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;
53 offset += file->f_pos;
57 if (offset < 0 || offset > inode->i_sb->s_maxbytes)
60 /* Special lock needed here? */
61 if (offset != file->f_pos) {
68 EXPORT_SYMBOL(generic_file_llseek_unlocked);
71 * generic_file_llseek - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @origin: type of seek
76 * This is a generic implemenation of ->llseek useable for all normal local
77 * filesystems. It just updates the file offset to the value specified by
78 * @offset and @origin under i_mutex.
80 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
84 mutex_lock(&file->f_dentry->d_inode->i_mutex);
85 rval = generic_file_llseek_unlocked(file, offset, origin);
86 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
90 EXPORT_SYMBOL(generic_file_llseek);
92 loff_t no_llseek(struct file *file, loff_t offset, int origin)
96 EXPORT_SYMBOL(no_llseek);
98 loff_t default_llseek(struct file *file, loff_t offset, int origin)
105 offset += i_size_read(file->f_path.dentry->d_inode);
108 offset += file->f_pos;
112 if (offset != file->f_pos) {
113 file->f_pos = offset;
121 EXPORT_SYMBOL(default_llseek);
123 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
125 loff_t (*fn)(struct file *, loff_t, int);
128 if (file->f_mode & FMODE_LSEEK) {
130 if (file->f_op && file->f_op->llseek)
131 fn = file->f_op->llseek;
133 return fn(file, offset, origin);
135 EXPORT_SYMBOL(vfs_llseek);
137 asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
144 file = fget_light(fd, &fput_needed);
149 if (origin <= SEEK_MAX) {
150 loff_t res = vfs_llseek(file, offset, origin);
152 if (res != (loff_t)retval)
153 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
155 fput_light(file, fput_needed);
160 #ifdef __ARCH_WANT_SYS_LLSEEK
161 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
162 unsigned long offset_low, loff_t __user * result,
171 file = fget_light(fd, &fput_needed);
176 if (origin > SEEK_MAX)
179 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
182 retval = (int)offset;
185 if (!copy_to_user(result, &offset, sizeof(offset)))
189 fput_light(file, fput_needed);
196 * rw_verify_area doesn't like huge counts. We limit
197 * them to something that fits in "int" so that others
198 * won't have to do range checks all the time.
200 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
202 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
206 int retval = -EINVAL;
208 inode = file->f_path.dentry->d_inode;
209 if (unlikely((ssize_t) count < 0))
212 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
215 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
216 retval = locks_mandatory_area(
217 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
218 inode, file, pos, count);
222 retval = security_file_permission(file,
223 read_write == READ ? MAY_READ : MAY_WRITE);
226 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
229 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
231 set_current_state(TASK_UNINTERRUPTIBLE);
232 if (!kiocbIsKicked(iocb))
235 kiocbClearKicked(iocb);
236 __set_current_state(TASK_RUNNING);
239 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
241 struct iovec iov = { .iov_base = buf, .iov_len = len };
245 init_sync_kiocb(&kiocb, filp);
246 kiocb.ki_pos = *ppos;
250 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
251 if (ret != -EIOCBRETRY)
253 wait_on_retry_sync_kiocb(&kiocb);
256 if (-EIOCBQUEUED == ret)
257 ret = wait_on_sync_kiocb(&kiocb);
258 *ppos = kiocb.ki_pos;
262 EXPORT_SYMBOL(do_sync_read);
264 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
268 if (!(file->f_mode & FMODE_READ))
270 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
272 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
275 ret = rw_verify_area(READ, file, pos, count);
278 if (file->f_op->read)
279 ret = file->f_op->read(file, buf, count, pos);
281 ret = do_sync_read(file, buf, count, pos);
283 fsnotify_access(file->f_path.dentry);
284 add_rchar(current, ret);
292 EXPORT_SYMBOL(vfs_read);
294 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
296 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
300 init_sync_kiocb(&kiocb, filp);
301 kiocb.ki_pos = *ppos;
305 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
306 if (ret != -EIOCBRETRY)
308 wait_on_retry_sync_kiocb(&kiocb);
311 if (-EIOCBQUEUED == ret)
312 ret = wait_on_sync_kiocb(&kiocb);
313 *ppos = kiocb.ki_pos;
317 EXPORT_SYMBOL(do_sync_write);
319 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
323 if (!(file->f_mode & FMODE_WRITE))
325 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
327 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
330 ret = rw_verify_area(WRITE, file, pos, count);
333 if (file->f_op->write)
334 ret = file->f_op->write(file, buf, count, pos);
336 ret = do_sync_write(file, buf, count, pos);
338 fsnotify_modify(file->f_path.dentry);
339 add_wchar(current, ret);
347 EXPORT_SYMBOL(vfs_write);
349 static inline loff_t file_pos_read(struct file *file)
354 static inline void file_pos_write(struct file *file, loff_t pos)
359 asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
362 ssize_t ret = -EBADF;
365 file = fget_light(fd, &fput_needed);
367 loff_t pos = file_pos_read(file);
368 ret = vfs_read(file, buf, count, &pos);
369 file_pos_write(file, pos);
370 fput_light(file, fput_needed);
376 asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
379 ssize_t ret = -EBADF;
382 file = fget_light(fd, &fput_needed);
384 loff_t pos = file_pos_read(file);
385 ret = vfs_write(file, buf, count, &pos);
386 file_pos_write(file, pos);
387 fput_light(file, fput_needed);
393 asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
394 size_t count, loff_t pos)
397 ssize_t ret = -EBADF;
403 file = fget_light(fd, &fput_needed);
406 if (file->f_mode & FMODE_PREAD)
407 ret = vfs_read(file, buf, count, &pos);
408 fput_light(file, fput_needed);
414 asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
415 size_t count, loff_t pos)
418 ssize_t ret = -EBADF;
424 file = fget_light(fd, &fput_needed);
427 if (file->f_mode & FMODE_PWRITE)
428 ret = vfs_write(file, buf, count, &pos);
429 fput_light(file, fput_needed);
436 * Reduce an iovec's length in-place. Return the resulting number of segments
438 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
440 unsigned long seg = 0;
443 while (seg < nr_segs) {
445 if (len + iov->iov_len >= to) {
446 iov->iov_len = to - len;
454 EXPORT_SYMBOL(iov_shorten);
456 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
457 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
462 init_sync_kiocb(&kiocb, filp);
463 kiocb.ki_pos = *ppos;
465 kiocb.ki_nbytes = len;
468 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
469 if (ret != -EIOCBRETRY)
471 wait_on_retry_sync_kiocb(&kiocb);
474 if (ret == -EIOCBQUEUED)
475 ret = wait_on_sync_kiocb(&kiocb);
476 *ppos = kiocb.ki_pos;
480 /* Do it by hand, with file-ops */
481 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
482 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
484 struct iovec *vector = iov;
487 while (nr_segs > 0) {
492 base = vector->iov_base;
493 len = vector->iov_len;
497 nr = fn(filp, base, len, ppos);
512 /* A write operation does a read from user space and vice versa */
513 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
515 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
516 unsigned long nr_segs, unsigned long fast_segs,
517 struct iovec *fast_pointer,
518 struct iovec **ret_pointer)
522 struct iovec *iov = fast_pointer;
525 * SuS says "The readv() function *may* fail if the iovcnt argument
526 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
527 * traditionally returned zero for zero segments, so...
535 * First get the "struct iovec" from user memory and
536 * verify all the pointers
538 if (nr_segs > UIO_MAXIOV) {
542 if (nr_segs > fast_segs) {
543 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
549 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
555 * According to the Single Unix Specification we should return EINVAL
556 * if an element length is < 0 when cast to ssize_t or if the
557 * total length would overflow the ssize_t return value of the
561 for (seg = 0; seg < nr_segs; seg++) {
562 void __user *buf = iov[seg].iov_base;
563 ssize_t len = (ssize_t)iov[seg].iov_len;
565 /* see if we we're about to use an invalid len or if
566 * it's about to overflow ssize_t */
567 if (len < 0 || (ret + len < ret)) {
571 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
583 static ssize_t do_readv_writev(int type, struct file *file,
584 const struct iovec __user * uvector,
585 unsigned long nr_segs, loff_t *pos)
588 struct iovec iovstack[UIO_FASTIOV];
589 struct iovec *iov = iovstack;
599 ret = rw_copy_check_uvector(type, uvector, nr_segs,
600 ARRAY_SIZE(iovstack), iovstack, &iov);
605 ret = rw_verify_area(type, file, pos, tot_len);
611 fn = file->f_op->read;
612 fnv = file->f_op->aio_read;
614 fn = (io_fn_t)file->f_op->write;
615 fnv = file->f_op->aio_write;
619 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
622 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
627 if ((ret + (type == READ)) > 0) {
629 fsnotify_access(file->f_path.dentry);
631 fsnotify_modify(file->f_path.dentry);
636 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
637 unsigned long vlen, loff_t *pos)
639 if (!(file->f_mode & FMODE_READ))
641 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
644 return do_readv_writev(READ, file, vec, vlen, pos);
647 EXPORT_SYMBOL(vfs_readv);
649 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
650 unsigned long vlen, loff_t *pos)
652 if (!(file->f_mode & FMODE_WRITE))
654 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
657 return do_readv_writev(WRITE, file, vec, vlen, pos);
660 EXPORT_SYMBOL(vfs_writev);
663 sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
666 ssize_t ret = -EBADF;
669 file = fget_light(fd, &fput_needed);
671 loff_t pos = file_pos_read(file);
672 ret = vfs_readv(file, vec, vlen, &pos);
673 file_pos_write(file, pos);
674 fput_light(file, fput_needed);
678 add_rchar(current, ret);
684 sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
687 ssize_t ret = -EBADF;
690 file = fget_light(fd, &fput_needed);
692 loff_t pos = file_pos_read(file);
693 ret = vfs_writev(file, vec, vlen, &pos);
694 file_pos_write(file, pos);
695 fput_light(file, fput_needed);
699 add_wchar(current, ret);
704 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
705 size_t count, loff_t max)
707 struct file * in_file, * out_file;
708 struct inode * in_inode, * out_inode;
711 int fput_needed_in, fput_needed_out, fl;
714 * Get input file, and verify that it is ok..
717 in_file = fget_light(in_fd, &fput_needed_in);
720 if (!(in_file->f_mode & FMODE_READ))
723 in_inode = in_file->f_path.dentry->d_inode;
726 if (!in_file->f_op || !in_file->f_op->splice_read)
730 ppos = &in_file->f_pos;
732 if (!(in_file->f_mode & FMODE_PREAD))
734 retval = rw_verify_area(READ, in_file, ppos, count);
740 * Get output file, and verify that it is ok..
743 out_file = fget_light(out_fd, &fput_needed_out);
746 if (!(out_file->f_mode & FMODE_WRITE))
749 if (!out_file->f_op || !out_file->f_op->sendpage)
751 out_inode = out_file->f_path.dentry->d_inode;
752 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
758 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
762 if (unlikely(pos < 0))
764 if (unlikely(pos + count > max)) {
774 * We need to debate whether we can enable this or not. The
775 * man page documents EAGAIN return for the output at least,
776 * and the application is arguably buggy if it doesn't expect
777 * EAGAIN on a non-blocking file descriptor.
779 if (in_file->f_flags & O_NONBLOCK)
780 fl = SPLICE_F_NONBLOCK;
782 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
785 add_rchar(current, retval);
786 add_wchar(current, retval);
795 fput_light(out_file, fput_needed_out);
797 fput_light(in_file, fput_needed_in);
802 asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
809 if (unlikely(get_user(off, offset)))
812 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
813 if (unlikely(put_user(pos, offset)))
818 return do_sendfile(out_fd, in_fd, NULL, count, 0);
821 asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
827 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
829 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
830 if (unlikely(put_user(pos, offset)))
835 return do_sendfile(out_fd, in_fd, NULL, count, 0);