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

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

Version: ~ [ linux-5.6 ] ~ [ linux-5.5.13 ] ~ [ linux-5.4.28 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.113 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.82 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  *  linux/fs/ext4/indirect.c
  4  *
  5  *  from
  6  *
  7  *  linux/fs/ext4/inode.c
  8  *
  9  * Copyright (C) 1992, 1993, 1994, 1995
 10  * Remy Card (card@masi.ibp.fr)
 11  * Laboratoire MASI - Institut Blaise Pascal
 12  * Universite Pierre et Marie Curie (Paris VI)
 13  *
 14  *  from
 15  *
 16  *  linux/fs/minix/inode.c
 17  *
 18  *  Copyright (C) 1991, 1992  Linus Torvalds
 19  *
 20  *  Goal-directed block allocation by Stephen Tweedie
 21  *      (sct@redhat.com), 1993, 1998
 22  */
 23 
 24 #include "ext4_jbd2.h"
 25 #include "truncate.h"
 26 #include <linux/dax.h>
 27 #include <linux/uio.h>
 28 
 29 #include <trace/events/ext4.h>
 30 
 31 typedef struct {
 32         __le32  *p;
 33         __le32  key;
 34         struct buffer_head *bh;
 35 } Indirect;
 36 
 37 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
 38 {
 39         p->key = *(p->p = v);
 40         p->bh = bh;
 41 }
 42 
 43 /**
 44  *      ext4_block_to_path - parse the block number into array of offsets
 45  *      @inode: inode in question (we are only interested in its superblock)
 46  *      @i_block: block number to be parsed
 47  *      @offsets: array to store the offsets in
 48  *      @boundary: set this non-zero if the referred-to block is likely to be
 49  *             followed (on disk) by an indirect block.
 50  *
 51  *      To store the locations of file's data ext4 uses a data structure common
 52  *      for UNIX filesystems - tree of pointers anchored in the inode, with
 53  *      data blocks at leaves and indirect blocks in intermediate nodes.
 54  *      This function translates the block number into path in that tree -
 55  *      return value is the path length and @offsets[n] is the offset of
 56  *      pointer to (n+1)th node in the nth one. If @block is out of range
 57  *      (negative or too large) warning is printed and zero returned.
 58  *
 59  *      Note: function doesn't find node addresses, so no IO is needed. All
 60  *      we need to know is the capacity of indirect blocks (taken from the
 61  *      inode->i_sb).
 62  */
 63 
 64 /*
 65  * Portability note: the last comparison (check that we fit into triple
 66  * indirect block) is spelled differently, because otherwise on an
 67  * architecture with 32-bit longs and 8Kb pages we might get into trouble
 68  * if our filesystem had 8Kb blocks. We might use long long, but that would
 69  * kill us on x86. Oh, well, at least the sign propagation does not matter -
 70  * i_block would have to be negative in the very beginning, so we would not
 71  * get there at all.
 72  */
 73 
 74 static int ext4_block_to_path(struct inode *inode,
 75                               ext4_lblk_t i_block,
 76                               ext4_lblk_t offsets[4], int *boundary)
 77 {
 78         int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
 79         int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
 80         const long direct_blocks = EXT4_NDIR_BLOCKS,
 81                 indirect_blocks = ptrs,
 82                 double_blocks = (1 << (ptrs_bits * 2));
 83         int n = 0;
 84         int final = 0;
 85 
 86         if (i_block < direct_blocks) {
 87                 offsets[n++] = i_block;
 88                 final = direct_blocks;
 89         } else if ((i_block -= direct_blocks) < indirect_blocks) {
 90                 offsets[n++] = EXT4_IND_BLOCK;
 91                 offsets[n++] = i_block;
 92                 final = ptrs;
 93         } else if ((i_block -= indirect_blocks) < double_blocks) {
 94                 offsets[n++] = EXT4_DIND_BLOCK;
 95                 offsets[n++] = i_block >> ptrs_bits;
 96                 offsets[n++] = i_block & (ptrs - 1);
 97                 final = ptrs;
 98         } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
 99                 offsets[n++] = EXT4_TIND_BLOCK;
100                 offsets[n++] = i_block >> (ptrs_bits * 2);
101                 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
102                 offsets[n++] = i_block & (ptrs - 1);
103                 final = ptrs;
104         } else {
105                 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
106                              i_block + direct_blocks +
107                              indirect_blocks + double_blocks, inode->i_ino);
108         }
109         if (boundary)
110                 *boundary = final - 1 - (i_block & (ptrs - 1));
111         return n;
112 }
113 
114 /**
115  *      ext4_get_branch - read the chain of indirect blocks leading to data
116  *      @inode: inode in question
117  *      @depth: depth of the chain (1 - direct pointer, etc.)
118  *      @offsets: offsets of pointers in inode/indirect blocks
119  *      @chain: place to store the result
120  *      @err: here we store the error value
121  *
122  *      Function fills the array of triples <key, p, bh> and returns %NULL
123  *      if everything went OK or the pointer to the last filled triple
124  *      (incomplete one) otherwise. Upon the return chain[i].key contains
125  *      the number of (i+1)-th block in the chain (as it is stored in memory,
126  *      i.e. little-endian 32-bit), chain[i].p contains the address of that
127  *      number (it points into struct inode for i==0 and into the bh->b_data
128  *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129  *      block for i>0 and NULL for i==0. In other words, it holds the block
130  *      numbers of the chain, addresses they were taken from (and where we can
131  *      verify that chain did not change) and buffer_heads hosting these
132  *      numbers.
133  *
134  *      Function stops when it stumbles upon zero pointer (absent block)
135  *              (pointer to last triple returned, *@err == 0)
136  *      or when it gets an IO error reading an indirect block
137  *              (ditto, *@err == -EIO)
138  *      or when it reads all @depth-1 indirect blocks successfully and finds
139  *      the whole chain, all way to the data (returns %NULL, *err == 0).
140  *
141  *      Need to be called with
142  *      down_read(&EXT4_I(inode)->i_data_sem)
143  */
144 static Indirect *ext4_get_branch(struct inode *inode, int depth,
145                                  ext4_lblk_t  *offsets,
146                                  Indirect chain[4], int *err)
147 {
148         struct super_block *sb = inode->i_sb;
149         Indirect *p = chain;
150         struct buffer_head *bh;
151         int ret = -EIO;
152 
153         *err = 0;
154         /* i_data is not going away, no lock needed */
155         add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
156         if (!p->key)
157                 goto no_block;
158         while (--depth) {
159                 bh = sb_getblk(sb, le32_to_cpu(p->key));
160                 if (unlikely(!bh)) {
161                         ret = -ENOMEM;
162                         goto failure;
163                 }
164 
165                 if (!bh_uptodate_or_lock(bh)) {
166                         if (bh_submit_read(bh) < 0) {
167                                 put_bh(bh);
168                                 goto failure;
169                         }
170                         /* validate block references */
171                         if (ext4_check_indirect_blockref(inode, bh)) {
172                                 put_bh(bh);
173                                 goto failure;
174                         }
175                 }
176 
177                 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
178                 /* Reader: end */
179                 if (!p->key)
180                         goto no_block;
181         }
182         return NULL;
183 
184 failure:
185         *err = ret;
186 no_block:
187         return p;
188 }
189 
190 /**
191  *      ext4_find_near - find a place for allocation with sufficient locality
192  *      @inode: owner
193  *      @ind: descriptor of indirect block.
194  *
195  *      This function returns the preferred place for block allocation.
196  *      It is used when heuristic for sequential allocation fails.
197  *      Rules are:
198  *        + if there is a block to the left of our position - allocate near it.
199  *        + if pointer will live in indirect block - allocate near that block.
200  *        + if pointer will live in inode - allocate in the same
201  *          cylinder group.
202  *
203  * In the latter case we colour the starting block by the callers PID to
204  * prevent it from clashing with concurrent allocations for a different inode
205  * in the same block group.   The PID is used here so that functionally related
206  * files will be close-by on-disk.
207  *
208  *      Caller must make sure that @ind is valid and will stay that way.
209  */
210 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
211 {
212         struct ext4_inode_info *ei = EXT4_I(inode);
213         __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
214         __le32 *p;
215 
216         /* Try to find previous block */
217         for (p = ind->p - 1; p >= start; p--) {
218                 if (*p)
219                         return le32_to_cpu(*p);
220         }
221 
222         /* No such thing, so let's try location of indirect block */
223         if (ind->bh)
224                 return ind->bh->b_blocknr;
225 
226         /*
227          * It is going to be referred to from the inode itself? OK, just put it
228          * into the same cylinder group then.
229          */
230         return ext4_inode_to_goal_block(inode);
231 }
232 
233 /**
234  *      ext4_find_goal - find a preferred place for allocation.
235  *      @inode: owner
236  *      @block:  block we want
237  *      @partial: pointer to the last triple within a chain
238  *
239  *      Normally this function find the preferred place for block allocation,
240  *      returns it.
241  *      Because this is only used for non-extent files, we limit the block nr
242  *      to 32 bits.
243  */
244 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
245                                    Indirect *partial)
246 {
247         ext4_fsblk_t goal;
248 
249         /*
250          * XXX need to get goal block from mballoc's data structures
251          */
252 
253         goal = ext4_find_near(inode, partial);
254         goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
255         return goal;
256 }
257 
258 /**
259  *      ext4_blks_to_allocate - Look up the block map and count the number
260  *      of direct blocks need to be allocated for the given branch.
261  *
262  *      @branch: chain of indirect blocks
263  *      @k: number of blocks need for indirect blocks
264  *      @blks: number of data blocks to be mapped.
265  *      @blocks_to_boundary:  the offset in the indirect block
266  *
267  *      return the total number of blocks to be allocate, including the
268  *      direct and indirect blocks.
269  */
270 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
271                                  int blocks_to_boundary)
272 {
273         unsigned int count = 0;
274 
275         /*
276          * Simple case, [t,d]Indirect block(s) has not allocated yet
277          * then it's clear blocks on that path have not allocated
278          */
279         if (k > 0) {
280                 /* right now we don't handle cross boundary allocation */
281                 if (blks < blocks_to_boundary + 1)
282                         count += blks;
283                 else
284                         count += blocks_to_boundary + 1;
285                 return count;
286         }
287 
288         count++;
289         while (count < blks && count <= blocks_to_boundary &&
290                 le32_to_cpu(*(branch[0].p + count)) == 0) {
291                 count++;
292         }
293         return count;
294 }
295 
296 /**
297  *      ext4_alloc_branch - allocate and set up a chain of blocks.
298  *      @handle: handle for this transaction
299  *      @inode: owner
300  *      @indirect_blks: number of allocated indirect blocks
301  *      @blks: number of allocated direct blocks
302  *      @goal: preferred place for allocation
303  *      @offsets: offsets (in the blocks) to store the pointers to next.
304  *      @branch: place to store the chain in.
305  *
306  *      This function allocates blocks, zeroes out all but the last one,
307  *      links them into chain and (if we are synchronous) writes them to disk.
308  *      In other words, it prepares a branch that can be spliced onto the
309  *      inode. It stores the information about that chain in the branch[], in
310  *      the same format as ext4_get_branch() would do. We are calling it after
311  *      we had read the existing part of chain and partial points to the last
312  *      triple of that (one with zero ->key). Upon the exit we have the same
313  *      picture as after the successful ext4_get_block(), except that in one
314  *      place chain is disconnected - *branch->p is still zero (we did not
315  *      set the last link), but branch->key contains the number that should
316  *      be placed into *branch->p to fill that gap.
317  *
318  *      If allocation fails we free all blocks we've allocated (and forget
319  *      their buffer_heads) and return the error value the from failed
320  *      ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
321  *      as described above and return 0.
322  */
323 static int ext4_alloc_branch(handle_t *handle,
324                              struct ext4_allocation_request *ar,
325                              int indirect_blks, ext4_lblk_t *offsets,
326                              Indirect *branch)
327 {
328         struct buffer_head *            bh;
329         ext4_fsblk_t                    b, new_blocks[4];
330         __le32                          *p;
331         int                             i, j, err, len = 1;
332 
333         for (i = 0; i <= indirect_blks; i++) {
334                 if (i == indirect_blks) {
335                         new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
336                 } else
337                         ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
338                                         ar->inode, ar->goal,
339                                         ar->flags & EXT4_MB_DELALLOC_RESERVED,
340                                         NULL, &err);
341                 if (err) {
342                         i--;
343                         goto failed;
344                 }
345                 branch[i].key = cpu_to_le32(new_blocks[i]);
346                 if (i == 0)
347                         continue;
348 
349                 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
350                 if (unlikely(!bh)) {
351                         err = -ENOMEM;
352                         goto failed;
353                 }
354                 lock_buffer(bh);
355                 BUFFER_TRACE(bh, "call get_create_access");
356                 err = ext4_journal_get_create_access(handle, bh);
357                 if (err) {
358                         unlock_buffer(bh);
359                         goto failed;
360                 }
361 
362                 memset(bh->b_data, 0, bh->b_size);
363                 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
364                 b = new_blocks[i];
365 
366                 if (i == indirect_blks)
367                         len = ar->len;
368                 for (j = 0; j < len; j++)
369                         *p++ = cpu_to_le32(b++);
370 
371                 BUFFER_TRACE(bh, "marking uptodate");
372                 set_buffer_uptodate(bh);
373                 unlock_buffer(bh);
374 
375                 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
376                 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
377                 if (err)
378                         goto failed;
379         }
380         return 0;
381 failed:
382         for (; i >= 0; i--) {
383                 /*
384                  * We want to ext4_forget() only freshly allocated indirect
385                  * blocks.  Buffer for new_blocks[i-1] is at branch[i].bh and
386                  * buffer at branch[0].bh is indirect block / inode already
387                  * existing before ext4_alloc_branch() was called.
388                  */
389                 if (i > 0 && i != indirect_blks && branch[i].bh)
390                         ext4_forget(handle, 1, ar->inode, branch[i].bh,
391                                     branch[i].bh->b_blocknr);
392                 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
393                                  (i == indirect_blks) ? ar->len : 1, 0);
394         }
395         return err;
396 }
397 
398 /**
399  * ext4_splice_branch - splice the allocated branch onto inode.
400  * @handle: handle for this transaction
401  * @inode: owner
402  * @block: (logical) number of block we are adding
403  * @chain: chain of indirect blocks (with a missing link - see
404  *      ext4_alloc_branch)
405  * @where: location of missing link
406  * @num:   number of indirect blocks we are adding
407  * @blks:  number of direct blocks we are adding
408  *
409  * This function fills the missing link and does all housekeeping needed in
410  * inode (->i_blocks, etc.). In case of success we end up with the full
411  * chain to new block and return 0.
412  */
413 static int ext4_splice_branch(handle_t *handle,
414                               struct ext4_allocation_request *ar,
415                               Indirect *where, int num)
416 {
417         int i;
418         int err = 0;
419         ext4_fsblk_t current_block;
420 
421         /*
422          * If we're splicing into a [td]indirect block (as opposed to the
423          * inode) then we need to get write access to the [td]indirect block
424          * before the splice.
425          */
426         if (where->bh) {
427                 BUFFER_TRACE(where->bh, "get_write_access");
428                 err = ext4_journal_get_write_access(handle, where->bh);
429                 if (err)
430                         goto err_out;
431         }
432         /* That's it */
433 
434         *where->p = where->key;
435 
436         /*
437          * Update the host buffer_head or inode to point to more just allocated
438          * direct blocks blocks
439          */
440         if (num == 0 && ar->len > 1) {
441                 current_block = le32_to_cpu(where->key) + 1;
442                 for (i = 1; i < ar->len; i++)
443                         *(where->p + i) = cpu_to_le32(current_block++);
444         }
445 
446         /* We are done with atomic stuff, now do the rest of housekeeping */
447         /* had we spliced it onto indirect block? */
448         if (where->bh) {
449                 /*
450                  * If we spliced it onto an indirect block, we haven't
451                  * altered the inode.  Note however that if it is being spliced
452                  * onto an indirect block at the very end of the file (the
453                  * file is growing) then we *will* alter the inode to reflect
454                  * the new i_size.  But that is not done here - it is done in
455                  * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
456                  */
457                 jbd_debug(5, "splicing indirect only\n");
458                 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
459                 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
460                 if (err)
461                         goto err_out;
462         } else {
463                 /*
464                  * OK, we spliced it into the inode itself on a direct block.
465                  */
466                 ext4_mark_inode_dirty(handle, ar->inode);
467                 jbd_debug(5, "splicing direct\n");
468         }
469         return err;
470 
471 err_out:
472         for (i = 1; i <= num; i++) {
473                 /*
474                  * branch[i].bh is newly allocated, so there is no
475                  * need to revoke the block, which is why we don't
476                  * need to set EXT4_FREE_BLOCKS_METADATA.
477                  */
478                 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
479                                  EXT4_FREE_BLOCKS_FORGET);
480         }
481         ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
482                          ar->len, 0);
483 
484         return err;
485 }
486 
487 /*
488  * The ext4_ind_map_blocks() function handles non-extents inodes
489  * (i.e., using the traditional indirect/double-indirect i_blocks
490  * scheme) for ext4_map_blocks().
491  *
492  * Allocation strategy is simple: if we have to allocate something, we will
493  * have to go the whole way to leaf. So let's do it before attaching anything
494  * to tree, set linkage between the newborn blocks, write them if sync is
495  * required, recheck the path, free and repeat if check fails, otherwise
496  * set the last missing link (that will protect us from any truncate-generated
497  * removals - all blocks on the path are immune now) and possibly force the
498  * write on the parent block.
499  * That has a nice additional property: no special recovery from the failed
500  * allocations is needed - we simply release blocks and do not touch anything
501  * reachable from inode.
502  *
503  * `handle' can be NULL if create == 0.
504  *
505  * return > 0, # of blocks mapped or allocated.
506  * return = 0, if plain lookup failed.
507  * return < 0, error case.
508  *
509  * The ext4_ind_get_blocks() function should be called with
510  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
511  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
512  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
513  * blocks.
514  */
515 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
516                         struct ext4_map_blocks *map,
517                         int flags)
518 {
519         struct ext4_allocation_request ar;
520         int err = -EIO;
521         ext4_lblk_t offsets[4];
522         Indirect chain[4];
523         Indirect *partial;
524         int indirect_blks;
525         int blocks_to_boundary = 0;
526         int depth;
527         int count = 0;
528         ext4_fsblk_t first_block = 0;
529 
530         trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
531         J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
532         J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
533         depth = ext4_block_to_path(inode, map->m_lblk, offsets,
534                                    &blocks_to_boundary);
535 
536         if (depth == 0)
537                 goto out;
538 
539         partial = ext4_get_branch(inode, depth, offsets, chain, &err);
540 
541         /* Simplest case - block found, no allocation needed */
542         if (!partial) {
543                 first_block = le32_to_cpu(chain[depth - 1].key);
544                 count++;
545                 /*map more blocks*/
546                 while (count < map->m_len && count <= blocks_to_boundary) {
547                         ext4_fsblk_t blk;
548 
549                         blk = le32_to_cpu(*(chain[depth-1].p + count));
550 
551                         if (blk == first_block + count)
552                                 count++;
553                         else
554                                 break;
555                 }
556                 goto got_it;
557         }
558 
559         /* Next simple case - plain lookup failed */
560         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
561                 unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
562                 int i;
563 
564                 /*
565                  * Count number blocks in a subtree under 'partial'. At each
566                  * level we count number of complete empty subtrees beyond
567                  * current offset and then descend into the subtree only
568                  * partially beyond current offset.
569                  */
570                 count = 0;
571                 for (i = partial - chain + 1; i < depth; i++)
572                         count = count * epb + (epb - offsets[i] - 1);
573                 count++;
574                 /* Fill in size of a hole we found */
575                 map->m_pblk = 0;
576                 map->m_len = min_t(unsigned int, map->m_len, count);
577                 goto cleanup;
578         }
579 
580         /* Failed read of indirect block */
581         if (err == -EIO)
582                 goto cleanup;
583 
584         /*
585          * Okay, we need to do block allocation.
586         */
587         if (ext4_has_feature_bigalloc(inode->i_sb)) {
588                 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
589                                  "non-extent mapped inodes with bigalloc");
590                 return -EFSCORRUPTED;
591         }
592 
593         /* Set up for the direct block allocation */
594         memset(&ar, 0, sizeof(ar));
595         ar.inode = inode;
596         ar.logical = map->m_lblk;
597         if (S_ISREG(inode->i_mode))
598                 ar.flags = EXT4_MB_HINT_DATA;
599         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
600                 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
601         if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
602                 ar.flags |= EXT4_MB_USE_RESERVED;
603 
604         ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
605 
606         /* the number of blocks need to allocate for [d,t]indirect blocks */
607         indirect_blks = (chain + depth) - partial - 1;
608 
609         /*
610          * Next look up the indirect map to count the totoal number of
611          * direct blocks to allocate for this branch.
612          */
613         ar.len = ext4_blks_to_allocate(partial, indirect_blks,
614                                        map->m_len, blocks_to_boundary);
615 
616         /*
617          * Block out ext4_truncate while we alter the tree
618          */
619         err = ext4_alloc_branch(handle, &ar, indirect_blks,
620                                 offsets + (partial - chain), partial);
621 
622         /*
623          * The ext4_splice_branch call will free and forget any buffers
624          * on the new chain if there is a failure, but that risks using
625          * up transaction credits, especially for bitmaps where the
626          * credits cannot be returned.  Can we handle this somehow?  We
627          * may need to return -EAGAIN upwards in the worst case.  --sct
628          */
629         if (!err)
630                 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
631         if (err)
632                 goto cleanup;
633 
634         map->m_flags |= EXT4_MAP_NEW;
635 
636         ext4_update_inode_fsync_trans(handle, inode, 1);
637         count = ar.len;
638 got_it:
639         map->m_flags |= EXT4_MAP_MAPPED;
640         map->m_pblk = le32_to_cpu(chain[depth-1].key);
641         map->m_len = count;
642         if (count > blocks_to_boundary)
643                 map->m_flags |= EXT4_MAP_BOUNDARY;
644         err = count;
645         /* Clean up and exit */
646         partial = chain + depth - 1;    /* the whole chain */
647 cleanup:
648         while (partial > chain) {
649                 BUFFER_TRACE(partial->bh, "call brelse");
650                 brelse(partial->bh);
651                 partial--;
652         }
653 out:
654         trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
655         return err;
656 }
657 
658 /*
659  * Calculate the number of metadata blocks need to reserve
660  * to allocate a new block at @lblocks for non extent file based file
661  */
662 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
663 {
664         struct ext4_inode_info *ei = EXT4_I(inode);
665         sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
666         int blk_bits;
667 
668         if (lblock < EXT4_NDIR_BLOCKS)
669                 return 0;
670 
671         lblock -= EXT4_NDIR_BLOCKS;
672 
673         if (ei->i_da_metadata_calc_len &&
674             (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
675                 ei->i_da_metadata_calc_len++;
676                 return 0;
677         }
678         ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
679         ei->i_da_metadata_calc_len = 1;
680         blk_bits = order_base_2(lblock);
681         return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
682 }
683 
684 /*
685  * Calculate number of indirect blocks touched by mapping @nrblocks logically
686  * contiguous blocks
687  */
688 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
689 {
690         /*
691          * With N contiguous data blocks, we need at most
692          * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
693          * 2 dindirect blocks, and 1 tindirect block
694          */
695         return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
696 }
697 
698 /*
699  * Truncate transactions can be complex and absolutely huge.  So we need to
700  * be able to restart the transaction at a conventient checkpoint to make
701  * sure we don't overflow the journal.
702  *
703  * Try to extend this transaction for the purposes of truncation.  If
704  * extend fails, we need to propagate the failure up and restart the
705  * transaction in the top-level truncate loop. --sct
706  *
707  * Returns 0 if we managed to create more room.  If we can't create more
708  * room, and the transaction must be restarted we return 1.
709  */
710 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
711 {
712         if (!ext4_handle_valid(handle))
713                 return 0;
714         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
715                 return 0;
716         if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
717                 return 0;
718         return 1;
719 }
720 
721 /*
722  * Probably it should be a library function... search for first non-zero word
723  * or memcmp with zero_page, whatever is better for particular architecture.
724  * Linus?
725  */
726 static inline int all_zeroes(__le32 *p, __le32 *q)
727 {
728         while (p < q)
729                 if (*p++)
730                         return 0;
731         return 1;
732 }
733 
734 /**
735  *      ext4_find_shared - find the indirect blocks for partial truncation.
736  *      @inode:   inode in question
737  *      @depth:   depth of the affected branch
738  *      @offsets: offsets of pointers in that branch (see ext4_block_to_path)
739  *      @chain:   place to store the pointers to partial indirect blocks
740  *      @top:     place to the (detached) top of branch
741  *
742  *      This is a helper function used by ext4_truncate().
743  *
744  *      When we do truncate() we may have to clean the ends of several
745  *      indirect blocks but leave the blocks themselves alive. Block is
746  *      partially truncated if some data below the new i_size is referred
747  *      from it (and it is on the path to the first completely truncated
748  *      data block, indeed).  We have to free the top of that path along
749  *      with everything to the right of the path. Since no allocation
750  *      past the truncation point is possible until ext4_truncate()
751  *      finishes, we may safely do the latter, but top of branch may
752  *      require special attention - pageout below the truncation point
753  *      might try to populate it.
754  *
755  *      We atomically detach the top of branch from the tree, store the
756  *      block number of its root in *@top, pointers to buffer_heads of
757  *      partially truncated blocks - in @chain[].bh and pointers to
758  *      their last elements that should not be removed - in
759  *      @chain[].p. Return value is the pointer to last filled element
760  *      of @chain.
761  *
762  *      The work left to caller to do the actual freeing of subtrees:
763  *              a) free the subtree starting from *@top
764  *              b) free the subtrees whose roots are stored in
765  *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
766  *              c) free the subtrees growing from the inode past the @chain[0].
767  *                      (no partially truncated stuff there).  */
768 
769 static Indirect *ext4_find_shared(struct inode *inode, int depth,
770                                   ext4_lblk_t offsets[4], Indirect chain[4],
771                                   __le32 *top)
772 {
773         Indirect *partial, *p;
774         int k, err;
775 
776         *top = 0;
777         /* Make k index the deepest non-null offset + 1 */
778         for (k = depth; k > 1 && !offsets[k-1]; k--)
779                 ;
780         partial = ext4_get_branch(inode, k, offsets, chain, &err);
781         /* Writer: pointers */
782         if (!partial)
783                 partial = chain + k-1;
784         /*
785          * If the branch acquired continuation since we've looked at it -
786          * fine, it should all survive and (new) top doesn't belong to us.
787          */
788         if (!partial->key && *partial->p)
789                 /* Writer: end */
790                 goto no_top;
791         for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
792                 ;
793         /*
794          * OK, we've found the last block that must survive. The rest of our
795          * branch should be detached before unlocking. However, if that rest
796          * of branch is all ours and does not grow immediately from the inode
797          * it's easier to cheat and just decrement partial->p.
798          */
799         if (p == chain + k - 1 && p > chain) {
800                 p->p--;
801         } else {
802                 *top = *p->p;
803                 /* Nope, don't do this in ext4.  Must leave the tree intact */
804 #if 0
805                 *p->p = 0;
806 #endif
807         }
808         /* Writer: end */
809 
810         while (partial > p) {
811                 brelse(partial->bh);
812                 partial--;
813         }
814 no_top:
815         return partial;
816 }
817 
818 /*
819  * Zero a number of block pointers in either an inode or an indirect block.
820  * If we restart the transaction we must again get write access to the
821  * indirect block for further modification.
822  *
823  * We release `count' blocks on disk, but (last - first) may be greater
824  * than `count' because there can be holes in there.
825  *
826  * Return 0 on success, 1 on invalid block range
827  * and < 0 on fatal error.
828  */
829 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
830                              struct buffer_head *bh,
831                              ext4_fsblk_t block_to_free,
832                              unsigned long count, __le32 *first,
833                              __le32 *last)
834 {
835         __le32 *p;
836         int     flags = EXT4_FREE_BLOCKS_VALIDATED;
837         int     err;
838 
839         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
840             ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
841                 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
842         else if (ext4_should_journal_data(inode))
843                 flags |= EXT4_FREE_BLOCKS_FORGET;
844 
845         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
846                                    count)) {
847                 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
848                                  "blocks %llu len %lu",
849                                  (unsigned long long) block_to_free, count);
850                 return 1;
851         }
852 
853         if (try_to_extend_transaction(handle, inode)) {
854                 if (bh) {
855                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
856                         err = ext4_handle_dirty_metadata(handle, inode, bh);
857                         if (unlikely(err))
858                                 goto out_err;
859                 }
860                 err = ext4_mark_inode_dirty(handle, inode);
861                 if (unlikely(err))
862                         goto out_err;
863                 err = ext4_truncate_restart_trans(handle, inode,
864                                         ext4_blocks_for_truncate(inode));
865                 if (unlikely(err))
866                         goto out_err;
867                 if (bh) {
868                         BUFFER_TRACE(bh, "retaking write access");
869                         err = ext4_journal_get_write_access(handle, bh);
870                         if (unlikely(err))
871                                 goto out_err;
872                 }
873         }
874 
875         for (p = first; p < last; p++)
876                 *p = 0;
877 
878         ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
879         return 0;
880 out_err:
881         ext4_std_error(inode->i_sb, err);
882         return err;
883 }
884 
885 /**
886  * ext4_free_data - free a list of data blocks
887  * @handle:     handle for this transaction
888  * @inode:      inode we are dealing with
889  * @this_bh:    indirect buffer_head which contains *@first and *@last
890  * @first:      array of block numbers
891  * @last:       points immediately past the end of array
892  *
893  * We are freeing all blocks referred from that array (numbers are stored as
894  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
895  *
896  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
897  * blocks are contiguous then releasing them at one time will only affect one
898  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
899  * actually use a lot of journal space.
900  *
901  * @this_bh will be %NULL if @first and @last point into the inode's direct
902  * block pointers.
903  */
904 static void ext4_free_data(handle_t *handle, struct inode *inode,
905                            struct buffer_head *this_bh,
906                            __le32 *first, __le32 *last)
907 {
908         ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
909         unsigned long count = 0;            /* Number of blocks in the run */
910         __le32 *block_to_free_p = NULL;     /* Pointer into inode/ind
911                                                corresponding to
912                                                block_to_free */
913         ext4_fsblk_t nr;                    /* Current block # */
914         __le32 *p;                          /* Pointer into inode/ind
915                                                for current block */
916         int err = 0;
917 
918         if (this_bh) {                          /* For indirect block */
919                 BUFFER_TRACE(this_bh, "get_write_access");
920                 err = ext4_journal_get_write_access(handle, this_bh);
921                 /* Important: if we can't update the indirect pointers
922                  * to the blocks, we can't free them. */
923                 if (err)
924                         return;
925         }
926 
927         for (p = first; p < last; p++) {
928                 nr = le32_to_cpu(*p);
929                 if (nr) {
930                         /* accumulate blocks to free if they're contiguous */
931                         if (count == 0) {
932                                 block_to_free = nr;
933                                 block_to_free_p = p;
934                                 count = 1;
935                         } else if (nr == block_to_free + count) {
936                                 count++;
937                         } else {
938                                 err = ext4_clear_blocks(handle, inode, this_bh,
939                                                         block_to_free, count,
940                                                         block_to_free_p, p);
941                                 if (err)
942                                         break;
943                                 block_to_free = nr;
944                                 block_to_free_p = p;
945                                 count = 1;
946                         }
947                 }
948         }
949 
950         if (!err && count > 0)
951                 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
952                                         count, block_to_free_p, p);
953         if (err < 0)
954                 /* fatal error */
955                 return;
956 
957         if (this_bh) {
958                 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
959 
960                 /*
961                  * The buffer head should have an attached journal head at this
962                  * point. However, if the data is corrupted and an indirect
963                  * block pointed to itself, it would have been detached when
964                  * the block was cleared. Check for this instead of OOPSing.
965                  */
966                 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
967                         ext4_handle_dirty_metadata(handle, inode, this_bh);
968                 else
969                         EXT4_ERROR_INODE(inode,
970                                          "circular indirect block detected at "
971                                          "block %llu",
972                                 (unsigned long long) this_bh->b_blocknr);
973         }
974 }
975 
976 /**
977  *      ext4_free_branches - free an array of branches
978  *      @handle: JBD handle for this transaction
979  *      @inode: inode we are dealing with
980  *      @parent_bh: the buffer_head which contains *@first and *@last
981  *      @first: array of block numbers
982  *      @last:  pointer immediately past the end of array
983  *      @depth: depth of the branches to free
984  *
985  *      We are freeing all blocks referred from these branches (numbers are
986  *      stored as little-endian 32-bit) and updating @inode->i_blocks
987  *      appropriately.
988  */
989 static void ext4_free_branches(handle_t *handle, struct inode *inode,
990                                struct buffer_head *parent_bh,
991                                __le32 *first, __le32 *last, int depth)
992 {
993         ext4_fsblk_t nr;
994         __le32 *p;
995 
996         if (ext4_handle_is_aborted(handle))
997                 return;
998 
999         if (depth--) {
1000                 struct buffer_head *bh;
1001                 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1002                 p = last;
1003                 while (--p >= first) {
1004                         nr = le32_to_cpu(*p);
1005                         if (!nr)
1006                                 continue;               /* A hole */
1007 
1008                         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1009                                                    nr, 1)) {
1010                                 EXT4_ERROR_INODE(inode,
1011                                                  "invalid indirect mapped "
1012                                                  "block %lu (level %d)",
1013                                                  (unsigned long) nr, depth);
1014                                 break;
1015                         }
1016 
1017                         /* Go read the buffer for the next level down */
1018                         bh = sb_bread(inode->i_sb, nr);
1019 
1020                         /*
1021                          * A read failure? Report error and clear slot
1022                          * (should be rare).
1023                          */
1024                         if (!bh) {
1025                                 EXT4_ERROR_INODE_BLOCK(inode, nr,
1026                                                        "Read failure");
1027                                 continue;
1028                         }
1029 
1030                         /* This zaps the entire block.  Bottom up. */
1031                         BUFFER_TRACE(bh, "free child branches");
1032                         ext4_free_branches(handle, inode, bh,
1033                                         (__le32 *) bh->b_data,
1034                                         (__le32 *) bh->b_data + addr_per_block,
1035                                         depth);
1036                         brelse(bh);
1037 
1038                         /*
1039                          * Everything below this this pointer has been
1040                          * released.  Now let this top-of-subtree go.
1041                          *
1042                          * We want the freeing of this indirect block to be
1043                          * atomic in the journal with the updating of the
1044                          * bitmap block which owns it.  So make some room in
1045                          * the journal.
1046                          *
1047                          * We zero the parent pointer *after* freeing its
1048                          * pointee in the bitmaps, so if extend_transaction()
1049                          * for some reason fails to put the bitmap changes and
1050                          * the release into the same transaction, recovery
1051                          * will merely complain about releasing a free block,
1052                          * rather than leaking blocks.
1053                          */
1054                         if (ext4_handle_is_aborted(handle))
1055                                 return;
1056                         if (try_to_extend_transaction(handle, inode)) {
1057                                 ext4_mark_inode_dirty(handle, inode);
1058                                 ext4_truncate_restart_trans(handle, inode,
1059                                             ext4_blocks_for_truncate(inode));
1060                         }
1061 
1062                         /*
1063                          * The forget flag here is critical because if
1064                          * we are journaling (and not doing data
1065                          * journaling), we have to make sure a revoke
1066                          * record is written to prevent the journal
1067                          * replay from overwriting the (former)
1068                          * indirect block if it gets reallocated as a
1069                          * data block.  This must happen in the same
1070                          * transaction where the data blocks are
1071                          * actually freed.
1072                          */
1073                         ext4_free_blocks(handle, inode, NULL, nr, 1,
1074                                          EXT4_FREE_BLOCKS_METADATA|
1075                                          EXT4_FREE_BLOCKS_FORGET);
1076 
1077                         if (parent_bh) {
1078                                 /*
1079                                  * The block which we have just freed is
1080                                  * pointed to by an indirect block: journal it
1081                                  */
1082                                 BUFFER_TRACE(parent_bh, "get_write_access");
1083                                 if (!ext4_journal_get_write_access(handle,
1084                                                                    parent_bh)){
1085                                         *p = 0;
1086                                         BUFFER_TRACE(parent_bh,
1087                                         "call ext4_handle_dirty_metadata");
1088                                         ext4_handle_dirty_metadata(handle,
1089                                                                    inode,
1090                                                                    parent_bh);
1091                                 }
1092                         }
1093                 }
1094         } else {
1095                 /* We have reached the bottom of the tree. */
1096                 BUFFER_TRACE(parent_bh, "free data blocks");
1097                 ext4_free_data(handle, inode, parent_bh, first, last);
1098         }
1099 }
1100 
1101 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1102 {
1103         struct ext4_inode_info *ei = EXT4_I(inode);
1104         __le32 *i_data = ei->i_data;
1105         int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1106         ext4_lblk_t offsets[4];
1107         Indirect chain[4];
1108         Indirect *partial;
1109         __le32 nr = 0;
1110         int n = 0;
1111         ext4_lblk_t last_block, max_block;
1112         unsigned blocksize = inode->i_sb->s_blocksize;
1113 
1114         last_block = (inode->i_size + blocksize-1)
1115                                         >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1116         max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1117                                         >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1118 
1119         if (last_block != max_block) {
1120                 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1121                 if (n == 0)
1122                         return;
1123         }
1124 
1125         ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1126 
1127         /*
1128          * The orphan list entry will now protect us from any crash which
1129          * occurs before the truncate completes, so it is now safe to propagate
1130          * the new, shorter inode size (held for now in i_size) into the
1131          * on-disk inode. We do this via i_disksize, which is the value which
1132          * ext4 *really* writes onto the disk inode.
1133          */
1134         ei->i_disksize = inode->i_size;
1135 
1136         if (last_block == max_block) {
1137                 /*
1138                  * It is unnecessary to free any data blocks if last_block is
1139                  * equal to the indirect block limit.
1140                  */
1141                 return;
1142         } else if (n == 1) {            /* direct blocks */
1143                 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1144                                i_data + EXT4_NDIR_BLOCKS);
1145                 goto do_indirects;
1146         }
1147 
1148         partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1149         /* Kill the top of shared branch (not detached) */
1150         if (nr) {
1151                 if (partial == chain) {
1152                         /* Shared branch grows from the inode */
1153                         ext4_free_branches(handle, inode, NULL,
1154                                            &nr, &nr+1, (chain+n-1) - partial);
1155                         *partial->p = 0;
1156                         /*
1157                          * We mark the inode dirty prior to restart,
1158                          * and prior to stop.  No need for it here.
1159                          */
1160                 } else {
1161                         /* Shared branch grows from an indirect block */
1162                         BUFFER_TRACE(partial->bh, "get_write_access");
1163                         ext4_free_branches(handle, inode, partial->bh,
1164                                         partial->p,
1165                                         partial->p+1, (chain+n-1) - partial);
1166                 }
1167         }
1168         /* Clear the ends of indirect blocks on the shared branch */
1169         while (partial > chain) {
1170                 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1171                                    (__le32*)partial->bh->b_data+addr_per_block,
1172                                    (chain+n-1) - partial);
1173                 BUFFER_TRACE(partial->bh, "call brelse");
1174                 brelse(partial->bh);
1175                 partial--;
1176         }
1177 do_indirects:
1178         /* Kill the remaining (whole) subtrees */
1179         switch (offsets[0]) {
1180         default:
1181                 nr = i_data[EXT4_IND_BLOCK];
1182                 if (nr) {
1183                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1184                         i_data[EXT4_IND_BLOCK] = 0;
1185                 }
1186         case EXT4_IND_BLOCK:
1187                 nr = i_data[EXT4_DIND_BLOCK];
1188                 if (nr) {
1189                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1190                         i_data[EXT4_DIND_BLOCK] = 0;
1191                 }
1192         case EXT4_DIND_BLOCK:
1193                 nr = i_data[EXT4_TIND_BLOCK];
1194                 if (nr) {
1195                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1196                         i_data[EXT4_TIND_BLOCK] = 0;
1197                 }
1198         case EXT4_TIND_BLOCK:
1199                 ;
1200         }
1201 }
1202 
1203 /**
1204  *      ext4_ind_remove_space - remove space from the range
1205  *      @handle: JBD handle for this transaction
1206  *      @inode: inode we are dealing with
1207  *      @start: First block to remove
1208  *      @end:   One block after the last block to remove (exclusive)
1209  *
1210  *      Free the blocks in the defined range (end is exclusive endpoint of
1211  *      range). This is used by ext4_punch_hole().
1212  */
1213 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1214                           ext4_lblk_t start, ext4_lblk_t end)
1215 {
1216         struct ext4_inode_info *ei = EXT4_I(inode);
1217         __le32 *i_data = ei->i_data;
1218         int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1219         ext4_lblk_t offsets[4], offsets2[4];
1220         Indirect chain[4], chain2[4];
1221         Indirect *partial, *partial2;
1222         ext4_lblk_t max_block;
1223         __le32 nr = 0, nr2 = 0;
1224         int n = 0, n2 = 0;
1225         unsigned blocksize = inode->i_sb->s_blocksize;
1226 
1227         max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1228                                         >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1229         if (end >= max_block)
1230                 end = max_block;
1231         if ((start >= end) || (start > max_block))
1232                 return 0;
1233 
1234         n = ext4_block_to_path(inode, start, offsets, NULL);
1235         n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1236 
1237         BUG_ON(n > n2);
1238 
1239         if ((n == 1) && (n == n2)) {
1240                 /* We're punching only within direct block range */
1241                 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1242                                i_data + offsets2[0]);
1243                 return 0;
1244         } else if (n2 > n) {
1245                 /*
1246                  * Start and end are on a different levels so we're going to
1247                  * free partial block at start, and partial block at end of
1248                  * the range. If there are some levels in between then
1249                  * do_indirects label will take care of that.
1250                  */
1251 
1252                 if (n == 1) {
1253                         /*
1254                          * Start is at the direct block level, free
1255                          * everything to the end of the level.
1256                          */
1257                         ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1258                                        i_data + EXT4_NDIR_BLOCKS);
1259                         goto end_range;
1260                 }
1261 
1262 
1263                 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1264                 if (nr) {
1265                         if (partial == chain) {
1266                                 /* Shared branch grows from the inode */
1267                                 ext4_free_branches(handle, inode, NULL,
1268                                            &nr, &nr+1, (chain+n-1) - partial);
1269                                 *partial->p = 0;
1270                         } else {
1271                                 /* Shared branch grows from an indirect block */
1272                                 BUFFER_TRACE(partial->bh, "get_write_access");
1273                                 ext4_free_branches(handle, inode, partial->bh,
1274                                         partial->p,
1275                                         partial->p+1, (chain+n-1) - partial);
1276                         }
1277                 }
1278 
1279                 /*
1280                  * Clear the ends of indirect blocks on the shared branch
1281                  * at the start of the range
1282                  */
1283                 while (partial > chain) {
1284                         ext4_free_branches(handle, inode, partial->bh,
1285                                 partial->p + 1,
1286                                 (__le32 *)partial->bh->b_data+addr_per_block,
1287                                 (chain+n-1) - partial);
1288                         BUFFER_TRACE(partial->bh, "call brelse");
1289                         brelse(partial->bh);
1290                         partial--;
1291                 }
1292 
1293 end_range:
1294                 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1295                 if (nr2) {
1296                         if (partial2 == chain2) {
1297                                 /*
1298                                  * Remember, end is exclusive so here we're at
1299                                  * the start of the next level we're not going
1300                                  * to free. Everything was covered by the start
1301                                  * of the range.
1302                                  */
1303                                 goto do_indirects;
1304                         }
1305                 } else {
1306                         /*
1307                          * ext4_find_shared returns Indirect structure which
1308                          * points to the last element which should not be
1309                          * removed by truncate. But this is end of the range
1310                          * in punch_hole so we need to point to the next element
1311                          */
1312                         partial2->p++;
1313                 }
1314 
1315                 /*
1316                  * Clear the ends of indirect blocks on the shared branch
1317                  * at the end of the range
1318                  */
1319                 while (partial2 > chain2) {
1320                         ext4_free_branches(handle, inode, partial2->bh,
1321                                            (__le32 *)partial2->bh->b_data,
1322                                            partial2->p,
1323                                            (chain2+n2-1) - partial2);
1324                         BUFFER_TRACE(partial2->bh, "call brelse");
1325                         brelse(partial2->bh);
1326                         partial2--;
1327                 }
1328                 goto do_indirects;
1329         }
1330 
1331         /* Punch happened within the same level (n == n2) */
1332         partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1333         partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1334 
1335         /* Free top, but only if partial2 isn't its subtree. */
1336         if (nr) {
1337                 int level = min(partial - chain, partial2 - chain2);
1338                 int i;
1339                 int subtree = 1;
1340 
1341                 for (i = 0; i <= level; i++) {
1342                         if (offsets[i] != offsets2[i]) {
1343                                 subtree = 0;
1344                                 break;
1345                         }
1346                 }
1347 
1348                 if (!subtree) {
1349                         if (partial == chain) {
1350                                 /* Shared branch grows from the inode */
1351                                 ext4_free_branches(handle, inode, NULL,
1352                                                    &nr, &nr+1,
1353                                                    (chain+n-1) - partial);
1354                                 *partial->p = 0;
1355                         } else {
1356                                 /* Shared branch grows from an indirect block */
1357                                 BUFFER_TRACE(partial->bh, "get_write_access");
1358                                 ext4_free_branches(handle, inode, partial->bh,
1359                                                    partial->p,
1360                                                    partial->p+1,
1361                                                    (chain+n-1) - partial);
1362                         }
1363                 }
1364         }
1365 
1366         if (!nr2) {
1367                 /*
1368                  * ext4_find_shared returns Indirect structure which
1369                  * points to the last element which should not be
1370                  * removed by truncate. But this is end of the range
1371                  * in punch_hole so we need to point to the next element
1372                  */
1373                 partial2->p++;
1374         }
1375 
1376         while (partial > chain || partial2 > chain2) {
1377                 int depth = (chain+n-1) - partial;
1378                 int depth2 = (chain2+n2-1) - partial2;
1379 
1380                 if (partial > chain && partial2 > chain2 &&
1381                     partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1382                         /*
1383                          * We've converged on the same block. Clear the range,
1384                          * then we're done.
1385                          */
1386                         ext4_free_branches(handle, inode, partial->bh,
1387                                            partial->p + 1,
1388                                            partial2->p,
1389                                            (chain+n-1) - partial);
1390                         BUFFER_TRACE(partial->bh, "call brelse");
1391                         brelse(partial->bh);
1392                         BUFFER_TRACE(partial2->bh, "call brelse");
1393                         brelse(partial2->bh);
1394                         return 0;
1395                 }
1396 
1397                 /*
1398                  * The start and end partial branches may not be at the same
1399                  * level even though the punch happened within one level. So, we
1400                  * give them a chance to arrive at the same level, then walk
1401                  * them in step with each other until we converge on the same
1402                  * block.
1403                  */
1404                 if (partial > chain && depth <= depth2) {
1405                         ext4_free_branches(handle, inode, partial->bh,
1406                                            partial->p + 1,
1407                                            (__le32 *)partial->bh->b_data+addr_per_block,
1408                                            (chain+n-1) - partial);
1409                         BUFFER_TRACE(partial->bh, "call brelse");
1410                         brelse(partial->bh);
1411                         partial--;
1412                 }
1413                 if (partial2 > chain2 && depth2 <= depth) {
1414                         ext4_free_branches(handle, inode, partial2->bh,
1415                                            (__le32 *)partial2->bh->b_data,
1416                                            partial2->p,
1417                                            (chain2+n2-1) - partial2);
1418                         BUFFER_TRACE(partial2->bh, "call brelse");
1419                         brelse(partial2->bh);
1420                         partial2--;
1421                 }
1422         }
1423         return 0;
1424 
1425 do_indirects:
1426         /* Kill the remaining (whole) subtrees */
1427         switch (offsets[0]) {
1428         default:
1429                 if (++n >= n2)
1430                         return 0;
1431                 nr = i_data[EXT4_IND_BLOCK];
1432                 if (nr) {
1433                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1434                         i_data[EXT4_IND_BLOCK] = 0;
1435                 }
1436         case EXT4_IND_BLOCK:
1437                 if (++n >= n2)
1438                         return 0;
1439                 nr = i_data[EXT4_DIND_BLOCK];
1440                 if (nr) {
1441                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1442                         i_data[EXT4_DIND_BLOCK] = 0;
1443                 }
1444         case EXT4_DIND_BLOCK:
1445                 if (++n >= n2)
1446                         return 0;
1447                 nr = i_data[EXT4_TIND_BLOCK];
1448                 if (nr) {
1449                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1450                         i_data[EXT4_TIND_BLOCK] = 0;
1451                 }
1452         case EXT4_TIND_BLOCK:
1453                 ;
1454         }
1455         return 0;
1456 }
1457 

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

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

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

osdn.jp