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

TOMOYO Linux Cross Reference
Linux/fs/btrfs/reflink.c

Version: ~ [ linux-6.12-rc7 ] ~ [ linux-6.11.7 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.60 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.116 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.171 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.229 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.285 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.323 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.12 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

Diff markup

Differences between /fs/btrfs/reflink.c (Version linux-6.12-rc7) and /fs/btrfs/reflink.c (Version linux-2.6.0)


  1 // SPDX-License-Identifier: GPL-2.0                 1 
  2                                                   
  3 #include <linux/blkdev.h>                         
  4 #include <linux/iversion.h>                       
  5 #include "ctree.h"                                
  6 #include "fs.h"                                   
  7 #include "messages.h"                             
  8 #include "compression.h"                          
  9 #include "delalloc-space.h"                       
 10 #include "disk-io.h"                              
 11 #include "reflink.h"                              
 12 #include "transaction.h"                          
 13 #include "subpage.h"                              
 14 #include "accessors.h"                            
 15 #include "file-item.h"                            
 16 #include "file.h"                                 
 17 #include "super.h"                                
 18                                                   
 19 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M            
 20                                                   
 21 static int clone_finish_inode_update(struct bt    
 22                                      struct in    
 23                                      u64 endof    
 24                                      const u64    
 25                                      const u64    
 26                                      int no_ti    
 27 {                                                 
 28         int ret;                                  
 29                                                   
 30         inode_inc_iversion(inode);                
 31         if (!no_time_update) {                    
 32                 inode_set_mtime_to_ts(inode, i    
 33         }                                         
 34         /*                                        
 35          * We round up to the block size at eo    
 36          * extents to clone above, but shouldn    
 37          */                                       
 38         if (endoff > destoff + olen)              
 39                 endoff = destoff + olen;          
 40         if (endoff > inode->i_size) {             
 41                 i_size_write(inode, endoff);      
 42                 btrfs_inode_safe_disk_i_size_w    
 43         }                                         
 44                                                   
 45         ret = btrfs_update_inode(trans, BTRFS_    
 46         if (ret) {                                
 47                 btrfs_abort_transaction(trans,    
 48                 btrfs_end_transaction(trans);     
 49                 goto out;                         
 50         }                                         
 51         ret = btrfs_end_transaction(trans);       
 52 out:                                              
 53         return ret;                               
 54 }                                                 
 55                                                   
 56 static int copy_inline_to_page(struct btrfs_in    
 57                                const u64 file_    
 58                                char *inline_da    
 59                                const u64 size,    
 60                                const u64 datal    
 61                                const u8 comp_t    
 62 {                                                 
 63         struct btrfs_fs_info *fs_info = inode-    
 64         const u32 block_size = fs_info->sector    
 65         const u64 range_end = file_offset + bl    
 66         const size_t inline_size = size - btrf    
 67         char *data_start = inline_data + btrfs    
 68         struct extent_changeset *data_reserved    
 69         struct folio *folio = NULL;               
 70         struct address_space *mapping = inode-    
 71         int ret;                                  
 72                                                   
 73         ASSERT(IS_ALIGNED(file_offset, block_s    
 74                                                   
 75         /*                                        
 76          * We have flushed and locked the rang    
 77          * inodes, we also have locked the ino    
 78          * reservation here. Also we must not     
 79          * a transaction open, otherwise we wo    
 80          */                                       
 81         ret = btrfs_delalloc_reserve_space(ino    
 82                                            blo    
 83         if (ret)                                  
 84                 goto out;                         
 85                                                   
 86         folio = __filemap_get_folio(mapping, f    
 87                                         FGP_LO    
 88                                         btrfs_    
 89         if (IS_ERR(folio)) {                      
 90                 ret = -ENOMEM;                    
 91                 goto out_unlock;                  
 92         }                                         
 93                                                   
 94         ret = set_folio_extent_mapped(folio);     
 95         if (ret < 0)                              
 96                 goto out_unlock;                  
 97                                                   
 98         clear_extent_bit(&inode->io_tree, file    
 99                          EXTENT_DELALLOC | EXT    
100                          NULL);                   
101         ret = btrfs_set_extent_delalloc(inode,    
102         if (ret)                                  
103                 goto out_unlock;                  
104                                                   
105         /*                                        
106          * After dirtying the page our caller     
107          * and if we are low on metadata free     
108          * delalloc for all inodes in order to    
109          * However we are holding the range lo    
110          * the clone/dedupe operation, so we m    
111          * other task releases enough space. S    
112          * possible to flush to avoid such dea    
113          * when we finish cloning all extents,    
114          * after finding each extent to clone.    
115          */                                       
116         set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH,    
117                                                   
118         if (comp_type == BTRFS_COMPRESS_NONE)     
119                 memcpy_to_folio(folio, offset_    
120                                         datal)    
121         } else {                                  
122                 ret = btrfs_decompress(comp_ty    
123                                        offset_    
124                                        inline_    
125                 if (ret)                          
126                         goto out_unlock;          
127                 flush_dcache_folio(folio);        
128         }                                         
129                                                   
130         /*                                        
131          * If our inline data is smaller then     
132          * remaining of the block/page is equi    
133          * like the following done:               
134          *                                        
135          * $ xfs_io -f -c "pwrite -S 0xab 0 50    
136          * $ sync  # (or fsync)                   
137          * $ xfs_io -c "falloc 0 4K" file         
138          * $ xfs_io -c "pwrite -S 0xcd 4K 4K"     
139          *                                        
140          * So what's in the range [500, 4095]     
141          */                                       
142         if (datal < block_size)                   
143                 folio_zero_range(folio, datal,    
144                                                   
145         btrfs_folio_set_uptodate(fs_info, foli    
146         btrfs_folio_clear_checked(fs_info, fol    
147         btrfs_folio_set_dirty(fs_info, folio,     
148 out_unlock:                                       
149         if (!IS_ERR(folio)) {                     
150                 folio_unlock(folio);              
151                 folio_put(folio);                 
152         }                                         
153         if (ret)                                  
154                 btrfs_delalloc_release_space(i    
155                                              b    
156         btrfs_delalloc_release_extents(inode,     
157 out:                                              
158         extent_changeset_free(data_reserved);     
159                                                   
160         return ret;                               
161 }                                                 
162                                                   
163 /*                                                
164  * Deal with cloning of inline extents. We try    
165  * the source inode to destination inode when     
166  * copy the inline extent's data into the resp    
167  */                                               
168 static int clone_copy_inline_extent(struct ino    
169                                     struct btr    
170                                     struct btr    
171                                     const u64     
172                                     const u64     
173                                     const u64     
174                                     const u8 c    
175                                     char *inli    
176                                     struct btr    
177 {                                                 
178         struct btrfs_fs_info *fs_info = inode_    
179         struct btrfs_root *root = BTRFS_I(dst)    
180         const u64 aligned_end = ALIGN(new_key-    
181                                       fs_info-    
182         struct btrfs_trans_handle *trans = NUL    
183         struct btrfs_drop_extents_args drop_ar    
184         int ret;                                  
185         struct btrfs_key key;                     
186                                                   
187         if (new_key->offset > 0) {                
188                 ret = copy_inline_to_page(BTRF    
189                                           inli    
190                 goto out;                         
191         }                                         
192                                                   
193         key.objectid = btrfs_ino(BTRFS_I(dst))    
194         key.type = BTRFS_EXTENT_DATA_KEY;         
195         key.offset = 0;                           
196         ret = btrfs_search_slot(NULL, root, &k    
197         if (ret < 0) {                            
198                 return ret;                       
199         } else if (ret > 0) {                     
200                 if (path->slots[0] >= btrfs_he    
201                         ret = btrfs_next_leaf(    
202                         if (ret < 0)              
203                                 return ret;       
204                         else if (ret > 0)         
205                                 goto copy_inli    
206                 }                                 
207                 btrfs_item_key_to_cpu(path->no    
208                 if (key.objectid == btrfs_ino(    
209                     key.type == BTRFS_EXTENT_D    
210                         /*                        
211                          * There's an implicit    
212                          * inline extent's dat    
213                          */                       
214                         ASSERT(key.offset > 0)    
215                         goto copy_to_page;        
216                 }                                 
217         } else if (i_size_read(dst) <= datal)     
218                 struct btrfs_file_extent_item     
219                                                   
220                 ei = btrfs_item_ptr(path->node    
221                                     struct btr    
222                 /*                                
223                  * If it's an inline extent re    
224                  * extent, otherwise copy the     
225                  * the respective page at the     
226                  */                               
227                 if (btrfs_file_extent_type(pat    
228                     BTRFS_FILE_EXTENT_INLINE)     
229                         goto copy_inline_exten    
230                                                   
231                 goto copy_to_page;                
232         }                                         
233                                                   
234 copy_inline_extent:                               
235         /*                                        
236          * We have no extent items, or we have    
237          * or may not be inlined. All these ca    
238          */                                       
239         if (i_size_read(dst) > datal) {           
240                 /*                                
241                  * At the destination offset 0    
242                  * extent or an inline extent     
243                  * clone. Deal with all these     
244                  * data into the respective pa    
245                  */                               
246                 goto copy_to_page;                
247         }                                         
248                                                   
249         /*                                        
250          * Release path before starting a new     
251          * that would confuse lockdep.            
252          */                                       
253         btrfs_release_path(path);                 
254         /*                                        
255          * If we end up here it means were cop    
256          * of the destination inode. We know w    
257          * extent item in the destination root    
258          *                                        
259          * 1 unit - adjusting old extent (we m    
260          * 1 unit - add new extent                
261          * 1 unit - inode update                  
262          */                                       
263         trans = btrfs_start_transaction(root,     
264         if (IS_ERR(trans)) {                      
265                 ret = PTR_ERR(trans);             
266                 trans = NULL;                     
267                 goto out;                         
268         }                                         
269         drop_args.path = path;                    
270         drop_args.start = drop_start;             
271         drop_args.end = aligned_end;              
272         drop_args.drop_cache = true;              
273         ret = btrfs_drop_extents(trans, root,     
274         if (ret)                                  
275                 goto out;                         
276         ret = btrfs_insert_empty_item(trans, r    
277         if (ret)                                  
278                 goto out;                         
279                                                   
280         write_extent_buffer(path->nodes[0], in    
281                             btrfs_item_ptr_off    
282                                                   
283                             size);                
284         btrfs_update_inode_bytes(BTRFS_I(dst),    
285         btrfs_set_inode_full_sync(BTRFS_I(dst)    
286         ret = btrfs_inode_set_file_extent_rang    
287 out:                                              
288         if (!ret && !trans) {                     
289                 /*                                
290                  * No transaction here means w    
291                  * page of the destination ino    
292                  *                                
293                  * 1 unit to update inode item    
294                  */                               
295                 trans = btrfs_start_transactio    
296                 if (IS_ERR(trans)) {              
297                         ret = PTR_ERR(trans);     
298                         trans = NULL;             
299                 }                                 
300         }                                         
301         if (ret && trans) {                       
302                 btrfs_abort_transaction(trans,    
303                 btrfs_end_transaction(trans);     
304         }                                         
305         if (!ret)                                 
306                 *trans_out = trans;               
307                                                   
308         return ret;                               
309                                                   
310 copy_to_page:                                     
311         /*                                        
312          * Release our path because we don't n    
313          * copy_inline_to_page() needs to rese    
314          * need to flush delalloc when we are     
315          * therefore cause a deadlock if write    
316          * write to the same leaf or an ordere    
317          * to the same leaf.                      
318          */                                       
319         btrfs_release_path(path);                 
320                                                   
321         ret = copy_inline_to_page(BTRFS_I(dst)    
322                                   inline_data,    
323         goto out;                                 
324 }                                                 
325                                                   
326 /*                                                
327  * Clone a range from inode file to another.      
328  *                                                
329  * @src:             Inode to clone from          
330  * @inode:           Inode to clone to            
331  * @off:             Offset within source to s    
332  * @olen:            Original length, passed b    
333  * @olen_aligned:    Block-aligned value of ol    
334  * @destoff:         Offset within @inode to s    
335  * @no_time_update:  Whether to update mtime/c    
336  */                                               
337 static int btrfs_clone(struct inode *src, stru    
338                        const u64 off, const u6    
339                        const u64 destoff, int     
340 {                                                 
341         struct btrfs_fs_info *fs_info = inode_    
342         struct btrfs_path *path = NULL;           
343         struct extent_buffer *leaf;               
344         struct btrfs_trans_handle *trans;         
345         char *buf = NULL;                         
346         struct btrfs_key key;                     
347         u32 nritems;                              
348         int slot;                                 
349         int ret;                                  
350         const u64 len = olen_aligned;             
351         u64 last_dest_end = destoff;              
352         u64 prev_extent_end = off;                
353                                                   
354         ret = -ENOMEM;                            
355         buf = kvmalloc(fs_info->nodesize, GFP_    
356         if (!buf)                                 
357                 return ret;                       
358                                                   
359         path = btrfs_alloc_path();                
360         if (!path) {                              
361                 kvfree(buf);                      
362                 return ret;                       
363         }                                         
364                                                   
365         path->reada = READA_FORWARD;              
366         /* Clone data */                          
367         key.objectid = btrfs_ino(BTRFS_I(src))    
368         key.type = BTRFS_EXTENT_DATA_KEY;         
369         key.offset = off;                         
370                                                   
371         while (1) {                               
372                 struct btrfs_file_extent_item     
373                 u64 extent_gen;                   
374                 int type;                         
375                 u32 size;                         
376                 struct btrfs_key new_key;         
377                 u64 disko = 0, diskl = 0;         
378                 u64 datao = 0, datal = 0;         
379                 u8 comp;                          
380                 u64 drop_start;                   
381                                                   
382                 /* Note the key will change ty    
383                 ret = btrfs_search_slot(NULL,     
384                                 0, 0);            
385                 if (ret < 0)                      
386                         goto out;                 
387                 /*                                
388                  * First search, if no extent     
389                  * found but the previous item    
390                  * it might overlap our target    
391                  */                               
392                 if (key.offset == off && ret >    
393                         btrfs_item_key_to_cpu(    
394                                                   
395                         if (key.type == BTRFS_    
396                                 path->slots[0]    
397                 }                                 
398                                                   
399                 nritems = btrfs_header_nritems    
400 process_slot:                                     
401                 if (path->slots[0] >= nritems)    
402                         ret = btrfs_next_leaf(    
403                         if (ret < 0)              
404                                 goto out;         
405                         if (ret > 0)              
406                                 break;            
407                         nritems = btrfs_header    
408                 }                                 
409                 leaf = path->nodes[0];            
410                 slot = path->slots[0];            
411                                                   
412                 btrfs_item_key_to_cpu(leaf, &k    
413                 if (key.type > BTRFS_EXTENT_DA    
414                     key.objectid != btrfs_ino(    
415                         break;                    
416                                                   
417                 ASSERT(key.type == BTRFS_EXTEN    
418                                                   
419                 extent = btrfs_item_ptr(leaf,     
420                                         struct    
421                 extent_gen = btrfs_file_extent    
422                 comp = btrfs_file_extent_compr    
423                 type = btrfs_file_extent_type(    
424                 if (type == BTRFS_FILE_EXTENT_    
425                     type == BTRFS_FILE_EXTENT_    
426                         disko = btrfs_file_ext    
427                         diskl = btrfs_file_ext    
428                         datao = btrfs_file_ext    
429                         datal = btrfs_file_ext    
430                 } else if (type == BTRFS_FILE_    
431                         /* Take upper bound, m    
432                         datal = btrfs_file_ext    
433                 }                                 
434                                                   
435                 /*                                
436                  * The first search might have    
437                  * ends before our target rang    
438                  * holes and NO_HOLES feature     
439                  *                                
440                  * Subsequent searches may lea    
441                  * processed before - this hap    
442                  * extent completion for a fil    
443                  * range, but that range was p    
444                  * also covered a leading part    
445                  */                               
446                 if (key.offset + datal <= prev    
447                         path->slots[0]++;         
448                         goto process_slot;        
449                 } else if (key.offset >= off +    
450                         break;                    
451                 }                                 
452                                                   
453                 prev_extent_end = key.offset +    
454                 size = btrfs_item_size(leaf, s    
455                 read_extent_buffer(leaf, buf,     
456                                    size);         
457                                                   
458                 btrfs_release_path(path);         
459                                                   
460                 memcpy(&new_key, &key, sizeof(    
461                 new_key.objectid = btrfs_ino(B    
462                 if (off <= key.offset)            
463                         new_key.offset = key.o    
464                 else                              
465                         new_key.offset = desto    
466                                                   
467                 /*                                
468                  * Deal with a hole that doesn    
469                  * represents it (NO_HOLES fea    
470                  * This hole is either in the     
471                  * the beginning (fully overla    
472                  */                               
473                 if (new_key.offset != last_des    
474                         drop_start = last_dest    
475                 else                              
476                         drop_start = new_key.o    
477                                                   
478                 if (type == BTRFS_FILE_EXTENT_    
479                     type == BTRFS_FILE_EXTENT_    
480                         struct btrfs_replace_e    
481                                                   
482                         /*                        
483                          *    a  | --- range t    
484                          * | ------------- ext    
485                          */                       
486                                                   
487                         /* Subtract range b */    
488                         if (key.offset + datal    
489                                 datal = off +     
490                                                   
491                         /* Subtract range a */    
492                         if (off > key.offset)     
493                                 datao += off -    
494                                 datal -= off -    
495                         }                         
496                                                   
497                         clone_info.disk_offset    
498                         clone_info.disk_len =     
499                         clone_info.data_offset    
500                         clone_info.data_len =     
501                         clone_info.file_offset    
502                         clone_info.extent_buf     
503                         clone_info.is_new_exte    
504                         clone_info.update_time    
505                         ret = btrfs_replace_fi    
506                                         drop_s    
507                                         &clone    
508                         if (ret)                  
509                                 goto out;         
510                 } else {                          
511                         ASSERT(type == BTRFS_F    
512                         /*                        
513                          * Inline extents alwa    
514                          * and can never be bi    
515                          * never clone only pa    
516                          * reflink operations     
517                          * offset, and the len    
518                          * the i_size (which i    
519                          */                       
520                         ASSERT(key.offset == 0    
521                         ASSERT(datal <= fs_inf    
522                         if (WARN_ON(type != BT    
523                             WARN_ON(key.offset    
524                             WARN_ON(datal > fs    
525                                 ret = -EUCLEAN    
526                                 goto out;         
527                         }                         
528                                                   
529                         ret = clone_copy_inlin    
530                                                   
531                                                   
532                         if (ret)                  
533                                 goto out;         
534                 }                                 
535                                                   
536                 btrfs_release_path(path);         
537                                                   
538                 /*                                
539                  * Whenever we share an extent    
540                  * of each inode to the curren    
541                  * make sure fsync does not lo    
542                  * overlapping ranges (because    
543                  * only to sections of the ori    
544                  * inode we do this regardless    
545                  * or even if they are inline     
546                  * sure a full fsync does not     
547                  * we only need to update last    
548                  * extent that is not a hole o    
549                  * the checksums problem on fs    
550                  */                               
551                 if (extent_gen == trans->trans    
552                         BTRFS_I(src)->last_ref    
553                                                   
554                 BTRFS_I(inode)->last_reflink_t    
555                                                   
556                 last_dest_end = ALIGN(new_key.    
557                                       fs_info-    
558                 ret = clone_finish_inode_updat    
559                                                   
560                 if (ret)                          
561                         goto out;                 
562                 if (new_key.offset + datal >=     
563                         break;                    
564                                                   
565                 btrfs_release_path(path);         
566                 key.offset = prev_extent_end;     
567                                                   
568                 if (fatal_signal_pending(curre    
569                         ret = -EINTR;             
570                         goto out;                 
571                 }                                 
572                                                   
573                 cond_resched();                   
574         }                                         
575         ret = 0;                                  
576                                                   
577         if (last_dest_end < destoff + len) {      
578                 /*                                
579                  * We have an implicit hole th    
580                  * cloning range at its end. T    
581                  * NO_HOLES feature enabled or    
582                  * mixing buffered and direct     
583                  */                               
584                 btrfs_release_path(path);         
585                                                   
586                 /*                                
587                  * When using NO_HOLES and we     
588                  * only a hole (no extents) in    
589                  * i_size, punching a hole in     
590                  * an extent map defining a ho    
591                  * beyond current i_size. If t    
592                  * greater than the new i_size    
593                  * need to make sure the next     
594                  * detects and logs a hole cov    
595                  * i_size to the new i_size. I    
596                  * besides a hole, then we kno    
597                  * set by previous calls to bt    
598                  * replaced file extent items.    
599                  */                               
600                 if (last_dest_end >= i_size_re    
601                         btrfs_set_inode_full_s    
602                                                   
603                 ret = btrfs_replace_file_exten    
604                                 last_dest_end,    
605                 if (ret)                          
606                         goto out;                 
607                                                   
608                 ret = clone_finish_inode_updat    
609                                                   
610         }                                         
611                                                   
612 out:                                              
613         btrfs_free_path(path);                    
614         kvfree(buf);                              
615         clear_bit(BTRFS_INODE_NO_DELALLOC_FLUS    
616                                                   
617         return ret;                               
618 }                                                 
619                                                   
620 static void btrfs_double_mmap_lock(struct inod    
621 {                                                 
622         if (inode1 < inode2)                      
623                 swap(inode1, inode2);             
624         down_write(&BTRFS_I(inode1)->i_mmap_lo    
625         down_write_nested(&BTRFS_I(inode2)->i_    
626 }                                                 
627                                                   
628 static void btrfs_double_mmap_unlock(struct in    
629 {                                                 
630         up_write(&BTRFS_I(inode1)->i_mmap_lock    
631         up_write(&BTRFS_I(inode2)->i_mmap_lock    
632 }                                                 
633                                                   
634 static int btrfs_extent_same_range(struct inod    
635                                    struct inod    
636 {                                                 
637         const u64 end = dst_loff + len - 1;       
638         struct extent_state *cached_state = NU    
639         struct btrfs_fs_info *fs_info = BTRFS_    
640         const u64 bs = fs_info->sectorsize;       
641         int ret;                                  
642                                                   
643         /*                                        
644          * Lock destination range to serialize    
645          * we are safe from concurrency with r    
646          * because we have already locked the     
647          * mode.                                  
648          */                                       
649         lock_extent(&BTRFS_I(dst)->io_tree, ds    
650         ret = btrfs_clone(src, dst, loff, len,    
651         unlock_extent(&BTRFS_I(dst)->io_tree,     
652                                                   
653         btrfs_btree_balance_dirty(fs_info);       
654                                                   
655         return ret;                               
656 }                                                 
657                                                   
658 static int btrfs_extent_same(struct inode *src    
659                              struct inode *dst    
660 {                                                 
661         int ret = 0;                              
662         u64 i, tail_len, chunk_count;             
663         struct btrfs_root *root_dst = BTRFS_I(    
664                                                   
665         spin_lock(&root_dst->root_item_lock);     
666         if (root_dst->send_in_progress) {         
667                 btrfs_warn_rl(root_dst->fs_inf    
668 "cannot deduplicate to root %llu while send op    
669                               btrfs_root_id(ro    
670                               root_dst->send_i    
671                 spin_unlock(&root_dst->root_it    
672                 return -EAGAIN;                   
673         }                                         
674         root_dst->dedupe_in_progress++;           
675         spin_unlock(&root_dst->root_item_lock)    
676                                                   
677         tail_len = olen % BTRFS_MAX_DEDUPE_LEN    
678         chunk_count = div_u64(olen, BTRFS_MAX_    
679                                                   
680         for (i = 0; i < chunk_count; i++) {       
681                 ret = btrfs_extent_same_range(    
682                                                   
683                 if (ret)                          
684                         goto out;                 
685                                                   
686                 loff += BTRFS_MAX_DEDUPE_LEN;     
687                 dst_loff += BTRFS_MAX_DEDUPE_L    
688         }                                         
689                                                   
690         if (tail_len > 0)                         
691                 ret = btrfs_extent_same_range(    
692 out:                                              
693         spin_lock(&root_dst->root_item_lock);     
694         root_dst->dedupe_in_progress--;           
695         spin_unlock(&root_dst->root_item_lock)    
696                                                   
697         return ret;                               
698 }                                                 
699                                                   
700 static noinline int btrfs_clone_files(struct f    
701                                         u64 of    
702 {                                                 
703         struct extent_state *cached_state = NU    
704         struct inode *inode = file_inode(file)    
705         struct inode *src = file_inode(file_sr    
706         struct btrfs_fs_info *fs_info = inode_    
707         int ret;                                  
708         int wb_ret;                               
709         u64 len = olen;                           
710         u64 bs = fs_info->sectorsize;             
711         u64 end;                                  
712                                                   
713         /*                                        
714          * VFS's generic_remap_file_range_prep    
715          * eof block into the middle of a file    
716          * if the file size is not blocksize a    
717          * for that case here.                    
718          */                                       
719         if (off + len == src->i_size)             
720                 len = ALIGN(src->i_size, bs) -    
721                                                   
722         if (destoff > inode->i_size) {            
723                 const u64 wb_start = ALIGN_DOW    
724                                                   
725                 ret = btrfs_cont_expand(BTRFS_    
726                 if (ret)                          
727                         return ret;               
728                 /*                                
729                  * We may have truncated the l    
730                  * not sector size aligned, so    
731                  * complete before proceeding     
732                  * with cloning and attempt to    
733                  * extent that no longer exist    
734                  * we found the previous exten    
735                  * attempted to increment its     
736                  */                               
737                 ret = btrfs_wait_ordered_range    
738                                                   
739                 if (ret)                          
740                         return ret;               
741         }                                         
742                                                   
743         /*                                        
744          * Lock destination range to serialize    
745          * we are safe from concurrency with r    
746          * because we have already locked the     
747          * mode.                                  
748          */                                       
749         end = destoff + len - 1;                  
750         lock_extent(&BTRFS_I(inode)->io_tree,     
751         ret = btrfs_clone(src, inode, off, ole    
752         unlock_extent(&BTRFS_I(inode)->io_tree    
753                                                   
754         /*                                        
755          * We may have copied an inline extent    
756          * range, so wait for writeback to com    
757          * from the page cache. This is a rare    
758          */                                       
759         wb_ret = btrfs_wait_ordered_range(BTRF    
760         ret = ret ? ret : wb_ret;                 
761         /*                                        
762          * Truncate page cache pages so that f    
763          * data immediately and not the previo    
764          */                                       
765         truncate_inode_pages_range(&inode->i_d    
766                                 round_down(des    
767                                 round_up(desto    
768                                                   
769         btrfs_btree_balance_dirty(fs_info);       
770                                                   
771         return ret;                               
772 }                                                 
773                                                   
774 static int btrfs_remap_file_range_prep(struct     
775                                        struct     
776                                        loff_t     
777 {                                                 
778         struct inode *inode_in = file_inode(fi    
779         struct inode *inode_out = file_inode(f    
780         u64 bs = BTRFS_I(inode_out)->root->fs_    
781         u64 wb_len;                               
782         int ret;                                  
783                                                   
784         if (!(remap_flags & REMAP_FILE_DEDUP))    
785                 struct btrfs_root *root_out =     
786                                                   
787                 if (btrfs_root_readonly(root_o    
788                         return -EROFS;            
789                                                   
790                 ASSERT(inode_in->i_sb == inode    
791         }                                         
792                                                   
793         /* Don't make the dst file partly chec    
794         if ((BTRFS_I(inode_in)->flags & BTRFS_    
795             (BTRFS_I(inode_out)->flags & BTRFS    
796                 return -EINVAL;                   
797         }                                         
798                                                   
799         /*                                        
800          * Now that the inodes are locked, we     
801          * and can not rely on the writeback f    
802          * generic_remap_file_range_prep() bec    
803          *                                        
804          * 1) For compression we must call fil    
805          *    twice (btrfs_fdatawrite_range()     
806          *    helper only calls it once;          
807          *                                        
808          * 2) filemap_fdatawrite_range(), call    
809          *    waits for the writeback to compl    
810          *    not for the ordered extents to c    
811          *    to complete so that new file ext    
812          */                                       
813         if (*len == 0 && !(remap_flags & REMAP    
814                 wb_len = ALIGN(inode_in->i_siz    
815         else                                      
816                 wb_len = ALIGN(*len, bs);         
817                                                   
818         /*                                        
819          * Workaround to make sure NOCOW buffe    
820          *                                        
821          * Btrfs' back references do not have     
822          * work at the whole extent level.        
823          * NOCOW buffered write without data s    
824          * to fall back to CoW due to lack of     
825          * data loss.                             
826          *                                        
827          * Here we take a shortcut by flushing    
828          * nocow write should reach disk as no    
829          * reference of the extent. We could d    
830          * data, but that needs extra accounti    
831          *                                        
832          * Also we don't need to check ASYNC_E    
833          * CoWed anyway, not affecting nocow p    
834          */                                       
835         ret = filemap_flush(inode_in->i_mappin    
836         if (ret < 0)                              
837                 return ret;                       
838                                                   
839         ret = btrfs_wait_ordered_range(BTRFS_I    
840                                        wb_len)    
841         if (ret < 0)                              
842                 return ret;                       
843         ret = btrfs_wait_ordered_range(BTRFS_I    
844                                        wb_len)    
845         if (ret < 0)                              
846                 return ret;                       
847                                                   
848         return generic_remap_file_range_prep(f    
849                                             le    
850 }                                                 
851                                                   
852 static bool file_sync_write(const struct file     
853 {                                                 
854         if (file->f_flags & (__O_SYNC | O_DSYN    
855                 return true;                      
856         if (IS_SYNC(file_inode(file)))            
857                 return true;                      
858                                                   
859         return false;                             
860 }                                                 
861                                                   
862 loff_t btrfs_remap_file_range(struct file *src    
863                 struct file *dst_file, loff_t     
864                 unsigned int remap_flags)         
865 {                                                 
866         struct inode *src_inode = file_inode(s    
867         struct inode *dst_inode = file_inode(d    
868         bool same_inode = dst_inode == src_ino    
869         int ret;                                  
870                                                   
871         if (remap_flags & ~(REMAP_FILE_DEDUP |    
872                 return -EINVAL;                   
873                                                   
874         if (same_inode) {                         
875                 btrfs_inode_lock(BTRFS_I(src_i    
876         } else {                                  
877                 lock_two_nondirectories(src_in    
878                 btrfs_double_mmap_lock(src_ino    
879         }                                         
880                                                   
881         ret = btrfs_remap_file_range_prep(src_    
882                                           &len    
883         if (ret < 0 || len == 0)                  
884                 goto out_unlock;                  
885                                                   
886         if (remap_flags & REMAP_FILE_DEDUP)       
887                 ret = btrfs_extent_same(src_in    
888         else                                      
889                 ret = btrfs_clone_files(dst_fi    
890                                                   
891 out_unlock:                                       
892         if (same_inode) {                         
893                 btrfs_inode_unlock(BTRFS_I(src    
894         } else {                                  
895                 btrfs_double_mmap_unlock(src_i    
896                 unlock_two_nondirectories(src_    
897         }                                         
898                                                   
899         /*                                        
900          * If either the source or the destina    
901          * O_DSYNC or has the S_SYNC attribute    
902          * source files/ranges, so that after     
903          * by a power failure results in the r    
904          * both files/ranges.                     
905          */                                       
906         if (ret == 0 && len > 0 &&                
907             (file_sync_write(src_file) || file    
908                 ret = btrfs_sync_file(src_file    
909                 if (ret == 0)                     
910                         ret = btrfs_sync_file(    
911                                                   
912         }                                         
913                                                   
914         return ret < 0 ? ret : len;               
915 }                                                 
916                                                   

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

kernel.org | git.kernel.org | LWN.net | Project Home | 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.

sflogo.php