE_SIGMASK */
1676
1677 long asmlinkage compat_sys_nfsservctl(int cmd, void *notused, void *notused2)
1678 {
1679 return sys_ni_syscall();
1680 }
1681
1682 #ifdef CONFIG_EPOLL
1683
1684 #ifdef HAVE_SET_RESTORE_SIGMASK
1685 asmlinkage long compat_sys_epoll_pwait(int epfd,
1686 struct compat_epoll_event __user *events,
1687 int maxevents, int timeout,
1688 const compat_sigset_t __user *sigmask,
1689 compat_size_t sigsetsize)
1690 {
1691 long err;
1692 compat_sigset_t csigmask;
1693 sigset_t ksigmask, sigsaved;
1694
1695 /*
1696 * If the caller wants a certain signal mask to be set during the wait,
1697 * we apply it here.
1698 */
1699 if (sigmask) {
1700 if (sigsetsize != sizeof(compat_sigset_t))
1701 return -EINVAL;
1702 if (copy_from_user(&csigmask, sigmask, sizeof(csigmask)))
1703 return -EFAULT;
1704 sigset_from_compat(&ksigmask, &csigmask);
1705 sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
1706 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1707 }
1708
1709 err = sys_epoll_wait(epfd, events, maxevents, timeout);
1710
1711 /*
1712 * If we changed the signal mask, we need to restore the original one.
1713 * In case we've got a signal while waiting, we do not restore the
1714 * signal mask yet, and we allow do_signal() to deliver the signal on
1715 * the way back to userspace, before the signal mask is restored.
1716 */
1717 if (sigmask) {
1718 if (err == -EINTR) {
1719 memcpy(&current->saved_sigmask, &sigsaved,
1720 sizeof(sigsaved));
1721 set_restore_sigmask();
1722 } else
1723 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1724 }
1725
1726 return err;
1727 }
1728 #endif /* HAVE_SET_RESTORE_SIGMASK */
1729
1730 #endif /* CONFIG_EPOLL */
1731
1732 #ifdef CONFIG_SIGNALFD
1733
1734 asmlinkage long compat_sys_signalfd4(int ufd,
1735 const compat_sigset_t __user *sigmask,
1736 compat_size_t sigsetsize, int flags)
1737 {
1738 compat_sigset_t ss32;
1739 sigset_t tmp;
1740 sigset_t __user *ksigmask;
1741
1742 if (sigsetsize != sizeof(compat_sigset_t))
1743 return -EINVAL;
1744 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1745 return -EFAULT;
1746 sigset_from_compat(&tmp, &ss32);
1747 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
1748 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
1749 return -EFAULT;
1750
1751 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
1752 }
1753
1754 asmlinkage long compat_sys_signalfd(int ufd,
1755 const compat_sigset_t __user *sigmask,
1756 compat_size_t sigsetsize)
1757 {
1758 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
1759 }
1760 #endif /* CONFIG_SIGNALFD */
1761
1762 #ifdef CONFIG_TIMERFD
1763
1764 asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
1765 const struct compat_itimerspec __user *utmr,
1766 struct compat_itimerspec __user *otmr)
1767 {
1768 int error;
1769 struct itimerspec t;
1770 struct itimerspec __user *ut;
1771
1772 if (get_compat_itimerspec(&t, utmr))
1773 return -EFAULT;
1774 ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
1775 if (copy_to_user(&ut[0], &t, sizeof(t)))
1776 return -EFAULT;
1777 error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
1778 if (!error && otmr)
1779 error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
1780 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1781
1782 return error;
1783 }
1784
1785 asmlinkage long compat_sys_timerfd_gettime(int ufd,
1786 struct compat_itimerspec __user *otmr)
1787 {
1788 int error;
1789 struct itimerspec t;
1790 struct itimerspec __user *ut;
1791
1792 ut = compat_alloc_user_space(sizeof(struct itimerspec));
1793 error = sys_timerfd_gettime(ufd, ut);
1794 if (!error)
1795 error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
1796 put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
1797
1798 return error;
1799 }
1800
1801 #endif /* CONFIG_TIMERFD */
1802
1803 #ifdef CONFIG_FHANDLE
1804 /*
1805 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1806 * doesn't set the O_LARGEFILE flag.
1807 */
1808 asmlinkage long
1809 compat_sys_open_by_handle_at(int mountdirfd,
1810 struct file_handle __user *handle, int flags)
1811 {
1812 return do_handle_open(mountdirfd, handle, flags);
1813 }
1814 #endif
Linux Btrfs Development