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 no_llseek(struct file *file, loff_t offset, int whence)
234 EXPORT_SYMBOL(no_llseek);
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 struct inode *inode = file_inode(file);
244 offset += i_size_read(inode);
248 retval = file->f_pos;
251 offset += file->f_pos;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset >= inode->i_size) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset >= inode->i_size) {
274 offset = inode->i_size;
278 if (offset >= 0 || unsigned_offsets(file)) {
279 if (offset != file->f_pos) {
280 file->f_pos = offset;
289 EXPORT_SYMBOL(default_llseek);
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 loff_t (*fn)(struct file *, loff_t, int);
296 if (file->f_mode & FMODE_LSEEK) {
297 if (file->f_op->llseek)
298 fn = file->f_op->llseek;
300 return fn(file, offset, whence);
302 EXPORT_SYMBOL(vfs_llseek);
304 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
307 struct fd f = fdget_pos(fd);
312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
324 return ksys_lseek(fd, offset, whence);
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
330 return ksys_lseek(fd, offset, whence);
334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
335 defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
337 unsigned long, offset_low, loff_t __user *, result,
338 unsigned int, whence)
341 struct fd f = fdget_pos(fd);
348 if (whence > SEEK_MAX)
351 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
354 retval = (int)offset;
357 if (!copy_to_user(result, &offset, sizeof(offset)))
366 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
368 if (unlikely((ssize_t) count < 0))
374 if (unlikely(pos < 0)) {
375 if (!unsigned_offsets(file))
377 if (count >= -pos) /* both values are in 0..LLONG_MAX */
379 } else if (unlikely((loff_t) (pos + count) < 0)) {
380 if (!unsigned_offsets(file))
385 return security_file_permission(file,
386 read_write == READ ? MAY_READ : MAY_WRITE);
388 EXPORT_SYMBOL(rw_verify_area);
390 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
392 struct iovec iov = { .iov_base = buf, .iov_len = len };
394 struct iov_iter iter;
397 init_sync_kiocb(&kiocb, filp);
398 kiocb.ki_pos = (ppos ? *ppos : 0);
399 iov_iter_init(&iter, READ, &iov, 1, len);
401 ret = call_read_iter(filp, &kiocb, &iter);
402 BUG_ON(ret == -EIOCBQUEUED);
404 *ppos = kiocb.ki_pos;
408 static int warn_unsupported(struct file *file, const char *op)
411 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
412 op, file, current->pid, current->comm);
416 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
420 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
423 struct iov_iter iter;
426 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
428 if (!(file->f_mode & FMODE_CAN_READ))
431 * Also fail if ->read_iter and ->read are both wired up as that
432 * implies very convoluted semantics.
434 if (unlikely(!file->f_op->read_iter || file->f_op->read))
435 return warn_unsupported(file, "read");
437 init_sync_kiocb(&kiocb, file);
438 kiocb.ki_pos = pos ? *pos : 0;
439 iov_iter_kvec(&iter, READ, &iov, 1, iov.iov_len);
440 ret = file->f_op->read_iter(&kiocb, &iter);
444 fsnotify_access(file);
445 add_rchar(current, ret);
451 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
455 ret = rw_verify_area(READ, file, pos, count);
458 return __kernel_read(file, buf, count, pos);
460 EXPORT_SYMBOL(kernel_read);
462 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
466 if (!(file->f_mode & FMODE_READ))
468 if (!(file->f_mode & FMODE_CAN_READ))
470 if (unlikely(!access_ok(buf, count)))
473 ret = rw_verify_area(READ, file, pos, count);
476 if (count > MAX_RW_COUNT)
477 count = MAX_RW_COUNT;
479 if (file->f_op->read)
480 ret = file->f_op->read(file, buf, count, pos);
481 else if (file->f_op->read_iter)
482 ret = new_sync_read(file, buf, count, pos);
486 fsnotify_access(file);
487 add_rchar(current, ret);
493 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
495 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
497 struct iov_iter iter;
500 init_sync_kiocb(&kiocb, filp);
501 kiocb.ki_pos = (ppos ? *ppos : 0);
502 iov_iter_init(&iter, WRITE, &iov, 1, len);
504 ret = call_write_iter(filp, &kiocb, &iter);
505 BUG_ON(ret == -EIOCBQUEUED);
507 *ppos = kiocb.ki_pos;
511 /* caller is responsible for file_start_write/file_end_write */
512 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
515 .iov_base = (void *)buf,
516 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
519 struct iov_iter iter;
522 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
524 if (!(file->f_mode & FMODE_CAN_WRITE))
527 * Also fail if ->write_iter and ->write are both wired up as that
528 * implies very convoluted semantics.
530 if (unlikely(!file->f_op->write_iter || file->f_op->write))
531 return warn_unsupported(file, "write");
533 init_sync_kiocb(&kiocb, file);
534 kiocb.ki_pos = pos ? *pos : 0;
535 iov_iter_kvec(&iter, WRITE, &iov, 1, iov.iov_len);
536 ret = file->f_op->write_iter(&kiocb, &iter);
540 fsnotify_modify(file);
541 add_wchar(current, ret);
547 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
548 * but autofs is one of the few internal kernel users that actually
549 * wants this _and_ can be built as a module. So we need to export
550 * this symbol for autofs, even though it really isn't appropriate
551 * for any other kernel modules.
553 EXPORT_SYMBOL_GPL(__kernel_write);
555 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
560 ret = rw_verify_area(WRITE, file, pos, count);
564 file_start_write(file);
565 ret = __kernel_write(file, buf, count, pos);
566 file_end_write(file);
569 EXPORT_SYMBOL(kernel_write);
571 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
575 if (!(file->f_mode & FMODE_WRITE))
577 if (!(file->f_mode & FMODE_CAN_WRITE))
579 if (unlikely(!access_ok(buf, count)))
582 ret = rw_verify_area(WRITE, file, pos, count);
585 if (count > MAX_RW_COUNT)
586 count = MAX_RW_COUNT;
587 file_start_write(file);
588 if (file->f_op->write)
589 ret = file->f_op->write(file, buf, count, pos);
590 else if (file->f_op->write_iter)
591 ret = new_sync_write(file, buf, count, pos);
595 fsnotify_modify(file);
596 add_wchar(current, ret);
599 file_end_write(file);
603 /* file_ppos returns &file->f_pos or NULL if file is stream */
604 static inline loff_t *file_ppos(struct file *file)
606 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
609 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
611 struct fd f = fdget_pos(fd);
612 ssize_t ret = -EBADF;
615 loff_t pos, *ppos = file_ppos(f.file);
620 ret = vfs_read(f.file, buf, count, ppos);
621 if (ret >= 0 && ppos)
628 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
630 return ksys_read(fd, buf, count);
633 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
635 struct fd f = fdget_pos(fd);
636 ssize_t ret = -EBADF;
639 loff_t pos, *ppos = file_ppos(f.file);
644 ret = vfs_write(f.file, buf, count, ppos);
645 if (ret >= 0 && ppos)
653 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
656 return ksys_write(fd, buf, count);
659 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
663 ssize_t ret = -EBADF;
671 if (f.file->f_mode & FMODE_PREAD)
672 ret = vfs_read(f.file, buf, count, &pos);
679 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
680 size_t, count, loff_t, pos)
682 return ksys_pread64(fd, buf, count, pos);
685 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
686 size_t count, loff_t pos)
689 ssize_t ret = -EBADF;
697 if (f.file->f_mode & FMODE_PWRITE)
698 ret = vfs_write(f.file, buf, count, &pos);
705 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
706 size_t, count, loff_t, pos)
708 return ksys_pwrite64(fd, buf, count, pos);
711 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
712 loff_t *ppos, int type, rwf_t flags)
717 init_sync_kiocb(&kiocb, filp);
718 ret = kiocb_set_rw_flags(&kiocb, flags);
721 kiocb.ki_pos = (ppos ? *ppos : 0);
724 ret = call_read_iter(filp, &kiocb, iter);
726 ret = call_write_iter(filp, &kiocb, iter);
727 BUG_ON(ret == -EIOCBQUEUED);
729 *ppos = kiocb.ki_pos;
733 /* Do it by hand, with file-ops */
734 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
735 loff_t *ppos, int type, rwf_t flags)
739 if (flags & ~RWF_HIPRI)
742 while (iov_iter_count(iter)) {
743 struct iovec iovec = iov_iter_iovec(iter);
747 nr = filp->f_op->read(filp, iovec.iov_base,
748 iovec.iov_len, ppos);
750 nr = filp->f_op->write(filp, iovec.iov_base,
751 iovec.iov_len, ppos);
760 if (nr != iovec.iov_len)
762 iov_iter_advance(iter, nr);
768 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
769 loff_t *pos, rwf_t flags)
774 if (!(file->f_mode & FMODE_READ))
776 if (!(file->f_mode & FMODE_CAN_READ))
779 tot_len = iov_iter_count(iter);
782 ret = rw_verify_area(READ, file, pos, tot_len);
786 if (file->f_op->read_iter)
787 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
789 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
792 fsnotify_access(file);
796 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
797 struct iov_iter *iter)
802 if (!file->f_op->read_iter)
804 if (!(file->f_mode & FMODE_READ))
806 if (!(file->f_mode & FMODE_CAN_READ))
809 tot_len = iov_iter_count(iter);
812 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
816 ret = call_read_iter(file, iocb, iter);
819 fsnotify_access(file);
822 EXPORT_SYMBOL(vfs_iocb_iter_read);
824 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
827 if (!file->f_op->read_iter)
829 return do_iter_read(file, iter, ppos, flags);
831 EXPORT_SYMBOL(vfs_iter_read);
833 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
834 loff_t *pos, rwf_t flags)
839 if (!(file->f_mode & FMODE_WRITE))
841 if (!(file->f_mode & FMODE_CAN_WRITE))
844 tot_len = iov_iter_count(iter);
847 ret = rw_verify_area(WRITE, file, pos, tot_len);
851 if (file->f_op->write_iter)
852 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
854 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
856 fsnotify_modify(file);
860 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
861 struct iov_iter *iter)
866 if (!file->f_op->write_iter)
868 if (!(file->f_mode & FMODE_WRITE))
870 if (!(file->f_mode & FMODE_CAN_WRITE))
873 tot_len = iov_iter_count(iter);
876 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
880 ret = call_write_iter(file, iocb, iter);
882 fsnotify_modify(file);
886 EXPORT_SYMBOL(vfs_iocb_iter_write);
888 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
891 if (!file->f_op->write_iter)
893 return do_iter_write(file, iter, ppos, flags);
895 EXPORT_SYMBOL(vfs_iter_write);
897 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
898 unsigned long vlen, loff_t *pos, rwf_t flags)
900 struct iovec iovstack[UIO_FASTIOV];
901 struct iovec *iov = iovstack;
902 struct iov_iter iter;
905 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
907 ret = do_iter_read(file, &iter, pos, flags);
914 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
915 unsigned long vlen, loff_t *pos, rwf_t flags)
917 struct iovec iovstack[UIO_FASTIOV];
918 struct iovec *iov = iovstack;
919 struct iov_iter iter;
922 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
924 file_start_write(file);
925 ret = do_iter_write(file, &iter, pos, flags);
926 file_end_write(file);
932 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
933 unsigned long vlen, rwf_t flags)
935 struct fd f = fdget_pos(fd);
936 ssize_t ret = -EBADF;
939 loff_t pos, *ppos = file_ppos(f.file);
944 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
945 if (ret >= 0 && ppos)
951 add_rchar(current, ret);
956 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
957 unsigned long vlen, rwf_t flags)
959 struct fd f = fdget_pos(fd);
960 ssize_t ret = -EBADF;
963 loff_t pos, *ppos = file_ppos(f.file);
968 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
969 if (ret >= 0 && ppos)
975 add_wchar(current, ret);
980 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
982 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
983 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
986 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
987 unsigned long vlen, loff_t pos, rwf_t flags)
990 ssize_t ret = -EBADF;
998 if (f.file->f_mode & FMODE_PREAD)
999 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1004 add_rchar(current, ret);
1009 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1010 unsigned long vlen, loff_t pos, rwf_t flags)
1013 ssize_t ret = -EBADF;
1021 if (f.file->f_mode & FMODE_PWRITE)
1022 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1027 add_wchar(current, ret);
1032 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1033 unsigned long, vlen)
1035 return do_readv(fd, vec, vlen, 0);
1038 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1039 unsigned long, vlen)
1041 return do_writev(fd, vec, vlen, 0);
1044 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1045 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1047 loff_t pos = pos_from_hilo(pos_h, pos_l);
1049 return do_preadv(fd, vec, vlen, pos, 0);
1052 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1053 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1056 loff_t pos = pos_from_hilo(pos_h, pos_l);
1059 return do_readv(fd, vec, vlen, flags);
1061 return do_preadv(fd, vec, vlen, pos, flags);
1064 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1065 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1067 loff_t pos = pos_from_hilo(pos_h, pos_l);
1069 return do_pwritev(fd, vec, vlen, pos, 0);
1072 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1073 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1076 loff_t pos = pos_from_hilo(pos_h, pos_l);
1079 return do_writev(fd, vec, vlen, flags);
1081 return do_pwritev(fd, vec, vlen, pos, flags);
1085 * Various compat syscalls. Note that they all pretend to take a native
1086 * iovec - import_iovec will properly treat those as compat_iovecs based on
1087 * in_compat_syscall().
1089 #ifdef CONFIG_COMPAT
1090 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1091 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1092 const struct iovec __user *, vec,
1093 unsigned long, vlen, loff_t, pos)
1095 return do_preadv(fd, vec, vlen, pos, 0);
1099 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1100 const struct iovec __user *, vec,
1101 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1103 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1105 return do_preadv(fd, vec, vlen, pos, 0);
1108 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1109 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1110 const struct iovec __user *, vec,
1111 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1114 return do_readv(fd, vec, vlen, flags);
1115 return do_preadv(fd, vec, vlen, pos, flags);
1119 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1120 const struct iovec __user *, vec,
1121 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1124 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1127 return do_readv(fd, vec, vlen, flags);
1128 return do_preadv(fd, vec, vlen, pos, flags);
1131 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1132 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1133 const struct iovec __user *, vec,
1134 unsigned long, vlen, loff_t, pos)
1136 return do_pwritev(fd, vec, vlen, pos, 0);
1140 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1141 const struct iovec __user *,vec,
1142 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1144 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1146 return do_pwritev(fd, vec, vlen, pos, 0);
1149 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1150 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1151 const struct iovec __user *, vec,
1152 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1155 return do_writev(fd, vec, vlen, flags);
1156 return do_pwritev(fd, vec, vlen, pos, flags);
1160 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1161 const struct iovec __user *,vec,
1162 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1164 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1167 return do_writev(fd, vec, vlen, flags);
1168 return do_pwritev(fd, vec, vlen, pos, flags);
1170 #endif /* CONFIG_COMPAT */
1172 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1173 size_t count, loff_t max)
1176 struct inode *in_inode, *out_inode;
1177 struct pipe_inode_info *opipe;
1184 * Get input file, and verify that it is ok..
1190 if (!(in.file->f_mode & FMODE_READ))
1194 pos = in.file->f_pos;
1197 if (!(in.file->f_mode & FMODE_PREAD))
1200 retval = rw_verify_area(READ, in.file, &pos, count);
1203 if (count > MAX_RW_COUNT)
1204 count = MAX_RW_COUNT;
1207 * Get output file, and verify that it is ok..
1210 out = fdget(out_fd);
1213 if (!(out.file->f_mode & FMODE_WRITE))
1215 in_inode = file_inode(in.file);
1216 out_inode = file_inode(out.file);
1217 out_pos = out.file->f_pos;
1220 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1222 if (unlikely(pos + count > max)) {
1223 retval = -EOVERFLOW;
1232 * We need to debate whether we can enable this or not. The
1233 * man page documents EAGAIN return for the output at least,
1234 * and the application is arguably buggy if it doesn't expect
1235 * EAGAIN on a non-blocking file descriptor.
1237 if (in.file->f_flags & O_NONBLOCK)
1238 fl = SPLICE_F_NONBLOCK;
1240 opipe = get_pipe_info(out.file, true);
1242 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1245 file_start_write(out.file);
1246 retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1248 file_end_write(out.file);
1250 retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1254 add_rchar(current, retval);
1255 add_wchar(current, retval);
1256 fsnotify_access(in.file);
1257 fsnotify_modify(out.file);
1258 out.file->f_pos = out_pos;
1262 in.file->f_pos = pos;
1268 retval = -EOVERFLOW;
1278 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1285 if (unlikely(get_user(off, offset)))
1288 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1289 if (unlikely(put_user(pos, offset)))
1294 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1297 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1303 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1305 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1306 if (unlikely(put_user(pos, offset)))
1311 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1314 #ifdef CONFIG_COMPAT
1315 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1316 compat_off_t __user *, offset, compat_size_t, count)
1323 if (unlikely(get_user(off, offset)))
1326 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1327 if (unlikely(put_user(pos, offset)))
1332 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1335 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1336 compat_loff_t __user *, offset, compat_size_t, count)
1342 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1344 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1345 if (unlikely(put_user(pos, offset)))
1350 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1355 * generic_copy_file_range - copy data between two files
1356 * @file_in: file structure to read from
1357 * @pos_in: file offset to read from
1358 * @file_out: file structure to write data to
1359 * @pos_out: file offset to write data to
1360 * @len: amount of data to copy
1361 * @flags: copy flags
1363 * This is a generic filesystem helper to copy data from one file to another.
1364 * It has no constraints on the source or destination file owners - the files
1365 * can belong to different superblocks and different filesystem types. Short
1366 * copies are allowed.
1368 * This should be called from the @file_out filesystem, as per the
1369 * ->copy_file_range() method.
1371 * Returns the number of bytes copied or a negative error indicating the
1375 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1376 struct file *file_out, loff_t pos_out,
1377 size_t len, unsigned int flags)
1379 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1380 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1382 EXPORT_SYMBOL(generic_copy_file_range);
1384 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1385 struct file *file_out, loff_t pos_out,
1386 size_t len, unsigned int flags)
1389 * Although we now allow filesystems to handle cross sb copy, passing
1390 * a file of the wrong filesystem type to filesystem driver can result
1391 * in an attempt to dereference the wrong type of ->private_data, so
1392 * avoid doing that until we really have a good reason. NFS defines
1393 * several different file_system_type structures, but they all end up
1394 * using the same ->copy_file_range() function pointer.
1396 if (file_out->f_op->copy_file_range &&
1397 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1398 return file_out->f_op->copy_file_range(file_in, pos_in,
1402 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1407 * Performs necessary checks before doing a file copy
1409 * Can adjust amount of bytes to copy via @req_count argument.
1410 * Returns appropriate error code that caller should return or
1411 * zero in case the copy should be allowed.
1413 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1414 struct file *file_out, loff_t pos_out,
1415 size_t *req_count, unsigned int flags)
1417 struct inode *inode_in = file_inode(file_in);
1418 struct inode *inode_out = file_inode(file_out);
1419 uint64_t count = *req_count;
1423 ret = generic_file_rw_checks(file_in, file_out);
1427 /* Don't touch certain kinds of inodes */
1428 if (IS_IMMUTABLE(inode_out))
1431 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1434 /* Ensure offsets don't wrap. */
1435 if (pos_in + count < pos_in || pos_out + count < pos_out)
1438 /* Shorten the copy to EOF */
1439 size_in = i_size_read(inode_in);
1440 if (pos_in >= size_in)
1443 count = min(count, size_in - (uint64_t)pos_in);
1445 ret = generic_write_check_limits(file_out, pos_out, &count);
1449 /* Don't allow overlapped copying within the same file. */
1450 if (inode_in == inode_out &&
1451 pos_out + count > pos_in &&
1452 pos_out < pos_in + count)
1460 * copy_file_range() differs from regular file read and write in that it
1461 * specifically allows return partial success. When it does so is up to
1462 * the copy_file_range method.
1464 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1465 struct file *file_out, loff_t pos_out,
1466 size_t len, unsigned int flags)
1473 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1478 ret = rw_verify_area(READ, file_in, &pos_in, len);
1482 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1489 file_start_write(file_out);
1492 * Try cloning first, this is supported by more file systems, and
1493 * more efficient if both clone and copy are supported (e.g. NFS).
1495 if (file_in->f_op->remap_file_range &&
1496 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1499 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1501 min_t(loff_t, MAX_RW_COUNT, len),
1502 REMAP_FILE_CAN_SHORTEN);
1509 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1511 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1514 fsnotify_access(file_in);
1515 add_rchar(current, ret);
1516 fsnotify_modify(file_out);
1517 add_wchar(current, ret);
1523 file_end_write(file_out);
1527 EXPORT_SYMBOL(vfs_copy_file_range);
1529 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1530 int, fd_out, loff_t __user *, off_out,
1531 size_t, len, unsigned int, flags)
1537 ssize_t ret = -EBADF;
1539 f_in = fdget(fd_in);
1543 f_out = fdget(fd_out);
1549 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1552 pos_in = f_in.file->f_pos;
1556 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1559 pos_out = f_out.file->f_pos;
1562 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1569 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1572 f_in.file->f_pos = pos_in;
1576 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1579 f_out.file->f_pos = pos_out;
1592 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1593 * LFS limits. If pos is under the limit it becomes a short access. If it
1594 * exceeds the limit we return -EFBIG.
1596 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1598 struct inode *inode = file->f_mapping->host;
1599 loff_t max_size = inode->i_sb->s_maxbytes;
1600 loff_t limit = rlimit(RLIMIT_FSIZE);
1602 if (limit != RLIM_INFINITY) {
1604 send_sig(SIGXFSZ, current, 0);
1607 *count = min(*count, limit - pos);
1610 if (!(file->f_flags & O_LARGEFILE))
1611 max_size = MAX_NON_LFS;
1613 if (unlikely(pos >= max_size))
1616 *count = min(*count, max_size - pos);
1621 /* Like generic_write_checks(), but takes size of write instead of iter. */
1622 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1624 struct file *file = iocb->ki_filp;
1625 struct inode *inode = file->f_mapping->host;
1627 if (IS_SWAPFILE(inode))
1633 /* FIXME: this is for backwards compatibility with 2.4 */
1634 if (iocb->ki_flags & IOCB_APPEND)
1635 iocb->ki_pos = i_size_read(inode);
1637 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
1640 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1642 EXPORT_SYMBOL(generic_write_checks_count);
1645 * Performs necessary checks before doing a write
1647 * Can adjust writing position or amount of bytes to write.
1648 * Returns appropriate error code that caller should return or
1649 * zero in case that write should be allowed.
1651 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1653 loff_t count = iov_iter_count(from);
1656 ret = generic_write_checks_count(iocb, &count);
1660 iov_iter_truncate(from, count);
1661 return iov_iter_count(from);
1663 EXPORT_SYMBOL(generic_write_checks);
1666 * Performs common checks before doing a file copy/clone
1667 * from @file_in to @file_out.
1669 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1671 struct inode *inode_in = file_inode(file_in);
1672 struct inode *inode_out = file_inode(file_out);
1674 /* Don't copy dirs, pipes, sockets... */
1675 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1677 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1680 if (!(file_in->f_mode & FMODE_READ) ||
1681 !(file_out->f_mode & FMODE_WRITE) ||
1682 (file_out->f_flags & O_APPEND))