1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops = {
29 .llseek = generic_file_llseek,
30 .read_iter = generic_file_read_iter,
31 .mmap = generic_file_readonly_mmap,
32 .splice_read = generic_file_splice_read,
35 EXPORT_SYMBOL(generic_ro_fops);
37 static inline bool unsigned_offsets(struct file *file)
39 return file->f_mode & FMODE_UNSIGNED_OFFSET;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 if (offset < 0 && !unsigned_offsets(file))
61 if (offset != file->f_pos) {
67 EXPORT_SYMBOL(vfs_setpos);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
87 loff_t maxsize, loff_t eof)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file->f_lock);
108 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109 spin_unlock(&file->f_lock);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset >= eof)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset >= eof)
130 return vfs_setpos(file, offset, maxsize);
132 EXPORT_SYMBOL(generic_file_llseek_size);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 struct inode *inode = file->f_mapping->host;
148 return generic_file_llseek_size(file, offset, whence,
149 inode->i_sb->s_maxbytes,
152 EXPORT_SYMBOL(generic_file_llseek);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
165 case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 return generic_file_llseek_size(file, offset, whence,
172 EXPORT_SYMBOL(fixed_size_llseek);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
184 case SEEK_SET: case SEEK_CUR:
185 return generic_file_llseek_size(file, offset, whence,
191 EXPORT_SYMBOL(no_seek_end_llseek);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
204 case SEEK_SET: case SEEK_CUR:
205 return generic_file_llseek_size(file, offset, whence,
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
228 EXPORT_SYMBOL(noop_llseek);
230 loff_t default_llseek(struct file *file, loff_t offset, int whence)
232 struct inode *inode = file_inode(file);
238 offset += i_size_read(inode);
242 retval = file->f_pos;
245 offset += file->f_pos;
249 * In the generic case the entire file is data, so as
250 * long as offset isn't at the end of the file then the
253 if (offset >= inode->i_size) {
260 * There is a virtual hole at the end of the file, so
261 * as long as offset isn't i_size or larger, return
264 if (offset >= inode->i_size) {
268 offset = inode->i_size;
272 if (offset >= 0 || unsigned_offsets(file)) {
273 if (offset != file->f_pos) {
274 file->f_pos = offset;
283 EXPORT_SYMBOL(default_llseek);
285 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
287 if (!(file->f_mode & FMODE_LSEEK))
289 return file->f_op->llseek(file, offset, whence);
291 EXPORT_SYMBOL(vfs_llseek);
293 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
296 struct fd f = fdget_pos(fd);
301 if (whence <= SEEK_MAX) {
302 loff_t res = vfs_llseek(f.file, offset, whence);
304 if (res != (loff_t)retval)
305 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
311 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
313 return ksys_lseek(fd, offset, whence);
317 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
319 return ksys_lseek(fd, offset, whence);
323 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
324 defined(__ARCH_WANT_SYS_LLSEEK)
325 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
326 unsigned long, offset_low, loff_t __user *, result,
327 unsigned int, whence)
330 struct fd f = fdget_pos(fd);
337 if (whence > SEEK_MAX)
340 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
343 retval = (int)offset;
346 if (!copy_to_user(result, &offset, sizeof(offset)))
355 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
357 if (unlikely((ssize_t) count < 0))
363 if (unlikely(pos < 0)) {
364 if (!unsigned_offsets(file))
366 if (count >= -pos) /* both values are in 0..LLONG_MAX */
368 } else if (unlikely((loff_t) (pos + count) < 0)) {
369 if (!unsigned_offsets(file))
374 return security_file_permission(file,
375 read_write == READ ? MAY_READ : MAY_WRITE);
377 EXPORT_SYMBOL(rw_verify_area);
379 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
382 struct iov_iter iter;
385 init_sync_kiocb(&kiocb, filp);
386 kiocb.ki_pos = (ppos ? *ppos : 0);
387 iov_iter_ubuf(&iter, ITER_DEST, buf, len);
389 ret = call_read_iter(filp, &kiocb, &iter);
390 BUG_ON(ret == -EIOCBQUEUED);
392 *ppos = kiocb.ki_pos;
396 static int warn_unsupported(struct file *file, const char *op)
399 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
400 op, file, current->pid, current->comm);
404 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
408 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
411 struct iov_iter iter;
414 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
416 if (!(file->f_mode & FMODE_CAN_READ))
419 * Also fail if ->read_iter and ->read are both wired up as that
420 * implies very convoluted semantics.
422 if (unlikely(!file->f_op->read_iter || file->f_op->read))
423 return warn_unsupported(file, "read");
425 init_sync_kiocb(&kiocb, file);
426 kiocb.ki_pos = pos ? *pos : 0;
427 iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
428 ret = file->f_op->read_iter(&kiocb, &iter);
432 fsnotify_access(file);
433 add_rchar(current, ret);
439 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
443 ret = rw_verify_area(READ, file, pos, count);
446 return __kernel_read(file, buf, count, pos);
448 EXPORT_SYMBOL(kernel_read);
450 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
454 if (!(file->f_mode & FMODE_READ))
456 if (!(file->f_mode & FMODE_CAN_READ))
458 if (unlikely(!access_ok(buf, count)))
461 ret = rw_verify_area(READ, file, pos, count);
464 if (count > MAX_RW_COUNT)
465 count = MAX_RW_COUNT;
467 if (file->f_op->read)
468 ret = file->f_op->read(file, buf, count, pos);
469 else if (file->f_op->read_iter)
470 ret = new_sync_read(file, buf, count, pos);
474 fsnotify_access(file);
475 add_rchar(current, ret);
481 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
484 struct iov_iter iter;
487 init_sync_kiocb(&kiocb, filp);
488 kiocb.ki_pos = (ppos ? *ppos : 0);
489 iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
491 ret = call_write_iter(filp, &kiocb, &iter);
492 BUG_ON(ret == -EIOCBQUEUED);
494 *ppos = kiocb.ki_pos;
498 /* caller is responsible for file_start_write/file_end_write */
499 ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
504 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
506 if (!(file->f_mode & FMODE_CAN_WRITE))
509 * Also fail if ->write_iter and ->write are both wired up as that
510 * implies very convoluted semantics.
512 if (unlikely(!file->f_op->write_iter || file->f_op->write))
513 return warn_unsupported(file, "write");
515 init_sync_kiocb(&kiocb, file);
516 kiocb.ki_pos = pos ? *pos : 0;
517 ret = file->f_op->write_iter(&kiocb, from);
521 fsnotify_modify(file);
522 add_wchar(current, ret);
528 /* caller is responsible for file_start_write/file_end_write */
529 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
532 .iov_base = (void *)buf,
533 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
535 struct iov_iter iter;
536 iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
537 return __kernel_write_iter(file, &iter, pos);
540 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
541 * but autofs is one of the few internal kernel users that actually
542 * wants this _and_ can be built as a module. So we need to export
543 * this symbol for autofs, even though it really isn't appropriate
544 * for any other kernel modules.
546 EXPORT_SYMBOL_GPL(__kernel_write);
548 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
553 ret = rw_verify_area(WRITE, file, pos, count);
557 file_start_write(file);
558 ret = __kernel_write(file, buf, count, pos);
559 file_end_write(file);
562 EXPORT_SYMBOL(kernel_write);
564 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
568 if (!(file->f_mode & FMODE_WRITE))
570 if (!(file->f_mode & FMODE_CAN_WRITE))
572 if (unlikely(!access_ok(buf, count)))
575 ret = rw_verify_area(WRITE, file, pos, count);
578 if (count > MAX_RW_COUNT)
579 count = MAX_RW_COUNT;
580 file_start_write(file);
581 if (file->f_op->write)
582 ret = file->f_op->write(file, buf, count, pos);
583 else if (file->f_op->write_iter)
584 ret = new_sync_write(file, buf, count, pos);
588 fsnotify_modify(file);
589 add_wchar(current, ret);
592 file_end_write(file);
596 /* file_ppos returns &file->f_pos or NULL if file is stream */
597 static inline loff_t *file_ppos(struct file *file)
599 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
602 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
604 struct fd f = fdget_pos(fd);
605 ssize_t ret = -EBADF;
608 loff_t pos, *ppos = file_ppos(f.file);
613 ret = vfs_read(f.file, buf, count, ppos);
614 if (ret >= 0 && ppos)
621 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
623 return ksys_read(fd, buf, count);
626 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
628 struct fd f = fdget_pos(fd);
629 ssize_t ret = -EBADF;
632 loff_t pos, *ppos = file_ppos(f.file);
637 ret = vfs_write(f.file, buf, count, ppos);
638 if (ret >= 0 && ppos)
646 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
649 return ksys_write(fd, buf, count);
652 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
656 ssize_t ret = -EBADF;
664 if (f.file->f_mode & FMODE_PREAD)
665 ret = vfs_read(f.file, buf, count, &pos);
672 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
673 size_t, count, loff_t, pos)
675 return ksys_pread64(fd, buf, count, pos);
678 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
679 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
680 size_t, count, compat_arg_u64_dual(pos))
682 return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
686 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
687 size_t count, loff_t pos)
690 ssize_t ret = -EBADF;
698 if (f.file->f_mode & FMODE_PWRITE)
699 ret = vfs_write(f.file, buf, count, &pos);
706 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
707 size_t, count, loff_t, pos)
709 return ksys_pwrite64(fd, buf, count, pos);
712 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
713 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
714 size_t, count, compat_arg_u64_dual(pos))
716 return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
720 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
721 loff_t *ppos, int type, rwf_t flags)
726 init_sync_kiocb(&kiocb, filp);
727 ret = kiocb_set_rw_flags(&kiocb, flags);
730 kiocb.ki_pos = (ppos ? *ppos : 0);
733 ret = call_read_iter(filp, &kiocb, iter);
735 ret = call_write_iter(filp, &kiocb, iter);
736 BUG_ON(ret == -EIOCBQUEUED);
738 *ppos = kiocb.ki_pos;
742 /* Do it by hand, with file-ops */
743 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
744 loff_t *ppos, int type, rwf_t flags)
748 if (flags & ~RWF_HIPRI)
751 while (iov_iter_count(iter)) {
755 nr = filp->f_op->read(filp, iter_iov_addr(iter),
756 iter_iov_len(iter), ppos);
758 nr = filp->f_op->write(filp, iter_iov_addr(iter),
759 iter_iov_len(iter), ppos);
768 if (nr != iter_iov_len(iter))
770 iov_iter_advance(iter, nr);
776 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
777 loff_t *pos, rwf_t flags)
782 if (!(file->f_mode & FMODE_READ))
784 if (!(file->f_mode & FMODE_CAN_READ))
787 tot_len = iov_iter_count(iter);
790 ret = rw_verify_area(READ, file, pos, tot_len);
794 if (file->f_op->read_iter)
795 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
797 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
800 fsnotify_access(file);
804 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
805 struct iov_iter *iter)
810 if (!file->f_op->read_iter)
812 if (!(file->f_mode & FMODE_READ))
814 if (!(file->f_mode & FMODE_CAN_READ))
817 tot_len = iov_iter_count(iter);
820 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
824 ret = call_read_iter(file, iocb, iter);
827 fsnotify_access(file);
830 EXPORT_SYMBOL(vfs_iocb_iter_read);
832 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
835 if (!file->f_op->read_iter)
837 return do_iter_read(file, iter, ppos, flags);
839 EXPORT_SYMBOL(vfs_iter_read);
841 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
842 loff_t *pos, rwf_t flags)
847 if (!(file->f_mode & FMODE_WRITE))
849 if (!(file->f_mode & FMODE_CAN_WRITE))
852 tot_len = iov_iter_count(iter);
855 ret = rw_verify_area(WRITE, file, pos, tot_len);
859 if (file->f_op->write_iter)
860 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
862 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
864 fsnotify_modify(file);
868 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
869 struct iov_iter *iter)
874 if (!file->f_op->write_iter)
876 if (!(file->f_mode & FMODE_WRITE))
878 if (!(file->f_mode & FMODE_CAN_WRITE))
881 tot_len = iov_iter_count(iter);
884 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
888 ret = call_write_iter(file, iocb, iter);
890 fsnotify_modify(file);
894 EXPORT_SYMBOL(vfs_iocb_iter_write);
896 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
899 if (!file->f_op->write_iter)
901 return do_iter_write(file, iter, ppos, flags);
903 EXPORT_SYMBOL(vfs_iter_write);
905 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
906 unsigned long vlen, loff_t *pos, rwf_t flags)
908 struct iovec iovstack[UIO_FASTIOV];
909 struct iovec *iov = iovstack;
910 struct iov_iter iter;
913 ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
915 ret = do_iter_read(file, &iter, pos, flags);
922 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
923 unsigned long vlen, loff_t *pos, rwf_t flags)
925 struct iovec iovstack[UIO_FASTIOV];
926 struct iovec *iov = iovstack;
927 struct iov_iter iter;
930 ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
932 file_start_write(file);
933 ret = do_iter_write(file, &iter, pos, flags);
934 file_end_write(file);
940 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
941 unsigned long vlen, rwf_t flags)
943 struct fd f = fdget_pos(fd);
944 ssize_t ret = -EBADF;
947 loff_t pos, *ppos = file_ppos(f.file);
952 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
953 if (ret >= 0 && ppos)
959 add_rchar(current, ret);
964 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
965 unsigned long vlen, rwf_t flags)
967 struct fd f = fdget_pos(fd);
968 ssize_t ret = -EBADF;
971 loff_t pos, *ppos = file_ppos(f.file);
976 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
977 if (ret >= 0 && ppos)
983 add_wchar(current, ret);
988 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
990 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
991 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
994 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
995 unsigned long vlen, loff_t pos, rwf_t flags)
998 ssize_t ret = -EBADF;
1006 if (f.file->f_mode & FMODE_PREAD)
1007 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1012 add_rchar(current, ret);
1017 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1018 unsigned long vlen, loff_t pos, rwf_t flags)
1021 ssize_t ret = -EBADF;
1029 if (f.file->f_mode & FMODE_PWRITE)
1030 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1035 add_wchar(current, ret);
1040 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1041 unsigned long, vlen)
1043 return do_readv(fd, vec, vlen, 0);
1046 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1047 unsigned long, vlen)
1049 return do_writev(fd, vec, vlen, 0);
1052 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1053 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1055 loff_t pos = pos_from_hilo(pos_h, pos_l);
1057 return do_preadv(fd, vec, vlen, pos, 0);
1060 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1061 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1064 loff_t pos = pos_from_hilo(pos_h, pos_l);
1067 return do_readv(fd, vec, vlen, flags);
1069 return do_preadv(fd, vec, vlen, pos, flags);
1072 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1073 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1075 loff_t pos = pos_from_hilo(pos_h, pos_l);
1077 return do_pwritev(fd, vec, vlen, pos, 0);
1080 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1081 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1084 loff_t pos = pos_from_hilo(pos_h, pos_l);
1087 return do_writev(fd, vec, vlen, flags);
1089 return do_pwritev(fd, vec, vlen, pos, flags);
1093 * Various compat syscalls. Note that they all pretend to take a native
1094 * iovec - import_iovec will properly treat those as compat_iovecs based on
1095 * in_compat_syscall().
1097 #ifdef CONFIG_COMPAT
1098 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1099 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1100 const struct iovec __user *, vec,
1101 unsigned long, vlen, loff_t, pos)
1103 return do_preadv(fd, vec, vlen, pos, 0);
1107 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1108 const struct iovec __user *, vec,
1109 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1111 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1113 return do_preadv(fd, vec, vlen, pos, 0);
1116 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1117 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1118 const struct iovec __user *, vec,
1119 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1122 return do_readv(fd, vec, vlen, flags);
1123 return do_preadv(fd, vec, vlen, pos, flags);
1127 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1128 const struct iovec __user *, vec,
1129 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1132 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1135 return do_readv(fd, vec, vlen, flags);
1136 return do_preadv(fd, vec, vlen, pos, flags);
1139 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1140 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1141 const struct iovec __user *, vec,
1142 unsigned long, vlen, loff_t, pos)
1144 return do_pwritev(fd, vec, vlen, pos, 0);
1148 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1149 const struct iovec __user *,vec,
1150 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1152 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1154 return do_pwritev(fd, vec, vlen, pos, 0);
1157 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1158 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1159 const struct iovec __user *, vec,
1160 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1163 return do_writev(fd, vec, vlen, flags);
1164 return do_pwritev(fd, vec, vlen, pos, flags);
1168 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1169 const struct iovec __user *,vec,
1170 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1172 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1175 return do_writev(fd, vec, vlen, flags);
1176 return do_pwritev(fd, vec, vlen, pos, flags);
1178 #endif /* CONFIG_COMPAT */
1180 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1181 size_t count, loff_t max)
1184 struct inode *in_inode, *out_inode;
1185 struct pipe_inode_info *opipe;
1192 * Get input file, and verify that it is ok..
1198 if (!(in.file->f_mode & FMODE_READ))
1202 pos = in.file->f_pos;
1205 if (!(in.file->f_mode & FMODE_PREAD))
1208 retval = rw_verify_area(READ, in.file, &pos, count);
1211 if (count > MAX_RW_COUNT)
1212 count = MAX_RW_COUNT;
1215 * Get output file, and verify that it is ok..
1218 out = fdget(out_fd);
1221 if (!(out.file->f_mode & FMODE_WRITE))
1223 in_inode = file_inode(in.file);
1224 out_inode = file_inode(out.file);
1225 out_pos = out.file->f_pos;
1228 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1230 if (unlikely(pos + count > max)) {
1231 retval = -EOVERFLOW;
1240 * We need to debate whether we can enable this or not. The
1241 * man page documents EAGAIN return for the output at least,
1242 * and the application is arguably buggy if it doesn't expect
1243 * EAGAIN on a non-blocking file descriptor.
1245 if (in.file->f_flags & O_NONBLOCK)
1246 fl = SPLICE_F_NONBLOCK;
1248 opipe = get_pipe_info(out.file, true);
1250 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1253 file_start_write(out.file);
1254 retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1256 file_end_write(out.file);
1258 if (out.file->f_flags & O_NONBLOCK)
1259 fl |= SPLICE_F_NONBLOCK;
1261 retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1265 add_rchar(current, retval);
1266 add_wchar(current, retval);
1267 fsnotify_access(in.file);
1268 fsnotify_modify(out.file);
1269 out.file->f_pos = out_pos;
1273 in.file->f_pos = pos;
1279 retval = -EOVERFLOW;
1289 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1296 if (unlikely(get_user(off, offset)))
1299 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1300 if (unlikely(put_user(pos, offset)))
1305 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1308 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1314 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1316 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1317 if (unlikely(put_user(pos, offset)))
1322 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1325 #ifdef CONFIG_COMPAT
1326 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1327 compat_off_t __user *, offset, compat_size_t, count)
1334 if (unlikely(get_user(off, offset)))
1337 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1338 if (unlikely(put_user(pos, offset)))
1343 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1346 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1347 compat_loff_t __user *, offset, compat_size_t, count)
1353 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1355 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1356 if (unlikely(put_user(pos, offset)))
1361 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1366 * generic_copy_file_range - copy data between two files
1367 * @file_in: file structure to read from
1368 * @pos_in: file offset to read from
1369 * @file_out: file structure to write data to
1370 * @pos_out: file offset to write data to
1371 * @len: amount of data to copy
1372 * @flags: copy flags
1374 * This is a generic filesystem helper to copy data from one file to another.
1375 * It has no constraints on the source or destination file owners - the files
1376 * can belong to different superblocks and different filesystem types. Short
1377 * copies are allowed.
1379 * This should be called from the @file_out filesystem, as per the
1380 * ->copy_file_range() method.
1382 * Returns the number of bytes copied or a negative error indicating the
1386 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1387 struct file *file_out, loff_t pos_out,
1388 size_t len, unsigned int flags)
1390 lockdep_assert(sb_write_started(file_inode(file_out)->i_sb));
1392 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1393 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1395 EXPORT_SYMBOL(generic_copy_file_range);
1398 * Performs necessary checks before doing a file copy
1400 * Can adjust amount of bytes to copy via @req_count argument.
1401 * Returns appropriate error code that caller should return or
1402 * zero in case the copy should be allowed.
1404 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1405 struct file *file_out, loff_t pos_out,
1406 size_t *req_count, unsigned int flags)
1408 struct inode *inode_in = file_inode(file_in);
1409 struct inode *inode_out = file_inode(file_out);
1410 uint64_t count = *req_count;
1414 ret = generic_file_rw_checks(file_in, file_out);
1419 * We allow some filesystems to handle cross sb copy, but passing
1420 * a file of the wrong filesystem type to filesystem driver can result
1421 * in an attempt to dereference the wrong type of ->private_data, so
1422 * avoid doing that until we really have a good reason.
1424 * nfs and cifs define several different file_system_type structures
1425 * and several different sets of file_operations, but they all end up
1426 * using the same ->copy_file_range() function pointer.
1428 if (flags & COPY_FILE_SPLICE) {
1429 /* cross sb splice is allowed */
1430 } else if (file_out->f_op->copy_file_range) {
1431 if (file_in->f_op->copy_file_range !=
1432 file_out->f_op->copy_file_range)
1434 } else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1438 /* Don't touch certain kinds of inodes */
1439 if (IS_IMMUTABLE(inode_out))
1442 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1445 /* Ensure offsets don't wrap. */
1446 if (pos_in + count < pos_in || pos_out + count < pos_out)
1449 /* Shorten the copy to EOF */
1450 size_in = i_size_read(inode_in);
1451 if (pos_in >= size_in)
1454 count = min(count, size_in - (uint64_t)pos_in);
1456 ret = generic_write_check_limits(file_out, pos_out, &count);
1460 /* Don't allow overlapped copying within the same file. */
1461 if (inode_in == inode_out &&
1462 pos_out + count > pos_in &&
1463 pos_out < pos_in + count)
1471 * copy_file_range() differs from regular file read and write in that it
1472 * specifically allows return partial success. When it does so is up to
1473 * the copy_file_range method.
1475 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1476 struct file *file_out, loff_t pos_out,
1477 size_t len, unsigned int flags)
1480 bool splice = flags & COPY_FILE_SPLICE;
1482 if (flags & ~COPY_FILE_SPLICE)
1485 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1490 ret = rw_verify_area(READ, file_in, &pos_in, len);
1494 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1501 file_start_write(file_out);
1504 * Cloning is supported by more file systems, so we implement copy on
1505 * same sb using clone, but for filesystems where both clone and copy
1506 * are supported (e.g. nfs,cifs), we only call the copy method.
1508 if (!splice && file_out->f_op->copy_file_range) {
1509 ret = file_out->f_op->copy_file_range(file_in, pos_in,
1515 if (!splice && file_in->f_op->remap_file_range &&
1516 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1517 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1519 min_t(loff_t, MAX_RW_COUNT, len),
1520 REMAP_FILE_CAN_SHORTEN);
1526 * We can get here for same sb copy of filesystems that do not implement
1527 * ->copy_file_range() in case filesystem does not support clone or in
1528 * case filesystem supports clone but rejected the clone request (e.g.
1529 * because it was not block aligned).
1531 * In both cases, fall back to kernel copy so we are able to maintain a
1532 * consistent story about which filesystems support copy_file_range()
1533 * and which filesystems do not, that will allow userspace tools to
1534 * make consistent desicions w.r.t using copy_file_range().
1536 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE.
1538 ret = generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1543 fsnotify_access(file_in);
1544 add_rchar(current, ret);
1545 fsnotify_modify(file_out);
1546 add_wchar(current, ret);
1552 file_end_write(file_out);
1556 EXPORT_SYMBOL(vfs_copy_file_range);
1558 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1559 int, fd_out, loff_t __user *, off_out,
1560 size_t, len, unsigned int, flags)
1566 ssize_t ret = -EBADF;
1568 f_in = fdget(fd_in);
1572 f_out = fdget(fd_out);
1578 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1581 pos_in = f_in.file->f_pos;
1585 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1588 pos_out = f_out.file->f_pos;
1595 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1602 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1605 f_in.file->f_pos = pos_in;
1609 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1612 f_out.file->f_pos = pos_out;
1625 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1626 * LFS limits. If pos is under the limit it becomes a short access. If it
1627 * exceeds the limit we return -EFBIG.
1629 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1631 struct inode *inode = file->f_mapping->host;
1632 loff_t max_size = inode->i_sb->s_maxbytes;
1633 loff_t limit = rlimit(RLIMIT_FSIZE);
1635 if (limit != RLIM_INFINITY) {
1637 send_sig(SIGXFSZ, current, 0);
1640 *count = min(*count, limit - pos);
1643 if (!(file->f_flags & O_LARGEFILE))
1644 max_size = MAX_NON_LFS;
1646 if (unlikely(pos >= max_size))
1649 *count = min(*count, max_size - pos);
1654 /* Like generic_write_checks(), but takes size of write instead of iter. */
1655 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1657 struct file *file = iocb->ki_filp;
1658 struct inode *inode = file->f_mapping->host;
1660 if (IS_SWAPFILE(inode))
1666 if (iocb->ki_flags & IOCB_APPEND)
1667 iocb->ki_pos = i_size_read(inode);
1669 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1670 !((iocb->ki_flags & IOCB_DIRECT) ||
1671 (file->f_mode & FMODE_BUF_WASYNC)))
1674 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1676 EXPORT_SYMBOL(generic_write_checks_count);
1679 * Performs necessary checks before doing a write
1681 * Can adjust writing position or amount of bytes to write.
1682 * Returns appropriate error code that caller should return or
1683 * zero in case that write should be allowed.
1685 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1687 loff_t count = iov_iter_count(from);
1690 ret = generic_write_checks_count(iocb, &count);
1694 iov_iter_truncate(from, count);
1695 return iov_iter_count(from);
1697 EXPORT_SYMBOL(generic_write_checks);
1700 * Performs common checks before doing a file copy/clone
1701 * from @file_in to @file_out.
1703 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1705 struct inode *inode_in = file_inode(file_in);
1706 struct inode *inode_out = file_inode(file_out);
1708 /* Don't copy dirs, pipes, sockets... */
1709 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1711 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1714 if (!(file_in->f_mode & FMODE_READ) ||
1715 !(file_out->f_mode & FMODE_WRITE) ||
1716 (file_out->f_flags & O_APPEND))