Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/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/aio.h>
 27 #include <linux/quotaops.h>
 28 #include <linux/pagevec.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 = file->f_flags & O_DIRECT;
 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 
136                 if (pos + length > sbi->s_bitmap_maxbytes)
137                         iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
138         }
139 
140         iocb->private = &overwrite;
141         if (o_direct) {
142                 blk_start_plug(&plug);
143 
144 
145                 /* check whether we do a DIO overwrite or not */
146                 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
147                     !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
148                         struct ext4_map_blocks map;
149                         unsigned int blkbits = inode->i_blkbits;
150                         int err, len;
151 
152                         map.m_lblk = pos >> blkbits;
153                         map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
154                                 - map.m_lblk;
155                         len = map.m_len;
156 
157                         err = ext4_map_blocks(NULL, inode, &map, 0);
158                         /*
159                          * 'err==len' means that all of blocks has
160                          * been preallocated no matter they are
161                          * initialized or not.  For excluding
162                          * unwritten extents, we need to check
163                          * m_flags.  There are two conditions that
164                          * indicate for initialized extents.  1) If we
165                          * hit extent cache, EXT4_MAP_MAPPED flag is
166                          * returned; 2) If we do a real lookup,
167                          * non-flags are returned.  So we should check
168                          * these two conditions.
169                          */
170                         if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
171                                 overwrite = 1;
172                 }
173         }
174 
175         ret = __generic_file_write_iter(iocb, from);
176         mutex_unlock(&inode->i_mutex);
177 
178         if (ret > 0) {
179                 ssize_t err;
180 
181                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
182                 if (err < 0)
183                         ret = err;
184         }
185         if (o_direct)
186                 blk_finish_plug(&plug);
187 
188 errout:
189         if (aio_mutex)
190                 mutex_unlock(aio_mutex);
191         return ret;
192 }
193 
194 static const struct vm_operations_struct ext4_file_vm_ops = {
195         .fault          = filemap_fault,
196         .map_pages      = filemap_map_pages,
197         .page_mkwrite   = ext4_page_mkwrite,
198         .remap_pages    = generic_file_remap_pages,
199 };
200 
201 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
202 {
203         file_accessed(file);
204         vma->vm_ops = &ext4_file_vm_ops;
205         return 0;
206 }
207 
208 static int ext4_file_open(struct inode * inode, struct file * filp)
209 {
210         struct super_block *sb = inode->i_sb;
211         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
212         struct vfsmount *mnt = filp->f_path.mnt;
213         struct path path;
214         char buf[64], *cp;
215 
216         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
217                      !(sb->s_flags & MS_RDONLY))) {
218                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
219                 /*
220                  * Sample where the filesystem has been mounted and
221                  * store it in the superblock for sysadmin convenience
222                  * when trying to sort through large numbers of block
223                  * devices or filesystem images.
224                  */
225                 memset(buf, 0, sizeof(buf));
226                 path.mnt = mnt;
227                 path.dentry = mnt->mnt_root;
228                 cp = d_path(&path, buf, sizeof(buf));
229                 if (!IS_ERR(cp)) {
230                         handle_t *handle;
231                         int err;
232 
233                         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
234                         if (IS_ERR(handle))
235                                 return PTR_ERR(handle);
236                         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
237                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
238                         if (err) {
239                                 ext4_journal_stop(handle);
240                                 return err;
241                         }
242                         strlcpy(sbi->s_es->s_last_mounted, cp,
243                                 sizeof(sbi->s_es->s_last_mounted));
244                         ext4_handle_dirty_super(handle, sb);
245                         ext4_journal_stop(handle);
246                 }
247         }
248         /*
249          * Set up the jbd2_inode if we are opening the inode for
250          * writing and the journal is present
251          */
252         if (filp->f_mode & FMODE_WRITE) {
253                 int ret = ext4_inode_attach_jinode(inode);
254                 if (ret < 0)
255                         return ret;
256         }
257         return dquot_file_open(inode, filp);
258 }
259 
260 /*
261  * Here we use ext4_map_blocks() to get a block mapping for a extent-based
262  * file rather than ext4_ext_walk_space() because we can introduce
263  * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
264  * function.  When extent status tree has been fully implemented, it will
265  * track all extent status for a file and we can directly use it to
266  * retrieve the offset for SEEK_DATA/SEEK_HOLE.
267  */
268 
269 /*
270  * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
271  * lookup page cache to check whether or not there has some data between
272  * [startoff, endoff] because, if this range contains an unwritten extent,
273  * we determine this extent as a data or a hole according to whether the
274  * page cache has data or not.
275  */
276 static int ext4_find_unwritten_pgoff(struct inode *inode,
277                                      int whence,
278                                      struct ext4_map_blocks *map,
279                                      loff_t *offset)
280 {
281         struct pagevec pvec;
282         unsigned int blkbits;
283         pgoff_t index;
284         pgoff_t end;
285         loff_t endoff;
286         loff_t startoff;
287         loff_t lastoff;
288         int found = 0;
289 
290         blkbits = inode->i_sb->s_blocksize_bits;
291         startoff = *offset;
292         lastoff = startoff;
293         endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
294 
295         index = startoff >> PAGE_CACHE_SHIFT;
296         end = endoff >> PAGE_CACHE_SHIFT;
297 
298         pagevec_init(&pvec, 0);
299         do {
300                 int i, num;
301                 unsigned long nr_pages;
302 
303                 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
304                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
305                                           (pgoff_t)num);
306                 if (nr_pages == 0) {
307                         if (whence == SEEK_DATA)
308                                 break;
309 
310                         BUG_ON(whence != SEEK_HOLE);
311                         /*
312                          * If this is the first time to go into the loop and
313                          * offset is not beyond the end offset, it will be a
314                          * hole at this offset
315                          */
316                         if (lastoff == startoff || lastoff < endoff)
317                                 found = 1;
318                         break;
319                 }
320 
321                 /*
322                  * If this is the first time to go into the loop and
323                  * offset is smaller than the first page offset, it will be a
324                  * hole at this offset.
325                  */
326                 if (lastoff == startoff && whence == SEEK_HOLE &&
327                     lastoff < page_offset(pvec.pages[0])) {
328                         found = 1;
329                         break;
330                 }
331 
332                 for (i = 0; i < nr_pages; i++) {
333                         struct page *page = pvec.pages[i];
334                         struct buffer_head *bh, *head;
335 
336                         /*
337                          * If the current offset is not beyond the end of given
338                          * range, it will be a hole.
339                          */
340                         if (lastoff < endoff && whence == SEEK_HOLE &&
341                             page->index > end) {
342                                 found = 1;
343                                 *offset = lastoff;
344                                 goto out;
345                         }
346 
347                         lock_page(page);
348 
349                         if (unlikely(page->mapping != inode->i_mapping)) {
350                                 unlock_page(page);
351                                 continue;
352                         }
353 
354                         if (!page_has_buffers(page)) {
355                                 unlock_page(page);
356                                 continue;
357                         }
358 
359                         if (page_has_buffers(page)) {
360                                 lastoff = page_offset(page);
361                                 bh = head = page_buffers(page);
362                                 do {
363                                         if (buffer_uptodate(bh) ||
364                                             buffer_unwritten(bh)) {
365                                                 if (whence == SEEK_DATA)
366                                                         found = 1;
367                                         } else {
368                                                 if (whence == SEEK_HOLE)
369                                                         found = 1;
370                                         }
371                                         if (found) {
372                                                 *offset = max_t(loff_t,
373                                                         startoff, lastoff);
374                                                 unlock_page(page);
375                                                 goto out;
376                                         }
377                                         lastoff += bh->b_size;
378                                         bh = bh->b_this_page;
379                                 } while (bh != head);
380                         }
381 
382                         lastoff = page_offset(page) + PAGE_SIZE;
383                         unlock_page(page);
384                 }
385 
386                 /*
387                  * The no. of pages is less than our desired, that would be a
388                  * hole in there.
389                  */
390                 if (nr_pages < num && whence == SEEK_HOLE) {
391                         found = 1;
392                         *offset = lastoff;
393                         break;
394                 }
395 
396                 index = pvec.pages[i - 1]->index + 1;
397                 pagevec_release(&pvec);
398         } while (index <= end);
399 
400 out:
401         pagevec_release(&pvec);
402         return found;
403 }
404 
405 /*
406  * ext4_seek_data() retrieves the offset for SEEK_DATA.
407  */
408 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
409 {
410         struct inode *inode = file->f_mapping->host;
411         struct ext4_map_blocks map;
412         struct extent_status es;
413         ext4_lblk_t start, last, end;
414         loff_t dataoff, isize;
415         int blkbits;
416         int ret = 0;
417 
418         mutex_lock(&inode->i_mutex);
419 
420         isize = i_size_read(inode);
421         if (offset >= isize) {
422                 mutex_unlock(&inode->i_mutex);
423                 return -ENXIO;
424         }
425 
426         blkbits = inode->i_sb->s_blocksize_bits;
427         start = offset >> blkbits;
428         last = start;
429         end = isize >> blkbits;
430         dataoff = offset;
431 
432         do {
433                 map.m_lblk = last;
434                 map.m_len = end - last + 1;
435                 ret = ext4_map_blocks(NULL, inode, &map, 0);
436                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
437                         if (last != start)
438                                 dataoff = (loff_t)last << blkbits;
439                         break;
440                 }
441 
442                 /*
443                  * If there is a delay extent at this offset,
444                  * it will be as a data.
445                  */
446                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
447                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
448                         if (last != start)
449                                 dataoff = (loff_t)last << blkbits;
450                         break;
451                 }
452 
453                 /*
454                  * If there is a unwritten extent at this offset,
455                  * it will be as a data or a hole according to page
456                  * cache that has data or not.
457                  */
458                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
459                         int unwritten;
460                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
461                                                               &map, &dataoff);
462                         if (unwritten)
463                                 break;
464                 }
465 
466                 last++;
467                 dataoff = (loff_t)last << blkbits;
468         } while (last <= end);
469 
470         mutex_unlock(&inode->i_mutex);
471 
472         if (dataoff > isize)
473                 return -ENXIO;
474 
475         return vfs_setpos(file, dataoff, maxsize);
476 }
477 
478 /*
479  * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
480  */
481 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
482 {
483         struct inode *inode = file->f_mapping->host;
484         struct ext4_map_blocks map;
485         struct extent_status es;
486         ext4_lblk_t start, last, end;
487         loff_t holeoff, isize;
488         int blkbits;
489         int ret = 0;
490 
491         mutex_lock(&inode->i_mutex);
492 
493         isize = i_size_read(inode);
494         if (offset >= isize) {
495                 mutex_unlock(&inode->i_mutex);
496                 return -ENXIO;
497         }
498 
499         blkbits = inode->i_sb->s_blocksize_bits;
500         start = offset >> blkbits;
501         last = start;
502         end = isize >> blkbits;
503         holeoff = offset;
504 
505         do {
506                 map.m_lblk = last;
507                 map.m_len = end - last + 1;
508                 ret = ext4_map_blocks(NULL, inode, &map, 0);
509                 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
510                         last += ret;
511                         holeoff = (loff_t)last << blkbits;
512                         continue;
513                 }
514 
515                 /*
516                  * If there is a delay extent at this offset,
517                  * we will skip this extent.
518                  */
519                 ext4_es_find_delayed_extent_range(inode, last, last, &es);
520                 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
521                         last = es.es_lblk + es.es_len;
522                         holeoff = (loff_t)last << blkbits;
523                         continue;
524                 }
525 
526                 /*
527                  * If there is a unwritten extent at this offset,
528                  * it will be as a data or a hole according to page
529                  * cache that has data or not.
530                  */
531                 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
532                         int unwritten;
533                         unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
534                                                               &map, &holeoff);
535                         if (!unwritten) {
536                                 last += ret;
537                                 holeoff = (loff_t)last << blkbits;
538                                 continue;
539                         }
540                 }
541 
542                 /* find a hole */
543                 break;
544         } while (last <= end);
545 
546         mutex_unlock(&inode->i_mutex);
547 
548         if (holeoff > isize)
549                 holeoff = isize;
550 
551         return vfs_setpos(file, holeoff, maxsize);
552 }
553 
554 /*
555  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
556  * by calling generic_file_llseek_size() with the appropriate maxbytes
557  * value for each.
558  */
559 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
560 {
561         struct inode *inode = file->f_mapping->host;
562         loff_t maxbytes;
563 
564         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
565                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
566         else
567                 maxbytes = inode->i_sb->s_maxbytes;
568 
569         switch (whence) {
570         case SEEK_SET:
571         case SEEK_CUR:
572         case SEEK_END:
573                 return generic_file_llseek_size(file, offset, whence,
574                                                 maxbytes, i_size_read(inode));
575         case SEEK_DATA:
576                 return ext4_seek_data(file, offset, maxbytes);
577         case SEEK_HOLE:
578                 return ext4_seek_hole(file, offset, maxbytes);
579         }
580 
581         return -EINVAL;
582 }
583 
584 const struct file_operations ext4_file_operations = {
585         .llseek         = ext4_llseek,
586         .read           = new_sync_read,
587         .write          = new_sync_write,
588         .read_iter      = generic_file_read_iter,
589         .write_iter     = ext4_file_write_iter,
590         .unlocked_ioctl = ext4_ioctl,
591 #ifdef CONFIG_COMPAT
592         .compat_ioctl   = ext4_compat_ioctl,
593 #endif
594         .mmap           = ext4_file_mmap,
595         .open           = ext4_file_open,
596         .release        = ext4_release_file,
597         .fsync          = ext4_sync_file,
598         .splice_read    = generic_file_splice_read,
599         .splice_write   = iter_file_splice_write,
600         .fallocate      = ext4_fallocate,
601 };
602 
603 const struct inode_operations ext4_file_inode_operations = {
604         .setattr        = ext4_setattr,
605         .getattr        = ext4_getattr,
606         .setxattr       = generic_setxattr,
607         .getxattr       = generic_getxattr,
608         .listxattr      = ext4_listxattr,
609         .removexattr    = generic_removexattr,
610         .get_acl        = ext4_get_acl,
611         .set_acl        = ext4_set_acl,
612         .fiemap         = ext4_fiemap,
613 };
614 
615 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us