generic_write_checks(): drop isblk argument
[firefly-linux-kernel-4.4.55.git] / fs / ext4 / file.c
1 /*
2  *  linux/fs/ext4/file.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/file.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  ext4 fs regular file handling primitives
16  *
17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
18  *      (jj@sunsite.ms.mff.cuni.cz)
19  */
20
21 #include <linux/time.h>
22 #include <linux/fs.h>
23 #include <linux/jbd2.h>
24 #include <linux/mount.h>
25 #include <linux/path.h>
26 #include <linux/quotaops.h>
27 #include <linux/pagevec.h>
28 #include <linux/uio.h>
29 #include "ext4.h"
30 #include "ext4_jbd2.h"
31 #include "xattr.h"
32 #include "acl.h"
33
34 /*
35  * Called when an inode is released. Note that this is different
36  * from ext4_file_open: open gets called at every open, but release
37  * gets called only when /all/ the files are closed.
38  */
39 static int ext4_release_file(struct inode *inode, struct file *filp)
40 {
41         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
42                 ext4_alloc_da_blocks(inode);
43                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
44         }
45         /* if we are the last writer on the inode, drop the block reservation */
46         if ((filp->f_mode & FMODE_WRITE) &&
47                         (atomic_read(&inode->i_writecount) == 1) &&
48                         !EXT4_I(inode)->i_reserved_data_blocks)
49         {
50                 down_write(&EXT4_I(inode)->i_data_sem);
51                 ext4_discard_preallocations(inode);
52                 up_write(&EXT4_I(inode)->i_data_sem);
53         }
54         if (is_dx(inode) && filp->private_data)
55                 ext4_htree_free_dir_info(filp->private_data);
56
57         return 0;
58 }
59
60 static void ext4_unwritten_wait(struct inode *inode)
61 {
62         wait_queue_head_t *wq = ext4_ioend_wq(inode);
63
64         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
65 }
66
67 /*
68  * This tests whether the IO in question is block-aligned or not.
69  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70  * are converted to written only after the IO is complete.  Until they are
71  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72  * it needs to zero out portions of the start and/or end block.  If 2 AIO
73  * threads are at work on the same unwritten block, they must be synchronized
74  * or one thread will zero the other's data, causing corruption.
75  */
76 static int
77 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
78 {
79         struct super_block *sb = inode->i_sb;
80         int blockmask = sb->s_blocksize - 1;
81
82         if (pos >= i_size_read(inode))
83                 return 0;
84
85         if ((pos | iov_iter_alignment(from)) & blockmask)
86                 return 1;
87
88         return 0;
89 }
90
91 static ssize_t
92 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
93 {
94         struct file *file = iocb->ki_filp;
95         struct inode *inode = file_inode(iocb->ki_filp);
96         struct mutex *aio_mutex = NULL;
97         struct blk_plug plug;
98         int o_direct = io_is_direct(file);
99         int overwrite = 0;
100         size_t length = iov_iter_count(from);
101         ssize_t ret;
102         loff_t pos = iocb->ki_pos;
103
104         /*
105          * Unaligned direct AIO must be serialized; see comment above
106          * In the case of O_APPEND, assume that we must always serialize
107          */
108         if (o_direct &&
109             ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
110             !is_sync_kiocb(iocb) &&
111             (file->f_flags & O_APPEND ||
112              ext4_unaligned_aio(inode, from, pos))) {
113                 aio_mutex = ext4_aio_mutex(inode);
114                 mutex_lock(aio_mutex);
115                 ext4_unwritten_wait(inode);
116         }
117
118         mutex_lock(&inode->i_mutex);
119         if (file->f_flags & O_APPEND)
120                 iocb->ki_pos = pos = i_size_read(inode);
121
122         /*
123          * If we have encountered a bitmap-format file, the size limit
124          * is smaller than s_maxbytes, which is for extent-mapped files.
125          */
126         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
127                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
128
129                 if ((pos > sbi->s_bitmap_maxbytes) ||
130                     (pos == sbi->s_bitmap_maxbytes && length > 0)) {
131                         mutex_unlock(&inode->i_mutex);
132                         ret = -EFBIG;
133                         goto errout;
134                 }
135                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
136                 length = iov_iter_count(from);
137         }
138
139         iocb->private = &overwrite;
140         if (o_direct) {
141                 blk_start_plug(&plug);
142
143
144                 /* check whether we do a DIO overwrite or not */
145                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
146                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
147                         struct ext4_map_blocks map;
148                         unsigned int blkbits = inode->i_blkbits;
149                         int err, len;
150
151                         map.m_lblk = pos >> blkbits;
152                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
153                                 - map.m_lblk;
154                         len = map.m_len;
155
156                         err = ext4_map_blocks(NULL, inode, &map, 0);
157                         /*
158                          * 'err==len' means that all of blocks has
159                          * been preallocated no matter they are
160                          * initialized or not.  For excluding
161                          * unwritten extents, we need to check
162                          * m_flags.  There are two conditions that
163                          * indicate for initialized extents.  1) If we
164                          * hit extent cache, EXT4_MAP_MAPPED flag is
165                          * returned; 2) If we do a real lookup,
166                          * non-flags are returned.  So we should check
167                          * these two conditions.
168                          */
169                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
170                                 overwrite = 1;
171                 }
172         }
173
174         ret = generic_write_checks(file, &iocb->ki_pos, &length);
175         if (ret)
176                 goto out;
177
178         if (length == 0)
179                 goto out;
180
181         iov_iter_truncate(from, length);
182         ret = __generic_file_write_iter(iocb, from);
183 out:
184         mutex_unlock(&inode->i_mutex);
185
186         if (ret > 0) {
187                 ssize_t err;
188
189                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
190                 if (err < 0)
191                         ret = err;
192         }
193         if (o_direct)
194                 blk_finish_plug(&plug);
195
196 errout:
197         if (aio_mutex)
198                 mutex_unlock(aio_mutex);
199         return ret;
200 }
201
202 #ifdef CONFIG_FS_DAX
203 static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
204 {
205         return dax_fault(vma, vmf, ext4_get_block);
206                                         /* Is this the right get_block? */
207 }
208
209 static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
210 {
211         return dax_mkwrite(vma, vmf, ext4_get_block);
212 }
213
214 static const struct vm_operations_struct ext4_dax_vm_ops = {
215         .fault          = ext4_dax_fault,
216         .page_mkwrite   = ext4_dax_mkwrite,
217 };
218 #else
219 #define ext4_dax_vm_ops ext4_file_vm_ops
220 #endif
221
222 static const struct vm_operations_struct ext4_file_vm_ops = {
223         .fault          = filemap_fault,
224         .map_pages      = filemap_map_pages,
225         .page_mkwrite   = ext4_page_mkwrite,
226 };
227
228 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
229 {
230         file_accessed(file);
231         if (IS_DAX(file_inode(file))) {
232                 vma->vm_ops = &ext4_dax_vm_ops;
233                 vma->vm_flags |= VM_MIXEDMAP;
234         } else {
235                 vma->vm_ops = &ext4_file_vm_ops;
236         }
237         return 0;
238 }
239
240 static int ext4_file_open(struct inode * inode, struct file * filp)
241 {
242         struct super_block *sb = inode->i_sb;
243         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
244         struct vfsmount *mnt = filp->f_path.mnt;
245         struct path path;
246         char buf[64], *cp;
247
248         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
249                      !(sb->s_flags & MS_RDONLY))) {
250                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
251                 /*
252                  * Sample where the filesystem has been mounted and
253                  * store it in the superblock for sysadmin convenience
254                  * when trying to sort through large numbers of block
255                  * devices or filesystem images.
256                  */
257                 memset(buf, 0, sizeof(buf));
258                 path.mnt = mnt;
259                 path.dentry = mnt->mnt_root;
260                 cp = d_path(&path, buf, sizeof(buf));
261                 if (!IS_ERR(cp)) {
262                         handle_t *handle;
263                         int err;
264
265                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
266                         if (IS_ERR(handle))
267                                 return PTR_ERR(handle);
268                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
269                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
270                         if (err) {
271                                 ext4_journal_stop(handle);
272                                 return err;
273                         }
274                         strlcpy(sbi->s_es->s_last_mounted, cp,
275                                 sizeof(sbi->s_es->s_last_mounted));
276                         ext4_handle_dirty_super(handle, sb);
277                         ext4_journal_stop(handle);
278                 }
279         }
280         /*
281          * Set up the jbd2_inode if we are opening the inode for
282          * writing and the journal is present
283          */
284         if (filp->f_mode & FMODE_WRITE) {
285                 int ret = ext4_inode_attach_jinode(inode);
286                 if (ret < 0)
287                         return ret;
288         }
289         return dquot_file_open(inode, filp);
290 }
291
292 /*
293  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
294  * file rather than ext4_ext_walk_space() because we can introduce
295  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
296  * function.  When extent status tree has been fully implemented, it will
297  * track all extent status for a file and we can directly use it to
298  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
299  */
300
301 /*
302  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
303  * lookup page cache to check whether or not there has some data between
304  * [startoff, endoff] because, if this range contains an unwritten extent,
305  * we determine this extent as a data or a hole according to whether the
306  * page cache has data or not.
307  */
308 static int ext4_find_unwritten_pgoff(struct inode *inode,
309                                      int whence,
310                                      struct ext4_map_blocks *map,
311                                      loff_t *offset)
312 {
313         struct pagevec pvec;
314         unsigned int blkbits;
315         pgoff_t index;
316         pgoff_t end;
317         loff_t endoff;
318         loff_t startoff;
319         loff_t lastoff;
320         int found = 0;
321
322         blkbits = inode->i_sb->s_blocksize_bits;
323         startoff = *offset;
324         lastoff = startoff;
325         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
326
327         index = startoff >> PAGE_CACHE_SHIFT;
328         end = endoff >> PAGE_CACHE_SHIFT;
329
330         pagevec_init(&pvec, 0);
331         do {
332                 int i, num;
333                 unsigned long nr_pages;
334
335                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
336                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
337                                           (pgoff_t)num);
338                 if (nr_pages == 0) {
339                         if (whence == SEEK_DATA)
340                                 break;
341
342                         BUG_ON(whence != SEEK_HOLE);
343                         /*
344                          * If this is the first time to go into the loop and
345                          * offset is not beyond the end offset, it will be a
346                          * hole at this offset
347                          */
348                         if (lastoff == startoff || lastoff < endoff)
349                                 found = 1;
350                         break;
351                 }
352
353                 /*
354                  * If this is the first time to go into the loop and
355                  * offset is smaller than the first page offset, it will be a
356                  * hole at this offset.
357                  */
358                 if (lastoff == startoff && whence == SEEK_HOLE &&
359                     lastoff < page_offset(pvec.pages[0])) {
360                         found = 1;
361                         break;
362                 }
363
364                 for (i = 0; i < nr_pages; i++) {
365                         struct page *page = pvec.pages[i];
366                         struct buffer_head *bh, *head;
367
368                         /*
369                          * If the current offset is not beyond the end of given
370                          * range, it will be a hole.
371                          */
372                         if (lastoff < endoff && whence == SEEK_HOLE &&
373                             page->index > end) {
374                                 found = 1;
375                                 *offset = lastoff;
376                                 goto out;
377                         }
378
379                         lock_page(page);
380
381                         if (unlikely(page->mapping != inode->i_mapping)) {
382                                 unlock_page(page);
383                                 continue;
384                         }
385
386                         if (!page_has_buffers(page)) {
387                                 unlock_page(page);
388                                 continue;
389                         }
390
391                         if (page_has_buffers(page)) {
392                                 lastoff = page_offset(page);
393                                 bh = head = page_buffers(page);
394                                 do {
395                                         if (buffer_uptodate(bh) ||
396                                             buffer_unwritten(bh)) {
397                                                 if (whence == SEEK_DATA)
398                                                         found = 1;
399                                         } else {
400                                                 if (whence == SEEK_HOLE)
401                                                         found = 1;
402                                         }
403                                         if (found) {
404                                                 *offset = max_t(loff_t,
405                                                         startoff, lastoff);
406                                                 unlock_page(page);
407                                                 goto out;
408                                         }
409                                         lastoff += bh->b_size;
410                                         bh = bh->b_this_page;
411                                 } while (bh != head);
412                         }
413
414                         lastoff = page_offset(page) + PAGE_SIZE;
415                         unlock_page(page);
416                 }
417
418                 /*
419                  * The no. of pages is less than our desired, that would be a
420                  * hole in there.
421                  */
422                 if (nr_pages < num && whence == SEEK_HOLE) {
423                         found = 1;
424                         *offset = lastoff;
425                         break;
426                 }
427
428                 index = pvec.pages[i - 1]->index + 1;
429                 pagevec_release(&pvec);
430         } while (index <= end);
431
432 out:
433         pagevec_release(&pvec);
434         return found;
435 }
436
437 /*
438  * ext4_seek_data() retrieves the offset for SEEK_DATA.
439  */
440 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
441 {
442         struct inode *inode = file->f_mapping->host;
443         struct ext4_map_blocks map;
444         struct extent_status es;
445         ext4_lblk_t start, last, end;
446         loff_t dataoff, isize;
447         int blkbits;
448         int ret = 0;
449
450         mutex_lock(&inode->i_mutex);
451
452         isize = i_size_read(inode);
453         if (offset >= isize) {
454                 mutex_unlock(&inode->i_mutex);
455                 return -ENXIO;
456         }
457
458         blkbits = inode->i_sb->s_blocksize_bits;
459         start = offset >> blkbits;
460         last = start;
461         end = isize >> blkbits;
462         dataoff = offset;
463
464         do {
465                 map.m_lblk = last;
466                 map.m_len = end - last + 1;
467                 ret = ext4_map_blocks(NULL, inode, &map, 0);
468                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
469                         if (last != start)
470                                 dataoff = (loff_t)last << blkbits;
471                         break;
472                 }
473
474                 /*
475                  * If there is a delay extent at this offset,
476                  * it will be as a data.
477                  */
478                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
479                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
480                         if (last != start)
481                                 dataoff = (loff_t)last << blkbits;
482                         break;
483                 }
484
485                 /*
486                  * If there is a unwritten extent at this offset,
487                  * it will be as a data or a hole according to page
488                  * cache that has data or not.
489                  */
490                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
491                         int unwritten;
492                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
493                                                               &map, &dataoff);
494                         if (unwritten)
495                                 break;
496                 }
497
498                 last++;
499                 dataoff = (loff_t)last << blkbits;
500         } while (last <= end);
501
502         mutex_unlock(&inode->i_mutex);
503
504         if (dataoff > isize)
505                 return -ENXIO;
506
507         return vfs_setpos(file, dataoff, maxsize);
508 }
509
510 /*
511  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
512  */
513 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
514 {
515         struct inode *inode = file->f_mapping->host;
516         struct ext4_map_blocks map;
517         struct extent_status es;
518         ext4_lblk_t start, last, end;
519         loff_t holeoff, isize;
520         int blkbits;
521         int ret = 0;
522
523         mutex_lock(&inode->i_mutex);
524
525         isize = i_size_read(inode);
526         if (offset >= isize) {
527                 mutex_unlock(&inode->i_mutex);
528                 return -ENXIO;
529         }
530
531         blkbits = inode->i_sb->s_blocksize_bits;
532         start = offset >> blkbits;
533         last = start;
534         end = isize >> blkbits;
535         holeoff = offset;
536
537         do {
538                 map.m_lblk = last;
539                 map.m_len = end - last + 1;
540                 ret = ext4_map_blocks(NULL, inode, &map, 0);
541                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
542                         last += ret;
543                         holeoff = (loff_t)last << blkbits;
544                         continue;
545                 }
546
547                 /*
548                  * If there is a delay extent at this offset,
549                  * we will skip this extent.
550                  */
551                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
552                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
553                         last = es.es_lblk + es.es_len;
554                         holeoff = (loff_t)last << blkbits;
555                         continue;
556                 }
557
558                 /*
559                  * If there is a unwritten extent at this offset,
560                  * it will be as a data or a hole according to page
561                  * cache that has data or not.
562                  */
563                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
564                         int unwritten;
565                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
566                                                               &map, &holeoff);
567                         if (!unwritten) {
568                                 last += ret;
569                                 holeoff = (loff_t)last << blkbits;
570                                 continue;
571                         }
572                 }
573
574                 /* find a hole */
575                 break;
576         } while (last <= end);
577
578         mutex_unlock(&inode->i_mutex);
579
580         if (holeoff > isize)
581                 holeoff = isize;
582
583         return vfs_setpos(file, holeoff, maxsize);
584 }
585
586 /*
587  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
588  * by calling generic_file_llseek_size() with the appropriate maxbytes
589  * value for each.
590  */
591 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
592 {
593         struct inode *inode = file->f_mapping->host;
594         loff_t maxbytes;
595
596         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
597                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
598         else
599                 maxbytes = inode->i_sb->s_maxbytes;
600
601         switch (whence) {
602         case SEEK_SET:
603         case SEEK_CUR:
604         case SEEK_END:
605                 return generic_file_llseek_size(file, offset, whence,
606                                                 maxbytes, i_size_read(inode));
607         case SEEK_DATA:
608                 return ext4_seek_data(file, offset, maxbytes);
609         case SEEK_HOLE:
610                 return ext4_seek_hole(file, offset, maxbytes);
611         }
612
613         return -EINVAL;
614 }
615
616 const struct file_operations ext4_file_operations = {
617         .llseek         = ext4_llseek,
618         .read_iter      = generic_file_read_iter,
619         .write_iter     = ext4_file_write_iter,
620         .unlocked_ioctl = ext4_ioctl,
621 #ifdef CONFIG_COMPAT
622         .compat_ioctl   = ext4_compat_ioctl,
623 #endif
624         .mmap           = ext4_file_mmap,
625         .open           = ext4_file_open,
626         .release        = ext4_release_file,
627         .fsync          = ext4_sync_file,
628         .splice_read    = generic_file_splice_read,
629         .splice_write   = iter_file_splice_write,
630         .fallocate      = ext4_fallocate,
631 };
632
633 #ifdef CONFIG_FS_DAX
634 const struct file_operations ext4_dax_file_operations = {
635         .llseek         = ext4_llseek,
636         .read_iter      = generic_file_read_iter,
637         .write_iter     = ext4_file_write_iter,
638         .unlocked_ioctl = ext4_ioctl,
639 #ifdef CONFIG_COMPAT
640         .compat_ioctl   = ext4_compat_ioctl,
641 #endif
642         .mmap           = ext4_file_mmap,
643         .open           = ext4_file_open,
644         .release        = ext4_release_file,
645         .fsync          = ext4_sync_file,
646         /* Splice not yet supported with DAX */
647         .fallocate      = ext4_fallocate,
648 };
649 #endif
650
651 const struct inode_operations ext4_file_inode_operations = {
652         .setattr        = ext4_setattr,
653         .getattr        = ext4_getattr,
654         .setxattr       = generic_setxattr,
655         .getxattr       = generic_getxattr,
656         .listxattr      = ext4_listxattr,
657         .removexattr    = generic_removexattr,
658         .get_acl        = ext4_get_acl,
659         .set_acl        = ext4_set_acl,
660         .fiemap         = ext4_fiemap,
661 };
662