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)
381 struct iovec iov = { .iov_base = buf, .iov_len = len };
383 struct iov_iter iter;
386 init_sync_kiocb(&kiocb, filp);
387 kiocb.ki_pos = (ppos ? *ppos : 0);
388 iov_iter_init(&iter, READ, &iov, 1, len);
390 ret = call_read_iter(filp, &kiocb, &iter);
391 BUG_ON(ret == -EIOCBQUEUED);
393 *ppos = kiocb.ki_pos;
397 static int warn_unsupported(struct file *file, const char *op)
400 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
401 op, file, current->pid, current->comm);
405 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
409 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
412 struct iov_iter iter;
415 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
417 if (!(file->f_mode & FMODE_CAN_READ))
420 * Also fail if ->read_iter and ->read are both wired up as that
421 * implies very convoluted semantics.
423 if (unlikely(!file->f_op->read_iter || file->f_op->read))
424 return warn_unsupported(file, "read");
426 init_sync_kiocb(&kiocb, file);
427 kiocb.ki_pos = pos ? *pos : 0;
428 iov_iter_kvec(&iter, READ, &iov, 1, iov.iov_len);
429 ret = file->f_op->read_iter(&kiocb, &iter);
433 fsnotify_access(file);
434 add_rchar(current, ret);
440 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
444 ret = rw_verify_area(READ, file, pos, count);
447 return __kernel_read(file, buf, count, pos);
449 EXPORT_SYMBOL(kernel_read);
451 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
455 if (!(file->f_mode & FMODE_READ))
457 if (!(file->f_mode & FMODE_CAN_READ))
459 if (unlikely(!access_ok(buf, count)))
462 ret = rw_verify_area(READ, file, pos, count);
465 if (count > MAX_RW_COUNT)
466 count = MAX_RW_COUNT;
468 if (file->f_op->read)
469 ret = file->f_op->read(file, buf, count, pos);
470 else if (file->f_op->read_iter)
471 ret = new_sync_read(file, buf, count, pos);
475 fsnotify_access(file);
476 add_rchar(current, ret);
482 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
484 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
486 struct iov_iter iter;
489 init_sync_kiocb(&kiocb, filp);
490 kiocb.ki_pos = (ppos ? *ppos : 0);
491 iov_iter_init(&iter, WRITE, &iov, 1, len);
493 ret = call_write_iter(filp, &kiocb, &iter);
494 BUG_ON(ret == -EIOCBQUEUED);
496 *ppos = kiocb.ki_pos;
500 /* caller is responsible for file_start_write/file_end_write */
501 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
504 .iov_base = (void *)buf,
505 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
508 struct iov_iter iter;
511 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
513 if (!(file->f_mode & FMODE_CAN_WRITE))
516 * Also fail if ->write_iter and ->write are both wired up as that
517 * implies very convoluted semantics.
519 if (unlikely(!file->f_op->write_iter || file->f_op->write))
520 return warn_unsupported(file, "write");
522 init_sync_kiocb(&kiocb, file);
523 kiocb.ki_pos = pos ? *pos : 0;
524 iov_iter_kvec(&iter, WRITE, &iov, 1, iov.iov_len);
525 ret = file->f_op->write_iter(&kiocb, &iter);
529 fsnotify_modify(file);
530 add_wchar(current, ret);
536 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
537 * but autofs is one of the few internal kernel users that actually
538 * wants this _and_ can be built as a module. So we need to export
539 * this symbol for autofs, even though it really isn't appropriate
540 * for any other kernel modules.
542 EXPORT_SYMBOL_GPL(__kernel_write);
544 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
549 ret = rw_verify_area(WRITE, file, pos, count);
553 file_start_write(file);
554 ret = __kernel_write(file, buf, count, pos);
555 file_end_write(file);
558 EXPORT_SYMBOL(kernel_write);
560 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
564 if (!(file->f_mode & FMODE_WRITE))
566 if (!(file->f_mode & FMODE_CAN_WRITE))
568 if (unlikely(!access_ok(buf, count)))
571 ret = rw_verify_area(WRITE, file, pos, count);
574 if (count > MAX_RW_COUNT)
575 count = MAX_RW_COUNT;
576 file_start_write(file);
577 if (file->f_op->write)
578 ret = file->f_op->write(file, buf, count, pos);
579 else if (file->f_op->write_iter)
580 ret = new_sync_write(file, buf, count, pos);
584 fsnotify_modify(file);
585 add_wchar(current, ret);
588 file_end_write(file);
592 /* file_ppos returns &file->f_pos or NULL if file is stream */
593 static inline loff_t *file_ppos(struct file *file)
595 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
598 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
600 struct fd f = fdget_pos(fd);
601 ssize_t ret = -EBADF;
604 loff_t pos, *ppos = file_ppos(f.file);
609 ret = vfs_read(f.file, buf, count, ppos);
610 if (ret >= 0 && ppos)
617 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
619 return ksys_read(fd, buf, count);
622 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
624 struct fd f = fdget_pos(fd);
625 ssize_t ret = -EBADF;
628 loff_t pos, *ppos = file_ppos(f.file);
633 ret = vfs_write(f.file, buf, count, ppos);
634 if (ret >= 0 && ppos)
642 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
645 return ksys_write(fd, buf, count);
648 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
652 ssize_t ret = -EBADF;
660 if (f.file->f_mode & FMODE_PREAD)
661 ret = vfs_read(f.file, buf, count, &pos);
668 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
669 size_t, count, loff_t, pos)
671 return ksys_pread64(fd, buf, count, pos);
674 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
675 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
676 size_t, count, compat_arg_u64_dual(pos))
678 return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
682 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
683 size_t count, loff_t pos)
686 ssize_t ret = -EBADF;
694 if (f.file->f_mode & FMODE_PWRITE)
695 ret = vfs_write(f.file, buf, count, &pos);
702 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
703 size_t, count, loff_t, pos)
705 return ksys_pwrite64(fd, buf, count, pos);
708 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
709 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
710 size_t, count, compat_arg_u64_dual(pos))
712 return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
716 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
717 loff_t *ppos, int type, rwf_t flags)
722 init_sync_kiocb(&kiocb, filp);
723 ret = kiocb_set_rw_flags(&kiocb, flags);
726 kiocb.ki_pos = (ppos ? *ppos : 0);
729 ret = call_read_iter(filp, &kiocb, iter);
731 ret = call_write_iter(filp, &kiocb, iter);
732 BUG_ON(ret == -EIOCBQUEUED);
734 *ppos = kiocb.ki_pos;
738 /* Do it by hand, with file-ops */
739 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
740 loff_t *ppos, int type, rwf_t flags)
744 if (flags & ~RWF_HIPRI)
747 while (iov_iter_count(iter)) {
748 struct iovec iovec = iov_iter_iovec(iter);
752 nr = filp->f_op->read(filp, iovec.iov_base,
753 iovec.iov_len, ppos);
755 nr = filp->f_op->write(filp, iovec.iov_base,
756 iovec.iov_len, ppos);
765 if (nr != iovec.iov_len)
767 iov_iter_advance(iter, nr);
773 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
774 loff_t *pos, rwf_t flags)
779 if (!(file->f_mode & FMODE_READ))
781 if (!(file->f_mode & FMODE_CAN_READ))
784 tot_len = iov_iter_count(iter);
787 ret = rw_verify_area(READ, file, pos, tot_len);
791 if (file->f_op->read_iter)
792 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
794 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
797 fsnotify_access(file);
801 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
802 struct iov_iter *iter)
807 if (!file->f_op->read_iter)
809 if (!(file->f_mode & FMODE_READ))
811 if (!(file->f_mode & FMODE_CAN_READ))
814 tot_len = iov_iter_count(iter);
817 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
821 ret = call_read_iter(file, iocb, iter);
824 fsnotify_access(file);
827 EXPORT_SYMBOL(vfs_iocb_iter_read);
829 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
832 if (!file->f_op->read_iter)
834 return do_iter_read(file, iter, ppos, flags);
836 EXPORT_SYMBOL(vfs_iter_read);
838 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
839 loff_t *pos, rwf_t flags)
844 if (!(file->f_mode & FMODE_WRITE))
846 if (!(file->f_mode & FMODE_CAN_WRITE))
849 tot_len = iov_iter_count(iter);
852 ret = rw_verify_area(WRITE, file, pos, tot_len);
856 if (file->f_op->write_iter)
857 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
859 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
861 fsnotify_modify(file);
865 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
866 struct iov_iter *iter)
871 if (!file->f_op->write_iter)
873 if (!(file->f_mode & FMODE_WRITE))
875 if (!(file->f_mode & FMODE_CAN_WRITE))
878 tot_len = iov_iter_count(iter);
881 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
885 ret = call_write_iter(file, iocb, iter);
887 fsnotify_modify(file);
891 EXPORT_SYMBOL(vfs_iocb_iter_write);
893 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
896 if (!file->f_op->write_iter)
898 return do_iter_write(file, iter, ppos, flags);
900 EXPORT_SYMBOL(vfs_iter_write);
902 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
903 unsigned long vlen, loff_t *pos, rwf_t flags)
905 struct iovec iovstack[UIO_FASTIOV];
906 struct iovec *iov = iovstack;
907 struct iov_iter iter;
910 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
912 ret = do_iter_read(file, &iter, pos, flags);
919 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
920 unsigned long vlen, loff_t *pos, rwf_t flags)
922 struct iovec iovstack[UIO_FASTIOV];
923 struct iovec *iov = iovstack;
924 struct iov_iter iter;
927 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
929 file_start_write(file);
930 ret = do_iter_write(file, &iter, pos, flags);
931 file_end_write(file);
937 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
938 unsigned long vlen, rwf_t flags)
940 struct fd f = fdget_pos(fd);
941 ssize_t ret = -EBADF;
944 loff_t pos, *ppos = file_ppos(f.file);
949 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
950 if (ret >= 0 && ppos)
956 add_rchar(current, ret);
961 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
962 unsigned long vlen, rwf_t flags)
964 struct fd f = fdget_pos(fd);
965 ssize_t ret = -EBADF;
968 loff_t pos, *ppos = file_ppos(f.file);
973 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
974 if (ret >= 0 && ppos)
980 add_wchar(current, ret);
985 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
987 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
988 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
991 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
992 unsigned long vlen, loff_t pos, rwf_t flags)
995 ssize_t ret = -EBADF;
1003 if (f.file->f_mode & FMODE_PREAD)
1004 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1009 add_rchar(current, ret);
1014 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1015 unsigned long vlen, loff_t pos, rwf_t flags)
1018 ssize_t ret = -EBADF;
1026 if (f.file->f_mode & FMODE_PWRITE)
1027 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1032 add_wchar(current, ret);
1037 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1038 unsigned long, vlen)
1040 return do_readv(fd, vec, vlen, 0);
1043 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1044 unsigned long, vlen)
1046 return do_writev(fd, vec, vlen, 0);
1049 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1050 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1052 loff_t pos = pos_from_hilo(pos_h, pos_l);
1054 return do_preadv(fd, vec, vlen, pos, 0);
1057 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1058 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1061 loff_t pos = pos_from_hilo(pos_h, pos_l);
1064 return do_readv(fd, vec, vlen, flags);
1066 return do_preadv(fd, vec, vlen, pos, flags);
1069 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1070 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1072 loff_t pos = pos_from_hilo(pos_h, pos_l);
1074 return do_pwritev(fd, vec, vlen, pos, 0);
1077 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1078 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1081 loff_t pos = pos_from_hilo(pos_h, pos_l);
1084 return do_writev(fd, vec, vlen, flags);
1086 return do_pwritev(fd, vec, vlen, pos, flags);
1090 * Various compat syscalls. Note that they all pretend to take a native
1091 * iovec - import_iovec will properly treat those as compat_iovecs based on
1092 * in_compat_syscall().
1094 #ifdef CONFIG_COMPAT
1095 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1096 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1097 const struct iovec __user *, vec,
1098 unsigned long, vlen, loff_t, pos)
1100 return do_preadv(fd, vec, vlen, pos, 0);
1104 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1105 const struct iovec __user *, vec,
1106 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1108 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1110 return do_preadv(fd, vec, vlen, pos, 0);
1113 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1114 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1115 const struct iovec __user *, vec,
1116 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1119 return do_readv(fd, vec, vlen, flags);
1120 return do_preadv(fd, vec, vlen, pos, flags);
1124 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1125 const struct iovec __user *, vec,
1126 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1129 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1132 return do_readv(fd, vec, vlen, flags);
1133 return do_preadv(fd, vec, vlen, pos, flags);
1136 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1137 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1138 const struct iovec __user *, vec,
1139 unsigned long, vlen, loff_t, pos)
1141 return do_pwritev(fd, vec, vlen, pos, 0);
1145 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1146 const struct iovec __user *,vec,
1147 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1149 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1151 return do_pwritev(fd, vec, vlen, pos, 0);
1154 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1155 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1156 const struct iovec __user *, vec,
1157 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1160 return do_writev(fd, vec, vlen, flags);
1161 return do_pwritev(fd, vec, vlen, pos, flags);
1165 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1166 const struct iovec __user *,vec,
1167 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1169 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1172 return do_writev(fd, vec, vlen, flags);
1173 return do_pwritev(fd, vec, vlen, pos, flags);
1175 #endif /* CONFIG_COMPAT */
1177 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1178 size_t count, loff_t max)
1181 struct inode *in_inode, *out_inode;
1182 struct pipe_inode_info *opipe;
1189 * Get input file, and verify that it is ok..
1195 if (!(in.file->f_mode & FMODE_READ))
1199 pos = in.file->f_pos;
1202 if (!(in.file->f_mode & FMODE_PREAD))
1205 retval = rw_verify_area(READ, in.file, &pos, count);
1208 if (count > MAX_RW_COUNT)
1209 count = MAX_RW_COUNT;
1212 * Get output file, and verify that it is ok..
1215 out = fdget(out_fd);
1218 if (!(out.file->f_mode & FMODE_WRITE))
1220 in_inode = file_inode(in.file);
1221 out_inode = file_inode(out.file);
1222 out_pos = out.file->f_pos;
1225 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1227 if (unlikely(pos + count > max)) {
1228 retval = -EOVERFLOW;
1237 * We need to debate whether we can enable this or not. The
1238 * man page documents EAGAIN return for the output at least,
1239 * and the application is arguably buggy if it doesn't expect
1240 * EAGAIN on a non-blocking file descriptor.
1242 if (in.file->f_flags & O_NONBLOCK)
1243 fl = SPLICE_F_NONBLOCK;
1245 opipe = get_pipe_info(out.file, true);
1247 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1250 file_start_write(out.file);
1251 retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1253 file_end_write(out.file);
1255 if (out.file->f_flags & O_NONBLOCK)
1256 fl |= SPLICE_F_NONBLOCK;
1258 retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1262 add_rchar(current, retval);
1263 add_wchar(current, retval);
1264 fsnotify_access(in.file);
1265 fsnotify_modify(out.file);
1266 out.file->f_pos = out_pos;
1270 in.file->f_pos = pos;
1276 retval = -EOVERFLOW;
1286 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1293 if (unlikely(get_user(off, offset)))
1296 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1297 if (unlikely(put_user(pos, offset)))
1302 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1305 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1311 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1313 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1314 if (unlikely(put_user(pos, offset)))
1319 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1322 #ifdef CONFIG_COMPAT
1323 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1324 compat_off_t __user *, offset, compat_size_t, count)
1331 if (unlikely(get_user(off, offset)))
1334 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1335 if (unlikely(put_user(pos, offset)))
1340 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1343 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1344 compat_loff_t __user *, offset, compat_size_t, count)
1350 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1352 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1353 if (unlikely(put_user(pos, offset)))
1358 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1363 * generic_copy_file_range - copy data between two files
1364 * @file_in: file structure to read from
1365 * @pos_in: file offset to read from
1366 * @file_out: file structure to write data to
1367 * @pos_out: file offset to write data to
1368 * @len: amount of data to copy
1369 * @flags: copy flags
1371 * This is a generic filesystem helper to copy data from one file to another.
1372 * It has no constraints on the source or destination file owners - the files
1373 * can belong to different superblocks and different filesystem types. Short
1374 * copies are allowed.
1376 * This should be called from the @file_out filesystem, as per the
1377 * ->copy_file_range() method.
1379 * Returns the number of bytes copied or a negative error indicating the
1383 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1384 struct file *file_out, loff_t pos_out,
1385 size_t len, unsigned int flags)
1387 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1388 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1390 EXPORT_SYMBOL(generic_copy_file_range);
1393 * Performs necessary checks before doing a file copy
1395 * Can adjust amount of bytes to copy via @req_count argument.
1396 * Returns appropriate error code that caller should return or
1397 * zero in case the copy should be allowed.
1399 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1400 struct file *file_out, loff_t pos_out,
1401 size_t *req_count, unsigned int flags)
1403 struct inode *inode_in = file_inode(file_in);
1404 struct inode *inode_out = file_inode(file_out);
1405 uint64_t count = *req_count;
1409 ret = generic_file_rw_checks(file_in, file_out);
1414 * We allow some filesystems to handle cross sb copy, but passing
1415 * a file of the wrong filesystem type to filesystem driver can result
1416 * in an attempt to dereference the wrong type of ->private_data, so
1417 * avoid doing that until we really have a good reason.
1419 * nfs and cifs define several different file_system_type structures
1420 * and several different sets of file_operations, but they all end up
1421 * using the same ->copy_file_range() function pointer.
1423 if (file_out->f_op->copy_file_range) {
1424 if (file_in->f_op->copy_file_range !=
1425 file_out->f_op->copy_file_range)
1427 } else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1431 /* Don't touch certain kinds of inodes */
1432 if (IS_IMMUTABLE(inode_out))
1435 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1438 /* Ensure offsets don't wrap. */
1439 if (pos_in + count < pos_in || pos_out + count < pos_out)
1442 /* Shorten the copy to EOF */
1443 size_in = i_size_read(inode_in);
1444 if (pos_in >= size_in)
1447 count = min(count, size_in - (uint64_t)pos_in);
1449 ret = generic_write_check_limits(file_out, pos_out, &count);
1453 /* Don't allow overlapped copying within the same file. */
1454 if (inode_in == inode_out &&
1455 pos_out + count > pos_in &&
1456 pos_out < pos_in + count)
1464 * copy_file_range() differs from regular file read and write in that it
1465 * specifically allows return partial success. When it does so is up to
1466 * the copy_file_range method.
1468 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1469 struct file *file_out, loff_t pos_out,
1470 size_t len, unsigned int flags)
1477 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1482 ret = rw_verify_area(READ, file_in, &pos_in, len);
1486 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1493 file_start_write(file_out);
1496 * Cloning is supported by more file systems, so we implement copy on
1497 * same sb using clone, but for filesystems where both clone and copy
1498 * are supported (e.g. nfs,cifs), we only call the copy method.
1500 if (file_out->f_op->copy_file_range) {
1501 ret = file_out->f_op->copy_file_range(file_in, pos_in,
1507 if (file_in->f_op->remap_file_range &&
1508 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1509 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1511 min_t(loff_t, MAX_RW_COUNT, len),
1512 REMAP_FILE_CAN_SHORTEN);
1518 * We can get here for same sb copy of filesystems that do not implement
1519 * ->copy_file_range() in case filesystem does not support clone or in
1520 * case filesystem supports clone but rejected the clone request (e.g.
1521 * because it was not block aligned).
1523 * In both cases, fall back to kernel copy so we are able to maintain a
1524 * consistent story about which filesystems support copy_file_range()
1525 * and which filesystems do not, that will allow userspace tools to
1526 * make consistent desicions w.r.t using copy_file_range().
1528 ret = generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1533 fsnotify_access(file_in);
1534 add_rchar(current, ret);
1535 fsnotify_modify(file_out);
1536 add_wchar(current, ret);
1542 file_end_write(file_out);
1546 EXPORT_SYMBOL(vfs_copy_file_range);
1548 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1549 int, fd_out, loff_t __user *, off_out,
1550 size_t, len, unsigned int, flags)
1556 ssize_t ret = -EBADF;
1558 f_in = fdget(fd_in);
1562 f_out = fdget(fd_out);
1568 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1571 pos_in = f_in.file->f_pos;
1575 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1578 pos_out = f_out.file->f_pos;
1581 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1588 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1591 f_in.file->f_pos = pos_in;
1595 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1598 f_out.file->f_pos = pos_out;
1611 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1612 * LFS limits. If pos is under the limit it becomes a short access. If it
1613 * exceeds the limit we return -EFBIG.
1615 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1617 struct inode *inode = file->f_mapping->host;
1618 loff_t max_size = inode->i_sb->s_maxbytes;
1619 loff_t limit = rlimit(RLIMIT_FSIZE);
1621 if (limit != RLIM_INFINITY) {
1623 send_sig(SIGXFSZ, current, 0);
1626 *count = min(*count, limit - pos);
1629 if (!(file->f_flags & O_LARGEFILE))
1630 max_size = MAX_NON_LFS;
1632 if (unlikely(pos >= max_size))
1635 *count = min(*count, max_size - pos);
1640 /* Like generic_write_checks(), but takes size of write instead of iter. */
1641 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1643 struct file *file = iocb->ki_filp;
1644 struct inode *inode = file->f_mapping->host;
1646 if (IS_SWAPFILE(inode))
1652 if (iocb->ki_flags & IOCB_APPEND)
1653 iocb->ki_pos = i_size_read(inode);
1655 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1656 !((iocb->ki_flags & IOCB_DIRECT) ||
1657 (file->f_mode & FMODE_BUF_WASYNC)))
1660 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1662 EXPORT_SYMBOL(generic_write_checks_count);
1665 * Performs necessary checks before doing a write
1667 * Can adjust writing position or amount of bytes to write.
1668 * Returns appropriate error code that caller should return or
1669 * zero in case that write should be allowed.
1671 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1673 loff_t count = iov_iter_count(from);
1676 ret = generic_write_checks_count(iocb, &count);
1680 iov_iter_truncate(from, count);
1681 return iov_iter_count(from);
1683 EXPORT_SYMBOL(generic_write_checks);
1686 * Performs common checks before doing a file copy/clone
1687 * from @file_in to @file_out.
1689 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1691 struct inode *inode_in = file_inode(file_in);
1692 struct inode *inode_out = file_inode(file_out);
1694 /* Don't copy dirs, pipes, sockets... */
1695 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1697 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1700 if (!(file_in->f_mode & FMODE_READ) ||
1701 !(file_out->f_mode & FMODE_WRITE) ||
1702 (file_out->f_flags & O_APPEND))