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);
101 * noop_llseek - No Operation Performed llseek implementation
102 * @file: file structure to seek on
103 * @offset: file offset to seek to
104 * @origin: type of seek
106 * This is an implementation of ->llseek useable for the rare special case when
107 * userspace expects the seek to succeed but the (device) file is actually not
108 * able to perform the seek. In this case you use noop_llseek() instead of
109 * falling back to the default implementation of ->llseek.
111 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
115 EXPORT_SYMBOL(noop_llseek);
117 loff_t no_llseek(struct file *file, loff_t offset, int origin)
121 EXPORT_SYMBOL(no_llseek);
123 loff_t default_llseek(struct file *file, loff_t offset, int origin)
130 offset += i_size_read(file->f_path.dentry->d_inode);
134 retval = file->f_pos;
137 offset += file->f_pos;
141 if (offset != file->f_pos) {
142 file->f_pos = offset;
151 EXPORT_SYMBOL(default_llseek);
153 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
155 loff_t (*fn)(struct file *, loff_t, int);
158 if (file->f_mode & FMODE_LSEEK) {
160 if (file->f_op && file->f_op->llseek)
161 fn = file->f_op->llseek;
163 return fn(file, offset, origin);
165 EXPORT_SYMBOL(vfs_llseek);
167 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
174 file = fget_light(fd, &fput_needed);
179 if (origin <= SEEK_MAX) {
180 loff_t res = vfs_llseek(file, offset, origin);
182 if (res != (loff_t)retval)
183 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
185 fput_light(file, fput_needed);
190 #ifdef __ARCH_WANT_SYS_LLSEEK
191 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
192 unsigned long, offset_low, loff_t __user *, result,
193 unsigned int, origin)
201 file = fget_light(fd, &fput_needed);
206 if (origin > SEEK_MAX)
209 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
212 retval = (int)offset;
215 if (!copy_to_user(result, &offset, sizeof(offset)))
219 fput_light(file, fput_needed);
226 * rw_verify_area doesn't like huge counts. We limit
227 * them to something that fits in "int" so that others
228 * won't have to do range checks all the time.
230 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
232 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
236 int retval = -EINVAL;
238 inode = file->f_path.dentry->d_inode;
239 if (unlikely((ssize_t) count < 0))
242 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
245 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
246 retval = locks_mandatory_area(
247 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
248 inode, file, pos, count);
252 retval = security_file_permission(file,
253 read_write == READ ? MAY_READ : MAY_WRITE);
256 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
259 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
261 set_current_state(TASK_UNINTERRUPTIBLE);
262 if (!kiocbIsKicked(iocb))
265 kiocbClearKicked(iocb);
266 __set_current_state(TASK_RUNNING);
269 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
271 struct iovec iov = { .iov_base = buf, .iov_len = len };
275 init_sync_kiocb(&kiocb, filp);
276 kiocb.ki_pos = *ppos;
278 kiocb.ki_nbytes = len;
281 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
282 if (ret != -EIOCBRETRY)
284 wait_on_retry_sync_kiocb(&kiocb);
287 if (-EIOCBQUEUED == ret)
288 ret = wait_on_sync_kiocb(&kiocb);
289 *ppos = kiocb.ki_pos;
293 EXPORT_SYMBOL(do_sync_read);
295 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
299 if (!(file->f_mode & FMODE_READ))
301 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
303 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
306 ret = rw_verify_area(READ, file, pos, count);
309 if (file->f_op->read)
310 ret = file->f_op->read(file, buf, count, pos);
312 ret = do_sync_read(file, buf, count, pos);
314 fsnotify_access(file->f_path.dentry);
315 add_rchar(current, ret);
323 EXPORT_SYMBOL(vfs_read);
325 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
327 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
331 init_sync_kiocb(&kiocb, filp);
332 kiocb.ki_pos = *ppos;
334 kiocb.ki_nbytes = len;
337 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
338 if (ret != -EIOCBRETRY)
340 wait_on_retry_sync_kiocb(&kiocb);
343 if (-EIOCBQUEUED == ret)
344 ret = wait_on_sync_kiocb(&kiocb);
345 *ppos = kiocb.ki_pos;
349 EXPORT_SYMBOL(do_sync_write);
351 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
355 if (!(file->f_mode & FMODE_WRITE))
357 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
359 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
362 ret = rw_verify_area(WRITE, file, pos, count);
365 if (file->f_op->write)
366 ret = file->f_op->write(file, buf, count, pos);
368 ret = do_sync_write(file, buf, count, pos);
370 fsnotify_modify(file->f_path.dentry);
371 add_wchar(current, ret);
379 EXPORT_SYMBOL(vfs_write);
381 static inline loff_t file_pos_read(struct file *file)
386 static inline void file_pos_write(struct file *file, loff_t pos)
391 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
394 ssize_t ret = -EBADF;
397 file = fget_light(fd, &fput_needed);
399 loff_t pos = file_pos_read(file);
400 ret = vfs_read(file, buf, count, &pos);
401 file_pos_write(file, pos);
402 fput_light(file, fput_needed);
408 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
412 ssize_t ret = -EBADF;
415 file = fget_light(fd, &fput_needed);
417 loff_t pos = file_pos_read(file);
418 ret = vfs_write(file, buf, count, &pos);
419 file_pos_write(file, pos);
420 fput_light(file, fput_needed);
426 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
427 size_t count, loff_t pos)
430 ssize_t ret = -EBADF;
436 file = fget_light(fd, &fput_needed);
439 if (file->f_mode & FMODE_PREAD)
440 ret = vfs_read(file, buf, count, &pos);
441 fput_light(file, fput_needed);
446 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
447 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
449 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
450 (size_t) count, pos);
452 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
455 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
456 size_t count, loff_t pos)
459 ssize_t ret = -EBADF;
465 file = fget_light(fd, &fput_needed);
468 if (file->f_mode & FMODE_PWRITE)
469 ret = vfs_write(file, buf, count, &pos);
470 fput_light(file, fput_needed);
475 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
476 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
478 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
479 (size_t) count, pos);
481 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
485 * Reduce an iovec's length in-place. Return the resulting number of segments
487 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
489 unsigned long seg = 0;
492 while (seg < nr_segs) {
494 if (len + iov->iov_len >= to) {
495 iov->iov_len = to - len;
503 EXPORT_SYMBOL(iov_shorten);
505 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
506 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
511 init_sync_kiocb(&kiocb, filp);
512 kiocb.ki_pos = *ppos;
514 kiocb.ki_nbytes = len;
517 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
518 if (ret != -EIOCBRETRY)
520 wait_on_retry_sync_kiocb(&kiocb);
523 if (ret == -EIOCBQUEUED)
524 ret = wait_on_sync_kiocb(&kiocb);
525 *ppos = kiocb.ki_pos;
529 /* Do it by hand, with file-ops */
530 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
531 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
533 struct iovec *vector = iov;
536 while (nr_segs > 0) {
541 base = vector->iov_base;
542 len = vector->iov_len;
546 nr = fn(filp, base, len, ppos);
561 /* A write operation does a read from user space and vice versa */
562 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
564 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
565 unsigned long nr_segs, unsigned long fast_segs,
566 struct iovec *fast_pointer,
567 struct iovec **ret_pointer)
571 struct iovec *iov = fast_pointer;
574 * SuS says "The readv() function *may* fail if the iovcnt argument
575 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
576 * traditionally returned zero for zero segments, so...
584 * First get the "struct iovec" from user memory and
585 * verify all the pointers
587 if (nr_segs > UIO_MAXIOV) {
591 if (nr_segs > fast_segs) {
592 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
598 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
604 * According to the Single Unix Specification we should return EINVAL
605 * if an element length is < 0 when cast to ssize_t or if the
606 * total length would overflow the ssize_t return value of the
610 for (seg = 0; seg < nr_segs; seg++) {
611 void __user *buf = iov[seg].iov_base;
612 ssize_t len = (ssize_t)iov[seg].iov_len;
614 /* see if we we're about to use an invalid len or if
615 * it's about to overflow ssize_t */
616 if (len < 0 || (ret + len < ret)) {
620 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
632 static ssize_t do_readv_writev(int type, struct file *file,
633 const struct iovec __user * uvector,
634 unsigned long nr_segs, loff_t *pos)
637 struct iovec iovstack[UIO_FASTIOV];
638 struct iovec *iov = iovstack;
648 ret = rw_copy_check_uvector(type, uvector, nr_segs,
649 ARRAY_SIZE(iovstack), iovstack, &iov);
654 ret = rw_verify_area(type, file, pos, tot_len);
660 fn = file->f_op->read;
661 fnv = file->f_op->aio_read;
663 fn = (io_fn_t)file->f_op->write;
664 fnv = file->f_op->aio_write;
668 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
671 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
676 if ((ret + (type == READ)) > 0) {
678 fsnotify_access(file->f_path.dentry);
680 fsnotify_modify(file->f_path.dentry);
685 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
686 unsigned long vlen, loff_t *pos)
688 if (!(file->f_mode & FMODE_READ))
690 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
693 return do_readv_writev(READ, file, vec, vlen, pos);
696 EXPORT_SYMBOL(vfs_readv);
698 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
699 unsigned long vlen, loff_t *pos)
701 if (!(file->f_mode & FMODE_WRITE))
703 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
706 return do_readv_writev(WRITE, file, vec, vlen, pos);
709 EXPORT_SYMBOL(vfs_writev);
711 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
715 ssize_t ret = -EBADF;
718 file = fget_light(fd, &fput_needed);
720 loff_t pos = file_pos_read(file);
721 ret = vfs_readv(file, vec, vlen, &pos);
722 file_pos_write(file, pos);
723 fput_light(file, fput_needed);
727 add_rchar(current, ret);
732 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
736 ssize_t ret = -EBADF;
739 file = fget_light(fd, &fput_needed);
741 loff_t pos = file_pos_read(file);
742 ret = vfs_writev(file, vec, vlen, &pos);
743 file_pos_write(file, pos);
744 fput_light(file, fput_needed);
748 add_wchar(current, ret);
753 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
755 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
756 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
759 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
760 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
762 loff_t pos = pos_from_hilo(pos_h, pos_l);
764 ssize_t ret = -EBADF;
770 file = fget_light(fd, &fput_needed);
773 if (file->f_mode & FMODE_PREAD)
774 ret = vfs_readv(file, vec, vlen, &pos);
775 fput_light(file, fput_needed);
779 add_rchar(current, ret);
784 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
785 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
787 loff_t pos = pos_from_hilo(pos_h, pos_l);
789 ssize_t ret = -EBADF;
795 file = fget_light(fd, &fput_needed);
798 if (file->f_mode & FMODE_PWRITE)
799 ret = vfs_writev(file, vec, vlen, &pos);
800 fput_light(file, fput_needed);
804 add_wchar(current, ret);
809 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
810 size_t count, loff_t max)
812 struct file * in_file, * out_file;
813 struct inode * in_inode, * out_inode;
816 int fput_needed_in, fput_needed_out, fl;
819 * Get input file, and verify that it is ok..
822 in_file = fget_light(in_fd, &fput_needed_in);
825 if (!(in_file->f_mode & FMODE_READ))
829 ppos = &in_file->f_pos;
831 if (!(in_file->f_mode & FMODE_PREAD))
833 retval = rw_verify_area(READ, in_file, ppos, count);
839 * Get output file, and verify that it is ok..
842 out_file = fget_light(out_fd, &fput_needed_out);
845 if (!(out_file->f_mode & FMODE_WRITE))
848 in_inode = in_file->f_path.dentry->d_inode;
849 out_inode = out_file->f_path.dentry->d_inode;
850 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
856 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
859 if (unlikely(pos + count > max)) {
869 * We need to debate whether we can enable this or not. The
870 * man page documents EAGAIN return for the output at least,
871 * and the application is arguably buggy if it doesn't expect
872 * EAGAIN on a non-blocking file descriptor.
874 if (in_file->f_flags & O_NONBLOCK)
875 fl = SPLICE_F_NONBLOCK;
877 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
880 add_rchar(current, retval);
881 add_wchar(current, retval);
890 fput_light(out_file, fput_needed_out);
892 fput_light(in_file, fput_needed_in);
897 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
904 if (unlikely(get_user(off, offset)))
907 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
908 if (unlikely(put_user(pos, offset)))
913 return do_sendfile(out_fd, in_fd, NULL, count, 0);
916 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
922 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
924 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
925 if (unlikely(put_user(pos, offset)))
930 return do_sendfile(out_fd, in_fd, NULL, count, 0);