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 __negative_fpos_check(struct file *file, loff_t pos, size_t count)
38 * pos or pos+count is negative here, check overflow.
39 * too big "count" will be caught in rw_verify_area().
41 if ((pos < 0) && (pos + count < pos))
43 if (file->f_mode & FMODE_UNSIGNED_OFFSET)
49 * generic_file_llseek_unlocked - lockless generic llseek implementation
50 * @file: file structure to seek on
51 * @offset: file offset to seek to
52 * @origin: type of seek
54 * Updates the file offset to the value specified by @offset and @origin.
55 * Locking must be provided by the caller.
58 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
60 struct inode *inode = file->f_mapping->host;
64 offset += inode->i_size;
68 * Here we special-case the lseek(fd, 0, SEEK_CUR)
69 * position-querying operation. Avoid rewriting the "same"
70 * f_pos value back to the file because a concurrent read(),
71 * write() or lseek() might have altered it
75 offset += file->f_pos;
79 if (offset < 0 && __negative_fpos_check(file, offset, 0))
81 if (offset > inode->i_sb->s_maxbytes)
84 /* Special lock needed here? */
85 if (offset != file->f_pos) {
92 EXPORT_SYMBOL(generic_file_llseek_unlocked);
95 * generic_file_llseek - generic llseek implementation for regular files
96 * @file: file structure to seek on
97 * @offset: file offset to seek to
98 * @origin: type of seek
100 * This is a generic implemenation of ->llseek useable for all normal local
101 * filesystems. It just updates the file offset to the value specified by
102 * @offset and @origin under i_mutex.
104 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
108 mutex_lock(&file->f_dentry->d_inode->i_mutex);
109 rval = generic_file_llseek_unlocked(file, offset, origin);
110 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
114 EXPORT_SYMBOL(generic_file_llseek);
117 * noop_llseek - No Operation Performed llseek implementation
118 * @file: file structure to seek on
119 * @offset: file offset to seek to
120 * @origin: type of seek
122 * This is an implementation of ->llseek useable for the rare special case when
123 * userspace expects the seek to succeed but the (device) file is actually not
124 * able to perform the seek. In this case you use noop_llseek() instead of
125 * falling back to the default implementation of ->llseek.
127 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
131 EXPORT_SYMBOL(noop_llseek);
133 loff_t no_llseek(struct file *file, loff_t offset, int origin)
137 EXPORT_SYMBOL(no_llseek);
139 loff_t default_llseek(struct file *file, loff_t offset, int origin)
143 mutex_lock(&file->f_dentry->d_inode->i_mutex);
146 offset += i_size_read(file->f_path.dentry->d_inode);
150 retval = file->f_pos;
153 offset += file->f_pos;
156 if (offset >= 0 || !__negative_fpos_check(file, offset, 0)) {
157 if (offset != file->f_pos) {
158 file->f_pos = offset;
164 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
167 EXPORT_SYMBOL(default_llseek);
169 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
171 loff_t (*fn)(struct file *, loff_t, int);
174 if (file->f_mode & FMODE_LSEEK) {
175 if (file->f_op && file->f_op->llseek)
176 fn = file->f_op->llseek;
178 return fn(file, offset, origin);
180 EXPORT_SYMBOL(vfs_llseek);
182 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
189 file = fget_light(fd, &fput_needed);
194 if (origin <= SEEK_MAX) {
195 loff_t res = vfs_llseek(file, offset, origin);
197 if (res != (loff_t)retval)
198 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
200 fput_light(file, fput_needed);
205 #ifdef __ARCH_WANT_SYS_LLSEEK
206 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
207 unsigned long, offset_low, loff_t __user *, result,
208 unsigned int, origin)
216 file = fget_light(fd, &fput_needed);
221 if (origin > SEEK_MAX)
224 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
227 retval = (int)offset;
230 if (!copy_to_user(result, &offset, sizeof(offset)))
234 fput_light(file, fput_needed);
242 * rw_verify_area doesn't like huge counts. We limit
243 * them to something that fits in "int" so that others
244 * won't have to do range checks all the time.
246 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
250 int retval = -EINVAL;
252 inode = file->f_path.dentry->d_inode;
253 if (unlikely((ssize_t) count < 0))
256 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0)) {
257 retval = __negative_fpos_check(file, pos, count);
262 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
263 retval = locks_mandatory_area(
264 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
265 inode, file, pos, count);
269 retval = security_file_permission(file,
270 read_write == READ ? MAY_READ : MAY_WRITE);
273 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
276 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
278 set_current_state(TASK_UNINTERRUPTIBLE);
279 if (!kiocbIsKicked(iocb))
282 kiocbClearKicked(iocb);
283 __set_current_state(TASK_RUNNING);
286 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
288 struct iovec iov = { .iov_base = buf, .iov_len = len };
292 init_sync_kiocb(&kiocb, filp);
293 kiocb.ki_pos = *ppos;
295 kiocb.ki_nbytes = len;
298 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
299 if (ret != -EIOCBRETRY)
301 wait_on_retry_sync_kiocb(&kiocb);
304 if (-EIOCBQUEUED == ret)
305 ret = wait_on_sync_kiocb(&kiocb);
306 *ppos = kiocb.ki_pos;
310 EXPORT_SYMBOL(do_sync_read);
312 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
316 if (!(file->f_mode & FMODE_READ))
318 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
320 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
323 ret = rw_verify_area(READ, file, pos, count);
326 if (file->f_op->read)
327 ret = file->f_op->read(file, buf, count, pos);
329 ret = do_sync_read(file, buf, count, pos);
331 fsnotify_access(file);
332 add_rchar(current, ret);
340 EXPORT_SYMBOL(vfs_read);
342 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
344 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
348 init_sync_kiocb(&kiocb, filp);
349 kiocb.ki_pos = *ppos;
351 kiocb.ki_nbytes = len;
354 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
355 if (ret != -EIOCBRETRY)
357 wait_on_retry_sync_kiocb(&kiocb);
360 if (-EIOCBQUEUED == ret)
361 ret = wait_on_sync_kiocb(&kiocb);
362 *ppos = kiocb.ki_pos;
366 EXPORT_SYMBOL(do_sync_write);
368 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
372 if (!(file->f_mode & FMODE_WRITE))
374 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
376 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
379 ret = rw_verify_area(WRITE, file, pos, count);
382 if (file->f_op->write)
383 ret = file->f_op->write(file, buf, count, pos);
385 ret = do_sync_write(file, buf, count, pos);
387 fsnotify_modify(file);
388 add_wchar(current, ret);
396 EXPORT_SYMBOL(vfs_write);
398 static inline loff_t file_pos_read(struct file *file)
403 static inline void file_pos_write(struct file *file, loff_t pos)
408 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
411 ssize_t ret = -EBADF;
414 file = fget_light(fd, &fput_needed);
416 loff_t pos = file_pos_read(file);
417 ret = vfs_read(file, buf, count, &pos);
418 file_pos_write(file, pos);
419 fput_light(file, fput_needed);
425 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
429 ssize_t ret = -EBADF;
432 file = fget_light(fd, &fput_needed);
434 loff_t pos = file_pos_read(file);
435 ret = vfs_write(file, buf, count, &pos);
436 file_pos_write(file, pos);
437 fput_light(file, fput_needed);
443 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
444 size_t count, loff_t pos)
447 ssize_t ret = -EBADF;
453 file = fget_light(fd, &fput_needed);
456 if (file->f_mode & FMODE_PREAD)
457 ret = vfs_read(file, buf, count, &pos);
458 fput_light(file, fput_needed);
463 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
464 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
466 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
467 (size_t) count, pos);
469 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
472 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
473 size_t count, loff_t pos)
476 ssize_t ret = -EBADF;
482 file = fget_light(fd, &fput_needed);
485 if (file->f_mode & FMODE_PWRITE)
486 ret = vfs_write(file, buf, count, &pos);
487 fput_light(file, fput_needed);
492 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
493 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
495 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
496 (size_t) count, pos);
498 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
502 * Reduce an iovec's length in-place. Return the resulting number of segments
504 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
506 unsigned long seg = 0;
509 while (seg < nr_segs) {
511 if (len + iov->iov_len >= to) {
512 iov->iov_len = to - len;
520 EXPORT_SYMBOL(iov_shorten);
522 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
523 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
528 init_sync_kiocb(&kiocb, filp);
529 kiocb.ki_pos = *ppos;
531 kiocb.ki_nbytes = len;
534 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
535 if (ret != -EIOCBRETRY)
537 wait_on_retry_sync_kiocb(&kiocb);
540 if (ret == -EIOCBQUEUED)
541 ret = wait_on_sync_kiocb(&kiocb);
542 *ppos = kiocb.ki_pos;
546 /* Do it by hand, with file-ops */
547 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
548 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
550 struct iovec *vector = iov;
553 while (nr_segs > 0) {
558 base = vector->iov_base;
559 len = vector->iov_len;
563 nr = fn(filp, base, len, ppos);
578 /* A write operation does a read from user space and vice versa */
579 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
581 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
582 unsigned long nr_segs, unsigned long fast_segs,
583 struct iovec *fast_pointer,
584 struct iovec **ret_pointer)
588 struct iovec *iov = fast_pointer;
591 * SuS says "The readv() function *may* fail if the iovcnt argument
592 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
593 * traditionally returned zero for zero segments, so...
601 * First get the "struct iovec" from user memory and
602 * verify all the pointers
604 if (nr_segs > UIO_MAXIOV) {
608 if (nr_segs > fast_segs) {
609 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
615 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
621 * According to the Single Unix Specification we should return EINVAL
622 * if an element length is < 0 when cast to ssize_t or if the
623 * total length would overflow the ssize_t return value of the
626 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
630 for (seg = 0; seg < nr_segs; seg++) {
631 void __user *buf = iov[seg].iov_base;
632 ssize_t len = (ssize_t)iov[seg].iov_len;
634 /* see if we we're about to use an invalid len or if
635 * it's about to overflow ssize_t */
640 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
644 if (len > MAX_RW_COUNT - ret) {
645 len = MAX_RW_COUNT - ret;
646 iov[seg].iov_len = len;
655 static ssize_t do_readv_writev(int type, struct file *file,
656 const struct iovec __user * uvector,
657 unsigned long nr_segs, loff_t *pos)
660 struct iovec iovstack[UIO_FASTIOV];
661 struct iovec *iov = iovstack;
671 ret = rw_copy_check_uvector(type, uvector, nr_segs,
672 ARRAY_SIZE(iovstack), iovstack, &iov);
677 ret = rw_verify_area(type, file, pos, tot_len);
683 fn = file->f_op->read;
684 fnv = file->f_op->aio_read;
686 fn = (io_fn_t)file->f_op->write;
687 fnv = file->f_op->aio_write;
691 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
694 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
699 if ((ret + (type == READ)) > 0) {
701 fsnotify_access(file);
703 fsnotify_modify(file);
708 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
709 unsigned long vlen, loff_t *pos)
711 if (!(file->f_mode & FMODE_READ))
713 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
716 return do_readv_writev(READ, file, vec, vlen, pos);
719 EXPORT_SYMBOL(vfs_readv);
721 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
722 unsigned long vlen, loff_t *pos)
724 if (!(file->f_mode & FMODE_WRITE))
726 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
729 return do_readv_writev(WRITE, file, vec, vlen, pos);
732 EXPORT_SYMBOL(vfs_writev);
734 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
738 ssize_t ret = -EBADF;
741 file = fget_light(fd, &fput_needed);
743 loff_t pos = file_pos_read(file);
744 ret = vfs_readv(file, vec, vlen, &pos);
745 file_pos_write(file, pos);
746 fput_light(file, fput_needed);
750 add_rchar(current, ret);
755 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
759 ssize_t ret = -EBADF;
762 file = fget_light(fd, &fput_needed);
764 loff_t pos = file_pos_read(file);
765 ret = vfs_writev(file, vec, vlen, &pos);
766 file_pos_write(file, pos);
767 fput_light(file, fput_needed);
771 add_wchar(current, ret);
776 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
778 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
779 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
782 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
783 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
785 loff_t pos = pos_from_hilo(pos_h, pos_l);
787 ssize_t ret = -EBADF;
793 file = fget_light(fd, &fput_needed);
796 if (file->f_mode & FMODE_PREAD)
797 ret = vfs_readv(file, vec, vlen, &pos);
798 fput_light(file, fput_needed);
802 add_rchar(current, ret);
807 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
808 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
810 loff_t pos = pos_from_hilo(pos_h, pos_l);
812 ssize_t ret = -EBADF;
818 file = fget_light(fd, &fput_needed);
821 if (file->f_mode & FMODE_PWRITE)
822 ret = vfs_writev(file, vec, vlen, &pos);
823 fput_light(file, fput_needed);
827 add_wchar(current, ret);
832 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
833 size_t count, loff_t max)
835 struct file * in_file, * out_file;
836 struct inode * in_inode, * out_inode;
839 int fput_needed_in, fput_needed_out, fl;
842 * Get input file, and verify that it is ok..
845 in_file = fget_light(in_fd, &fput_needed_in);
848 if (!(in_file->f_mode & FMODE_READ))
852 ppos = &in_file->f_pos;
854 if (!(in_file->f_mode & FMODE_PREAD))
856 retval = rw_verify_area(READ, in_file, ppos, count);
862 * Get output file, and verify that it is ok..
865 out_file = fget_light(out_fd, &fput_needed_out);
868 if (!(out_file->f_mode & FMODE_WRITE))
871 in_inode = in_file->f_path.dentry->d_inode;
872 out_inode = out_file->f_path.dentry->d_inode;
873 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
879 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
882 if (unlikely(pos + count > max)) {
892 * We need to debate whether we can enable this or not. The
893 * man page documents EAGAIN return for the output at least,
894 * and the application is arguably buggy if it doesn't expect
895 * EAGAIN on a non-blocking file descriptor.
897 if (in_file->f_flags & O_NONBLOCK)
898 fl = SPLICE_F_NONBLOCK;
900 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
903 add_rchar(current, retval);
904 add_wchar(current, retval);
913 fput_light(out_file, fput_needed_out);
915 fput_light(in_file, fput_needed_in);
920 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
927 if (unlikely(get_user(off, offset)))
930 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
931 if (unlikely(put_user(pos, offset)))
936 return do_sendfile(out_fd, in_fd, NULL, count, 0);
939 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
945 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
947 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
948 if (unlikely(put_user(pos, offset)))
953 return do_sendfile(out_fd, in_fd, NULL, count, 0);