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/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/module.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include "read_write.h"
20 #include <asm/uaccess.h>
21 #include <asm/unistd.h>
23 const struct file_operations generic_ro_fops = {
24 .llseek = generic_file_llseek,
26 .aio_read = generic_file_aio_read,
27 .mmap = generic_file_readonly_mmap,
28 .splice_read = generic_file_splice_read,
31 EXPORT_SYMBOL(generic_ro_fops);
33 static inline int unsigned_offsets(struct file *file)
35 return file->f_mode & FMODE_UNSIGNED_OFFSET;
39 * generic_file_llseek_unlocked - lockless generic llseek implementation
40 * @file: file structure to seek on
41 * @offset: file offset to seek to
42 * @origin: type of seek
44 * Updates the file offset to the value specified by @offset and @origin.
45 * Locking must be provided by the caller.
48 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
50 struct inode *inode = file->f_mapping->host;
54 offset += inode->i_size;
58 * Here we special-case the lseek(fd, 0, SEEK_CUR)
59 * position-querying operation. Avoid rewriting the "same"
60 * f_pos value back to the file because a concurrent read(),
61 * write() or lseek() might have altered it
65 offset += file->f_pos;
69 * In the generic case the entire file is data, so as long as
70 * offset isn't at the end of the file then the offset is data.
72 if (offset >= inode->i_size)
77 * There is a virtual hole at the end of the file, so as long as
78 * offset isn't i_size or larger, return i_size.
80 if (offset >= inode->i_size)
82 offset = inode->i_size;
86 if (offset < 0 && !unsigned_offsets(file))
88 if (offset > inode->i_sb->s_maxbytes)
91 /* Special lock needed here? */
92 if (offset != file->f_pos) {
99 EXPORT_SYMBOL(generic_file_llseek_unlocked);
102 * generic_file_llseek - generic llseek implementation for regular files
103 * @file: file structure to seek on
104 * @offset: file offset to seek to
105 * @origin: type of seek
107 * This is a generic implemenation of ->llseek useable for all normal local
108 * filesystems. It just updates the file offset to the value specified by
109 * @offset and @origin under i_mutex.
111 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
115 mutex_lock(&file->f_dentry->d_inode->i_mutex);
116 rval = generic_file_llseek_unlocked(file, offset, origin);
117 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
121 EXPORT_SYMBOL(generic_file_llseek);
124 * noop_llseek - No Operation Performed llseek implementation
125 * @file: file structure to seek on
126 * @offset: file offset to seek to
127 * @origin: type of seek
129 * This is an implementation of ->llseek useable for the rare special case when
130 * userspace expects the seek to succeed but the (device) file is actually not
131 * able to perform the seek. In this case you use noop_llseek() instead of
132 * falling back to the default implementation of ->llseek.
134 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
138 EXPORT_SYMBOL(noop_llseek);
140 loff_t no_llseek(struct file *file, loff_t offset, int origin)
144 EXPORT_SYMBOL(no_llseek);
146 loff_t default_llseek(struct file *file, loff_t offset, int origin)
148 struct inode *inode = file->f_path.dentry->d_inode;
151 mutex_lock(&inode->i_mutex);
154 offset += i_size_read(inode);
158 retval = file->f_pos;
161 offset += file->f_pos;
165 * In the generic case the entire file is data, so as
166 * long as offset isn't at the end of the file then the
169 if (offset >= inode->i_size)
174 * There is a virtual hole at the end of the file, so
175 * as long as offset isn't i_size or larger, return
178 if (offset >= inode->i_size)
180 offset = inode->i_size;
184 if (offset >= 0 || unsigned_offsets(file)) {
185 if (offset != file->f_pos) {
186 file->f_pos = offset;
192 mutex_unlock(&inode->i_mutex);
195 EXPORT_SYMBOL(default_llseek);
197 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
199 loff_t (*fn)(struct file *, loff_t, int);
202 if (file->f_mode & FMODE_LSEEK) {
203 if (file->f_op && file->f_op->llseek)
204 fn = file->f_op->llseek;
206 return fn(file, offset, origin);
208 EXPORT_SYMBOL(vfs_llseek);
210 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
217 file = fget_light(fd, &fput_needed);
222 if (origin <= SEEK_MAX) {
223 loff_t res = vfs_llseek(file, offset, origin);
225 if (res != (loff_t)retval)
226 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
228 fput_light(file, fput_needed);
233 #ifdef __ARCH_WANT_SYS_LLSEEK
234 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
235 unsigned long, offset_low, loff_t __user *, result,
236 unsigned int, origin)
244 file = fget_light(fd, &fput_needed);
249 if (origin > SEEK_MAX)
252 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
255 retval = (int)offset;
258 if (!copy_to_user(result, &offset, sizeof(offset)))
262 fput_light(file, fput_needed);
270 * rw_verify_area doesn't like huge counts. We limit
271 * them to something that fits in "int" so that others
272 * won't have to do range checks all the time.
274 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
278 int retval = -EINVAL;
280 inode = file->f_path.dentry->d_inode;
281 if (unlikely((ssize_t) count < 0))
284 if (unlikely(pos < 0)) {
285 if (!unsigned_offsets(file))
287 if (count >= -pos) /* both values are in 0..LLONG_MAX */
289 } else if (unlikely((loff_t) (pos + count) < 0)) {
290 if (!unsigned_offsets(file))
294 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
295 retval = locks_mandatory_area(
296 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
297 inode, file, pos, count);
301 retval = security_file_permission(file,
302 read_write == READ ? MAY_READ : MAY_WRITE);
305 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
308 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
310 set_current_state(TASK_UNINTERRUPTIBLE);
311 if (!kiocbIsKicked(iocb))
314 kiocbClearKicked(iocb);
315 __set_current_state(TASK_RUNNING);
318 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
320 struct iovec iov = { .iov_base = buf, .iov_len = len };
324 init_sync_kiocb(&kiocb, filp);
325 kiocb.ki_pos = *ppos;
327 kiocb.ki_nbytes = len;
330 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
331 if (ret != -EIOCBRETRY)
333 wait_on_retry_sync_kiocb(&kiocb);
336 if (-EIOCBQUEUED == ret)
337 ret = wait_on_sync_kiocb(&kiocb);
338 *ppos = kiocb.ki_pos;
342 EXPORT_SYMBOL(do_sync_read);
344 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
348 if (!(file->f_mode & FMODE_READ))
350 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
352 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
355 ret = rw_verify_area(READ, file, pos, count);
358 if (file->f_op->read)
359 ret = file->f_op->read(file, buf, count, pos);
361 ret = do_sync_read(file, buf, count, pos);
363 fsnotify_access(file);
364 add_rchar(current, ret);
372 EXPORT_SYMBOL(vfs_read);
374 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
376 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
380 init_sync_kiocb(&kiocb, filp);
381 kiocb.ki_pos = *ppos;
383 kiocb.ki_nbytes = len;
386 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
387 if (ret != -EIOCBRETRY)
389 wait_on_retry_sync_kiocb(&kiocb);
392 if (-EIOCBQUEUED == ret)
393 ret = wait_on_sync_kiocb(&kiocb);
394 *ppos = kiocb.ki_pos;
398 EXPORT_SYMBOL(do_sync_write);
400 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
404 if (!(file->f_mode & FMODE_WRITE))
406 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
408 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
411 ret = rw_verify_area(WRITE, file, pos, count);
414 if (file->f_op->write)
415 ret = file->f_op->write(file, buf, count, pos);
417 ret = do_sync_write(file, buf, count, pos);
419 fsnotify_modify(file);
420 add_wchar(current, ret);
428 EXPORT_SYMBOL(vfs_write);
430 static inline loff_t file_pos_read(struct file *file)
435 static inline void file_pos_write(struct file *file, loff_t pos)
440 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
443 ssize_t ret = -EBADF;
446 file = fget_light(fd, &fput_needed);
448 loff_t pos = file_pos_read(file);
449 ret = vfs_read(file, buf, count, &pos);
450 file_pos_write(file, pos);
451 fput_light(file, fput_needed);
457 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
461 ssize_t ret = -EBADF;
464 file = fget_light(fd, &fput_needed);
466 loff_t pos = file_pos_read(file);
467 ret = vfs_write(file, buf, count, &pos);
468 file_pos_write(file, pos);
469 fput_light(file, fput_needed);
475 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
476 size_t count, loff_t pos)
479 ssize_t ret = -EBADF;
485 file = fget_light(fd, &fput_needed);
488 if (file->f_mode & FMODE_PREAD)
489 ret = vfs_read(file, buf, count, &pos);
490 fput_light(file, fput_needed);
495 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
496 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
498 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
499 (size_t) count, pos);
501 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
504 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
505 size_t count, loff_t pos)
508 ssize_t ret = -EBADF;
514 file = fget_light(fd, &fput_needed);
517 if (file->f_mode & FMODE_PWRITE)
518 ret = vfs_write(file, buf, count, &pos);
519 fput_light(file, fput_needed);
524 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
525 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
527 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
528 (size_t) count, pos);
530 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
534 * Reduce an iovec's length in-place. Return the resulting number of segments
536 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
538 unsigned long seg = 0;
541 while (seg < nr_segs) {
543 if (len + iov->iov_len >= to) {
544 iov->iov_len = to - len;
552 EXPORT_SYMBOL(iov_shorten);
554 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
555 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
560 init_sync_kiocb(&kiocb, filp);
561 kiocb.ki_pos = *ppos;
563 kiocb.ki_nbytes = len;
566 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
567 if (ret != -EIOCBRETRY)
569 wait_on_retry_sync_kiocb(&kiocb);
572 if (ret == -EIOCBQUEUED)
573 ret = wait_on_sync_kiocb(&kiocb);
574 *ppos = kiocb.ki_pos;
578 /* Do it by hand, with file-ops */
579 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
580 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
582 struct iovec *vector = iov;
585 while (nr_segs > 0) {
590 base = vector->iov_base;
591 len = vector->iov_len;
595 nr = fn(filp, base, len, ppos);
610 /* A write operation does a read from user space and vice versa */
611 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
613 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
614 unsigned long nr_segs, unsigned long fast_segs,
615 struct iovec *fast_pointer,
616 struct iovec **ret_pointer)
620 struct iovec *iov = fast_pointer;
623 * SuS says "The readv() function *may* fail if the iovcnt argument
624 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
625 * traditionally returned zero for zero segments, so...
633 * First get the "struct iovec" from user memory and
634 * verify all the pointers
636 if (nr_segs > UIO_MAXIOV) {
640 if (nr_segs > fast_segs) {
641 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
647 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
653 * According to the Single Unix Specification we should return EINVAL
654 * if an element length is < 0 when cast to ssize_t or if the
655 * total length would overflow the ssize_t return value of the
658 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
662 for (seg = 0; seg < nr_segs; seg++) {
663 void __user *buf = iov[seg].iov_base;
664 ssize_t len = (ssize_t)iov[seg].iov_len;
666 /* see if we we're about to use an invalid len or if
667 * it's about to overflow ssize_t */
672 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
676 if (len > MAX_RW_COUNT - ret) {
677 len = MAX_RW_COUNT - ret;
678 iov[seg].iov_len = len;
687 static ssize_t do_readv_writev(int type, struct file *file,
688 const struct iovec __user * uvector,
689 unsigned long nr_segs, loff_t *pos)
692 struct iovec iovstack[UIO_FASTIOV];
693 struct iovec *iov = iovstack;
703 ret = rw_copy_check_uvector(type, uvector, nr_segs,
704 ARRAY_SIZE(iovstack), iovstack, &iov);
709 ret = rw_verify_area(type, file, pos, tot_len);
715 fn = file->f_op->read;
716 fnv = file->f_op->aio_read;
718 fn = (io_fn_t)file->f_op->write;
719 fnv = file->f_op->aio_write;
723 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
726 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
731 if ((ret + (type == READ)) > 0) {
733 fsnotify_access(file);
735 fsnotify_modify(file);
740 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
741 unsigned long vlen, loff_t *pos)
743 if (!(file->f_mode & FMODE_READ))
745 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
748 return do_readv_writev(READ, file, vec, vlen, pos);
751 EXPORT_SYMBOL(vfs_readv);
753 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
754 unsigned long vlen, loff_t *pos)
756 if (!(file->f_mode & FMODE_WRITE))
758 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
761 return do_readv_writev(WRITE, file, vec, vlen, pos);
764 EXPORT_SYMBOL(vfs_writev);
766 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
770 ssize_t ret = -EBADF;
773 file = fget_light(fd, &fput_needed);
775 loff_t pos = file_pos_read(file);
776 ret = vfs_readv(file, vec, vlen, &pos);
777 file_pos_write(file, pos);
778 fput_light(file, fput_needed);
782 add_rchar(current, ret);
787 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
791 ssize_t ret = -EBADF;
794 file = fget_light(fd, &fput_needed);
796 loff_t pos = file_pos_read(file);
797 ret = vfs_writev(file, vec, vlen, &pos);
798 file_pos_write(file, pos);
799 fput_light(file, fput_needed);
803 add_wchar(current, ret);
808 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
810 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
811 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
814 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
815 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
817 loff_t pos = pos_from_hilo(pos_h, pos_l);
819 ssize_t ret = -EBADF;
825 file = fget_light(fd, &fput_needed);
828 if (file->f_mode & FMODE_PREAD)
829 ret = vfs_readv(file, vec, vlen, &pos);
830 fput_light(file, fput_needed);
834 add_rchar(current, ret);
839 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
840 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
842 loff_t pos = pos_from_hilo(pos_h, pos_l);
844 ssize_t ret = -EBADF;
850 file = fget_light(fd, &fput_needed);
853 if (file->f_mode & FMODE_PWRITE)
854 ret = vfs_writev(file, vec, vlen, &pos);
855 fput_light(file, fput_needed);
859 add_wchar(current, ret);
864 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
865 size_t count, loff_t max)
867 struct file * in_file, * out_file;
868 struct inode * in_inode, * out_inode;
871 int fput_needed_in, fput_needed_out, fl;
874 * Get input file, and verify that it is ok..
877 in_file = fget_light(in_fd, &fput_needed_in);
880 if (!(in_file->f_mode & FMODE_READ))
884 ppos = &in_file->f_pos;
886 if (!(in_file->f_mode & FMODE_PREAD))
888 retval = rw_verify_area(READ, in_file, ppos, count);
894 * Get output file, and verify that it is ok..
897 out_file = fget_light(out_fd, &fput_needed_out);
900 if (!(out_file->f_mode & FMODE_WRITE))
903 in_inode = in_file->f_path.dentry->d_inode;
904 out_inode = out_file->f_path.dentry->d_inode;
905 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
911 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
914 if (unlikely(pos + count > max)) {
924 * We need to debate whether we can enable this or not. The
925 * man page documents EAGAIN return for the output at least,
926 * and the application is arguably buggy if it doesn't expect
927 * EAGAIN on a non-blocking file descriptor.
929 if (in_file->f_flags & O_NONBLOCK)
930 fl = SPLICE_F_NONBLOCK;
932 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
935 add_rchar(current, retval);
936 add_wchar(current, retval);
945 fput_light(out_file, fput_needed_out);
947 fput_light(in_file, fput_needed_in);
952 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
959 if (unlikely(get_user(off, offset)))
962 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
963 if (unlikely(put_user(pos, offset)))
968 return do_sendfile(out_fd, in_fd, NULL, count, 0);
971 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
977 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
979 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
980 if (unlikely(put_user(pos, offset)))
985 return do_sendfile(out_fd, in_fd, NULL, count, 0);