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/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
23 #include <asm/uaccess.h>
24 #include <asm/unistd.h>
26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
36 EXPORT_SYMBOL(generic_ro_fops);
38 static inline int unsigned_offsets(struct file *file)
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
57 if (offset < 0 && !unsigned_offsets(file))
62 if (offset != file->f_pos) {
68 EXPORT_SYMBOL(vfs_setpos);
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
131 return vfs_setpos(file, offset, maxsize);
133 EXPORT_SYMBOL(generic_file_llseek_size);
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
147 struct inode *inode = file->f_mapping->host;
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
153 EXPORT_SYMBOL(generic_file_llseek);
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
173 EXPORT_SYMBOL(fixed_size_llseek);
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
192 EXPORT_SYMBOL(no_seek_end_llseek);
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
212 EXPORT_SYMBOL(no_seek_end_llseek_size);
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
225 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
229 EXPORT_SYMBOL(noop_llseek);
231 loff_t no_llseek(struct file *file, loff_t offset, int whence)
235 EXPORT_SYMBOL(no_llseek);
237 loff_t default_llseek(struct file *file, loff_t offset, int whence)
239 struct inode *inode = file_inode(file);
245 offset += i_size_read(inode);
249 retval = file->f_pos;
252 offset += file->f_pos;
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
260 if (offset >= inode->i_size) {
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
271 if (offset >= inode->i_size) {
275 offset = inode->i_size;
279 if (offset >= 0 || unsigned_offsets(file)) {
280 if (offset != file->f_pos) {
281 file->f_pos = offset;
290 EXPORT_SYMBOL(default_llseek);
292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
294 loff_t (*fn)(struct file *, loff_t, int);
297 if (file->f_mode & FMODE_LSEEK) {
298 if (file->f_op->llseek)
299 fn = file->f_op->llseek;
301 return fn(file, offset, whence);
303 EXPORT_SYMBOL(vfs_llseek);
305 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
308 struct fd f = fdget_pos(fd);
313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
324 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
326 return sys_lseek(fd, offset, whence);
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
333 unsigned int, whence)
336 struct fd f = fdget_pos(fd);
343 if (whence > SEEK_MAX)
346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
349 retval = (int)offset;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
361 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
366 if (!file->f_op->read_iter)
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
373 ret = file->f_op->read_iter(&kiocb, iter);
374 BUG_ON(ret == -EIOCBQUEUED);
376 *ppos = kiocb.ki_pos;
379 EXPORT_SYMBOL(vfs_iter_read);
381 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
386 if (!file->f_op->write_iter)
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
393 ret = file->f_op->write_iter(&kiocb, iter);
394 BUG_ON(ret == -EIOCBQUEUED);
396 *ppos = kiocb.ki_pos;
399 EXPORT_SYMBOL(vfs_iter_write);
401 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
405 int retval = -EINVAL;
407 inode = file_inode(file);
408 if (unlikely((ssize_t) count < 0))
411 if (unlikely(pos < 0)) {
412 if (!unsigned_offsets(file))
414 if (count >= -pos) /* both values are in 0..LLONG_MAX */
416 } else if (unlikely((loff_t) (pos + count) < 0)) {
417 if (!unsigned_offsets(file))
421 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
422 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
423 read_write == READ ? F_RDLCK : F_WRLCK);
427 return security_file_permission(file,
428 read_write == READ ? MAY_READ : MAY_WRITE);
431 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
433 struct iovec iov = { .iov_base = buf, .iov_len = len };
435 struct iov_iter iter;
438 init_sync_kiocb(&kiocb, filp);
439 kiocb.ki_pos = *ppos;
440 iov_iter_init(&iter, READ, &iov, 1, len);
442 ret = filp->f_op->read_iter(&kiocb, &iter);
443 BUG_ON(ret == -EIOCBQUEUED);
444 *ppos = kiocb.ki_pos;
448 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
451 if (file->f_op->read)
452 return file->f_op->read(file, buf, count, pos);
453 else if (file->f_op->read_iter)
454 return new_sync_read(file, buf, count, pos);
458 EXPORT_SYMBOL(__vfs_read);
460 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
464 if (!(file->f_mode & FMODE_READ))
466 if (!(file->f_mode & FMODE_CAN_READ))
468 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
471 ret = rw_verify_area(READ, file, pos, count);
473 if (count > MAX_RW_COUNT)
474 count = MAX_RW_COUNT;
475 ret = __vfs_read(file, buf, count, pos);
477 fsnotify_access(file);
478 add_rchar(current, ret);
486 EXPORT_SYMBOL(vfs_read);
488 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
490 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
492 struct iov_iter iter;
495 init_sync_kiocb(&kiocb, filp);
496 kiocb.ki_pos = *ppos;
497 iov_iter_init(&iter, WRITE, &iov, 1, len);
499 ret = filp->f_op->write_iter(&kiocb, &iter);
500 BUG_ON(ret == -EIOCBQUEUED);
502 *ppos = kiocb.ki_pos;
506 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
509 if (file->f_op->write)
510 return file->f_op->write(file, p, count, pos);
511 else if (file->f_op->write_iter)
512 return new_sync_write(file, p, count, pos);
516 EXPORT_SYMBOL(__vfs_write);
518 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
521 const char __user *p;
524 if (!(file->f_mode & FMODE_CAN_WRITE))
529 p = (__force const char __user *)buf;
530 if (count > MAX_RW_COUNT)
531 count = MAX_RW_COUNT;
532 ret = __vfs_write(file, p, count, pos);
535 fsnotify_modify(file);
536 add_wchar(current, ret);
542 EXPORT_SYMBOL(__kernel_write);
544 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
548 if (!(file->f_mode & FMODE_WRITE))
550 if (!(file->f_mode & FMODE_CAN_WRITE))
552 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
555 ret = rw_verify_area(WRITE, file, pos, count);
557 if (count > MAX_RW_COUNT)
558 count = MAX_RW_COUNT;
559 file_start_write(file);
560 ret = __vfs_write(file, buf, count, pos);
562 fsnotify_modify(file);
563 add_wchar(current, ret);
566 file_end_write(file);
572 EXPORT_SYMBOL(vfs_write);
574 static inline loff_t file_pos_read(struct file *file)
579 static inline void file_pos_write(struct file *file, loff_t pos)
584 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
586 struct fd f = fdget_pos(fd);
587 ssize_t ret = -EBADF;
590 loff_t pos = file_pos_read(f.file);
591 ret = vfs_read(f.file, buf, count, &pos);
593 file_pos_write(f.file, pos);
599 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
602 struct fd f = fdget_pos(fd);
603 ssize_t ret = -EBADF;
606 loff_t pos = file_pos_read(f.file);
607 ret = vfs_write(f.file, buf, count, &pos);
609 file_pos_write(f.file, pos);
616 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
617 size_t, count, loff_t, pos)
620 ssize_t ret = -EBADF;
628 if (f.file->f_mode & FMODE_PREAD)
629 ret = vfs_read(f.file, buf, count, &pos);
636 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
637 size_t, count, loff_t, pos)
640 ssize_t ret = -EBADF;
648 if (f.file->f_mode & FMODE_PWRITE)
649 ret = vfs_write(f.file, buf, count, &pos);
657 * Reduce an iovec's length in-place. Return the resulting number of segments
659 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
661 unsigned long seg = 0;
664 while (seg < nr_segs) {
666 if (len + iov->iov_len >= to) {
667 iov->iov_len = to - len;
675 EXPORT_SYMBOL(iov_shorten);
677 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
678 loff_t *ppos, iter_fn_t fn, int flags)
683 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
686 init_sync_kiocb(&kiocb, filp);
687 if (flags & RWF_HIPRI)
688 kiocb.ki_flags |= IOCB_HIPRI;
689 if (flags & RWF_DSYNC)
690 kiocb.ki_flags |= IOCB_DSYNC;
691 if (flags & RWF_SYNC)
692 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
693 kiocb.ki_pos = *ppos;
695 ret = fn(&kiocb, iter);
696 BUG_ON(ret == -EIOCBQUEUED);
697 *ppos = kiocb.ki_pos;
701 /* Do it by hand, with file-ops */
702 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
703 loff_t *ppos, io_fn_t fn, int flags)
707 if (flags & ~RWF_HIPRI)
710 while (iov_iter_count(iter)) {
711 struct iovec iovec = iov_iter_iovec(iter);
714 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
722 if (nr != iovec.iov_len)
724 iov_iter_advance(iter, nr);
730 /* A write operation does a read from user space and vice versa */
731 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
733 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
734 unsigned long nr_segs, unsigned long fast_segs,
735 struct iovec *fast_pointer,
736 struct iovec **ret_pointer)
740 struct iovec *iov = fast_pointer;
743 * SuS says "The readv() function *may* fail if the iovcnt argument
744 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
745 * traditionally returned zero for zero segments, so...
753 * First get the "struct iovec" from user memory and
754 * verify all the pointers
756 if (nr_segs > UIO_MAXIOV) {
760 if (nr_segs > fast_segs) {
761 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
767 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
773 * According to the Single Unix Specification we should return EINVAL
774 * if an element length is < 0 when cast to ssize_t or if the
775 * total length would overflow the ssize_t return value of the
778 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
782 for (seg = 0; seg < nr_segs; seg++) {
783 void __user *buf = iov[seg].iov_base;
784 ssize_t len = (ssize_t)iov[seg].iov_len;
786 /* see if we we're about to use an invalid len or if
787 * it's about to overflow ssize_t */
793 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
797 if (len > MAX_RW_COUNT - ret) {
798 len = MAX_RW_COUNT - ret;
799 iov[seg].iov_len = len;
808 static ssize_t do_readv_writev(int type, struct file *file,
809 const struct iovec __user * uvector,
810 unsigned long nr_segs, loff_t *pos,
814 struct iovec iovstack[UIO_FASTIOV];
815 struct iovec *iov = iovstack;
816 struct iov_iter iter;
821 ret = import_iovec(type, uvector, nr_segs,
822 ARRAY_SIZE(iovstack), &iov, &iter);
826 tot_len = iov_iter_count(&iter);
829 ret = rw_verify_area(type, file, pos, tot_len);
834 fn = file->f_op->read;
835 iter_fn = file->f_op->read_iter;
837 fn = (io_fn_t)file->f_op->write;
838 iter_fn = file->f_op->write_iter;
839 file_start_write(file);
843 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
845 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
848 file_end_write(file);
852 if ((ret + (type == READ)) > 0) {
854 fsnotify_access(file);
856 fsnotify_modify(file);
861 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
862 unsigned long vlen, loff_t *pos, int flags)
864 if (!(file->f_mode & FMODE_READ))
866 if (!(file->f_mode & FMODE_CAN_READ))
869 return do_readv_writev(READ, file, vec, vlen, pos, flags);
872 EXPORT_SYMBOL(vfs_readv);
874 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
875 unsigned long vlen, loff_t *pos, int flags)
877 if (!(file->f_mode & FMODE_WRITE))
879 if (!(file->f_mode & FMODE_CAN_WRITE))
882 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
885 EXPORT_SYMBOL(vfs_writev);
887 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
888 unsigned long vlen, int flags)
890 struct fd f = fdget_pos(fd);
891 ssize_t ret = -EBADF;
894 loff_t pos = file_pos_read(f.file);
895 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
897 file_pos_write(f.file, pos);
902 add_rchar(current, ret);
907 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
908 unsigned long vlen, int flags)
910 struct fd f = fdget_pos(fd);
911 ssize_t ret = -EBADF;
914 loff_t pos = file_pos_read(f.file);
915 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
917 file_pos_write(f.file, pos);
922 add_wchar(current, ret);
927 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
929 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
930 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
933 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
934 unsigned long vlen, loff_t pos, int flags)
937 ssize_t ret = -EBADF;
945 if (f.file->f_mode & FMODE_PREAD)
946 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
951 add_rchar(current, ret);
956 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
957 unsigned long vlen, loff_t pos, int flags)
960 ssize_t ret = -EBADF;
968 if (f.file->f_mode & FMODE_PWRITE)
969 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
974 add_wchar(current, ret);
979 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
982 return do_readv(fd, vec, vlen, 0);
985 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
988 return do_writev(fd, vec, vlen, 0);
991 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
992 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
994 loff_t pos = pos_from_hilo(pos_h, pos_l);
996 return do_preadv(fd, vec, vlen, pos, 0);
999 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1000 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1003 loff_t pos = pos_from_hilo(pos_h, pos_l);
1006 return do_readv(fd, vec, vlen, flags);
1008 return do_preadv(fd, vec, vlen, pos, flags);
1011 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1012 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1014 loff_t pos = pos_from_hilo(pos_h, pos_l);
1016 return do_pwritev(fd, vec, vlen, pos, 0);
1019 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1020 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1023 loff_t pos = pos_from_hilo(pos_h, pos_l);
1026 return do_writev(fd, vec, vlen, flags);
1028 return do_pwritev(fd, vec, vlen, pos, flags);
1031 #ifdef CONFIG_COMPAT
1033 static ssize_t compat_do_readv_writev(int type, struct file *file,
1034 const struct compat_iovec __user *uvector,
1035 unsigned long nr_segs, loff_t *pos,
1038 compat_ssize_t tot_len;
1039 struct iovec iovstack[UIO_FASTIOV];
1040 struct iovec *iov = iovstack;
1041 struct iov_iter iter;
1046 ret = compat_import_iovec(type, uvector, nr_segs,
1047 UIO_FASTIOV, &iov, &iter);
1051 tot_len = iov_iter_count(&iter);
1054 ret = rw_verify_area(type, file, pos, tot_len);
1059 fn = file->f_op->read;
1060 iter_fn = file->f_op->read_iter;
1062 fn = (io_fn_t)file->f_op->write;
1063 iter_fn = file->f_op->write_iter;
1064 file_start_write(file);
1068 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1070 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1073 file_end_write(file);
1077 if ((ret + (type == READ)) > 0) {
1079 fsnotify_access(file);
1081 fsnotify_modify(file);
1086 static size_t compat_readv(struct file *file,
1087 const struct compat_iovec __user *vec,
1088 unsigned long vlen, loff_t *pos, int flags)
1090 ssize_t ret = -EBADF;
1092 if (!(file->f_mode & FMODE_READ))
1096 if (!(file->f_mode & FMODE_CAN_READ))
1099 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1103 add_rchar(current, ret);
1108 static size_t do_compat_readv(compat_ulong_t fd,
1109 const struct compat_iovec __user *vec,
1110 compat_ulong_t vlen, int flags)
1112 struct fd f = fdget_pos(fd);
1118 pos = f.file->f_pos;
1119 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1121 f.file->f_pos = pos;
1127 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1128 const struct compat_iovec __user *,vec,
1129 compat_ulong_t, vlen)
1131 return do_compat_readv(fd, vec, vlen, 0);
1134 static long do_compat_preadv64(unsigned long fd,
1135 const struct compat_iovec __user *vec,
1136 unsigned long vlen, loff_t pos, int flags)
1147 if (f.file->f_mode & FMODE_PREAD)
1148 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1153 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1154 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1155 const struct compat_iovec __user *,vec,
1156 unsigned long, vlen, loff_t, pos)
1158 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1162 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1163 const struct compat_iovec __user *,vec,
1164 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1166 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1168 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1171 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1172 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1173 const struct compat_iovec __user *,vec,
1174 unsigned long, vlen, loff_t, pos, int, flags)
1176 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1180 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1181 const struct compat_iovec __user *,vec,
1182 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1185 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1188 return do_compat_readv(fd, vec, vlen, flags);
1190 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1193 static size_t compat_writev(struct file *file,
1194 const struct compat_iovec __user *vec,
1195 unsigned long vlen, loff_t *pos, int flags)
1197 ssize_t ret = -EBADF;
1199 if (!(file->f_mode & FMODE_WRITE))
1203 if (!(file->f_mode & FMODE_CAN_WRITE))
1206 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1210 add_wchar(current, ret);
1215 static size_t do_compat_writev(compat_ulong_t fd,
1216 const struct compat_iovec __user* vec,
1217 compat_ulong_t vlen, int flags)
1219 struct fd f = fdget_pos(fd);
1225 pos = f.file->f_pos;
1226 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1228 f.file->f_pos = pos;
1233 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1234 const struct compat_iovec __user *, vec,
1235 compat_ulong_t, vlen)
1237 return do_compat_writev(fd, vec, vlen, 0);
1240 static long do_compat_pwritev64(unsigned long fd,
1241 const struct compat_iovec __user *vec,
1242 unsigned long vlen, loff_t pos, int flags)
1253 if (f.file->f_mode & FMODE_PWRITE)
1254 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1259 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1260 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1261 const struct compat_iovec __user *,vec,
1262 unsigned long, vlen, loff_t, pos)
1264 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1268 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1269 const struct compat_iovec __user *,vec,
1270 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1272 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1274 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1277 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1278 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1279 const struct compat_iovec __user *,vec,
1280 unsigned long, vlen, loff_t, pos, int, flags)
1282 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1286 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1287 const struct compat_iovec __user *,vec,
1288 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1290 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1293 return do_compat_writev(fd, vec, vlen, flags);
1295 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1300 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1301 size_t count, loff_t max)
1304 struct inode *in_inode, *out_inode;
1311 * Get input file, and verify that it is ok..
1317 if (!(in.file->f_mode & FMODE_READ))
1321 pos = in.file->f_pos;
1324 if (!(in.file->f_mode & FMODE_PREAD))
1327 retval = rw_verify_area(READ, in.file, &pos, count);
1330 if (count > MAX_RW_COUNT)
1331 count = MAX_RW_COUNT;
1334 * Get output file, and verify that it is ok..
1337 out = fdget(out_fd);
1340 if (!(out.file->f_mode & FMODE_WRITE))
1343 in_inode = file_inode(in.file);
1344 out_inode = file_inode(out.file);
1345 out_pos = out.file->f_pos;
1346 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1351 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1353 if (unlikely(pos + count > max)) {
1354 retval = -EOVERFLOW;
1363 * We need to debate whether we can enable this or not. The
1364 * man page documents EAGAIN return for the output at least,
1365 * and the application is arguably buggy if it doesn't expect
1366 * EAGAIN on a non-blocking file descriptor.
1368 if (in.file->f_flags & O_NONBLOCK)
1369 fl = SPLICE_F_NONBLOCK;
1371 file_start_write(out.file);
1372 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1373 file_end_write(out.file);
1376 add_rchar(current, retval);
1377 add_wchar(current, retval);
1378 fsnotify_access(in.file);
1379 fsnotify_modify(out.file);
1380 out.file->f_pos = out_pos;
1384 in.file->f_pos = pos;
1390 retval = -EOVERFLOW;
1400 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1407 if (unlikely(get_user(off, offset)))
1410 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1411 if (unlikely(put_user(pos, offset)))
1416 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1419 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1425 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1427 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1428 if (unlikely(put_user(pos, offset)))
1433 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1436 #ifdef CONFIG_COMPAT
1437 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1438 compat_off_t __user *, offset, compat_size_t, count)
1445 if (unlikely(get_user(off, offset)))
1448 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1449 if (unlikely(put_user(pos, offset)))
1454 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1457 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1458 compat_loff_t __user *, offset, compat_size_t, count)
1464 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1466 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1467 if (unlikely(put_user(pos, offset)))
1472 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1477 * copy_file_range() differs from regular file read and write in that it
1478 * specifically allows return partial success. When it does so is up to
1479 * the copy_file_range method.
1481 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1482 struct file *file_out, loff_t pos_out,
1483 size_t len, unsigned int flags)
1485 struct inode *inode_in = file_inode(file_in);
1486 struct inode *inode_out = file_inode(file_out);
1492 ret = rw_verify_area(READ, file_in, &pos_in, len);
1496 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1500 if (!(file_in->f_mode & FMODE_READ) ||
1501 !(file_out->f_mode & FMODE_WRITE) ||
1502 (file_out->f_flags & O_APPEND))
1505 /* this could be relaxed once a method supports cross-fs copies */
1506 if (inode_in->i_sb != inode_out->i_sb)
1512 ret = mnt_want_write_file(file_out);
1517 if (file_out->f_op->copy_file_range)
1518 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1519 pos_out, len, flags);
1520 if (ret == -EOPNOTSUPP)
1521 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1522 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1525 fsnotify_access(file_in);
1526 add_rchar(current, ret);
1527 fsnotify_modify(file_out);
1528 add_wchar(current, ret);
1533 mnt_drop_write_file(file_out);
1537 EXPORT_SYMBOL(vfs_copy_file_range);
1539 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1540 int, fd_out, loff_t __user *, off_out,
1541 size_t, len, unsigned int, flags)
1547 ssize_t ret = -EBADF;
1549 f_in = fdget(fd_in);
1553 f_out = fdget(fd_out);
1559 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1562 pos_in = f_in.file->f_pos;
1566 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1569 pos_out = f_out.file->f_pos;
1572 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1579 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1582 f_in.file->f_pos = pos_in;
1586 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1589 f_out.file->f_pos = pos_out;
1601 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1603 struct inode *inode = file_inode(file);
1605 if (unlikely(pos < 0))
1608 if (unlikely((loff_t) (pos + len) < 0))
1611 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1612 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1615 retval = locks_mandatory_area(inode, file, pos, end,
1616 write ? F_WRLCK : F_RDLCK);
1621 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1624 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1625 struct file *file_out, loff_t pos_out, u64 len)
1627 struct inode *inode_in = file_inode(file_in);
1628 struct inode *inode_out = file_inode(file_out);
1631 if (inode_in->i_sb != inode_out->i_sb ||
1632 file_in->f_path.mnt != file_out->f_path.mnt)
1635 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1637 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1640 if (!(file_in->f_mode & FMODE_READ) ||
1641 !(file_out->f_mode & FMODE_WRITE) ||
1642 (file_out->f_flags & O_APPEND))
1645 if (!file_in->f_op->clone_file_range)
1648 ret = clone_verify_area(file_in, pos_in, len, false);
1652 ret = clone_verify_area(file_out, pos_out, len, true);
1656 if (pos_in + len > i_size_read(inode_in))
1659 ret = mnt_want_write_file(file_out);
1663 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1664 file_out, pos_out, len);
1666 fsnotify_access(file_in);
1667 fsnotify_modify(file_out);
1670 mnt_drop_write_file(file_out);
1673 EXPORT_SYMBOL(vfs_clone_file_range);
1675 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1677 struct file_dedupe_range_info *info;
1678 struct inode *src = file_inode(file);
1683 bool is_admin = capable(CAP_SYS_ADMIN);
1684 u16 count = same->dest_count;
1685 struct file *dst_file;
1689 if (!(file->f_mode & FMODE_READ))
1692 if (same->reserved1 || same->reserved2)
1695 off = same->src_offset;
1696 len = same->src_length;
1699 if (S_ISDIR(src->i_mode))
1703 if (!S_ISREG(src->i_mode))
1706 ret = clone_verify_area(file, off, len, false);
1711 /* pre-format output fields to sane values */
1712 for (i = 0; i < count; i++) {
1713 same->info[i].bytes_deduped = 0ULL;
1714 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1717 for (i = 0, info = same->info; i < count; i++, info++) {
1719 struct fd dst_fd = fdget(info->dest_fd);
1721 dst_file = dst_fd.file;
1723 info->status = -EBADF;
1726 dst = file_inode(dst_file);
1728 ret = mnt_want_write_file(dst_file);
1734 dst_off = info->dest_offset;
1735 ret = clone_verify_area(dst_file, dst_off, len, true);
1742 if (info->reserved) {
1743 info->status = -EINVAL;
1744 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1745 info->status = -EINVAL;
1746 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1747 info->status = -EXDEV;
1748 } else if (S_ISDIR(dst->i_mode)) {
1749 info->status = -EISDIR;
1750 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1751 info->status = -EINVAL;
1753 deduped = dst_file->f_op->dedupe_file_range(file, off,
1756 if (deduped == -EBADE)
1757 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1758 else if (deduped < 0)
1759 info->status = deduped;
1761 info->bytes_deduped += deduped;
1765 mnt_drop_write_file(dst_file);
1769 if (fatal_signal_pending(current))
1776 EXPORT_SYMBOL(vfs_dedupe_file_range);