~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/read_write.c

Version: ~ [ linux-5.10-rc6 ] ~ [ linux-5.9.12 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.81 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.161 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.210 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.247 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.247 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  *  linux/fs/read_write.c
  4  *
  5  *  Copyright (C) 1991, 1992  Linus Torvalds
  6  */
  7 
  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>
 22 #include <linux/fs.h>
 23 #include "internal.h"
 24 
 25 #include <linux/uaccess.h>
 26 #include <asm/unistd.h>
 27 
 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,
 33 };
 34 
 35 EXPORT_SYMBOL(generic_ro_fops);
 36 
 37 static inline bool unsigned_offsets(struct file *file)
 38 {
 39         return file->f_mode & FMODE_UNSIGNED_OFFSET;
 40 }
 41 
 42 /**
 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
 47  *
 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.
 51  *
 52  * Return the specified offset on success and -EINVAL on invalid offset.
 53  */
 54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
 55 {
 56         if (offset < 0 && !unsigned_offsets(file))
 57                 return -EINVAL;
 58         if (offset > maxsize)
 59                 return -EINVAL;
 60 
 61         if (offset != file->f_pos) {
 62                 file->f_pos = offset;
 63                 file->f_version = 0;
 64         }
 65         return offset;
 66 }
 67 EXPORT_SYMBOL(vfs_setpos);
 68 
 69 /**
 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
 76  *
 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
 79  *
 80  * Synchronization:
 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.
 84  */
 85 loff_t
 86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
 87                 loff_t maxsize, loff_t eof)
 88 {
 89         switch (whence) {
 90         case SEEK_END:
 91                 offset += eof;
 92                 break;
 93         case SEEK_CUR:
 94                 /*
 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
 99                  */
100                 if (offset == 0)
101                         return file->f_pos;
102                 /*
103                  * f_lock protects against read/modify/write race with other
104                  * SEEK_CURs. Note that parallel writes and reads behave
105                  * like SEEK_SET.
106                  */
107                 spin_lock(&file->f_lock);
108                 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109                 spin_unlock(&file->f_lock);
110                 return offset;
111         case SEEK_DATA:
112                 /*
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.
115                  */
116                 if ((unsigned long long)offset >= eof)
117                         return -ENXIO;
118                 break;
119         case SEEK_HOLE:
120                 /*
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.
123                  */
124                 if ((unsigned long long)offset >= eof)
125                         return -ENXIO;
126                 offset = eof;
127                 break;
128         }
129 
130         return vfs_setpos(file, offset, maxsize);
131 }
132 EXPORT_SYMBOL(generic_file_llseek_size);
133 
134 /**
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
139  *
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.
143  */
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
145 {
146         struct inode *inode = file->f_mapping->host;
147 
148         return generic_file_llseek_size(file, offset, whence,
149                                         inode->i_sb->s_maxbytes,
150                                         i_size_read(inode));
151 }
152 EXPORT_SYMBOL(generic_file_llseek);
153 
154 /**
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
160  *
161  */
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
163 {
164         switch (whence) {
165         case SEEK_SET: case SEEK_CUR: case SEEK_END:
166                 return generic_file_llseek_size(file, offset, whence,
167                                                 size, size);
168         default:
169                 return -EINVAL;
170         }
171 }
172 EXPORT_SYMBOL(fixed_size_llseek);
173 
174 /**
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
179  *
180  */
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
182 {
183         switch (whence) {
184         case SEEK_SET: case SEEK_CUR:
185                 return generic_file_llseek_size(file, offset, whence,
186                                                 OFFSET_MAX, 0);
187         default:
188                 return -EINVAL;
189         }
190 }
191 EXPORT_SYMBOL(no_seek_end_llseek);
192 
193 /**
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
199  *
200  */
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
202 {
203         switch (whence) {
204         case SEEK_SET: case SEEK_CUR:
205                 return generic_file_llseek_size(file, offset, whence,
206                                                 size, 0);
207         default:
208                 return -EINVAL;
209         }
210 }
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
212 
213 /**
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
218  *
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.
223  */
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
225 {
226         return file->f_pos;
227 }
228 EXPORT_SYMBOL(noop_llseek);
229 
230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
231 {
232         return -ESPIPE;
233 }
234 EXPORT_SYMBOL(no_llseek);
235 
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
237 {
238         struct inode *inode = file_inode(file);
239         loff_t retval;
240 
241         inode_lock(inode);
242         switch (whence) {
243                 case SEEK_END:
244                         offset += i_size_read(inode);
245                         break;
246                 case SEEK_CUR:
247                         if (offset == 0) {
248                                 retval = file->f_pos;
249                                 goto out;
250                         }
251                         offset += file->f_pos;
252                         break;
253                 case SEEK_DATA:
254                         /*
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
257                          * offset is data.
258                          */
259                         if (offset >= inode->i_size) {
260                                 retval = -ENXIO;
261                                 goto out;
262                         }
263                         break;
264                 case SEEK_HOLE:
265                         /*
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
268                          * i_size.
269                          */
270                         if (offset >= inode->i_size) {
271                                 retval = -ENXIO;
272                                 goto out;
273                         }
274                         offset = inode->i_size;
275                         break;
276         }
277         retval = -EINVAL;
278         if (offset >= 0 || unsigned_offsets(file)) {
279                 if (offset != file->f_pos) {
280                         file->f_pos = offset;
281                         file->f_version = 0;
282                 }
283                 retval = offset;
284         }
285 out:
286         inode_unlock(inode);
287         return retval;
288 }
289 EXPORT_SYMBOL(default_llseek);
290 
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
292 {
293         loff_t (*fn)(struct file *, loff_t, int);
294 
295         fn = no_llseek;
296         if (file->f_mode & FMODE_LSEEK) {
297                 if (file->f_op->llseek)
298                         fn = file->f_op->llseek;
299         }
300         return fn(file, offset, whence);
301 }
302 EXPORT_SYMBOL(vfs_llseek);
303 
304 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
305 {
306         off_t retval;
307         struct fd f = fdget_pos(fd);
308         if (!f.file)
309                 return -EBADF;
310 
311         retval = -EINVAL;
312         if (whence <= SEEK_MAX) {
313                 loff_t res = vfs_llseek(f.file, offset, whence);
314                 retval = res;
315                 if (res != (loff_t)retval)
316                         retval = -EOVERFLOW;    /* LFS: should only happen on 32 bit platforms */
317         }
318         fdput_pos(f);
319         return retval;
320 }
321 
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
323 {
324         return ksys_lseek(fd, offset, whence);
325 }
326 
327 #ifdef CONFIG_COMPAT
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
329 {
330         return ksys_lseek(fd, offset, whence);
331 }
332 #endif
333 
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)
339 {
340         int retval;
341         struct fd f = fdget_pos(fd);
342         loff_t offset;
343 
344         if (!f.file)
345                 return -EBADF;
346 
347         retval = -EINVAL;
348         if (whence > SEEK_MAX)
349                 goto out_putf;
350 
351         offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
352                         whence);
353 
354         retval = (int)offset;
355         if (offset >= 0) {
356                 retval = -EFAULT;
357                 if (!copy_to_user(result, &offset, sizeof(offset)))
358                         retval = 0;
359         }
360 out_putf:
361         fdput_pos(f);
362         return retval;
363 }
364 #endif
365 
366 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
367 {
368         struct inode *inode;
369         int retval = -EINVAL;
370 
371         inode = file_inode(file);
372         if (unlikely((ssize_t) count < 0))
373                 return retval;
374 
375         /*
376          * ranged mandatory locking does not apply to streams - it makes sense
377          * only for files where position has a meaning.
378          */
379         if (ppos) {
380                 loff_t pos = *ppos;
381 
382                 if (unlikely(pos < 0)) {
383                         if (!unsigned_offsets(file))
384                                 return retval;
385                         if (count >= -pos) /* both values are in 0..LLONG_MAX */
386                                 return -EOVERFLOW;
387                 } else if (unlikely((loff_t) (pos + count) < 0)) {
388                         if (!unsigned_offsets(file))
389                                 return retval;
390                 }
391 
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);
395                         if (retval < 0)
396                                 return retval;
397                 }
398         }
399 
400         return security_file_permission(file,
401                                 read_write == READ ? MAY_READ : MAY_WRITE);
402 }
403 
404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
405 {
406         struct iovec iov = { .iov_base = buf, .iov_len = len };
407         struct kiocb kiocb;
408         struct iov_iter iter;
409         ssize_t ret;
410 
411         init_sync_kiocb(&kiocb, filp);
412         kiocb.ki_pos = (ppos ? *ppos : 0);
413         iov_iter_init(&iter, READ, &iov, 1, len);
414 
415         ret = call_read_iter(filp, &kiocb, &iter);
416         BUG_ON(ret == -EIOCBQUEUED);
417         if (ppos)
418                 *ppos = kiocb.ki_pos;
419         return ret;
420 }
421 
422 static int warn_unsupported(struct file *file, const char *op)
423 {
424         pr_warn_ratelimited(
425                 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
426                 op, file, current->pid, current->comm);
427         return -EINVAL;
428 }
429 
430 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
431 {
432         struct kvec iov = {
433                 .iov_base       = buf,
434                 .iov_len        = min_t(size_t, count, MAX_RW_COUNT),
435         };
436         struct kiocb kiocb;
437         struct iov_iter iter;
438         ssize_t ret;
439 
440         if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
441                 return -EINVAL;
442         if (!(file->f_mode & FMODE_CAN_READ))
443                 return -EINVAL;
444         /*
445          * Also fail if ->read_iter and ->read are both wired up as that
446          * implies very convoluted semantics.
447          */
448         if (unlikely(!file->f_op->read_iter || file->f_op->read))
449                 return warn_unsupported(file, "read");
450 
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);
455         if (ret > 0) {
456                 if (pos)
457                         *pos = kiocb.ki_pos;
458                 fsnotify_access(file);
459                 add_rchar(current, ret);
460         }
461         inc_syscr(current);
462         return ret;
463 }
464 
465 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
466 {
467         ssize_t ret;
468 
469         ret = rw_verify_area(READ, file, pos, count);
470         if (ret)
471                 return ret;
472         return __kernel_read(file, buf, count, pos);
473 }
474 EXPORT_SYMBOL(kernel_read);
475 
476 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
477 {
478         ssize_t ret;
479 
480         if (!(file->f_mode & FMODE_READ))
481                 return -EBADF;
482         if (!(file->f_mode & FMODE_CAN_READ))
483                 return -EINVAL;
484         if (unlikely(!access_ok(buf, count)))
485                 return -EFAULT;
486 
487         ret = rw_verify_area(READ, file, pos, count);
488         if (ret)
489                 return ret;
490         if (count > MAX_RW_COUNT)
491                 count =  MAX_RW_COUNT;
492 
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);
497         else
498                 ret = -EINVAL;
499         if (ret > 0) {
500                 fsnotify_access(file);
501                 add_rchar(current, ret);
502         }
503         inc_syscr(current);
504         return ret;
505 }
506 
507 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
508 {
509         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
510         struct kiocb kiocb;
511         struct iov_iter iter;
512         ssize_t ret;
513 
514         init_sync_kiocb(&kiocb, filp);
515         kiocb.ki_pos = (ppos ? *ppos : 0);
516         iov_iter_init(&iter, WRITE, &iov, 1, len);
517 
518         ret = call_write_iter(filp, &kiocb, &iter);
519         BUG_ON(ret == -EIOCBQUEUED);
520         if (ret > 0 && ppos)
521                 *ppos = kiocb.ki_pos;
522         return ret;
523 }
524 
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)
527 {
528         struct kvec iov = {
529                 .iov_base       = (void *)buf,
530                 .iov_len        = min_t(size_t, count, MAX_RW_COUNT),
531         };
532         struct kiocb kiocb;
533         struct iov_iter iter;
534         ssize_t ret;
535 
536         if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
537                 return -EBADF;
538         if (!(file->f_mode & FMODE_CAN_WRITE))
539                 return -EINVAL;
540         /*
541          * Also fail if ->write_iter and ->write are both wired up as that
542          * implies very convoluted semantics.
543          */
544         if (unlikely(!file->f_op->write_iter || file->f_op->write))
545                 return warn_unsupported(file, "write");
546 
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);
551         if (ret > 0) {
552                 if (pos)
553                         *pos = kiocb.ki_pos;
554                 fsnotify_modify(file);
555                 add_wchar(current, ret);
556         }
557         inc_syscw(current);
558         return ret;
559 }
560 /*
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.
566  */
567 EXPORT_SYMBOL_GPL(__kernel_write);
568 
569 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
570                             loff_t *pos)
571 {
572         ssize_t ret;
573 
574         ret = rw_verify_area(WRITE, file, pos, count);
575         if (ret)
576                 return ret;
577 
578         file_start_write(file);
579         ret =  __kernel_write(file, buf, count, pos);
580         file_end_write(file);
581         return ret;
582 }
583 EXPORT_SYMBOL(kernel_write);
584 
585 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
586 {
587         ssize_t ret;
588 
589         if (!(file->f_mode & FMODE_WRITE))
590                 return -EBADF;
591         if (!(file->f_mode & FMODE_CAN_WRITE))
592                 return -EINVAL;
593         if (unlikely(!access_ok(buf, count)))
594                 return -EFAULT;
595 
596         ret = rw_verify_area(WRITE, file, pos, count);
597         if (ret)
598                 return ret;
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);
606         else
607                 ret = -EINVAL;
608         if (ret > 0) {
609                 fsnotify_modify(file);
610                 add_wchar(current, ret);
611         }
612         inc_syscw(current);
613         file_end_write(file);
614         return ret;
615 }
616 
617 /* file_ppos returns &file->f_pos or NULL if file is stream */
618 static inline loff_t *file_ppos(struct file *file)
619 {
620         return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
621 }
622 
623 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
624 {
625         struct fd f = fdget_pos(fd);
626         ssize_t ret = -EBADF;
627 
628         if (f.file) {
629                 loff_t pos, *ppos = file_ppos(f.file);
630                 if (ppos) {
631                         pos = *ppos;
632                         ppos = &pos;
633                 }
634                 ret = vfs_read(f.file, buf, count, ppos);
635                 if (ret >= 0 && ppos)
636                         f.file->f_pos = pos;
637                 fdput_pos(f);
638         }
639         return ret;
640 }
641 
642 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
643 {
644         return ksys_read(fd, buf, count);
645 }
646 
647 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
648 {
649         struct fd f = fdget_pos(fd);
650         ssize_t ret = -EBADF;
651 
652         if (f.file) {
653                 loff_t pos, *ppos = file_ppos(f.file);
654                 if (ppos) {
655                         pos = *ppos;
656                         ppos = &pos;
657                 }
658                 ret = vfs_write(f.file, buf, count, ppos);
659                 if (ret >= 0 && ppos)
660                         f.file->f_pos = pos;
661                 fdput_pos(f);
662         }
663 
664         return ret;
665 }
666 
667 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
668                 size_t, count)
669 {
670         return ksys_write(fd, buf, count);
671 }
672 
673 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
674                      loff_t pos)
675 {
676         struct fd f;
677         ssize_t ret = -EBADF;
678 
679         if (pos < 0)
680                 return -EINVAL;
681 
682         f = fdget(fd);
683         if (f.file) {
684                 ret = -ESPIPE;
685                 if (f.file->f_mode & FMODE_PREAD)
686                         ret = vfs_read(f.file, buf, count, &pos);
687                 fdput(f);
688         }
689 
690         return ret;
691 }
692 
693 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
694                         size_t, count, loff_t, pos)
695 {
696         return ksys_pread64(fd, buf, count, pos);
697 }
698 
699 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
700                       size_t count, loff_t pos)
701 {
702         struct fd f;
703         ssize_t ret = -EBADF;
704 
705         if (pos < 0)
706                 return -EINVAL;
707 
708         f = fdget(fd);
709         if (f.file) {
710                 ret = -ESPIPE;
711                 if (f.file->f_mode & FMODE_PWRITE)  
712                         ret = vfs_write(f.file, buf, count, &pos);
713                 fdput(f);
714         }
715 
716         return ret;
717 }
718 
719 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
720                          size_t, count, loff_t, pos)
721 {
722         return ksys_pwrite64(fd, buf, count, pos);
723 }
724 
725 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
726                 loff_t *ppos, int type, rwf_t flags)
727 {
728         struct kiocb kiocb;
729         ssize_t ret;
730 
731         init_sync_kiocb(&kiocb, filp);
732         ret = kiocb_set_rw_flags(&kiocb, flags);
733         if (ret)
734                 return ret;
735         kiocb.ki_pos = (ppos ? *ppos : 0);
736 
737         if (type == READ)
738                 ret = call_read_iter(filp, &kiocb, iter);
739         else
740                 ret = call_write_iter(filp, &kiocb, iter);
741         BUG_ON(ret == -EIOCBQUEUED);
742         if (ppos)
743                 *ppos = kiocb.ki_pos;
744         return ret;
745 }
746 
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)
750 {
751         ssize_t ret = 0;
752 
753         if (flags & ~RWF_HIPRI)
754                 return -EOPNOTSUPP;
755 
756         while (iov_iter_count(iter)) {
757                 struct iovec iovec = iov_iter_iovec(iter);
758                 ssize_t nr;
759 
760                 if (type == READ) {
761                         nr = filp->f_op->read(filp, iovec.iov_base,
762                                               iovec.iov_len, ppos);
763                 } else {
764                         nr = filp->f_op->write(filp, iovec.iov_base,
765                                                iovec.iov_len, ppos);
766                 }
767 
768                 if (nr < 0) {
769                         if (!ret)
770                                 ret = nr;
771                         break;
772                 }
773                 ret += nr;
774                 if (nr != iovec.iov_len)
775                         break;
776                 iov_iter_advance(iter, nr);
777         }
778 
779         return ret;
780 }
781 
782 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
783                 loff_t *pos, rwf_t flags)
784 {
785         size_t tot_len;
786         ssize_t ret = 0;
787 
788         if (!(file->f_mode & FMODE_READ))
789                 return -EBADF;
790         if (!(file->f_mode & FMODE_CAN_READ))
791                 return -EINVAL;
792 
793         tot_len = iov_iter_count(iter);
794         if (!tot_len)
795                 goto out;
796         ret = rw_verify_area(READ, file, pos, tot_len);
797         if (ret < 0)
798                 return ret;
799 
800         if (file->f_op->read_iter)
801                 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
802         else
803                 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
804 out:
805         if (ret >= 0)
806                 fsnotify_access(file);
807         return ret;
808 }
809 
810 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
811                            struct iov_iter *iter)
812 {
813         size_t tot_len;
814         ssize_t ret = 0;
815 
816         if (!file->f_op->read_iter)
817                 return -EINVAL;
818         if (!(file->f_mode & FMODE_READ))
819                 return -EBADF;
820         if (!(file->f_mode & FMODE_CAN_READ))
821                 return -EINVAL;
822 
823         tot_len = iov_iter_count(iter);
824         if (!tot_len)
825                 goto out;
826         ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
827         if (ret < 0)
828                 return ret;
829 
830         ret = call_read_iter(file, iocb, iter);
831 out:
832         if (ret >= 0)
833                 fsnotify_access(file);
834         return ret;
835 }
836 EXPORT_SYMBOL(vfs_iocb_iter_read);
837 
838 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
839                 rwf_t flags)
840 {
841         if (!file->f_op->read_iter)
842                 return -EINVAL;
843         return do_iter_read(file, iter, ppos, flags);
844 }
845 EXPORT_SYMBOL(vfs_iter_read);
846 
847 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
848                 loff_t *pos, rwf_t flags)
849 {
850         size_t tot_len;
851         ssize_t ret = 0;
852 
853         if (!(file->f_mode & FMODE_WRITE))
854                 return -EBADF;
855         if (!(file->f_mode & FMODE_CAN_WRITE))
856                 return -EINVAL;
857 
858         tot_len = iov_iter_count(iter);
859         if (!tot_len)
860                 return 0;
861         ret = rw_verify_area(WRITE, file, pos, tot_len);
862         if (ret < 0)
863                 return ret;
864 
865         if (file->f_op->write_iter)
866                 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
867         else
868                 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
869         if (ret > 0)
870                 fsnotify_modify(file);
871         return ret;
872 }
873 
874 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
875                             struct iov_iter *iter)
876 {
877         size_t tot_len;
878         ssize_t ret = 0;
879 
880         if (!file->f_op->write_iter)
881                 return -EINVAL;
882         if (!(file->f_mode & FMODE_WRITE))
883                 return -EBADF;
884         if (!(file->f_mode & FMODE_CAN_WRITE))
885                 return -EINVAL;
886 
887         tot_len = iov_iter_count(iter);
888         if (!tot_len)
889                 return 0;
890         ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
891         if (ret < 0)
892                 return ret;
893 
894         ret = call_write_iter(file, iocb, iter);
895         if (ret > 0)
896                 fsnotify_modify(file);
897 
898         return ret;
899 }
900 EXPORT_SYMBOL(vfs_iocb_iter_write);
901 
902 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
903                 rwf_t flags)
904 {
905         if (!file->f_op->write_iter)
906                 return -EINVAL;
907         return do_iter_write(file, iter, ppos, flags);
908 }
909 EXPORT_SYMBOL(vfs_iter_write);
910 
911 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
912                   unsigned long vlen, loff_t *pos, rwf_t flags)
913 {
914         struct iovec iovstack[UIO_FASTIOV];
915         struct iovec *iov = iovstack;
916         struct iov_iter iter;
917         ssize_t ret;
918 
919         ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
920         if (ret >= 0) {
921                 ret = do_iter_read(file, &iter, pos, flags);
922                 kfree(iov);
923         }
924 
925         return ret;
926 }
927 
928 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
929                    unsigned long vlen, loff_t *pos, rwf_t flags)
930 {
931         struct iovec iovstack[UIO_FASTIOV];
932         struct iovec *iov = iovstack;
933         struct iov_iter iter;
934         ssize_t ret;
935 
936         ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
937         if (ret >= 0) {
938                 file_start_write(file);
939                 ret = do_iter_write(file, &iter, pos, flags);
940                 file_end_write(file);
941                 kfree(iov);
942         }
943         return ret;
944 }
945 
946 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
947                         unsigned long vlen, rwf_t flags)
948 {
949         struct fd f = fdget_pos(fd);
950         ssize_t ret = -EBADF;
951 
952         if (f.file) {
953                 loff_t pos, *ppos = file_ppos(f.file);
954                 if (ppos) {
955                         pos = *ppos;
956                         ppos = &pos;
957                 }
958                 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
959                 if (ret >= 0 && ppos)
960                         f.file->f_pos = pos;
961                 fdput_pos(f);
962         }
963 
964         if (ret > 0)
965                 add_rchar(current, ret);
966         inc_syscr(current);
967         return ret;
968 }
969 
970 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
971                          unsigned long vlen, rwf_t flags)
972 {
973         struct fd f = fdget_pos(fd);
974         ssize_t ret = -EBADF;
975 
976         if (f.file) {
977                 loff_t pos, *ppos = file_ppos(f.file);
978                 if (ppos) {
979                         pos = *ppos;
980                         ppos = &pos;
981                 }
982                 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
983                 if (ret >= 0 && ppos)
984                         f.file->f_pos = pos;
985                 fdput_pos(f);
986         }
987 
988         if (ret > 0)
989                 add_wchar(current, ret);
990         inc_syscw(current);
991         return ret;
992 }
993 
994 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
995 {
996 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
997         return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
998 }
999 
1000 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1001                          unsigned long vlen, loff_t pos, rwf_t flags)
1002 {
1003         struct fd f;
1004         ssize_t ret = -EBADF;
1005 
1006         if (pos < 0)
1007                 return -EINVAL;
1008 
1009         f = fdget(fd);
1010         if (f.file) {
1011                 ret = -ESPIPE;
1012                 if (f.file->f_mode & FMODE_PREAD)
1013                         ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1014                 fdput(f);
1015         }
1016 
1017         if (ret > 0)
1018                 add_rchar(current, ret);
1019         inc_syscr(current);
1020         return ret;
1021 }
1022 
1023 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1024                           unsigned long vlen, loff_t pos, rwf_t flags)
1025 {
1026         struct fd f;
1027         ssize_t ret = -EBADF;
1028 
1029         if (pos < 0)
1030                 return -EINVAL;
1031 
1032         f = fdget(fd);
1033         if (f.file) {
1034                 ret = -ESPIPE;
1035                 if (f.file->f_mode & FMODE_PWRITE)
1036                         ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1037                 fdput(f);
1038         }
1039 
1040         if (ret > 0)
1041                 add_wchar(current, ret);
1042         inc_syscw(current);
1043         return ret;
1044 }
1045 
1046 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1047                 unsigned long, vlen)
1048 {
1049         return do_readv(fd, vec, vlen, 0);
1050 }
1051 
1052 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1053                 unsigned long, vlen)
1054 {
1055         return do_writev(fd, vec, vlen, 0);
1056 }
1057 
1058 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1059                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1060 {
1061         loff_t pos = pos_from_hilo(pos_h, pos_l);
1062 
1063         return do_preadv(fd, vec, vlen, pos, 0);
1064 }
1065 
1066 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1067                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1068                 rwf_t, flags)
1069 {
1070         loff_t pos = pos_from_hilo(pos_h, pos_l);
1071 
1072         if (pos == -1)
1073                 return do_readv(fd, vec, vlen, flags);
1074 
1075         return do_preadv(fd, vec, vlen, pos, flags);
1076 }
1077 
1078 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1079                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1080 {
1081         loff_t pos = pos_from_hilo(pos_h, pos_l);
1082 
1083         return do_pwritev(fd, vec, vlen, pos, 0);
1084 }
1085 
1086 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1087                 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1088                 rwf_t, flags)
1089 {
1090         loff_t pos = pos_from_hilo(pos_h, pos_l);
1091 
1092         if (pos == -1)
1093                 return do_writev(fd, vec, vlen, flags);
1094 
1095         return do_pwritev(fd, vec, vlen, pos, flags);
1096 }
1097 
1098 /*
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().
1102  */
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)
1108 {
1109         return do_preadv(fd, vec, vlen, pos, 0);
1110 }
1111 #endif
1112 
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)
1116 {
1117         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1118 
1119         return do_preadv(fd, vec, vlen, pos, 0);
1120 }
1121 
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)
1126 {
1127         if (pos == -1)
1128                 return do_readv(fd, vec, vlen, flags);
1129         return do_preadv(fd, vec, vlen, pos, flags);
1130 }
1131 #endif
1132 
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,
1136                 rwf_t, flags)
1137 {
1138         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1139 
1140         if (pos == -1)
1141                 return do_readv(fd, vec, vlen, flags);
1142         return do_preadv(fd, vec, vlen, pos, flags);
1143 }
1144 
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)
1149 {
1150         return do_pwritev(fd, vec, vlen, pos, 0);
1151 }
1152 #endif
1153 
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)
1157 {
1158         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1159 
1160         return do_pwritev(fd, vec, vlen, pos, 0);
1161 }
1162 
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)
1167 {
1168         if (pos == -1)
1169                 return do_writev(fd, vec, vlen, flags);
1170         return do_pwritev(fd, vec, vlen, pos, flags);
1171 }
1172 #endif
1173 
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)
1177 {
1178         loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1179 
1180         if (pos == -1)
1181                 return do_writev(fd, vec, vlen, flags);
1182         return do_pwritev(fd, vec, vlen, pos, flags);
1183 }
1184 #endif /* CONFIG_COMPAT */
1185 
1186 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1187                            size_t count, loff_t max)
1188 {
1189         struct fd in, out;
1190         struct inode *in_inode, *out_inode;
1191         loff_t pos;
1192         loff_t out_pos;
1193         ssize_t retval;
1194         int fl;
1195 
1196         /*
1197          * Get input file, and verify that it is ok..
1198          */
1199         retval = -EBADF;
1200         in = fdget(in_fd);
1201         if (!in.file)
1202                 goto out;
1203         if (!(in.file->f_mode & FMODE_READ))
1204                 goto fput_in;
1205         retval = -ESPIPE;
1206         if (!ppos) {
1207                 pos = in.file->f_pos;
1208         } else {
1209                 pos = *ppos;
1210                 if (!(in.file->f_mode & FMODE_PREAD))
1211                         goto fput_in;
1212         }
1213         retval = rw_verify_area(READ, in.file, &pos, count);
1214         if (retval < 0)
1215                 goto fput_in;
1216         if (count > MAX_RW_COUNT)
1217                 count =  MAX_RW_COUNT;
1218 
1219         /*
1220          * Get output file, and verify that it is ok..
1221          */
1222         retval = -EBADF;
1223         out = fdget(out_fd);
1224         if (!out.file)
1225                 goto fput_in;
1226         if (!(out.file->f_mode & FMODE_WRITE))
1227                 goto fput_out;
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);
1232         if (retval < 0)
1233                 goto fput_out;
1234 
1235         if (!max)
1236                 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1237 
1238         if (unlikely(pos + count > max)) {
1239                 retval = -EOVERFLOW;
1240                 if (pos >= max)
1241                         goto fput_out;
1242                 count = max - pos;
1243         }
1244 
1245         fl = 0;
1246 #if 0
1247         /*
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.
1252          */
1253         if (in.file->f_flags & O_NONBLOCK)
1254                 fl = SPLICE_F_NONBLOCK;
1255 #endif
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);
1259 
1260         if (retval > 0) {
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;
1266                 if (ppos)
1267                         *ppos = pos;
1268                 else
1269                         in.file->f_pos = pos;
1270         }
1271 
1272         inc_syscr(current);
1273         inc_syscw(current);
1274         if (pos > max)
1275                 retval = -EOVERFLOW;
1276 
1277 fput_out:
1278         fdput(out);
1279 fput_in:
1280         fdput(in);
1281 out:
1282         return retval;
1283 }
1284 
1285 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1286 {
1287         loff_t pos;
1288         off_t off;
1289         ssize_t ret;
1290 
1291         if (offset) {
1292                 if (unlikely(get_user(off, offset)))
1293                         return -EFAULT;
1294                 pos = off;
1295                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1296                 if (unlikely(put_user(pos, offset)))
1297                         return -EFAULT;
1298                 return ret;
1299         }
1300 
1301         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1302 }
1303 
1304 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1305 {
1306         loff_t pos;
1307         ssize_t ret;
1308 
1309         if (offset) {
1310                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1311                         return -EFAULT;
1312                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1313                 if (unlikely(put_user(pos, offset)))
1314                         return -EFAULT;
1315                 return ret;
1316         }
1317 
1318         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1319 }
1320 
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)
1324 {
1325         loff_t pos;
1326         off_t off;
1327         ssize_t ret;
1328 
1329         if (offset) {
1330                 if (unlikely(get_user(off, offset)))
1331                         return -EFAULT;
1332                 pos = off;
1333                 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1334                 if (unlikely(put_user(pos, offset)))
1335                         return -EFAULT;
1336                 return ret;
1337         }
1338 
1339         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1340 }
1341 
1342 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1343                 compat_loff_t __user *, offset, compat_size_t, count)
1344 {
1345         loff_t pos;
1346         ssize_t ret;
1347 
1348         if (offset) {
1349                 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1350                         return -EFAULT;
1351                 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1352                 if (unlikely(put_user(pos, offset)))
1353                         return -EFAULT;
1354                 return ret;
1355         }
1356 
1357         return do_sendfile(out_fd, in_fd, NULL, count, 0);
1358 }
1359 #endif
1360 
1361 /**
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
1369  *
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.
1374  *
1375  * This should be called from the @file_out filesystem, as per the
1376  * ->copy_file_range() method.
1377  *
1378  * Returns the number of bytes copied or a negative error indicating the
1379  * failure.
1380  */
1381 
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)
1385 {
1386         return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1387                                 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1388 }
1389 EXPORT_SYMBOL(generic_copy_file_range);
1390 
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)
1394 {
1395         /*
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.
1402          */
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,
1406                                                        file_out, pos_out,
1407                                                        len, flags);
1408 
1409         return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1410                                        flags);
1411 }
1412 
1413 /*
1414  * Performs necessary checks before doing a file copy
1415  *
1416  * Can adjust amount of bytes to copy via @req_count argument.
1417  * Returns appropriate error code that caller should return or
1418  * zero in case the copy should be allowed.
1419  */
1420 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1421                                     struct file *file_out, loff_t pos_out,
1422                                     size_t *req_count, unsigned int flags)
1423 {
1424         struct inode *inode_in = file_inode(file_in);
1425         struct inode *inode_out = file_inode(file_out);
1426         uint64_t count = *req_count;
1427         loff_t size_in;
1428         int ret;
1429 
1430         ret = generic_file_rw_checks(file_in, file_out);
1431         if (ret)
1432                 return ret;
1433 
1434         /* Don't touch certain kinds of inodes */
1435         if (IS_IMMUTABLE(inode_out))
1436                 return -EPERM;
1437 
1438         if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1439                 return -ETXTBSY;
1440 
1441         /* Ensure offsets don't wrap. */
1442         if (pos_in + count < pos_in || pos_out + count < pos_out)
1443                 return -EOVERFLOW;
1444 
1445         /* Shorten the copy to EOF */
1446         size_in = i_size_read(inode_in);
1447         if (pos_in >= size_in)
1448                 count = 0;
1449         else
1450                 count = min(count, size_in - (uint64_t)pos_in);
1451 
1452         ret = generic_write_check_limits(file_out, pos_out, &count);
1453         if (ret)
1454                 return ret;
1455 
1456         /* Don't allow overlapped copying within the same file. */
1457         if (inode_in == inode_out &&
1458             pos_out + count > pos_in &&
1459             pos_out < pos_in + count)
1460                 return -EINVAL;
1461 
1462         *req_count = count;
1463         return 0;
1464 }
1465 
1466 /*
1467  * copy_file_range() differs from regular file read and write in that it
1468  * specifically allows return partial success.  When it does so is up to
1469  * the copy_file_range method.
1470  */
1471 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1472                             struct file *file_out, loff_t pos_out,
1473                             size_t len, unsigned int flags)
1474 {
1475         ssize_t ret;
1476 
1477         if (flags != 0)
1478                 return -EINVAL;
1479 
1480         ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1481                                        flags);
1482         if (unlikely(ret))
1483                 return ret;
1484 
1485         ret = rw_verify_area(READ, file_in, &pos_in, len);
1486         if (unlikely(ret))
1487                 return ret;
1488 
1489         ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1490         if (unlikely(ret))
1491                 return ret;
1492 
1493         if (len == 0)
1494                 return 0;
1495 
1496         file_start_write(file_out);
1497 
1498         /*
1499          * Try cloning first, this is supported by more file systems, and
1500          * more efficient if both clone and copy are supported (e.g. NFS).
1501          */
1502         if (file_in->f_op->remap_file_range &&
1503             file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1504                 loff_t cloned;
1505 
1506                 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1507                                 file_out, pos_out,
1508                                 min_t(loff_t, MAX_RW_COUNT, len),
1509                                 REMAP_FILE_CAN_SHORTEN);
1510                 if (cloned > 0) {
1511                         ret = cloned;
1512                         goto done;
1513                 }
1514         }
1515 
1516         ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1517                                 flags);
1518         WARN_ON_ONCE(ret == -EOPNOTSUPP);
1519 done:
1520         if (ret > 0) {
1521                 fsnotify_access(file_in);
1522                 add_rchar(current, ret);
1523                 fsnotify_modify(file_out);
1524                 add_wchar(current, ret);
1525         }
1526 
1527         inc_syscr(current);
1528         inc_syscw(current);
1529 
1530         file_end_write(file_out);
1531 
1532         return ret;
1533 }
1534 EXPORT_SYMBOL(vfs_copy_file_range);
1535 
1536 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1537                 int, fd_out, loff_t __user *, off_out,
1538                 size_t, len, unsigned int, flags)
1539 {
1540         loff_t pos_in;
1541         loff_t pos_out;
1542         struct fd f_in;
1543         struct fd f_out;
1544         ssize_t ret = -EBADF;
1545 
1546         f_in = fdget(fd_in);
1547         if (!f_in.file)
1548                 goto out2;
1549 
1550         f_out = fdget(fd_out);
1551         if (!f_out.file)
1552                 goto out1;
1553 
1554         ret = -EFAULT;
1555         if (off_in) {
1556                 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1557                         goto out;
1558         } else {
1559                 pos_in = f_in.file->f_pos;
1560         }
1561 
1562         if (off_out) {
1563                 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1564                         goto out;
1565         } else {
1566                 pos_out = f_out.file->f_pos;
1567         }
1568 
1569         ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1570                                   flags);
1571         if (ret > 0) {
1572                 pos_in += ret;
1573                 pos_out += ret;
1574 
1575                 if (off_in) {
1576                         if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1577                                 ret = -EFAULT;
1578                 } else {
1579                         f_in.file->f_pos = pos_in;
1580                 }
1581 
1582                 if (off_out) {
1583                         if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1584                                 ret = -EFAULT;
1585                 } else {
1586                         f_out.file->f_pos = pos_out;
1587                 }
1588         }
1589 
1590 out:
1591         fdput(f_out);
1592 out1:
1593         fdput(f_in);
1594 out2:
1595         return ret;
1596 }
1597 
1598 /*
1599  * Don't operate on ranges the page cache doesn't support, and don't exceed the
1600  * LFS limits.  If pos is under the limit it becomes a short access.  If it
1601  * exceeds the limit we return -EFBIG.
1602  */
1603 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1604 {
1605         struct inode *inode = file->f_mapping->host;
1606         loff_t max_size = inode->i_sb->s_maxbytes;
1607         loff_t limit = rlimit(RLIMIT_FSIZE);
1608 
1609         if (limit != RLIM_INFINITY) {
1610                 if (pos >= limit) {
1611                         send_sig(SIGXFSZ, current, 0);
1612                         return -EFBIG;
1613                 }
1614                 *count = min(*count, limit - pos);
1615         }
1616 
1617         if (!(file->f_flags & O_LARGEFILE))
1618                 max_size = MAX_NON_LFS;
1619 
1620         if (unlikely(pos >= max_size))
1621                 return -EFBIG;
1622 
1623         *count = min(*count, max_size - pos);
1624 
1625         return 0;
1626 }
1627 
1628 /*
1629  * Performs necessary checks before doing a write
1630  *
1631  * Can adjust writing position or amount of bytes to write.
1632  * Returns appropriate error code that caller should return or
1633  * zero in case that write should be allowed.
1634  */
1635 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1636 {
1637         struct file *file = iocb->ki_filp;
1638         struct inode *inode = file->f_mapping->host;
1639         loff_t count;
1640         int ret;
1641 
1642         if (IS_SWAPFILE(inode))
1643                 return -ETXTBSY;
1644 
1645         if (!iov_iter_count(from))
1646                 return 0;
1647 
1648         /* FIXME: this is for backwards compatibility with 2.4 */
1649         if (iocb->ki_flags & IOCB_APPEND)
1650                 iocb->ki_pos = i_size_read(inode);
1651 
1652         if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
1653                 return -EINVAL;
1654 
1655         count = iov_iter_count(from);
1656         ret = generic_write_check_limits(file, iocb->ki_pos, &count);
1657         if (ret)
1658                 return ret;
1659 
1660         iov_iter_truncate(from, count);
1661         return iov_iter_count(from);
1662 }
1663 EXPORT_SYMBOL(generic_write_checks);
1664 
1665 /*
1666  * Performs common checks before doing a file copy/clone
1667  * from @file_in to @file_out.
1668  */
1669 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1670 {
1671         struct inode *inode_in = file_inode(file_in);
1672         struct inode *inode_out = file_inode(file_out);
1673 
1674         /* Don't copy dirs, pipes, sockets... */
1675         if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1676                 return -EISDIR;
1677         if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1678                 return -EINVAL;
1679 
1680         if (!(file_in->f_mode & FMODE_READ) ||
1681             !(file_out->f_mode & FMODE_WRITE) ||
1682             (file_out->f_flags & O_APPEND))
1683                 return -EBADF;
1684 
1685         return 0;
1686 }
1687 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp