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

TOMOYO Linux Cross Reference
Linux/fs/ext4/extents.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
  4  * Written by Alex Tomas <alex@clusterfs.com>
  5  *
  6  * Architecture independence:
  7  *   Copyright (c) 2005, Bull S.A.
  8  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
  9  */
 10 
 11 /*
 12  * Extents support for EXT4
 13  *
 14  * TODO:
 15  *   - ext4*_error() should be used in some situations
 16  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
 17  *   - smart tree reduction
 18  */
 19 
 20 #include <linux/fs.h>
 21 #include <linux/time.h>
 22 #include <linux/jbd2.h>
 23 #include <linux/highuid.h>
 24 #include <linux/pagemap.h>
 25 #include <linux/quotaops.h>
 26 #include <linux/string.h>
 27 #include <linux/slab.h>
 28 #include <linux/uaccess.h>
 29 #include <linux/fiemap.h>
 30 #include <linux/backing-dev.h>
 31 #include "ext4_jbd2.h"
 32 #include "ext4_extents.h"
 33 #include "xattr.h"
 34 
 35 #include <trace/events/ext4.h>
 36 
 37 /*
 38  * used by extent splitting.
 39  */
 40 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
 41                                         due to ENOSPC */
 42 #define EXT4_EXT_MARK_UNWRIT1   0x2  /* mark first half unwritten */
 43 #define EXT4_EXT_MARK_UNWRIT2   0x4  /* mark second half unwritten */
 44 
 45 #define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
 46 #define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
 47 
 48 static __le32 ext4_extent_block_csum(struct inode *inode,
 49                                      struct ext4_extent_header *eh)
 50 {
 51         struct ext4_inode_info *ei = EXT4_I(inode);
 52         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 53         __u32 csum;
 54 
 55         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
 56                            EXT4_EXTENT_TAIL_OFFSET(eh));
 57         return cpu_to_le32(csum);
 58 }
 59 
 60 static int ext4_extent_block_csum_verify(struct inode *inode,
 61                                          struct ext4_extent_header *eh)
 62 {
 63         struct ext4_extent_tail *et;
 64 
 65         if (!ext4_has_metadata_csum(inode->i_sb))
 66                 return 1;
 67 
 68         et = find_ext4_extent_tail(eh);
 69         if (et->et_checksum != ext4_extent_block_csum(inode, eh))
 70                 return 0;
 71         return 1;
 72 }
 73 
 74 static void ext4_extent_block_csum_set(struct inode *inode,
 75                                        struct ext4_extent_header *eh)
 76 {
 77         struct ext4_extent_tail *et;
 78 
 79         if (!ext4_has_metadata_csum(inode->i_sb))
 80                 return;
 81 
 82         et = find_ext4_extent_tail(eh);
 83         et->et_checksum = ext4_extent_block_csum(inode, eh);
 84 }
 85 
 86 static int ext4_split_extent(handle_t *handle,
 87                                 struct inode *inode,
 88                                 struct ext4_ext_path **ppath,
 89                                 struct ext4_map_blocks *map,
 90                                 int split_flag,
 91                                 int flags);
 92 
 93 static int ext4_split_extent_at(handle_t *handle,
 94                              struct inode *inode,
 95                              struct ext4_ext_path **ppath,
 96                              ext4_lblk_t split,
 97                              int split_flag,
 98                              int flags);
 99 
100 static int ext4_find_delayed_extent(struct inode *inode,
101                                     struct extent_status *newes);
102 
103 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
104 {
105         /*
106          * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
107          * moment, get_block can be called only for blocks inside i_size since
108          * page cache has been already dropped and writes are blocked by
109          * i_mutex. So we can safely drop the i_data_sem here.
110          */
111         BUG_ON(EXT4_JOURNAL(inode) == NULL);
112         ext4_discard_preallocations(inode);
113         up_write(&EXT4_I(inode)->i_data_sem);
114         *dropped = 1;
115         return 0;
116 }
117 
118 /*
119  * Make sure 'handle' has at least 'check_cred' credits. If not, restart
120  * transaction with 'restart_cred' credits. The function drops i_data_sem
121  * when restarting transaction and gets it after transaction is restarted.
122  *
123  * The function returns 0 on success, 1 if transaction had to be restarted,
124  * and < 0 in case of fatal error.
125  */
126 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
127                                 int check_cred, int restart_cred,
128                                 int revoke_cred)
129 {
130         int ret;
131         int dropped = 0;
132 
133         ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
134                 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
135         if (dropped)
136                 down_write(&EXT4_I(inode)->i_data_sem);
137         return ret;
138 }
139 
140 /*
141  * could return:
142  *  - EROFS
143  *  - ENOMEM
144  */
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146                                 struct ext4_ext_path *path)
147 {
148         if (path->p_bh) {
149                 /* path points to block */
150                 BUFFER_TRACE(path->p_bh, "get_write_access");
151                 return ext4_journal_get_write_access(handle, path->p_bh);
152         }
153         /* path points to leaf/index in inode body */
154         /* we use in-core data, no need to protect them */
155         return 0;
156 }
157 
158 /*
159  * could return:
160  *  - EROFS
161  *  - ENOMEM
162  *  - EIO
163  */
164 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
165                      struct inode *inode, struct ext4_ext_path *path)
166 {
167         int err;
168 
169         WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
170         if (path->p_bh) {
171                 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
172                 /* path points to block */
173                 err = __ext4_handle_dirty_metadata(where, line, handle,
174                                                    inode, path->p_bh);
175         } else {
176                 /* path points to leaf/index in inode body */
177                 err = ext4_mark_inode_dirty(handle, inode);
178         }
179         return err;
180 }
181 
182 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
183                               struct ext4_ext_path *path,
184                               ext4_lblk_t block)
185 {
186         if (path) {
187                 int depth = path->p_depth;
188                 struct ext4_extent *ex;
189 
190                 /*
191                  * Try to predict block placement assuming that we are
192                  * filling in a file which will eventually be
193                  * non-sparse --- i.e., in the case of libbfd writing
194                  * an ELF object sections out-of-order but in a way
195                  * the eventually results in a contiguous object or
196                  * executable file, or some database extending a table
197                  * space file.  However, this is actually somewhat
198                  * non-ideal if we are writing a sparse file such as
199                  * qemu or KVM writing a raw image file that is going
200                  * to stay fairly sparse, since it will end up
201                  * fragmenting the file system's free space.  Maybe we
202                  * should have some hueristics or some way to allow
203                  * userspace to pass a hint to file system,
204                  * especially if the latter case turns out to be
205                  * common.
206                  */
207                 ex = path[depth].p_ext;
208                 if (ex) {
209                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
210                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
211 
212                         if (block > ext_block)
213                                 return ext_pblk + (block - ext_block);
214                         else
215                                 return ext_pblk - (ext_block - block);
216                 }
217 
218                 /* it looks like index is empty;
219                  * try to find starting block from index itself */
220                 if (path[depth].p_bh)
221                         return path[depth].p_bh->b_blocknr;
222         }
223 
224         /* OK. use inode's group */
225         return ext4_inode_to_goal_block(inode);
226 }
227 
228 /*
229  * Allocation for a meta data block
230  */
231 static ext4_fsblk_t
232 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
233                         struct ext4_ext_path *path,
234                         struct ext4_extent *ex, int *err, unsigned int flags)
235 {
236         ext4_fsblk_t goal, newblock;
237 
238         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
239         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
240                                         NULL, err);
241         return newblock;
242 }
243 
244 static inline int ext4_ext_space_block(struct inode *inode, int check)
245 {
246         int size;
247 
248         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
249                         / sizeof(struct ext4_extent);
250 #ifdef AGGRESSIVE_TEST
251         if (!check && size > 6)
252                 size = 6;
253 #endif
254         return size;
255 }
256 
257 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
258 {
259         int size;
260 
261         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
262                         / sizeof(struct ext4_extent_idx);
263 #ifdef AGGRESSIVE_TEST
264         if (!check && size > 5)
265                 size = 5;
266 #endif
267         return size;
268 }
269 
270 static inline int ext4_ext_space_root(struct inode *inode, int check)
271 {
272         int size;
273 
274         size = sizeof(EXT4_I(inode)->i_data);
275         size -= sizeof(struct ext4_extent_header);
276         size /= sizeof(struct ext4_extent);
277 #ifdef AGGRESSIVE_TEST
278         if (!check && size > 3)
279                 size = 3;
280 #endif
281         return size;
282 }
283 
284 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
285 {
286         int size;
287 
288         size = sizeof(EXT4_I(inode)->i_data);
289         size -= sizeof(struct ext4_extent_header);
290         size /= sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292         if (!check && size > 4)
293                 size = 4;
294 #endif
295         return size;
296 }
297 
298 static inline int
299 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
300                            struct ext4_ext_path **ppath, ext4_lblk_t lblk,
301                            int nofail)
302 {
303         struct ext4_ext_path *path = *ppath;
304         int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
305 
306         return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
307                         EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
308                         EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
309                         (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
310 }
311 
312 /*
313  * Calculate the number of metadata blocks needed
314  * to allocate @blocks
315  * Worse case is one block per extent
316  */
317 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
318 {
319         struct ext4_inode_info *ei = EXT4_I(inode);
320         int idxs;
321 
322         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
323                 / sizeof(struct ext4_extent_idx));
324 
325         /*
326          * If the new delayed allocation block is contiguous with the
327          * previous da block, it can share index blocks with the
328          * previous block, so we only need to allocate a new index
329          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
330          * an additional index block, and at ldxs**3 blocks, yet
331          * another index blocks.
332          */
333         if (ei->i_da_metadata_calc_len &&
334             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
335                 int num = 0;
336 
337                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
338                         num++;
339                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
340                         num++;
341                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
342                         num++;
343                         ei->i_da_metadata_calc_len = 0;
344                 } else
345                         ei->i_da_metadata_calc_len++;
346                 ei->i_da_metadata_calc_last_lblock++;
347                 return num;
348         }
349 
350         /*
351          * In the worst case we need a new set of index blocks at
352          * every level of the inode's extent tree.
353          */
354         ei->i_da_metadata_calc_len = 1;
355         ei->i_da_metadata_calc_last_lblock = lblock;
356         return ext_depth(inode) + 1;
357 }
358 
359 static int
360 ext4_ext_max_entries(struct inode *inode, int depth)
361 {
362         int max;
363 
364         if (depth == ext_depth(inode)) {
365                 if (depth == 0)
366                         max = ext4_ext_space_root(inode, 1);
367                 else
368                         max = ext4_ext_space_root_idx(inode, 1);
369         } else {
370                 if (depth == 0)
371                         max = ext4_ext_space_block(inode, 1);
372                 else
373                         max = ext4_ext_space_block_idx(inode, 1);
374         }
375 
376         return max;
377 }
378 
379 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
380 {
381         ext4_fsblk_t block = ext4_ext_pblock(ext);
382         int len = ext4_ext_get_actual_len(ext);
383         ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
384 
385         /*
386          * We allow neither:
387          *  - zero length
388          *  - overflow/wrap-around
389          */
390         if (lblock + len <= lblock)
391                 return 0;
392         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
393 }
394 
395 static int ext4_valid_extent_idx(struct inode *inode,
396                                 struct ext4_extent_idx *ext_idx)
397 {
398         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
399 
400         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
401 }
402 
403 static int ext4_valid_extent_entries(struct inode *inode,
404                                 struct ext4_extent_header *eh,
405                                 int depth)
406 {
407         unsigned short entries;
408         if (eh->eh_entries == 0)
409                 return 1;
410 
411         entries = le16_to_cpu(eh->eh_entries);
412 
413         if (depth == 0) {
414                 /* leaf entries */
415                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
416                 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
417                 ext4_fsblk_t pblock = 0;
418                 ext4_lblk_t lblock = 0;
419                 ext4_lblk_t prev = 0;
420                 int len = 0;
421                 while (entries) {
422                         if (!ext4_valid_extent(inode, ext))
423                                 return 0;
424 
425                         /* Check for overlapping extents */
426                         lblock = le32_to_cpu(ext->ee_block);
427                         len = ext4_ext_get_actual_len(ext);
428                         if ((lblock <= prev) && prev) {
429                                 pblock = ext4_ext_pblock(ext);
430                                 es->s_last_error_block = cpu_to_le64(pblock);
431                                 return 0;
432                         }
433                         ext++;
434                         entries--;
435                         prev = lblock + len - 1;
436                 }
437         } else {
438                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
439                 while (entries) {
440                         if (!ext4_valid_extent_idx(inode, ext_idx))
441                                 return 0;
442                         ext_idx++;
443                         entries--;
444                 }
445         }
446         return 1;
447 }
448 
449 static int __ext4_ext_check(const char *function, unsigned int line,
450                             struct inode *inode, struct ext4_extent_header *eh,
451                             int depth, ext4_fsblk_t pblk)
452 {
453         const char *error_msg;
454         int max = 0, err = -EFSCORRUPTED;
455 
456         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
457                 error_msg = "invalid magic";
458                 goto corrupted;
459         }
460         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
461                 error_msg = "unexpected eh_depth";
462                 goto corrupted;
463         }
464         if (unlikely(eh->eh_max == 0)) {
465                 error_msg = "invalid eh_max";
466                 goto corrupted;
467         }
468         max = ext4_ext_max_entries(inode, depth);
469         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
470                 error_msg = "too large eh_max";
471                 goto corrupted;
472         }
473         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
474                 error_msg = "invalid eh_entries";
475                 goto corrupted;
476         }
477         if (!ext4_valid_extent_entries(inode, eh, depth)) {
478                 error_msg = "invalid extent entries";
479                 goto corrupted;
480         }
481         if (unlikely(depth > 32)) {
482                 error_msg = "too large eh_depth";
483                 goto corrupted;
484         }
485         /* Verify checksum on non-root extent tree nodes */
486         if (ext_depth(inode) != depth &&
487             !ext4_extent_block_csum_verify(inode, eh)) {
488                 error_msg = "extent tree corrupted";
489                 err = -EFSBADCRC;
490                 goto corrupted;
491         }
492         return 0;
493 
494 corrupted:
495         ext4_error_inode(inode, function, line, 0,
496                          "pblk %llu bad header/extent: %s - magic %x, "
497                          "entries %u, max %u(%u), depth %u(%u)",
498                          (unsigned long long) pblk, error_msg,
499                          le16_to_cpu(eh->eh_magic),
500                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
501                          max, le16_to_cpu(eh->eh_depth), depth);
502         return err;
503 }
504 
505 #define ext4_ext_check(inode, eh, depth, pblk)                  \
506         __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
507 
508 int ext4_ext_check_inode(struct inode *inode)
509 {
510         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
511 }
512 
513 static struct buffer_head *
514 __read_extent_tree_block(const char *function, unsigned int line,
515                          struct inode *inode, ext4_fsblk_t pblk, int depth,
516                          int flags)
517 {
518         struct buffer_head              *bh;
519         int                             err;
520 
521         bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
522         if (unlikely(!bh))
523                 return ERR_PTR(-ENOMEM);
524 
525         if (!bh_uptodate_or_lock(bh)) {
526                 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
527                 err = bh_submit_read(bh);
528                 if (err < 0)
529                         goto errout;
530         }
531         if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
532                 return bh;
533         if (!ext4_has_feature_journal(inode->i_sb) ||
534             (inode->i_ino !=
535              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
536                 err = __ext4_ext_check(function, line, inode,
537                                        ext_block_hdr(bh), depth, pblk);
538                 if (err)
539                         goto errout;
540         }
541         set_buffer_verified(bh);
542         /*
543          * If this is a leaf block, cache all of its entries
544          */
545         if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
546                 struct ext4_extent_header *eh = ext_block_hdr(bh);
547                 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
548                 ext4_lblk_t prev = 0;
549                 int i;
550 
551                 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
552                         unsigned int status = EXTENT_STATUS_WRITTEN;
553                         ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
554                         int len = ext4_ext_get_actual_len(ex);
555 
556                         if (prev && (prev != lblk))
557                                 ext4_es_cache_extent(inode, prev,
558                                                      lblk - prev, ~0,
559                                                      EXTENT_STATUS_HOLE);
560 
561                         if (ext4_ext_is_unwritten(ex))
562                                 status = EXTENT_STATUS_UNWRITTEN;
563                         ext4_es_cache_extent(inode, lblk, len,
564                                              ext4_ext_pblock(ex), status);
565                         prev = lblk + len;
566                 }
567         }
568         return bh;
569 errout:
570         put_bh(bh);
571         return ERR_PTR(err);
572 
573 }
574 
575 #define read_extent_tree_block(inode, pblk, depth, flags)               \
576         __read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
577                                  (depth), (flags))
578 
579 /*
580  * This function is called to cache a file's extent information in the
581  * extent status tree
582  */
583 int ext4_ext_precache(struct inode *inode)
584 {
585         struct ext4_inode_info *ei = EXT4_I(inode);
586         struct ext4_ext_path *path = NULL;
587         struct buffer_head *bh;
588         int i = 0, depth, ret = 0;
589 
590         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
591                 return 0;       /* not an extent-mapped inode */
592 
593         down_read(&ei->i_data_sem);
594         depth = ext_depth(inode);
595 
596         path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
597                        GFP_NOFS);
598         if (path == NULL) {
599                 up_read(&ei->i_data_sem);
600                 return -ENOMEM;
601         }
602 
603         /* Don't cache anything if there are no external extent blocks */
604         if (depth == 0)
605                 goto out;
606         path[0].p_hdr = ext_inode_hdr(inode);
607         ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
608         if (ret)
609                 goto out;
610         path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
611         while (i >= 0) {
612                 /*
613                  * If this is a leaf block or we've reached the end of
614                  * the index block, go up
615                  */
616                 if ((i == depth) ||
617                     path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
618                         brelse(path[i].p_bh);
619                         path[i].p_bh = NULL;
620                         i--;
621                         continue;
622                 }
623                 bh = read_extent_tree_block(inode,
624                                             ext4_idx_pblock(path[i].p_idx++),
625                                             depth - i - 1,
626                                             EXT4_EX_FORCE_CACHE);
627                 if (IS_ERR(bh)) {
628                         ret = PTR_ERR(bh);
629                         break;
630                 }
631                 i++;
632                 path[i].p_bh = bh;
633                 path[i].p_hdr = ext_block_hdr(bh);
634                 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
635         }
636         ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
637 out:
638         up_read(&ei->i_data_sem);
639         ext4_ext_drop_refs(path);
640         kfree(path);
641         return ret;
642 }
643 
644 #ifdef EXT_DEBUG
645 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
646 {
647         int k, l = path->p_depth;
648 
649         ext_debug("path:");
650         for (k = 0; k <= l; k++, path++) {
651                 if (path->p_idx) {
652                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
653                             ext4_idx_pblock(path->p_idx));
654                 } else if (path->p_ext) {
655                         ext_debug("  %d:[%d]%d:%llu ",
656                                   le32_to_cpu(path->p_ext->ee_block),
657                                   ext4_ext_is_unwritten(path->p_ext),
658                                   ext4_ext_get_actual_len(path->p_ext),
659                                   ext4_ext_pblock(path->p_ext));
660                 } else
661                         ext_debug("  []");
662         }
663         ext_debug("\n");
664 }
665 
666 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
667 {
668         int depth = ext_depth(inode);
669         struct ext4_extent_header *eh;
670         struct ext4_extent *ex;
671         int i;
672 
673         if (!path)
674                 return;
675 
676         eh = path[depth].p_hdr;
677         ex = EXT_FIRST_EXTENT(eh);
678 
679         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
680 
681         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
682                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
683                           ext4_ext_is_unwritten(ex),
684                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
685         }
686         ext_debug("\n");
687 }
688 
689 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
690                         ext4_fsblk_t newblock, int level)
691 {
692         int depth = ext_depth(inode);
693         struct ext4_extent *ex;
694 
695         if (depth != level) {
696                 struct ext4_extent_idx *idx;
697                 idx = path[level].p_idx;
698                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
699                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
700                                         le32_to_cpu(idx->ei_block),
701                                         ext4_idx_pblock(idx),
702                                         newblock);
703                         idx++;
704                 }
705 
706                 return;
707         }
708 
709         ex = path[depth].p_ext;
710         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
711                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
712                                 le32_to_cpu(ex->ee_block),
713                                 ext4_ext_pblock(ex),
714                                 ext4_ext_is_unwritten(ex),
715                                 ext4_ext_get_actual_len(ex),
716                                 newblock);
717                 ex++;
718         }
719 }
720 
721 #else
722 #define ext4_ext_show_path(inode, path)
723 #define ext4_ext_show_leaf(inode, path)
724 #define ext4_ext_show_move(inode, path, newblock, level)
725 #endif
726 
727 void ext4_ext_drop_refs(struct ext4_ext_path *path)
728 {
729         int depth, i;
730 
731         if (!path)
732                 return;
733         depth = path->p_depth;
734         for (i = 0; i <= depth; i++, path++)
735                 if (path->p_bh) {
736                         brelse(path->p_bh);
737                         path->p_bh = NULL;
738                 }
739 }
740 
741 /*
742  * ext4_ext_binsearch_idx:
743  * binary search for the closest index of the given block
744  * the header must be checked before calling this
745  */
746 static void
747 ext4_ext_binsearch_idx(struct inode *inode,
748                         struct ext4_ext_path *path, ext4_lblk_t block)
749 {
750         struct ext4_extent_header *eh = path->p_hdr;
751         struct ext4_extent_idx *r, *l, *m;
752 
753 
754         ext_debug("binsearch for %u(idx):  ", block);
755 
756         l = EXT_FIRST_INDEX(eh) + 1;
757         r = EXT_LAST_INDEX(eh);
758         while (l <= r) {
759                 m = l + (r - l) / 2;
760                 if (block < le32_to_cpu(m->ei_block))
761                         r = m - 1;
762                 else
763                         l = m + 1;
764                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
765                                 m, le32_to_cpu(m->ei_block),
766                                 r, le32_to_cpu(r->ei_block));
767         }
768 
769         path->p_idx = l - 1;
770         ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
771                   ext4_idx_pblock(path->p_idx));
772 
773 #ifdef CHECK_BINSEARCH
774         {
775                 struct ext4_extent_idx *chix, *ix;
776                 int k;
777 
778                 chix = ix = EXT_FIRST_INDEX(eh);
779                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
780                   if (k != 0 &&
781                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
782                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
783                                        "first=0x%p\n", k,
784                                        ix, EXT_FIRST_INDEX(eh));
785                                 printk(KERN_DEBUG "%u <= %u\n",
786                                        le32_to_cpu(ix->ei_block),
787                                        le32_to_cpu(ix[-1].ei_block));
788                         }
789                         BUG_ON(k && le32_to_cpu(ix->ei_block)
790                                            <= le32_to_cpu(ix[-1].ei_block));
791                         if (block < le32_to_cpu(ix->ei_block))
792                                 break;
793                         chix = ix;
794                 }
795                 BUG_ON(chix != path->p_idx);
796         }
797 #endif
798 
799 }
800 
801 /*
802  * ext4_ext_binsearch:
803  * binary search for closest extent of the given block
804  * the header must be checked before calling this
805  */
806 static void
807 ext4_ext_binsearch(struct inode *inode,
808                 struct ext4_ext_path *path, ext4_lblk_t block)
809 {
810         struct ext4_extent_header *eh = path->p_hdr;
811         struct ext4_extent *r, *l, *m;
812 
813         if (eh->eh_entries == 0) {
814                 /*
815                  * this leaf is empty:
816                  * we get such a leaf in split/add case
817                  */
818                 return;
819         }
820 
821         ext_debug("binsearch for %u:  ", block);
822 
823         l = EXT_FIRST_EXTENT(eh) + 1;
824         r = EXT_LAST_EXTENT(eh);
825 
826         while (l <= r) {
827                 m = l + (r - l) / 2;
828                 if (block < le32_to_cpu(m->ee_block))
829                         r = m - 1;
830                 else
831                         l = m + 1;
832                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
833                                 m, le32_to_cpu(m->ee_block),
834                                 r, le32_to_cpu(r->ee_block));
835         }
836 
837         path->p_ext = l - 1;
838         ext_debug("  -> %d:%llu:[%d]%d ",
839                         le32_to_cpu(path->p_ext->ee_block),
840                         ext4_ext_pblock(path->p_ext),
841                         ext4_ext_is_unwritten(path->p_ext),
842                         ext4_ext_get_actual_len(path->p_ext));
843 
844 #ifdef CHECK_BINSEARCH
845         {
846                 struct ext4_extent *chex, *ex;
847                 int k;
848 
849                 chex = ex = EXT_FIRST_EXTENT(eh);
850                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
851                         BUG_ON(k && le32_to_cpu(ex->ee_block)
852                                           <= le32_to_cpu(ex[-1].ee_block));
853                         if (block < le32_to_cpu(ex->ee_block))
854                                 break;
855                         chex = ex;
856                 }
857                 BUG_ON(chex != path->p_ext);
858         }
859 #endif
860 
861 }
862 
863 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
864 {
865         struct ext4_extent_header *eh;
866 
867         eh = ext_inode_hdr(inode);
868         eh->eh_depth = 0;
869         eh->eh_entries = 0;
870         eh->eh_magic = EXT4_EXT_MAGIC;
871         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
872         ext4_mark_inode_dirty(handle, inode);
873         return 0;
874 }
875 
876 struct ext4_ext_path *
877 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
878                  struct ext4_ext_path **orig_path, int flags)
879 {
880         struct ext4_extent_header *eh;
881         struct buffer_head *bh;
882         struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
883         short int depth, i, ppos = 0;
884         int ret;
885 
886         eh = ext_inode_hdr(inode);
887         depth = ext_depth(inode);
888         if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
889                 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
890                                  depth);
891                 ret = -EFSCORRUPTED;
892                 goto err;
893         }
894 
895         if (path) {
896                 ext4_ext_drop_refs(path);
897                 if (depth > path[0].p_maxdepth) {
898                         kfree(path);
899                         *orig_path = path = NULL;
900                 }
901         }
902         if (!path) {
903                 /* account possible depth increase */
904                 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
905                                 GFP_NOFS);
906                 if (unlikely(!path))
907                         return ERR_PTR(-ENOMEM);
908                 path[0].p_maxdepth = depth + 1;
909         }
910         path[0].p_hdr = eh;
911         path[0].p_bh = NULL;
912 
913         i = depth;
914         /* walk through the tree */
915         while (i) {
916                 ext_debug("depth %d: num %d, max %d\n",
917                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
918 
919                 ext4_ext_binsearch_idx(inode, path + ppos, block);
920                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
921                 path[ppos].p_depth = i;
922                 path[ppos].p_ext = NULL;
923 
924                 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
925                                             flags);
926                 if (IS_ERR(bh)) {
927                         ret = PTR_ERR(bh);
928                         goto err;
929                 }
930 
931                 eh = ext_block_hdr(bh);
932                 ppos++;
933                 path[ppos].p_bh = bh;
934                 path[ppos].p_hdr = eh;
935         }
936 
937         path[ppos].p_depth = i;
938         path[ppos].p_ext = NULL;
939         path[ppos].p_idx = NULL;
940 
941         /* find extent */
942         ext4_ext_binsearch(inode, path + ppos, block);
943         /* if not an empty leaf */
944         if (path[ppos].p_ext)
945                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
946 
947         ext4_ext_show_path(inode, path);
948 
949         return path;
950 
951 err:
952         ext4_ext_drop_refs(path);
953         kfree(path);
954         if (orig_path)
955                 *orig_path = NULL;
956         return ERR_PTR(ret);
957 }
958 
959 /*
960  * ext4_ext_insert_index:
961  * insert new index [@logical;@ptr] into the block at @curp;
962  * check where to insert: before @curp or after @curp
963  */
964 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
965                                  struct ext4_ext_path *curp,
966                                  int logical, ext4_fsblk_t ptr)
967 {
968         struct ext4_extent_idx *ix;
969         int len, err;
970 
971         err = ext4_ext_get_access(handle, inode, curp);
972         if (err)
973                 return err;
974 
975         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
976                 EXT4_ERROR_INODE(inode,
977                                  "logical %d == ei_block %d!",
978                                  logical, le32_to_cpu(curp->p_idx->ei_block));
979                 return -EFSCORRUPTED;
980         }
981 
982         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
983                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
984                 EXT4_ERROR_INODE(inode,
985                                  "eh_entries %d >= eh_max %d!",
986                                  le16_to_cpu(curp->p_hdr->eh_entries),
987                                  le16_to_cpu(curp->p_hdr->eh_max));
988                 return -EFSCORRUPTED;
989         }
990 
991         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
992                 /* insert after */
993                 ext_debug("insert new index %d after: %llu\n", logical, ptr);
994                 ix = curp->p_idx + 1;
995         } else {
996                 /* insert before */
997                 ext_debug("insert new index %d before: %llu\n", logical, ptr);
998                 ix = curp->p_idx;
999         }
1000 
1001         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1002         BUG_ON(len < 0);
1003         if (len > 0) {
1004                 ext_debug("insert new index %d: "
1005                                 "move %d indices from 0x%p to 0x%p\n",
1006                                 logical, len, ix, ix + 1);
1007                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1008         }
1009 
1010         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1011                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1012                 return -EFSCORRUPTED;
1013         }
1014 
1015         ix->ei_block = cpu_to_le32(logical);
1016         ext4_idx_store_pblock(ix, ptr);
1017         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1018 
1019         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1020                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1021                 return -EFSCORRUPTED;
1022         }
1023 
1024         err = ext4_ext_dirty(handle, inode, curp);
1025         ext4_std_error(inode->i_sb, err);
1026 
1027         return err;
1028 }
1029 
1030 /*
1031  * ext4_ext_split:
1032  * inserts new subtree into the path, using free index entry
1033  * at depth @at:
1034  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1035  * - makes decision where to split
1036  * - moves remaining extents and index entries (right to the split point)
1037  *   into the newly allocated blocks
1038  * - initializes subtree
1039  */
1040 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1041                           unsigned int flags,
1042                           struct ext4_ext_path *path,
1043                           struct ext4_extent *newext, int at)
1044 {
1045         struct buffer_head *bh = NULL;
1046         int depth = ext_depth(inode);
1047         struct ext4_extent_header *neh;
1048         struct ext4_extent_idx *fidx;
1049         int i = at, k, m, a;
1050         ext4_fsblk_t newblock, oldblock;
1051         __le32 border;
1052         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1053         int err = 0;
1054         size_t ext_size = 0;
1055 
1056         /* make decision: where to split? */
1057         /* FIXME: now decision is simplest: at current extent */
1058 
1059         /* if current leaf will be split, then we should use
1060          * border from split point */
1061         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1062                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1063                 return -EFSCORRUPTED;
1064         }
1065         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1066                 border = path[depth].p_ext[1].ee_block;
1067                 ext_debug("leaf will be split."
1068                                 " next leaf starts at %d\n",
1069                                   le32_to_cpu(border));
1070         } else {
1071                 border = newext->ee_block;
1072                 ext_debug("leaf will be added."
1073                                 " next leaf starts at %d\n",
1074                                 le32_to_cpu(border));
1075         }
1076 
1077         /*
1078          * If error occurs, then we break processing
1079          * and mark filesystem read-only. index won't
1080          * be inserted and tree will be in consistent
1081          * state. Next mount will repair buffers too.
1082          */
1083 
1084         /*
1085          * Get array to track all allocated blocks.
1086          * We need this to handle errors and free blocks
1087          * upon them.
1088          */
1089         ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1090         if (!ablocks)
1091                 return -ENOMEM;
1092 
1093         /* allocate all needed blocks */
1094         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1095         for (a = 0; a < depth - at; a++) {
1096                 newblock = ext4_ext_new_meta_block(handle, inode, path,
1097                                                    newext, &err, flags);
1098                 if (newblock == 0)
1099                         goto cleanup;
1100                 ablocks[a] = newblock;
1101         }
1102 
1103         /* initialize new leaf */
1104         newblock = ablocks[--a];
1105         if (unlikely(newblock == 0)) {
1106                 EXT4_ERROR_INODE(inode, "newblock == 0!");
1107                 err = -EFSCORRUPTED;
1108                 goto cleanup;
1109         }
1110         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1111         if (unlikely(!bh)) {
1112                 err = -ENOMEM;
1113                 goto cleanup;
1114         }
1115         lock_buffer(bh);
1116 
1117         err = ext4_journal_get_create_access(handle, bh);
1118         if (err)
1119                 goto cleanup;
1120 
1121         neh = ext_block_hdr(bh);
1122         neh->eh_entries = 0;
1123         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1124         neh->eh_magic = EXT4_EXT_MAGIC;
1125         neh->eh_depth = 0;
1126 
1127         /* move remainder of path[depth] to the new leaf */
1128         if (unlikely(path[depth].p_hdr->eh_entries !=
1129                      path[depth].p_hdr->eh_max)) {
1130                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1131                                  path[depth].p_hdr->eh_entries,
1132                                  path[depth].p_hdr->eh_max);
1133                 err = -EFSCORRUPTED;
1134                 goto cleanup;
1135         }
1136         /* start copy from next extent */
1137         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1138         ext4_ext_show_move(inode, path, newblock, depth);
1139         if (m) {
1140                 struct ext4_extent *ex;
1141                 ex = EXT_FIRST_EXTENT(neh);
1142                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1143                 le16_add_cpu(&neh->eh_entries, m);
1144         }
1145 
1146         /* zero out unused area in the extent block */
1147         ext_size = sizeof(struct ext4_extent_header) +
1148                 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1149         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1150         ext4_extent_block_csum_set(inode, neh);
1151         set_buffer_uptodate(bh);
1152         unlock_buffer(bh);
1153 
1154         err = ext4_handle_dirty_metadata(handle, inode, bh);
1155         if (err)
1156                 goto cleanup;
1157         brelse(bh);
1158         bh = NULL;
1159 
1160         /* correct old leaf */
1161         if (m) {
1162                 err = ext4_ext_get_access(handle, inode, path + depth);
1163                 if (err)
1164                         goto cleanup;
1165                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1166                 err = ext4_ext_dirty(handle, inode, path + depth);
1167                 if (err)
1168                         goto cleanup;
1169 
1170         }
1171 
1172         /* create intermediate indexes */
1173         k = depth - at - 1;
1174         if (unlikely(k < 0)) {
1175                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1176                 err = -EFSCORRUPTED;
1177                 goto cleanup;
1178         }
1179         if (k)
1180                 ext_debug("create %d intermediate indices\n", k);
1181         /* insert new index into current index block */
1182         /* current depth stored in i var */
1183         i = depth - 1;
1184         while (k--) {
1185                 oldblock = newblock;
1186                 newblock = ablocks[--a];
1187                 bh = sb_getblk(inode->i_sb, newblock);
1188                 if (unlikely(!bh)) {
1189                         err = -ENOMEM;
1190                         goto cleanup;
1191                 }
1192                 lock_buffer(bh);
1193 
1194                 err = ext4_journal_get_create_access(handle, bh);
1195                 if (err)
1196                         goto cleanup;
1197 
1198                 neh = ext_block_hdr(bh);
1199                 neh->eh_entries = cpu_to_le16(1);
1200                 neh->eh_magic = EXT4_EXT_MAGIC;
1201                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1202                 neh->eh_depth = cpu_to_le16(depth - i);
1203                 fidx = EXT_FIRST_INDEX(neh);
1204                 fidx->ei_block = border;
1205                 ext4_idx_store_pblock(fidx, oldblock);
1206 
1207                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1208                                 i, newblock, le32_to_cpu(border), oldblock);
1209 
1210                 /* move remainder of path[i] to the new index block */
1211                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1212                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1213                         EXT4_ERROR_INODE(inode,
1214                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1215                                          le32_to_cpu(path[i].p_ext->ee_block));
1216                         err = -EFSCORRUPTED;
1217                         goto cleanup;
1218                 }
1219                 /* start copy indexes */
1220                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1221                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1222                                 EXT_MAX_INDEX(path[i].p_hdr));
1223                 ext4_ext_show_move(inode, path, newblock, i);
1224                 if (m) {
1225                         memmove(++fidx, path[i].p_idx,
1226                                 sizeof(struct ext4_extent_idx) * m);
1227                         le16_add_cpu(&neh->eh_entries, m);
1228                 }
1229                 /* zero out unused area in the extent block */
1230                 ext_size = sizeof(struct ext4_extent_header) +
1231                    (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1232                 memset(bh->b_data + ext_size, 0,
1233                         inode->i_sb->s_blocksize - ext_size);
1234                 ext4_extent_block_csum_set(inode, neh);
1235                 set_buffer_uptodate(bh);
1236                 unlock_buffer(bh);
1237 
1238                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1239                 if (err)
1240                         goto cleanup;
1241                 brelse(bh);
1242                 bh = NULL;
1243 
1244                 /* correct old index */
1245                 if (m) {
1246                         err = ext4_ext_get_access(handle, inode, path + i);
1247                         if (err)
1248                                 goto cleanup;
1249                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1250                         err = ext4_ext_dirty(handle, inode, path + i);
1251                         if (err)
1252                                 goto cleanup;
1253                 }
1254 
1255                 i--;
1256         }
1257 
1258         /* insert new index */
1259         err = ext4_ext_insert_index(handle, inode, path + at,
1260                                     le32_to_cpu(border), newblock);
1261 
1262 cleanup:
1263         if (bh) {
1264                 if (buffer_locked(bh))
1265                         unlock_buffer(bh);
1266                 brelse(bh);
1267         }
1268 
1269         if (err) {
1270                 /* free all allocated blocks in error case */
1271                 for (i = 0; i < depth; i++) {
1272                         if (!ablocks[i])
1273                                 continue;
1274                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1275                                          EXT4_FREE_BLOCKS_METADATA);
1276                 }
1277         }
1278         kfree(ablocks);
1279 
1280         return err;
1281 }
1282 
1283 /*
1284  * ext4_ext_grow_indepth:
1285  * implements tree growing procedure:
1286  * - allocates new block
1287  * - moves top-level data (index block or leaf) into the new block
1288  * - initializes new top-level, creating index that points to the
1289  *   just created block
1290  */
1291 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1292                                  unsigned int flags)
1293 {
1294         struct ext4_extent_header *neh;
1295         struct buffer_head *bh;
1296         ext4_fsblk_t newblock, goal = 0;
1297         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1298         int err = 0;
1299         size_t ext_size = 0;
1300 
1301         /* Try to prepend new index to old one */
1302         if (ext_depth(inode))
1303                 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1304         if (goal > le32_to_cpu(es->s_first_data_block)) {
1305                 flags |= EXT4_MB_HINT_TRY_GOAL;
1306                 goal--;
1307         } else
1308                 goal = ext4_inode_to_goal_block(inode);
1309         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1310                                         NULL, &err);
1311         if (newblock == 0)
1312                 return err;
1313 
1314         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1315         if (unlikely(!bh))
1316                 return -ENOMEM;
1317         lock_buffer(bh);
1318 
1319         err = ext4_journal_get_create_access(handle, bh);
1320         if (err) {
1321                 unlock_buffer(bh);
1322                 goto out;
1323         }
1324 
1325         ext_size = sizeof(EXT4_I(inode)->i_data);
1326         /* move top-level index/leaf into new block */
1327         memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1328         /* zero out unused area in the extent block */
1329         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1330 
1331         /* set size of new block */
1332         neh = ext_block_hdr(bh);
1333         /* old root could have indexes or leaves
1334          * so calculate e_max right way */
1335         if (ext_depth(inode))
1336                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1337         else
1338                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1339         neh->eh_magic = EXT4_EXT_MAGIC;
1340         ext4_extent_block_csum_set(inode, neh);
1341         set_buffer_uptodate(bh);
1342         unlock_buffer(bh);
1343 
1344         err = ext4_handle_dirty_metadata(handle, inode, bh);
1345         if (err)
1346                 goto out;
1347 
1348         /* Update top-level index: num,max,pointer */
1349         neh = ext_inode_hdr(inode);
1350         neh->eh_entries = cpu_to_le16(1);
1351         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1352         if (neh->eh_depth == 0) {
1353                 /* Root extent block becomes index block */
1354                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1355                 EXT_FIRST_INDEX(neh)->ei_block =
1356                         EXT_FIRST_EXTENT(neh)->ee_block;
1357         }
1358         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1359                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1360                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1361                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1362 
1363         le16_add_cpu(&neh->eh_depth, 1);
1364         ext4_mark_inode_dirty(handle, inode);
1365 out:
1366         brelse(bh);
1367 
1368         return err;
1369 }
1370 
1371 /*
1372  * ext4_ext_create_new_leaf:
1373  * finds empty index and adds new leaf.
1374  * if no free index is found, then it requests in-depth growing.
1375  */
1376 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1377                                     unsigned int mb_flags,
1378                                     unsigned int gb_flags,
1379                                     struct ext4_ext_path **ppath,
1380                                     struct ext4_extent *newext)
1381 {
1382         struct ext4_ext_path *path = *ppath;
1383         struct ext4_ext_path *curp;
1384         int depth, i, err = 0;
1385 
1386 repeat:
1387         i = depth = ext_depth(inode);
1388 
1389         /* walk up to the tree and look for free index entry */
1390         curp = path + depth;
1391         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1392                 i--;
1393                 curp--;
1394         }
1395 
1396         /* we use already allocated block for index block,
1397          * so subsequent data blocks should be contiguous */
1398         if (EXT_HAS_FREE_INDEX(curp)) {
1399                 /* if we found index with free entry, then use that
1400                  * entry: create all needed subtree and add new leaf */
1401                 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1402                 if (err)
1403                         goto out;
1404 
1405                 /* refill path */
1406                 path = ext4_find_extent(inode,
1407                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1408                                     ppath, gb_flags);
1409                 if (IS_ERR(path))
1410                         err = PTR_ERR(path);
1411         } else {
1412                 /* tree is full, time to grow in depth */
1413                 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1414                 if (err)
1415                         goto out;
1416 
1417                 /* refill path */
1418                 path = ext4_find_extent(inode,
1419                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1420                                     ppath, gb_flags);
1421                 if (IS_ERR(path)) {
1422                         err = PTR_ERR(path);
1423                         goto out;
1424                 }
1425 
1426                 /*
1427                  * only first (depth 0 -> 1) produces free space;
1428                  * in all other cases we have to split the grown tree
1429                  */
1430                 depth = ext_depth(inode);
1431                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1432                         /* now we need to split */
1433                         goto repeat;
1434                 }
1435         }
1436 
1437 out:
1438         return err;
1439 }
1440 
1441 /*
1442  * search the closest allocated block to the left for *logical
1443  * and returns it at @logical + it's physical address at @phys
1444  * if *logical is the smallest allocated block, the function
1445  * returns 0 at @phys
1446  * return value contains 0 (success) or error code
1447  */
1448 static int ext4_ext_search_left(struct inode *inode,
1449                                 struct ext4_ext_path *path,
1450                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1451 {
1452         struct ext4_extent_idx *ix;
1453         struct ext4_extent *ex;
1454         int depth, ee_len;
1455 
1456         if (unlikely(path == NULL)) {
1457                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1458                 return -EFSCORRUPTED;
1459         }
1460         depth = path->p_depth;
1461         *phys = 0;
1462 
1463         if (depth == 0 && path->p_ext == NULL)
1464                 return 0;
1465 
1466         /* usually extent in the path covers blocks smaller
1467          * then *logical, but it can be that extent is the
1468          * first one in the file */
1469 
1470         ex = path[depth].p_ext;
1471         ee_len = ext4_ext_get_actual_len(ex);
1472         if (*logical < le32_to_cpu(ex->ee_block)) {
1473                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1474                         EXT4_ERROR_INODE(inode,
1475                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1476                                          *logical, le32_to_cpu(ex->ee_block));
1477                         return -EFSCORRUPTED;
1478                 }
1479                 while (--depth >= 0) {
1480                         ix = path[depth].p_idx;
1481                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1482                                 EXT4_ERROR_INODE(inode,
1483                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1484                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1485                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1486                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1487                                   depth);
1488                                 return -EFSCORRUPTED;
1489                         }
1490                 }
1491                 return 0;
1492         }
1493 
1494         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1495                 EXT4_ERROR_INODE(inode,
1496                                  "logical %d < ee_block %d + ee_len %d!",
1497                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1498                 return -EFSCORRUPTED;
1499         }
1500 
1501         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1502         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1503         return 0;
1504 }
1505 
1506 /*
1507  * search the closest allocated block to the right for *logical
1508  * and returns it at @logical + it's physical address at @phys
1509  * if *logical is the largest allocated block, the function
1510  * returns 0 at @phys
1511  * return value contains 0 (success) or error code
1512  */
1513 static int ext4_ext_search_right(struct inode *inode,
1514                                  struct ext4_ext_path *path,
1515                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1516                                  struct ext4_extent **ret_ex)
1517 {
1518         struct buffer_head *bh = NULL;
1519         struct ext4_extent_header *eh;
1520         struct ext4_extent_idx *ix;
1521         struct ext4_extent *ex;
1522         ext4_fsblk_t block;
1523         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1524         int ee_len;
1525 
1526         if (unlikely(path == NULL)) {
1527                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1528                 return -EFSCORRUPTED;
1529         }
1530         depth = path->p_depth;
1531         *phys = 0;
1532 
1533         if (depth == 0 && path->p_ext == NULL)
1534                 return 0;
1535 
1536         /* usually extent in the path covers blocks smaller
1537          * then *logical, but it can be that extent is the
1538          * first one in the file */
1539 
1540         ex = path[depth].p_ext;
1541         ee_len = ext4_ext_get_actual_len(ex);
1542         if (*logical < le32_to_cpu(ex->ee_block)) {
1543                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1544                         EXT4_ERROR_INODE(inode,
1545                                          "first_extent(path[%d].p_hdr) != ex",
1546                                          depth);
1547                         return -EFSCORRUPTED;
1548                 }
1549                 while (--depth >= 0) {
1550                         ix = path[depth].p_idx;
1551                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1552                                 EXT4_ERROR_INODE(inode,
1553                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1554                                                  *logical);
1555                                 return -EFSCORRUPTED;
1556                         }
1557                 }
1558                 goto found_extent;
1559         }
1560 
1561         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1562                 EXT4_ERROR_INODE(inode,
1563                                  "logical %d < ee_block %d + ee_len %d!",
1564                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1565                 return -EFSCORRUPTED;
1566         }
1567 
1568         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1569                 /* next allocated block in this leaf */
1570                 ex++;
1571                 goto found_extent;
1572         }
1573 
1574         /* go up and search for index to the right */
1575         while (--depth >= 0) {
1576                 ix = path[depth].p_idx;
1577                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1578                         goto got_index;
1579         }
1580 
1581         /* we've gone up to the root and found no index to the right */
1582         return 0;
1583 
1584 got_index:
1585         /* we've found index to the right, let's
1586          * follow it and find the closest allocated
1587          * block to the right */
1588         ix++;
1589         block = ext4_idx_pblock(ix);
1590         while (++depth < path->p_depth) {
1591                 /* subtract from p_depth to get proper eh_depth */
1592                 bh = read_extent_tree_block(inode, block,
1593                                             path->p_depth - depth, 0);
1594                 if (IS_ERR(bh))
1595                         return PTR_ERR(bh);
1596                 eh = ext_block_hdr(bh);
1597                 ix = EXT_FIRST_INDEX(eh);
1598                 block = ext4_idx_pblock(ix);
1599                 put_bh(bh);
1600         }
1601 
1602         bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1603         if (IS_ERR(bh))
1604                 return PTR_ERR(bh);
1605         eh = ext_block_hdr(bh);
1606         ex = EXT_FIRST_EXTENT(eh);
1607 found_extent:
1608         *logical = le32_to_cpu(ex->ee_block);
1609         *phys = ext4_ext_pblock(ex);
1610         *ret_ex = ex;
1611         if (bh)
1612                 put_bh(bh);
1613         return 0;
1614 }
1615 
1616 /*
1617  * ext4_ext_next_allocated_block:
1618  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1619  * NOTE: it considers block number from index entry as
1620  * allocated block. Thus, index entries have to be consistent
1621  * with leaves.
1622  */
1623 ext4_lblk_t
1624 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1625 {
1626         int depth;
1627 
1628         BUG_ON(path == NULL);
1629         depth = path->p_depth;
1630 
1631         if (depth == 0 && path->p_ext == NULL)
1632                 return EXT_MAX_BLOCKS;
1633 
1634         while (depth >= 0) {
1635                 if (depth == path->p_depth) {
1636                         /* leaf */
1637                         if (path[depth].p_ext &&
1638                                 path[depth].p_ext !=
1639                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1640                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1641                 } else {
1642                         /* index */
1643                         if (path[depth].p_idx !=
1644                                         EXT_LAST_INDEX(path[depth].p_hdr))
1645                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1646                 }
1647                 depth--;
1648         }
1649 
1650         return EXT_MAX_BLOCKS;
1651 }
1652 
1653 /*
1654  * ext4_ext_next_leaf_block:
1655  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1656  */
1657 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1658 {
1659         int depth;
1660 
1661         BUG_ON(path == NULL);
1662         depth = path->p_depth;
1663 
1664         /* zero-tree has no leaf blocks at all */
1665         if (depth == 0)
1666                 return EXT_MAX_BLOCKS;
1667 
1668         /* go to index block */
1669         depth--;
1670 
1671         while (depth >= 0) {
1672                 if (path[depth].p_idx !=
1673                                 EXT_LAST_INDEX(path[depth].p_hdr))
1674                         return (ext4_lblk_t)
1675                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1676                 depth--;
1677         }
1678 
1679         return EXT_MAX_BLOCKS;
1680 }
1681 
1682 /*
1683  * ext4_ext_correct_indexes:
1684  * if leaf gets modified and modified extent is first in the leaf,
1685  * then we have to correct all indexes above.
1686  * TODO: do we need to correct tree in all cases?
1687  */
1688 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1689                                 struct ext4_ext_path *path)
1690 {
1691         struct ext4_extent_header *eh;
1692         int depth = ext_depth(inode);
1693         struct ext4_extent *ex;
1694         __le32 border;
1695         int k, err = 0;
1696 
1697         eh = path[depth].p_hdr;
1698         ex = path[depth].p_ext;
1699 
1700         if (unlikely(ex == NULL || eh == NULL)) {
1701                 EXT4_ERROR_INODE(inode,
1702                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1703                 return -EFSCORRUPTED;
1704         }
1705 
1706         if (depth == 0) {
1707                 /* there is no tree at all */
1708                 return 0;
1709         }
1710 
1711         if (ex != EXT_FIRST_EXTENT(eh)) {
1712                 /* we correct tree if first leaf got modified only */
1713                 return 0;
1714         }
1715 
1716         /*
1717          * TODO: we need correction if border is smaller than current one
1718          */
1719         k = depth - 1;
1720         border = path[depth].p_ext->ee_block;
1721         err = ext4_ext_get_access(handle, inode, path + k);
1722         if (err)
1723                 return err;
1724         path[k].p_idx->ei_block = border;
1725         err = ext4_ext_dirty(handle, inode, path + k);
1726         if (err)
1727                 return err;
1728 
1729         while (k--) {
1730                 /* change all left-side indexes */
1731                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1732                         break;
1733                 err = ext4_ext_get_access(handle, inode, path + k);
1734                 if (err)
1735                         break;
1736                 path[k].p_idx->ei_block = border;
1737                 err = ext4_ext_dirty(handle, inode, path + k);
1738                 if (err)
1739                         break;
1740         }
1741 
1742         return err;
1743 }
1744 
1745 int
1746 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1747                                 struct ext4_extent *ex2)
1748 {
1749         unsigned short ext1_ee_len, ext2_ee_len;
1750 
1751         if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1752                 return 0;
1753 
1754         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1755         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1756 
1757         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1758                         le32_to_cpu(ex2->ee_block))
1759                 return 0;
1760 
1761         /*
1762          * To allow future support for preallocated extents to be added
1763          * as an RO_COMPAT feature, refuse to merge to extents if
1764          * this can result in the top bit of ee_len being set.
1765          */
1766         if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1767                 return 0;
1768 
1769         if (ext4_ext_is_unwritten(ex1) &&
1770             ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1771                 return 0;
1772 #ifdef AGGRESSIVE_TEST
1773         if (ext1_ee_len >= 4)
1774                 return 0;
1775 #endif
1776 
1777         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1778                 return 1;
1779         return 0;
1780 }
1781 
1782 /*
1783  * This function tries to merge the "ex" extent to the next extent in the tree.
1784  * It always tries to merge towards right. If you want to merge towards
1785  * left, pass "ex - 1" as argument instead of "ex".
1786  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1787  * 1 if they got merged.
1788  */
1789 static int ext4_ext_try_to_merge_right(struct inode *inode,
1790                                  struct ext4_ext_path *path,
1791                                  struct ext4_extent *ex)
1792 {
1793         struct ext4_extent_header *eh;
1794         unsigned int depth, len;
1795         int merge_done = 0, unwritten;
1796 
1797         depth = ext_depth(inode);
1798         BUG_ON(path[depth].p_hdr == NULL);
1799         eh = path[depth].p_hdr;
1800 
1801         while (ex < EXT_LAST_EXTENT(eh)) {
1802                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1803                         break;
1804                 /* merge with next extent! */
1805                 unwritten = ext4_ext_is_unwritten(ex);
1806                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1807                                 + ext4_ext_get_actual_len(ex + 1));
1808                 if (unwritten)
1809                         ext4_ext_mark_unwritten(ex);
1810 
1811                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1812                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1813                                 * sizeof(struct ext4_extent);
1814                         memmove(ex + 1, ex + 2, len);
1815                 }
1816                 le16_add_cpu(&eh->eh_entries, -1);
1817                 merge_done = 1;
1818                 WARN_ON(eh->eh_entries == 0);
1819                 if (!eh->eh_entries)
1820                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1821         }
1822 
1823         return merge_done;
1824 }
1825 
1826 /*
1827  * This function does a very simple check to see if we can collapse
1828  * an extent tree with a single extent tree leaf block into the inode.
1829  */
1830 static void ext4_ext_try_to_merge_up(handle_t *handle,
1831                                      struct inode *inode,
1832                                      struct ext4_ext_path *path)
1833 {
1834         size_t s;
1835         unsigned max_root = ext4_ext_space_root(inode, 0);
1836         ext4_fsblk_t blk;
1837 
1838         if ((path[0].p_depth != 1) ||
1839             (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1840             (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1841                 return;
1842 
1843         /*
1844          * We need to modify the block allocation bitmap and the block
1845          * group descriptor to release the extent tree block.  If we
1846          * can't get the journal credits, give up.
1847          */
1848         if (ext4_journal_extend(handle, 2,
1849                         ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1850                 return;
1851 
1852         /*
1853          * Copy the extent data up to the inode
1854          */
1855         blk = ext4_idx_pblock(path[0].p_idx);
1856         s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1857                 sizeof(struct ext4_extent_idx);
1858         s += sizeof(struct ext4_extent_header);
1859 
1860         path[1].p_maxdepth = path[0].p_maxdepth;
1861         memcpy(path[0].p_hdr, path[1].p_hdr, s);
1862         path[0].p_depth = 0;
1863         path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1864                 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1865         path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1866 
1867         brelse(path[1].p_bh);
1868         ext4_free_blocks(handle, inode, NULL, blk, 1,
1869                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1870 }
1871 
1872 /*
1873  * This function tries to merge the @ex extent to neighbours in the tree.
1874  * return 1 if merge left else 0.
1875  */
1876 static void ext4_ext_try_to_merge(handle_t *handle,
1877                                   struct inode *inode,
1878                                   struct ext4_ext_path *path,
1879                                   struct ext4_extent *ex) {
1880         struct ext4_extent_header *eh;
1881         unsigned int depth;
1882         int merge_done = 0;
1883 
1884         depth = ext_depth(inode);
1885         BUG_ON(path[depth].p_hdr == NULL);
1886         eh = path[depth].p_hdr;
1887 
1888         if (ex > EXT_FIRST_EXTENT(eh))
1889                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1890 
1891         if (!merge_done)
1892                 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1893 
1894         ext4_ext_try_to_merge_up(handle, inode, path);
1895 }
1896 
1897 /*
1898  * check if a portion of the "newext" extent overlaps with an
1899  * existing extent.
1900  *
1901  * If there is an overlap discovered, it updates the length of the newext
1902  * such that there will be no overlap, and then returns 1.
1903  * If there is no overlap found, it returns 0.
1904  */
1905 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1906                                            struct inode *inode,
1907                                            struct ext4_extent *newext,
1908                                            struct ext4_ext_path *path)
1909 {
1910         ext4_lblk_t b1, b2;
1911         unsigned int depth, len1;
1912         unsigned int ret = 0;
1913 
1914         b1 = le32_to_cpu(newext->ee_block);
1915         len1 = ext4_ext_get_actual_len(newext);
1916         depth = ext_depth(inode);
1917         if (!path[depth].p_ext)
1918                 goto out;
1919         b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1920 
1921         /*
1922          * get the next allocated block if the extent in the path
1923          * is before the requested block(s)
1924          */
1925         if (b2 < b1) {
1926                 b2 = ext4_ext_next_allocated_block(path);
1927                 if (b2 == EXT_MAX_BLOCKS)
1928                         goto out;
1929                 b2 = EXT4_LBLK_CMASK(sbi, b2);
1930         }
1931 
1932         /* check for wrap through zero on extent logical start block*/
1933         if (b1 + len1 < b1) {
1934                 len1 = EXT_MAX_BLOCKS - b1;
1935                 newext->ee_len = cpu_to_le16(len1);
1936                 ret = 1;
1937         }
1938 
1939         /* check for overlap */
1940         if (b1 + len1 > b2) {
1941                 newext->ee_len = cpu_to_le16(b2 - b1);
1942                 ret = 1;
1943         }
1944 out:
1945         return ret;
1946 }
1947 
1948 /*
1949  * ext4_ext_insert_extent:
1950  * tries to merge requsted extent into the existing extent or
1951  * inserts requested extent as new one into the tree,
1952  * creating new leaf in the no-space case.
1953  */
1954 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1955                                 struct ext4_ext_path **ppath,
1956                                 struct ext4_extent *newext, int gb_flags)
1957 {
1958         struct ext4_ext_path *path = *ppath;
1959         struct ext4_extent_header *eh;
1960         struct ext4_extent *ex, *fex;
1961         struct ext4_extent *nearex; /* nearest extent */
1962         struct ext4_ext_path *npath = NULL;
1963         int depth, len, err;
1964         ext4_lblk_t next;
1965         int mb_flags = 0, unwritten;
1966 
1967         if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1968                 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1969         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1970                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1971                 return -EFSCORRUPTED;
1972         }
1973         depth = ext_depth(inode);
1974         ex = path[depth].p_ext;
1975         eh = path[depth].p_hdr;
1976         if (unlikely(path[depth].p_hdr == NULL)) {
1977                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1978                 return -EFSCORRUPTED;
1979         }
1980 
1981         /* try to insert block into found extent and return */
1982         if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1983 
1984                 /*
1985                  * Try to see whether we should rather test the extent on
1986                  * right from ex, or from the left of ex. This is because
1987                  * ext4_find_extent() can return either extent on the
1988                  * left, or on the right from the searched position. This
1989                  * will make merging more effective.
1990                  */
1991                 if (ex < EXT_LAST_EXTENT(eh) &&
1992                     (le32_to_cpu(ex->ee_block) +
1993                     ext4_ext_get_actual_len(ex) <
1994                     le32_to_cpu(newext->ee_block))) {
1995                         ex += 1;
1996                         goto prepend;
1997                 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1998                            (le32_to_cpu(newext->ee_block) +
1999                            ext4_ext_get_actual_len(newext) <
2000                            le32_to_cpu(ex->ee_block)))
2001                         ex -= 1;
2002 
2003                 /* Try to append newex to the ex */
2004                 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2005                         ext_debug("append [%d]%d block to %u:[%d]%d"
2006                                   "(from %llu)\n",
2007                                   ext4_ext_is_unwritten(newext),
2008                                   ext4_ext_get_actual_len(newext),
2009                                   le32_to_cpu(ex->ee_block),
2010                                   ext4_ext_is_unwritten(ex),
2011                                   ext4_ext_get_actual_len(ex),
2012                                   ext4_ext_pblock(ex));
2013                         err = ext4_ext_get_access(handle, inode,
2014                                                   path + depth);
2015                         if (err)
2016                                 return err;
2017                         unwritten = ext4_ext_is_unwritten(ex);
2018                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2019                                         + ext4_ext_get_actual_len(newext));
2020                         if (unwritten)
2021                                 ext4_ext_mark_unwritten(ex);
2022                         eh = path[depth].p_hdr;
2023                         nearex = ex;
2024                         goto merge;
2025                 }
2026 
2027 prepend:
2028                 /* Try to prepend newex to the ex */
2029                 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2030                         ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2031                                   "(from %llu)\n",
2032                                   le32_to_cpu(newext->ee_block),
2033                                   ext4_ext_is_unwritten(newext),
2034                                   ext4_ext_get_actual_len(newext),
2035                                   le32_to_cpu(ex->ee_block),
2036                                   ext4_ext_is_unwritten(ex),
2037                                   ext4_ext_get_actual_len(ex),
2038                                   ext4_ext_pblock(ex));
2039                         err = ext4_ext_get_access(handle, inode,
2040                                                   path + depth);
2041                         if (err)
2042                                 return err;
2043 
2044                         unwritten = ext4_ext_is_unwritten(ex);
2045                         ex->ee_block = newext->ee_block;
2046                         ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2047                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2048                                         + ext4_ext_get_actual_len(newext));
2049                         if (unwritten)
2050                                 ext4_ext_mark_unwritten(ex);
2051                         eh = path[depth].p_hdr;
2052                         nearex = ex;
2053                         goto merge;
2054                 }
2055         }
2056 
2057         depth = ext_depth(inode);
2058         eh = path[depth].p_hdr;
2059         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2060                 goto has_space;
2061 
2062         /* probably next leaf has space for us? */
2063         fex = EXT_LAST_EXTENT(eh);
2064         next = EXT_MAX_BLOCKS;
2065         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2066                 next = ext4_ext_next_leaf_block(path);
2067         if (next != EXT_MAX_BLOCKS) {
2068                 ext_debug("next leaf block - %u\n", next);
2069                 BUG_ON(npath != NULL);
2070                 npath = ext4_find_extent(inode, next, NULL, 0);
2071                 if (IS_ERR(npath))
2072                         return PTR_ERR(npath);
2073                 BUG_ON(npath->p_depth != path->p_depth);
2074                 eh = npath[depth].p_hdr;
2075                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2076                         ext_debug("next leaf isn't full(%d)\n",
2077                                   le16_to_cpu(eh->eh_entries));
2078                         path = npath;
2079                         goto has_space;
2080                 }
2081                 ext_debug("next leaf has no free space(%d,%d)\n",
2082                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2083         }
2084 
2085         /*
2086          * There is no free space in the found leaf.
2087          * We're gonna add a new leaf in the tree.
2088          */
2089         if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2090                 mb_flags |= EXT4_MB_USE_RESERVED;
2091         err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2092                                        ppath, newext);
2093         if (err)
2094                 goto cleanup;
2095         depth = ext_depth(inode);
2096         eh = path[depth].p_hdr;
2097 
2098 has_space:
2099         nearex = path[depth].p_ext;
2100 
2101         err = ext4_ext_get_access(handle, inode, path + depth);
2102         if (err)
2103                 goto cleanup;
2104 
2105         if (!nearex) {
2106                 /* there is no extent in this leaf, create first one */
2107                 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2108                                 le32_to_cpu(newext->ee_block),
2109                                 ext4_ext_pblock(newext),
2110                                 ext4_ext_is_unwritten(newext),
2111                                 ext4_ext_get_actual_len(newext));
2112                 nearex = EXT_FIRST_EXTENT(eh);
2113         } else {
2114                 if (le32_to_cpu(newext->ee_block)
2115                            > le32_to_cpu(nearex->ee_block)) {
2116                         /* Insert after */
2117                         ext_debug("insert %u:%llu:[%d]%d before: "
2118                                         "nearest %p\n",
2119                                         le32_to_cpu(newext->ee_block),
2120                                         ext4_ext_pblock(newext),
2121                                         ext4_ext_is_unwritten(newext),
2122                                         ext4_ext_get_actual_len(newext),
2123                                         nearex);
2124                         nearex++;
2125                 } else {
2126                         /* Insert before */
2127                         BUG_ON(newext->ee_block == nearex->ee_block);
2128                         ext_debug("insert %u:%llu:[%d]%d after: "
2129                                         "nearest %p\n",
2130                                         le32_to_cpu(newext->ee_block),
2131                                         ext4_ext_pblock(newext),
2132                                         ext4_ext_is_unwritten(newext),
2133                                         ext4_ext_get_actual_len(newext),
2134                                         nearex);
2135                 }
2136                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2137                 if (len > 0) {
2138                         ext_debug("insert %u:%llu:[%d]%d: "
2139                                         "move %d extents from 0x%p to 0x%p\n",
2140                                         le32_to_cpu(newext->ee_block),
2141                                         ext4_ext_pblock(newext),
2142                                         ext4_ext_is_unwritten(newext),
2143                                         ext4_ext_get_actual_len(newext),
2144                                         len, nearex, nearex + 1);
2145                         memmove(nearex + 1, nearex,
2146                                 len * sizeof(struct ext4_extent));
2147                 }
2148         }
2149 
2150         le16_add_cpu(&eh->eh_entries, 1);
2151         path[depth].p_ext = nearex;
2152         nearex->ee_block = newext->ee_block;
2153         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2154         nearex->ee_len = newext->ee_len;
2155 
2156 merge:
2157         /* try to merge extents */
2158         if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2159                 ext4_ext_try_to_merge(handle, inode, path, nearex);
2160 
2161 
2162         /* time to correct all indexes above */
2163         err = ext4_ext_correct_indexes(handle, inode, path);
2164         if (err)
2165                 goto cleanup;
2166 
2167         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2168 
2169 cleanup:
2170         ext4_ext_drop_refs(npath);
2171         kfree(npath);
2172         return err;
2173 }
2174 
2175 static int ext4_fill_fiemap_extents(struct inode *inode,
2176                                     ext4_lblk_t block, ext4_lblk_t num,
2177                                     struct fiemap_extent_info *fieinfo)
2178 {
2179         struct ext4_ext_path *path = NULL;
2180         struct ext4_extent *ex;
2181         struct extent_status es;
2182         ext4_lblk_t next, next_del, start = 0, end = 0;
2183         ext4_lblk_t last = block + num;
2184         int exists, depth = 0, err = 0;
2185         unsigned int flags = 0;
2186         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2187 
2188         while (block < last && block != EXT_MAX_BLOCKS) {
2189                 num = last - block;
2190                 /* find extent for this block */
2191                 down_read(&EXT4_I(inode)->i_data_sem);
2192 
2193                 path = ext4_find_extent(inode, block, &path, 0);
2194                 if (IS_ERR(path)) {
2195                         up_read(&EXT4_I(inode)->i_data_sem);
2196                         err = PTR_ERR(path);
2197                         path = NULL;
2198                         break;
2199                 }
2200 
2201                 depth = ext_depth(inode);
2202                 if (unlikely(path[depth].p_hdr == NULL)) {
2203                         up_read(&EXT4_I(inode)->i_data_sem);
2204                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2205                         err = -EFSCORRUPTED;
2206                         break;
2207                 }
2208                 ex = path[depth].p_ext;
2209                 next = ext4_ext_next_allocated_block(path);
2210 
2211                 flags = 0;
2212                 exists = 0;
2213                 if (!ex) {
2214                         /* there is no extent yet, so try to allocate
2215                          * all requested space */
2216                         start = block;
2217                         end = block + num;
2218                 } else if (le32_to_cpu(ex->ee_block) > block) {
2219                         /* need to allocate space before found extent */
2220                         start = block;
2221                         end = le32_to_cpu(ex->ee_block);
2222                         if (block + num < end)
2223                                 end = block + num;
2224                 } else if (block >= le32_to_cpu(ex->ee_block)
2225                                         + ext4_ext_get_actual_len(ex)) {
2226                         /* need to allocate space after found extent */
2227                         start = block;
2228                         end = block + num;
2229                         if (end >= next)
2230                                 end = next;
2231                 } else if (block >= le32_to_cpu(ex->ee_block)) {
2232                         /*
2233                          * some part of requested space is covered
2234                          * by found extent
2235                          */
2236                         start = block;
2237                         end = le32_to_cpu(ex->ee_block)
2238                                 + ext4_ext_get_actual_len(ex);
2239                         if (block + num < end)
2240                                 end = block + num;
2241                         exists = 1;
2242                 } else {
2243                         BUG();
2244                 }
2245                 BUG_ON(end <= start);
2246 
2247                 if (!exists) {
2248                         es.es_lblk = start;
2249                         es.es_len = end - start;
2250                         es.es_pblk = 0;
2251                 } else {
2252                         es.es_lblk = le32_to_cpu(ex->ee_block);
2253                         es.es_len = ext4_ext_get_actual_len(ex);
2254                         es.es_pblk = ext4_ext_pblock(ex);
2255                         if (ext4_ext_is_unwritten(ex))
2256                                 flags |= FIEMAP_EXTENT_UNWRITTEN;
2257                 }
2258 
2259                 /*
2260                  * Find delayed extent and update es accordingly. We call
2261                  * it even in !exists case to find out whether es is the
2262                  * last existing extent or not.
2263                  */
2264                 next_del = ext4_find_delayed_extent(inode, &es);
2265                 if (!exists && next_del) {
2266                         exists = 1;
2267                         flags |= (FIEMAP_EXTENT_DELALLOC |
2268                                   FIEMAP_EXTENT_UNKNOWN);
2269                 }
2270                 up_read(&EXT4_I(inode)->i_data_sem);
2271 
2272                 if (unlikely(es.es_len == 0)) {
2273                         EXT4_ERROR_INODE(inode, "es.es_len == 0");
2274                         err = -EFSCORRUPTED;
2275                         break;
2276                 }
2277 
2278                 /*
2279                  * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2280                  * we need to check next == EXT_MAX_BLOCKS because it is
2281                  * possible that an extent is with unwritten and delayed
2282                  * status due to when an extent is delayed allocated and
2283                  * is allocated by fallocate status tree will track both of
2284                  * them in a extent.
2285                  *
2286                  * So we could return a unwritten and delayed extent, and
2287                  * its block is equal to 'next'.
2288                  */
2289                 if (next == next_del && next == EXT_MAX_BLOCKS) {
2290                         flags |= FIEMAP_EXTENT_LAST;
2291                         if (unlikely(next_del != EXT_MAX_BLOCKS ||
2292                                      next != EXT_MAX_BLOCKS)) {
2293                                 EXT4_ERROR_INODE(inode,
2294                                                  "next extent == %u, next "
2295                                                  "delalloc extent = %u",
2296                                                  next, next_del);
2297                                 err = -EFSCORRUPTED;
2298                                 break;
2299                         }
2300                 }
2301 
2302                 if (exists) {
2303                         err = fiemap_fill_next_extent(fieinfo,
2304                                 (__u64)es.es_lblk << blksize_bits,
2305                                 (__u64)es.es_pblk << blksize_bits,
2306                                 (__u64)es.es_len << blksize_bits,
2307                                 flags);
2308                         if (err < 0)
2309                                 break;
2310                         if (err == 1) {
2311                                 err = 0;
2312                                 break;
2313                         }
2314                 }
2315 
2316                 block = es.es_lblk + es.es_len;
2317         }
2318 
2319         ext4_ext_drop_refs(path);
2320         kfree(path);
2321         return err;
2322 }
2323 
2324 static int ext4_fill_es_cache_info(struct inode *inode,
2325                                    ext4_lblk_t block, ext4_lblk_t num,
2326                                    struct fiemap_extent_info *fieinfo)
2327 {
2328         ext4_lblk_t next, end = block + num - 1;
2329         struct extent_status es;
2330         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2331         unsigned int flags;
2332         int err;
2333 
2334         while (block <= end) {
2335                 next = 0;
2336                 flags = 0;
2337                 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2338                         break;
2339                 if (ext4_es_is_unwritten(&es))
2340                         flags |= FIEMAP_EXTENT_UNWRITTEN;
2341                 if (ext4_es_is_delayed(&es))
2342                         flags |= (FIEMAP_EXTENT_DELALLOC |
2343                                   FIEMAP_EXTENT_UNKNOWN);
2344                 if (ext4_es_is_hole(&es))
2345                         flags |= EXT4_FIEMAP_EXTENT_HOLE;
2346                 if (next == 0)
2347                         flags |= FIEMAP_EXTENT_LAST;
2348                 if (flags & (FIEMAP_EXTENT_DELALLOC|
2349                              EXT4_FIEMAP_EXTENT_HOLE))
2350                         es.es_pblk = 0;
2351                 else
2352                         es.es_pblk = ext4_es_pblock(&es);
2353                 err = fiemap_fill_next_extent(fieinfo,
2354                                 (__u64)es.es_lblk << blksize_bits,
2355                                 (__u64)es.es_pblk << blksize_bits,
2356                                 (__u64)es.es_len << blksize_bits,
2357                                 flags);
2358                 if (next == 0)
2359                         break;
2360                 block = next;
2361                 if (err < 0)
2362                         return err;
2363                 if (err == 1)
2364                         return 0;
2365         }
2366         return 0;
2367 }
2368 
2369 
2370 /*
2371  * ext4_ext_determine_hole - determine hole around given block
2372  * @inode:      inode we lookup in
2373  * @path:       path in extent tree to @lblk
2374  * @lblk:       pointer to logical block around which we want to determine hole
2375  *
2376  * Determine hole length (and start if easily possible) around given logical
2377  * block. We don't try too hard to find the beginning of the hole but @path
2378  * actually points to extent before @lblk, we provide it.
2379  *
2380  * The function returns the length of a hole starting at @lblk. We update @lblk
2381  * to the beginning of the hole if we managed to find it.
2382  */
2383 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2384                                            struct ext4_ext_path *path,
2385                                            ext4_lblk_t *lblk)
2386 {
2387         int depth = ext_depth(inode);
2388         struct ext4_extent *ex;
2389         ext4_lblk_t len;
2390 
2391         ex = path[depth].p_ext;
2392         if (ex == NULL) {
2393                 /* there is no extent yet, so gap is [0;-] */
2394                 *lblk = 0;
2395                 len = EXT_MAX_BLOCKS;
2396         } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2397                 len = le32_to_cpu(ex->ee_block) - *lblk;
2398         } else if (*lblk >= le32_to_cpu(ex->ee_block)
2399                         + ext4_ext_get_actual_len(ex)) {
2400                 ext4_lblk_t next;
2401 
2402                 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2403                 next = ext4_ext_next_allocated_block(path);
2404                 BUG_ON(next == *lblk);
2405                 len = next - *lblk;
2406         } else {
2407                 BUG();
2408         }
2409         return len;
2410 }
2411 
2412 /*
2413  * ext4_ext_put_gap_in_cache:
2414  * calculate boundaries of the gap that the requested block fits into
2415  * and cache this gap
2416  */
2417 static void
2418 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2419                           ext4_lblk_t hole_len)
2420 {
2421         struct extent_status es;
2422 
2423         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2424                                   hole_start + hole_len - 1, &es);
2425         if (es.es_len) {
2426                 /* There's delayed extent containing lblock? */
2427                 if (es.es_lblk <= hole_start)
2428                         return;
2429                 hole_len = min(es.es_lblk - hole_start, hole_len);
2430         }
2431         ext_debug(" -> %u:%u\n", hole_start, hole_len);
2432         ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2433                               EXTENT_STATUS_HOLE);
2434 }
2435 
2436 /*
2437  * ext4_ext_rm_idx:
2438  * removes index from the index block.
2439  */
2440 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2441                         struct ext4_ext_path *path, int depth)
2442 {
2443         int err;
2444         ext4_fsblk_t leaf;
2445 
2446         /* free index block */
2447         depth--;
2448         path = path + depth;
2449         leaf = ext4_idx_pblock(path->p_idx);
2450         if (unlikely(path->p_hdr->eh_entries == 0)) {
2451                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2452                 return -EFSCORRUPTED;
2453         }
2454         err = ext4_ext_get_access(handle, inode, path);
2455         if (err)
2456                 return err;
2457 
2458         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2459                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2460                 len *= sizeof(struct ext4_extent_idx);
2461                 memmove(path->p_idx, path->p_idx + 1, len);
2462         }
2463 
2464         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2465         err = ext4_ext_dirty(handle, inode, path);
2466         if (err)
2467                 return err;
2468         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2469         trace_ext4_ext_rm_idx(inode, leaf);
2470 
2471         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2472                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2473 
2474         while (--depth >= 0) {
2475                 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2476                         break;
2477                 path--;
2478                 err = ext4_ext_get_access(handle, inode, path);
2479                 if (err)
2480                         break;
2481                 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2482                 err = ext4_ext_dirty(handle, inode, path);
2483                 if (err)
2484                         break;
2485         }
2486         return err;
2487 }
2488 
2489 /*
2490  * ext4_ext_calc_credits_for_single_extent:
2491  * This routine returns max. credits that needed to insert an extent
2492  * to the extent tree.
2493  * When pass the actual path, the caller should calculate credits
2494  * under i_data_sem.
2495  */
2496 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2497                                                 struct ext4_ext_path *path)
2498 {
2499         if (path) {
2500                 int depth = ext_depth(inode);
2501                 int ret = 0;
2502 
2503                 /* probably there is space in leaf? */
2504                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2505                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2506 
2507                         /*
2508                          *  There are some space in the leaf tree, no
2509                          *  need to account for leaf block credit
2510                          *
2511                          *  bitmaps and block group descriptor blocks
2512                          *  and other metadata blocks still need to be
2513                          *  accounted.
2514                          */
2515                         /* 1 bitmap, 1 block group descriptor */
2516                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2517                         return ret;
2518                 }
2519         }
2520 
2521         return ext4_chunk_trans_blocks(inode, nrblocks);
2522 }
2523 
2524 /*
2525  * How many index/leaf blocks need to change/allocate to add @extents extents?
2526  *
2527  * If we add a single extent, then in the worse case, each tree level
2528  * index/leaf need to be changed in case of the tree split.
2529  *
2530  * If more extents are inserted, they could cause the whole tree split more
2531  * than once, but this is really rare.
2532  */
2533 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2534 {
2535         int index;
2536         int depth;
2537 
2538         /* If we are converting the inline data, only one is needed here. */
2539         if (ext4_has_inline_data(inode))
2540                 return 1;
2541 
2542         depth = ext_depth(inode);
2543 
2544         if (extents <= 1)
2545                 index = depth * 2;
2546         else
2547                 index = depth * 3;
2548 
2549         return index;
2550 }
2551 
2552 static inline int get_default_free_blocks_flags(struct inode *inode)
2553 {
2554         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2555             ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2556                 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2557         else if (ext4_should_journal_data(inode))
2558                 return EXT4_FREE_BLOCKS_FORGET;
2559         return 0;
2560 }
2561 
2562 /*
2563  * ext4_rereserve_cluster - increment the reserved cluster count when
2564  *                          freeing a cluster with a pending reservation
2565  *
2566  * @inode - file containing the cluster
2567  * @lblk - logical block in cluster to be reserved
2568  *
2569  * Increments the reserved cluster count and adjusts quota in a bigalloc
2570  * file system when freeing a partial cluster containing at least one
2571  * delayed and unwritten block.  A partial cluster meeting that
2572  * requirement will have a pending reservation.  If so, the
2573  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2574  * defer reserved and allocated space accounting to a subsequent call
2575  * to this function.
2576  */
2577 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2578 {
2579         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2580         struct ext4_inode_info *ei = EXT4_I(inode);
2581 
2582         dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2583 
2584         spin_lock(&ei->i_block_reservation_lock);
2585         ei->i_reserved_data_blocks++;
2586         percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2587         spin_unlock(&ei->i_block_reservation_lock);
2588 
2589         percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2590         ext4_remove_pending(inode, lblk);
2591 }
2592 
2593 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2594                               struct ext4_extent *ex,
2595                               struct partial_cluster *partial,
2596                               ext4_lblk_t from, ext4_lblk_t to)
2597 {
2598         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2599         unsigned short ee_len = ext4_ext_get_actual_len(ex);
2600         ext4_fsblk_t last_pblk, pblk;
2601         ext4_lblk_t num;
2602         int flags;
2603 
2604         /* only extent tail removal is allowed */
2605         if (from < le32_to_cpu(ex->ee_block) ||
2606             to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2607                 ext4_error(sbi->s_sb,
2608                            "strange request: removal(2) %u-%u from %u:%u",
2609                            from, to, le32_to_cpu(ex->ee_block), ee_len);
2610                 return 0;
2611         }
2612 
2613 #ifdef EXTENTS_STATS
2614         spin_lock(&sbi->s_ext_stats_lock);
2615         sbi->s_ext_blocks += ee_len;
2616         sbi->s_ext_extents++;
2617         if (ee_len < sbi->s_ext_min)
2618                 sbi->s_ext_min = ee_len;
2619         if (ee_len > sbi->s_ext_max)
2620                 sbi->s_ext_max = ee_len;
2621         if (ext_depth(inode) > sbi->s_depth_max)
2622                 sbi->s_depth_max = ext_depth(inode);
2623         spin_unlock(&sbi->s_ext_stats_lock);
2624 #endif
2625 
2626         trace_ext4_remove_blocks(inode, ex, from, to, partial);
2627 
2628         /*
2629          * if we have a partial cluster, and it's different from the
2630          * cluster of the last block in the extent, we free it
2631          */
2632         last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2633 
2634         if (partial->state != initial &&
2635             partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2636                 if (partial->state == tofree) {
2637                         flags = get_default_free_blocks_flags(inode);
2638                         if (ext4_is_pending(inode, partial->lblk))
2639                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2640                         ext4_free_blocks(handle, inode, NULL,
2641                                          EXT4_C2B(sbi, partial->pclu),
2642                                          sbi->s_cluster_ratio, flags);
2643                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2644                                 ext4_rereserve_cluster(inode, partial->lblk);
2645                 }
2646                 partial->state = initial;
2647         }
2648 
2649         num = le32_to_cpu(ex->ee_block) + ee_len - from;
2650         pblk = ext4_ext_pblock(ex) + ee_len - num;
2651 
2652         /*
2653          * We free the partial cluster at the end of the extent (if any),
2654          * unless the cluster is used by another extent (partial_cluster
2655          * state is nofree).  If a partial cluster exists here, it must be
2656          * shared with the last block in the extent.
2657          */
2658         flags = get_default_free_blocks_flags(inode);
2659 
2660         /* partial, left end cluster aligned, right end unaligned */
2661         if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2662             (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2663             (partial->state != nofree)) {
2664                 if (ext4_is_pending(inode, to))
2665                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2666                 ext4_free_blocks(handle, inode, NULL,
2667                                  EXT4_PBLK_CMASK(sbi, last_pblk),
2668                                  sbi->s_cluster_ratio, flags);
2669                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2670                         ext4_rereserve_cluster(inode, to);
2671                 partial->state = initial;
2672                 flags = get_default_free_blocks_flags(inode);
2673         }
2674 
2675         flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2676 
2677         /*
2678          * For bigalloc file systems, we never free a partial cluster
2679          * at the beginning of the extent.  Instead, we check to see if we
2680          * need to free it on a subsequent call to ext4_remove_blocks,
2681          * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2682          */
2683         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2684         ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2685 
2686         /* reset the partial cluster if we've freed past it */
2687         if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2688                 partial->state = initial;
2689 
2690         /*
2691          * If we've freed the entire extent but the beginning is not left
2692          * cluster aligned and is not marked as ineligible for freeing we
2693          * record the partial cluster at the beginning of the extent.  It
2694          * wasn't freed by the preceding ext4_free_blocks() call, and we
2695          * need to look farther to the left to determine if it's to be freed
2696          * (not shared with another extent). Else, reset the partial
2697          * cluster - we're either  done freeing or the beginning of the
2698          * extent is left cluster aligned.
2699          */
2700         if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2701                 if (partial->state == initial) {
2702                         partial->pclu = EXT4_B2C(sbi, pblk);
2703                         partial->lblk = from;
2704                         partial->state = tofree;
2705                 }
2706         } else {
2707                 partial->state = initial;
2708         }
2709 
2710         return 0;
2711 }
2712 
2713 /*
2714  * ext4_ext_rm_leaf() Removes the extents associated with the
2715  * blocks appearing between "start" and "end".  Both "start"
2716  * and "end" must appear in the same extent or EIO is returned.
2717  *
2718  * @handle: The journal handle
2719  * @inode:  The files inode
2720  * @path:   The path to the leaf
2721  * @partial_cluster: The cluster which we'll have to free if all extents
2722  *                   has been released from it.  However, if this value is
2723  *                   negative, it's a cluster just to the right of the
2724  *                   punched region and it must not be freed.
2725  * @start:  The first block to remove
2726  * @end:   The last block to remove
2727  */
2728 static int
2729 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2730                  struct ext4_ext_path *path,
2731                  struct partial_cluster *partial,
2732                  ext4_lblk_t start, ext4_lblk_t end)
2733 {
2734         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2735         int err = 0, correct_index = 0;
2736         int depth = ext_depth(inode), credits, revoke_credits;
2737         struct ext4_extent_header *eh;
2738         ext4_lblk_t a, b;
2739         unsigned num;
2740         ext4_lblk_t ex_ee_block;
2741         unsigned short ex_ee_len;
2742         unsigned unwritten = 0;
2743         struct ext4_extent *ex;
2744         ext4_fsblk_t pblk;
2745 
2746         /* the header must be checked already in ext4_ext_remove_space() */
2747         ext_debug("truncate since %u in leaf to %u\n", start, end);
2748         if (!path[depth].p_hdr)
2749                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2750         eh = path[depth].p_hdr;
2751         if (unlikely(path[depth].p_hdr == NULL)) {
2752                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2753                 return -EFSCORRUPTED;
2754         }
2755         /* find where to start removing */
2756         ex = path[depth].p_ext;
2757         if (!ex)
2758                 ex = EXT_LAST_EXTENT(eh);
2759 
2760         ex_ee_block = le32_to_cpu(ex->ee_block);
2761         ex_ee_len = ext4_ext_get_actual_len(ex);
2762 
2763         trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2764 
2765         while (ex >= EXT_FIRST_EXTENT(eh) &&
2766                         ex_ee_block + ex_ee_len > start) {
2767 
2768                 if (ext4_ext_is_unwritten(ex))
2769                         unwritten = 1;
2770                 else
2771                         unwritten = 0;
2772 
2773                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2774                           unwritten, ex_ee_len);
2775                 path[depth].p_ext = ex;
2776 
2777                 a = ex_ee_block > start ? ex_ee_block : start;
2778                 b = ex_ee_block+ex_ee_len - 1 < end ?
2779                         ex_ee_block+ex_ee_len - 1 : end;
2780 
2781                 ext_debug("  border %u:%u\n", a, b);
2782 
2783                 /* If this extent is beyond the end of the hole, skip it */
2784                 if (end < ex_ee_block) {
2785                         /*
2786                          * We're going to skip this extent and move to another,
2787                          * so note that its first cluster is in use to avoid
2788                          * freeing it when removing blocks.  Eventually, the
2789                          * right edge of the truncated/punched region will
2790                          * be just to the left.
2791                          */
2792                         if (sbi->s_cluster_ratio > 1) {
2793                                 pblk = ext4_ext_pblock(ex);
2794                                 partial->pclu = EXT4_B2C(sbi, pblk);
2795                                 partial->state = nofree;
2796                         }
2797                         ex--;
2798                         ex_ee_block = le32_to_cpu(ex->ee_block);
2799                         ex_ee_len = ext4_ext_get_actual_len(ex);
2800                         continue;
2801                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2802                         EXT4_ERROR_INODE(inode,
2803                                          "can not handle truncate %u:%u "
2804                                          "on extent %u:%u",
2805                                          start, end, ex_ee_block,
2806                                          ex_ee_block + ex_ee_len - 1);
2807                         err = -EFSCORRUPTED;
2808                         goto out;
2809                 } else if (a != ex_ee_block) {
2810                         /* remove tail of the extent */
2811                         num = a - ex_ee_block;
2812                 } else {
2813                         /* remove whole extent: excellent! */
2814                         num = 0;
2815                 }
2816                 /*
2817                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2818                  * descriptor) for each block group; assume two block
2819                  * groups plus ex_ee_len/blocks_per_block_group for
2820                  * the worst case
2821                  */
2822                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2823                 if (ex == EXT_FIRST_EXTENT(eh)) {
2824                         correct_index = 1;
2825                         credits += (ext_depth(inode)) + 1;
2826                 }
2827                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2828                 /*
2829                  * We may end up freeing some index blocks and data from the
2830                  * punched range. Note that partial clusters are accounted for
2831                  * by ext4_free_data_revoke_credits().
2832                  */
2833                 revoke_credits =
2834                         ext4_free_metadata_revoke_credits(inode->i_sb,
2835                                                           ext_depth(inode)) +
2836                         ext4_free_data_revoke_credits(inode, b - a + 1);
2837 
2838                 err = ext4_datasem_ensure_credits(handle, inode, credits,
2839                                                   credits, revoke_credits);
2840                 if (err) {
2841                         if (err > 0)
2842                                 err = -EAGAIN;
2843                         goto out;
2844                 }
2845 
2846                 err = ext4_ext_get_access(handle, inode, path + depth);
2847                 if (err)
2848                         goto out;
2849 
2850                 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2851                 if (err)
2852                         goto out;
2853 
2854                 if (num == 0)
2855                         /* this extent is removed; mark slot entirely unused */
2856                         ext4_ext_store_pblock(ex, 0);
2857 
2858                 ex->ee_len = cpu_to_le16(num);
2859                 /*
2860                  * Do not mark unwritten if all the blocks in the
2861                  * extent have been removed.
2862                  */
2863                 if (unwritten && num)
2864                         ext4_ext_mark_unwritten(ex);
2865                 /*
2866                  * If the extent was completely released,
2867                  * we need to remove it from the leaf
2868                  */
2869                 if (num == 0) {
2870                         if (end != EXT_MAX_BLOCKS - 1) {
2871                                 /*
2872                                  * For hole punching, we need to scoot all the
2873                                  * extents up when an extent is removed so that
2874                                  * we dont have blank extents in the middle
2875                                  */
2876                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2877                                         sizeof(struct ext4_extent));
2878 
2879                                 /* Now get rid of the one at the end */
2880                                 memset(EXT_LAST_EXTENT(eh), 0,
2881                                         sizeof(struct ext4_extent));
2882                         }
2883                         le16_add_cpu(&eh->eh_entries, -1);
2884                 }
2885 
2886                 err = ext4_ext_dirty(handle, inode, path + depth);
2887                 if (err)
2888                         goto out;
2889 
2890                 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2891                                 ext4_ext_pblock(ex));
2892                 ex--;
2893                 ex_ee_block = le32_to_cpu(ex->ee_block);
2894                 ex_ee_len = ext4_ext_get_actual_len(ex);
2895         }
2896 
2897         if (correct_index && eh->eh_entries)
2898                 err = ext4_ext_correct_indexes(handle, inode, path);
2899 
2900         /*
2901          * If there's a partial cluster and at least one extent remains in
2902          * the leaf, free the partial cluster if it isn't shared with the
2903          * current extent.  If it is shared with the current extent
2904          * we reset the partial cluster because we've reached the start of the
2905          * truncated/punched region and we're done removing blocks.
2906          */
2907         if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2908                 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2909                 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2910                         int flags = get_default_free_blocks_flags(inode);
2911 
2912                         if (ext4_is_pending(inode, partial->lblk))
2913                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2914                         ext4_free_blocks(handle, inode, NULL,
2915                                          EXT4_C2B(sbi, partial->pclu),
2916                                          sbi->s_cluster_ratio, flags);
2917                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2918                                 ext4_rereserve_cluster(inode, partial->lblk);
2919                 }
2920                 partial->state = initial;
2921         }
2922 
2923         /* if this leaf is free, then we should
2924          * remove it from index block above */
2925         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2926                 err = ext4_ext_rm_idx(handle, inode, path, depth);
2927 
2928 out:
2929         return err;
2930 }
2931 
2932 /*
2933  * ext4_ext_more_to_rm:
2934  * returns 1 if current index has to be freed (even partial)
2935  */
2936 static int
2937 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2938 {
2939         BUG_ON(path->p_idx == NULL);
2940 
2941         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2942                 return 0;
2943 
2944         /*
2945          * if truncate on deeper level happened, it wasn't partial,
2946          * so we have to consider current index for truncation
2947          */
2948         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2949                 return 0;
2950         return 1;
2951 }
2952 
2953 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2954                           ext4_lblk_t end)
2955 {
2956         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2957         int depth = ext_depth(inode);
2958         struct ext4_ext_path *path = NULL;
2959         struct partial_cluster partial;
2960         handle_t *handle;
2961         int i = 0, err = 0;
2962 
2963         partial.pclu = 0;
2964         partial.lblk = 0;
2965         partial.state = initial;
2966 
2967         ext_debug("truncate since %u to %u\n", start, end);
2968 
2969         /* probably first extent we're gonna free will be last in block */
2970         handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2971                         depth + 1,
2972                         ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2973         if (IS_ERR(handle))
2974                 return PTR_ERR(handle);
2975 
2976 again:
2977         trace_ext4_ext_remove_space(inode, start, end, depth);
2978 
2979         /*
2980          * Check if we are removing extents inside the extent tree. If that
2981          * is the case, we are going to punch a hole inside the extent tree
2982          * so we have to check whether we need to split the extent covering
2983          * the last block to remove so we can easily remove the part of it
2984          * in ext4_ext_rm_leaf().
2985          */
2986         if (end < EXT_MAX_BLOCKS - 1) {
2987                 struct ext4_extent *ex;
2988                 ext4_lblk_t ee_block, ex_end, lblk;
2989                 ext4_fsblk_t pblk;
2990 
2991                 /* find extent for or closest extent to this block */
2992                 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2993                 if (IS_ERR(path)) {
2994                         ext4_journal_stop(handle);
2995                         return PTR_ERR(path);
2996                 }
2997                 depth = ext_depth(inode);
2998                 /* Leaf not may not exist only if inode has no blocks at all */
2999                 ex = path[depth].p_ext;
3000                 if (!ex) {
3001                         if (depth) {
3002                                 EXT4_ERROR_INODE(inode,
3003                                                  "path[%d].p_hdr == NULL",
3004                                                  depth);
3005                                 err = -EFSCORRUPTED;
3006                         }
3007                         goto out;
3008                 }
3009 
3010                 ee_block = le32_to_cpu(ex->ee_block);
3011                 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
3012 
3013                 /*
3014                  * See if the last block is inside the extent, if so split
3015                  * the extent at 'end' block so we can easily remove the
3016                  * tail of the first part of the split extent in
3017                  * ext4_ext_rm_leaf().
3018                  */
3019                 if (end >= ee_block && end < ex_end) {
3020 
3021                         /*
3022                          * If we're going to split the extent, note that
3023                          * the cluster containing the block after 'end' is
3024                          * in use to avoid freeing it when removing blocks.
3025                          */
3026                         if (sbi->s_cluster_ratio > 1) {
3027                                 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
3028                                 partial.pclu = EXT4_B2C(sbi, pblk);
3029                                 partial.state = nofree;
3030                         }
3031 
3032                         /*
3033                          * Split the extent in two so that 'end' is the last
3034                          * block in the first new extent. Also we should not
3035                          * fail removing space due to ENOSPC so try to use
3036                          * reserved block if that happens.
3037                          */
3038                         err = ext4_force_split_extent_at(handle, inode, &path,
3039                                                          end + 1, 1);
3040                         if (err < 0)
3041                                 goto out;
3042 
3043                 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3044                            partial.state == initial) {
3045                         /*
3046                          * If we're punching, there's an extent to the right.
3047                          * If the partial cluster hasn't been set, set it to
3048                          * that extent's first cluster and its state to nofree
3049                          * so it won't be freed should it contain blocks to be
3050                          * removed. If it's already set (tofree/nofree), we're
3051                          * retrying and keep the original partial cluster info
3052                          * so a cluster marked tofree as a result of earlier
3053                          * extent removal is not lost.
3054                          */
3055                         lblk = ex_end + 1;
3056                         err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3057                                                     &ex);
3058                         if (err)
3059                                 goto out;
3060                         if (pblk) {
3061                                 partial.pclu = EXT4_B2C(sbi, pblk);
3062                                 partial.state = nofree;
3063                         }
3064                 }
3065         }
3066         /*
3067          * We start scanning from right side, freeing all the blocks
3068          * after i_size and walking into the tree depth-wise.
3069          */
3070         depth = ext_depth(inode);
3071         if (path) {
3072                 int k = i = depth;
3073                 while (--k > 0)
3074                         path[k].p_block =
3075                                 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3076         } else {
3077                 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3078                                GFP_NOFS);
3079                 if (path == NULL) {
3080                         ext4_journal_stop(handle);
3081                         return -ENOMEM;
3082                 }
3083                 path[0].p_maxdepth = path[0].p_depth = depth;
3084                 path[0].p_hdr = ext_inode_hdr(inode);
3085                 i = 0;
3086 
3087                 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3088                         err = -EFSCORRUPTED;
3089                         goto out;
3090                 }
3091         }
3092         err = 0;
3093 
3094         while (i >= 0 && err == 0) {
3095                 if (i == depth) {
3096                         /* this is leaf block */
3097                         err = ext4_ext_rm_leaf(handle, inode, path,
3098                                                &partial, start, end);
3099                         /* root level has p_bh == NULL, brelse() eats this */
3100                         brelse(path[i].p_bh);
3101                         path[i].p_bh = NULL;
3102                         i--;
3103                         continue;
3104                 }
3105 
3106                 /* this is index block */
3107                 if (!path[i].p_hdr) {
3108                         ext_debug("initialize header\n");
3109                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3110                 }
3111 
3112                 if (!path[i].p_idx) {
3113                         /* this level hasn't been touched yet */
3114                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3115                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3116                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
3117                                   path[i].p_hdr,
3118                                   le16_to_cpu(path[i].p_hdr->eh_entries));
3119                 } else {
3120                         /* we were already here, see at next index */
3121                         path[i].p_idx--;
3122                 }
3123 
3124                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3125                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
3126                                 path[i].p_idx);
3127                 if (ext4_ext_more_to_rm(path + i)) {
3128                         struct buffer_head *bh;
3129                         /* go to the next level */
3130                         ext_debug("move to level %d (block %llu)\n",
3131                                   i + 1, ext4_idx_pblock(path[i].p_idx));
3132                         memset(path + i + 1, 0, sizeof(*path));
3133                         bh = read_extent_tree_block(inode,
3134                                 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3135                                 EXT4_EX_NOCACHE);
3136                         if (IS_ERR(bh)) {
3137                                 /* should we reset i_size? */
3138                                 err = PTR_ERR(bh);
3139                                 break;
3140                         }
3141                         /* Yield here to deal with large extent trees.
3142                          * Should be a no-op if we did IO above. */
3143                         cond_resched();
3144                         if (WARN_ON(i + 1 > depth)) {
3145                                 err = -EFSCORRUPTED;
3146                                 break;
3147                         }
3148                         path[i + 1].p_bh = bh;
3149 
3150                         /* save actual number of indexes since this
3151                          * number is changed at the next iteration */
3152                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3153                         i++;
3154                 } else {
3155                         /* we finished processing this index, go up */
3156                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3157                                 /* index is empty, remove it;
3158                                  * handle must be already prepared by the
3159                                  * truncatei_leaf() */
3160                                 err = ext4_ext_rm_idx(handle, inode, path, i);
3161                         }
3162                         /* root level has p_bh == NULL, brelse() eats this */
3163                         brelse(path[i].p_bh);
3164                         path[i].p_bh = NULL;
3165                         i--;
3166                         ext_debug("return to level %d\n", i);
3167                 }
3168         }
3169 
3170         trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3171                                          path->p_hdr->eh_entries);
3172 
3173         /*
3174          * if there's a partial cluster and we have removed the first extent
3175          * in the file, then we also free the partial cluster, if any
3176          */
3177         if (partial.state == tofree && err == 0) {
3178                 int flags = get_default_free_blocks_flags(inode);
3179 
3180                 if (ext4_is_pending(inode, partial.lblk))
3181                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3182                 ext4_free_blocks(handle, inode, NULL,
3183                                  EXT4_C2B(sbi, partial.pclu),
3184                                  sbi->s_cluster_ratio, flags);
3185                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3186                         ext4_rereserve_cluster(inode, partial.lblk);
3187                 partial.state = initial;
3188         }
3189 
3190         /* TODO: flexible tree reduction should be here */
3191         if (path->p_hdr->eh_entries == 0) {
3192                 /*
3193                  * truncate to zero freed all the tree,
3194                  * so we need to correct eh_depth
3195                  */
3196                 err = ext4_ext_get_access(handle, inode, path);
3197                 if (err == 0) {
3198                         ext_inode_hdr(inode)->eh_depth = 0;
3199                         ext_inode_hdr(inode)->eh_max =
3200                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
3201                         err = ext4_ext_dirty(handle, inode, path);
3202                 }
3203         }
3204 out:
3205         ext4_ext_drop_refs(path);
3206         kfree(path);
3207         path = NULL;
3208         if (err == -EAGAIN)
3209                 goto again;
3210         ext4_journal_stop(handle);
3211 
3212         return err;
3213 }
3214 
3215 /*
3216  * called at mount time
3217  */
3218 void ext4_ext_init(struct super_block *sb)
3219 {
3220         /*
3221          * possible initialization would be here
3222          */
3223 
3224         if (ext4_has_feature_extents(sb)) {
3225 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3226                 printk(KERN_INFO "EXT4-fs: file extents enabled"
3227 #ifdef AGGRESSIVE_TEST
3228                        ", aggressive tests"
3229 #endif
3230 #ifdef CHECK_BINSEARCH
3231                        ", check binsearch"
3232 #endif
3233 #ifdef EXTENTS_STATS
3234                        ", stats"
3235 #endif
3236                        "\n");
3237 #endif
3238 #ifdef EXTENTS_STATS
3239                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3240                 EXT4_SB(sb)->s_ext_min = 1 << 30;
3241                 EXT4_SB(sb)->s_ext_max = 0;
3242 #endif
3243         }
3244 }
3245 
3246 /*
3247  * called at umount time
3248  */
3249 void ext4_ext_release(struct super_block *sb)
3250 {
3251         if (!ext4_has_feature_extents(sb))
3252                 return;
3253 
3254 #ifdef EXTENTS_STATS
3255         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3256                 struct ext4_sb_info *sbi = EXT4_SB(sb);
3257                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3258                         sbi->s_ext_blocks, sbi->s_ext_extents,
3259                         sbi->s_ext_blocks / sbi->s_ext_extents);
3260                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3261                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3262         }
3263 #endif
3264 }
3265 
3266 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3267 {
3268         ext4_lblk_t  ee_block;
3269         ext4_fsblk_t ee_pblock;
3270         unsigned int ee_len;
3271 
3272         ee_block  = le32_to_cpu(ex->ee_block);
3273         ee_len    = ext4_ext_get_actual_len(ex);
3274         ee_pblock = ext4_ext_pblock(ex);
3275 
3276         if (ee_len == 0)
3277                 return 0;
3278 
3279         return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3280                                      EXTENT_STATUS_WRITTEN);
3281 }
3282 
3283 /* FIXME!! we need to try to merge to left or right after zero-out  */
3284 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3285 {
3286         ext4_fsblk_t ee_pblock;
3287         unsigned int ee_len;
3288 
3289         ee_len    = ext4_ext_get_actual_len(ex);
3290         ee_pblock = ext4_ext_pblock(ex);
3291         return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3292                                   ee_len);
3293 }
3294 
3295 /*
3296  * ext4_split_extent_at() splits an extent at given block.
3297  *
3298  * @handle: the journal handle
3299  * @inode: the file inode
3300  * @path: the path to the extent
3301  * @split: the logical block where the extent is splitted.
3302  * @split_flags: indicates if the extent could be zeroout if split fails, and
3303  *               the states(init or unwritten) of new extents.
3304  * @flags: flags used to insert new extent to extent tree.
3305  *
3306  *
3307  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3308  * of which are deterimined by split_flag.
3309  *
3310  * There are two cases:
3311  *  a> the extent are splitted into two extent.
3312  *  b> split is not needed, and just mark the extent.
3313  *
3314  * return 0 on success.
3315  */
3316 static int ext4_split_extent_at(handle_t *handle,
3317                              struct inode *inode,
3318                              struct ext4_ext_path **ppath,
3319                              ext4_lblk_t split,
3320                              int split_flag,
3321                              int flags)
3322 {
3323         struct ext4_ext_path *path = *ppath;
3324         ext4_fsblk_t newblock;
3325         ext4_lblk_t ee_block;
3326         struct ext4_extent *ex, newex, orig_ex, zero_ex;
3327         struct ext4_extent *ex2 = NULL;
3328         unsigned int ee_len, depth;
3329         int err = 0;
3330 
3331         BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3332                (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3333 
3334         ext_debug("ext4_split_extents_at: inode %lu, logical"
3335                 "block %llu\n", inode->i_ino, (unsigned long long)split);
3336 
3337         ext4_ext_show_leaf(inode, path);
3338 
3339         depth = ext_depth(inode);
3340         ex = path[depth].p_ext;
3341         ee_block = le32_to_cpu(ex->ee_block);
3342         ee_len = ext4_ext_get_actual_len(ex);
3343         newblock = split - ee_block + ext4_ext_pblock(ex);
3344 
3345         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3346         BUG_ON(!ext4_ext_is_unwritten(ex) &&
3347                split_flag & (EXT4_EXT_MAY_ZEROOUT |
3348                              EXT4_EXT_MARK_UNWRIT1 |
3349                              EXT4_EXT_MARK_UNWRIT2));
3350 
3351         err = ext4_ext_get_access(handle, inode, path + depth);
3352         if (err)
3353                 goto out;
3354 
3355         if (split == ee_block) {
3356                 /*
3357                  * case b: block @split is the block that the extent begins with
3358                  * then we just change the state of the extent, and splitting
3359                  * is not needed.
3360                  */
3361                 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3362                         ext4_ext_mark_unwritten(ex);
3363                 else
3364                         ext4_ext_mark_initialized(ex);
3365 
3366                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3367                         ext4_ext_try_to_merge(handle, inode, path, ex);
3368 
3369                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3370                 goto out;
3371         }
3372 
3373         /* case a */
3374         memcpy(&orig_ex, ex, sizeof(orig_ex));
3375         ex->ee_len = cpu_to_le16(split - ee_block);
3376         if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3377                 ext4_ext_mark_unwritten(ex);
3378 
3379         /*
3380          * path may lead to new leaf, not to original leaf any more
3381          * after ext4_ext_insert_extent() returns,
3382          */
3383         err = ext4_ext_dirty(handle, inode, path + depth);
3384         if (err)
3385                 goto fix_extent_len;
3386 
3387         ex2 = &newex;
3388         ex2->ee_block = cpu_to_le32(split);
3389         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3390         ext4_ext_store_pblock(ex2, newblock);
3391         if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3392                 ext4_ext_mark_unwritten(ex2);
3393 
3394         err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3395         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3396                 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3397                         if (split_flag & EXT4_EXT_DATA_VALID1) {
3398                                 err = ext4_ext_zeroout(inode, ex2);
3399                                 zero_ex.ee_block = ex2->ee_block;
3400                                 zero_ex.ee_len = cpu_to_le16(
3401                                                 ext4_ext_get_actual_len(ex2));
3402                                 ext4_ext_store_pblock(&zero_ex,
3403                                                       ext4_ext_pblock(ex2));
3404                         } else {
3405                                 err = ext4_ext_zeroout(inode, ex);
3406                                 zero_ex.ee_block = ex->ee_block;
3407                                 zero_ex.ee_len = cpu_to_le16(
3408                                                 ext4_ext_get_actual_len(ex));
3409                                 ext4_ext_store_pblock(&zero_ex,
3410                                                       ext4_ext_pblock(ex));
3411                         }
3412                 } else {
3413                         err = ext4_ext_zeroout(inode, &orig_ex);
3414                         zero_ex.ee_block = orig_ex.ee_block;
3415                         zero_ex.ee_len = cpu_to_le16(
3416                                                 ext4_ext_get_actual_len(&orig_ex));
3417                         ext4_ext_store_pblock(&zero_ex,
3418                                               ext4_ext_pblock(&orig_ex));
3419                 }
3420 
3421                 if (err)
3422                         goto fix_extent_len;
3423                 /* update the extent length and mark as initialized */
3424                 ex->ee_len = cpu_to_le16(ee_len);
3425                 ext4_ext_try_to_merge(handle, inode, path, ex);
3426                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3427                 if (err)
3428                         goto fix_extent_len;
3429 
3430                 /* update extent status tree */
3431                 err = ext4_zeroout_es(inode, &zero_ex);
3432 
3433                 goto out;
3434         } else if (err)
3435                 goto fix_extent_len;
3436 
3437 out:
3438         ext4_ext_show_leaf(inode, path);
3439         return err;
3440 
3441 fix_extent_len:
3442         ex->ee_len = orig_ex.ee_len;
3443         ext4_ext_dirty(handle, inode, path + path->p_depth);
3444         return err;
3445 }
3446 
3447 /*
3448  * ext4_split_extents() splits an extent and mark extent which is covered
3449  * by @map as split_flags indicates
3450  *
3451  * It may result in splitting the extent into multiple extents (up to three)
3452  * There are three possibilities:
3453  *   a> There is no split required
3454  *   b> Splits in two extents: Split is happening at either end of the extent
3455  *   c> Splits in three extents: Somone is splitting in middle of the extent
3456  *
3457  */
3458 static int ext4_split_extent(handle_t *handle,
3459                               struct inode *inode,
3460                               struct ext4_ext_path **ppath,
3461                               struct ext4_map_blocks *map,
3462                               int split_flag,
3463                               int flags)
3464 {
3465         struct ext4_ext_path *path = *ppath;
3466         ext4_lblk_t ee_block;
3467         struct ext4_extent *ex;
3468         unsigned int ee_len, depth;
3469         int err = 0;
3470         int unwritten;
3471         int split_flag1, flags1;
3472         int allocated = map->m_len;
3473 
3474         depth = ext_depth(inode);
3475         ex = path[depth].p_ext;
3476         ee_block = le32_to_cpu(ex->ee_block);
3477         ee_len = ext4_ext_get_actual_len(ex);
3478         unwritten = ext4_ext_is_unwritten(ex);
3479 
3480         if (map->m_lblk + map->m_len < ee_block + ee_len) {
3481                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3482                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3483                 if (unwritten)
3484                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3485                                        EXT4_EXT_MARK_UNWRIT2;
3486                 if (split_flag & EXT4_EXT_DATA_VALID2)
3487                         split_flag1 |= EXT4_EXT_DATA_VALID1;
3488                 err = ext4_split_extent_at(handle, inode, ppath,
3489                                 map->m_lblk + map->m_len, split_flag1, flags1);
3490                 if (err)
3491                         goto out;
3492         } else {
3493                 allocated = ee_len - (map->m_lblk - ee_block);
3494         }
3495         /*
3496          * Update path is required because previous ext4_split_extent_at() may
3497          * result in split of original leaf or extent zeroout.
3498          */
3499         path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3500         if (IS_ERR(path))
3501                 return PTR_ERR(path);
3502         depth = ext_depth(inode);
3503         ex = path[depth].p_ext;
3504         if (!ex) {
3505                 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3506                                  (unsigned long) map->m_lblk);
3507                 return -EFSCORRUPTED;
3508         }
3509         unwritten = ext4_ext_is_unwritten(ex);
3510         split_flag1 = 0;
3511 
3512         if (map->m_lblk >= ee_block) {
3513                 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3514                 if (unwritten) {
3515                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3516                         split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3517                                                      EXT4_EXT_MARK_UNWRIT2);
3518                 }
3519                 err = ext4_split_extent_at(handle, inode, ppath,
3520                                 map->m_lblk, split_flag1, flags);
3521                 if (err)
3522                         goto out;
3523         }
3524 
3525         ext4_ext_show_leaf(inode, path);
3526 out:
3527         return err ? err : allocated;
3528 }
3529 
3530 /*
3531  * This function is called by ext4_ext_map_blocks() if someone tries to write
3532  * to an unwritten extent. It may result in splitting the unwritten
3533  * extent into multiple extents (up to three - one initialized and two
3534  * unwritten).
3535  * There are three possibilities:
3536  *   a> There is no split required: Entire extent should be initialized
3537  *   b> Splits in two extents: Write is happening at either end of the extent
3538  *   c> Splits in three extents: Somone is writing in middle of the extent
3539  *
3540  * Pre-conditions:
3541  *  - The extent pointed to by 'path' is unwritten.
3542  *  - The extent pointed to by 'path' contains a superset
3543  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3544  *
3545  * Post-conditions on success:
3546  *  - the returned value is the number of blocks beyond map->l_lblk
3547  *    that are allocated and initialized.
3548  *    It is guaranteed to be >= map->m_len.
3549  */
3550 static int ext4_ext_convert_to_initialized(handle_t *handle,
3551                                            struct inode *inode,
3552                                            struct ext4_map_blocks *map,
3553                                            struct ext4_ext_path **ppath,
3554                                            int flags)
3555 {
3556         struct ext4_ext_path *path = *ppath;
3557         struct ext4_sb_info *sbi;
3558         struct ext4_extent_header *eh;
3559         struct ext4_map_blocks split_map;
3560         struct ext4_extent zero_ex1, zero_ex2;
3561         struct ext4_extent *ex, *abut_ex;
3562         ext4_lblk_t ee_block, eof_block;
3563         unsigned int ee_len, depth, map_len = map->m_len;
3564         int allocated = 0, max_zeroout = 0;
3565         int err = 0;
3566         int split_flag = EXT4_EXT_DATA_VALID2;
3567 
3568         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3569                 "block %llu, max_blocks %u\n", inode->i_ino,
3570                 (unsigned long long)map->m_lblk, map_len);
3571 
3572         sbi = EXT4_SB(inode->i_sb);
3573         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3574                         >> inode->i_sb->s_blocksize_bits;
3575         if (eof_block < map->m_lblk + map_len)
3576                 eof_block = map->m_lblk + map_len;
3577 
3578         depth = ext_depth(inode);
3579         eh = path[depth].p_hdr;
3580         ex = path[depth].p_ext;
3581         ee_block = le32_to_cpu(ex->ee_block);
3582         ee_len = ext4_ext_get_actual_len(ex);
3583         zero_ex1.ee_len = 0;
3584         zero_ex2.ee_len = 0;
3585 
3586         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3587 
3588         /* Pre-conditions */
3589         BUG_ON(!ext4_ext_is_unwritten(ex));
3590         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3591 
3592         /*
3593          * Attempt to transfer newly initialized blocks from the currently
3594          * unwritten extent to its neighbor. This is much cheaper
3595          * than an insertion followed by a merge as those involve costly
3596          * memmove() calls. Transferring to the left is the common case in
3597          * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3598          * followed by append writes.
3599          *
3600          * Limitations of the current logic:
3601          *  - L1: we do not deal with writes covering the whole extent.
3602          *    This would require removing the extent if the transfer
3603          *    is possible.
3604          *  - L2: we only attempt to merge with an extent stored in the
3605          *    same extent tree node.
3606          */
3607         if ((map->m_lblk == ee_block) &&
3608                 /* See if we can merge left */
3609                 (map_len < ee_len) &&           /*L1*/
3610                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3611                 ext4_lblk_t prev_lblk;
3612                 ext4_fsblk_t prev_pblk, ee_pblk;
3613                 unsigned int prev_len;
3614 
3615                 abut_ex = ex - 1;
3616                 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3617                 prev_len = ext4_ext_get_actual_len(abut_ex);
3618                 prev_pblk = ext4_ext_pblock(abut_ex);
3619                 ee_pblk = ext4_ext_pblock(ex);
3620 
3621                 /*
3622                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3623                  * upon those conditions:
3624                  * - C1: abut_ex is initialized,
3625                  * - C2: abut_ex is logically abutting ex,
3626                  * - C3: abut_ex is physically abutting ex,
3627                  * - C4: abut_ex can receive the additional blocks without
3628                  *   overflowing the (initialized) length limit.
3629                  */
3630                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3631                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3632                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3633                         (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3634                         err = ext4_ext_get_access(handle, inode, path + depth);
3635                         if (err)
3636                                 goto out;
3637 
3638                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3639                                 map, ex, abut_ex);
3640 
3641                         /* Shift the start of ex by 'map_len' blocks */
3642                         ex->ee_block = cpu_to_le32(ee_block + map_len);
3643                         ext4_ext_store_pblock(ex, ee_pblk + map_len);
3644                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3645                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3646 
3647                         /* Extend abut_ex by 'map_len' blocks */
3648                         abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3649 
3650                         /* Result: number of initialized blocks past m_lblk */
3651                         allocated = map_len;
3652                 }
3653         } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3654                    (map_len < ee_len) &&        /*L1*/
3655                    ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3656                 /* See if we can merge right */
3657                 ext4_lblk_t next_lblk;
3658                 ext4_fsblk_t next_pblk, ee_pblk;
3659                 unsigned int next_len;
3660 
3661                 abut_ex = ex + 1;
3662                 next_lblk = le32_to_cpu(abut_ex->ee_block);
3663                 next_len = ext4_ext_get_actual_len(abut_ex);
3664                 next_pblk = ext4_ext_pblock(abut_ex);
3665                 ee_pblk = ext4_ext_pblock(ex);
3666 
3667                 /*
3668                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3669                  * upon those conditions:
3670                  * - C1: abut_ex is initialized,
3671                  * - C2: abut_ex is logically abutting ex,
3672                  * - C3: abut_ex is physically abutting ex,
3673                  * - C4: abut_ex can receive the additional blocks without
3674                  *   overflowing the (initialized) length limit.
3675                  */
3676                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3677                     ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3678                     ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3679                     (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3680                         err = ext4_ext_get_access(handle, inode, path + depth);
3681                         if (err)
3682                                 goto out;
3683 
3684                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3685                                 map, ex, abut_ex);
3686 
3687                         /* Shift the start of abut_ex by 'map_len' blocks */
3688                         abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3689                         ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3690                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3691                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3692 
3693                         /* Extend abut_ex by 'map_len' blocks */
3694                         abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3695 
3696                         /* Result: number of initialized blocks past m_lblk */
3697                         allocated = map_len;
3698                 }
3699         }
3700         if (allocated) {
3701                 /* Mark the block containing both extents as dirty */
3702                 ext4_ext_dirty(handle, inode, path + depth);
3703 
3704                 /* Update path to point to the right extent */
3705                 path[depth].p_ext = abut_ex;
3706                 goto out;
3707         } else
3708                 allocated = ee_len - (map->m_lblk - ee_block);
3709 
3710         WARN_ON(map->m_lblk < ee_block);
3711         /*
3712          * It is safe to convert extent to initialized via explicit
3713          * zeroout only if extent is fully inside i_size or new_size.
3714          */
3715         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3716 
3717         if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3718                 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3719                         (inode->i_sb->s_blocksize_bits - 10);
3720 
3721         if (IS_ENCRYPTED(inode))
3722                 max_zeroout = 0;
3723 
3724         /*
3725          * five cases:
3726          * 1. split the extent into three extents.
3727          * 2. split the extent into two extents, zeroout the head of the first
3728          *    extent.
3729          * 3. split the extent into two extents, zeroout the tail of the second
3730          *    extent.
3731          * 4. split the extent into two extents with out zeroout.
3732          * 5. no splitting needed, just possibly zeroout the head and / or the
3733          *    tail of the extent.
3734          */
3735         split_map.m_lblk = map->m_lblk;
3736         split_map.m_len = map->m_len;
3737 
3738         if (max_zeroout && (allocated > split_map.m_len)) {
3739                 if (allocated <= max_zeroout) {
3740                         /* case 3 or 5 */
3741                         zero_ex1.ee_block =
3742                                  cpu_to_le32(split_map.m_lblk +
3743                                              split_map.m_len);
3744                         zero_ex1.ee_len =
3745                                 cpu_to_le16(allocated - split_map.m_len);
3746                         ext4_ext_store_pblock(&zero_ex1,
3747                                 ext4_ext_pblock(ex) + split_map.m_lblk +
3748                                 split_map.m_len - ee_block);
3749                         err = ext4_ext_zeroout(inode, &zero_ex1);
3750                         if (err)
3751                                 goto out;
3752                         split_map.m_len = allocated;
3753                 }
3754                 if (split_map.m_lblk - ee_block + split_map.m_len <
3755                                                                 max_zeroout) {
3756                         /* case 2 or 5 */
3757                         if (split_map.m_lblk != ee_block) {
3758                                 zero_ex2.ee_block = ex->ee_block;
3759                                 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3760                                                         ee_block);
3761                                 ext4_ext_store_pblock(&zero_ex2,
3762                                                       ext4_ext_pblock(ex));
3763                                 err = ext4_ext_zeroout(inode, &zero_ex2);
3764                                 if (err)
3765                                         goto out;
3766                         }
3767 
3768                         split_map.m_len += split_map.m_lblk - ee_block;
3769                         split_map.m_lblk = ee_block;
3770                         allocated = map->m_len;
3771                 }
3772         }
3773 
3774         err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3775                                 flags);
3776         if (err > 0)
3777                 err = 0;
3778 out:
3779         /* If we have gotten a failure, don't zero out status tree */
3780         if (!err) {
3781                 err = ext4_zeroout_es(inode, &zero_ex1);
3782                 if (!err)
3783                         err = ext4_zeroout_es(inode, &zero_ex2);
3784         }
3785         return err ? err : allocated;
3786 }
3787 
3788 /*
3789  * This function is called by ext4_ext_map_blocks() from
3790  * ext4_get_blocks_dio_write() when DIO to write
3791  * to an unwritten extent.
3792  *
3793  * Writing to an unwritten extent may result in splitting the unwritten
3794  * extent into multiple initialized/unwritten extents (up to three)
3795  * There are three possibilities:
3796  *   a> There is no split required: Entire extent should be unwritten
3797  *   b> Splits in two extents: Write is happening at either end of the extent
3798  *   c> Splits in three extents: Somone is writing in middle of the extent
3799  *
3800  * This works the same way in the case of initialized -> unwritten conversion.
3801  *
3802  * One of more index blocks maybe needed if the extent tree grow after
3803  * the unwritten extent split. To prevent ENOSPC occur at the IO
3804  * complete, we need to split the unwritten extent before DIO submit
3805  * the IO. The unwritten extent called at this time will be split
3806  * into three unwritten extent(at most). After IO complete, the part
3807  * being filled will be convert to initialized by the end_io callback function
3808  * via ext4_convert_unwritten_extents().
3809  *
3810  * Returns the size of unwritten extent to be written on success.
3811  */
3812 static int ext4_split_convert_extents(handle_t *handle,
3813                                         struct inode *inode,
3814                                         struct ext4_map_blocks *map,
3815                                         struct ext4_ext_path **ppath,
3816                                         int flags)
3817 {
3818         struct ext4_ext_path *path = *ppath;
3819         ext4_lblk_t eof_block;
3820         ext4_lblk_t ee_block;
3821         struct ext4_extent *ex;
3822         unsigned int ee_len;
3823         int split_flag = 0, depth;
3824 
3825         ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3826                   __func__, inode->i_ino,
3827                   (unsigned long long)map->m_lblk, map->m_len);
3828 
3829         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3830                         >> inode->i_sb->s_blocksize_bits;
3831         if (eof_block < map->m_lblk + map->m_len)
3832                 eof_block = map->m_lblk + map->m_len;
3833         /*
3834          * It is safe to convert extent to initialized via explicit
3835          * zeroout only if extent is fully insde i_size or new_size.
3836          */
3837         depth = ext_depth(inode);
3838         ex = path[depth].p_ext;
3839         ee_block = le32_to_cpu(ex->ee_block);
3840         ee_len = ext4_ext_get_actual_len(ex);
3841 
3842         /* Convert to unwritten */
3843         if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3844                 split_flag |= EXT4_EXT_DATA_VALID1;
3845         /* Convert to initialized */
3846         } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3847                 split_flag |= ee_block + ee_len <= eof_block ?
3848                               EXT4_EXT_MAY_ZEROOUT : 0;
3849                 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3850         }
3851         flags |= EXT4_GET_BLOCKS_PRE_IO;
3852         return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3853 }
3854 
3855 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3856                                                 struct inode *inode,
3857                                                 struct ext4_map_blocks *map,
3858                                                 struct ext4_ext_path **ppath)
3859 {
3860         struct ext4_ext_path *path = *ppath;
3861         struct ext4_extent *ex;
3862         ext4_lblk_t ee_block;
3863         unsigned int ee_len;
3864         int depth;
3865         int err = 0;
3866 
3867         depth = ext_depth(inode);
3868         ex = path[depth].p_ext;
3869         ee_block = le32_to_cpu(ex->ee_block);
3870         ee_len = ext4_ext_get_actual_len(ex);
3871 
3872         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3873                 "block %llu, max_blocks %u\n", inode->i_ino,
3874                   (unsigned long long)ee_block, ee_len);
3875 
3876         /* If extent is larger than requested it is a clear sign that we still
3877          * have some extent state machine issues left. So extent_split is still
3878          * required.
3879          * TODO: Once all related issues will be fixed this situation should be
3880          * illegal.
3881          */
3882         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3883 #ifdef CONFIG_EXT4_DEBUG
3884                 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3885                              " len %u; IO logical block %llu, len %u",
3886                              inode->i_ino, (unsigned long long)ee_block, ee_len,
3887                              (unsigned long long)map->m_lblk, map->m_len);
3888 #endif
3889                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3890                                                  EXT4_GET_BLOCKS_CONVERT);
3891                 if (err < 0)
3892                         return err;
3893                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3894                 if (IS_ERR(path))
3895                         return PTR_ERR(path);
3896                 depth = ext_depth(inode);
3897                 ex = path[depth].p_ext;
3898         }
3899 
3900         err = ext4_ext_get_access(handle, inode, path + depth);
3901         if (err)
3902                 goto out;
3903         /* first mark the extent as initialized */
3904         ext4_ext_mark_initialized(ex);
3905 
3906         /* note: ext4_ext_correct_indexes() isn't needed here because
3907          * borders are not changed
3908          */
3909         ext4_ext_try_to_merge(handle, inode, path, ex);
3910 
3911         /* Mark modified extent as dirty */
3912         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3913 out:
3914         ext4_ext_show_leaf(inode, path);
3915         return err;
3916 }
3917 
3918 /*
3919  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3920  */
3921 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3922                               ext4_lblk_t lblk,
3923                               struct ext4_ext_path *path,
3924                               unsigned int len)
3925 {
3926         int i, depth;
3927         struct ext4_extent_header *eh;
3928         struct ext4_extent *last_ex;
3929 
3930         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3931                 return 0;
3932 
3933         depth = ext_depth(inode);
3934         eh = path[depth].p_hdr;
3935 
3936         /*
3937          * We're going to remove EOFBLOCKS_FL entirely in future so we
3938          * do not care for this case anymore. Simply remove the flag
3939          * if there are no extents.
3940          */
3941         if (unlikely(!eh->eh_entries))
3942                 goto out;
3943         last_ex = EXT_LAST_EXTENT(eh);
3944         /*
3945          * We should clear the EOFBLOCKS_FL flag if we are writing the
3946          * last block in the last extent in the file.  We test this by
3947          * first checking to see if the caller to
3948          * ext4_ext_get_blocks() was interested in the last block (or
3949          * a block beyond the last block) in the current extent.  If
3950          * this turns out to be false, we can bail out from this
3951          * function immediately.
3952          */
3953         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3954             ext4_ext_get_actual_len(last_ex))
3955                 return 0;
3956         /*
3957          * If the caller does appear to be planning to write at or
3958          * beyond the end of the current extent, we then test to see
3959          * if the current extent is the last extent in the file, by
3960          * checking to make sure it was reached via the rightmost node
3961          * at each level of the tree.
3962          */
3963         for (i = depth-1; i >= 0; i--)
3964                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3965                         return 0;
3966 out:
3967         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3968         return ext4_mark_inode_dirty(handle, inode);
3969 }
3970 
3971 static int
3972 convert_initialized_extent(handle_t *handle, struct inode *inode,
3973                            struct ext4_map_blocks *map,
3974                            struct ext4_ext_path **ppath,
3975                            unsigned int allocated)
3976 {
3977         struct ext4_ext_path *path = *ppath;
3978         struct ext4_extent *ex;
3979         ext4_lblk_t ee_block;
3980         unsigned int ee_len;
3981         int depth;
3982         int err = 0;
3983 
3984         /*
3985          * Make sure that the extent is no bigger than we support with
3986          * unwritten extent
3987          */
3988         if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3989                 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3990 
3991         depth = ext_depth(inode);
3992         ex = path[depth].p_ext;
3993         ee_block = le32_to_cpu(ex->ee_block);
3994         ee_len = ext4_ext_get_actual_len(ex);
3995 
3996         ext_debug("%s: inode %lu, logical"
3997                 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3998                   (unsigned long long)ee_block, ee_len);
3999 
4000         if (ee_block != map->m_lblk || ee_len > map->m_len) {
4001                 err = ext4_split_convert_extents(handle, inode, map, ppath,
4002                                 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
4003                 if (err < 0)
4004                         return err;
4005                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
4006                 if (IS_ERR(path))
4007                         return PTR_ERR(path);
4008                 depth = ext_depth(inode);
4009                 ex = path[depth].p_ext;
4010                 if (!ex) {
4011                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
4012                                          (unsigned long) map->m_lblk);
4013                         return -EFSCORRUPTED;
4014                 }
4015         }
4016 
4017         err = ext4_ext_get_access(handle, inode, path + depth);
4018         if (err)
4019                 return err;
4020         /* first mark the extent as unwritten */
4021         ext4_ext_mark_unwritten(ex);
4022 
4023         /* note: ext4_ext_correct_indexes() isn't needed here because
4024          * borders are not changed
4025          */
4026         ext4_ext_try_to_merge(handle, inode, path, ex);
4027 
4028         /* Mark modified extent as dirty */
4029         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4030         if (err)
4031                 return err;
4032         ext4_ext_show_leaf(inode, path);
4033 
4034         ext4_update_inode_fsync_trans(handle, inode, 1);
4035         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4036         if (err)
4037                 return err;
4038         map->m_flags |= EXT4_MAP_UNWRITTEN;
4039         if (allocated > map->m_len)
4040                 allocated = map->m_len;
4041         map->m_len = allocated;
4042         return allocated;
4043 }
4044 
4045 static int
4046 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4047                         struct ext4_map_blocks *map,
4048                         struct ext4_ext_path **ppath, int flags,
4049                         unsigned int allocated, ext4_fsblk_t newblock)
4050 {
4051         struct ext4_ext_path *path = *ppath;
4052         int ret = 0;
4053         int err = 0;
4054 
4055         ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4056                   "block %llu, max_blocks %u, flags %x, allocated %u\n",
4057                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4058                   flags, allocated);
4059         ext4_ext_show_leaf(inode, path);
4060 
4061         /*
4062          * When writing into unwritten space, we should not fail to
4063          * allocate metadata blocks for the new extent block if needed.
4064          */
4065         flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4066 
4067         trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4068                                                     allocated, newblock);
4069 
4070         /* get_block() before submit the IO, split the extent */
4071         if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4072                 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4073                                          flags | EXT4_GET_BLOCKS_CONVERT);
4074                 if (ret <= 0)
4075                         goto out;
4076                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4077                 goto out;
4078         }
4079         /* IO end_io complete, convert the filled extent to written */
4080         if (flags & EXT4_GET_BLOCKS_CONVERT) {
4081                 if (flags & EXT4_GET_BLOCKS_ZERO) {
4082                         if (allocated > map->m_len)
4083                                 allocated = map->m_len;
4084                         err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4085                                                  allocated);
4086                         if (err < 0)
4087                                 goto out2;
4088                 }
4089                 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4090                                                            ppath);
4091                 if (ret >= 0) {
4092                         ext4_update_inode_fsync_trans(handle, inode, 1);
4093                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
4094                                                  path, map->m_len);
4095                 } else
4096                         err = ret;
4097                 map->m_flags |= EXT4_MAP_MAPPED;
4098                 map->m_pblk = newblock;
4099                 if (allocated > map->m_len)
4100                         allocated = map->m_len;
4101                 map->m_len = allocated;
4102                 goto out2;
4103         }
4104         /* buffered IO case */
4105         /*
4106          * repeat fallocate creation request
4107          * we already have an unwritten extent
4108          */
4109         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4110                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4111                 goto map_out;
4112         }
4113 
4114         /* buffered READ or buffered write_begin() lookup */
4115         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4116                 /*
4117                  * We have blocks reserved already.  We
4118                  * return allocated blocks so that delalloc
4119                  * won't do block reservation for us.  But
4120                  * the buffer head will be unmapped so that
4121                  * a read from the block returns 0s.
4122                  */
4123                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4124                 goto out1;
4125         }
4126 
4127         /* buffered write, writepage time, convert*/
4128         ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4129         if (ret >= 0)
4130                 ext4_update_inode_fsync_trans(handle, inode, 1);
4131 out:
4132         if (ret <= 0) {
4133                 err = ret;
4134                 goto out2;
4135         } else
4136                 allocated = ret;
4137         map->m_flags |= EXT4_MAP_NEW;
4138         if (allocated > map->m_len)
4139                 allocated = map->m_len;
4140         map->m_len = allocated;
4141 
4142 map_out:
4143         map->m_flags |= EXT4_MAP_MAPPED;
4144         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4145                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4146                                          map->m_len);
4147                 if (err < 0)
4148                         goto out2;
4149         }
4150 out1:
4151         if (allocated > map->m_len)
4152                 allocated = map->m_len;
4153         ext4_ext_show_leaf(inode, path);
4154         map->m_pblk = newblock;
4155         map->m_len = allocated;
4156 out2:
4157         return err ? err : allocated;
4158 }
4159 
4160 /*
4161  * get_implied_cluster_alloc - check to see if the requested
4162  * allocation (in the map structure) overlaps with a cluster already
4163  * allocated in an extent.
4164  *      @sb     The filesystem superblock structure
4165  *      @map    The requested lblk->pblk mapping
4166  *      @ex     The extent structure which might contain an implied
4167  *                      cluster allocation
4168  *
4169  * This function is called by ext4_ext_map_blocks() after we failed to
4170  * find blocks that were already in the inode's extent tree.  Hence,
4171  * we know that the beginning of the requested region cannot overlap
4172  * the extent from the inode's extent tree.  There are three cases we
4173  * want to catch.  The first is this case:
4174  *
4175  *               |--- cluster # N--|
4176  *    |--- extent ---|  |---- requested region ---|
4177  *                      |==========|
4178  *
4179  * The second case that we need to test for is this one:
4180  *
4181  *   |--------- cluster # N ----------------|
4182  *         |--- requested region --|   |------- extent ----|
4183  *         |=======================|
4184  *
4185  * The third case is when the requested region lies between two extents
4186  * within the same cluster:
4187  *          |------------- cluster # N-------------|
4188  * |----- ex -----|                  |---- ex_right ----|
4189  *                  |------ requested region ------|
4190  *                  |================|
4191  *
4192  * In each of the above cases, we need to set the map->m_pblk and
4193  * map->m_len so it corresponds to the return the extent labelled as
4194  * "|====|" from cluster #N, since it is already in use for data in
4195  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
4196  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4197  * as a new "allocated" block region.  Otherwise, we will return 0 and
4198  * ext4_ext_map_blocks() will then allocate one or more new clusters
4199  * by calling ext4_mb_new_blocks().
4200  */
4201 static int get_implied_cluster_alloc(struct super_block *sb,
4202                                      struct ext4_map_blocks *map,
4203                                      struct ext4_extent *ex,
4204                                      struct ext4_ext_path *path)
4205 {
4206         struct ext4_sb_info *sbi = EXT4_SB(sb);
4207         ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4208         ext4_lblk_t ex_cluster_start, ex_cluster_end;
4209         ext4_lblk_t rr_cluster_start;
4210         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4211         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4212         unsigned short ee_len = ext4_ext_get_actual_len(ex);
4213 
4214         /* The extent passed in that we are trying to match */
4215         ex_cluster_start = EXT4_B2C(sbi, ee_block);
4216         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4217 
4218         /* The requested region passed into ext4_map_blocks() */
4219         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4220 
4221         if ((rr_cluster_start == ex_cluster_end) ||
4222             (rr_cluster_start == ex_cluster_start)) {
4223                 if (rr_cluster_start == ex_cluster_end)
4224                         ee_start += ee_len - 1;
4225                 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4226                 map->m_len = min(map->m_len,
4227                                  (unsigned) sbi->s_cluster_ratio - c_offset);
4228                 /*
4229                  * Check for and handle this case:
4230                  *
4231                  *   |--------- cluster # N-------------|
4232                  *                     |------- extent ----|
4233                  *         |--- requested region ---|
4234                  *         |===========|
4235                  */
4236 
4237                 if (map->m_lblk < ee_block)
4238                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
4239 
4240                 /*
4241                  * Check for the case where there is already another allocated
4242                  * block to the right of 'ex' but before the end of the cluster.
4243                  *
4244                  *          |------------- cluster # N-------------|
4245                  * |----- ex -----|                  |---- ex_right ----|
4246                  *                  |------ requested region ------|
4247                  *                  |================|
4248                  */
4249                 if (map->m_lblk > ee_block) {
4250                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4251                         map->m_len = min(map->m_len, next - map->m_lblk);
4252                 }
4253 
4254                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4255                 return 1;
4256         }
4257 
4258         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4259         return 0;
4260 }
4261 
4262 
4263 /*
4264  * Block allocation/map/preallocation routine for extents based files
4265  *
4266  *
4267  * Need to be called with
4268  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4269  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4270  *
4271  * return > 0, number of of blocks already mapped/allocated
4272  *          if create == 0 and these are pre-allocated blocks
4273  *              buffer head is unmapped
4274  *          otherwise blocks are mapped
4275  *
4276  * return = 0, if plain look up failed (blocks have not been allocated)
4277  *          buffer head is unmapped
4278  *
4279  * return < 0, error case.
4280  */
4281 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4282                         struct ext4_map_blocks *map, int flags)
4283 {
4284         struct ext4_ext_path *path = NULL;
4285         struct ext4_extent newex, *ex, *ex2;
4286         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4287         ext4_fsblk_t newblock = 0;
4288         int free_on_err = 0, err = 0, depth, ret;
4289         unsigned int allocated = 0, offset = 0;
4290         unsigned int allocated_clusters = 0;
4291         struct ext4_allocation_request ar;
4292         ext4_lblk_t cluster_offset;
4293         bool map_from_cluster = false;
4294 
4295         ext_debug("blocks %u/%u requested for inode %lu\n",
4296                   map->m_lblk, map->m_len, inode->i_ino);
4297         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4298 
4299         /* find extent for this block */
4300         path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4301         if (IS_ERR(path)) {
4302                 err = PTR_ERR(path);
4303                 path = NULL;
4304                 goto out2;
4305         }
4306 
4307         depth = ext_depth(inode);
4308 
4309         /*
4310          * consistent leaf must not be empty;
4311          * this situation is possible, though, _during_ tree modification;
4312          * this is why assert can't be put in ext4_find_extent()
4313          */
4314         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4315                 EXT4_ERROR_INODE(inode, "bad extent address "
4316                                  "lblock: %lu, depth: %d pblock %lld",
4317                                  (unsigned long) map->m_lblk, depth,
4318                                  path[depth].p_block);
4319                 err = -EFSCORRUPTED;
4320                 goto out2;
4321         }
4322 
4323         ex = path[depth].p_ext;
4324         if (ex) {
4325                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4326                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4327                 unsigned short ee_len;
4328 
4329 
4330                 /*
4331                  * unwritten extents are treated as holes, except that
4332                  * we split out initialized portions during a write.
4333                  */
4334                 ee_len = ext4_ext_get_actual_len(ex);
4335 
4336                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4337 
4338                 /* if found extent covers block, simply return it */
4339                 if (in_range(map->m_lblk, ee_block, ee_len)) {
4340                         newblock = map->m_lblk - ee_block + ee_start;
4341                         /* number of remaining blocks in the extent */
4342                         allocated = ee_len - (map->m_lblk - ee_block);
4343                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4344                                   ee_block, ee_len, newblock);
4345 
4346                         /*
4347                          * If the extent is initialized check whether the
4348                          * caller wants to convert it to unwritten.
4349                          */
4350                         if ((!ext4_ext_is_unwritten(ex)) &&
4351                             (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4352                                 allocated = convert_initialized_extent(
4353                                                 handle, inode, map, &path,
4354                                                 allocated);
4355                                 goto out2;
4356                         } else if (!ext4_ext_is_unwritten(ex))
4357                                 goto out;
4358 
4359                         ret = ext4_ext_handle_unwritten_extents(
4360                                 handle, inode, map, &path, flags,
4361                                 allocated, newblock);
4362                         if (ret < 0)
4363                                 err = ret;
4364                         else
4365                                 allocated = ret;
4366                         goto out2;
4367                 }
4368         }
4369 
4370         /*
4371          * requested block isn't allocated yet;
4372          * we couldn't try to create block if create flag is zero
4373          */
4374         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4375                 ext4_lblk_t hole_start, hole_len;
4376 
4377                 hole_start = map->m_lblk;
4378                 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4379                 /*
4380                  * put just found gap into cache to speed up
4381                  * subsequent requests
4382                  */
4383                 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4384 
4385                 /* Update hole_len to reflect hole size after map->m_lblk */
4386                 if (hole_start != map->m_lblk)
4387                         hole_len -= map->m_lblk - hole_start;
4388                 map->m_pblk = 0;
4389                 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4390 
4391                 goto out2;
4392         }
4393 
4394         /*
4395          * Okay, we need to do block allocation.
4396          */
4397         newex.ee_block = cpu_to_le32(map->m_lblk);
4398         cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4399 
4400         /*
4401          * If we are doing bigalloc, check to see if the extent returned
4402          * by ext4_find_extent() implies a cluster we can use.
4403          */
4404         if (cluster_offset && ex &&
4405             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4406                 ar.len = allocated = map->m_len;
4407                 newblock = map->m_pblk;
4408                 map_from_cluster = true;
4409                 goto got_allocated_blocks;
4410         }
4411 
4412         /* find neighbour allocated blocks */
4413         ar.lleft = map->m_lblk;
4414         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4415         if (err)
4416                 goto out2;
4417         ar.lright = map->m_lblk;
4418         ex2 = NULL;
4419         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4420         if (err)
4421                 goto out2;
4422 
4423         /* Check if the extent after searching to the right implies a
4424          * cluster we can use. */
4425         if ((sbi->s_cluster_ratio > 1) && ex2 &&
4426             get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4427                 ar.len = allocated = map->m_len;
4428                 newblock = map->m_pblk;
4429                 map_from_cluster = true;
4430                 goto got_allocated_blocks;
4431         }
4432 
4433         /*
4434          * See if request is beyond maximum number of blocks we can have in
4435          * a single extent. For an initialized extent this limit is
4436          * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4437          * EXT_UNWRITTEN_MAX_LEN.
4438          */
4439         if (map->m_len > EXT_INIT_MAX_LEN &&
4440             !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4441                 map->m_len = EXT_INIT_MAX_LEN;
4442         else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4443                  (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4444                 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4445 
4446         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4447         newex.ee_len = cpu_to_le16(map->m_len);
4448         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4449         if (err)
4450                 allocated = ext4_ext_get_actual_len(&newex);
4451         else
4452                 allocated = map->m_len;
4453 
4454         /* allocate new block */
4455         ar.inode = inode;
4456         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4457         ar.logical = map->m_lblk;
4458         /*
4459          * We calculate the offset from the beginning of the cluster
4460          * for the logical block number, since when we allocate a
4461          * physical cluster, the physical block should start at the
4462          * same offset from the beginning of the cluster.  This is
4463          * needed so that future calls to get_implied_cluster_alloc()
4464          * work correctly.
4465          */
4466         offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4467         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4468         ar.goal -= offset;
4469         ar.logical -= offset;
4470         if (S_ISREG(inode->i_mode))
4471                 ar.flags = EXT4_MB_HINT_DATA;
4472         else
4473                 /* disable in-core preallocation for non-regular files */
4474                 ar.flags = 0;
4475         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4476                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4477         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4478                 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4479         if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4480                 ar.flags |= EXT4_MB_USE_RESERVED;
4481         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4482         if (!newblock)
4483                 goto out2;
4484         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4485                   ar.goal, newblock, allocated);
4486         free_on_err = 1;
4487         allocated_clusters = ar.len;
4488         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4489         if (ar.len > allocated)
4490                 ar.len = allocated;
4491 
4492 got_allocated_blocks:
4493         /* try to insert new extent into found leaf and return */
4494         ext4_ext_store_pblock(&newex, newblock + offset);
4495         newex.ee_len = cpu_to_le16(ar.len);
4496         /* Mark unwritten */
4497         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4498                 ext4_ext_mark_unwritten(&newex);
4499                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4500         }
4501 
4502         err = 0;
4503         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4504                 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4505                                          path, ar.len);
4506         if (!err)
4507                 err = ext4_ext_insert_extent(handle, inode, &path,
4508                                              &newex, flags);
4509 
4510         if (err && free_on_err) {
4511                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4512                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4513                 /* free data blocks we just allocated */
4514                 /* not a good idea to call discard here directly,
4515                  * but otherwise we'd need to call it every free() */
4516                 ext4_discard_preallocations(inode);
4517                 ext4_free_blocks(handle, inode, NULL, newblock,
4518                                  EXT4_C2B(sbi, allocated_clusters), fb_flags);
4519                 goto out2;
4520         }
4521 
4522         /* previous routine could use block we allocated */
4523         newblock = ext4_ext_pblock(&newex);
4524         allocated = ext4_ext_get_actual_len(&newex);
4525         if (allocated > map->m_len)
4526                 allocated = map->m_len;
4527         map->m_flags |= EXT4_MAP_NEW;
4528 
4529         /*
4530          * Reduce the reserved cluster count to reflect successful deferred
4531          * allocation of delayed allocated clusters or direct allocation of
4532          * clusters discovered to be delayed allocated.  Once allocated, a
4533          * cluster is not included in the reserved count.
4534          */
4535         if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4536                 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4537                         /*
4538                          * When allocating delayed allocated clusters, simply
4539                          * reduce the reserved cluster count and claim quota
4540                          */
4541                         ext4_da_update_reserve_space(inode, allocated_clusters,
4542                                                         1);
4543                 } else {
4544                         ext4_lblk_t lblk, len;
4545                         unsigned int n;
4546 
4547                         /*
4548                          * When allocating non-delayed allocated clusters
4549                          * (from fallocate, filemap, DIO, or clusters
4550                          * allocated when delalloc has been disabled by
4551                          * ext4_nonda_switch), reduce the reserved cluster
4552                          * count by the number of allocated clusters that
4553                          * have previously been delayed allocated.  Quota
4554                          * has been claimed by ext4_mb_new_blocks() above,
4555                          * so release the quota reservations made for any
4556                          * previously delayed allocated clusters.
4557                          */
4558                         lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4559                         len = allocated_clusters << sbi->s_cluster_bits;
4560                         n = ext4_es_delayed_clu(inode, lblk, len);
4561                         if (n > 0)
4562                                 ext4_da_update_reserve_space(inode, (int) n, 0);
4563                 }
4564         }
4565 
4566         /*
4567          * Cache the extent and update transaction to commit on fdatasync only
4568          * when it is _not_ an unwritten extent.
4569          */
4570         if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4571                 ext4_update_inode_fsync_trans(handle, inode, 1);
4572         else
4573                 ext4_update_inode_fsync_trans(handle, inode, 0);
4574 out:
4575         if (allocated > map->m_len)
4576                 allocated = map->m_len;
4577         ext4_ext_show_leaf(inode, path);
4578         map->m_flags |= EXT4_MAP_MAPPED;
4579         map->m_pblk = newblock;
4580         map->m_len = allocated;
4581 out2:
4582         ext4_ext_drop_refs(path);
4583         kfree(path);
4584 
4585         trace_ext4_ext_map_blocks_exit(inode, flags, map,
4586                                        err ? err : allocated);
4587         return err ? err : allocated;
4588 }
4589 
4590 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4591 {
4592         struct super_block *sb = inode->i_sb;
4593         ext4_lblk_t last_block;
4594         int err = 0;
4595 
4596         /*
4597          * TODO: optimization is possible here.
4598          * Probably we need not scan at all,
4599          * because page truncation is enough.
4600          */
4601 
4602         /* we have to know where to truncate from in crash case */
4603         EXT4_I(inode)->i_disksize = inode->i_size;
4604         err = ext4_mark_inode_dirty(handle, inode);
4605         if (err)
4606                 return err;
4607 
4608         last_block = (inode->i_size + sb->s_blocksize - 1)
4609                         >> EXT4_BLOCK_SIZE_BITS(sb);
4610 retry:
4611         err = ext4_es_remove_extent(inode, last_block,
4612                                     EXT_MAX_BLOCKS - last_block);
4613         if (err == -ENOMEM) {
4614                 cond_resched();
4615                 congestion_wait(BLK_RW_ASYNC, HZ/50);
4616                 goto retry;
4617         }
4618         if (err)
4619                 return err;
4620         return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4621 }
4622 
4623 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4624                                   ext4_lblk_t len, loff_t new_size,
4625                                   int flags)
4626 {
4627         struct inode *inode = file_inode(file);
4628         handle_t *handle;
4629         int ret = 0;
4630         int ret2 = 0;
4631         int retries = 0;
4632         int depth = 0;
4633         struct ext4_map_blocks map;
4634         unsigned int credits;
4635         loff_t epos;
4636 
4637         BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4638         map.m_lblk = offset;
4639         map.m_len = len;
4640         /*
4641          * Don't normalize the request if it can fit in one extent so
4642          * that it doesn't get unnecessarily split into multiple
4643          * extents.
4644          */
4645         if (len <= EXT_UNWRITTEN_MAX_LEN)
4646                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4647 
4648         /*
4649          * credits to insert 1 extent into extent tree
4650          */
4651         credits = ext4_chunk_trans_blocks(inode, len);
4652         depth = ext_depth(inode);
4653 
4654 retry:
4655         while (ret >= 0 && len) {
4656                 /*
4657                  * Recalculate credits when extent tree depth changes.
4658                  */
4659                 if (depth != ext_depth(inode)) {
4660                         credits = ext4_chunk_trans_blocks(inode, len);
4661                         depth = ext_depth(inode);
4662                 }
4663 
4664                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4665                                             credits);
4666                 if (IS_ERR(handle)) {
4667                         ret = PTR_ERR(handle);
4668                         break;
4669                 }
4670                 ret = ext4_map_blocks(handle, inode, &map, flags);
4671                 if (ret <= 0) {
4672                         ext4_debug("inode #%lu: block %u: len %u: "
4673                                    "ext4_ext_map_blocks returned %d",
4674                                    inode->i_ino, map.m_lblk,
4675                                    map.m_len, ret);
4676                         ext4_mark_inode_dirty(handle, inode);
4677                         ret2 = ext4_journal_stop(handle);
4678                         break;
4679                 }
4680                 map.m_lblk += ret;
4681                 map.m_len = len = len - ret;
4682                 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4683                 inode->i_ctime = current_time(inode);
4684                 if (new_size) {
4685                         if (epos > new_size)
4686                                 epos = new_size;
4687                         if (ext4_update_inode_size(inode, epos) & 0x1)
4688                                 inode->i_mtime = inode->i_ctime;
4689                 } else {
4690                         if (epos > inode->i_size)
4691                                 ext4_set_inode_flag(inode,
4692                                                     EXT4_INODE_EOFBLOCKS);
4693                 }
4694                 ext4_mark_inode_dirty(handle, inode);
4695                 ext4_update_inode_fsync_trans(handle, inode, 1);
4696                 ret2 = ext4_journal_stop(handle);
4697                 if (ret2)
4698                         break;
4699         }
4700         if (ret == -ENOSPC &&
4701                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4702                 ret = 0;
4703                 goto retry;
4704         }
4705 
4706         return ret > 0 ? ret2 : ret;
4707 }
4708 
4709 static long ext4_zero_range(struct file *file, loff_t offset,
4710                             loff_t len, int mode)
4711 {
4712         struct inode *inode = file_inode(file);
4713         handle_t *handle = NULL;
4714         unsigned int max_blocks;
4715         loff_t new_size = 0;
4716         int ret = 0;
4717         int flags;
4718         int credits;
4719         int partial_begin, partial_end;
4720         loff_t start, end;
4721         ext4_lblk_t lblk;
4722         unsigned int blkbits = inode->i_blkbits;
4723 
4724         trace_ext4_zero_range(inode, offset, len, mode);
4725 
4726         if (!S_ISREG(inode->i_mode))
4727                 return -EINVAL;
4728 
4729         /* Call ext4_force_commit to flush all data in case of data=journal. */
4730         if (ext4_should_journal_data(inode)) {
4731                 ret = ext4_force_commit(inode->i_sb);
4732                 if (ret)
4733                         return ret;
4734         }
4735 
4736         /*
4737          * Round up offset. This is not fallocate, we neet to zero out
4738          * blocks, so convert interior block aligned part of the range to
4739          * unwritten and possibly manually zero out unaligned parts of the
4740          * range.
4741          */
4742         start = round_up(offset, 1 << blkbits);
4743         end = round_down((offset + len), 1 << blkbits);
4744 
4745         if (start < offset || end > offset + len)
4746                 return -EINVAL;
4747         partial_begin = offset & ((1 << blkbits) - 1);
4748         partial_end = (offset + len) & ((1 << blkbits) - 1);
4749 
4750         lblk = start >> blkbits;
4751         max_blocks = (end >> blkbits);
4752         if (max_blocks < lblk)
4753                 max_blocks = 0;
4754         else
4755                 max_blocks -= lblk;
4756 
4757         inode_lock(inode);
4758 
4759         /*
4760          * Indirect files do not support unwritten extnets
4761          */
4762         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4763                 ret = -EOPNOTSUPP;
4764                 goto out_mutex;
4765         }
4766 
4767         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4768             (offset + len > i_size_read(inode) ||
4769              offset + len > EXT4_I(inode)->i_disksize)) {
4770                 new_size = offset + len;
4771                 ret = inode_newsize_ok(inode, new_size);
4772                 if (ret)
4773                         goto out_mutex;
4774         }
4775 
4776         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4777         if (mode & FALLOC_FL_KEEP_SIZE)
4778                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4779 
4780         /* Wait all existing dio workers, newcomers will block on i_mutex */
4781         inode_dio_wait(inode);
4782 
4783         /* Preallocate the range including the unaligned edges */
4784         if (partial_begin || partial_end) {
4785                 ret = ext4_alloc_file_blocks(file,
4786                                 round_down(offset, 1 << blkbits) >> blkbits,
4787                                 (round_up((offset + len), 1 << blkbits) -
4788                                  round_down(offset, 1 << blkbits)) >> blkbits,
4789                                 new_size, flags);
4790                 if (ret)
4791                         goto out_mutex;
4792 
4793         }
4794 
4795         /* Zero range excluding the unaligned edges */
4796         if (max_blocks > 0) {
4797                 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4798                           EXT4_EX_NOCACHE);
4799 
4800                 /*
4801                  * Prevent page faults from reinstantiating pages we have
4802                  * released from page cache.
4803                  */
4804                 down_write(&EXT4_I(inode)->i_mmap_sem);
4805 
4806                 ret = ext4_break_layouts(inode);
4807                 if (ret) {
4808                         up_write(&EXT4_I(inode)->i_mmap_sem);
4809                         goto out_mutex;
4810                 }
4811 
4812                 ret = ext4_update_disksize_before_punch(inode, offset, len);
4813                 if (ret) {
4814                         up_write(&EXT4_I(inode)->i_mmap_sem);
4815                         goto out_mutex;
4816                 }
4817                 /* Now release the pages and zero block aligned part of pages */
4818                 truncate_pagecache_range(inode, start, end - 1);
4819                 inode->i_mtime = inode->i_ctime = current_time(inode);
4820 
4821                 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4822                                              flags);
4823                 up_write(&EXT4_I(inode)->i_mmap_sem);
4824                 if (ret)
4825                         goto out_mutex;
4826         }
4827         if (!partial_begin && !partial_end)
4828                 goto out_mutex;
4829 
4830         /*
4831          * In worst case we have to writeout two nonadjacent unwritten
4832          * blocks and update the inode
4833          */
4834         credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4835         if (ext4_should_journal_data(inode))
4836                 credits += 2;
4837         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4838         if (IS_ERR(handle)) {
4839                 ret = PTR_ERR(handle);
4840                 ext4_std_error(inode->i_sb, ret);
4841                 goto out_mutex;
4842         }
4843 
4844         inode->i_mtime = inode->i_ctime = current_time(inode);
4845         if (new_size) {
4846                 ext4_update_inode_size(inode, new_size);
4847         } else {
4848                 /*
4849                 * Mark that we allocate beyond EOF so the subsequent truncate
4850                 * can proceed even if the new size is the same as i_size.
4851                 */
4852                 if ((offset + len) > i_size_read(inode))
4853                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4854         }
4855         ext4_mark_inode_dirty(handle, inode);
4856 
4857         /* Zero out partial block at the edges of the range */
4858         ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4859         if (ret >= 0)
4860                 ext4_update_inode_fsync_trans(handle, inode, 1);
4861 
4862         if (file->f_flags & O_SYNC)
4863                 ext4_handle_sync(handle);
4864 
4865         ext4_journal_stop(handle);
4866 out_mutex:
4867         inode_unlock(inode);
4868         return ret;
4869 }
4870 
4871 /*
4872  * preallocate space for a file. This implements ext4's fallocate file
4873  * operation, which gets called from sys_fallocate system call.
4874  * For block-mapped files, posix_fallocate should fall back to the method
4875  * of writing zeroes to the required new blocks (the same behavior which is
4876  * expected for file systems which do not support fallocate() system call).
4877  */
4878 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4879 {
4880         struct inode *inode = file_inode(file);
4881         loff_t new_size = 0;
4882         unsigned int max_blocks;
4883         int ret = 0;
4884         int flags;
4885         ext4_lblk_t lblk;
4886         unsigned int blkbits = inode->i_blkbits;
4887 
4888         /*
4889          * Encrypted inodes can't handle collapse range or insert
4890          * range since we would need to re-encrypt blocks with a
4891          * different IV or XTS tweak (which are based on the logical
4892          * block number).
4893          *
4894          * XXX It's not clear why zero range isn't working, but we'll
4895          * leave it disabled for encrypted inodes for now.  This is a
4896          * bug we should fix....
4897          */
4898         if (IS_ENCRYPTED(inode) &&
4899             (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4900                      FALLOC_FL_ZERO_RANGE)))
4901                 return -EOPNOTSUPP;
4902 
4903         /* Return error if mode is not supported */
4904         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4905                      FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4906                      FALLOC_FL_INSERT_RANGE))
4907                 return -EOPNOTSUPP;
4908 
4909         if (mode & FALLOC_FL_PUNCH_HOLE)
4910                 return ext4_punch_hole(inode, offset, len);
4911 
4912         ret = ext4_convert_inline_data(inode);
4913         if (ret)
4914                 return ret;
4915 
4916         if (mode & FALLOC_FL_COLLAPSE_RANGE)
4917                 return ext4_collapse_range(inode, offset, len);
4918 
4919         if (mode & FALLOC_FL_INSERT_RANGE)
4920                 return ext4_insert_range(inode, offset, len);
4921 
4922         if (mode & FALLOC_FL_ZERO_RANGE)
4923                 return ext4_zero_range(file, offset, len, mode);
4924 
4925         trace_ext4_fallocate_enter(inode, offset, len, mode);
4926         lblk = offset >> blkbits;
4927 
4928         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4929         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4930         if (mode & FALLOC_FL_KEEP_SIZE)
4931                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4932 
4933         inode_lock(inode);
4934 
4935         /*
4936          * We only support preallocation for extent-based files only
4937          */
4938         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4939                 ret = -EOPNOTSUPP;
4940                 goto out;
4941         }
4942 
4943         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4944             (offset + len > i_size_read(inode) ||
4945              offset + len > EXT4_I(inode)->i_disksize)) {
4946                 new_size = offset + len;
4947                 ret = inode_newsize_ok(inode, new_size);
4948                 if (ret)
4949                         goto out;
4950         }
4951 
4952         /* Wait all existing dio workers, newcomers will block on i_mutex */
4953         inode_dio_wait(inode);
4954 
4955         ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4956         if (ret)
4957                 goto out;
4958 
4959         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4960                 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4961                                                 EXT4_I(inode)->i_sync_tid);
4962         }
4963 out:
4964         inode_unlock(inode);
4965         trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4966         return ret;
4967 }
4968 
4969 /*
4970  * This function convert a range of blocks to written extents
4971  * The caller of this function will pass the start offset and the size.
4972  * all unwritten extents within this range will be converted to
4973  * written extents.
4974  *
4975  * This function is called from the direct IO end io call back
4976  * function, to convert the fallocated extents after IO is completed.
4977  * Returns 0 on success.
4978  */
4979 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4980                                    loff_t offset, ssize_t len)
4981 {
4982         unsigned int max_blocks;
4983         int ret = 0;
4984         int ret2 = 0;
4985         struct ext4_map_blocks map;
4986         unsigned int blkbits = inode->i_blkbits;
4987         unsigned int credits = 0;
4988 
4989         map.m_lblk = offset >> blkbits;
4990         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4991 
4992         if (!handle) {
4993                 /*
4994                  * credits to insert 1 extent into extent tree
4995                  */
4996                 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4997         }
4998         while (ret >= 0 && ret < max_blocks) {
4999                 map.m_lblk += ret;
5000                 map.m_len = (max_blocks -= ret);
5001                 if (credits) {
5002                         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5003                                                     credits);
5004                         if (IS_ERR(handle)) {
5005                                 ret = PTR_ERR(handle);
5006                                 break;
5007                         }
5008                 }
5009                 ret = ext4_map_blocks(handle, inode, &map,
5010                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5011                 if (ret <= 0)
5012                         ext4_warning(inode->i_sb,
5013                                      "inode #%lu: block %u: len %u: "
5014                                      "ext4_ext_map_blocks returned %d",
5015                                      inode->i_ino, map.m_lblk,
5016                                      map.m_len, ret);
5017                 ext4_mark_inode_dirty(handle, inode);
5018                 if (credits)
5019                         ret2 = ext4_journal_stop(handle);
5020                 if (ret <= 0 || ret2)
5021                         break;
5022         }
5023         return ret > 0 ? ret2 : ret;
5024 }
5025 
5026 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
5027 {
5028         int ret, err = 0;
5029         struct ext4_io_end_vec *io_end_vec;
5030 
5031         /*
5032          * This is somewhat ugly but the idea is clear: When transaction is
5033          * reserved, everything goes into it. Otherwise we rather start several
5034          * smaller transactions for conversion of each extent separately.
5035          */
5036         if (handle) {
5037                 handle = ext4_journal_start_reserved(handle,
5038                                                      EXT4_HT_EXT_CONVERT);
5039                 if (IS_ERR(handle))
5040                         return PTR_ERR(handle);
5041         }
5042 
5043         list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
5044                 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
5045                                                      io_end_vec->offset,
5046                                                      io_end_vec->size);
5047                 if (ret)
5048                         break;
5049         }
5050 
5051         if (handle)
5052                 err = ext4_journal_stop(handle);
5053 
5054         return ret < 0 ? ret : err;
5055 }
5056 
5057 /*
5058  * If newes is not existing extent (newes->ec_pblk equals zero) find
5059  * delayed extent at start of newes and update newes accordingly and
5060  * return start of the next delayed extent.
5061  *
5062  * If newes is existing extent (newes->ec_pblk is not equal zero)
5063  * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5064  * extent found. Leave newes unmodified.
5065  */
5066 static int ext4_find_delayed_extent(struct inode *inode,
5067                                     struct extent_status *newes)
5068 {
5069         struct extent_status es;
5070         ext4_lblk_t block, next_del;
5071 
5072         if (newes->es_pblk == 0) {
5073                 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5074                                           newes->es_lblk,
5075                                           newes->es_lblk + newes->es_len - 1,
5076                                           &es);
5077 
5078                 /*
5079                  * No extent in extent-tree contains block @newes->es_pblk,
5080                  * then the block may stay in 1)a hole or 2)delayed-extent.
5081                  */
5082                 if (es.es_len == 0)
5083                         /* A hole found. */
5084                         return 0;
5085 
5086                 if (es.es_lblk > newes->es_lblk) {
5087                         /* A hole found. */
5088                         newes->es_len = min(es.es_lblk - newes->es_lblk,
5089                                             newes->es_len);
5090                         return 0;
5091                 }
5092 
5093                 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5094         }
5095 
5096         block = newes->es_lblk + newes->es_len;
5097         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5098                                   EXT_MAX_BLOCKS, &es);
5099         if (es.es_len == 0)
5100                 next_del = EXT_MAX_BLOCKS;
5101         else
5102                 next_del = es.es_lblk;
5103 
5104         return next_del;
5105 }
5106 
5107 static int ext4_xattr_fiemap(struct inode *inode,
5108                                 struct fiemap_extent_info *fieinfo)
5109 {
5110         __u64 physical = 0;
5111         __u64 length;
5112         __u32 flags = FIEMAP_EXTENT_LAST;
5113         int blockbits = inode->i_sb->s_blocksize_bits;
5114         int error = 0;
5115 
5116         /* in-inode? */
5117         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5118                 struct ext4_iloc iloc;
5119                 int offset;     /* offset of xattr in inode */
5120 
5121                 error = ext4_get_inode_loc(inode, &iloc);
5122                 if (error)
5123                         return error;
5124                 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5125                 offset = EXT4_GOOD_OLD_INODE_SIZE +
5126                                 EXT4_I(inode)->i_extra_isize;
5127                 physical += offset;
5128                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5129                 flags |= FIEMAP_EXTENT_DATA_INLINE;
5130                 brelse(iloc.bh);
5131         } else { /* external block */
5132                 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5133                 length = inode->i_sb->s_blocksize;
5134         }
5135 
5136         if (physical)
5137                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5138                                                 length, flags);
5139         return (error < 0 ? error : 0);
5140 }
5141 
5142 static int _ext4_fiemap(struct inode *inode,
5143                         struct fiemap_extent_info *fieinfo,
5144                         __u64 start, __u64 len,
5145                         int (*fill)(struct inode *, ext4_lblk_t,
5146                                     ext4_lblk_t,
5147                                     struct fiemap_extent_info *))
5148 {
5149         ext4_lblk_t start_blk;
5150         u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5151 
5152         int error = 0;
5153 
5154         if (ext4_has_inline_data(inode)) {
5155                 int has_inline = 1;
5156 
5157                 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5158                                                 start, len);
5159 
5160                 if (has_inline)
5161                         return error;
5162         }
5163 
5164         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5165                 error = ext4_ext_precache(inode);
5166                 if (error)
5167                         return error;
5168                 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5169         }
5170 
5171         /* fallback to generic here if not in extents fmt */
5172         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5173             fill == ext4_fill_fiemap_extents)
5174                 return generic_block_fiemap(inode, fieinfo, start, len,
5175                         ext4_get_block);
5176 
5177         if (fill == ext4_fill_es_cache_info)
5178                 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5179         if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5180                 return -EBADR;
5181 
5182         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5183                 error = ext4_xattr_fiemap(inode, fieinfo);
5184         } else {
5185                 ext4_lblk_t len_blks;
5186                 __u64 last_blk;
5187 
5188                 start_blk = start >> inode->i_sb->s_blocksize_bits;
5189                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5190                 if (last_blk >= EXT_MAX_BLOCKS)
5191                         last_blk = EXT_MAX_BLOCKS-1;
5192                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5193 
5194                 /*
5195                  * Walk the extent tree gathering extent information
5196                  * and pushing extents back to the user.
5197                  */
5198                 error = fill(inode, start_blk, len_blks, fieinfo);
5199         }
5200         return error;
5201 }
5202 
5203 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5204                 __u64 start, __u64 len)
5205 {
5206         return _ext4_fiemap(inode, fieinfo, start, len,
5207                             ext4_fill_fiemap_extents);
5208 }
5209 
5210 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5211                       __u64 start, __u64 len)
5212 {
5213         if (ext4_has_inline_data(inode)) {
5214                 int has_inline;
5215 
5216                 down_read(&EXT4_I(inode)->xattr_sem);
5217                 has_inline = ext4_has_inline_data(inode);
5218                 up_read(&EXT4_I(inode)->xattr_sem);
5219                 if (has_inline)
5220                         return 0;
5221         }
5222 
5223         return _ext4_fiemap(inode, fieinfo, start, len,
5224                             ext4_fill_es_cache_info);
5225 }
5226 
5227 
5228 /*
5229  * ext4_access_path:
5230  * Function to access the path buffer for marking it dirty.
5231  * It also checks if there are sufficient credits left in the journal handle
5232  * to update path.
5233  */
5234 static int
5235 ext4_access_path(handle_t *handle, struct inode *inode,
5236                 struct ext4_ext_path *path)
5237 {
5238         int credits, err;
5239 
5240         if (!ext4_handle_valid(handle))
5241                 return 0;
5242 
5243         /*
5244          * Check if need to extend journal credits
5245          * 3 for leaf, sb, and inode plus 2 (bmap and group
5246          * descriptor) for each block group; assume two block
5247          * groups
5248          */
5249         credits = ext4_writepage_trans_blocks(inode);
5250         err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
5251         if (err < 0)
5252                 return err;
5253 
5254         err = ext4_ext_get_access(handle, inode, path);
5255         return err;
5256 }
5257 
5258 /*
5259  * ext4_ext_shift_path_extents:
5260  * Shift the extents of a path structure lying between path[depth].p_ext
5261  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5262  * if it is right shift or left shift operation.
5263  */
5264 static int
5265 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5266                             struct inode *inode, handle_t *handle,
5267                             enum SHIFT_DIRECTION SHIFT)
5268 {
5269         int depth, err = 0;
5270         struct ext4_extent *ex_start, *ex_last;
5271         bool update = 0;
5272         depth = path->p_depth;
5273 
5274         while (depth >= 0) {
5275                 if (depth == path->p_depth) {
5276                         ex_start = path[depth].p_ext;
5277                         if (!ex_start)
5278                                 return -EFSCORRUPTED;
5279 
5280                         ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5281 
5282                         err = ext4_access_path(handle, inode, path + depth);
5283                         if (err)
5284                                 goto out;
5285 
5286                         if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5287                                 update = 1;
5288 
5289                         while (ex_start <= ex_last) {
5290                                 if (SHIFT == SHIFT_LEFT) {
5291                                         le32_add_cpu(&ex_start->ee_block,
5292                                                 -shift);
5293                                         /* Try to merge to the left. */
5294                                         if ((ex_start >
5295                                             EXT_FIRST_EXTENT(path[depth].p_hdr))
5296                                             &&
5297                                             ext4_ext_try_to_merge_right(inode,
5298                                             path, ex_start - 1))
5299                                                 ex_last--;
5300                                         else
5301                                                 ex_start++;
5302                                 } else {
5303                                         le32_add_cpu(&ex_last->ee_block, shift);
5304                                         ext4_ext_try_to_merge_right(inode, path,
5305                                                 ex_last);
5306                                         ex_last--;
5307                                 }
5308                         }
5309                         err = ext4_ext_dirty(handle, inode, path + depth);
5310                         if (err)
5311                                 goto out;
5312 
5313                         if (--depth < 0 || !update)
5314                                 break;
5315                 }
5316 
5317                 /* Update index too */
5318                 err = ext4_access_path(handle, inode, path + depth);
5319                 if (err)
5320                         goto out;
5321 
5322                 if (SHIFT == SHIFT_LEFT)
5323                         le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5324                 else
5325                         le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5326                 err = ext4_ext_dirty(handle, inode, path + depth);
5327                 if (err)
5328                         goto out;
5329 
5330                 /* we are done if current index is not a starting index */
5331                 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5332                         break;
5333 
5334                 depth--;
5335         }
5336 
5337 out:
5338         return err;
5339 }
5340 
5341 /*
5342  * ext4_ext_shift_extents:
5343  * All the extents which lies in the range from @start to the last allocated
5344  * block for the @inode are shifted either towards left or right (depending
5345  * upon @SHIFT) by @shift blocks.
5346  * On success, 0 is returned, error otherwise.
5347  */
5348 static int
5349 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5350                        ext4_lblk_t start, ext4_lblk_t shift,
5351                        enum SHIFT_DIRECTION SHIFT)
5352 {
5353         struct ext4_ext_path *path;
5354         int ret = 0, depth;
5355         struct ext4_extent *extent;
5356         ext4_lblk_t stop, *iterator, ex_start, ex_end;
5357 
5358         /* Let path point to the last extent */
5359         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5360                                 EXT4_EX_NOCACHE);
5361         if (IS_ERR(path))
5362                 return PTR_ERR(path);
5363 
5364         depth = path->p_depth;
5365         extent = path[depth].p_ext;
5366         if (!extent)
5367                 goto out;
5368 
5369         stop = le32_to_cpu(extent->ee_block);
5370 
5371        /*
5372         * For left shifts, make sure the hole on the left is big enough to
5373         * accommodate the shift.  For right shifts, make sure the last extent
5374         * won't be shifted beyond EXT_MAX_BLOCKS.
5375         */
5376         if (SHIFT == SHIFT_LEFT) {
5377                 path = ext4_find_extent(inode, start - 1, &path,
5378                                         EXT4_EX_NOCACHE);
5379                 if (IS_ERR(path))
5380                         return PTR_ERR(path);
5381                 depth = path->p_depth;
5382                 extent =  path[depth].p_ext;
5383                 if (extent) {
5384                         ex_start = le32_to_cpu(extent->ee_block);
5385                         ex_end = le32_to_cpu(extent->ee_block) +
5386                                 ext4_ext_get_actual_len(extent);
5387                 } else {
5388                         ex_start = 0;
5389                         ex_end = 0;
5390                 }
5391 
5392                 if ((start == ex_start && shift > ex_start) ||
5393                     (shift > start - ex_end)) {
5394                         ret = -EINVAL;
5395                         goto out;
5396                 }
5397         } else {
5398                 if (shift > EXT_MAX_BLOCKS -
5399                     (stop + ext4_ext_get_actual_len(extent))) {
5400                         ret = -EINVAL;
5401                         goto out;
5402                 }
5403         }
5404 
5405         /*
5406          * In case of left shift, iterator points to start and it is increased
5407          * till we reach stop. In case of right shift, iterator points to stop
5408          * and it is decreased till we reach start.
5409          */
5410         if (SHIFT == SHIFT_LEFT)
5411                 iterator = &start;
5412         else
5413                 iterator = &stop;
5414 
5415         /*
5416          * Its safe to start updating extents.  Start and stop are unsigned, so
5417          * in case of right shift if extent with 0 block is reached, iterator
5418          * becomes NULL to indicate the end of the loop.
5419          */
5420         while (iterator && start <= stop) {
5421                 path = ext4_find_extent(inode, *iterator, &path,
5422                                         EXT4_EX_NOCACHE);
5423                 if (IS_ERR(path))
5424                         return PTR_ERR(path);
5425                 depth = path->p_depth;
5426                 extent = path[depth].p_ext;
5427                 if (!extent) {
5428                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5429                                          (unsigned long) *iterator);
5430                         return -EFSCORRUPTED;
5431                 }
5432                 if (SHIFT == SHIFT_LEFT && *iterator >
5433                     le32_to_cpu(extent->ee_block)) {
5434                         /* Hole, move to the next extent */
5435                         if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5436                                 path[depth].p_ext++;
5437                         } else {
5438                                 *iterator = ext4_ext_next_allocated_block(path);
5439                                 continue;
5440                         }
5441                 }
5442 
5443                 if (SHIFT == SHIFT_LEFT) {
5444                         extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5445                         *iterator = le32_to_cpu(extent->ee_block) +
5446                                         ext4_ext_get_actual_len(extent);
5447                 } else {
5448                         extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5449                         if (le32_to_cpu(extent->ee_block) > 0)
5450                                 *iterator = le32_to_cpu(extent->ee_block) - 1;
5451                         else
5452                                 /* Beginning is reached, end of the loop */
5453                                 iterator = NULL;
5454                         /* Update path extent in case we need to stop */
5455                         while (le32_to_cpu(extent->ee_block) < start)
5456                                 extent++;
5457                         path[depth].p_ext = extent;
5458                 }
5459                 ret = ext4_ext_shift_path_extents(path, shift, inode,
5460                                 handle, SHIFT);
5461                 if (ret)
5462                         break;
5463         }
5464 out:
5465         ext4_ext_drop_refs(path);
5466         kfree(path);
5467         return ret;
5468 }
5469 
5470 /*
5471  * ext4_collapse_range:
5472  * This implements the fallocate's collapse range functionality for ext4
5473  * Returns: 0 and non-zero on error.
5474  */
5475 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5476 {
5477         struct super_block *sb = inode->i_sb;
5478         ext4_lblk_t punch_start, punch_stop;
5479         handle_t *handle;
5480         unsigned int credits;
5481         loff_t new_size, ioffset;
5482         int ret;
5483 
5484         /*
5485          * We need to test this early because xfstests assumes that a
5486          * collapse range of (0, 1) will return EOPNOTSUPP if the file
5487          * system does not support collapse range.
5488          */
5489         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5490                 return -EOPNOTSUPP;
5491 
5492         /* Collapse range works only on fs block size aligned offsets. */
5493         if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5494             len & (EXT4_CLUSTER_SIZE(sb) - 1))
5495                 return -EINVAL;
5496 
5497         if (!S_ISREG(inode->i_mode))
5498                 return -EINVAL;
5499 
5500         trace_ext4_collapse_range(inode, offset, len);
5501 
5502         punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5503         punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5504 
5505         /* Call ext4_force_commit to flush all data in case of data=journal. */
5506         if (ext4_should_journal_data(inode)) {
5507                 ret = ext4_force_commit(inode->i_sb);
5508                 if (ret)
5509                         return ret;
5510         }
5511 
5512         inode_lock(inode);
5513         /*
5514          * There is no need to overlap collapse range with EOF, in which case
5515          * it is effectively a truncate operation
5516          */
5517         if (offset + len >= i_size_read(inode)) {
5518                 ret = -EINVAL;
5519                 goto out_mutex;
5520         }
5521 
5522         /* Currently just for extent based files */
5523         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5524                 ret = -EOPNOTSUPP;
5525                 goto out_mutex;
5526         }
5527 
5528         /* Wait for existing dio to complete */
5529         inode_dio_wait(inode);
5530 
5531         /*
5532          * Prevent page faults from reinstantiating pages we have released from
5533          * page cache.
5534          */
5535         down_write(&EXT4_I(inode)->i_mmap_sem);
5536 
5537         ret = ext4_break_layouts(inode);
5538         if (ret)
5539                 goto out_mmap;
5540 
5541         /*
5542          * Need to round down offset to be aligned with page size boundary
5543          * for page size > block size.
5544          */
5545         ioffset = round_down(offset, PAGE_SIZE);
5546         /*
5547          * Write tail of the last page before removed range since it will get
5548          * removed from the page cache below.
5549          */
5550         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5551         if (ret)
5552                 goto out_mmap;
5553         /*
5554          * Write data that will be shifted to preserve them when discarding
5555          * page cache below. We are also protected from pages becoming dirty
5556          * by i_mmap_sem.
5557          */
5558         ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5559                                            LLONG_MAX);
5560         if (ret)
5561                 goto out_mmap;
5562         truncate_pagecache(inode, ioffset);
5563 
5564         credits = ext4_writepage_trans_blocks(inode);
5565         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5566         if (IS_ERR(handle)) {
5567                 ret = PTR_ERR(handle);
5568                 goto out_mmap;
5569         }
5570 
5571         down_write(&EXT4_I(inode)->i_data_sem);
5572         ext4_discard_preallocations(inode);
5573 
5574         ret = ext4_es_remove_extent(inode, punch_start,
5575                                     EXT_MAX_BLOCKS - punch_start);
5576         if (ret) {
5577                 up_write(&EXT4_I(inode)->i_data_sem);
5578                 goto out_stop;
5579         }
5580 
5581         ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5582         if (ret) {
5583                 up_write(&EXT4_I(inode)->i_data_sem);
5584                 goto out_stop;
5585         }
5586         ext4_discard_preallocations(inode);
5587 
5588         ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5589                                      punch_stop - punch_start, SHIFT_LEFT);
5590         if (ret) {
5591                 up_write(&EXT4_I(inode)->i_data_sem);
5592                 goto out_stop;
5593         }
5594 
5595         new_size = i_size_read(inode) - len;
5596         i_size_write(inode, new_size);
5597         EXT4_I(inode)->i_disksize = new_size;
5598 
5599         up_write(&EXT4_I(inode)->i_data_sem);
5600         if (IS_SYNC(inode))
5601                 ext4_handle_sync(handle);
5602         inode->i_mtime = inode->i_ctime = current_time(inode);
5603         ext4_mark_inode_dirty(handle, inode);
5604         ext4_update_inode_fsync_trans(handle, inode, 1);
5605 
5606 out_stop:
5607         ext4_journal_stop(handle);
5608 out_mmap:
5609         up_write(&EXT4_I(inode)->i_mmap_sem);
5610 out_mutex:
5611         inode_unlock(inode);
5612         return ret;
5613 }
5614 
5615 /*
5616  * ext4_insert_range:
5617  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5618  * The data blocks starting from @offset to the EOF are shifted by @len
5619  * towards right to create a hole in the @inode. Inode size is increased
5620  * by len bytes.
5621  * Returns 0 on success, error otherwise.
5622  */