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)
369 int retval = -EINVAL;
371 inode = file_inode(file);
372 if (unlikely((ssize_t) count < 0))
376 * ranged mandatory locking does not apply to streams - it makes sense
377 * only for files where position has a meaning.
382 if (unlikely(pos < 0)) {
383 if (!unsigned_offsets(file))
385 if (count >= -pos) /* both values are in 0..LLONG_MAX */
387 } else if (unlikely((loff_t) (pos + count) < 0)) {
388 if (!unsigned_offsets(file))
392 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
393 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
394 read_write == READ ? F_RDLCK : F_WRLCK);
400 return security_file_permission(file,
401 read_write == READ ? MAY_READ : MAY_WRITE);
404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
406 struct iovec iov = { .iov_base = buf, .iov_len = len };
408 struct iov_iter iter;
411 init_sync_kiocb(&kiocb, filp);
412 kiocb.ki_pos = (ppos ? *ppos : 0);
413 iov_iter_init(&iter, READ, &iov, 1, len);
415 ret = call_read_iter(filp, &kiocb, &iter);
416 BUG_ON(ret == -EIOCBQUEUED);
418 *ppos = kiocb.ki_pos;
422 static int warn_unsupported(struct file *file, const char *op)
425 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
426 op, file, current->pid, current->comm);
430 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
434 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
437 struct iov_iter iter;
440 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
442 if (!(file->f_mode & FMODE_CAN_READ))
445 * Also fail if ->read_iter and ->read are both wired up as that
446 * implies very convoluted semantics.
448 if (unlikely(!file->f_op->read_iter || file->f_op->read))
449 return warn_unsupported(file, "read");
451 init_sync_kiocb(&kiocb, file);
452 kiocb.ki_pos = pos ? *pos : 0;
453 iov_iter_kvec(&iter, READ, &iov, 1, iov.iov_len);
454 ret = file->f_op->read_iter(&kiocb, &iter);
458 fsnotify_access(file);
459 add_rchar(current, ret);
465 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
469 ret = rw_verify_area(READ, file, pos, count);
472 return __kernel_read(file, buf, count, pos);
474 EXPORT_SYMBOL(kernel_read);
476 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
480 if (!(file->f_mode & FMODE_READ))
482 if (!(file->f_mode & FMODE_CAN_READ))
484 if (unlikely(!access_ok(buf, count)))
487 ret = rw_verify_area(READ, file, pos, count);
490 if (count > MAX_RW_COUNT)
491 count = MAX_RW_COUNT;
493 if (file->f_op->read)
494 ret = file->f_op->read(file, buf, count, pos);
495 else if (file->f_op->read_iter)
496 ret = new_sync_read(file, buf, count, pos);
500 fsnotify_access(file);
501 add_rchar(current, ret);
507 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
509 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
511 struct iov_iter iter;
514 init_sync_kiocb(&kiocb, filp);
515 kiocb.ki_pos = (ppos ? *ppos : 0);
516 iov_iter_init(&iter, WRITE, &iov, 1, len);
518 ret = call_write_iter(filp, &kiocb, &iter);
519 BUG_ON(ret == -EIOCBQUEUED);
521 *ppos = kiocb.ki_pos;
525 /* caller is responsible for file_start_write/file_end_write */
526 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
529 .iov_base = (void *)buf,
530 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
533 struct iov_iter iter;
536 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
538 if (!(file->f_mode & FMODE_CAN_WRITE))
541 * Also fail if ->write_iter and ->write are both wired up as that
542 * implies very convoluted semantics.
544 if (unlikely(!file->f_op->write_iter || file->f_op->write))
545 return warn_unsupported(file, "write");
547 init_sync_kiocb(&kiocb, file);
548 kiocb.ki_pos = pos ? *pos : 0;
549 iov_iter_kvec(&iter, WRITE, &iov, 1, iov.iov_len);
550 ret = file->f_op->write_iter(&kiocb, &iter);
554 fsnotify_modify(file);
555 add_wchar(current, ret);
561 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
562 * but autofs is one of the few internal kernel users that actually
563 * wants this _and_ can be built as a module. So we need to export
564 * this symbol for autofs, even though it really isn't appropriate
565 * for any other kernel modules.
567 EXPORT_SYMBOL_GPL(__kernel_write);
569 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
574 ret = rw_verify_area(WRITE, file, pos, count);
578 file_start_write(file);
579 ret = __kernel_write(file, buf, count, pos);
580 file_end_write(file);
583 EXPORT_SYMBOL(kernel_write);
585 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
589 if (!(file->f_mode & FMODE_WRITE))
591 if (!(file->f_mode & FMODE_CAN_WRITE))
593 if (unlikely(!access_ok(buf, count)))
596 ret = rw_verify_area(WRITE, file, pos, count);
599 if (count > MAX_RW_COUNT)
600 count = MAX_RW_COUNT;
601 file_start_write(file);
602 if (file->f_op->write)
603 ret = file->f_op->write(file, buf, count, pos);
604 else if (file->f_op->write_iter)
605 ret = new_sync_write(file, buf, count, pos);
609 fsnotify_modify(file);
610 add_wchar(current, ret);
613 file_end_write(file);
617 /* file_ppos returns &file->f_pos or NULL if file is stream */
618 static inline loff_t *file_ppos(struct file *file)
620 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
623 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
625 struct fd f = fdget_pos(fd);
626 ssize_t ret = -EBADF;
629 loff_t pos, *ppos = file_ppos(f.file);
634 ret = vfs_read(f.file, buf, count, ppos);
635 if (ret >= 0 && ppos)
642 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
644 return ksys_read(fd, buf, count);
647 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
649 struct fd f = fdget_pos(fd);
650 ssize_t ret = -EBADF;
653 loff_t pos, *ppos = file_ppos(f.file);
658 ret = vfs_write(f.file, buf, count, ppos);
659 if (ret >= 0 && ppos)
667 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
670 return ksys_write(fd, buf, count);
673 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
677 ssize_t ret = -EBADF;
685 if (f.file->f_mode & FMODE_PREAD)
686 ret = vfs_read(f.file, buf, count, &pos);
693 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
694 size_t, count, loff_t, pos)
696 return ksys_pread64(fd, buf, count, pos);
699 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
700 size_t count, loff_t pos)
703 ssize_t ret = -EBADF;
711 if (f.file->f_mode & FMODE_PWRITE)
712 ret = vfs_write(f.file, buf, count, &pos);
719 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
720 size_t, count, loff_t, pos)
722 return ksys_pwrite64(fd, buf, count, pos);
725 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
726 loff_t *ppos, int type, rwf_t flags)
731 init_sync_kiocb(&kiocb, filp);
732 ret = kiocb_set_rw_flags(&kiocb, flags);
735 kiocb.ki_pos = (ppos ? *ppos : 0);
738 ret = call_read_iter(filp, &kiocb, iter);
740 ret = call_write_iter(filp, &kiocb, iter);
741 BUG_ON(ret == -EIOCBQUEUED);
743 *ppos = kiocb.ki_pos;
747 /* Do it by hand, with file-ops */
748 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
749 loff_t *ppos, int type, rwf_t flags)
753 if (flags & ~RWF_HIPRI)
756 while (iov_iter_count(iter)) {
757 struct iovec iovec = iov_iter_iovec(iter);
761 nr = filp->f_op->read(filp, iovec.iov_base,
762 iovec.iov_len, ppos);
764 nr = filp->f_op->write(filp, iovec.iov_base,
765 iovec.iov_len, ppos);
774 if (nr != iovec.iov_len)
776 iov_iter_advance(iter, nr);
782 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
783 loff_t *pos, rwf_t flags)
788 if (!(file->f_mode & FMODE_READ))
790 if (!(file->f_mode & FMODE_CAN_READ))
793 tot_len = iov_iter_count(iter);
796 ret = rw_verify_area(READ, file, pos, tot_len);
800 if (file->f_op->read_iter)
801 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
803 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
806 fsnotify_access(file);
810 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
811 struct iov_iter *iter)
816 if (!file->f_op->read_iter)
818 if (!(file->f_mode & FMODE_READ))
820 if (!(file->f_mode & FMODE_CAN_READ))
823 tot_len = iov_iter_count(iter);
826 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
830 ret = call_read_iter(file, iocb, iter);
833 fsnotify_access(file);
836 EXPORT_SYMBOL(vfs_iocb_iter_read);
838 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
841 if (!file->f_op->read_iter)
843 return do_iter_read(file, iter, ppos, flags);
845 EXPORT_SYMBOL(vfs_iter_read);
847 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
848 loff_t *pos, rwf_t flags)
853 if (!(file->f_mode & FMODE_WRITE))
855 if (!(file->f_mode & FMODE_CAN_WRITE))
858 tot_len = iov_iter_count(iter);
861 ret = rw_verify_area(WRITE, file, pos, tot_len);
865 if (file->f_op->write_iter)
866 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
868 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
870 fsnotify_modify(file);
874 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
875 struct iov_iter *iter)
880 if (!file->f_op->write_iter)
882 if (!(file->f_mode & FMODE_WRITE))
884 if (!(file->f_mode & FMODE_CAN_WRITE))
887 tot_len = iov_iter_count(iter);
890 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
894 ret = call_write_iter(file, iocb, iter);
896 fsnotify_modify(file);
900 EXPORT_SYMBOL(vfs_iocb_iter_write);
902 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
905 if (!file->f_op->write_iter)
907 return do_iter_write(file, iter, ppos, flags);
909 EXPORT_SYMBOL(vfs_iter_write);
911 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
912 unsigned long vlen, loff_t *pos, rwf_t flags)
914 struct iovec iovstack[UIO_FASTIOV];
915 struct iovec *iov = iovstack;
916 struct iov_iter iter;
919 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
921 ret = do_iter_read(file, &iter, pos, flags);
928 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
929 unsigned long vlen, loff_t *pos, rwf_t flags)
931 struct iovec iovstack[UIO_FASTIOV];
932 struct iovec *iov = iovstack;
933 struct iov_iter iter;
936 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
938 file_start_write(file);
939 ret = do_iter_write(file, &iter, pos, flags);
940 file_end_write(file);
946 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
947 unsigned long vlen, rwf_t flags)
949 struct fd f = fdget_pos(fd);
950 ssize_t ret = -EBADF;
953 loff_t pos, *ppos = file_ppos(f.file);
958 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
959 if (ret >= 0 && ppos)
965 add_rchar(current, ret);
970 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
971 unsigned long vlen, rwf_t flags)
973 struct fd f = fdget_pos(fd);
974 ssize_t ret = -EBADF;
977 loff_t pos, *ppos = file_ppos(f.file);
982 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
983 if (ret >= 0 && ppos)
989 add_wchar(current, ret);
994 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
996 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
997 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1000 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1001 unsigned long vlen, loff_t pos, rwf_t flags)
1004 ssize_t ret = -EBADF;
1012 if (f.file->f_mode & FMODE_PREAD)
1013 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1018 add_rchar(current, ret);
1023 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1024 unsigned long vlen, loff_t pos, rwf_t flags)
1027 ssize_t ret = -EBADF;
1035 if (f.file->f_mode & FMODE_PWRITE)
1036 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1041 add_wchar(current, ret);
1046 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1047 unsigned long, vlen)
1049 return do_readv(fd, vec, vlen, 0);
1052 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1053 unsigned long, vlen)
1055 return do_writev(fd, vec, vlen, 0);
1058 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1059 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1061 loff_t pos = pos_from_hilo(pos_h, pos_l);
1063 return do_preadv(fd, vec, vlen, pos, 0);
1066 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1067 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1070 loff_t pos = pos_from_hilo(pos_h, pos_l);
1073 return do_readv(fd, vec, vlen, flags);
1075 return do_preadv(fd, vec, vlen, pos, flags);
1078 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1079 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1081 loff_t pos = pos_from_hilo(pos_h, pos_l);
1083 return do_pwritev(fd, vec, vlen, pos, 0);
1086 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1087 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1090 loff_t pos = pos_from_hilo(pos_h, pos_l);
1093 return do_writev(fd, vec, vlen, flags);
1095 return do_pwritev(fd, vec, vlen, pos, flags);
1099 * Various compat syscalls. Note that they all pretend to take a native
1100 * iovec - import_iovec will properly treat those as compat_iovecs based on
1101 * in_compat_syscall().
1103 #ifdef CONFIG_COMPAT
1104 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1105 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1106 const struct iovec __user *, vec,
1107 unsigned long, vlen, loff_t, pos)
1109 return do_preadv(fd, vec, vlen, pos, 0);
1113 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1114 const struct iovec __user *, vec,
1115 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1117 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1119 return do_preadv(fd, vec, vlen, pos, 0);
1122 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1123 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1124 const struct iovec __user *, vec,
1125 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1128 return do_readv(fd, vec, vlen, flags);
1129 return do_preadv(fd, vec, vlen, pos, flags);
1133 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1134 const struct iovec __user *, vec,
1135 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1138 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1141 return do_readv(fd, vec, vlen, flags);
1142 return do_preadv(fd, vec, vlen, pos, flags);
1145 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1146 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1147 const struct iovec __user *, vec,
1148 unsigned long, vlen, loff_t, pos)
1150 return do_pwritev(fd, vec, vlen, pos, 0);
1154 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1155 const struct iovec __user *,vec,
1156 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1158 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1160 return do_pwritev(fd, vec, vlen, pos, 0);
1163 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1164 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1165 const struct iovec __user *, vec,
1166 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1169 return do_writev(fd, vec, vlen, flags);
1170 return do_pwritev(fd, vec, vlen, pos, flags);
1174 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1175 const struct iovec __user *,vec,
1176 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1178 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1181 return do_writev(fd, vec, vlen, flags);
1182 return do_pwritev(fd, vec, vlen, pos, flags);
1184 #endif /* CONFIG_COMPAT */
1186 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1187 size_t count, loff_t max)
1190 struct inode *in_inode, *out_inode;
1197 * Get input file, and verify that it is ok..
1203 if (!(in.file->f_mode & FMODE_READ))
1207 pos = in.file->f_pos;
1210 if (!(in.file->f_mode & FMODE_PREAD))
1213 retval = rw_verify_area(READ, in.file, &pos, count);
1216 if (count > MAX_RW_COUNT)
1217 count = MAX_RW_COUNT;
1220 * Get output file, and verify that it is ok..
1223 out = fdget(out_fd);
1226 if (!(out.file->f_mode & FMODE_WRITE))
1228 in_inode = file_inode(in.file);
1229 out_inode = file_inode(out.file);
1230 out_pos = out.file->f_pos;
1231 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1236 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1238 if (unlikely(pos + count > max)) {
1239 retval = -EOVERFLOW;
1248 * We need to debate whether we can enable this or not. The
1249 * man page documents EAGAIN return for the output at least,
1250 * and the application is arguably buggy if it doesn't expect
1251 * EAGAIN on a non-blocking file descriptor.
1253 if (in.file->f_flags & O_NONBLOCK)
1254 fl = SPLICE_F_NONBLOCK;
1256 file_start_write(out.file);
1257 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1258 file_end_write(out.file);
1261 add_rchar(current, retval);
1262 add_wchar(current, retval);
1263 fsnotify_access(in.file);
1264 fsnotify_modify(out.file);
1265 out.file->f_pos = out_pos;
1269 in.file->f_pos = pos;
1275 retval = -EOVERFLOW;
1285 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1292 if (unlikely(get_user(off, offset)))
1295 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1296 if (unlikely(put_user(pos, offset)))
1301 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1304 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1310 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1312 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1313 if (unlikely(put_user(pos, offset)))
1318 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1321 #ifdef CONFIG_COMPAT
1322 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1323 compat_off_t __user *, offset, compat_size_t, count)
1330 if (unlikely(get_user(off, offset)))
1333 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1334 if (unlikely(put_user(pos, offset)))
1339 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1342 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1343 compat_loff_t __user *, offset, compat_size_t, count)
1349 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1351 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1352 if (unlikely(put_user(pos, offset)))
1357 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1362 * generic_copy_file_range - copy data between two files
1363 * @file_in: file structure to read from
1364 * @pos_in: file offset to read from
1365 * @file_out: file structure to write data to
1366 * @pos_out: file offset to write data to
1367 * @len: amount of data to copy
1368 * @flags: copy flags
1370 * This is a generic filesystem helper to copy data from one file to another.
1371 * It has no constraints on the source or destination file owners - the files
1372 * can belong to different superblocks and different filesystem types. Short
1373 * copies are allowed.
1375 * This should be called from the @file_out filesystem, as per the
1376 * ->copy_file_range() method.
1378 * Returns the number of bytes copied or a negative error indicating the
1382 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1383 struct file *file_out, loff_t pos_out,
1384 size_t len, unsigned int flags)
1386 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1387 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1389 EXPORT_SYMBOL(generic_copy_file_range);
1391 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1392 struct file *file_out, loff_t pos_out,
1393 size_t len, unsigned int flags)
1396 * Although we now allow filesystems to handle cross sb copy, passing
1397 * a file of the wrong filesystem type to filesystem driver can result
1398 * in an attempt to dereference the wrong type of ->private_data, so
1399 * avoid doing that until we really have a good reason. NFS defines
1400 * several different file_system_type structures, but they all end up
1401 * using the same ->copy_file_range() function pointer.
1403 if (file_out->f_op->copy_file_range &&
1404 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1405 return file_out->f_op->copy_file_range(file_in, pos_in,
1409 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1414 * copy_file_range() differs from regular file read and write in that it
1415 * specifically allows return partial success. When it does so is up to
1416 * the copy_file_range method.
1418 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1419 struct file *file_out, loff_t pos_out,
1420 size_t len, unsigned int flags)
1427 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1432 ret = rw_verify_area(READ, file_in, &pos_in, len);
1436 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1443 file_start_write(file_out);
1446 * Try cloning first, this is supported by more file systems, and
1447 * more efficient if both clone and copy are supported (e.g. NFS).
1449 if (file_in->f_op->remap_file_range &&
1450 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1453 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1455 min_t(loff_t, MAX_RW_COUNT, len),
1456 REMAP_FILE_CAN_SHORTEN);
1463 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1465 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1468 fsnotify_access(file_in);
1469 add_rchar(current, ret);
1470 fsnotify_modify(file_out);
1471 add_wchar(current, ret);
1477 file_end_write(file_out);
1481 EXPORT_SYMBOL(vfs_copy_file_range);
1483 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1484 int, fd_out, loff_t __user *, off_out,
1485 size_t, len, unsigned int, flags)
1491 ssize_t ret = -EBADF;
1493 f_in = fdget(fd_in);
1497 f_out = fdget(fd_out);
1503 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1506 pos_in = f_in.file->f_pos;
1510 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1513 pos_out = f_out.file->f_pos;
1516 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1523 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1526 f_in.file->f_pos = pos_in;
1530 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1533 f_out.file->f_pos = pos_out;
1545 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1548 struct inode *inode = file_inode(file);
1550 if (unlikely(pos < 0 || len < 0))
1553 if (unlikely((loff_t) (pos + len) < 0))
1556 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1557 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1560 retval = locks_mandatory_area(inode, file, pos, end,
1561 write ? F_WRLCK : F_RDLCK);
1566 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1569 * Ensure that we don't remap a partial EOF block in the middle of something
1570 * else. Assume that the offsets have already been checked for block
1573 * For clone we only link a partial EOF block above or at the destination file's
1574 * EOF. For deduplication we accept a partial EOF block only if it ends at the
1575 * destination file's EOF (can not link it into the middle of a file).
1577 * Shorten the request if possible.
1579 static int generic_remap_check_len(struct inode *inode_in,
1580 struct inode *inode_out,
1583 unsigned int remap_flags)
1585 u64 blkmask = i_blocksize(inode_in) - 1;
1586 loff_t new_len = *len;
1588 if ((*len & blkmask) == 0)
1591 if (pos_out + *len < i_size_read(inode_out))
1592 new_len &= ~blkmask;
1594 if (new_len == *len)
1597 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1602 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1605 /* Read a page's worth of file data into the page cache. */
1606 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1610 page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1613 if (!PageUptodate(page)) {
1615 return ERR_PTR(-EIO);
1621 * Lock two pages, ensuring that we lock in offset order if the pages are from
1624 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1626 /* Always lock in order of increasing index. */
1627 if (page1->index > page2->index)
1635 /* Unlock two pages, being careful not to unlock the same page twice. */
1636 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1644 * Compare extents of two files to see if they are the same.
1645 * Caller must have locked both inodes to prevent write races.
1647 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1648 struct inode *dest, loff_t destoff,
1649 loff_t len, bool *is_same)
1655 struct page *src_page;
1656 struct page *dest_page;
1664 src_poff = srcoff & (PAGE_SIZE - 1);
1665 dest_poff = destoff & (PAGE_SIZE - 1);
1666 cmp_len = min(PAGE_SIZE - src_poff,
1667 PAGE_SIZE - dest_poff);
1668 cmp_len = min(cmp_len, len);
1672 src_page = vfs_dedupe_get_page(src, srcoff);
1673 if (IS_ERR(src_page)) {
1674 error = PTR_ERR(src_page);
1677 dest_page = vfs_dedupe_get_page(dest, destoff);
1678 if (IS_ERR(dest_page)) {
1679 error = PTR_ERR(dest_page);
1684 vfs_lock_two_pages(src_page, dest_page);
1687 * Now that we've locked both pages, make sure they're still
1688 * mapped to the file data we're interested in. If not,
1689 * someone is invalidating pages on us and we lose.
1691 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1692 src_page->mapping != src->i_mapping ||
1693 dest_page->mapping != dest->i_mapping) {
1698 src_addr = kmap_atomic(src_page);
1699 dest_addr = kmap_atomic(dest_page);
1701 flush_dcache_page(src_page);
1702 flush_dcache_page(dest_page);
1704 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1707 kunmap_atomic(dest_addr);
1708 kunmap_atomic(src_addr);
1710 vfs_unlock_two_pages(src_page, dest_page);
1711 put_page(dest_page);
1730 * Check that the two inodes are eligible for cloning, the ranges make
1731 * sense, and then flush all dirty data. Caller must ensure that the
1732 * inodes have been locked against any other modifications.
1734 * If there's an error, then the usual negative error code is returned.
1735 * Otherwise returns 0 with *len set to the request length.
1737 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
1738 struct file *file_out, loff_t pos_out,
1739 loff_t *len, unsigned int remap_flags)
1741 struct inode *inode_in = file_inode(file_in);
1742 struct inode *inode_out = file_inode(file_out);
1743 bool same_inode = (inode_in == inode_out);
1746 /* Don't touch certain kinds of inodes */
1747 if (IS_IMMUTABLE(inode_out))
1750 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1753 /* Don't reflink dirs, pipes, sockets... */
1754 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1756 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1759 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1761 loff_t isize = i_size_read(inode_in);
1763 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
1767 *len = isize - pos_in;
1772 /* Check that we don't violate system file offset limits. */
1773 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
1778 /* Wait for the completion of any pending IOs on both files */
1779 inode_dio_wait(inode_in);
1781 inode_dio_wait(inode_out);
1783 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1784 pos_in, pos_in + *len - 1);
1788 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1789 pos_out, pos_out + *len - 1);
1794 * Check that the extents are the same.
1796 if (remap_flags & REMAP_FILE_DEDUP) {
1797 bool is_same = false;
1799 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1800 inode_out, pos_out, *len, &is_same);
1807 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
1812 /* If can't alter the file contents, we're done. */
1813 if (!(remap_flags & REMAP_FILE_DEDUP))
1814 ret = file_modified(file_out);
1818 EXPORT_SYMBOL(generic_remap_file_range_prep);
1820 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
1821 struct file *file_out, loff_t pos_out,
1822 loff_t len, unsigned int remap_flags)
1826 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
1829 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1830 * the same mount. Practically, they only need to be on the same file
1833 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
1836 ret = generic_file_rw_checks(file_in, file_out);
1840 if (!file_in->f_op->remap_file_range)
1843 ret = remap_verify_area(file_in, pos_in, len, false);
1847 ret = remap_verify_area(file_out, pos_out, len, true);
1851 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1852 file_out, pos_out, len, remap_flags);
1856 fsnotify_access(file_in);
1857 fsnotify_modify(file_out);
1860 EXPORT_SYMBOL(do_clone_file_range);
1862 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1863 struct file *file_out, loff_t pos_out,
1864 loff_t len, unsigned int remap_flags)
1868 file_start_write(file_out);
1869 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
1871 file_end_write(file_out);
1875 EXPORT_SYMBOL(vfs_clone_file_range);
1877 /* Check whether we are allowed to dedupe the destination file */
1878 static bool allow_file_dedupe(struct file *file)
1880 if (capable(CAP_SYS_ADMIN))
1882 if (file->f_mode & FMODE_WRITE)
1884 if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
1886 if (!inode_permission(file_inode(file), MAY_WRITE))
1891 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
1892 struct file *dst_file, loff_t dst_pos,
1893 loff_t len, unsigned int remap_flags)
1897 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
1898 REMAP_FILE_CAN_SHORTEN));
1900 ret = mnt_want_write_file(dst_file);
1904 ret = remap_verify_area(dst_file, dst_pos, len, true);
1906 goto out_drop_write;
1909 if (!allow_file_dedupe(dst_file))
1910 goto out_drop_write;
1913 if (src_file->f_path.mnt != dst_file->f_path.mnt)
1914 goto out_drop_write;
1917 if (S_ISDIR(file_inode(dst_file)->i_mode))
1918 goto out_drop_write;
1921 if (!dst_file->f_op->remap_file_range)
1922 goto out_drop_write;
1926 goto out_drop_write;
1929 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
1930 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
1932 mnt_drop_write_file(dst_file);
1936 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
1938 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1940 struct file_dedupe_range_info *info;
1941 struct inode *src = file_inode(file);
1946 u16 count = same->dest_count;
1949 if (!(file->f_mode & FMODE_READ))
1952 if (same->reserved1 || same->reserved2)
1955 off = same->src_offset;
1956 len = same->src_length;
1958 if (S_ISDIR(src->i_mode))
1961 if (!S_ISREG(src->i_mode))
1964 if (!file->f_op->remap_file_range)
1967 ret = remap_verify_area(file, off, len, false);
1972 if (off + len > i_size_read(src))
1975 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
1976 len = min_t(u64, len, 1 << 30);
1978 /* pre-format output fields to sane values */
1979 for (i = 0; i < count; i++) {
1980 same->info[i].bytes_deduped = 0ULL;
1981 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1984 for (i = 0, info = same->info; i < count; i++, info++) {
1985 struct fd dst_fd = fdget(info->dest_fd);
1986 struct file *dst_file = dst_fd.file;
1989 info->status = -EBADF;
1993 if (info->reserved) {
1994 info->status = -EINVAL;
1998 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
1999 info->dest_offset, len,
2000 REMAP_FILE_CAN_SHORTEN);
2001 if (deduped == -EBADE)
2002 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2003 else if (deduped < 0)
2004 info->status = deduped;
2006 info->bytes_deduped = len;
2011 if (fatal_signal_pending(current))
2016 EXPORT_SYMBOL(vfs_dedupe_file_range);