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/aio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
22 #include <asm/uaccess.h>
23 #include <asm/unistd.h>
25 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
26 typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *,
27 unsigned long, loff_t);
28 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
30 const struct file_operations generic_ro_fops = {
31 .llseek = generic_file_llseek,
32 .read = new_sync_read,
33 .read_iter = generic_file_read_iter,
34 .mmap = generic_file_readonly_mmap,
35 .splice_read = generic_file_splice_read,
38 EXPORT_SYMBOL(generic_ro_fops);
40 static inline int unsigned_offsets(struct file *file)
42 return file->f_mode & FMODE_UNSIGNED_OFFSET;
46 * vfs_setpos - update the file offset for lseek
47 * @file: file structure in question
48 * @offset: file offset to seek to
49 * @maxsize: maximum file size
51 * This is a low-level filesystem helper for updating the file offset to
52 * the value specified by @offset if the given offset is valid and it is
53 * not equal to the current file offset.
55 * Return the specified offset on success and -EINVAL on invalid offset.
57 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
59 if (offset < 0 && !unsigned_offsets(file))
64 if (offset != file->f_pos) {
70 EXPORT_SYMBOL(vfs_setpos);
73 * generic_file_llseek_size - generic llseek implementation for regular files
74 * @file: file structure to seek on
75 * @offset: file offset to seek to
76 * @whence: type of seek
77 * @size: max size of this file in file system
78 * @eof: offset used for SEEK_END position
80 * This is a variant of generic_file_llseek that allows passing in a custom
81 * maximum file size and a custom EOF position, for e.g. hashed directories
84 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
85 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
86 * read/writes behave like SEEK_SET against seeks.
89 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
90 loff_t maxsize, loff_t eof)
98 * Here we special-case the lseek(fd, 0, SEEK_CUR)
99 * position-querying operation. Avoid rewriting the "same"
100 * f_pos value back to the file because a concurrent read(),
101 * write() or lseek() might have altered it
106 * f_lock protects against read/modify/write race with other
107 * SEEK_CURs. Note that parallel writes and reads behave
110 spin_lock(&file->f_lock);
111 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
112 spin_unlock(&file->f_lock);
116 * In the generic case the entire file is data, so as long as
117 * offset isn't at the end of the file then the offset is data.
124 * There is a virtual hole at the end of the file, so as long as
125 * offset isn't i_size or larger, return i_size.
133 return vfs_setpos(file, offset, maxsize);
135 EXPORT_SYMBOL(generic_file_llseek_size);
138 * generic_file_llseek - generic llseek implementation for regular files
139 * @file: file structure to seek on
140 * @offset: file offset to seek to
141 * @whence: type of seek
143 * This is a generic implemenation of ->llseek useable for all normal local
144 * filesystems. It just updates the file offset to the value specified by
145 * @offset and @whence.
147 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
149 struct inode *inode = file->f_mapping->host;
151 return generic_file_llseek_size(file, offset, whence,
152 inode->i_sb->s_maxbytes,
155 EXPORT_SYMBOL(generic_file_llseek);
158 * fixed_size_llseek - llseek implementation for fixed-sized devices
159 * @file: file structure to seek on
160 * @offset: file offset to seek to
161 * @whence: type of seek
162 * @size: size of the file
165 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
168 case SEEK_SET: case SEEK_CUR: case SEEK_END:
169 return generic_file_llseek_size(file, offset, whence,
175 EXPORT_SYMBOL(fixed_size_llseek);
178 * noop_llseek - No Operation Performed llseek implementation
179 * @file: file structure to seek on
180 * @offset: file offset to seek to
181 * @whence: type of seek
183 * This is an implementation of ->llseek useable for the rare special case when
184 * userspace expects the seek to succeed but the (device) file is actually not
185 * able to perform the seek. In this case you use noop_llseek() instead of
186 * falling back to the default implementation of ->llseek.
188 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
192 EXPORT_SYMBOL(noop_llseek);
194 loff_t no_llseek(struct file *file, loff_t offset, int whence)
198 EXPORT_SYMBOL(no_llseek);
200 loff_t default_llseek(struct file *file, loff_t offset, int whence)
202 struct inode *inode = file_inode(file);
205 mutex_lock(&inode->i_mutex);
208 offset += i_size_read(inode);
212 retval = file->f_pos;
215 offset += file->f_pos;
219 * In the generic case the entire file is data, so as
220 * long as offset isn't at the end of the file then the
223 if (offset >= inode->i_size) {
230 * There is a virtual hole at the end of the file, so
231 * as long as offset isn't i_size or larger, return
234 if (offset >= inode->i_size) {
238 offset = inode->i_size;
242 if (offset >= 0 || unsigned_offsets(file)) {
243 if (offset != file->f_pos) {
244 file->f_pos = offset;
250 mutex_unlock(&inode->i_mutex);
253 EXPORT_SYMBOL(default_llseek);
255 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
257 loff_t (*fn)(struct file *, loff_t, int);
260 if (file->f_mode & FMODE_LSEEK) {
261 if (file->f_op->llseek)
262 fn = file->f_op->llseek;
264 return fn(file, offset, whence);
266 EXPORT_SYMBOL(vfs_llseek);
268 static inline struct fd fdget_pos(int fd)
270 return __to_fd(__fdget_pos(fd));
273 static inline void fdput_pos(struct fd f)
275 if (f.flags & FDPUT_POS_UNLOCK)
276 mutex_unlock(&f.file->f_pos_lock);
280 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
283 struct fd f = fdget_pos(fd);
288 if (whence <= SEEK_MAX) {
289 loff_t res = vfs_llseek(f.file, offset, whence);
291 if (res != (loff_t)retval)
292 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
299 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
301 return sys_lseek(fd, offset, whence);
305 #ifdef __ARCH_WANT_SYS_LLSEEK
306 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
307 unsigned long, offset_low, loff_t __user *, result,
308 unsigned int, whence)
311 struct fd f = fdget_pos(fd);
318 if (whence > SEEK_MAX)
321 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
324 retval = (int)offset;
327 if (!copy_to_user(result, &offset, sizeof(offset)))
336 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
341 if (!file->f_op->read_iter)
344 init_sync_kiocb(&kiocb, file);
345 kiocb.ki_pos = *ppos;
348 ret = file->f_op->read_iter(&kiocb, iter);
349 if (ret == -EIOCBQUEUED)
350 ret = wait_on_sync_kiocb(&kiocb);
353 *ppos = kiocb.ki_pos;
356 EXPORT_SYMBOL(vfs_iter_read);
358 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
363 if (!file->f_op->write_iter)
366 init_sync_kiocb(&kiocb, file);
367 kiocb.ki_pos = *ppos;
370 ret = file->f_op->write_iter(&kiocb, iter);
371 if (ret == -EIOCBQUEUED)
372 ret = wait_on_sync_kiocb(&kiocb);
375 *ppos = kiocb.ki_pos;
378 EXPORT_SYMBOL(vfs_iter_write);
381 * rw_verify_area doesn't like huge counts. We limit
382 * them to something that fits in "int" so that others
383 * won't have to do range checks all the time.
385 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
389 int retval = -EINVAL;
391 inode = file_inode(file);
392 if (unlikely((ssize_t) count < 0))
395 if (unlikely(pos < 0)) {
396 if (!unsigned_offsets(file))
398 if (count >= -pos) /* both values are in 0..LLONG_MAX */
400 } else if (unlikely((loff_t) (pos + count) < 0)) {
401 if (!unsigned_offsets(file))
405 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
406 retval = locks_mandatory_area(
407 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
408 inode, file, pos, count);
412 retval = security_file_permission(file,
413 read_write == READ ? MAY_READ : MAY_WRITE);
416 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
419 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
421 struct iovec iov = { .iov_base = buf, .iov_len = len };
425 init_sync_kiocb(&kiocb, filp);
426 kiocb.ki_pos = *ppos;
428 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
429 if (-EIOCBQUEUED == ret)
430 ret = wait_on_sync_kiocb(&kiocb);
431 *ppos = kiocb.ki_pos;
435 EXPORT_SYMBOL(do_sync_read);
437 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
439 struct iovec iov = { .iov_base = buf, .iov_len = len };
441 struct iov_iter iter;
444 init_sync_kiocb(&kiocb, filp);
445 kiocb.ki_pos = *ppos;
446 iov_iter_init(&iter, READ, &iov, 1, len);
448 ret = filp->f_op->read_iter(&kiocb, &iter);
449 if (-EIOCBQUEUED == ret)
450 ret = wait_on_sync_kiocb(&kiocb);
451 *ppos = kiocb.ki_pos;
455 EXPORT_SYMBOL(new_sync_read);
457 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
462 if (file->f_op->read)
463 ret = file->f_op->read(file, buf, count, pos);
464 else if (file->f_op->aio_read)
465 ret = do_sync_read(file, buf, count, pos);
466 else if (file->f_op->read_iter)
467 ret = new_sync_read(file, buf, count, pos);
474 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
478 if (!(file->f_mode & FMODE_READ))
480 if (!(file->f_mode & FMODE_CAN_READ))
482 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
485 ret = rw_verify_area(READ, file, pos, count);
488 ret = __vfs_read(file, buf, count, pos);
490 fsnotify_access(file);
491 add_rchar(current, ret);
499 EXPORT_SYMBOL(vfs_read);
501 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
503 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
507 init_sync_kiocb(&kiocb, filp);
508 kiocb.ki_pos = *ppos;
510 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
511 if (-EIOCBQUEUED == ret)
512 ret = wait_on_sync_kiocb(&kiocb);
513 *ppos = kiocb.ki_pos;
517 EXPORT_SYMBOL(do_sync_write);
519 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
521 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
523 struct iov_iter iter;
526 init_sync_kiocb(&kiocb, filp);
527 kiocb.ki_pos = *ppos;
528 iov_iter_init(&iter, WRITE, &iov, 1, len);
530 ret = filp->f_op->write_iter(&kiocb, &iter);
531 if (-EIOCBQUEUED == ret)
532 ret = wait_on_sync_kiocb(&kiocb);
533 *ppos = kiocb.ki_pos;
537 EXPORT_SYMBOL(new_sync_write);
539 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
542 const char __user *p;
545 if (!(file->f_mode & FMODE_CAN_WRITE))
550 p = (__force const char __user *)buf;
551 if (count > MAX_RW_COUNT)
552 count = MAX_RW_COUNT;
553 if (file->f_op->write)
554 ret = file->f_op->write(file, p, count, pos);
555 else if (file->f_op->aio_write)
556 ret = do_sync_write(file, p, count, pos);
558 ret = new_sync_write(file, p, count, pos);
561 fsnotify_modify(file);
562 add_wchar(current, ret);
568 EXPORT_SYMBOL(__kernel_write);
570 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
574 if (!(file->f_mode & FMODE_WRITE))
576 if (!(file->f_mode & FMODE_CAN_WRITE))
578 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
581 ret = rw_verify_area(WRITE, file, pos, count);
584 file_start_write(file);
585 if (file->f_op->write)
586 ret = file->f_op->write(file, buf, count, pos);
587 else if (file->f_op->aio_write)
588 ret = do_sync_write(file, buf, count, pos);
590 ret = new_sync_write(file, buf, count, pos);
592 fsnotify_modify(file);
593 add_wchar(current, ret);
596 file_end_write(file);
602 EXPORT_SYMBOL(vfs_write);
604 static inline loff_t file_pos_read(struct file *file)
609 static inline void file_pos_write(struct file *file, loff_t pos)
614 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
616 struct fd f = fdget_pos(fd);
617 ssize_t ret = -EBADF;
620 loff_t pos = file_pos_read(f.file);
621 ret = vfs_read(f.file, buf, count, &pos);
623 file_pos_write(f.file, pos);
629 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
632 struct fd f = fdget_pos(fd);
633 ssize_t ret = -EBADF;
636 loff_t pos = file_pos_read(f.file);
637 ret = vfs_write(f.file, buf, count, &pos);
639 file_pos_write(f.file, pos);
646 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
647 size_t, count, loff_t, pos)
650 ssize_t ret = -EBADF;
658 if (f.file->f_mode & FMODE_PREAD)
659 ret = vfs_read(f.file, buf, count, &pos);
666 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
667 size_t, count, loff_t, pos)
670 ssize_t ret = -EBADF;
678 if (f.file->f_mode & FMODE_PWRITE)
679 ret = vfs_write(f.file, buf, count, &pos);
687 * Reduce an iovec's length in-place. Return the resulting number of segments
689 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
691 unsigned long seg = 0;
694 while (seg < nr_segs) {
696 if (len + iov->iov_len >= to) {
697 iov->iov_len = to - len;
705 EXPORT_SYMBOL(iov_shorten);
707 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
708 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
711 struct iov_iter iter;
714 init_sync_kiocb(&kiocb, filp);
715 kiocb.ki_pos = *ppos;
717 iov_iter_init(&iter, rw, iov, nr_segs, len);
718 ret = fn(&kiocb, &iter);
719 if (ret == -EIOCBQUEUED)
720 ret = wait_on_sync_kiocb(&kiocb);
721 *ppos = kiocb.ki_pos;
725 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
726 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
731 init_sync_kiocb(&kiocb, filp);
732 kiocb.ki_pos = *ppos;
734 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
735 if (ret == -EIOCBQUEUED)
736 ret = wait_on_sync_kiocb(&kiocb);
737 *ppos = kiocb.ki_pos;
741 /* Do it by hand, with file-ops */
742 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
743 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
745 struct iovec *vector = iov;
748 while (nr_segs > 0) {
753 base = vector->iov_base;
754 len = vector->iov_len;
758 nr = fn(filp, base, len, ppos);
773 /* A write operation does a read from user space and vice versa */
774 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
776 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
777 unsigned long nr_segs, unsigned long fast_segs,
778 struct iovec *fast_pointer,
779 struct iovec **ret_pointer)
783 struct iovec *iov = fast_pointer;
786 * SuS says "The readv() function *may* fail if the iovcnt argument
787 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
788 * traditionally returned zero for zero segments, so...
796 * First get the "struct iovec" from user memory and
797 * verify all the pointers
799 if (nr_segs > UIO_MAXIOV) {
803 if (nr_segs > fast_segs) {
804 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
810 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
816 * According to the Single Unix Specification we should return EINVAL
817 * if an element length is < 0 when cast to ssize_t or if the
818 * total length would overflow the ssize_t return value of the
821 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
825 for (seg = 0; seg < nr_segs; seg++) {
826 void __user *buf = iov[seg].iov_base;
827 ssize_t len = (ssize_t)iov[seg].iov_len;
829 /* see if we we're about to use an invalid len or if
830 * it's about to overflow ssize_t */
836 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
840 if (len > MAX_RW_COUNT - ret) {
841 len = MAX_RW_COUNT - ret;
842 iov[seg].iov_len = len;
851 static ssize_t do_readv_writev(int type, struct file *file,
852 const struct iovec __user * uvector,
853 unsigned long nr_segs, loff_t *pos)
856 struct iovec iovstack[UIO_FASTIOV];
857 struct iovec *iov = iovstack;
863 ret = rw_copy_check_uvector(type, uvector, nr_segs,
864 ARRAY_SIZE(iovstack), iovstack, &iov);
869 ret = rw_verify_area(type, file, pos, tot_len);
875 fn = file->f_op->read;
876 fnv = file->f_op->aio_read;
877 iter_fn = file->f_op->read_iter;
879 fn = (io_fn_t)file->f_op->write;
880 fnv = file->f_op->aio_write;
881 iter_fn = file->f_op->write_iter;
882 file_start_write(file);
886 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
889 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
892 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
895 file_end_write(file);
900 if ((ret + (type == READ)) > 0) {
902 fsnotify_access(file);
904 fsnotify_modify(file);
909 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
910 unsigned long vlen, loff_t *pos)
912 if (!(file->f_mode & FMODE_READ))
914 if (!(file->f_mode & FMODE_CAN_READ))
917 return do_readv_writev(READ, file, vec, vlen, pos);
920 EXPORT_SYMBOL(vfs_readv);
922 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
923 unsigned long vlen, loff_t *pos)
925 if (!(file->f_mode & FMODE_WRITE))
927 if (!(file->f_mode & FMODE_CAN_WRITE))
930 return do_readv_writev(WRITE, file, vec, vlen, pos);
933 EXPORT_SYMBOL(vfs_writev);
935 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
938 struct fd f = fdget_pos(fd);
939 ssize_t ret = -EBADF;
942 loff_t pos = file_pos_read(f.file);
943 ret = vfs_readv(f.file, vec, vlen, &pos);
945 file_pos_write(f.file, pos);
950 add_rchar(current, ret);
955 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
958 struct fd f = fdget_pos(fd);
959 ssize_t ret = -EBADF;
962 loff_t pos = file_pos_read(f.file);
963 ret = vfs_writev(f.file, vec, vlen, &pos);
965 file_pos_write(f.file, pos);
970 add_wchar(current, ret);
975 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
977 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
978 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
981 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
982 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
984 loff_t pos = pos_from_hilo(pos_h, pos_l);
986 ssize_t ret = -EBADF;
994 if (f.file->f_mode & FMODE_PREAD)
995 ret = vfs_readv(f.file, vec, vlen, &pos);
1000 add_rchar(current, ret);
1005 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1006 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1008 loff_t pos = pos_from_hilo(pos_h, pos_l);
1010 ssize_t ret = -EBADF;
1018 if (f.file->f_mode & FMODE_PWRITE)
1019 ret = vfs_writev(f.file, vec, vlen, &pos);
1024 add_wchar(current, ret);
1029 #ifdef CONFIG_COMPAT
1031 static ssize_t compat_do_readv_writev(int type, struct file *file,
1032 const struct compat_iovec __user *uvector,
1033 unsigned long nr_segs, loff_t *pos)
1035 compat_ssize_t tot_len;
1036 struct iovec iovstack[UIO_FASTIOV];
1037 struct iovec *iov = iovstack;
1043 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1044 UIO_FASTIOV, iovstack, &iov);
1049 ret = rw_verify_area(type, file, pos, tot_len);
1055 fn = file->f_op->read;
1056 fnv = file->f_op->aio_read;
1057 iter_fn = file->f_op->read_iter;
1059 fn = (io_fn_t)file->f_op->write;
1060 fnv = file->f_op->aio_write;
1061 iter_fn = file->f_op->write_iter;
1062 file_start_write(file);
1066 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1069 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1072 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1075 file_end_write(file);
1078 if (iov != iovstack)
1080 if ((ret + (type == READ)) > 0) {
1082 fsnotify_access(file);
1084 fsnotify_modify(file);
1089 static size_t compat_readv(struct file *file,
1090 const struct compat_iovec __user *vec,
1091 unsigned long vlen, loff_t *pos)
1093 ssize_t ret = -EBADF;
1095 if (!(file->f_mode & FMODE_READ))
1099 if (!(file->f_mode & FMODE_CAN_READ))
1102 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1106 add_rchar(current, ret);
1111 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1112 const struct compat_iovec __user *,vec,
1113 compat_ulong_t, vlen)
1115 struct fd f = fdget_pos(fd);
1121 pos = f.file->f_pos;
1122 ret = compat_readv(f.file, vec, vlen, &pos);
1124 f.file->f_pos = pos;
1129 static long __compat_sys_preadv64(unsigned long fd,
1130 const struct compat_iovec __user *vec,
1131 unsigned long vlen, loff_t pos)
1142 if (f.file->f_mode & FMODE_PREAD)
1143 ret = compat_readv(f.file, vec, vlen, &pos);
1148 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1149 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1150 const struct compat_iovec __user *,vec,
1151 unsigned long, vlen, loff_t, pos)
1153 return __compat_sys_preadv64(fd, vec, vlen, pos);
1157 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1158 const struct compat_iovec __user *,vec,
1159 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1161 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1163 return __compat_sys_preadv64(fd, vec, vlen, pos);
1166 static size_t compat_writev(struct file *file,
1167 const struct compat_iovec __user *vec,
1168 unsigned long vlen, loff_t *pos)
1170 ssize_t ret = -EBADF;
1172 if (!(file->f_mode & FMODE_WRITE))
1176 if (!(file->f_mode & FMODE_CAN_WRITE))
1179 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1183 add_wchar(current, ret);
1188 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1189 const struct compat_iovec __user *, vec,
1190 compat_ulong_t, vlen)
1192 struct fd f = fdget_pos(fd);
1198 pos = f.file->f_pos;
1199 ret = compat_writev(f.file, vec, vlen, &pos);
1201 f.file->f_pos = pos;
1206 static long __compat_sys_pwritev64(unsigned long fd,
1207 const struct compat_iovec __user *vec,
1208 unsigned long vlen, loff_t pos)
1219 if (f.file->f_mode & FMODE_PWRITE)
1220 ret = compat_writev(f.file, vec, vlen, &pos);
1225 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1226 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1227 const struct compat_iovec __user *,vec,
1228 unsigned long, vlen, loff_t, pos)
1230 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1234 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1235 const struct compat_iovec __user *,vec,
1236 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1238 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1240 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1244 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1245 size_t count, loff_t max)
1248 struct inode *in_inode, *out_inode;
1255 * Get input file, and verify that it is ok..
1261 if (!(in.file->f_mode & FMODE_READ))
1265 pos = in.file->f_pos;
1268 if (!(in.file->f_mode & FMODE_PREAD))
1271 retval = rw_verify_area(READ, in.file, &pos, count);
1277 * Get output file, and verify that it is ok..
1280 out = fdget(out_fd);
1283 if (!(out.file->f_mode & FMODE_WRITE))
1286 in_inode = file_inode(in.file);
1287 out_inode = file_inode(out.file);
1288 out_pos = out.file->f_pos;
1289 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1295 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1297 if (unlikely(pos + count > max)) {
1298 retval = -EOVERFLOW;
1307 * We need to debate whether we can enable this or not. The
1308 * man page documents EAGAIN return for the output at least,
1309 * and the application is arguably buggy if it doesn't expect
1310 * EAGAIN on a non-blocking file descriptor.
1312 if (in.file->f_flags & O_NONBLOCK)
1313 fl = SPLICE_F_NONBLOCK;
1315 file_start_write(out.file);
1316 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1317 file_end_write(out.file);
1320 add_rchar(current, retval);
1321 add_wchar(current, retval);
1322 fsnotify_access(in.file);
1323 fsnotify_modify(out.file);
1324 out.file->f_pos = out_pos;
1328 in.file->f_pos = pos;
1334 retval = -EOVERFLOW;
1344 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1351 if (unlikely(get_user(off, offset)))
1354 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1355 if (unlikely(put_user(pos, offset)))
1360 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1363 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1369 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1371 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1372 if (unlikely(put_user(pos, offset)))
1377 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1380 #ifdef CONFIG_COMPAT
1381 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1382 compat_off_t __user *, offset, compat_size_t, count)
1389 if (unlikely(get_user(off, offset)))
1392 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1393 if (unlikely(put_user(pos, offset)))
1398 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1401 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1402 compat_loff_t __user *, offset, compat_size_t, count)
1408 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1410 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1411 if (unlikely(put_user(pos, offset)))
1416 return do_sendfile(out_fd, in_fd, NULL, count, 0);