Loading...
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/fs_struct.h>
48#include <linux/slab.h>
49#include <linux/pagemap.h>
50#include <linux/aio.h>
51
52#include <linux/uaccess.h>
53#include <asm/mmu_context.h>
54#include <asm/ioctls.h>
55#include "internal.h"
56
57/*
58 * Not all architectures have sys_utime, so implement this in terms
59 * of sys_utimes.
60 */
61COMPAT_SYSCALL_DEFINE2(utime, const char __user *, filename,
62 struct compat_utimbuf __user *, t)
63{
64 struct timespec tv[2];
65
66 if (t) {
67 if (get_user(tv[0].tv_sec, &t->actime) ||
68 get_user(tv[1].tv_sec, &t->modtime))
69 return -EFAULT;
70 tv[0].tv_nsec = 0;
71 tv[1].tv_nsec = 0;
72 }
73 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
74}
75
76COMPAT_SYSCALL_DEFINE4(utimensat, unsigned int, dfd, const char __user *, filename, struct compat_timespec __user *, t, int, flags)
77{
78 struct timespec tv[2];
79
80 if (t) {
81 if (compat_get_timespec(&tv[0], &t[0]) ||
82 compat_get_timespec(&tv[1], &t[1]))
83 return -EFAULT;
84
85 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
86 return 0;
87 }
88 return do_utimes(dfd, filename, t ? tv : NULL, flags);
89}
90
91COMPAT_SYSCALL_DEFINE3(futimesat, unsigned int, dfd, const char __user *, filename, struct compat_timeval __user *, t)
92{
93 struct timespec tv[2];
94
95 if (t) {
96 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
97 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
98 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
99 get_user(tv[1].tv_nsec, &t[1].tv_usec))
100 return -EFAULT;
101 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
102 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
103 return -EINVAL;
104 tv[0].tv_nsec *= 1000;
105 tv[1].tv_nsec *= 1000;
106 }
107 return do_utimes(dfd, filename, t ? tv : NULL, 0);
108}
109
110COMPAT_SYSCALL_DEFINE2(utimes, const char __user *, filename, struct compat_timeval __user *, t)
111{
112 return compat_sys_futimesat(AT_FDCWD, filename, t);
113}
114
115static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
116{
117 struct compat_stat tmp;
118
119 if (!old_valid_dev(stat->dev) || !old_valid_dev(stat->rdev))
120 return -EOVERFLOW;
121
122 memset(&tmp, 0, sizeof(tmp));
123 tmp.st_dev = old_encode_dev(stat->dev);
124 tmp.st_ino = stat->ino;
125 if (sizeof(tmp.st_ino) < sizeof(stat->ino) && tmp.st_ino != stat->ino)
126 return -EOVERFLOW;
127 tmp.st_mode = stat->mode;
128 tmp.st_nlink = stat->nlink;
129 if (tmp.st_nlink != stat->nlink)
130 return -EOVERFLOW;
131 SET_UID(tmp.st_uid, from_kuid_munged(current_user_ns(), stat->uid));
132 SET_GID(tmp.st_gid, from_kgid_munged(current_user_ns(), stat->gid));
133 tmp.st_rdev = old_encode_dev(stat->rdev);
134 if ((u64) stat->size > MAX_NON_LFS)
135 return -EOVERFLOW;
136 tmp.st_size = stat->size;
137 tmp.st_atime = stat->atime.tv_sec;
138 tmp.st_atime_nsec = stat->atime.tv_nsec;
139 tmp.st_mtime = stat->mtime.tv_sec;
140 tmp.st_mtime_nsec = stat->mtime.tv_nsec;
141 tmp.st_ctime = stat->ctime.tv_sec;
142 tmp.st_ctime_nsec = stat->ctime.tv_nsec;
143 tmp.st_blocks = stat->blocks;
144 tmp.st_blksize = stat->blksize;
145 return copy_to_user(ubuf, &tmp, sizeof(tmp)) ? -EFAULT : 0;
146}
147
148COMPAT_SYSCALL_DEFINE2(newstat, const char __user *, filename,
149 struct compat_stat __user *, statbuf)
150{
151 struct kstat stat;
152 int error;
153
154 error = vfs_stat(filename, &stat);
155 if (error)
156 return error;
157 return cp_compat_stat(&stat, statbuf);
158}
159
160COMPAT_SYSCALL_DEFINE2(newlstat, const char __user *, filename,
161 struct compat_stat __user *, statbuf)
162{
163 struct kstat stat;
164 int error;
165
166 error = vfs_lstat(filename, &stat);
167 if (error)
168 return error;
169 return cp_compat_stat(&stat, statbuf);
170}
171
172#ifndef __ARCH_WANT_STAT64
173COMPAT_SYSCALL_DEFINE4(newfstatat, unsigned int, dfd,
174 const char __user *, filename,
175 struct compat_stat __user *, statbuf, int, flag)
176{
177 struct kstat stat;
178 int error;
179
180 error = vfs_fstatat(dfd, filename, &stat, flag);
181 if (error)
182 return error;
183 return cp_compat_stat(&stat, statbuf);
184}
185#endif
186
187COMPAT_SYSCALL_DEFINE2(newfstat, unsigned int, fd,
188 struct compat_stat __user *, statbuf)
189{
190 struct kstat stat;
191 int error = vfs_fstat(fd, &stat);
192
193 if (!error)
194 error = cp_compat_stat(&stat, statbuf);
195 return error;
196}
197
198static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
199{
200
201 if (sizeof ubuf->f_blocks == 4) {
202 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
203 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
204 return -EOVERFLOW;
205 /* f_files and f_ffree may be -1; it's okay
206 * to stuff that into 32 bits */
207 if (kbuf->f_files != 0xffffffffffffffffULL
208 && (kbuf->f_files & 0xffffffff00000000ULL))
209 return -EOVERFLOW;
210 if (kbuf->f_ffree != 0xffffffffffffffffULL
211 && (kbuf->f_ffree & 0xffffffff00000000ULL))
212 return -EOVERFLOW;
213 }
214 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
215 __put_user(kbuf->f_type, &ubuf->f_type) ||
216 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
217 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
218 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
219 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
220 __put_user(kbuf->f_files, &ubuf->f_files) ||
221 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
222 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
223 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
224 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
225 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
226 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
227 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
228 return -EFAULT;
229 return 0;
230}
231
232/*
233 * The following statfs calls are copies of code from fs/statfs.c and
234 * should be checked against those from time to time
235 */
236COMPAT_SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct compat_statfs __user *, buf)
237{
238 struct kstatfs tmp;
239 int error = user_statfs(pathname, &tmp);
240 if (!error)
241 error = put_compat_statfs(buf, &tmp);
242 return error;
243}
244
245COMPAT_SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct compat_statfs __user *, buf)
246{
247 struct kstatfs tmp;
248 int error = fd_statfs(fd, &tmp);
249 if (!error)
250 error = put_compat_statfs(buf, &tmp);
251 return error;
252}
253
254static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
255{
256 if (sizeof(ubuf->f_bsize) == 4) {
257 if ((kbuf->f_type | kbuf->f_bsize | kbuf->f_namelen |
258 kbuf->f_frsize | kbuf->f_flags) & 0xffffffff00000000ULL)
259 return -EOVERFLOW;
260 /* f_files and f_ffree may be -1; it's okay
261 * to stuff that into 32 bits */
262 if (kbuf->f_files != 0xffffffffffffffffULL
263 && (kbuf->f_files & 0xffffffff00000000ULL))
264 return -EOVERFLOW;
265 if (kbuf->f_ffree != 0xffffffffffffffffULL
266 && (kbuf->f_ffree & 0xffffffff00000000ULL))
267 return -EOVERFLOW;
268 }
269 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
270 __put_user(kbuf->f_type, &ubuf->f_type) ||
271 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
272 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
273 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
274 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
275 __put_user(kbuf->f_files, &ubuf->f_files) ||
276 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
277 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
278 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
279 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
280 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
281 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
282 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
283 return -EFAULT;
284 return 0;
285}
286
287COMPAT_SYSCALL_DEFINE3(statfs64, const char __user *, pathname, compat_size_t, sz, struct compat_statfs64 __user *, buf)
288{
289 struct kstatfs tmp;
290 int error;
291
292 if (sz != sizeof(*buf))
293 return -EINVAL;
294
295 error = user_statfs(pathname, &tmp);
296 if (!error)
297 error = put_compat_statfs64(buf, &tmp);
298 return error;
299}
300
301COMPAT_SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, compat_size_t, sz, struct compat_statfs64 __user *, buf)
302{
303 struct kstatfs tmp;
304 int error;
305
306 if (sz != sizeof(*buf))
307 return -EINVAL;
308
309 error = fd_statfs(fd, &tmp);
310 if (!error)
311 error = put_compat_statfs64(buf, &tmp);
312 return error;
313}
314
315/*
316 * This is a copy of sys_ustat, just dealing with a structure layout.
317 * Given how simple this syscall is that apporach is more maintainable
318 * than the various conversion hacks.
319 */
320COMPAT_SYSCALL_DEFINE2(ustat, unsigned, dev, struct compat_ustat __user *, u)
321{
322 struct compat_ustat tmp;
323 struct kstatfs sbuf;
324 int err = vfs_ustat(new_decode_dev(dev), &sbuf);
325 if (err)
326 return err;
327
328 memset(&tmp, 0, sizeof(struct compat_ustat));
329 tmp.f_tfree = sbuf.f_bfree;
330 tmp.f_tinode = sbuf.f_ffree;
331 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
332 return -EFAULT;
333 return 0;
334}
335
336static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
337{
338 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
339 __get_user(kfl->l_type, &ufl->l_type) ||
340 __get_user(kfl->l_whence, &ufl->l_whence) ||
341 __get_user(kfl->l_start, &ufl->l_start) ||
342 __get_user(kfl->l_len, &ufl->l_len) ||
343 __get_user(kfl->l_pid, &ufl->l_pid))
344 return -EFAULT;
345 return 0;
346}
347
348static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
349{
350 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
351 __put_user(kfl->l_type, &ufl->l_type) ||
352 __put_user(kfl->l_whence, &ufl->l_whence) ||
353 __put_user(kfl->l_start, &ufl->l_start) ||
354 __put_user(kfl->l_len, &ufl->l_len) ||
355 __put_user(kfl->l_pid, &ufl->l_pid))
356 return -EFAULT;
357 return 0;
358}
359
360#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
361static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
362{
363 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
364 __get_user(kfl->l_type, &ufl->l_type) ||
365 __get_user(kfl->l_whence, &ufl->l_whence) ||
366 __get_user(kfl->l_start, &ufl->l_start) ||
367 __get_user(kfl->l_len, &ufl->l_len) ||
368 __get_user(kfl->l_pid, &ufl->l_pid))
369 return -EFAULT;
370 return 0;
371}
372#endif
373
374#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
375static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
376{
377 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
378 __put_user(kfl->l_type, &ufl->l_type) ||
379 __put_user(kfl->l_whence, &ufl->l_whence) ||
380 __put_user(kfl->l_start, &ufl->l_start) ||
381 __put_user(kfl->l_len, &ufl->l_len) ||
382 __put_user(kfl->l_pid, &ufl->l_pid))
383 return -EFAULT;
384 return 0;
385}
386#endif
387
388static unsigned int
389convert_fcntl_cmd(unsigned int cmd)
390{
391 switch (cmd) {
392 case F_GETLK64:
393 return F_GETLK;
394 case F_SETLK64:
395 return F_SETLK;
396 case F_SETLKW64:
397 return F_SETLKW;
398 }
399
400 return cmd;
401}
402
403COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
404 compat_ulong_t, arg)
405{
406 mm_segment_t old_fs;
407 struct flock f;
408 long ret;
409 unsigned int conv_cmd;
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 case F_OFD_GETLK:
447 case F_OFD_SETLK:
448 case F_OFD_SETLKW:
449 ret = get_compat_flock64(&f, compat_ptr(arg));
450 if (ret != 0)
451 break;
452 old_fs = get_fs();
453 set_fs(KERNEL_DS);
454 conv_cmd = convert_fcntl_cmd(cmd);
455 ret = sys_fcntl(fd, conv_cmd, (unsigned long)&f);
456 set_fs(old_fs);
457 if ((conv_cmd == F_GETLK || conv_cmd == F_OFD_GETLK) && ret == 0) {
458 /* need to return lock information - see above for commentary */
459 if (f.l_start > COMPAT_LOFF_T_MAX)
460 ret = -EOVERFLOW;
461 if (f.l_len > COMPAT_LOFF_T_MAX)
462 f.l_len = COMPAT_LOFF_T_MAX;
463 if (ret == 0)
464 ret = put_compat_flock64(&f, compat_ptr(arg));
465 }
466 break;
467
468 default:
469 ret = sys_fcntl(fd, cmd, arg);
470 break;
471 }
472 return ret;
473}
474
475COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
476 compat_ulong_t, arg)
477{
478 switch (cmd) {
479 case F_GETLK64:
480 case F_SETLK64:
481 case F_SETLKW64:
482 case F_OFD_GETLK:
483 case F_OFD_SETLK:
484 case F_OFD_SETLKW:
485 return -EINVAL;
486 }
487 return compat_sys_fcntl64(fd, cmd, arg);
488}
489
490/* A write operation does a read from user space and vice versa */
491#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
492
493ssize_t compat_rw_copy_check_uvector(int type,
494 const struct compat_iovec __user *uvector, unsigned long nr_segs,
495 unsigned long fast_segs, struct iovec *fast_pointer,
496 struct iovec **ret_pointer)
497{
498 compat_ssize_t tot_len;
499 struct iovec *iov = *ret_pointer = fast_pointer;
500 ssize_t ret = 0;
501 int seg;
502
503 /*
504 * SuS says "The readv() function *may* fail if the iovcnt argument
505 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
506 * traditionally returned zero for zero segments, so...
507 */
508 if (nr_segs == 0)
509 goto out;
510
511 ret = -EINVAL;
512 if (nr_segs > UIO_MAXIOV)
513 goto out;
514 if (nr_segs > fast_segs) {
515 ret = -ENOMEM;
516 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
517 if (iov == NULL)
518 goto out;
519 }
520 *ret_pointer = iov;
521
522 ret = -EFAULT;
523 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
524 goto out;
525
526 /*
527 * Single unix specification:
528 * We should -EINVAL if an element length is not >= 0 and fitting an
529 * ssize_t.
530 *
531 * In Linux, the total length is limited to MAX_RW_COUNT, there is
532 * no overflow possibility.
533 */
534 tot_len = 0;
535 ret = -EINVAL;
536 for (seg = 0; seg < nr_segs; seg++) {
537 compat_uptr_t buf;
538 compat_ssize_t len;
539
540 if (__get_user(len, &uvector->iov_len) ||
541 __get_user(buf, &uvector->iov_base)) {
542 ret = -EFAULT;
543 goto out;
544 }
545 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
546 goto out;
547 if (type >= 0 &&
548 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
549 ret = -EFAULT;
550 goto out;
551 }
552 if (len > MAX_RW_COUNT - tot_len)
553 len = MAX_RW_COUNT - tot_len;
554 tot_len += len;
555 iov->iov_base = compat_ptr(buf);
556 iov->iov_len = (compat_size_t) len;
557 uvector++;
558 iov++;
559 }
560 ret = tot_len;
561
562out:
563 return ret;
564}
565
566struct compat_ncp_mount_data {
567 compat_int_t version;
568 compat_uint_t ncp_fd;
569 __compat_uid_t mounted_uid;
570 compat_pid_t wdog_pid;
571 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
572 compat_uint_t time_out;
573 compat_uint_t retry_count;
574 compat_uint_t flags;
575 __compat_uid_t uid;
576 __compat_gid_t gid;
577 compat_mode_t file_mode;
578 compat_mode_t dir_mode;
579};
580
581struct compat_ncp_mount_data_v4 {
582 compat_int_t version;
583 compat_ulong_t flags;
584 compat_ulong_t mounted_uid;
585 compat_long_t wdog_pid;
586 compat_uint_t ncp_fd;
587 compat_uint_t time_out;
588 compat_uint_t retry_count;
589 compat_ulong_t uid;
590 compat_ulong_t gid;
591 compat_ulong_t file_mode;
592 compat_ulong_t dir_mode;
593};
594
595static void *do_ncp_super_data_conv(void *raw_data)
596{
597 int version = *(unsigned int *)raw_data;
598
599 if (version == 3) {
600 struct compat_ncp_mount_data *c_n = raw_data;
601 struct ncp_mount_data *n = raw_data;
602
603 n->dir_mode = c_n->dir_mode;
604 n->file_mode = c_n->file_mode;
605 n->gid = c_n->gid;
606 n->uid = c_n->uid;
607 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
608 n->wdog_pid = c_n->wdog_pid;
609 n->mounted_uid = c_n->mounted_uid;
610 } else if (version == 4) {
611 struct compat_ncp_mount_data_v4 *c_n = raw_data;
612 struct ncp_mount_data_v4 *n = raw_data;
613
614 n->dir_mode = c_n->dir_mode;
615 n->file_mode = c_n->file_mode;
616 n->gid = c_n->gid;
617 n->uid = c_n->uid;
618 n->retry_count = c_n->retry_count;
619 n->time_out = c_n->time_out;
620 n->ncp_fd = c_n->ncp_fd;
621 n->wdog_pid = c_n->wdog_pid;
622 n->mounted_uid = c_n->mounted_uid;
623 n->flags = c_n->flags;
624 } else if (version != 5) {
625 return NULL;
626 }
627
628 return raw_data;
629}
630
631
632struct compat_nfs_string {
633 compat_uint_t len;
634 compat_uptr_t data;
635};
636
637static inline void compat_nfs_string(struct nfs_string *dst,
638 struct compat_nfs_string *src)
639{
640 dst->data = compat_ptr(src->data);
641 dst->len = src->len;
642}
643
644struct compat_nfs4_mount_data_v1 {
645 compat_int_t version;
646 compat_int_t flags;
647 compat_int_t rsize;
648 compat_int_t wsize;
649 compat_int_t timeo;
650 compat_int_t retrans;
651 compat_int_t acregmin;
652 compat_int_t acregmax;
653 compat_int_t acdirmin;
654 compat_int_t acdirmax;
655 struct compat_nfs_string client_addr;
656 struct compat_nfs_string mnt_path;
657 struct compat_nfs_string hostname;
658 compat_uint_t host_addrlen;
659 compat_uptr_t host_addr;
660 compat_int_t proto;
661 compat_int_t auth_flavourlen;
662 compat_uptr_t auth_flavours;
663};
664
665static int do_nfs4_super_data_conv(void *raw_data)
666{
667 int version = *(compat_uint_t *) raw_data;
668
669 if (version == 1) {
670 struct compat_nfs4_mount_data_v1 *raw = raw_data;
671 struct nfs4_mount_data *real = raw_data;
672
673 /* copy the fields backwards */
674 real->auth_flavours = compat_ptr(raw->auth_flavours);
675 real->auth_flavourlen = raw->auth_flavourlen;
676 real->proto = raw->proto;
677 real->host_addr = compat_ptr(raw->host_addr);
678 real->host_addrlen = raw->host_addrlen;
679 compat_nfs_string(&real->hostname, &raw->hostname);
680 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
681 compat_nfs_string(&real->client_addr, &raw->client_addr);
682 real->acdirmax = raw->acdirmax;
683 real->acdirmin = raw->acdirmin;
684 real->acregmax = raw->acregmax;
685 real->acregmin = raw->acregmin;
686 real->retrans = raw->retrans;
687 real->timeo = raw->timeo;
688 real->wsize = raw->wsize;
689 real->rsize = raw->rsize;
690 real->flags = raw->flags;
691 real->version = raw->version;
692 }
693
694 return 0;
695}
696
697#define NCPFS_NAME "ncpfs"
698#define NFS4_NAME "nfs4"
699
700COMPAT_SYSCALL_DEFINE5(mount, const char __user *, dev_name,
701 const char __user *, dir_name,
702 const char __user *, type, compat_ulong_t, flags,
703 const void __user *, data)
704{
705 char *kernel_type;
706 void *options;
707 char *kernel_dev;
708 int retval;
709
710 kernel_type = copy_mount_string(type);
711 retval = PTR_ERR(kernel_type);
712 if (IS_ERR(kernel_type))
713 goto out;
714
715 kernel_dev = copy_mount_string(dev_name);
716 retval = PTR_ERR(kernel_dev);
717 if (IS_ERR(kernel_dev))
718 goto out1;
719
720 options = copy_mount_options(data);
721 retval = PTR_ERR(options);
722 if (IS_ERR(options))
723 goto out2;
724
725 if (kernel_type && options) {
726 if (!strcmp(kernel_type, NCPFS_NAME)) {
727 do_ncp_super_data_conv(options);
728 } else if (!strcmp(kernel_type, NFS4_NAME)) {
729 retval = -EINVAL;
730 if (do_nfs4_super_data_conv(options))
731 goto out3;
732 }
733 }
734
735 retval = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
736
737 out3:
738 kfree(options);
739 out2:
740 kfree(kernel_dev);
741 out1:
742 kfree(kernel_type);
743 out:
744 return retval;
745}
746
747struct compat_old_linux_dirent {
748 compat_ulong_t d_ino;
749 compat_ulong_t d_offset;
750 unsigned short d_namlen;
751 char d_name[1];
752};
753
754struct compat_readdir_callback {
755 struct dir_context ctx;
756 struct compat_old_linux_dirent __user *dirent;
757 int result;
758};
759
760static int compat_fillonedir(struct dir_context *ctx, const char *name,
761 int namlen, loff_t offset, u64 ino,
762 unsigned int d_type)
763{
764 struct compat_readdir_callback *buf =
765 container_of(ctx, struct compat_readdir_callback, ctx);
766 struct compat_old_linux_dirent __user *dirent;
767 compat_ulong_t d_ino;
768
769 if (buf->result)
770 return -EINVAL;
771 d_ino = ino;
772 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
773 buf->result = -EOVERFLOW;
774 return -EOVERFLOW;
775 }
776 buf->result++;
777 dirent = buf->dirent;
778 if (!access_ok(VERIFY_WRITE, dirent,
779 (unsigned long)(dirent->d_name + namlen + 1) -
780 (unsigned long)dirent))
781 goto efault;
782 if ( __put_user(d_ino, &dirent->d_ino) ||
783 __put_user(offset, &dirent->d_offset) ||
784 __put_user(namlen, &dirent->d_namlen) ||
785 __copy_to_user(dirent->d_name, name, namlen) ||
786 __put_user(0, dirent->d_name + namlen))
787 goto efault;
788 return 0;
789efault:
790 buf->result = -EFAULT;
791 return -EFAULT;
792}
793
794COMPAT_SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
795 struct compat_old_linux_dirent __user *, dirent, unsigned int, count)
796{
797 int error;
798 struct fd f = fdget_pos(fd);
799 struct compat_readdir_callback buf = {
800 .ctx.actor = compat_fillonedir,
801 .dirent = dirent
802 };
803
804 if (!f.file)
805 return -EBADF;
806
807 error = iterate_dir(f.file, &buf.ctx);
808 if (buf.result)
809 error = buf.result;
810
811 fdput_pos(f);
812 return error;
813}
814
815struct compat_linux_dirent {
816 compat_ulong_t d_ino;
817 compat_ulong_t d_off;
818 unsigned short d_reclen;
819 char d_name[1];
820};
821
822struct compat_getdents_callback {
823 struct dir_context ctx;
824 struct compat_linux_dirent __user *current_dir;
825 struct compat_linux_dirent __user *previous;
826 int count;
827 int error;
828};
829
830static int compat_filldir(struct dir_context *ctx, const char *name, int namlen,
831 loff_t offset, u64 ino, unsigned int d_type)
832{
833 struct compat_linux_dirent __user * dirent;
834 struct compat_getdents_callback *buf =
835 container_of(ctx, struct compat_getdents_callback, ctx);
836 compat_ulong_t d_ino;
837 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
838 namlen + 2, sizeof(compat_long_t));
839
840 buf->error = -EINVAL; /* only used if we fail.. */
841 if (reclen > buf->count)
842 return -EINVAL;
843 d_ino = ino;
844 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
845 buf->error = -EOVERFLOW;
846 return -EOVERFLOW;
847 }
848 dirent = buf->previous;
849 if (dirent) {
850 if (signal_pending(current))
851 return -EINTR;
852 if (__put_user(offset, &dirent->d_off))
853 goto efault;
854 }
855 dirent = buf->current_dir;
856 if (__put_user(d_ino, &dirent->d_ino))
857 goto efault;
858 if (__put_user(reclen, &dirent->d_reclen))
859 goto efault;
860 if (copy_to_user(dirent->d_name, name, namlen))
861 goto efault;
862 if (__put_user(0, dirent->d_name + namlen))
863 goto efault;
864 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
865 goto efault;
866 buf->previous = dirent;
867 dirent = (void __user *)dirent + reclen;
868 buf->current_dir = dirent;
869 buf->count -= reclen;
870 return 0;
871efault:
872 buf->error = -EFAULT;
873 return -EFAULT;
874}
875
876COMPAT_SYSCALL_DEFINE3(getdents, unsigned int, fd,
877 struct compat_linux_dirent __user *, dirent, unsigned int, count)
878{
879 struct fd f;
880 struct compat_linux_dirent __user * lastdirent;
881 struct compat_getdents_callback buf = {
882 .ctx.actor = compat_filldir,
883 .current_dir = dirent,
884 .count = count
885 };
886 int error;
887
888 if (!access_ok(VERIFY_WRITE, dirent, count))
889 return -EFAULT;
890
891 f = fdget_pos(fd);
892 if (!f.file)
893 return -EBADF;
894
895 error = iterate_dir(f.file, &buf.ctx);
896 if (error >= 0)
897 error = buf.error;
898 lastdirent = buf.previous;
899 if (lastdirent) {
900 if (put_user(buf.ctx.pos, &lastdirent->d_off))
901 error = -EFAULT;
902 else
903 error = count - buf.count;
904 }
905 fdput_pos(f);
906 return error;
907}
908
909#ifdef __ARCH_WANT_COMPAT_SYS_GETDENTS64
910
911struct compat_getdents_callback64 {
912 struct dir_context ctx;
913 struct linux_dirent64 __user *current_dir;
914 struct linux_dirent64 __user *previous;
915 int count;
916 int error;
917};
918
919static int compat_filldir64(struct dir_context *ctx, const char *name,
920 int namlen, loff_t offset, u64 ino,
921 unsigned int d_type)
922{
923 struct linux_dirent64 __user *dirent;
924 struct compat_getdents_callback64 *buf =
925 container_of(ctx, struct compat_getdents_callback64, ctx);
926 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
927 sizeof(u64));
928 u64 off;
929
930 buf->error = -EINVAL; /* only used if we fail.. */
931 if (reclen > buf->count)
932 return -EINVAL;
933 dirent = buf->previous;
934
935 if (dirent) {
936 if (signal_pending(current))
937 return -EINTR;
938 if (__put_user_unaligned(offset, &dirent->d_off))
939 goto efault;
940 }
941 dirent = buf->current_dir;
942 if (__put_user_unaligned(ino, &dirent->d_ino))
943 goto efault;
944 off = 0;
945 if (__put_user_unaligned(off, &dirent->d_off))
946 goto efault;
947 if (__put_user(reclen, &dirent->d_reclen))
948 goto efault;
949 if (__put_user(d_type, &dirent->d_type))
950 goto efault;
951 if (copy_to_user(dirent->d_name, name, namlen))
952 goto efault;
953 if (__put_user(0, dirent->d_name + namlen))
954 goto efault;
955 buf->previous = dirent;
956 dirent = (void __user *)dirent + reclen;
957 buf->current_dir = dirent;
958 buf->count -= reclen;
959 return 0;
960efault:
961 buf->error = -EFAULT;
962 return -EFAULT;
963}
964
965COMPAT_SYSCALL_DEFINE3(getdents64, unsigned int, fd,
966 struct linux_dirent64 __user *, dirent, unsigned int, count)
967{
968 struct fd f;
969 struct linux_dirent64 __user * lastdirent;
970 struct compat_getdents_callback64 buf = {
971 .ctx.actor = compat_filldir64,
972 .current_dir = dirent,
973 .count = count
974 };
975 int error;
976
977 if (!access_ok(VERIFY_WRITE, dirent, count))
978 return -EFAULT;
979
980 f = fdget_pos(fd);
981 if (!f.file)
982 return -EBADF;
983
984 error = iterate_dir(f.file, &buf.ctx);
985 if (error >= 0)
986 error = buf.error;
987 lastdirent = buf.previous;
988 if (lastdirent) {
989 typeof(lastdirent->d_off) d_off = buf.ctx.pos;
990 if (__put_user_unaligned(d_off, &lastdirent->d_off))
991 error = -EFAULT;
992 else
993 error = count - buf.count;
994 }
995 fdput_pos(f);
996 return error;
997}
998#endif /* __ARCH_WANT_COMPAT_SYS_GETDENTS64 */
999
1000/*
1001 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1002 * O_LARGEFILE flag.
1003 */
1004COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1005{
1006 return do_sys_open(AT_FDCWD, filename, flags, mode);
1007}
1008
1009/*
1010 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1011 * O_LARGEFILE flag.
1012 */
1013COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1014{
1015 return do_sys_open(dfd, filename, flags, mode);
1016}
1017
1018#define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1019
1020static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1021 int timeval, int ret)
1022{
1023 struct timespec ts;
1024
1025 if (!p)
1026 return ret;
1027
1028 if (current->personality & STICKY_TIMEOUTS)
1029 goto sticky;
1030
1031 /* No update for zero timeout */
1032 if (!end_time->tv_sec && !end_time->tv_nsec)
1033 return ret;
1034
1035 ktime_get_ts(&ts);
1036 ts = timespec_sub(*end_time, ts);
1037 if (ts.tv_sec < 0)
1038 ts.tv_sec = ts.tv_nsec = 0;
1039
1040 if (timeval) {
1041 struct compat_timeval rtv;
1042
1043 rtv.tv_sec = ts.tv_sec;
1044 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1045
1046 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1047 return ret;
1048 } else {
1049 struct compat_timespec rts;
1050
1051 rts.tv_sec = ts.tv_sec;
1052 rts.tv_nsec = ts.tv_nsec;
1053
1054 if (!copy_to_user(p, &rts, sizeof(rts)))
1055 return ret;
1056 }
1057 /*
1058 * If an application puts its timeval in read-only memory, we
1059 * don't want the Linux-specific update to the timeval to
1060 * cause a fault after the select has completed
1061 * successfully. However, because we're not updating the
1062 * timeval, we can't restart the system call.
1063 */
1064
1065sticky:
1066 if (ret == -ERESTARTNOHAND)
1067 ret = -EINTR;
1068 return ret;
1069}
1070
1071/*
1072 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1073 * 64-bit unsigned longs.
1074 */
1075static
1076int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1077 unsigned long *fdset)
1078{
1079 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1080 if (ufdset) {
1081 unsigned long odd;
1082
1083 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1084 return -EFAULT;
1085
1086 odd = nr & 1UL;
1087 nr &= ~1UL;
1088 while (nr) {
1089 unsigned long h, l;
1090 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1091 return -EFAULT;
1092 ufdset += 2;
1093 *fdset++ = h << 32 | l;
1094 nr -= 2;
1095 }
1096 if (odd && __get_user(*fdset, ufdset))
1097 return -EFAULT;
1098 } else {
1099 /* Tricky, must clear full unsigned long in the
1100 * kernel fdset at the end, this makes sure that
1101 * actually happens.
1102 */
1103 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1104 }
1105 return 0;
1106}
1107
1108static
1109int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1110 unsigned long *fdset)
1111{
1112 unsigned long odd;
1113 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1114
1115 if (!ufdset)
1116 return 0;
1117
1118 odd = nr & 1UL;
1119 nr &= ~1UL;
1120 while (nr) {
1121 unsigned long h, l;
1122 l = *fdset++;
1123 h = l >> 32;
1124 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1125 return -EFAULT;
1126 ufdset += 2;
1127 nr -= 2;
1128 }
1129 if (odd && __put_user(*fdset, ufdset))
1130 return -EFAULT;
1131 return 0;
1132}
1133
1134
1135/*
1136 * This is a virtual copy of sys_select from fs/select.c and probably
1137 * should be compared to it from time to time
1138 */
1139
1140/*
1141 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1142 * like to be certain this leads to no problems. So I return
1143 * EINTR just for safety.
1144 *
1145 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1146 * I'm trying ERESTARTNOHAND which restart only when you want to.
1147 */
1148int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1149 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1150 struct timespec *end_time)
1151{
1152 fd_set_bits fds;
1153 void *bits;
1154 int size, max_fds, ret = -EINVAL;
1155 struct fdtable *fdt;
1156 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1157
1158 if (n < 0)
1159 goto out_nofds;
1160
1161 /* max_fds can increase, so grab it once to avoid race */
1162 rcu_read_lock();
1163 fdt = files_fdtable(current->files);
1164 max_fds = fdt->max_fds;
1165 rcu_read_unlock();
1166 if (n > max_fds)
1167 n = max_fds;
1168
1169 /*
1170 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1171 * since we used fdset we need to allocate memory in units of
1172 * long-words.
1173 */
1174 size = FDS_BYTES(n);
1175 bits = stack_fds;
1176 if (size > sizeof(stack_fds) / 6) {
1177 bits = kmalloc(6 * size, GFP_KERNEL);
1178 ret = -ENOMEM;
1179 if (!bits)
1180 goto out_nofds;
1181 }
1182 fds.in = (unsigned long *) bits;
1183 fds.out = (unsigned long *) (bits + size);
1184 fds.ex = (unsigned long *) (bits + 2*size);
1185 fds.res_in = (unsigned long *) (bits + 3*size);
1186 fds.res_out = (unsigned long *) (bits + 4*size);
1187 fds.res_ex = (unsigned long *) (bits + 5*size);
1188
1189 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1190 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1191 (ret = compat_get_fd_set(n, exp, fds.ex)))
1192 goto out;
1193 zero_fd_set(n, fds.res_in);
1194 zero_fd_set(n, fds.res_out);
1195 zero_fd_set(n, fds.res_ex);
1196
1197 ret = do_select(n, &fds, end_time);
1198
1199 if (ret < 0)
1200 goto out;
1201 if (!ret) {
1202 ret = -ERESTARTNOHAND;
1203 if (signal_pending(current))
1204 goto out;
1205 ret = 0;
1206 }
1207
1208 if (compat_set_fd_set(n, inp, fds.res_in) ||
1209 compat_set_fd_set(n, outp, fds.res_out) ||
1210 compat_set_fd_set(n, exp, fds.res_ex))
1211 ret = -EFAULT;
1212out:
1213 if (bits != stack_fds)
1214 kfree(bits);
1215out_nofds:
1216 return ret;
1217}
1218
1219COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1220 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1221 struct compat_timeval __user *, tvp)
1222{
1223 struct timespec end_time, *to = NULL;
1224 struct compat_timeval tv;
1225 int ret;
1226
1227 if (tvp) {
1228 if (copy_from_user(&tv, tvp, sizeof(tv)))
1229 return -EFAULT;
1230
1231 to = &end_time;
1232 if (poll_select_set_timeout(to,
1233 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1234 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1235 return -EINVAL;
1236 }
1237
1238 ret = compat_core_sys_select(n, inp, outp, exp, to);
1239 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1240
1241 return ret;
1242}
1243
1244struct compat_sel_arg_struct {
1245 compat_ulong_t n;
1246 compat_uptr_t inp;
1247 compat_uptr_t outp;
1248 compat_uptr_t exp;
1249 compat_uptr_t tvp;
1250};
1251
1252COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1253{
1254 struct compat_sel_arg_struct a;
1255
1256 if (copy_from_user(&a, arg, sizeof(a)))
1257 return -EFAULT;
1258 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1259 compat_ptr(a.exp), compat_ptr(a.tvp));
1260}
1261
1262static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1263 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1264 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1265 compat_size_t sigsetsize)
1266{
1267 compat_sigset_t ss32;
1268 sigset_t ksigmask, sigsaved;
1269 struct compat_timespec ts;
1270 struct timespec end_time, *to = NULL;
1271 int ret;
1272
1273 if (tsp) {
1274 if (copy_from_user(&ts, tsp, sizeof(ts)))
1275 return -EFAULT;
1276
1277 to = &end_time;
1278 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1279 return -EINVAL;
1280 }
1281
1282 if (sigmask) {
1283 if (sigsetsize != sizeof(compat_sigset_t))
1284 return -EINVAL;
1285 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1286 return -EFAULT;
1287 sigset_from_compat(&ksigmask, &ss32);
1288
1289 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1290 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1291 }
1292
1293 ret = compat_core_sys_select(n, inp, outp, exp, to);
1294 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1295
1296 if (ret == -ERESTARTNOHAND) {
1297 /*
1298 * Don't restore the signal mask yet. Let do_signal() deliver
1299 * the signal on the way back to userspace, before the signal
1300 * mask is restored.
1301 */
1302 if (sigmask) {
1303 memcpy(¤t->saved_sigmask, &sigsaved,
1304 sizeof(sigsaved));
1305 set_restore_sigmask();
1306 }
1307 } else if (sigmask)
1308 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1309
1310 return ret;
1311}
1312
1313COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1314 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1315 struct compat_timespec __user *, tsp, void __user *, sig)
1316{
1317 compat_size_t sigsetsize = 0;
1318 compat_uptr_t up = 0;
1319
1320 if (sig) {
1321 if (!access_ok(VERIFY_READ, sig,
1322 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1323 __get_user(up, (compat_uptr_t __user *)sig) ||
1324 __get_user(sigsetsize,
1325 (compat_size_t __user *)(sig+sizeof(up))))
1326 return -EFAULT;
1327 }
1328 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1329 sigsetsize);
1330}
1331
1332COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1333 unsigned int, nfds, struct compat_timespec __user *, tsp,
1334 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1335{
1336 compat_sigset_t ss32;
1337 sigset_t ksigmask, sigsaved;
1338 struct compat_timespec ts;
1339 struct timespec end_time, *to = NULL;
1340 int ret;
1341
1342 if (tsp) {
1343 if (copy_from_user(&ts, tsp, sizeof(ts)))
1344 return -EFAULT;
1345
1346 to = &end_time;
1347 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1348 return -EINVAL;
1349 }
1350
1351 if (sigmask) {
1352 if (sigsetsize != sizeof(compat_sigset_t))
1353 return -EINVAL;
1354 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1355 return -EFAULT;
1356 sigset_from_compat(&ksigmask, &ss32);
1357
1358 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1359 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1360 }
1361
1362 ret = do_sys_poll(ufds, nfds, to);
1363
1364 /* We can restart this syscall, usually */
1365 if (ret == -EINTR) {
1366 /*
1367 * Don't restore the signal mask yet. Let do_signal() deliver
1368 * the signal on the way back to userspace, before the signal
1369 * mask is restored.
1370 */
1371 if (sigmask) {
1372 memcpy(¤t->saved_sigmask, &sigsaved,
1373 sizeof(sigsaved));
1374 set_restore_sigmask();
1375 }
1376 ret = -ERESTARTNOHAND;
1377 } else if (sigmask)
1378 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1379
1380 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1381
1382 return ret;
1383}
1384
1385#ifdef CONFIG_FHANDLE
1386/*
1387 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1388 * doesn't set the O_LARGEFILE flag.
1389 */
1390COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
1391 struct file_handle __user *, handle, int, flags)
1392{
1393 return do_handle_open(mountdirfd, handle, flags);
1394}
1395#endif
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/fs_struct.h>
48#include <linux/slab.h>
49#include <linux/pagemap.h>
50#include <linux/aio.h>
51
52#include <asm/uaccess.h>
53#include <asm/mmu_context.h>
54#include <asm/ioctls.h>
55#include "internal.h"
56
57int compat_log = 1;
58
59int 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/*
72 * Not all architectures have sys_utime, so implement this in terms
73 * of sys_utimes.
74 */
75COMPAT_SYSCALL_DEFINE2(utime, const char __user *, filename,
76 struct compat_utimbuf __user *, t)
77{
78 struct timespec tv[2];
79
80 if (t) {
81 if (get_user(tv[0].tv_sec, &t->actime) ||
82 get_user(tv[1].tv_sec, &t->modtime))
83 return -EFAULT;
84 tv[0].tv_nsec = 0;
85 tv[1].tv_nsec = 0;
86 }
87 return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
88}
89
90COMPAT_SYSCALL_DEFINE4(utimensat, unsigned int, dfd, const char __user *, filename, struct compat_timespec __user *, t, int, flags)
91{
92 struct timespec tv[2];
93
94 if (t) {
95 if (compat_get_timespec(&tv[0], &t[0]) ||
96 compat_get_timespec(&tv[1], &t[1]))
97 return -EFAULT;
98
99 if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
100 return 0;
101 }
102 return do_utimes(dfd, filename, t ? tv : NULL, flags);
103}
104
105COMPAT_SYSCALL_DEFINE3(futimesat, unsigned int, dfd, const char __user *, filename, struct compat_timeval __user *, t)
106{
107 struct timespec tv[2];
108
109 if (t) {
110 if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
111 get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
112 get_user(tv[1].tv_sec, &t[1].tv_sec) ||
113 get_user(tv[1].tv_nsec, &t[1].tv_usec))
114 return -EFAULT;
115 if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
116 tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
117 return -EINVAL;
118 tv[0].tv_nsec *= 1000;
119 tv[1].tv_nsec *= 1000;
120 }
121 return do_utimes(dfd, filename, t ? tv : NULL, 0);
122}
123
124COMPAT_SYSCALL_DEFINE2(utimes, const char __user *, filename, struct compat_timeval __user *, t)
125{
126 return compat_sys_futimesat(AT_FDCWD, filename, t);
127}
128
129static int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
130{
131 struct compat_stat tmp;
132
133 if (!old_valid_dev(stat->dev) || !old_valid_dev(stat->rdev))
134 return -EOVERFLOW;
135
136 memset(&tmp, 0, sizeof(tmp));
137 tmp.st_dev = old_encode_dev(stat->dev);
138 tmp.st_ino = stat->ino;
139 if (sizeof(tmp.st_ino) < sizeof(stat->ino) && tmp.st_ino != stat->ino)
140 return -EOVERFLOW;
141 tmp.st_mode = stat->mode;
142 tmp.st_nlink = stat->nlink;
143 if (tmp.st_nlink != stat->nlink)
144 return -EOVERFLOW;
145 SET_UID(tmp.st_uid, from_kuid_munged(current_user_ns(), stat->uid));
146 SET_GID(tmp.st_gid, from_kgid_munged(current_user_ns(), stat->gid));
147 tmp.st_rdev = old_encode_dev(stat->rdev);
148 if ((u64) stat->size > MAX_NON_LFS)
149 return -EOVERFLOW;
150 tmp.st_size = stat->size;
151 tmp.st_atime = stat->atime.tv_sec;
152 tmp.st_atime_nsec = stat->atime.tv_nsec;
153 tmp.st_mtime = stat->mtime.tv_sec;
154 tmp.st_mtime_nsec = stat->mtime.tv_nsec;
155 tmp.st_ctime = stat->ctime.tv_sec;
156 tmp.st_ctime_nsec = stat->ctime.tv_nsec;
157 tmp.st_blocks = stat->blocks;
158 tmp.st_blksize = stat->blksize;
159 return copy_to_user(ubuf, &tmp, sizeof(tmp)) ? -EFAULT : 0;
160}
161
162COMPAT_SYSCALL_DEFINE2(newstat, const char __user *, filename,
163 struct compat_stat __user *, statbuf)
164{
165 struct kstat stat;
166 int error;
167
168 error = vfs_stat(filename, &stat);
169 if (error)
170 return error;
171 return cp_compat_stat(&stat, statbuf);
172}
173
174COMPAT_SYSCALL_DEFINE2(newlstat, const char __user *, filename,
175 struct compat_stat __user *, statbuf)
176{
177 struct kstat stat;
178 int error;
179
180 error = vfs_lstat(filename, &stat);
181 if (error)
182 return error;
183 return cp_compat_stat(&stat, statbuf);
184}
185
186#ifndef __ARCH_WANT_STAT64
187COMPAT_SYSCALL_DEFINE4(newfstatat, unsigned int, dfd,
188 const char __user *, filename,
189 struct compat_stat __user *, statbuf, int, flag)
190{
191 struct kstat stat;
192 int error;
193
194 error = vfs_fstatat(dfd, filename, &stat, flag);
195 if (error)
196 return error;
197 return cp_compat_stat(&stat, statbuf);
198}
199#endif
200
201COMPAT_SYSCALL_DEFINE2(newfstat, unsigned int, fd,
202 struct compat_stat __user *, statbuf)
203{
204 struct kstat stat;
205 int error = vfs_fstat(fd, &stat);
206
207 if (!error)
208 error = cp_compat_stat(&stat, statbuf);
209 return error;
210}
211
212static int put_compat_statfs(struct compat_statfs __user *ubuf, struct kstatfs *kbuf)
213{
214
215 if (sizeof ubuf->f_blocks == 4) {
216 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
217 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
218 return -EOVERFLOW;
219 /* f_files and f_ffree may be -1; it's okay
220 * to stuff that into 32 bits */
221 if (kbuf->f_files != 0xffffffffffffffffULL
222 && (kbuf->f_files & 0xffffffff00000000ULL))
223 return -EOVERFLOW;
224 if (kbuf->f_ffree != 0xffffffffffffffffULL
225 && (kbuf->f_ffree & 0xffffffff00000000ULL))
226 return -EOVERFLOW;
227 }
228 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
229 __put_user(kbuf->f_type, &ubuf->f_type) ||
230 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
231 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
232 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
233 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
234 __put_user(kbuf->f_files, &ubuf->f_files) ||
235 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
236 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
237 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
238 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
239 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
240 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
241 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
242 return -EFAULT;
243 return 0;
244}
245
246/*
247 * The following statfs calls are copies of code from fs/statfs.c and
248 * should be checked against those from time to time
249 */
250COMPAT_SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct compat_statfs __user *, buf)
251{
252 struct kstatfs tmp;
253 int error = user_statfs(pathname, &tmp);
254 if (!error)
255 error = put_compat_statfs(buf, &tmp);
256 return error;
257}
258
259COMPAT_SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct compat_statfs __user *, buf)
260{
261 struct kstatfs tmp;
262 int error = fd_statfs(fd, &tmp);
263 if (!error)
264 error = put_compat_statfs(buf, &tmp);
265 return error;
266}
267
268static int put_compat_statfs64(struct compat_statfs64 __user *ubuf, struct kstatfs *kbuf)
269{
270 if (sizeof ubuf->f_blocks == 4) {
271 if ((kbuf->f_blocks | kbuf->f_bfree | kbuf->f_bavail |
272 kbuf->f_bsize | kbuf->f_frsize) & 0xffffffff00000000ULL)
273 return -EOVERFLOW;
274 /* f_files and f_ffree may be -1; it's okay
275 * to stuff that into 32 bits */
276 if (kbuf->f_files != 0xffffffffffffffffULL
277 && (kbuf->f_files & 0xffffffff00000000ULL))
278 return -EOVERFLOW;
279 if (kbuf->f_ffree != 0xffffffffffffffffULL
280 && (kbuf->f_ffree & 0xffffffff00000000ULL))
281 return -EOVERFLOW;
282 }
283 if (!access_ok(VERIFY_WRITE, ubuf, sizeof(*ubuf)) ||
284 __put_user(kbuf->f_type, &ubuf->f_type) ||
285 __put_user(kbuf->f_bsize, &ubuf->f_bsize) ||
286 __put_user(kbuf->f_blocks, &ubuf->f_blocks) ||
287 __put_user(kbuf->f_bfree, &ubuf->f_bfree) ||
288 __put_user(kbuf->f_bavail, &ubuf->f_bavail) ||
289 __put_user(kbuf->f_files, &ubuf->f_files) ||
290 __put_user(kbuf->f_ffree, &ubuf->f_ffree) ||
291 __put_user(kbuf->f_namelen, &ubuf->f_namelen) ||
292 __put_user(kbuf->f_fsid.val[0], &ubuf->f_fsid.val[0]) ||
293 __put_user(kbuf->f_fsid.val[1], &ubuf->f_fsid.val[1]) ||
294 __put_user(kbuf->f_frsize, &ubuf->f_frsize) ||
295 __put_user(kbuf->f_flags, &ubuf->f_flags) ||
296 __clear_user(ubuf->f_spare, sizeof(ubuf->f_spare)))
297 return -EFAULT;
298 return 0;
299}
300
301COMPAT_SYSCALL_DEFINE3(statfs64, const char __user *, pathname, compat_size_t, sz, struct compat_statfs64 __user *, buf)
302{
303 struct kstatfs tmp;
304 int error;
305
306 if (sz != sizeof(*buf))
307 return -EINVAL;
308
309 error = user_statfs(pathname, &tmp);
310 if (!error)
311 error = put_compat_statfs64(buf, &tmp);
312 return error;
313}
314
315COMPAT_SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, compat_size_t, sz, struct compat_statfs64 __user *, buf)
316{
317 struct kstatfs tmp;
318 int error;
319
320 if (sz != sizeof(*buf))
321 return -EINVAL;
322
323 error = fd_statfs(fd, &tmp);
324 if (!error)
325 error = put_compat_statfs64(buf, &tmp);
326 return error;
327}
328
329/*
330 * This is a copy of sys_ustat, just dealing with a structure layout.
331 * Given how simple this syscall is that apporach is more maintainable
332 * than the various conversion hacks.
333 */
334COMPAT_SYSCALL_DEFINE2(ustat, unsigned, dev, struct compat_ustat __user *, u)
335{
336 struct compat_ustat tmp;
337 struct kstatfs sbuf;
338 int err = vfs_ustat(new_decode_dev(dev), &sbuf);
339 if (err)
340 return err;
341
342 memset(&tmp, 0, sizeof(struct compat_ustat));
343 tmp.f_tfree = sbuf.f_bfree;
344 tmp.f_tinode = sbuf.f_ffree;
345 if (copy_to_user(u, &tmp, sizeof(struct compat_ustat)))
346 return -EFAULT;
347 return 0;
348}
349
350static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
351{
352 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
353 __get_user(kfl->l_type, &ufl->l_type) ||
354 __get_user(kfl->l_whence, &ufl->l_whence) ||
355 __get_user(kfl->l_start, &ufl->l_start) ||
356 __get_user(kfl->l_len, &ufl->l_len) ||
357 __get_user(kfl->l_pid, &ufl->l_pid))
358 return -EFAULT;
359 return 0;
360}
361
362static int put_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
363{
364 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
365 __put_user(kfl->l_type, &ufl->l_type) ||
366 __put_user(kfl->l_whence, &ufl->l_whence) ||
367 __put_user(kfl->l_start, &ufl->l_start) ||
368 __put_user(kfl->l_len, &ufl->l_len) ||
369 __put_user(kfl->l_pid, &ufl->l_pid))
370 return -EFAULT;
371 return 0;
372}
373
374#ifndef HAVE_ARCH_GET_COMPAT_FLOCK64
375static int get_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
376{
377 if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
378 __get_user(kfl->l_type, &ufl->l_type) ||
379 __get_user(kfl->l_whence, &ufl->l_whence) ||
380 __get_user(kfl->l_start, &ufl->l_start) ||
381 __get_user(kfl->l_len, &ufl->l_len) ||
382 __get_user(kfl->l_pid, &ufl->l_pid))
383 return -EFAULT;
384 return 0;
385}
386#endif
387
388#ifndef HAVE_ARCH_PUT_COMPAT_FLOCK64
389static int put_compat_flock64(struct flock *kfl, struct compat_flock64 __user *ufl)
390{
391 if (!access_ok(VERIFY_WRITE, ufl, sizeof(*ufl)) ||
392 __put_user(kfl->l_type, &ufl->l_type) ||
393 __put_user(kfl->l_whence, &ufl->l_whence) ||
394 __put_user(kfl->l_start, &ufl->l_start) ||
395 __put_user(kfl->l_len, &ufl->l_len) ||
396 __put_user(kfl->l_pid, &ufl->l_pid))
397 return -EFAULT;
398 return 0;
399}
400#endif
401
402static unsigned int
403convert_fcntl_cmd(unsigned int cmd)
404{
405 switch (cmd) {
406 case F_GETLK64:
407 return F_GETLK;
408 case F_SETLK64:
409 return F_SETLK;
410 case F_SETLKW64:
411 return F_SETLKW;
412 }
413
414 return cmd;
415}
416
417COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
418 compat_ulong_t, arg)
419{
420 mm_segment_t old_fs;
421 struct flock f;
422 long ret;
423 unsigned int conv_cmd;
424
425 switch (cmd) {
426 case F_GETLK:
427 case F_SETLK:
428 case F_SETLKW:
429 ret = get_compat_flock(&f, compat_ptr(arg));
430 if (ret != 0)
431 break;
432 old_fs = get_fs();
433 set_fs(KERNEL_DS);
434 ret = sys_fcntl(fd, cmd, (unsigned long)&f);
435 set_fs(old_fs);
436 if (cmd == F_GETLK && ret == 0) {
437 /* GETLK was successful and we need to return the data...
438 * but it needs to fit in the compat structure.
439 * l_start shouldn't be too big, unless the original
440 * start + end is greater than COMPAT_OFF_T_MAX, in which
441 * case the app was asking for trouble, so we return
442 * -EOVERFLOW in that case.
443 * l_len could be too big, in which case we just truncate it,
444 * and only allow the app to see that part of the conflicting
445 * lock that might make sense to it anyway
446 */
447
448 if (f.l_start > COMPAT_OFF_T_MAX)
449 ret = -EOVERFLOW;
450 if (f.l_len > COMPAT_OFF_T_MAX)
451 f.l_len = COMPAT_OFF_T_MAX;
452 if (ret == 0)
453 ret = put_compat_flock(&f, compat_ptr(arg));
454 }
455 break;
456
457 case F_GETLK64:
458 case F_SETLK64:
459 case F_SETLKW64:
460 case F_OFD_GETLK:
461 case F_OFD_SETLK:
462 case F_OFD_SETLKW:
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 conv_cmd = convert_fcntl_cmd(cmd);
469 ret = sys_fcntl(fd, conv_cmd, (unsigned long)&f);
470 set_fs(old_fs);
471 if ((conv_cmd == F_GETLK || conv_cmd == F_OFD_GETLK) && ret == 0) {
472 /* need to return lock information - see above for commentary */
473 if (f.l_start > COMPAT_LOFF_T_MAX)
474 ret = -EOVERFLOW;
475 if (f.l_len > COMPAT_LOFF_T_MAX)
476 f.l_len = COMPAT_LOFF_T_MAX;
477 if (ret == 0)
478 ret = put_compat_flock64(&f, compat_ptr(arg));
479 }
480 break;
481
482 default:
483 ret = sys_fcntl(fd, cmd, arg);
484 break;
485 }
486 return ret;
487}
488
489COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
490 compat_ulong_t, arg)
491{
492 switch (cmd) {
493 case F_GETLK64:
494 case F_SETLK64:
495 case F_SETLKW64:
496 case F_OFD_GETLK:
497 case F_OFD_SETLK:
498 case F_OFD_SETLKW:
499 return -EINVAL;
500 }
501 return compat_sys_fcntl64(fd, cmd, arg);
502}
503
504COMPAT_SYSCALL_DEFINE2(io_setup, unsigned, nr_reqs, u32 __user *, ctx32p)
505{
506 long ret;
507 aio_context_t ctx64;
508
509 mm_segment_t oldfs = get_fs();
510 if (unlikely(get_user(ctx64, ctx32p)))
511 return -EFAULT;
512
513 set_fs(KERNEL_DS);
514 /* The __user pointer cast is valid because of the set_fs() */
515 ret = sys_io_setup(nr_reqs, (aio_context_t __user *) &ctx64);
516 set_fs(oldfs);
517 /* truncating is ok because it's a user address */
518 if (!ret)
519 ret = put_user((u32) ctx64, ctx32p);
520 return ret;
521}
522
523COMPAT_SYSCALL_DEFINE5(io_getevents, compat_aio_context_t, ctx_id,
524 compat_long_t, min_nr,
525 compat_long_t, nr,
526 struct io_event __user *, events,
527 struct compat_timespec __user *, timeout)
528{
529 struct timespec t;
530 struct timespec __user *ut = NULL;
531
532 if (timeout) {
533 if (compat_get_timespec(&t, timeout))
534 return -EFAULT;
535
536 ut = compat_alloc_user_space(sizeof(*ut));
537 if (copy_to_user(ut, &t, sizeof(t)) )
538 return -EFAULT;
539 }
540 return sys_io_getevents(ctx_id, min_nr, nr, events, ut);
541}
542
543/* A write operation does a read from user space and vice versa */
544#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
545
546ssize_t compat_rw_copy_check_uvector(int type,
547 const struct compat_iovec __user *uvector, unsigned long nr_segs,
548 unsigned long fast_segs, struct iovec *fast_pointer,
549 struct iovec **ret_pointer)
550{
551 compat_ssize_t tot_len;
552 struct iovec *iov = *ret_pointer = fast_pointer;
553 ssize_t ret = 0;
554 int seg;
555
556 /*
557 * SuS says "The readv() function *may* fail if the iovcnt argument
558 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
559 * traditionally returned zero for zero segments, so...
560 */
561 if (nr_segs == 0)
562 goto out;
563
564 ret = -EINVAL;
565 if (nr_segs > UIO_MAXIOV || nr_segs < 0)
566 goto out;
567 if (nr_segs > fast_segs) {
568 ret = -ENOMEM;
569 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
570 if (iov == NULL)
571 goto out;
572 }
573 *ret_pointer = iov;
574
575 ret = -EFAULT;
576 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
577 goto out;
578
579 /*
580 * Single unix specification:
581 * We should -EINVAL if an element length is not >= 0 and fitting an
582 * ssize_t.
583 *
584 * In Linux, the total length is limited to MAX_RW_COUNT, there is
585 * no overflow possibility.
586 */
587 tot_len = 0;
588 ret = -EINVAL;
589 for (seg = 0; seg < nr_segs; seg++) {
590 compat_uptr_t buf;
591 compat_ssize_t len;
592
593 if (__get_user(len, &uvector->iov_len) ||
594 __get_user(buf, &uvector->iov_base)) {
595 ret = -EFAULT;
596 goto out;
597 }
598 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
599 goto out;
600 if (type >= 0 &&
601 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
602 ret = -EFAULT;
603 goto out;
604 }
605 if (len > MAX_RW_COUNT - tot_len)
606 len = MAX_RW_COUNT - tot_len;
607 tot_len += len;
608 iov->iov_base = compat_ptr(buf);
609 iov->iov_len = (compat_size_t) len;
610 uvector++;
611 iov++;
612 }
613 ret = tot_len;
614
615out:
616 return ret;
617}
618
619static inline long
620copy_iocb(long nr, u32 __user *ptr32, struct iocb __user * __user *ptr64)
621{
622 compat_uptr_t uptr;
623 int i;
624
625 for (i = 0; i < nr; ++i) {
626 if (get_user(uptr, ptr32 + i))
627 return -EFAULT;
628 if (put_user(compat_ptr(uptr), ptr64 + i))
629 return -EFAULT;
630 }
631 return 0;
632}
633
634#define MAX_AIO_SUBMITS (PAGE_SIZE/sizeof(struct iocb *))
635
636COMPAT_SYSCALL_DEFINE3(io_submit, compat_aio_context_t, ctx_id,
637 int, nr, u32 __user *, iocb)
638{
639 struct iocb __user * __user *iocb64;
640 long ret;
641
642 if (unlikely(nr < 0))
643 return -EINVAL;
644
645 if (nr > MAX_AIO_SUBMITS)
646 nr = MAX_AIO_SUBMITS;
647
648 iocb64 = compat_alloc_user_space(nr * sizeof(*iocb64));
649 ret = copy_iocb(nr, iocb, iocb64);
650 if (!ret)
651 ret = do_io_submit(ctx_id, nr, iocb64, 1);
652 return ret;
653}
654
655struct compat_ncp_mount_data {
656 compat_int_t version;
657 compat_uint_t ncp_fd;
658 __compat_uid_t mounted_uid;
659 compat_pid_t wdog_pid;
660 unsigned char mounted_vol[NCP_VOLNAME_LEN + 1];
661 compat_uint_t time_out;
662 compat_uint_t retry_count;
663 compat_uint_t flags;
664 __compat_uid_t uid;
665 __compat_gid_t gid;
666 compat_mode_t file_mode;
667 compat_mode_t dir_mode;
668};
669
670struct compat_ncp_mount_data_v4 {
671 compat_int_t version;
672 compat_ulong_t flags;
673 compat_ulong_t mounted_uid;
674 compat_long_t wdog_pid;
675 compat_uint_t ncp_fd;
676 compat_uint_t time_out;
677 compat_uint_t retry_count;
678 compat_ulong_t uid;
679 compat_ulong_t gid;
680 compat_ulong_t file_mode;
681 compat_ulong_t dir_mode;
682};
683
684static void *do_ncp_super_data_conv(void *raw_data)
685{
686 int version = *(unsigned int *)raw_data;
687
688 if (version == 3) {
689 struct compat_ncp_mount_data *c_n = raw_data;
690 struct ncp_mount_data *n = raw_data;
691
692 n->dir_mode = c_n->dir_mode;
693 n->file_mode = c_n->file_mode;
694 n->gid = c_n->gid;
695 n->uid = c_n->uid;
696 memmove (n->mounted_vol, c_n->mounted_vol, (sizeof (c_n->mounted_vol) + 3 * sizeof (unsigned int)));
697 n->wdog_pid = c_n->wdog_pid;
698 n->mounted_uid = c_n->mounted_uid;
699 } else if (version == 4) {
700 struct compat_ncp_mount_data_v4 *c_n = raw_data;
701 struct ncp_mount_data_v4 *n = raw_data;
702
703 n->dir_mode = c_n->dir_mode;
704 n->file_mode = c_n->file_mode;
705 n->gid = c_n->gid;
706 n->uid = c_n->uid;
707 n->retry_count = c_n->retry_count;
708 n->time_out = c_n->time_out;
709 n->ncp_fd = c_n->ncp_fd;
710 n->wdog_pid = c_n->wdog_pid;
711 n->mounted_uid = c_n->mounted_uid;
712 n->flags = c_n->flags;
713 } else if (version != 5) {
714 return NULL;
715 }
716
717 return raw_data;
718}
719
720
721struct compat_nfs_string {
722 compat_uint_t len;
723 compat_uptr_t data;
724};
725
726static inline void compat_nfs_string(struct nfs_string *dst,
727 struct compat_nfs_string *src)
728{
729 dst->data = compat_ptr(src->data);
730 dst->len = src->len;
731}
732
733struct compat_nfs4_mount_data_v1 {
734 compat_int_t version;
735 compat_int_t flags;
736 compat_int_t rsize;
737 compat_int_t wsize;
738 compat_int_t timeo;
739 compat_int_t retrans;
740 compat_int_t acregmin;
741 compat_int_t acregmax;
742 compat_int_t acdirmin;
743 compat_int_t acdirmax;
744 struct compat_nfs_string client_addr;
745 struct compat_nfs_string mnt_path;
746 struct compat_nfs_string hostname;
747 compat_uint_t host_addrlen;
748 compat_uptr_t host_addr;
749 compat_int_t proto;
750 compat_int_t auth_flavourlen;
751 compat_uptr_t auth_flavours;
752};
753
754static int do_nfs4_super_data_conv(void *raw_data)
755{
756 int version = *(compat_uint_t *) raw_data;
757
758 if (version == 1) {
759 struct compat_nfs4_mount_data_v1 *raw = raw_data;
760 struct nfs4_mount_data *real = raw_data;
761
762 /* copy the fields backwards */
763 real->auth_flavours = compat_ptr(raw->auth_flavours);
764 real->auth_flavourlen = raw->auth_flavourlen;
765 real->proto = raw->proto;
766 real->host_addr = compat_ptr(raw->host_addr);
767 real->host_addrlen = raw->host_addrlen;
768 compat_nfs_string(&real->hostname, &raw->hostname);
769 compat_nfs_string(&real->mnt_path, &raw->mnt_path);
770 compat_nfs_string(&real->client_addr, &raw->client_addr);
771 real->acdirmax = raw->acdirmax;
772 real->acdirmin = raw->acdirmin;
773 real->acregmax = raw->acregmax;
774 real->acregmin = raw->acregmin;
775 real->retrans = raw->retrans;
776 real->timeo = raw->timeo;
777 real->wsize = raw->wsize;
778 real->rsize = raw->rsize;
779 real->flags = raw->flags;
780 real->version = raw->version;
781 }
782
783 return 0;
784}
785
786#define NCPFS_NAME "ncpfs"
787#define NFS4_NAME "nfs4"
788
789COMPAT_SYSCALL_DEFINE5(mount, const char __user *, dev_name,
790 const char __user *, dir_name,
791 const char __user *, type, compat_ulong_t, flags,
792 const void __user *, data)
793{
794 char *kernel_type;
795 void *options;
796 char *kernel_dev;
797 int retval;
798
799 kernel_type = copy_mount_string(type);
800 retval = PTR_ERR(kernel_type);
801 if (IS_ERR(kernel_type))
802 goto out;
803
804 kernel_dev = copy_mount_string(dev_name);
805 retval = PTR_ERR(kernel_dev);
806 if (IS_ERR(kernel_dev))
807 goto out1;
808
809 options = copy_mount_options(data);
810 retval = PTR_ERR(options);
811 if (IS_ERR(options))
812 goto out2;
813
814 if (kernel_type && options) {
815 if (!strcmp(kernel_type, NCPFS_NAME)) {
816 do_ncp_super_data_conv(options);
817 } else if (!strcmp(kernel_type, NFS4_NAME)) {
818 retval = -EINVAL;
819 if (do_nfs4_super_data_conv(options))
820 goto out3;
821 }
822 }
823
824 retval = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
825
826 out3:
827 kfree(options);
828 out2:
829 kfree(kernel_dev);
830 out1:
831 kfree(kernel_type);
832 out:
833 return retval;
834}
835
836struct compat_old_linux_dirent {
837 compat_ulong_t d_ino;
838 compat_ulong_t d_offset;
839 unsigned short d_namlen;
840 char d_name[1];
841};
842
843struct compat_readdir_callback {
844 struct dir_context ctx;
845 struct compat_old_linux_dirent __user *dirent;
846 int result;
847};
848
849static int compat_fillonedir(struct dir_context *ctx, const char *name,
850 int namlen, loff_t offset, u64 ino,
851 unsigned int d_type)
852{
853 struct compat_readdir_callback *buf =
854 container_of(ctx, struct compat_readdir_callback, ctx);
855 struct compat_old_linux_dirent __user *dirent;
856 compat_ulong_t d_ino;
857
858 if (buf->result)
859 return -EINVAL;
860 d_ino = ino;
861 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
862 buf->result = -EOVERFLOW;
863 return -EOVERFLOW;
864 }
865 buf->result++;
866 dirent = buf->dirent;
867 if (!access_ok(VERIFY_WRITE, dirent,
868 (unsigned long)(dirent->d_name + namlen + 1) -
869 (unsigned long)dirent))
870 goto efault;
871 if ( __put_user(d_ino, &dirent->d_ino) ||
872 __put_user(offset, &dirent->d_offset) ||
873 __put_user(namlen, &dirent->d_namlen) ||
874 __copy_to_user(dirent->d_name, name, namlen) ||
875 __put_user(0, dirent->d_name + namlen))
876 goto efault;
877 return 0;
878efault:
879 buf->result = -EFAULT;
880 return -EFAULT;
881}
882
883COMPAT_SYSCALL_DEFINE3(old_readdir, unsigned int, fd,
884 struct compat_old_linux_dirent __user *, dirent, unsigned int, count)
885{
886 int error;
887 struct fd f = fdget(fd);
888 struct compat_readdir_callback buf = {
889 .ctx.actor = compat_fillonedir,
890 .dirent = dirent
891 };
892
893 if (!f.file)
894 return -EBADF;
895
896 error = iterate_dir(f.file, &buf.ctx);
897 if (buf.result)
898 error = buf.result;
899
900 fdput(f);
901 return error;
902}
903
904struct compat_linux_dirent {
905 compat_ulong_t d_ino;
906 compat_ulong_t d_off;
907 unsigned short d_reclen;
908 char d_name[1];
909};
910
911struct compat_getdents_callback {
912 struct dir_context ctx;
913 struct compat_linux_dirent __user *current_dir;
914 struct compat_linux_dirent __user *previous;
915 int count;
916 int error;
917};
918
919static int compat_filldir(struct dir_context *ctx, const char *name, int namlen,
920 loff_t offset, u64 ino, unsigned int d_type)
921{
922 struct compat_linux_dirent __user * dirent;
923 struct compat_getdents_callback *buf =
924 container_of(ctx, struct compat_getdents_callback, ctx);
925 compat_ulong_t d_ino;
926 int reclen = ALIGN(offsetof(struct compat_linux_dirent, d_name) +
927 namlen + 2, sizeof(compat_long_t));
928
929 buf->error = -EINVAL; /* only used if we fail.. */
930 if (reclen > buf->count)
931 return -EINVAL;
932 d_ino = ino;
933 if (sizeof(d_ino) < sizeof(ino) && d_ino != ino) {
934 buf->error = -EOVERFLOW;
935 return -EOVERFLOW;
936 }
937 dirent = buf->previous;
938 if (dirent) {
939 if (__put_user(offset, &dirent->d_off))
940 goto efault;
941 }
942 dirent = buf->current_dir;
943 if (__put_user(d_ino, &dirent->d_ino))
944 goto efault;
945 if (__put_user(reclen, &dirent->d_reclen))
946 goto efault;
947 if (copy_to_user(dirent->d_name, name, namlen))
948 goto efault;
949 if (__put_user(0, dirent->d_name + namlen))
950 goto efault;
951 if (__put_user(d_type, (char __user *) dirent + reclen - 1))
952 goto efault;
953 buf->previous = dirent;
954 dirent = (void __user *)dirent + reclen;
955 buf->current_dir = dirent;
956 buf->count -= reclen;
957 return 0;
958efault:
959 buf->error = -EFAULT;
960 return -EFAULT;
961}
962
963COMPAT_SYSCALL_DEFINE3(getdents, unsigned int, fd,
964 struct compat_linux_dirent __user *, dirent, unsigned int, count)
965{
966 struct fd f;
967 struct compat_linux_dirent __user * lastdirent;
968 struct compat_getdents_callback buf = {
969 .ctx.actor = compat_filldir,
970 .current_dir = dirent,
971 .count = count
972 };
973 int error;
974
975 if (!access_ok(VERIFY_WRITE, dirent, count))
976 return -EFAULT;
977
978 f = fdget(fd);
979 if (!f.file)
980 return -EBADF;
981
982 error = iterate_dir(f.file, &buf.ctx);
983 if (error >= 0)
984 error = buf.error;
985 lastdirent = buf.previous;
986 if (lastdirent) {
987 if (put_user(buf.ctx.pos, &lastdirent->d_off))
988 error = -EFAULT;
989 else
990 error = count - buf.count;
991 }
992 fdput(f);
993 return error;
994}
995
996#ifdef __ARCH_WANT_COMPAT_SYS_GETDENTS64
997
998struct compat_getdents_callback64 {
999 struct dir_context ctx;
1000 struct linux_dirent64 __user *current_dir;
1001 struct linux_dirent64 __user *previous;
1002 int count;
1003 int error;
1004};
1005
1006static int compat_filldir64(struct dir_context *ctx, const char *name,
1007 int namlen, loff_t offset, u64 ino,
1008 unsigned int d_type)
1009{
1010 struct linux_dirent64 __user *dirent;
1011 struct compat_getdents_callback64 *buf =
1012 container_of(ctx, struct compat_getdents_callback64, ctx);
1013 int reclen = ALIGN(offsetof(struct linux_dirent64, d_name) + namlen + 1,
1014 sizeof(u64));
1015 u64 off;
1016
1017 buf->error = -EINVAL; /* only used if we fail.. */
1018 if (reclen > buf->count)
1019 return -EINVAL;
1020 dirent = buf->previous;
1021
1022 if (dirent) {
1023 if (__put_user_unaligned(offset, &dirent->d_off))
1024 goto efault;
1025 }
1026 dirent = buf->current_dir;
1027 if (__put_user_unaligned(ino, &dirent->d_ino))
1028 goto efault;
1029 off = 0;
1030 if (__put_user_unaligned(off, &dirent->d_off))
1031 goto efault;
1032 if (__put_user(reclen, &dirent->d_reclen))
1033 goto efault;
1034 if (__put_user(d_type, &dirent->d_type))
1035 goto efault;
1036 if (copy_to_user(dirent->d_name, name, namlen))
1037 goto efault;
1038 if (__put_user(0, dirent->d_name + namlen))
1039 goto efault;
1040 buf->previous = dirent;
1041 dirent = (void __user *)dirent + reclen;
1042 buf->current_dir = dirent;
1043 buf->count -= reclen;
1044 return 0;
1045efault:
1046 buf->error = -EFAULT;
1047 return -EFAULT;
1048}
1049
1050COMPAT_SYSCALL_DEFINE3(getdents64, unsigned int, fd,
1051 struct linux_dirent64 __user *, dirent, unsigned int, count)
1052{
1053 struct fd f;
1054 struct linux_dirent64 __user * lastdirent;
1055 struct compat_getdents_callback64 buf = {
1056 .ctx.actor = compat_filldir64,
1057 .current_dir = dirent,
1058 .count = count
1059 };
1060 int error;
1061
1062 if (!access_ok(VERIFY_WRITE, dirent, count))
1063 return -EFAULT;
1064
1065 f = fdget(fd);
1066 if (!f.file)
1067 return -EBADF;
1068
1069 error = iterate_dir(f.file, &buf.ctx);
1070 if (error >= 0)
1071 error = buf.error;
1072 lastdirent = buf.previous;
1073 if (lastdirent) {
1074 typeof(lastdirent->d_off) d_off = buf.ctx.pos;
1075 if (__put_user_unaligned(d_off, &lastdirent->d_off))
1076 error = -EFAULT;
1077 else
1078 error = count - buf.count;
1079 }
1080 fdput(f);
1081 return error;
1082}
1083#endif /* __ARCH_WANT_COMPAT_SYS_GETDENTS64 */
1084
1085/*
1086 * Exactly like fs/open.c:sys_open(), except that it doesn't set the
1087 * O_LARGEFILE flag.
1088 */
1089COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1090{
1091 return do_sys_open(AT_FDCWD, filename, flags, mode);
1092}
1093
1094/*
1095 * Exactly like fs/open.c:sys_openat(), except that it doesn't set the
1096 * O_LARGEFILE flag.
1097 */
1098COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1099{
1100 return do_sys_open(dfd, filename, flags, mode);
1101}
1102
1103#define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1104
1105static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
1106 int timeval, int ret)
1107{
1108 struct timespec ts;
1109
1110 if (!p)
1111 return ret;
1112
1113 if (current->personality & STICKY_TIMEOUTS)
1114 goto sticky;
1115
1116 /* No update for zero timeout */
1117 if (!end_time->tv_sec && !end_time->tv_nsec)
1118 return ret;
1119
1120 ktime_get_ts(&ts);
1121 ts = timespec_sub(*end_time, ts);
1122 if (ts.tv_sec < 0)
1123 ts.tv_sec = ts.tv_nsec = 0;
1124
1125 if (timeval) {
1126 struct compat_timeval rtv;
1127
1128 rtv.tv_sec = ts.tv_sec;
1129 rtv.tv_usec = ts.tv_nsec / NSEC_PER_USEC;
1130
1131 if (!copy_to_user(p, &rtv, sizeof(rtv)))
1132 return ret;
1133 } else {
1134 struct compat_timespec rts;
1135
1136 rts.tv_sec = ts.tv_sec;
1137 rts.tv_nsec = ts.tv_nsec;
1138
1139 if (!copy_to_user(p, &rts, sizeof(rts)))
1140 return ret;
1141 }
1142 /*
1143 * If an application puts its timeval in read-only memory, we
1144 * don't want the Linux-specific update to the timeval to
1145 * cause a fault after the select has completed
1146 * successfully. However, because we're not updating the
1147 * timeval, we can't restart the system call.
1148 */
1149
1150sticky:
1151 if (ret == -ERESTARTNOHAND)
1152 ret = -EINTR;
1153 return ret;
1154}
1155
1156/*
1157 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1158 * 64-bit unsigned longs.
1159 */
1160static
1161int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1162 unsigned long *fdset)
1163{
1164 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1165 if (ufdset) {
1166 unsigned long odd;
1167
1168 if (!access_ok(VERIFY_WRITE, ufdset, nr*sizeof(compat_ulong_t)))
1169 return -EFAULT;
1170
1171 odd = nr & 1UL;
1172 nr &= ~1UL;
1173 while (nr) {
1174 unsigned long h, l;
1175 if (__get_user(l, ufdset) || __get_user(h, ufdset+1))
1176 return -EFAULT;
1177 ufdset += 2;
1178 *fdset++ = h << 32 | l;
1179 nr -= 2;
1180 }
1181 if (odd && __get_user(*fdset, ufdset))
1182 return -EFAULT;
1183 } else {
1184 /* Tricky, must clear full unsigned long in the
1185 * kernel fdset at the end, this makes sure that
1186 * actually happens.
1187 */
1188 memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
1189 }
1190 return 0;
1191}
1192
1193static
1194int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1195 unsigned long *fdset)
1196{
1197 unsigned long odd;
1198 nr = DIV_ROUND_UP(nr, __COMPAT_NFDBITS);
1199
1200 if (!ufdset)
1201 return 0;
1202
1203 odd = nr & 1UL;
1204 nr &= ~1UL;
1205 while (nr) {
1206 unsigned long h, l;
1207 l = *fdset++;
1208 h = l >> 32;
1209 if (__put_user(l, ufdset) || __put_user(h, ufdset+1))
1210 return -EFAULT;
1211 ufdset += 2;
1212 nr -= 2;
1213 }
1214 if (odd && __put_user(*fdset, ufdset))
1215 return -EFAULT;
1216 return 0;
1217}
1218
1219
1220/*
1221 * This is a virtual copy of sys_select from fs/select.c and probably
1222 * should be compared to it from time to time
1223 */
1224
1225/*
1226 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1227 * like to be certain this leads to no problems. So I return
1228 * EINTR just for safety.
1229 *
1230 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1231 * I'm trying ERESTARTNOHAND which restart only when you want to.
1232 */
1233int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1234 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1235 struct timespec *end_time)
1236{
1237 fd_set_bits fds;
1238 void *bits;
1239 int size, max_fds, ret = -EINVAL;
1240 struct fdtable *fdt;
1241 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1242
1243 if (n < 0)
1244 goto out_nofds;
1245
1246 /* max_fds can increase, so grab it once to avoid race */
1247 rcu_read_lock();
1248 fdt = files_fdtable(current->files);
1249 max_fds = fdt->max_fds;
1250 rcu_read_unlock();
1251 if (n > max_fds)
1252 n = max_fds;
1253
1254 /*
1255 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1256 * since we used fdset we need to allocate memory in units of
1257 * long-words.
1258 */
1259 size = FDS_BYTES(n);
1260 bits = stack_fds;
1261 if (size > sizeof(stack_fds) / 6) {
1262 bits = kmalloc(6 * size, GFP_KERNEL);
1263 ret = -ENOMEM;
1264 if (!bits)
1265 goto out_nofds;
1266 }
1267 fds.in = (unsigned long *) bits;
1268 fds.out = (unsigned long *) (bits + size);
1269 fds.ex = (unsigned long *) (bits + 2*size);
1270 fds.res_in = (unsigned long *) (bits + 3*size);
1271 fds.res_out = (unsigned long *) (bits + 4*size);
1272 fds.res_ex = (unsigned long *) (bits + 5*size);
1273
1274 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1275 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1276 (ret = compat_get_fd_set(n, exp, fds.ex)))
1277 goto out;
1278 zero_fd_set(n, fds.res_in);
1279 zero_fd_set(n, fds.res_out);
1280 zero_fd_set(n, fds.res_ex);
1281
1282 ret = do_select(n, &fds, end_time);
1283
1284 if (ret < 0)
1285 goto out;
1286 if (!ret) {
1287 ret = -ERESTARTNOHAND;
1288 if (signal_pending(current))
1289 goto out;
1290 ret = 0;
1291 }
1292
1293 if (compat_set_fd_set(n, inp, fds.res_in) ||
1294 compat_set_fd_set(n, outp, fds.res_out) ||
1295 compat_set_fd_set(n, exp, fds.res_ex))
1296 ret = -EFAULT;
1297out:
1298 if (bits != stack_fds)
1299 kfree(bits);
1300out_nofds:
1301 return ret;
1302}
1303
1304COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1305 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1306 struct compat_timeval __user *, tvp)
1307{
1308 struct timespec end_time, *to = NULL;
1309 struct compat_timeval tv;
1310 int ret;
1311
1312 if (tvp) {
1313 if (copy_from_user(&tv, tvp, sizeof(tv)))
1314 return -EFAULT;
1315
1316 to = &end_time;
1317 if (poll_select_set_timeout(to,
1318 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1319 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1320 return -EINVAL;
1321 }
1322
1323 ret = compat_core_sys_select(n, inp, outp, exp, to);
1324 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
1325
1326 return ret;
1327}
1328
1329struct compat_sel_arg_struct {
1330 compat_ulong_t n;
1331 compat_uptr_t inp;
1332 compat_uptr_t outp;
1333 compat_uptr_t exp;
1334 compat_uptr_t tvp;
1335};
1336
1337COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1338{
1339 struct compat_sel_arg_struct a;
1340
1341 if (copy_from_user(&a, arg, sizeof(a)))
1342 return -EFAULT;
1343 return compat_sys_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1344 compat_ptr(a.exp), compat_ptr(a.tvp));
1345}
1346
1347static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1348 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1349 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1350 compat_size_t sigsetsize)
1351{
1352 compat_sigset_t ss32;
1353 sigset_t ksigmask, sigsaved;
1354 struct compat_timespec ts;
1355 struct timespec end_time, *to = NULL;
1356 int ret;
1357
1358 if (tsp) {
1359 if (copy_from_user(&ts, tsp, sizeof(ts)))
1360 return -EFAULT;
1361
1362 to = &end_time;
1363 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1364 return -EINVAL;
1365 }
1366
1367 if (sigmask) {
1368 if (sigsetsize != sizeof(compat_sigset_t))
1369 return -EINVAL;
1370 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1371 return -EFAULT;
1372 sigset_from_compat(&ksigmask, &ss32);
1373
1374 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1375 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1376 }
1377
1378 ret = compat_core_sys_select(n, inp, outp, exp, to);
1379 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1380
1381 if (ret == -ERESTARTNOHAND) {
1382 /*
1383 * Don't restore the signal mask yet. Let do_signal() deliver
1384 * the signal on the way back to userspace, before the signal
1385 * mask is restored.
1386 */
1387 if (sigmask) {
1388 memcpy(¤t->saved_sigmask, &sigsaved,
1389 sizeof(sigsaved));
1390 set_restore_sigmask();
1391 }
1392 } else if (sigmask)
1393 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1394
1395 return ret;
1396}
1397
1398COMPAT_SYSCALL_DEFINE6(pselect6, int, n, compat_ulong_t __user *, inp,
1399 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1400 struct compat_timespec __user *, tsp, void __user *, sig)
1401{
1402 compat_size_t sigsetsize = 0;
1403 compat_uptr_t up = 0;
1404
1405 if (sig) {
1406 if (!access_ok(VERIFY_READ, sig,
1407 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1408 __get_user(up, (compat_uptr_t __user *)sig) ||
1409 __get_user(sigsetsize,
1410 (compat_size_t __user *)(sig+sizeof(up))))
1411 return -EFAULT;
1412 }
1413 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1414 sigsetsize);
1415}
1416
1417COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds,
1418 unsigned int, nfds, struct compat_timespec __user *, tsp,
1419 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1420{
1421 compat_sigset_t ss32;
1422 sigset_t ksigmask, sigsaved;
1423 struct compat_timespec ts;
1424 struct timespec end_time, *to = NULL;
1425 int ret;
1426
1427 if (tsp) {
1428 if (copy_from_user(&ts, tsp, sizeof(ts)))
1429 return -EFAULT;
1430
1431 to = &end_time;
1432 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1433 return -EINVAL;
1434 }
1435
1436 if (sigmask) {
1437 if (sigsetsize != sizeof(compat_sigset_t))
1438 return -EINVAL;
1439 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1440 return -EFAULT;
1441 sigset_from_compat(&ksigmask, &ss32);
1442
1443 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1444 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1445 }
1446
1447 ret = do_sys_poll(ufds, nfds, to);
1448
1449 /* We can restart this syscall, usually */
1450 if (ret == -EINTR) {
1451 /*
1452 * Don't restore the signal mask yet. Let do_signal() deliver
1453 * the signal on the way back to userspace, before the signal
1454 * mask is restored.
1455 */
1456 if (sigmask) {
1457 memcpy(¤t->saved_sigmask, &sigsaved,
1458 sizeof(sigsaved));
1459 set_restore_sigmask();
1460 }
1461 ret = -ERESTARTNOHAND;
1462 } else if (sigmask)
1463 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1464
1465 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
1466
1467 return ret;
1468}
1469
1470#ifdef CONFIG_FHANDLE
1471/*
1472 * Exactly like fs/open.c:sys_open_by_handle_at(), except that it
1473 * doesn't set the O_LARGEFILE flag.
1474 */
1475COMPAT_SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
1476 struct file_handle __user *, handle, int, flags)
1477{
1478 return do_handle_open(mountdirfd, handle, flags);
1479}
1480#endif