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

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

Version: ~ [ linux-5.16-rc1 ] ~ [ linux-5.15.2 ] ~ [ linux-5.14.18 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.79 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.159 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.217 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.255 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.290 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.292 ] ~ [ 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 /*
  2  *  linux/fs/ext4/indirect.c
  3  *
  4  *  from
  5  *
  6  *  linux/fs/ext4/inode.c
  7  *
  8  * Copyright (C) 1992, 1993, 1994, 1995
  9  * Remy Card (card@masi.ibp.fr)
 10  * Laboratoire MASI - Institut Blaise Pascal
 11  * Universite Pierre et Marie Curie (Paris VI)
 12  *
 13  *  from
 14  *
 15  *  linux/fs/minix/inode.c
 16  *
 17  *  Copyright (C) 1991, 1992  Linus Torvalds
 18  *
 19  *  Goal-directed block allocation by Stephen Tweedie
 20  *      (sct@redhat.com), 1993, 1998
 21  */
 22 
 23 #include <linux/aio.h>
 24 #include "ext4_jbd2.h"
 25 #include "truncate.h"
 26 #include "ext4_extents.h"       /* Needed for EXT_MAX_BLOCKS */
 27 
 28 #include <trace/events/ext4.h>
 29 
 30 typedef struct {
 31         __le32  *p;
 32         __le32  key;
 33         struct buffer_head *bh;
 34 } Indirect;
 35 
 36 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
 37 {
 38         p->key = *(p->p = v);
 39         p->bh = bh;
 40 }
 41 
 42 /**
 43  *      ext4_block_to_path - parse the block number into array of offsets
 44  *      @inode: inode in question (we are only interested in its superblock)
 45  *      @i_block: block number to be parsed
 46  *      @offsets: array to store the offsets in
 47  *      @boundary: set this non-zero if the referred-to block is likely to be
 48  *             followed (on disk) by an indirect block.
 49  *
 50  *      To store the locations of file's data ext4 uses a data structure common
 51  *      for UNIX filesystems - tree of pointers anchored in the inode, with
 52  *      data blocks at leaves and indirect blocks in intermediate nodes.
 53  *      This function translates the block number into path in that tree -
 54  *      return value is the path length and @offsets[n] is the offset of
 55  *      pointer to (n+1)th node in the nth one. If @block is out of range
 56  *      (negative or too large) warning is printed and zero returned.
 57  *
 58  *      Note: function doesn't find node addresses, so no IO is needed. All
 59  *      we need to know is the capacity of indirect blocks (taken from the
 60  *      inode->i_sb).
 61  */
 62 
 63 /*
 64  * Portability note: the last comparison (check that we fit into triple
 65  * indirect block) is spelled differently, because otherwise on an
 66  * architecture with 32-bit longs and 8Kb pages we might get into trouble
 67  * if our filesystem had 8Kb blocks. We might use long long, but that would
 68  * kill us on x86. Oh, well, at least the sign propagation does not matter -
 69  * i_block would have to be negative in the very beginning, so we would not
 70  * get there at all.
 71  */
 72 
 73 static int ext4_block_to_path(struct inode *inode,
 74                               ext4_lblk_t i_block,
 75                               ext4_lblk_t offsets[4], int *boundary)
 76 {
 77         int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
 78         int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
 79         const long direct_blocks = EXT4_NDIR_BLOCKS,
 80                 indirect_blocks = ptrs,
 81                 double_blocks = (1 << (ptrs_bits * 2));
 82         int n = 0;
 83         int final = 0;
 84 
 85         if (i_block < direct_blocks) {
 86                 offsets[n++] = i_block;
 87                 final = direct_blocks;
 88         } else if ((i_block -= direct_blocks) < indirect_blocks) {
 89                 offsets[n++] = EXT4_IND_BLOCK;
 90                 offsets[n++] = i_block;
 91                 final = ptrs;
 92         } else if ((i_block -= indirect_blocks) < double_blocks) {
 93                 offsets[n++] = EXT4_DIND_BLOCK;
 94                 offsets[n++] = i_block >> ptrs_bits;
 95                 offsets[n++] = i_block & (ptrs - 1);
 96                 final = ptrs;
 97         } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
 98                 offsets[n++] = EXT4_TIND_BLOCK;
 99                 offsets[n++] = i_block >> (ptrs_bits * 2);
100                 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
101                 offsets[n++] = i_block & (ptrs - 1);
102                 final = ptrs;
103         } else {
104                 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
105                              i_block + direct_blocks +
106                              indirect_blocks + double_blocks, inode->i_ino);
107         }
108         if (boundary)
109                 *boundary = final - 1 - (i_block & (ptrs - 1));
110         return n;
111 }
112 
113 /**
114  *      ext4_get_branch - read the chain of indirect blocks leading to data
115  *      @inode: inode in question
116  *      @depth: depth of the chain (1 - direct pointer, etc.)
117  *      @offsets: offsets of pointers in inode/indirect blocks
118  *      @chain: place to store the result
119  *      @err: here we store the error value
120  *
121  *      Function fills the array of triples <key, p, bh> and returns %NULL
122  *      if everything went OK or the pointer to the last filled triple
123  *      (incomplete one) otherwise. Upon the return chain[i].key contains
124  *      the number of (i+1)-th block in the chain (as it is stored in memory,
125  *      i.e. little-endian 32-bit), chain[i].p contains the address of that
126  *      number (it points into struct inode for i==0 and into the bh->b_data
127  *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
128  *      block for i>0 and NULL for i==0. In other words, it holds the block
129  *      numbers of the chain, addresses they were taken from (and where we can
130  *      verify that chain did not change) and buffer_heads hosting these
131  *      numbers.
132  *
133  *      Function stops when it stumbles upon zero pointer (absent block)
134  *              (pointer to last triple returned, *@err == 0)
135  *      or when it gets an IO error reading an indirect block
136  *              (ditto, *@err == -EIO)
137  *      or when it reads all @depth-1 indirect blocks successfully and finds
138  *      the whole chain, all way to the data (returns %NULL, *err == 0).
139  *
140  *      Need to be called with
141  *      down_read(&EXT4_I(inode)->i_data_sem)
142  */
143 static Indirect *ext4_get_branch(struct inode *inode, int depth,
144                                  ext4_lblk_t  *offsets,
145                                  Indirect chain[4], int *err)
146 {
147         struct super_block *sb = inode->i_sb;
148         Indirect *p = chain;
149         struct buffer_head *bh;
150         int ret = -EIO;
151 
152         *err = 0;
153         /* i_data is not going away, no lock needed */
154         add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
155         if (!p->key)
156                 goto no_block;
157         while (--depth) {
158                 bh = sb_getblk(sb, le32_to_cpu(p->key));
159                 if (unlikely(!bh)) {
160                         ret = -ENOMEM;
161                         goto failure;
162                 }
163 
164                 if (!bh_uptodate_or_lock(bh)) {
165                         if (bh_submit_read(bh) < 0) {
166                                 put_bh(bh);
167                                 goto failure;
168                         }
169                         /* validate block references */
170                         if (ext4_check_indirect_blockref(inode, bh)) {
171                                 put_bh(bh);
172                                 goto failure;
173                         }
174                 }
175 
176                 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
177                 /* Reader: end */
178                 if (!p->key)
179                         goto no_block;
180         }
181         return NULL;
182 
183 failure:
184         *err = ret;
185 no_block:
186         return p;
187 }
188 
189 /**
190  *      ext4_find_near - find a place for allocation with sufficient locality
191  *      @inode: owner
192  *      @ind: descriptor of indirect block.
193  *
194  *      This function returns the preferred place for block allocation.
195  *      It is used when heuristic for sequential allocation fails.
196  *      Rules are:
197  *        + if there is a block to the left of our position - allocate near it.
198  *        + if pointer will live in indirect block - allocate near that block.
199  *        + if pointer will live in inode - allocate in the same
200  *          cylinder group.
201  *
202  * In the latter case we colour the starting block by the callers PID to
203  * prevent it from clashing with concurrent allocations for a different inode
204  * in the same block group.   The PID is used here so that functionally related
205  * files will be close-by on-disk.
206  *
207  *      Caller must make sure that @ind is valid and will stay that way.
208  */
209 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
210 {
211         struct ext4_inode_info *ei = EXT4_I(inode);
212         __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
213         __le32 *p;
214 
215         /* Try to find previous block */
216         for (p = ind->p - 1; p >= start; p--) {
217                 if (*p)
218                         return le32_to_cpu(*p);
219         }
220 
221         /* No such thing, so let's try location of indirect block */
222         if (ind->bh)
223                 return ind->bh->b_blocknr;
224 
225         /*
226          * It is going to be referred to from the inode itself? OK, just put it
227          * into the same cylinder group then.
228          */
229         return ext4_inode_to_goal_block(inode);
230 }
231 
232 /**
233  *      ext4_find_goal - find a preferred place for allocation.
234  *      @inode: owner
235  *      @block:  block we want
236  *      @partial: pointer to the last triple within a chain
237  *
238  *      Normally this function find the preferred place for block allocation,
239  *      returns it.
240  *      Because this is only used for non-extent files, we limit the block nr
241  *      to 32 bits.
242  */
243 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
244                                    Indirect *partial)
245 {
246         ext4_fsblk_t goal;
247 
248         /*
249          * XXX need to get goal block from mballoc's data structures
250          */
251 
252         goal = ext4_find_near(inode, partial);
253         goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
254         return goal;
255 }
256 
257 /**
258  *      ext4_blks_to_allocate - Look up the block map and count the number
259  *      of direct blocks need to be allocated for the given branch.
260  *
261  *      @branch: chain of indirect blocks
262  *      @k: number of blocks need for indirect blocks
263  *      @blks: number of data blocks to be mapped.
264  *      @blocks_to_boundary:  the offset in the indirect block
265  *
266  *      return the total number of blocks to be allocate, including the
267  *      direct and indirect blocks.
268  */
269 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
270                                  int blocks_to_boundary)
271 {
272         unsigned int count = 0;
273 
274         /*
275          * Simple case, [t,d]Indirect block(s) has not allocated yet
276          * then it's clear blocks on that path have not allocated
277          */
278         if (k > 0) {
279                 /* right now we don't handle cross boundary allocation */
280                 if (blks < blocks_to_boundary + 1)
281                         count += blks;
282                 else
283                         count += blocks_to_boundary + 1;
284                 return count;
285         }
286 
287         count++;
288         while (count < blks && count <= blocks_to_boundary &&
289                 le32_to_cpu(*(branch[0].p + count)) == 0) {
290                 count++;
291         }
292         return count;
293 }
294 
295 /**
296  *      ext4_alloc_branch - allocate and set up a chain of blocks.
297  *      @handle: handle for this transaction
298  *      @inode: owner
299  *      @indirect_blks: number of allocated indirect blocks
300  *      @blks: number of allocated direct blocks
301  *      @goal: preferred place for allocation
302  *      @offsets: offsets (in the blocks) to store the pointers to next.
303  *      @branch: place to store the chain in.
304  *
305  *      This function allocates blocks, zeroes out all but the last one,
306  *      links them into chain and (if we are synchronous) writes them to disk.
307  *      In other words, it prepares a branch that can be spliced onto the
308  *      inode. It stores the information about that chain in the branch[], in
309  *      the same format as ext4_get_branch() would do. We are calling it after
310  *      we had read the existing part of chain and partial points to the last
311  *      triple of that (one with zero ->key). Upon the exit we have the same
312  *      picture as after the successful ext4_get_block(), except that in one
313  *      place chain is disconnected - *branch->p is still zero (we did not
314  *      set the last link), but branch->key contains the number that should
315  *      be placed into *branch->p to fill that gap.
316  *
317  *      If allocation fails we free all blocks we've allocated (and forget
318  *      their buffer_heads) and return the error value the from failed
319  *      ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
320  *      as described above and return 0.
321  */
322 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
323                              ext4_lblk_t iblock, int indirect_blks,
324                              int *blks, ext4_fsblk_t goal,
325                              ext4_lblk_t *offsets, Indirect *branch)
326 {
327         struct ext4_allocation_request  ar;
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         /*
334          * Set up for the direct block allocation
335          */
336         memset(&ar, 0, sizeof(ar));
337         ar.inode = inode;
338         ar.len = *blks;
339         ar.logical = iblock;
340         if (S_ISREG(inode->i_mode))
341                 ar.flags = EXT4_MB_HINT_DATA;
342 
343         for (i = 0; i <= indirect_blks; i++) {
344                 if (i == indirect_blks) {
345                         ar.goal = goal;
346                         new_blocks[i] = ext4_mb_new_blocks(handle, &ar, &err);
347                 } else
348                         goal = new_blocks[i] = ext4_new_meta_blocks(handle, inode,
349                                                         goal, 0, NULL, &err);
350                 if (err) {
351                         i--;
352                         goto failed;
353                 }
354                 branch[i].key = cpu_to_le32(new_blocks[i]);
355                 if (i == 0)
356                         continue;
357 
358                 bh = branch[i].bh = sb_getblk(inode->i_sb, new_blocks[i-1]);
359                 if (unlikely(!bh)) {
360                         err = -ENOMEM;
361                         goto failed;
362                 }
363                 lock_buffer(bh);
364                 BUFFER_TRACE(bh, "call get_create_access");
365                 err = ext4_journal_get_create_access(handle, bh);
366                 if (err) {
367                         unlock_buffer(bh);
368                         goto failed;
369                 }
370 
371                 memset(bh->b_data, 0, bh->b_size);
372                 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
373                 b = new_blocks[i];
374 
375                 if (i == indirect_blks)
376                         len = ar.len;
377                 for (j = 0; j < len; j++)
378                         *p++ = cpu_to_le32(b++);
379 
380                 BUFFER_TRACE(bh, "marking uptodate");
381                 set_buffer_uptodate(bh);
382                 unlock_buffer(bh);
383 
384                 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
385                 err = ext4_handle_dirty_metadata(handle, inode, bh);
386                 if (err)
387                         goto failed;
388         }
389         *blks = ar.len;
390         return 0;
391 failed:
392         for (; i >= 0; i--) {
393                 /*
394                  * We want to ext4_forget() only freshly allocated indirect
395                  * blocks.  Buffer for new_blocks[i-1] is at branch[i].bh and
396                  * buffer at branch[0].bh is indirect block / inode already
397                  * existing before ext4_alloc_branch() was called.
398                  */
399                 if (i > 0 && i != indirect_blks && branch[i].bh)
400                         ext4_forget(handle, 1, inode, branch[i].bh,
401                                     branch[i].bh->b_blocknr);
402                 ext4_free_blocks(handle, inode, NULL, new_blocks[i],
403                                  (i == indirect_blks) ? ar.len : 1, 0);
404         }
405         return err;
406 }
407 
408 /**
409  * ext4_splice_branch - splice the allocated branch onto inode.
410  * @handle: handle for this transaction
411  * @inode: owner
412  * @block: (logical) number of block we are adding
413  * @chain: chain of indirect blocks (with a missing link - see
414  *      ext4_alloc_branch)
415  * @where: location of missing link
416  * @num:   number of indirect blocks we are adding
417  * @blks:  number of direct blocks we are adding
418  *
419  * This function fills the missing link and does all housekeeping needed in
420  * inode (->i_blocks, etc.). In case of success we end up with the full
421  * chain to new block and return 0.
422  */
423 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
424                               ext4_lblk_t block, Indirect *where, int num,
425                               int blks)
426 {
427         int i;
428         int err = 0;
429         ext4_fsblk_t current_block;
430 
431         /*
432          * If we're splicing into a [td]indirect block (as opposed to the
433          * inode) then we need to get write access to the [td]indirect block
434          * before the splice.
435          */
436         if (where->bh) {
437                 BUFFER_TRACE(where->bh, "get_write_access");
438                 err = ext4_journal_get_write_access(handle, where->bh);
439                 if (err)
440                         goto err_out;
441         }
442         /* That's it */
443 
444         *where->p = where->key;
445 
446         /*
447          * Update the host buffer_head or inode to point to more just allocated
448          * direct blocks blocks
449          */
450         if (num == 0 && blks > 1) {
451                 current_block = le32_to_cpu(where->key) + 1;
452                 for (i = 1; i < blks; i++)
453                         *(where->p + i) = cpu_to_le32(current_block++);
454         }
455 
456         /* We are done with atomic stuff, now do the rest of housekeeping */
457         /* had we spliced it onto indirect block? */
458         if (where->bh) {
459                 /*
460                  * If we spliced it onto an indirect block, we haven't
461                  * altered the inode.  Note however that if it is being spliced
462                  * onto an indirect block at the very end of the file (the
463                  * file is growing) then we *will* alter the inode to reflect
464                  * the new i_size.  But that is not done here - it is done in
465                  * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
466                  */
467                 jbd_debug(5, "splicing indirect only\n");
468                 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
469                 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
470                 if (err)
471                         goto err_out;
472         } else {
473                 /*
474                  * OK, we spliced it into the inode itself on a direct block.
475                  */
476                 ext4_mark_inode_dirty(handle, inode);
477                 jbd_debug(5, "splicing direct\n");
478         }
479         return err;
480 
481 err_out:
482         for (i = 1; i <= num; i++) {
483                 /*
484                  * branch[i].bh is newly allocated, so there is no
485                  * need to revoke the block, which is why we don't
486                  * need to set EXT4_FREE_BLOCKS_METADATA.
487                  */
488                 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
489                                  EXT4_FREE_BLOCKS_FORGET);
490         }
491         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
492                          blks, 0);
493 
494         return err;
495 }
496 
497 /*
498  * The ext4_ind_map_blocks() function handles non-extents inodes
499  * (i.e., using the traditional indirect/double-indirect i_blocks
500  * scheme) for ext4_map_blocks().
501  *
502  * Allocation strategy is simple: if we have to allocate something, we will
503  * have to go the whole way to leaf. So let's do it before attaching anything
504  * to tree, set linkage between the newborn blocks, write them if sync is
505  * required, recheck the path, free and repeat if check fails, otherwise
506  * set the last missing link (that will protect us from any truncate-generated
507  * removals - all blocks on the path are immune now) and possibly force the
508  * write on the parent block.
509  * That has a nice additional property: no special recovery from the failed
510  * allocations is needed - we simply release blocks and do not touch anything
511  * reachable from inode.
512  *
513  * `handle' can be NULL if create == 0.
514  *
515  * return > 0, # of blocks mapped or allocated.
516  * return = 0, if plain lookup failed.
517  * return < 0, error case.
518  *
519  * The ext4_ind_get_blocks() function should be called with
520  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
521  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
522  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
523  * blocks.
524  */
525 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
526                         struct ext4_map_blocks *map,
527                         int flags)
528 {
529         int err = -EIO;
530         ext4_lblk_t offsets[4];
531         Indirect chain[4];
532         Indirect *partial;
533         ext4_fsblk_t goal;
534         int indirect_blks;
535         int blocks_to_boundary = 0;
536         int depth;
537         int count = 0;
538         ext4_fsblk_t first_block = 0;
539 
540         trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
541         J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
542         J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
543         depth = ext4_block_to_path(inode, map->m_lblk, offsets,
544                                    &blocks_to_boundary);
545 
546         if (depth == 0)
547                 goto out;
548 
549         partial = ext4_get_branch(inode, depth, offsets, chain, &err);
550 
551         /* Simplest case - block found, no allocation needed */
552         if (!partial) {
553                 first_block = le32_to_cpu(chain[depth - 1].key);
554                 count++;
555                 /*map more blocks*/
556                 while (count < map->m_len && count <= blocks_to_boundary) {
557                         ext4_fsblk_t blk;
558 
559                         blk = le32_to_cpu(*(chain[depth-1].p + count));
560 
561                         if (blk == first_block + count)
562                                 count++;
563                         else
564                                 break;
565                 }
566                 goto got_it;
567         }
568 
569         /* Next simple case - plain lookup or failed read of indirect block */
570         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
571                 goto cleanup;
572 
573         /*
574          * Okay, we need to do block allocation.
575         */
576         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
577                                        EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
578                 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
579                                  "non-extent mapped inodes with bigalloc");
580                 return -EUCLEAN;
581         }
582 
583         goal = ext4_find_goal(inode, map->m_lblk, partial);
584 
585         /* the number of blocks need to allocate for [d,t]indirect blocks */
586         indirect_blks = (chain + depth) - partial - 1;
587 
588         /*
589          * Next look up the indirect map to count the totoal number of
590          * direct blocks to allocate for this branch.
591          */
592         count = ext4_blks_to_allocate(partial, indirect_blks,
593                                       map->m_len, blocks_to_boundary);
594         /*
595          * Block out ext4_truncate while we alter the tree
596          */
597         err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
598                                 &count, goal,
599                                 offsets + (partial - chain), partial);
600 
601         /*
602          * The ext4_splice_branch call will free and forget any buffers
603          * on the new chain if there is a failure, but that risks using
604          * up transaction credits, especially for bitmaps where the
605          * credits cannot be returned.  Can we handle this somehow?  We
606          * may need to return -EAGAIN upwards in the worst case.  --sct
607          */
608         if (!err)
609                 err = ext4_splice_branch(handle, inode, map->m_lblk,
610                                          partial, indirect_blks, count);
611         if (err)
612                 goto cleanup;
613 
614         map->m_flags |= EXT4_MAP_NEW;
615 
616         ext4_update_inode_fsync_trans(handle, inode, 1);
617 got_it:
618         map->m_flags |= EXT4_MAP_MAPPED;
619         map->m_pblk = le32_to_cpu(chain[depth-1].key);
620         map->m_len = count;
621         if (count > blocks_to_boundary)
622                 map->m_flags |= EXT4_MAP_BOUNDARY;
623         err = count;
624         /* Clean up and exit */
625         partial = chain + depth - 1;    /* the whole chain */
626 cleanup:
627         while (partial > chain) {
628                 BUFFER_TRACE(partial->bh, "call brelse");
629                 brelse(partial->bh);
630                 partial--;
631         }
632 out:
633         trace_ext4_ind_map_blocks_exit(inode, map, err);
634         return err;
635 }
636 
637 /*
638  * O_DIRECT for ext3 (or indirect map) based files
639  *
640  * If the O_DIRECT write will extend the file then add this inode to the
641  * orphan list.  So recovery will truncate it back to the original size
642  * if the machine crashes during the write.
643  *
644  * If the O_DIRECT write is intantiating holes inside i_size and the machine
645  * crashes then stale disk data _may_ be exposed inside the file. But current
646  * VFS code falls back into buffered path in that case so we are safe.
647  */
648 ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
649                            const struct iovec *iov, loff_t offset,
650                            unsigned long nr_segs)
651 {
652         struct file *file = iocb->ki_filp;
653         struct inode *inode = file->f_mapping->host;
654         struct ext4_inode_info *ei = EXT4_I(inode);
655         handle_t *handle;
656         ssize_t ret;
657         int orphan = 0;
658         size_t count = iov_length(iov, nr_segs);
659         int retries = 0;
660 
661         if (rw == WRITE) {
662                 loff_t final_size = offset + count;
663 
664                 if (final_size > inode->i_size) {
665                         /* Credits for sb + inode write */
666                         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
667                         if (IS_ERR(handle)) {
668                                 ret = PTR_ERR(handle);
669                                 goto out;
670                         }
671                         ret = ext4_orphan_add(handle, inode);
672                         if (ret) {
673                                 ext4_journal_stop(handle);
674                                 goto out;
675                         }
676                         orphan = 1;
677                         ei->i_disksize = inode->i_size;
678                         ext4_journal_stop(handle);
679                 }
680         }
681 
682 retry:
683         if (rw == READ && ext4_should_dioread_nolock(inode)) {
684                 if (unlikely(atomic_read(&EXT4_I(inode)->i_unwritten))) {
685                         mutex_lock(&inode->i_mutex);
686                         ext4_flush_unwritten_io(inode);
687                         mutex_unlock(&inode->i_mutex);
688                 }
689                 /*
690                  * Nolock dioread optimization may be dynamically disabled
691                  * via ext4_inode_block_unlocked_dio(). Check inode's state
692                  * while holding extra i_dio_count ref.
693                  */
694                 atomic_inc(&inode->i_dio_count);
695                 smp_mb();
696                 if (unlikely(ext4_test_inode_state(inode,
697                                                     EXT4_STATE_DIOREAD_LOCK))) {
698                         inode_dio_done(inode);
699                         goto locked;
700                 }
701                 ret = __blockdev_direct_IO(rw, iocb, inode,
702                                  inode->i_sb->s_bdev, iov,
703                                  offset, nr_segs,
704                                  ext4_get_block, NULL, NULL, 0);
705                 inode_dio_done(inode);
706         } else {
707 locked:
708                 ret = blockdev_direct_IO(rw, iocb, inode, iov,
709                                  offset, nr_segs, ext4_get_block);
710 
711                 if (unlikely((rw & WRITE) && ret < 0)) {
712                         loff_t isize = i_size_read(inode);
713                         loff_t end = offset + iov_length(iov, nr_segs);
714 
715                         if (end > isize)
716                                 ext4_truncate_failed_write(inode);
717                 }
718         }
719         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
720                 goto retry;
721 
722         if (orphan) {
723                 int err;
724 
725                 /* Credits for sb + inode write */
726                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
727                 if (IS_ERR(handle)) {
728                         /* This is really bad luck. We've written the data
729                          * but cannot extend i_size. Bail out and pretend
730                          * the write failed... */
731                         ret = PTR_ERR(handle);
732                         if (inode->i_nlink)
733                                 ext4_orphan_del(NULL, inode);
734 
735                         goto out;
736                 }
737                 if (inode->i_nlink)
738                         ext4_orphan_del(handle, inode);
739                 if (ret > 0) {
740                         loff_t end = offset + ret;
741                         if (end > inode->i_size) {
742                                 ei->i_disksize = end;
743                                 i_size_write(inode, end);
744                                 /*
745                                  * We're going to return a positive `ret'
746                                  * here due to non-zero-length I/O, so there's
747                                  * no way of reporting error returns from
748                                  * ext4_mark_inode_dirty() to userspace.  So
749                                  * ignore it.
750                                  */
751                                 ext4_mark_inode_dirty(handle, inode);
752                         }
753                 }
754                 err = ext4_journal_stop(handle);
755                 if (ret == 0)
756                         ret = err;
757         }
758 out:
759         return ret;
760 }
761 
762 /*
763  * Calculate the number of metadata blocks need to reserve
764  * to allocate a new block at @lblocks for non extent file based file
765  */
766 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
767 {
768         struct ext4_inode_info *ei = EXT4_I(inode);
769         sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
770         int blk_bits;
771 
772         if (lblock < EXT4_NDIR_BLOCKS)
773                 return 0;
774 
775         lblock -= EXT4_NDIR_BLOCKS;
776 
777         if (ei->i_da_metadata_calc_len &&
778             (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
779                 ei->i_da_metadata_calc_len++;
780                 return 0;
781         }
782         ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
783         ei->i_da_metadata_calc_len = 1;
784         blk_bits = order_base_2(lblock);
785         return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
786 }
787 
788 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
789 {
790         int indirects;
791 
792         /* if nrblocks are contiguous */
793         if (chunk) {
794                 /*
795                  * With N contiguous data blocks, we need at most
796                  * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
797                  * 2 dindirect blocks, and 1 tindirect block
798                  */
799                 return DIV_ROUND_UP(nrblocks,
800                                     EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
801         }
802         /*
803          * if nrblocks are not contiguous, worse case, each block touch
804          * a indirect block, and each indirect block touch a double indirect
805          * block, plus a triple indirect block
806          */
807         indirects = nrblocks * 2 + 1;
808         return indirects;
809 }
810 
811 /*
812  * Truncate transactions can be complex and absolutely huge.  So we need to
813  * be able to restart the transaction at a conventient checkpoint to make
814  * sure we don't overflow the journal.
815  *
816  * Try to extend this transaction for the purposes of truncation.  If
817  * extend fails, we need to propagate the failure up and restart the
818  * transaction in the top-level truncate loop. --sct
819  *
820  * Returns 0 if we managed to create more room.  If we can't create more
821  * room, and the transaction must be restarted we return 1.
822  */
823 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
824 {
825         if (!ext4_handle_valid(handle))
826                 return 0;
827         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
828                 return 0;
829         if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
830                 return 0;
831         return 1;
832 }
833 
834 /*
835  * Probably it should be a library function... search for first non-zero word
836  * or memcmp with zero_page, whatever is better for particular architecture.
837  * Linus?
838  */
839 static inline int all_zeroes(__le32 *p, __le32 *q)
840 {
841         while (p < q)
842                 if (*p++)
843                         return 0;
844         return 1;
845 }
846 
847 /**
848  *      ext4_find_shared - find the indirect blocks for partial truncation.
849  *      @inode:   inode in question
850  *      @depth:   depth of the affected branch
851  *      @offsets: offsets of pointers in that branch (see ext4_block_to_path)
852  *      @chain:   place to store the pointers to partial indirect blocks
853  *      @top:     place to the (detached) top of branch
854  *
855  *      This is a helper function used by ext4_truncate().
856  *
857  *      When we do truncate() we may have to clean the ends of several
858  *      indirect blocks but leave the blocks themselves alive. Block is
859  *      partially truncated if some data below the new i_size is referred
860  *      from it (and it is on the path to the first completely truncated
861  *      data block, indeed).  We have to free the top of that path along
862  *      with everything to the right of the path. Since no allocation
863  *      past the truncation point is possible until ext4_truncate()
864  *      finishes, we may safely do the latter, but top of branch may
865  *      require special attention - pageout below the truncation point
866  *      might try to populate it.
867  *
868  *      We atomically detach the top of branch from the tree, store the
869  *      block number of its root in *@top, pointers to buffer_heads of
870  *      partially truncated blocks - in @chain[].bh and pointers to
871  *      their last elements that should not be removed - in
872  *      @chain[].p. Return value is the pointer to last filled element
873  *      of @chain.
874  *
875  *      The work left to caller to do the actual freeing of subtrees:
876  *              a) free the subtree starting from *@top
877  *              b) free the subtrees whose roots are stored in
878  *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
879  *              c) free the subtrees growing from the inode past the @chain[0].
880  *                      (no partially truncated stuff there).  */
881 
882 static Indirect *ext4_find_shared(struct inode *inode, int depth,
883                                   ext4_lblk_t offsets[4], Indirect chain[4],
884                                   __le32 *top)
885 {
886         Indirect *partial, *p;
887         int k, err;
888 
889         *top = 0;
890         /* Make k index the deepest non-null offset + 1 */
891         for (k = depth; k > 1 && !offsets[k-1]; k--)
892                 ;
893         partial = ext4_get_branch(inode, k, offsets, chain, &err);
894         /* Writer: pointers */
895         if (!partial)
896                 partial = chain + k-1;
897         /*
898          * If the branch acquired continuation since we've looked at it -
899          * fine, it should all survive and (new) top doesn't belong to us.
900          */
901         if (!partial->key && *partial->p)
902                 /* Writer: end */
903                 goto no_top;
904         for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
905                 ;
906         /*
907          * OK, we've found the last block that must survive. The rest of our
908          * branch should be detached before unlocking. However, if that rest
909          * of branch is all ours and does not grow immediately from the inode
910          * it's easier to cheat and just decrement partial->p.
911          */
912         if (p == chain + k - 1 && p > chain) {
913                 p->p--;
914         } else {
915                 *top = *p->p;
916                 /* Nope, don't do this in ext4.  Must leave the tree intact */
917 #if 0
918                 *p->p = 0;
919 #endif
920         }
921         /* Writer: end */
922 
923         while (partial > p) {
924                 brelse(partial->bh);
925                 partial--;
926         }
927 no_top:
928         return partial;
929 }
930 
931 /*
932  * Zero a number of block pointers in either an inode or an indirect block.
933  * If we restart the transaction we must again get write access to the
934  * indirect block for further modification.
935  *
936  * We release `count' blocks on disk, but (last - first) may be greater
937  * than `count' because there can be holes in there.
938  *
939  * Return 0 on success, 1 on invalid block range
940  * and < 0 on fatal error.
941  */
942 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
943                              struct buffer_head *bh,
944                              ext4_fsblk_t block_to_free,
945                              unsigned long count, __le32 *first,
946                              __le32 *last)
947 {
948         __le32 *p;
949         int     flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
950         int     err;
951 
952         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
953                 flags |= EXT4_FREE_BLOCKS_METADATA;
954 
955         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
956                                    count)) {
957                 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
958                                  "blocks %llu len %lu",
959                                  (unsigned long long) block_to_free, count);
960                 return 1;
961         }
962 
963         if (try_to_extend_transaction(handle, inode)) {
964                 if (bh) {
965                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
966                         err = ext4_handle_dirty_metadata(handle, inode, bh);
967                         if (unlikely(err))
968                                 goto out_err;
969                 }
970                 err = ext4_mark_inode_dirty(handle, inode);
971                 if (unlikely(err))
972                         goto out_err;
973                 err = ext4_truncate_restart_trans(handle, inode,
974                                         ext4_blocks_for_truncate(inode));
975                 if (unlikely(err))
976                         goto out_err;
977                 if (bh) {
978                         BUFFER_TRACE(bh, "retaking write access");
979                         err = ext4_journal_get_write_access(handle, bh);
980                         if (unlikely(err))
981                                 goto out_err;
982                 }
983         }
984 
985         for (p = first; p < last; p++)
986                 *p = 0;
987 
988         ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
989         return 0;
990 out_err:
991         ext4_std_error(inode->i_sb, err);
992         return err;
993 }
994 
995 /**
996  * ext4_free_data - free a list of data blocks
997  * @handle:     handle for this transaction
998  * @inode:      inode we are dealing with
999  * @this_bh:    indirect buffer_head which contains *@first and *@last
1000  * @first:      array of block numbers
1001  * @last:       points immediately past the end of array
1002  *
1003  * We are freeing all blocks referred from that array (numbers are stored as
1004  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1005  *
1006  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
1007  * blocks are contiguous then releasing them at one time will only affect one
1008  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1009  * actually use a lot of journal space.
1010  *
1011  * @this_bh will be %NULL if @first and @last point into the inode's direct
1012  * block pointers.
1013  */
1014 static void ext4_free_data(handle_t *handle, struct inode *inode,
1015                            struct buffer_head *this_bh,
1016                            __le32 *first, __le32 *last)
1017 {
1018         ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
1019         unsigned long count = 0;            /* Number of blocks in the run */
1020         __le32 *block_to_free_p = NULL;     /* Pointer into inode/ind
1021                                                corresponding to
1022                                                block_to_free */
1023         ext4_fsblk_t nr;                    /* Current block # */
1024         __le32 *p;                          /* Pointer into inode/ind
1025                                                for current block */
1026         int err = 0;
1027 
1028         if (this_bh) {                          /* For indirect block */
1029                 BUFFER_TRACE(this_bh, "get_write_access");
1030                 err = ext4_journal_get_write_access(handle, this_bh);
1031                 /* Important: if we can't update the indirect pointers
1032                  * to the blocks, we can't free them. */
1033                 if (err)
1034                         return;
1035         }
1036 
1037         for (p = first; p < last; p++) {
1038                 nr = le32_to_cpu(*p);
1039                 if (nr) {
1040                         /* accumulate blocks to free if they're contiguous */
1041                         if (count == 0) {
1042                                 block_to_free = nr;
1043                                 block_to_free_p = p;
1044                                 count = 1;
1045                         } else if (nr == block_to_free + count) {
1046                                 count++;
1047                         } else {
1048                                 err = ext4_clear_blocks(handle, inode, this_bh,
1049                                                         block_to_free, count,
1050                                                         block_to_free_p, p);
1051                                 if (err)
1052                                         break;
1053                                 block_to_free = nr;
1054                                 block_to_free_p = p;
1055                                 count = 1;
1056                         }
1057                 }
1058         }
1059 
1060         if (!err && count > 0)
1061                 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1062                                         count, block_to_free_p, p);
1063         if (err < 0)
1064                 /* fatal error */
1065                 return;
1066 
1067         if (this_bh) {
1068                 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1069 
1070                 /*
1071                  * The buffer head should have an attached journal head at this
1072                  * point. However, if the data is corrupted and an indirect
1073                  * block pointed to itself, it would have been detached when
1074                  * the block was cleared. Check for this instead of OOPSing.
1075                  */
1076                 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1077                         ext4_handle_dirty_metadata(handle, inode, this_bh);
1078                 else
1079                         EXT4_ERROR_INODE(inode,
1080                                          "circular indirect block detected at "
1081                                          "block %llu",
1082                                 (unsigned long long) this_bh->b_blocknr);
1083         }
1084 }
1085 
1086 /**
1087  *      ext4_free_branches - free an array of branches
1088  *      @handle: JBD handle for this transaction
1089  *      @inode: inode we are dealing with
1090  *      @parent_bh: the buffer_head which contains *@first and *@last
1091  *      @first: array of block numbers
1092  *      @last:  pointer immediately past the end of array
1093  *      @depth: depth of the branches to free
1094  *
1095  *      We are freeing all blocks referred from these branches (numbers are
1096  *      stored as little-endian 32-bit) and updating @inode->i_blocks
1097  *      appropriately.
1098  */
1099 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1100                                struct buffer_head *parent_bh,
1101                                __le32 *first, __le32 *last, int depth)
1102 {
1103         ext4_fsblk_t nr;
1104         __le32 *p;
1105 
1106         if (ext4_handle_is_aborted(handle))
1107                 return;
1108 
1109         if (depth--) {
1110                 struct buffer_head *bh;
1111                 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1112                 p = last;
1113                 while (--p >= first) {
1114                         nr = le32_to_cpu(*p);
1115                         if (!nr)
1116                                 continue;               /* A hole */
1117 
1118                         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1119                                                    nr, 1)) {
1120                                 EXT4_ERROR_INODE(inode,
1121                                                  "invalid indirect mapped "
1122                                                  "block %lu (level %d)",
1123                                                  (unsigned long) nr, depth);
1124                                 break;
1125                         }
1126 
1127                         /* Go read the buffer for the next level down */
1128                         bh = sb_bread(inode->i_sb, nr);
1129 
1130                         /*
1131                          * A read failure? Report error and clear slot
1132                          * (should be rare).
1133                          */
1134                         if (!bh) {
1135                                 EXT4_ERROR_INODE_BLOCK(inode, nr,
1136                                                        "Read failure");
1137                                 continue;
1138                         }
1139 
1140                         /* This zaps the entire block.  Bottom up. */
1141                         BUFFER_TRACE(bh, "free child branches");
1142                         ext4_free_branches(handle, inode, bh,
1143                                         (__le32 *) bh->b_data,
1144                                         (__le32 *) bh->b_data + addr_per_block,
1145                                         depth);
1146                         brelse(bh);
1147 
1148                         /*
1149                          * Everything below this this pointer has been
1150                          * released.  Now let this top-of-subtree go.
1151                          *
1152                          * We want the freeing of this indirect block to be
1153                          * atomic in the journal with the updating of the
1154                          * bitmap block which owns it.  So make some room in
1155                          * the journal.
1156                          *
1157                          * We zero the parent pointer *after* freeing its
1158                          * pointee in the bitmaps, so if extend_transaction()
1159                          * for some reason fails to put the bitmap changes and
1160                          * the release into the same transaction, recovery
1161                          * will merely complain about releasing a free block,
1162                          * rather than leaking blocks.
1163                          */
1164                         if (ext4_handle_is_aborted(handle))
1165                                 return;
1166                         if (try_to_extend_transaction(handle, inode)) {
1167                                 ext4_mark_inode_dirty(handle, inode);
1168                                 ext4_truncate_restart_trans(handle, inode,
1169                                             ext4_blocks_for_truncate(inode));
1170                         }
1171 
1172                         /*
1173                          * The forget flag here is critical because if
1174                          * we are journaling (and not doing data
1175                          * journaling), we have to make sure a revoke
1176                          * record is written to prevent the journal
1177                          * replay from overwriting the (former)
1178                          * indirect block if it gets reallocated as a
1179                          * data block.  This must happen in the same
1180                          * transaction where the data blocks are
1181                          * actually freed.
1182                          */
1183                         ext4_free_blocks(handle, inode, NULL, nr, 1,
1184                                          EXT4_FREE_BLOCKS_METADATA|
1185                                          EXT4_FREE_BLOCKS_FORGET);
1186 
1187                         if (parent_bh) {
1188                                 /*
1189                                  * The block which we have just freed is
1190                                  * pointed to by an indirect block: journal it
1191                                  */
1192                                 BUFFER_TRACE(parent_bh, "get_write_access");
1193                                 if (!ext4_journal_get_write_access(handle,
1194                                                                    parent_bh)){
1195                                         *p = 0;
1196                                         BUFFER_TRACE(parent_bh,
1197                                         "call ext4_handle_dirty_metadata");
1198                                         ext4_handle_dirty_metadata(handle,
1199                                                                    inode,
1200                                                                    parent_bh);
1201                                 }
1202                         }
1203                 }
1204         } else {
1205                 /* We have reached the bottom of the tree. */
1206                 BUFFER_TRACE(parent_bh, "free data blocks");
1207                 ext4_free_data(handle, inode, parent_bh, first, last);
1208         }
1209 }
1210 
1211 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1212 {
1213         struct ext4_inode_info *ei = EXT4_I(inode);
1214         __le32 *i_data = ei->i_data;
1215         int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1216         ext4_lblk_t offsets[4];
1217         Indirect chain[4];
1218         Indirect *partial;
1219         __le32 nr = 0;
1220         int n = 0;
1221         ext4_lblk_t last_block, max_block;
1222         unsigned blocksize = inode->i_sb->s_blocksize;
1223 
1224         last_block = (inode->i_size + blocksize-1)
1225                                         >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1226         max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1227                                         >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1228 
1229         if (last_block != max_block) {
1230                 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1231                 if (n == 0)
1232                         return;
1233         }
1234 
1235         ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1236 
1237         /*
1238          * The orphan list entry will now protect us from any crash which
1239          * occurs before the truncate completes, so it is now safe to propagate
1240          * the new, shorter inode size (held for now in i_size) into the
1241          * on-disk inode. We do this via i_disksize, which is the value which
1242          * ext4 *really* writes onto the disk inode.
1243          */
1244         ei->i_disksize = inode->i_size;
1245 
1246         if (last_block == max_block) {
1247                 /*
1248                  * It is unnecessary to free any data blocks if last_block is
1249                  * equal to the indirect block limit.
1250                  */
1251                 return;
1252         } else if (n == 1) {            /* direct blocks */
1253                 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1254                                i_data + EXT4_NDIR_BLOCKS);
1255                 goto do_indirects;
1256         }
1257 
1258         partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1259         /* Kill the top of shared branch (not detached) */
1260         if (nr) {
1261                 if (partial == chain) {
1262                         /* Shared branch grows from the inode */
1263                         ext4_free_branches(handle, inode, NULL,
1264                                            &nr, &nr+1, (chain+n-1) - partial);
1265                         *partial->p = 0;
1266                         /*
1267                          * We mark the inode dirty prior to restart,
1268                          * and prior to stop.  No need for it here.
1269                          */
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         /* Clear the ends of indirect blocks on the shared branch */
1279         while (partial > chain) {
1280                 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1281                                    (__le32*)partial->bh->b_data+addr_per_block,
1282                                    (chain+n-1) - partial);
1283                 BUFFER_TRACE(partial->bh, "call brelse");
1284                 brelse(partial->bh);
1285                 partial--;
1286         }
1287 do_indirects:
1288         /* Kill the remaining (whole) subtrees */
1289         switch (offsets[0]) {
1290         default:
1291                 nr = i_data[EXT4_IND_BLOCK];
1292                 if (nr) {
1293                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1294                         i_data[EXT4_IND_BLOCK] = 0;
1295                 }
1296         case EXT4_IND_BLOCK:
1297                 nr = i_data[EXT4_DIND_BLOCK];
1298                 if (nr) {
1299                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1300                         i_data[EXT4_DIND_BLOCK] = 0;
1301                 }
1302         case EXT4_DIND_BLOCK:
1303                 nr = i_data[EXT4_TIND_BLOCK];
1304                 if (nr) {
1305                         ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1306                         i_data[EXT4_TIND_BLOCK] = 0;
1307                 }
1308         case EXT4_TIND_BLOCK:
1309                 ;
1310         }
1311 }
1312 
1313 static int free_hole_blocks(handle_t *handle, struct inode *inode,
1314                             struct buffer_head *parent_bh, __le32 *i_data,
1315                             int level, ext4_lblk_t first,
1316                             ext4_lblk_t count, int max)
1317 {
1318         struct buffer_head *bh = NULL;
1319         int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1320         int ret = 0;
1321         int i, inc;
1322         ext4_lblk_t offset;
1323         __le32 blk;
1324 
1325         inc = 1 << ((EXT4_BLOCK_SIZE_BITS(inode->i_sb) - 2) * level);
1326         for (i = 0, offset = 0; i < max; i++, i_data++, offset += inc) {
1327                 if (offset >= count + first)
1328                         break;
1329                 if (*i_data == 0 || (offset + inc) <= first)
1330                         continue;
1331                 blk = *i_data;
1332                 if (level > 0) {
1333                         ext4_lblk_t first2;
1334                         ext4_lblk_t count2;
1335 
1336                         bh = sb_bread(inode->i_sb, le32_to_cpu(blk));
1337                         if (!bh) {
1338                                 EXT4_ERROR_INODE_BLOCK(inode, le32_to_cpu(blk),
1339                                                        "Read failure");
1340                                 return -EIO;
1341                         }
1342                         if (first > offset) {
1343                                 first2 = first - offset;
1344                                 count2 = count;
1345                         } else {
1346                                 first2 = 0;
1347                                 count2 = count - (offset - first);
1348                         }
1349                         ret = free_hole_blocks(handle, inode, bh,
1350                                                (__le32 *)bh->b_data, level - 1,
1351                                                first2, count2,
1352                                                inode->i_sb->s_blocksize >> 2);
1353                         if (ret) {
1354                                 brelse(bh);
1355                                 goto err;
1356                         }
1357                 }
1358                 if (level == 0 ||
1359                     (bh && all_zeroes((__le32 *)bh->b_data,
1360                                       (__le32 *)bh->b_data + addr_per_block))) {
1361                         ext4_free_data(handle, inode, parent_bh, &blk, &blk+1);
1362                         *i_data = 0;
1363                 }
1364                 brelse(bh);
1365                 bh = NULL;
1366         }
1367 
1368 err:
1369         return ret;
1370 }
1371 
1372 int ext4_free_hole_blocks(handle_t *handle, struct inode *inode,
1373                           ext4_lblk_t first, ext4_lblk_t stop)
1374 {
1375         int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1376         int level, ret = 0;
1377         int num = EXT4_NDIR_BLOCKS;
1378         ext4_lblk_t count, max = EXT4_NDIR_BLOCKS;
1379         __le32 *i_data = EXT4_I(inode)->i_data;
1380 
1381         count = stop - first;
1382         for (level = 0; level < 4; level++, max *= addr_per_block) {
1383                 if (first < max) {
1384                         ret = free_hole_blocks(handle, inode, NULL, i_data,
1385                                                level, first, count, num);
1386                         if (ret)
1387                                 goto err;
1388                         if (count > max - first)
1389                                 count -= max - first;
1390                         else
1391                                 break;
1392                         first = 0;
1393                 } else {
1394                         first -= max;
1395                 }
1396                 i_data += num;
1397                 if (level == 0) {
1398                         num = 1;
1399                         max = 1;
1400                 }
1401         }
1402 
1403 err:
1404         return ret;
1405 }
1406 
1407 

~ [ 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