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) {
176 * There is a virtual hole at the end of the file, so
177 * as long as offset isn't i_size or larger, return
180 if (offset >= inode->i_size) {
184 offset = inode->i_size;
188 if (offset >= 0 || unsigned_offsets(file)) {
189 if (offset != file->f_pos) {
190 file->f_pos = offset;
196 mutex_unlock(&inode->i_mutex);
199 EXPORT_SYMBOL(default_llseek);
201 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
203 loff_t (*fn)(struct file *, loff_t, int);
206 if (file->f_mode & FMODE_LSEEK) {
207 if (file->f_op && file->f_op->llseek)
208 fn = file->f_op->llseek;
210 return fn(file, offset, origin);
212 EXPORT_SYMBOL(vfs_llseek);
214 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
221 file = fget_light(fd, &fput_needed);
226 if (origin <= SEEK_MAX) {
227 loff_t res = vfs_llseek(file, offset, origin);
229 if (res != (loff_t)retval)
230 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
232 fput_light(file, fput_needed);
237 #ifdef __ARCH_WANT_SYS_LLSEEK
238 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
239 unsigned long, offset_low, loff_t __user *, result,
240 unsigned int, origin)
248 file = fget_light(fd, &fput_needed);
253 if (origin > SEEK_MAX)
256 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
259 retval = (int)offset;
262 if (!copy_to_user(result, &offset, sizeof(offset)))
266 fput_light(file, fput_needed);
274 * rw_verify_area doesn't like huge counts. We limit
275 * them to something that fits in "int" so that others
276 * won't have to do range checks all the time.
278 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
282 int retval = -EINVAL;
284 inode = file->f_path.dentry->d_inode;
285 if (unlikely((ssize_t) count < 0))
288 if (unlikely(pos < 0)) {
289 if (!unsigned_offsets(file))
291 if (count >= -pos) /* both values are in 0..LLONG_MAX */
293 } else if (unlikely((loff_t) (pos + count) < 0)) {
294 if (!unsigned_offsets(file))
298 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
299 retval = locks_mandatory_area(
300 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
301 inode, file, pos, count);
305 retval = security_file_permission(file,
306 read_write == READ ? MAY_READ : MAY_WRITE);
309 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
312 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
314 set_current_state(TASK_UNINTERRUPTIBLE);
315 if (!kiocbIsKicked(iocb))
318 kiocbClearKicked(iocb);
319 __set_current_state(TASK_RUNNING);
322 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
324 struct iovec iov = { .iov_base = buf, .iov_len = len };
328 init_sync_kiocb(&kiocb, filp);
329 kiocb.ki_pos = *ppos;
331 kiocb.ki_nbytes = len;
334 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
335 if (ret != -EIOCBRETRY)
337 wait_on_retry_sync_kiocb(&kiocb);
340 if (-EIOCBQUEUED == ret)
341 ret = wait_on_sync_kiocb(&kiocb);
342 *ppos = kiocb.ki_pos;
346 EXPORT_SYMBOL(do_sync_read);
348 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
352 if (!(file->f_mode & FMODE_READ))
354 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
356 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
359 ret = rw_verify_area(READ, file, pos, count);
362 if (file->f_op->read)
363 ret = file->f_op->read(file, buf, count, pos);
365 ret = do_sync_read(file, buf, count, pos);
367 fsnotify_access(file);
368 add_rchar(current, ret);
376 EXPORT_SYMBOL(vfs_read);
378 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
380 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
384 init_sync_kiocb(&kiocb, filp);
385 kiocb.ki_pos = *ppos;
387 kiocb.ki_nbytes = len;
390 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
391 if (ret != -EIOCBRETRY)
393 wait_on_retry_sync_kiocb(&kiocb);
396 if (-EIOCBQUEUED == ret)
397 ret = wait_on_sync_kiocb(&kiocb);
398 *ppos = kiocb.ki_pos;
402 EXPORT_SYMBOL(do_sync_write);
404 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
408 if (!(file->f_mode & FMODE_WRITE))
410 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
412 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
415 ret = rw_verify_area(WRITE, file, pos, count);
418 if (file->f_op->write)
419 ret = file->f_op->write(file, buf, count, pos);
421 ret = do_sync_write(file, buf, count, pos);
423 fsnotify_modify(file);
424 add_wchar(current, ret);
432 EXPORT_SYMBOL(vfs_write);
434 static inline loff_t file_pos_read(struct file *file)
439 static inline void file_pos_write(struct file *file, loff_t pos)
444 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
447 ssize_t ret = -EBADF;
450 file = fget_light(fd, &fput_needed);
452 loff_t pos = file_pos_read(file);
453 ret = vfs_read(file, buf, count, &pos);
454 file_pos_write(file, pos);
455 fput_light(file, fput_needed);
461 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
465 ssize_t ret = -EBADF;
468 file = fget_light(fd, &fput_needed);
470 loff_t pos = file_pos_read(file);
471 ret = vfs_write(file, buf, count, &pos);
472 file_pos_write(file, pos);
473 fput_light(file, fput_needed);
479 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
480 size_t count, loff_t pos)
483 ssize_t ret = -EBADF;
489 file = fget_light(fd, &fput_needed);
492 if (file->f_mode & FMODE_PREAD)
493 ret = vfs_read(file, buf, count, &pos);
494 fput_light(file, fput_needed);
499 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
500 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
502 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
503 (size_t) count, pos);
505 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
508 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
509 size_t count, loff_t pos)
512 ssize_t ret = -EBADF;
518 file = fget_light(fd, &fput_needed);
521 if (file->f_mode & FMODE_PWRITE)
522 ret = vfs_write(file, buf, count, &pos);
523 fput_light(file, fput_needed);
528 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
529 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
531 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
532 (size_t) count, pos);
534 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
538 * Reduce an iovec's length in-place. Return the resulting number of segments
540 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
542 unsigned long seg = 0;
545 while (seg < nr_segs) {
547 if (len + iov->iov_len >= to) {
548 iov->iov_len = to - len;
556 EXPORT_SYMBOL(iov_shorten);
558 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
559 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
564 init_sync_kiocb(&kiocb, filp);
565 kiocb.ki_pos = *ppos;
567 kiocb.ki_nbytes = len;
570 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
571 if (ret != -EIOCBRETRY)
573 wait_on_retry_sync_kiocb(&kiocb);
576 if (ret == -EIOCBQUEUED)
577 ret = wait_on_sync_kiocb(&kiocb);
578 *ppos = kiocb.ki_pos;
582 /* Do it by hand, with file-ops */
583 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
584 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
586 struct iovec *vector = iov;
589 while (nr_segs > 0) {
594 base = vector->iov_base;
595 len = vector->iov_len;
599 nr = fn(filp, base, len, ppos);
614 /* A write operation does a read from user space and vice versa */
615 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
617 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
618 unsigned long nr_segs, unsigned long fast_segs,
619 struct iovec *fast_pointer,
620 struct iovec **ret_pointer)
624 struct iovec *iov = fast_pointer;
627 * SuS says "The readv() function *may* fail if the iovcnt argument
628 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
629 * traditionally returned zero for zero segments, so...
637 * First get the "struct iovec" from user memory and
638 * verify all the pointers
640 if (nr_segs > UIO_MAXIOV) {
644 if (nr_segs > fast_segs) {
645 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
651 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
657 * According to the Single Unix Specification we should return EINVAL
658 * if an element length is < 0 when cast to ssize_t or if the
659 * total length would overflow the ssize_t return value of the
662 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
666 for (seg = 0; seg < nr_segs; seg++) {
667 void __user *buf = iov[seg].iov_base;
668 ssize_t len = (ssize_t)iov[seg].iov_len;
670 /* see if we we're about to use an invalid len or if
671 * it's about to overflow ssize_t */
676 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
680 if (len > MAX_RW_COUNT - ret) {
681 len = MAX_RW_COUNT - ret;
682 iov[seg].iov_len = len;
691 static ssize_t do_readv_writev(int type, struct file *file,
692 const struct iovec __user * uvector,
693 unsigned long nr_segs, loff_t *pos)
696 struct iovec iovstack[UIO_FASTIOV];
697 struct iovec *iov = iovstack;
707 ret = rw_copy_check_uvector(type, uvector, nr_segs,
708 ARRAY_SIZE(iovstack), iovstack, &iov);
713 ret = rw_verify_area(type, file, pos, tot_len);
719 fn = file->f_op->read;
720 fnv = file->f_op->aio_read;
722 fn = (io_fn_t)file->f_op->write;
723 fnv = file->f_op->aio_write;
727 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
730 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
735 if ((ret + (type == READ)) > 0) {
737 fsnotify_access(file);
739 fsnotify_modify(file);
744 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
745 unsigned long vlen, loff_t *pos)
747 if (!(file->f_mode & FMODE_READ))
749 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
752 return do_readv_writev(READ, file, vec, vlen, pos);
755 EXPORT_SYMBOL(vfs_readv);
757 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
758 unsigned long vlen, loff_t *pos)
760 if (!(file->f_mode & FMODE_WRITE))
762 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
765 return do_readv_writev(WRITE, file, vec, vlen, pos);
768 EXPORT_SYMBOL(vfs_writev);
770 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
774 ssize_t ret = -EBADF;
777 file = fget_light(fd, &fput_needed);
779 loff_t pos = file_pos_read(file);
780 ret = vfs_readv(file, vec, vlen, &pos);
781 file_pos_write(file, pos);
782 fput_light(file, fput_needed);
786 add_rchar(current, ret);
791 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
795 ssize_t ret = -EBADF;
798 file = fget_light(fd, &fput_needed);
800 loff_t pos = file_pos_read(file);
801 ret = vfs_writev(file, vec, vlen, &pos);
802 file_pos_write(file, pos);
803 fput_light(file, fput_needed);
807 add_wchar(current, ret);
812 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
814 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
815 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
818 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
819 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
821 loff_t pos = pos_from_hilo(pos_h, pos_l);
823 ssize_t ret = -EBADF;
829 file = fget_light(fd, &fput_needed);
832 if (file->f_mode & FMODE_PREAD)
833 ret = vfs_readv(file, vec, vlen, &pos);
834 fput_light(file, fput_needed);
838 add_rchar(current, ret);
843 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
844 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
846 loff_t pos = pos_from_hilo(pos_h, pos_l);
848 ssize_t ret = -EBADF;
854 file = fget_light(fd, &fput_needed);
857 if (file->f_mode & FMODE_PWRITE)
858 ret = vfs_writev(file, vec, vlen, &pos);
859 fput_light(file, fput_needed);
863 add_wchar(current, ret);
868 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
869 size_t count, loff_t max)
871 struct file * in_file, * out_file;
872 struct inode * in_inode, * out_inode;
875 int fput_needed_in, fput_needed_out, fl;
878 * Get input file, and verify that it is ok..
881 in_file = fget_light(in_fd, &fput_needed_in);
884 if (!(in_file->f_mode & FMODE_READ))
888 ppos = &in_file->f_pos;
890 if (!(in_file->f_mode & FMODE_PREAD))
892 retval = rw_verify_area(READ, in_file, ppos, count);
898 * Get output file, and verify that it is ok..
901 out_file = fget_light(out_fd, &fput_needed_out);
904 if (!(out_file->f_mode & FMODE_WRITE))
907 in_inode = in_file->f_path.dentry->d_inode;
908 out_inode = out_file->f_path.dentry->d_inode;
909 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
915 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
918 if (unlikely(pos + count > max)) {
928 * We need to debate whether we can enable this or not. The
929 * man page documents EAGAIN return for the output at least,
930 * and the application is arguably buggy if it doesn't expect
931 * EAGAIN on a non-blocking file descriptor.
933 if (in_file->f_flags & O_NONBLOCK)
934 fl = SPLICE_F_NONBLOCK;
936 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
939 add_rchar(current, retval);
940 add_wchar(current, retval);
949 fput_light(out_file, fput_needed_out);
951 fput_light(in_file, fput_needed_in);
956 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
963 if (unlikely(get_user(off, offset)))
966 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
967 if (unlikely(put_user(pos, offset)))
972 return do_sendfile(out_fd, in_fd, NULL, count, 0);
975 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
981 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
983 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
984 if (unlikely(put_user(pos, offset)))
989 return do_sendfile(out_fd, in_fd, NULL, count, 0);