2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/smp_lock.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/module.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include "read_write.h"
21 #include <asm/uaccess.h>
22 #include <asm/unistd.h>
24 const struct file_operations generic_ro_fops = {
25 .llseek = generic_file_llseek,
27 .aio_read = generic_file_aio_read,
28 .mmap = generic_file_readonly_mmap,
29 .splice_read = generic_file_splice_read,
32 EXPORT_SYMBOL(generic_ro_fops);
35 * generic_file_llseek_unlocked - lockless generic llseek implementation
36 * @file: file structure to seek on
37 * @offset: file offset to seek to
38 * @origin: type of seek
40 * Updates the file offset to the value specified by @offset and @origin.
41 * Locking must be provided by the caller.
44 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
46 struct inode *inode = file->f_mapping->host;
50 offset += inode->i_size;
54 * Here we special-case the lseek(fd, 0, SEEK_CUR)
55 * position-querying operation. Avoid rewriting the "same"
56 * f_pos value back to the file because a concurrent read(),
57 * write() or lseek() might have altered it
61 offset += file->f_pos;
65 if (offset < 0 || offset > inode->i_sb->s_maxbytes)
68 /* Special lock needed here? */
69 if (offset != file->f_pos) {
76 EXPORT_SYMBOL(generic_file_llseek_unlocked);
79 * generic_file_llseek - generic llseek implementation for regular files
80 * @file: file structure to seek on
81 * @offset: file offset to seek to
82 * @origin: type of seek
84 * This is a generic implemenation of ->llseek useable for all normal local
85 * filesystems. It just updates the file offset to the value specified by
86 * @offset and @origin under i_mutex.
88 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
92 mutex_lock(&file->f_dentry->d_inode->i_mutex);
93 rval = generic_file_llseek_unlocked(file, offset, origin);
94 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
98 EXPORT_SYMBOL(generic_file_llseek);
100 loff_t no_llseek(struct file *file, loff_t offset, int origin)
104 EXPORT_SYMBOL(no_llseek);
106 loff_t default_llseek(struct file *file, loff_t offset, int origin)
113 offset += i_size_read(file->f_path.dentry->d_inode);
117 retval = file->f_pos;
120 offset += file->f_pos;
124 if (offset != file->f_pos) {
125 file->f_pos = offset;
134 EXPORT_SYMBOL(default_llseek);
136 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
138 loff_t (*fn)(struct file *, loff_t, int);
141 if (file->f_mode & FMODE_LSEEK) {
143 if (file->f_op && file->f_op->llseek)
144 fn = file->f_op->llseek;
146 return fn(file, offset, origin);
148 EXPORT_SYMBOL(vfs_llseek);
150 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
157 file = fget_light(fd, &fput_needed);
162 if (origin <= SEEK_MAX) {
163 loff_t res = vfs_llseek(file, offset, origin);
165 if (res != (loff_t)retval)
166 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
168 fput_light(file, fput_needed);
173 #ifdef __ARCH_WANT_SYS_LLSEEK
174 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
175 unsigned long, offset_low, loff_t __user *, result,
176 unsigned int, origin)
184 file = fget_light(fd, &fput_needed);
189 if (origin > SEEK_MAX)
192 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
195 retval = (int)offset;
198 if (!copy_to_user(result, &offset, sizeof(offset)))
202 fput_light(file, fput_needed);
209 * rw_verify_area doesn't like huge counts. We limit
210 * them to something that fits in "int" so that others
211 * won't have to do range checks all the time.
213 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
215 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
219 int retval = -EINVAL;
221 inode = file->f_path.dentry->d_inode;
222 if (unlikely((ssize_t) count < 0))
225 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
228 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
229 retval = locks_mandatory_area(
230 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
231 inode, file, pos, count);
235 retval = security_file_permission(file,
236 read_write == READ ? MAY_READ : MAY_WRITE);
239 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
242 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
244 set_current_state(TASK_UNINTERRUPTIBLE);
245 if (!kiocbIsKicked(iocb))
248 kiocbClearKicked(iocb);
249 __set_current_state(TASK_RUNNING);
252 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
254 struct iovec iov = { .iov_base = buf, .iov_len = len };
258 init_sync_kiocb(&kiocb, filp);
259 kiocb.ki_pos = *ppos;
263 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
264 if (ret != -EIOCBRETRY)
266 wait_on_retry_sync_kiocb(&kiocb);
269 if (-EIOCBQUEUED == ret)
270 ret = wait_on_sync_kiocb(&kiocb);
271 *ppos = kiocb.ki_pos;
275 EXPORT_SYMBOL(do_sync_read);
277 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
281 if (!(file->f_mode & FMODE_READ))
283 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
285 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
288 ret = rw_verify_area(READ, file, pos, count);
291 if (file->f_op->read)
292 ret = file->f_op->read(file, buf, count, pos);
294 ret = do_sync_read(file, buf, count, pos);
296 fsnotify_access(file->f_path.dentry);
297 add_rchar(current, ret);
305 EXPORT_SYMBOL(vfs_read);
307 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
309 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
313 init_sync_kiocb(&kiocb, filp);
314 kiocb.ki_pos = *ppos;
318 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
319 if (ret != -EIOCBRETRY)
321 wait_on_retry_sync_kiocb(&kiocb);
324 if (-EIOCBQUEUED == ret)
325 ret = wait_on_sync_kiocb(&kiocb);
326 *ppos = kiocb.ki_pos;
330 EXPORT_SYMBOL(do_sync_write);
332 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
336 if (!(file->f_mode & FMODE_WRITE))
338 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
340 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
343 ret = rw_verify_area(WRITE, file, pos, count);
346 if (file->f_op->write)
347 ret = file->f_op->write(file, buf, count, pos);
349 ret = do_sync_write(file, buf, count, pos);
351 fsnotify_modify(file->f_path.dentry);
352 add_wchar(current, ret);
360 EXPORT_SYMBOL(vfs_write);
362 static inline loff_t file_pos_read(struct file *file)
367 static inline void file_pos_write(struct file *file, loff_t pos)
372 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
375 ssize_t ret = -EBADF;
378 file = fget_light(fd, &fput_needed);
380 loff_t pos = file_pos_read(file);
381 ret = vfs_read(file, buf, count, &pos);
382 file_pos_write(file, pos);
383 fput_light(file, fput_needed);
389 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
393 ssize_t ret = -EBADF;
396 file = fget_light(fd, &fput_needed);
398 loff_t pos = file_pos_read(file);
399 ret = vfs_write(file, buf, count, &pos);
400 file_pos_write(file, pos);
401 fput_light(file, fput_needed);
407 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
408 size_t count, loff_t pos)
411 ssize_t ret = -EBADF;
417 file = fget_light(fd, &fput_needed);
420 if (file->f_mode & FMODE_PREAD)
421 ret = vfs_read(file, buf, count, &pos);
422 fput_light(file, fput_needed);
427 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
428 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
430 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
431 (size_t) count, pos);
433 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
436 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
437 size_t count, loff_t pos)
440 ssize_t ret = -EBADF;
446 file = fget_light(fd, &fput_needed);
449 if (file->f_mode & FMODE_PWRITE)
450 ret = vfs_write(file, buf, count, &pos);
451 fput_light(file, fput_needed);
456 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
457 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
459 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
460 (size_t) count, pos);
462 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
466 * Reduce an iovec's length in-place. Return the resulting number of segments
468 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
470 unsigned long seg = 0;
473 while (seg < nr_segs) {
475 if (len + iov->iov_len >= to) {
476 iov->iov_len = to - len;
484 EXPORT_SYMBOL(iov_shorten);
486 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
487 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
492 init_sync_kiocb(&kiocb, filp);
493 kiocb.ki_pos = *ppos;
495 kiocb.ki_nbytes = len;
498 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
499 if (ret != -EIOCBRETRY)
501 wait_on_retry_sync_kiocb(&kiocb);
504 if (ret == -EIOCBQUEUED)
505 ret = wait_on_sync_kiocb(&kiocb);
506 *ppos = kiocb.ki_pos;
510 /* Do it by hand, with file-ops */
511 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
512 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
514 struct iovec *vector = iov;
517 while (nr_segs > 0) {
522 base = vector->iov_base;
523 len = vector->iov_len;
527 nr = fn(filp, base, len, ppos);
542 /* A write operation does a read from user space and vice versa */
543 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
545 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
546 unsigned long nr_segs, unsigned long fast_segs,
547 struct iovec *fast_pointer,
548 struct iovec **ret_pointer)
552 struct iovec *iov = fast_pointer;
555 * SuS says "The readv() function *may* fail if the iovcnt argument
556 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
557 * traditionally returned zero for zero segments, so...
565 * First get the "struct iovec" from user memory and
566 * verify all the pointers
568 if (nr_segs > UIO_MAXIOV) {
572 if (nr_segs > fast_segs) {
573 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
579 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
585 * According to the Single Unix Specification we should return EINVAL
586 * if an element length is < 0 when cast to ssize_t or if the
587 * total length would overflow the ssize_t return value of the
591 for (seg = 0; seg < nr_segs; seg++) {
592 void __user *buf = iov[seg].iov_base;
593 ssize_t len = (ssize_t)iov[seg].iov_len;
595 /* see if we we're about to use an invalid len or if
596 * it's about to overflow ssize_t */
597 if (len < 0 || (ret + len < ret)) {
601 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
613 static ssize_t do_readv_writev(int type, struct file *file,
614 const struct iovec __user * uvector,
615 unsigned long nr_segs, loff_t *pos)
618 struct iovec iovstack[UIO_FASTIOV];
619 struct iovec *iov = iovstack;
629 ret = rw_copy_check_uvector(type, uvector, nr_segs,
630 ARRAY_SIZE(iovstack), iovstack, &iov);
635 ret = rw_verify_area(type, file, pos, tot_len);
641 fn = file->f_op->read;
642 fnv = file->f_op->aio_read;
644 fn = (io_fn_t)file->f_op->write;
645 fnv = file->f_op->aio_write;
649 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
652 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
657 if ((ret + (type == READ)) > 0) {
659 fsnotify_access(file->f_path.dentry);
661 fsnotify_modify(file->f_path.dentry);
666 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
667 unsigned long vlen, loff_t *pos)
669 if (!(file->f_mode & FMODE_READ))
671 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
674 return do_readv_writev(READ, file, vec, vlen, pos);
677 EXPORT_SYMBOL(vfs_readv);
679 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
680 unsigned long vlen, loff_t *pos)
682 if (!(file->f_mode & FMODE_WRITE))
684 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
687 return do_readv_writev(WRITE, file, vec, vlen, pos);
690 EXPORT_SYMBOL(vfs_writev);
692 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
696 ssize_t ret = -EBADF;
699 file = fget_light(fd, &fput_needed);
701 loff_t pos = file_pos_read(file);
702 ret = vfs_readv(file, vec, vlen, &pos);
703 file_pos_write(file, pos);
704 fput_light(file, fput_needed);
708 add_rchar(current, ret);
713 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
717 ssize_t ret = -EBADF;
720 file = fget_light(fd, &fput_needed);
722 loff_t pos = file_pos_read(file);
723 ret = vfs_writev(file, vec, vlen, &pos);
724 file_pos_write(file, pos);
725 fput_light(file, fput_needed);
729 add_wchar(current, ret);
734 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
736 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
737 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
740 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
741 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
743 loff_t pos = pos_from_hilo(pos_h, pos_l);
745 ssize_t ret = -EBADF;
751 file = fget_light(fd, &fput_needed);
754 if (file->f_mode & FMODE_PREAD)
755 ret = vfs_readv(file, vec, vlen, &pos);
756 fput_light(file, fput_needed);
760 add_rchar(current, ret);
765 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
766 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
768 loff_t pos = pos_from_hilo(pos_h, pos_l);
770 ssize_t ret = -EBADF;
776 file = fget_light(fd, &fput_needed);
779 if (file->f_mode & FMODE_PWRITE)
780 ret = vfs_writev(file, vec, vlen, &pos);
781 fput_light(file, fput_needed);
785 add_wchar(current, ret);
790 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
791 size_t count, loff_t max)
793 struct file * in_file, * out_file;
794 struct inode * in_inode, * out_inode;
797 int fput_needed_in, fput_needed_out, fl;
800 * Get input file, and verify that it is ok..
803 in_file = fget_light(in_fd, &fput_needed_in);
806 if (!(in_file->f_mode & FMODE_READ))
810 ppos = &in_file->f_pos;
812 if (!(in_file->f_mode & FMODE_PREAD))
814 retval = rw_verify_area(READ, in_file, ppos, count);
820 * Get output file, and verify that it is ok..
823 out_file = fget_light(out_fd, &fput_needed_out);
826 if (!(out_file->f_mode & FMODE_WRITE))
829 in_inode = in_file->f_path.dentry->d_inode;
830 out_inode = out_file->f_path.dentry->d_inode;
831 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
837 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
840 if (unlikely(pos + count > max)) {
850 * We need to debate whether we can enable this or not. The
851 * man page documents EAGAIN return for the output at least,
852 * and the application is arguably buggy if it doesn't expect
853 * EAGAIN on a non-blocking file descriptor.
855 if (in_file->f_flags & O_NONBLOCK)
856 fl = SPLICE_F_NONBLOCK;
858 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
861 add_rchar(current, retval);
862 add_wchar(current, retval);
871 fput_light(out_file, fput_needed_out);
873 fput_light(in_file, fput_needed_in);
878 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
885 if (unlikely(get_user(off, offset)))
888 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
889 if (unlikely(put_user(pos, offset)))
894 return do_sendfile(out_fd, in_fd, NULL, count, 0);
897 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
903 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
905 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
906 if (unlikely(put_user(pos, offset)))
911 return do_sendfile(out_fd, in_fd, NULL, count, 0);