2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include "read_write.h"
20 #include <asm/uaccess.h>
21 #include <asm/unistd.h>
23 const struct file_operations generic_ro_fops = {
24 .llseek = generic_file_llseek,
26 .aio_read = generic_file_aio_read,
27 .mmap = generic_file_readonly_mmap,
28 .splice_read = generic_file_splice_read,
31 EXPORT_SYMBOL(generic_ro_fops);
33 static inline int unsigned_offsets(struct file *file)
35 return file->f_mode & FMODE_UNSIGNED_OFFSET;
38 static loff_t lseek_execute(struct file *file, struct inode *inode,
39 loff_t offset, loff_t maxsize)
41 if (offset < 0 && !unsigned_offsets(file))
46 if (offset != file->f_pos) {
54 * generic_file_llseek_size - generic llseek implementation for regular files
55 * @file: file structure to seek on
56 * @offset: file offset to seek to
57 * @whence: type of seek
58 * @size: max size of this file in file system
59 * @eof: offset used for SEEK_END position
61 * This is a variant of generic_file_llseek that allows passing in a custom
62 * maximum file size and a custom EOF position, for e.g. hashed directories
65 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
66 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
67 * read/writes behave like SEEK_SET against seeks.
70 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
71 loff_t maxsize, loff_t eof)
73 struct inode *inode = file->f_mapping->host;
81 * Here we special-case the lseek(fd, 0, SEEK_CUR)
82 * position-querying operation. Avoid rewriting the "same"
83 * f_pos value back to the file because a concurrent read(),
84 * write() or lseek() might have altered it
89 * f_lock protects against read/modify/write race with other
90 * SEEK_CURs. Note that parallel writes and reads behave
93 spin_lock(&file->f_lock);
94 offset = lseek_execute(file, inode, file->f_pos + offset,
96 spin_unlock(&file->f_lock);
100 * In the generic case the entire file is data, so as long as
101 * offset isn't at the end of the file then the offset is data.
108 * There is a virtual hole at the end of the file, so as long as
109 * offset isn't i_size or larger, return i_size.
117 return lseek_execute(file, inode, offset, maxsize);
119 EXPORT_SYMBOL(generic_file_llseek_size);
122 * generic_file_llseek - generic llseek implementation for regular files
123 * @file: file structure to seek on
124 * @offset: file offset to seek to
125 * @whence: type of seek
127 * This is a generic implemenation of ->llseek useable for all normal local
128 * filesystems. It just updates the file offset to the value specified by
129 * @offset and @whence under i_mutex.
131 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
133 struct inode *inode = file->f_mapping->host;
135 return generic_file_llseek_size(file, offset, whence,
136 inode->i_sb->s_maxbytes,
139 EXPORT_SYMBOL(generic_file_llseek);
142 * noop_llseek - No Operation Performed llseek implementation
143 * @file: file structure to seek on
144 * @offset: file offset to seek to
145 * @whence: type of seek
147 * This is an implementation of ->llseek useable for the rare special case when
148 * userspace expects the seek to succeed but the (device) file is actually not
149 * able to perform the seek. In this case you use noop_llseek() instead of
150 * falling back to the default implementation of ->llseek.
152 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
156 EXPORT_SYMBOL(noop_llseek);
158 loff_t no_llseek(struct file *file, loff_t offset, int whence)
162 EXPORT_SYMBOL(no_llseek);
164 loff_t default_llseek(struct file *file, loff_t offset, int whence)
166 struct inode *inode = file->f_path.dentry->d_inode;
169 mutex_lock(&inode->i_mutex);
172 offset += i_size_read(inode);
176 retval = file->f_pos;
179 offset += file->f_pos;
183 * In the generic case the entire file is data, so as
184 * long as offset isn't at the end of the file then the
187 if (offset >= inode->i_size) {
194 * There is a virtual hole at the end of the file, so
195 * as long as offset isn't i_size or larger, return
198 if (offset >= inode->i_size) {
202 offset = inode->i_size;
206 if (offset >= 0 || unsigned_offsets(file)) {
207 if (offset != file->f_pos) {
208 file->f_pos = offset;
214 mutex_unlock(&inode->i_mutex);
217 EXPORT_SYMBOL(default_llseek);
219 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
221 loff_t (*fn)(struct file *, loff_t, int);
224 if (file->f_mode & FMODE_LSEEK) {
225 if (file->f_op && file->f_op->llseek)
226 fn = file->f_op->llseek;
228 return fn(file, offset, whence);
230 EXPORT_SYMBOL(vfs_llseek);
232 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
235 struct fd f = fdget(fd);
240 if (whence <= SEEK_MAX) {
241 loff_t res = vfs_llseek(f.file, offset, whence);
243 if (res != (loff_t)retval)
244 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
250 #ifdef __ARCH_WANT_SYS_LLSEEK
251 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
252 unsigned long, offset_low, loff_t __user *, result,
253 unsigned int, whence)
256 struct fd f = fdget(fd);
263 if (whence > SEEK_MAX)
266 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
269 retval = (int)offset;
272 if (!copy_to_user(result, &offset, sizeof(offset)))
283 * rw_verify_area doesn't like huge counts. We limit
284 * them to something that fits in "int" so that others
285 * won't have to do range checks all the time.
287 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
291 int retval = -EINVAL;
293 inode = file->f_path.dentry->d_inode;
294 if (unlikely((ssize_t) count < 0))
297 if (unlikely(pos < 0)) {
298 if (!unsigned_offsets(file))
300 if (count >= -pos) /* both values are in 0..LLONG_MAX */
302 } else if (unlikely((loff_t) (pos + count) < 0)) {
303 if (!unsigned_offsets(file))
307 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
308 retval = locks_mandatory_area(
309 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
310 inode, file, pos, count);
314 retval = security_file_permission(file,
315 read_write == READ ? MAY_READ : MAY_WRITE);
318 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
321 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
323 set_current_state(TASK_UNINTERRUPTIBLE);
324 if (!kiocbIsKicked(iocb))
327 kiocbClearKicked(iocb);
328 __set_current_state(TASK_RUNNING);
331 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
333 struct iovec iov = { .iov_base = buf, .iov_len = len };
337 init_sync_kiocb(&kiocb, filp);
338 kiocb.ki_pos = *ppos;
340 kiocb.ki_nbytes = len;
343 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
344 if (ret != -EIOCBRETRY)
346 wait_on_retry_sync_kiocb(&kiocb);
349 if (-EIOCBQUEUED == ret)
350 ret = wait_on_sync_kiocb(&kiocb);
351 *ppos = kiocb.ki_pos;
355 EXPORT_SYMBOL(do_sync_read);
357 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
361 if (!(file->f_mode & FMODE_READ))
363 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
365 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
368 ret = rw_verify_area(READ, file, pos, count);
371 if (file->f_op->read)
372 ret = file->f_op->read(file, buf, count, pos);
374 ret = do_sync_read(file, buf, count, pos);
376 fsnotify_access(file);
377 add_rchar(current, ret);
385 EXPORT_SYMBOL(vfs_read);
387 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
389 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
393 init_sync_kiocb(&kiocb, filp);
394 kiocb.ki_pos = *ppos;
396 kiocb.ki_nbytes = len;
399 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
400 if (ret != -EIOCBRETRY)
402 wait_on_retry_sync_kiocb(&kiocb);
405 if (-EIOCBQUEUED == ret)
406 ret = wait_on_sync_kiocb(&kiocb);
407 *ppos = kiocb.ki_pos;
411 EXPORT_SYMBOL(do_sync_write);
413 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
417 if (!(file->f_mode & FMODE_WRITE))
419 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
421 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
424 ret = rw_verify_area(WRITE, file, pos, count);
427 if (file->f_op->write)
428 ret = file->f_op->write(file, buf, count, pos);
430 ret = do_sync_write(file, buf, count, pos);
432 fsnotify_modify(file);
433 add_wchar(current, ret);
441 EXPORT_SYMBOL(vfs_write);
443 static inline loff_t file_pos_read(struct file *file)
448 static inline void file_pos_write(struct file *file, loff_t pos)
453 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
455 struct fd f = fdget(fd);
456 ssize_t ret = -EBADF;
459 loff_t pos = file_pos_read(f.file);
460 ret = vfs_read(f.file, buf, count, &pos);
461 file_pos_write(f.file, pos);
467 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
470 struct fd f = fdget(fd);
471 ssize_t ret = -EBADF;
474 loff_t pos = file_pos_read(f.file);
475 ret = vfs_write(f.file, buf, count, &pos);
476 file_pos_write(f.file, pos);
483 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
484 size_t count, loff_t pos)
487 ssize_t ret = -EBADF;
495 if (f.file->f_mode & FMODE_PREAD)
496 ret = vfs_read(f.file, buf, count, &pos);
502 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
503 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
505 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
506 (size_t) count, pos);
508 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
511 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
512 size_t count, loff_t pos)
515 ssize_t ret = -EBADF;
523 if (f.file->f_mode & FMODE_PWRITE)
524 ret = vfs_write(f.file, buf, count, &pos);
530 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
531 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
533 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
534 (size_t) count, pos);
536 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
540 * Reduce an iovec's length in-place. Return the resulting number of segments
542 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
544 unsigned long seg = 0;
547 while (seg < nr_segs) {
549 if (len + iov->iov_len >= to) {
550 iov->iov_len = to - len;
558 EXPORT_SYMBOL(iov_shorten);
560 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
561 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
566 init_sync_kiocb(&kiocb, filp);
567 kiocb.ki_pos = *ppos;
569 kiocb.ki_nbytes = len;
572 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
573 if (ret != -EIOCBRETRY)
575 wait_on_retry_sync_kiocb(&kiocb);
578 if (ret == -EIOCBQUEUED)
579 ret = wait_on_sync_kiocb(&kiocb);
580 *ppos = kiocb.ki_pos;
584 /* Do it by hand, with file-ops */
585 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
586 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
588 struct iovec *vector = iov;
591 while (nr_segs > 0) {
596 base = vector->iov_base;
597 len = vector->iov_len;
601 nr = fn(filp, base, len, ppos);
616 /* A write operation does a read from user space and vice versa */
617 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
619 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
620 unsigned long nr_segs, unsigned long fast_segs,
621 struct iovec *fast_pointer,
622 struct iovec **ret_pointer)
626 struct iovec *iov = fast_pointer;
629 * SuS says "The readv() function *may* fail if the iovcnt argument
630 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
631 * traditionally returned zero for zero segments, so...
639 * First get the "struct iovec" from user memory and
640 * verify all the pointers
642 if (nr_segs > UIO_MAXIOV) {
646 if (nr_segs > fast_segs) {
647 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
653 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
659 * According to the Single Unix Specification we should return EINVAL
660 * if an element length is < 0 when cast to ssize_t or if the
661 * total length would overflow the ssize_t return value of the
664 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
668 for (seg = 0; seg < nr_segs; seg++) {
669 void __user *buf = iov[seg].iov_base;
670 ssize_t len = (ssize_t)iov[seg].iov_len;
672 /* see if we we're about to use an invalid len or if
673 * it's about to overflow ssize_t */
679 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
683 if (len > MAX_RW_COUNT - ret) {
684 len = MAX_RW_COUNT - ret;
685 iov[seg].iov_len = len;
694 static ssize_t do_readv_writev(int type, struct file *file,
695 const struct iovec __user * uvector,
696 unsigned long nr_segs, loff_t *pos)
699 struct iovec iovstack[UIO_FASTIOV];
700 struct iovec *iov = iovstack;
710 ret = rw_copy_check_uvector(type, uvector, nr_segs,
711 ARRAY_SIZE(iovstack), iovstack, &iov);
716 ret = rw_verify_area(type, file, pos, tot_len);
722 fn = file->f_op->read;
723 fnv = file->f_op->aio_read;
725 fn = (io_fn_t)file->f_op->write;
726 fnv = file->f_op->aio_write;
730 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
733 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
738 if ((ret + (type == READ)) > 0) {
740 fsnotify_access(file);
742 fsnotify_modify(file);
747 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
748 unsigned long vlen, loff_t *pos)
750 if (!(file->f_mode & FMODE_READ))
752 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
755 return do_readv_writev(READ, file, vec, vlen, pos);
758 EXPORT_SYMBOL(vfs_readv);
760 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
761 unsigned long vlen, loff_t *pos)
763 if (!(file->f_mode & FMODE_WRITE))
765 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
768 return do_readv_writev(WRITE, file, vec, vlen, pos);
771 EXPORT_SYMBOL(vfs_writev);
773 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
776 struct fd f = fdget(fd);
777 ssize_t ret = -EBADF;
780 loff_t pos = file_pos_read(f.file);
781 ret = vfs_readv(f.file, vec, vlen, &pos);
782 file_pos_write(f.file, pos);
787 add_rchar(current, ret);
792 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
795 struct fd f = fdget(fd);
796 ssize_t ret = -EBADF;
799 loff_t pos = file_pos_read(f.file);
800 ret = vfs_writev(f.file, vec, vlen, &pos);
801 file_pos_write(f.file, pos);
806 add_wchar(current, ret);
811 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
813 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
814 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
817 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
818 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
820 loff_t pos = pos_from_hilo(pos_h, pos_l);
822 ssize_t ret = -EBADF;
830 if (f.file->f_mode & FMODE_PREAD)
831 ret = vfs_readv(f.file, vec, vlen, &pos);
836 add_rchar(current, ret);
841 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
842 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
844 loff_t pos = pos_from_hilo(pos_h, pos_l);
846 ssize_t ret = -EBADF;
854 if (f.file->f_mode & FMODE_PWRITE)
855 ret = vfs_writev(f.file, vec, vlen, &pos);
860 add_wchar(current, ret);
865 ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos, size_t count,
869 struct inode *in_inode, *out_inode;
875 * Get input file, and verify that it is ok..
881 if (!(in.file->f_mode & FMODE_READ))
885 ppos = &in.file->f_pos;
887 if (!(in.file->f_mode & FMODE_PREAD))
889 retval = rw_verify_area(READ, in.file, ppos, count);
895 * Get output file, and verify that it is ok..
901 if (!(out.file->f_mode & FMODE_WRITE))
904 in_inode = in.file->f_path.dentry->d_inode;
905 out_inode = out.file->f_path.dentry->d_inode;
906 retval = rw_verify_area(WRITE, out.file, &out.file->f_pos, count);
912 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
915 if (unlikely(pos + count > max)) {
925 * We need to debate whether we can enable this or not. The
926 * man page documents EAGAIN return for the output at least,
927 * and the application is arguably buggy if it doesn't expect
928 * EAGAIN on a non-blocking file descriptor.
930 if (in.file->f_flags & O_NONBLOCK)
931 fl = SPLICE_F_NONBLOCK;
933 retval = do_splice_direct(in.file, ppos, out.file, count, fl);
936 add_rchar(current, retval);
937 add_wchar(current, retval);
953 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
960 if (unlikely(get_user(off, offset)))
963 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
964 if (unlikely(put_user(pos, offset)))
969 return do_sendfile(out_fd, in_fd, NULL, count, 0);
972 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
978 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
980 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
981 if (unlikely(put_user(pos, offset)))
986 return do_sendfile(out_fd, in_fd, NULL, count, 0);