Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
xfs_dir2_block.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_log.h"
22 #include "xfs_trans.h"
23 #include "xfs_sb.h"
24 #include "xfs_ag.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_dinode.h"
29 #include "xfs_inode.h"
30 #include "xfs_inode_item.h"
31 #include "xfs_dir2.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
36 
37 /*
38  * Local function prototypes.
39  */
40 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
41  int first, int last);
42 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
43 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
44  int *entno);
45 static int xfs_dir2_block_sort(const void *a, const void *b);
46 
47 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
48 
49 /*
50  * One-time startup routine called from xfs_init().
51  */
52 void
54 {
55  xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
56  xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
57 }
58 
59 /*
60  * Add an entry to a block directory.
61  */
62 int /* error */
64  xfs_da_args_t *args) /* directory op arguments */
65 {
66  xfs_dir2_data_free_t *bf; /* bestfree table in block */
67  xfs_dir2_data_hdr_t *hdr; /* block header */
68  xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
69  struct xfs_buf *bp; /* buffer for block */
70  xfs_dir2_block_tail_t *btp; /* block tail */
71  int compact; /* need to compact leaf ents */
72  xfs_dir2_data_entry_t *dep; /* block data entry */
73  xfs_inode_t *dp; /* directory inode */
74  xfs_dir2_data_unused_t *dup; /* block unused entry */
75  int error; /* error return value */
76  xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
77  xfs_dahash_t hash; /* hash value of found entry */
78  int high; /* high index for binary srch */
79  int highstale; /* high stale index */
80  int lfloghigh=0; /* last final leaf to log */
81  int lfloglow=0; /* first final leaf to log */
82  int len; /* length of the new entry */
83  int low; /* low index for binary srch */
84  int lowstale; /* low stale index */
85  int mid=0; /* midpoint for binary srch */
86  xfs_mount_t *mp; /* filesystem mount point */
87  int needlog; /* need to log header */
88  int needscan; /* need to rescan freespace */
89  __be16 *tagp; /* pointer to tag value */
90  xfs_trans_t *tp; /* transaction structure */
91 
92  trace_xfs_dir2_block_addname(args);
93 
94  dp = args->dp;
95  tp = args->trans;
96  mp = dp->i_mount;
97  /*
98  * Read the (one and only) directory block into dabuf bp.
99  */
100  if ((error =
101  xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
102  return error;
103  }
104  ASSERT(bp != NULL);
105  hdr = bp->b_addr;
106  /*
107  * Check the magic number, corrupted if wrong.
108  */
110  XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
111  XFS_ERRLEVEL_LOW, mp, hdr);
112  xfs_trans_brelse(tp, bp);
113  return XFS_ERROR(EFSCORRUPTED);
114  }
115  len = xfs_dir2_data_entsize(args->namelen);
116  /*
117  * Set up pointers to parts of the block.
118  */
119  bf = hdr->bestfree;
120  btp = xfs_dir2_block_tail_p(mp, hdr);
121  blp = xfs_dir2_block_leaf_p(btp);
122  /*
123  * No stale entries? Need space for entry and new leaf.
124  */
125  if (!btp->stale) {
126  /*
127  * Tag just before the first leaf entry.
128  */
129  tagp = (__be16 *)blp - 1;
130  /*
131  * Data object just before the first leaf entry.
132  */
133  enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
134  /*
135  * If it's not free then can't do this add without cleaning up:
136  * the space before the first leaf entry needs to be free so it
137  * can be expanded to hold the pointer to the new entry.
138  */
140  dup = enddup = NULL;
141  /*
142  * Check out the biggest freespace and see if it's the same one.
143  */
144  else {
145  dup = (xfs_dir2_data_unused_t *)
146  ((char *)hdr + be16_to_cpu(bf[0].offset));
147  if (dup == enddup) {
148  /*
149  * It is the biggest freespace, is it too small
150  * to hold the new leaf too?
151  */
152  if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
153  /*
154  * Yes, we use the second-largest
155  * entry instead if it works.
156  */
157  if (be16_to_cpu(bf[1].length) >= len)
158  dup = (xfs_dir2_data_unused_t *)
159  ((char *)hdr +
160  be16_to_cpu(bf[1].offset));
161  else
162  dup = NULL;
163  }
164  } else {
165  /*
166  * Not the same free entry,
167  * just check its length.
168  */
169  if (be16_to_cpu(dup->length) < len) {
170  dup = NULL;
171  }
172  }
173  }
174  compact = 0;
175  }
176  /*
177  * If there are stale entries we'll use one for the leaf.
178  * Is the biggest entry enough to avoid compaction?
179  */
180  else if (be16_to_cpu(bf[0].length) >= len) {
181  dup = (xfs_dir2_data_unused_t *)
182  ((char *)hdr + be16_to_cpu(bf[0].offset));
183  compact = 0;
184  }
185  /*
186  * Will need to compact to make this work.
187  */
188  else {
189  /*
190  * Tag just before the first leaf entry.
191  */
192  tagp = (__be16 *)blp - 1;
193  /*
194  * Data object just before the first leaf entry.
195  */
196  dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
197  /*
198  * If it's not free then the data will go where the
199  * leaf data starts now, if it works at all.
200  */
202  if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
203  (uint)sizeof(*blp) < len)
204  dup = NULL;
205  } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
206  dup = NULL;
207  else
208  dup = (xfs_dir2_data_unused_t *)blp;
209  compact = 1;
210  }
211  /*
212  * If this isn't a real add, we're done with the buffer.
213  */
214  if (args->op_flags & XFS_DA_OP_JUSTCHECK)
215  xfs_trans_brelse(tp, bp);
216  /*
217  * If we don't have space for the new entry & leaf ...
218  */
219  if (!dup) {
220  /*
221  * Not trying to actually do anything, or don't have
222  * a space reservation: return no-space.
223  */
224  if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
225  return XFS_ERROR(ENOSPC);
226  /*
227  * Convert to the next larger format.
228  * Then add the new entry in that format.
229  */
230  error = xfs_dir2_block_to_leaf(args, bp);
231  if (error)
232  return error;
233  return xfs_dir2_leaf_addname(args);
234  }
235  /*
236  * Just checking, and it would work, so say so.
237  */
238  if (args->op_flags & XFS_DA_OP_JUSTCHECK)
239  return 0;
240  needlog = needscan = 0;
241  /*
242  * If need to compact the leaf entries, do it now.
243  * Leave the highest-numbered stale entry stale.
244  * XXX should be the one closest to mid but mid is not yet computed.
245  */
246  if (compact) {
247  int fromidx; /* source leaf index */
248  int toidx; /* target leaf index */
249 
250  for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
251  highstale = lfloghigh = -1;
252  fromidx >= 0;
253  fromidx--) {
254  if (blp[fromidx].address ==
256  if (highstale == -1)
257  highstale = toidx;
258  else {
259  if (lfloghigh == -1)
260  lfloghigh = toidx;
261  continue;
262  }
263  }
264  if (fromidx < toidx)
265  blp[toidx] = blp[fromidx];
266  toidx--;
267  }
268  lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
269  lfloghigh -= be32_to_cpu(btp->stale) - 1;
270  be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
272  (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
273  (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
274  &needlog, &needscan);
275  blp += be32_to_cpu(btp->stale) - 1;
276  btp->stale = cpu_to_be32(1);
277  /*
278  * If we now need to rebuild the bestfree map, do so.
279  * This needs to happen before the next call to use_free.
280  */
281  if (needscan) {
282  xfs_dir2_data_freescan(mp, hdr, &needlog);
283  needscan = 0;
284  }
285  }
286  /*
287  * Set leaf logging boundaries to impossible state.
288  * For the no-stale case they're set explicitly.
289  */
290  else if (btp->stale) {
291  lfloglow = be32_to_cpu(btp->count);
292  lfloghigh = -1;
293  }
294  /*
295  * Find the slot that's first lower than our hash value, -1 if none.
296  */
297  for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
298  mid = (low + high) >> 1;
299  if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
300  break;
301  if (hash < args->hashval)
302  low = mid + 1;
303  else
304  high = mid - 1;
305  }
306  while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
307  mid--;
308  }
309  /*
310  * No stale entries, will use enddup space to hold new leaf.
311  */
312  if (!btp->stale) {
313  /*
314  * Mark the space needed for the new leaf entry, now in use.
315  */
316  xfs_dir2_data_use_free(tp, bp, enddup,
318  ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
319  sizeof(*blp)),
320  (xfs_dir2_data_aoff_t)sizeof(*blp),
321  &needlog, &needscan);
322  /*
323  * Update the tail (entry count).
324  */
325  be32_add_cpu(&btp->count, 1);
326  /*
327  * If we now need to rebuild the bestfree map, do so.
328  * This needs to happen before the next call to use_free.
329  */
330  if (needscan) {
331  xfs_dir2_data_freescan(mp, hdr, &needlog);
332  needscan = 0;
333  }
334  /*
335  * Adjust pointer to the first leaf entry, we're about to move
336  * the table up one to open up space for the new leaf entry.
337  * Then adjust our index to match.
338  */
339  blp--;
340  mid++;
341  if (mid)
342  memmove(blp, &blp[1], mid * sizeof(*blp));
343  lfloglow = 0;
344  lfloghigh = mid;
345  }
346  /*
347  * Use a stale leaf for our new entry.
348  */
349  else {
350  for (lowstale = mid;
351  lowstale >= 0 &&
352  blp[lowstale].address !=
354  lowstale--)
355  continue;
356  for (highstale = mid + 1;
357  highstale < be32_to_cpu(btp->count) &&
358  blp[highstale].address !=
360  (lowstale < 0 || mid - lowstale > highstale - mid);
361  highstale++)
362  continue;
363  /*
364  * Move entries toward the low-numbered stale entry.
365  */
366  if (lowstale >= 0 &&
367  (highstale == be32_to_cpu(btp->count) ||
368  mid - lowstale <= highstale - mid)) {
369  if (mid - lowstale)
370  memmove(&blp[lowstale], &blp[lowstale + 1],
371  (mid - lowstale) * sizeof(*blp));
372  lfloglow = MIN(lowstale, lfloglow);
373  lfloghigh = MAX(mid, lfloghigh);
374  }
375  /*
376  * Move entries toward the high-numbered stale entry.
377  */
378  else {
379  ASSERT(highstale < be32_to_cpu(btp->count));
380  mid++;
381  if (highstale - mid)
382  memmove(&blp[mid + 1], &blp[mid],
383  (highstale - mid) * sizeof(*blp));
384  lfloglow = MIN(mid, lfloglow);
385  lfloghigh = MAX(highstale, lfloghigh);
386  }
387  be32_add_cpu(&btp->stale, -1);
388  }
389  /*
390  * Point to the new data entry.
391  */
392  dep = (xfs_dir2_data_entry_t *)dup;
393  /*
394  * Fill in the leaf entry.
395  */
396  blp[mid].hashval = cpu_to_be32(args->hashval);
397  blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
398  (char *)dep - (char *)hdr));
399  xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
400  /*
401  * Mark space for the data entry used.
402  */
403  xfs_dir2_data_use_free(tp, bp, dup,
404  (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
405  (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
406  /*
407  * Create the new data entry.
408  */
409  dep->inumber = cpu_to_be64(args->inumber);
410  dep->namelen = args->namelen;
411  memcpy(dep->name, args->name, args->namelen);
412  tagp = xfs_dir2_data_entry_tag_p(dep);
413  *tagp = cpu_to_be16((char *)dep - (char *)hdr);
414  /*
415  * Clean up the bestfree array and log the header, tail, and entry.
416  */
417  if (needscan)
418  xfs_dir2_data_freescan(mp, hdr, &needlog);
419  if (needlog)
420  xfs_dir2_data_log_header(tp, bp);
421  xfs_dir2_block_log_tail(tp, bp);
422  xfs_dir2_data_log_entry(tp, bp, dep);
423  xfs_dir2_data_check(dp, bp);
424  return 0;
425 }
426 
427 /*
428  * Readdir for block directories.
429  */
430 int /* error */
432  xfs_inode_t *dp, /* incore inode */
433  void *dirent,
434  xfs_off_t *offset,
435  filldir_t filldir)
436 {
437  xfs_dir2_data_hdr_t *hdr; /* block header */
438  struct xfs_buf *bp; /* buffer for block */
439  xfs_dir2_block_tail_t *btp; /* block tail */
440  xfs_dir2_data_entry_t *dep; /* block data entry */
441  xfs_dir2_data_unused_t *dup; /* block unused entry */
442  char *endptr; /* end of the data entries */
443  int error; /* error return value */
444  xfs_mount_t *mp; /* filesystem mount point */
445  char *ptr; /* current data entry */
446  int wantoff; /* starting block offset */
447  xfs_off_t cook;
448 
449  mp = dp->i_mount;
450  /*
451  * If the block number in the offset is out of range, we're done.
452  */
453  if (xfs_dir2_dataptr_to_db(mp, *offset) > mp->m_dirdatablk) {
454  return 0;
455  }
456  /*
457  * Can't read the block, give up, else get dabuf in bp.
458  */
459  error = xfs_da_read_buf(NULL, dp, mp->m_dirdatablk, -1,
460  &bp, XFS_DATA_FORK);
461  if (error)
462  return error;
463 
464  ASSERT(bp != NULL);
465  /*
466  * Extract the byte offset we start at from the seek pointer.
467  * We'll skip entries before this.
468  */
469  wantoff = xfs_dir2_dataptr_to_off(mp, *offset);
470  hdr = bp->b_addr;
471  xfs_dir2_data_check(dp, bp);
472  /*
473  * Set up values for the loop.
474  */
475  btp = xfs_dir2_block_tail_p(mp, hdr);
476  ptr = (char *)(hdr + 1);
477  endptr = (char *)xfs_dir2_block_leaf_p(btp);
478 
479  /*
480  * Loop over the data portion of the block.
481  * Each object is a real entry (dep) or an unused one (dup).
482  */
483  while (ptr < endptr) {
484  dup = (xfs_dir2_data_unused_t *)ptr;
485  /*
486  * Unused, skip it.
487  */
489  ptr += be16_to_cpu(dup->length);
490  continue;
491  }
492 
493  dep = (xfs_dir2_data_entry_t *)ptr;
494 
495  /*
496  * Bump pointer for the next iteration.
497  */
498  ptr += xfs_dir2_data_entsize(dep->namelen);
499  /*
500  * The entry is before the desired starting point, skip it.
501  */
502  if ((char *)dep - (char *)hdr < wantoff)
503  continue;
504 
505  cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
506  (char *)dep - (char *)hdr);
507 
508  /*
509  * If it didn't fit, set the final offset to here & return.
510  */
511  if (filldir(dirent, (char *)dep->name, dep->namelen,
512  cook & 0x7fffffff, be64_to_cpu(dep->inumber),
513  DT_UNKNOWN)) {
514  *offset = cook & 0x7fffffff;
515  xfs_trans_brelse(NULL, bp);
516  return 0;
517  }
518  }
519 
520  /*
521  * Reached the end of the block.
522  * Set the offset to a non-existent block 1 and return.
523  */
524  *offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
525  0x7fffffff;
526  xfs_trans_brelse(NULL, bp);
527  return 0;
528 }
529 
530 /*
531  * Log leaf entries from the block.
532  */
533 static void
534 xfs_dir2_block_log_leaf(
535  xfs_trans_t *tp, /* transaction structure */
536  struct xfs_buf *bp, /* block buffer */
537  int first, /* index of first logged leaf */
538  int last) /* index of last logged leaf */
539 {
543 
544  btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
545  blp = xfs_dir2_block_leaf_p(btp);
546  xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
547  (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
548 }
549 
550 /*
551  * Log the block tail.
552  */
553 static void
554 xfs_dir2_block_log_tail(
555  xfs_trans_t *tp, /* transaction structure */
556  struct xfs_buf *bp) /* block buffer */
557 {
558  xfs_dir2_data_hdr_t *hdr = bp->b_addr;
560 
561  btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
562  xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
563  (uint)((char *)(btp + 1) - (char *)hdr - 1));
564 }
565 
566 /*
567  * Look up an entry in the block. This is the external routine,
568  * xfs_dir2_block_lookup_int does the real work.
569  */
570 int /* error */
572  xfs_da_args_t *args) /* dir lookup arguments */
573 {
574  xfs_dir2_data_hdr_t *hdr; /* block header */
575  xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
576  struct xfs_buf *bp; /* block buffer */
577  xfs_dir2_block_tail_t *btp; /* block tail */
578  xfs_dir2_data_entry_t *dep; /* block data entry */
579  xfs_inode_t *dp; /* incore inode */
580  int ent; /* entry index */
581  int error; /* error return value */
582  xfs_mount_t *mp; /* filesystem mount point */
583 
584  trace_xfs_dir2_block_lookup(args);
585 
586  /*
587  * Get the buffer, look up the entry.
588  * If not found (ENOENT) then return, have no buffer.
589  */
590  if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
591  return error;
592  dp = args->dp;
593  mp = dp->i_mount;
594  hdr = bp->b_addr;
595  xfs_dir2_data_check(dp, bp);
596  btp = xfs_dir2_block_tail_p(mp, hdr);
597  blp = xfs_dir2_block_leaf_p(btp);
598  /*
599  * Get the offset from the leaf entry, to point to the data.
600  */
601  dep = (xfs_dir2_data_entry_t *)((char *)hdr +
602  xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
603  /*
604  * Fill in inode number, CI name if appropriate, release the block.
605  */
606  args->inumber = be64_to_cpu(dep->inumber);
607  error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
608  xfs_trans_brelse(args->trans, bp);
609  return XFS_ERROR(error);
610 }
611 
612 /*
613  * Internal block lookup routine.
614  */
615 static int /* error */
616 xfs_dir2_block_lookup_int(
617  xfs_da_args_t *args, /* dir lookup arguments */
618  struct xfs_buf **bpp, /* returned block buffer */
619  int *entno) /* returned entry number */
620 {
621  xfs_dir2_dataptr_t addr; /* data entry address */
622  xfs_dir2_data_hdr_t *hdr; /* block header */
623  xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
624  struct xfs_buf *bp; /* block buffer */
625  xfs_dir2_block_tail_t *btp; /* block tail */
626  xfs_dir2_data_entry_t *dep; /* block data entry */
627  xfs_inode_t *dp; /* incore inode */
628  int error; /* error return value */
629  xfs_dahash_t hash; /* found hash value */
630  int high; /* binary search high index */
631  int low; /* binary search low index */
632  int mid; /* binary search current idx */
633  xfs_mount_t *mp; /* filesystem mount point */
634  xfs_trans_t *tp; /* transaction pointer */
635  enum xfs_dacmp cmp; /* comparison result */
636 
637  dp = args->dp;
638  tp = args->trans;
639  mp = dp->i_mount;
640  /*
641  * Read the buffer, return error if we can't get it.
642  */
643  if ((error =
644  xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
645  return error;
646  }
647  ASSERT(bp != NULL);
648  hdr = bp->b_addr;
649  xfs_dir2_data_check(dp, bp);
650  btp = xfs_dir2_block_tail_p(mp, hdr);
651  blp = xfs_dir2_block_leaf_p(btp);
652  /*
653  * Loop doing a binary search for our hash value.
654  * Find our entry, ENOENT if it's not there.
655  */
656  for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
657  ASSERT(low <= high);
658  mid = (low + high) >> 1;
659  if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
660  break;
661  if (hash < args->hashval)
662  low = mid + 1;
663  else
664  high = mid - 1;
665  if (low > high) {
667  xfs_trans_brelse(tp, bp);
668  return XFS_ERROR(ENOENT);
669  }
670  }
671  /*
672  * Back up to the first one with the right hash value.
673  */
674  while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
675  mid--;
676  }
677  /*
678  * Now loop forward through all the entries with the
679  * right hash value looking for our name.
680  */
681  do {
682  if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
683  continue;
684  /*
685  * Get pointer to the entry from the leaf.
686  */
687  dep = (xfs_dir2_data_entry_t *)
688  ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
689  /*
690  * Compare name and if it's an exact match, return the index
691  * and buffer. If it's the first case-insensitive match, store
692  * the index and buffer and continue looking for an exact match.
693  */
694  cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
695  if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
696  args->cmpresult = cmp;
697  *bpp = bp;
698  *entno = mid;
699  if (cmp == XFS_CMP_EXACT)
700  return 0;
701  }
702  } while (++mid < be32_to_cpu(btp->count) &&
703  be32_to_cpu(blp[mid].hashval) == hash);
704 
706  /*
707  * Here, we can only be doing a lookup (not a rename or replace).
708  * If a case-insensitive match was found earlier, return success.
709  */
710  if (args->cmpresult == XFS_CMP_CASE)
711  return 0;
712  /*
713  * No match, release the buffer and return ENOENT.
714  */
715  xfs_trans_brelse(tp, bp);
716  return XFS_ERROR(ENOENT);
717 }
718 
719 /*
720  * Remove an entry from a block format directory.
721  * If that makes the block small enough to fit in shortform, transform it.
722  */
723 int /* error */
725  xfs_da_args_t *args) /* directory operation args */
726 {
727  xfs_dir2_data_hdr_t *hdr; /* block header */
728  xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
729  struct xfs_buf *bp; /* block buffer */
730  xfs_dir2_block_tail_t *btp; /* block tail */
731  xfs_dir2_data_entry_t *dep; /* block data entry */
732  xfs_inode_t *dp; /* incore inode */
733  int ent; /* block leaf entry index */
734  int error; /* error return value */
735  xfs_mount_t *mp; /* filesystem mount point */
736  int needlog; /* need to log block header */
737  int needscan; /* need to fixup bestfree */
738  xfs_dir2_sf_hdr_t sfh; /* shortform header */
739  int size; /* shortform size */
740  xfs_trans_t *tp; /* transaction pointer */
741 
742  trace_xfs_dir2_block_removename(args);
743 
744  /*
745  * Look up the entry in the block. Gets the buffer and entry index.
746  * It will always be there, the vnodeops level does a lookup first.
747  */
748  if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
749  return error;
750  }
751  dp = args->dp;
752  tp = args->trans;
753  mp = dp->i_mount;
754  hdr = bp->b_addr;
755  btp = xfs_dir2_block_tail_p(mp, hdr);
756  blp = xfs_dir2_block_leaf_p(btp);
757  /*
758  * Point to the data entry using the leaf entry.
759  */
760  dep = (xfs_dir2_data_entry_t *)
761  ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
762  /*
763  * Mark the data entry's space free.
764  */
765  needlog = needscan = 0;
767  (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
768  xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
769  /*
770  * Fix up the block tail.
771  */
772  be32_add_cpu(&btp->stale, 1);
773  xfs_dir2_block_log_tail(tp, bp);
774  /*
775  * Remove the leaf entry by marking it stale.
776  */
778  xfs_dir2_block_log_leaf(tp, bp, ent, ent);
779  /*
780  * Fix up bestfree, log the header if necessary.
781  */
782  if (needscan)
783  xfs_dir2_data_freescan(mp, hdr, &needlog);
784  if (needlog)
785  xfs_dir2_data_log_header(tp, bp);
786  xfs_dir2_data_check(dp, bp);
787  /*
788  * See if the size as a shortform is good enough.
789  */
790  size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
791  if (size > XFS_IFORK_DSIZE(dp))
792  return 0;
793 
794  /*
795  * If it works, do the conversion.
796  */
797  return xfs_dir2_block_to_sf(args, bp, size, &sfh);
798 }
799 
800 /*
801  * Replace an entry in a V2 block directory.
802  * Change the inode number to the new value.
803  */
804 int /* error */
806  xfs_da_args_t *args) /* directory operation args */
807 {
808  xfs_dir2_data_hdr_t *hdr; /* block header */
809  xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
810  struct xfs_buf *bp; /* block buffer */
811  xfs_dir2_block_tail_t *btp; /* block tail */
812  xfs_dir2_data_entry_t *dep; /* block data entry */
813  xfs_inode_t *dp; /* incore inode */
814  int ent; /* leaf entry index */
815  int error; /* error return value */
816  xfs_mount_t *mp; /* filesystem mount point */
817 
818  trace_xfs_dir2_block_replace(args);
819 
820  /*
821  * Lookup the entry in the directory. Get buffer and entry index.
822  * This will always succeed since the caller has already done a lookup.
823  */
824  if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
825  return error;
826  }
827  dp = args->dp;
828  mp = dp->i_mount;
829  hdr = bp->b_addr;
830  btp = xfs_dir2_block_tail_p(mp, hdr);
831  blp = xfs_dir2_block_leaf_p(btp);
832  /*
833  * Point to the data entry we need to change.
834  */
835  dep = (xfs_dir2_data_entry_t *)
836  ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
837  ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
838  /*
839  * Change the inode number to the new value.
840  */
841  dep->inumber = cpu_to_be64(args->inumber);
842  xfs_dir2_data_log_entry(args->trans, bp, dep);
843  xfs_dir2_data_check(dp, bp);
844  return 0;
845 }
846 
847 /*
848  * Qsort comparison routine for the block leaf entries.
849  */
850 static int /* sort order */
851 xfs_dir2_block_sort(
852  const void *a, /* first leaf entry */
853  const void *b) /* second leaf entry */
854 {
855  const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
856  const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
857 
858  la = a;
859  lb = b;
860  return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
861  (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
862 }
863 
864 /*
865  * Convert a V2 leaf directory to a V2 block directory if possible.
866  */
867 int /* error */
869  xfs_da_args_t *args, /* operation arguments */
870  struct xfs_buf *lbp, /* leaf buffer */
871  struct xfs_buf *dbp) /* data buffer */
872 {
873  __be16 *bestsp; /* leaf bests table */
874  xfs_dir2_data_hdr_t *hdr; /* block header */
875  xfs_dir2_block_tail_t *btp; /* block tail */
876  xfs_inode_t *dp; /* incore directory inode */
877  xfs_dir2_data_unused_t *dup; /* unused data entry */
878  int error; /* error return value */
879  int from; /* leaf from index */
880  xfs_dir2_leaf_t *leaf; /* leaf structure */
881  xfs_dir2_leaf_entry_t *lep; /* leaf entry */
882  xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
883  xfs_mount_t *mp; /* file system mount point */
884  int needlog; /* need to log data header */
885  int needscan; /* need to scan for bestfree */
886  xfs_dir2_sf_hdr_t sfh; /* shortform header */
887  int size; /* bytes used */
888  __be16 *tagp; /* end of entry (tag) */
889  int to; /* block/leaf to index */
890  xfs_trans_t *tp; /* transaction pointer */
891 
892  trace_xfs_dir2_leaf_to_block(args);
893 
894  dp = args->dp;
895  tp = args->trans;
896  mp = dp->i_mount;
897  leaf = lbp->b_addr;
899  ltp = xfs_dir2_leaf_tail_p(mp, leaf);
900  /*
901  * If there are data blocks other than the first one, take this
902  * opportunity to remove trailing empty data blocks that may have
903  * been left behind during no-space-reservation operations.
904  * These will show up in the leaf bests table.
905  */
906  while (dp->i_d.di_size > mp->m_dirblksize) {
907  bestsp = xfs_dir2_leaf_bests_p(ltp);
908  if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
909  mp->m_dirblksize - (uint)sizeof(*hdr)) {
910  if ((error =
911  xfs_dir2_leaf_trim_data(args, lbp,
912  (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
913  return error;
914  } else
915  return 0;
916  }
917  /*
918  * Read the data block if we don't already have it, give up if it fails.
919  */
920  if (dbp == NULL &&
921  (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
922  XFS_DATA_FORK))) {
923  return error;
924  }
925  hdr = dbp->b_addr;
927  /*
928  * Size of the "leaf" area in the block.
929  */
930  size = (uint)sizeof(xfs_dir2_block_tail_t) +
931  (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
932  /*
933  * Look at the last data entry.
934  */
935  tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
936  dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
937  /*
938  * If it's not free or is too short we can't do it.
939  */
941  be16_to_cpu(dup->length) < size)
942  return 0;
943 
944  /*
945  * Start converting it to block form.
946  */
948  needlog = 1;
949  needscan = 0;
950  /*
951  * Use up the space at the end of the block (blp/btp).
952  */
953  xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
954  &needlog, &needscan);
955  /*
956  * Initialize the block tail.
957  */
958  btp = xfs_dir2_block_tail_p(mp, hdr);
959  btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
960  btp->stale = 0;
961  xfs_dir2_block_log_tail(tp, dbp);
962  /*
963  * Initialize the block leaf area. We compact out stale entries.
964  */
965  lep = xfs_dir2_block_leaf_p(btp);
966  for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
967  if (leaf->ents[from].address ==
969  continue;
970  lep[to++] = leaf->ents[from];
971  }
972  ASSERT(to == be32_to_cpu(btp->count));
973  xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
974  /*
975  * Scan the bestfree if we need it and log the data block header.
976  */
977  if (needscan)
978  xfs_dir2_data_freescan(mp, hdr, &needlog);
979  if (needlog)
980  xfs_dir2_data_log_header(tp, dbp);
981  /*
982  * Pitch the old leaf block.
983  */
984  error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
985  if (error)
986  return error;
987 
988  /*
989  * Now see if the resulting block can be shrunken to shortform.
990  */
991  size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
992  if (size > XFS_IFORK_DSIZE(dp))
993  return 0;
994 
995  return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
996 }
997 
998 /*
999  * Convert the shortform directory to block form.
1000  */
1001 int /* error */
1003  xfs_da_args_t *args) /* operation arguments */
1004 {
1005  xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1006  xfs_dir2_data_hdr_t *hdr; /* block header */
1007  xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1008  struct xfs_buf *bp; /* block buffer */
1009  xfs_dir2_block_tail_t *btp; /* block tail pointer */
1010  xfs_dir2_data_entry_t *dep; /* data entry pointer */
1011  xfs_inode_t *dp; /* incore directory inode */
1012  int dummy; /* trash */
1013  xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1014  int endoffset; /* end of data objects */
1015  int error; /* error return value */
1016  int i; /* index */
1017  xfs_mount_t *mp; /* filesystem mount point */
1018  int needlog; /* need to log block header */
1019  int needscan; /* need to scan block freespc */
1020  int newoffset; /* offset from current entry */
1021  int offset; /* target block offset */
1022  xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1023  xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1024  xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1025  __be16 *tagp; /* end of data entry */
1026  xfs_trans_t *tp; /* transaction pointer */
1027  struct xfs_name name;
1028 
1029  trace_xfs_dir2_sf_to_block(args);
1030 
1031  dp = args->dp;
1032  tp = args->trans;
1033  mp = dp->i_mount;
1034  ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1035  /*
1036  * Bomb out if the shortform directory is way too short.
1037  */
1038  if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1039  ASSERT(XFS_FORCED_SHUTDOWN(mp));
1040  return XFS_ERROR(EIO);
1041  }
1042 
1043  oldsfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
1044 
1045  ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1046  ASSERT(dp->i_df.if_u1.if_data != NULL);
1047  ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1048 
1049  /*
1050  * Copy the directory into a temporary buffer.
1051  * Then pitch the incore inode data so we can make extents.
1052  */
1053  sfp = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1054  memcpy(sfp, oldsfp, dp->i_df.if_bytes);
1055 
1056  xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1057  dp->i_d.di_size = 0;
1059 
1060  /*
1061  * Add block 0 to the inode.
1062  */
1063  error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1064  if (error) {
1065  kmem_free(sfp);
1066  return error;
1067  }
1068  /*
1069  * Initialize the data block.
1070  */
1071  error = xfs_dir2_data_init(args, blkno, &bp);
1072  if (error) {
1073  kmem_free(sfp);
1074  return error;
1075  }
1076  hdr = bp->b_addr;
1078  /*
1079  * Compute size of block "tail" area.
1080  */
1081  i = (uint)sizeof(*btp) +
1082  (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1083  /*
1084  * The whole thing is initialized to free by the init routine.
1085  * Say we're using the leaf and tail area.
1086  */
1087  dup = (xfs_dir2_data_unused_t *)(hdr + 1);
1088  needlog = needscan = 0;
1089  xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1090  &needscan);
1091  ASSERT(needscan == 0);
1092  /*
1093  * Fill in the tail.
1094  */
1095  btp = xfs_dir2_block_tail_p(mp, hdr);
1096  btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1097  btp->stale = 0;
1098  blp = xfs_dir2_block_leaf_p(btp);
1099  endoffset = (uint)((char *)blp - (char *)hdr);
1100  /*
1101  * Remove the freespace, we'll manage it.
1102  */
1103  xfs_dir2_data_use_free(tp, bp, dup,
1104  (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1105  be16_to_cpu(dup->length), &needlog, &needscan);
1106  /*
1107  * Create entry for .
1108  */
1109  dep = (xfs_dir2_data_entry_t *)
1110  ((char *)hdr + XFS_DIR2_DATA_DOT_OFFSET);
1111  dep->inumber = cpu_to_be64(dp->i_ino);
1112  dep->namelen = 1;
1113  dep->name[0] = '.';
1114  tagp = xfs_dir2_data_entry_tag_p(dep);
1115  *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1116  xfs_dir2_data_log_entry(tp, bp, dep);
1117  blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1118  blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1119  (char *)dep - (char *)hdr));
1120  /*
1121  * Create entry for ..
1122  */
1123  dep = (xfs_dir2_data_entry_t *)
1124  ((char *)hdr + XFS_DIR2_DATA_DOTDOT_OFFSET);
1126  dep->namelen = 2;
1127  dep->name[0] = dep->name[1] = '.';
1128  tagp = xfs_dir2_data_entry_tag_p(dep);
1129  *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1130  xfs_dir2_data_log_entry(tp, bp, dep);
1131  blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1132  blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1133  (char *)dep - (char *)hdr));
1134  offset = XFS_DIR2_DATA_FIRST_OFFSET;
1135  /*
1136  * Loop over existing entries, stuff them in.
1137  */
1138  i = 0;
1139  if (!sfp->count)
1140  sfep = NULL;
1141  else
1142  sfep = xfs_dir2_sf_firstentry(sfp);
1143  /*
1144  * Need to preserve the existing offset values in the sf directory.
1145  * Insert holes (unused entries) where necessary.
1146  */
1147  while (offset < endoffset) {
1148  /*
1149  * sfep is null when we reach the end of the list.
1150  */
1151  if (sfep == NULL)
1152  newoffset = endoffset;
1153  else
1154  newoffset = xfs_dir2_sf_get_offset(sfep);
1155  /*
1156  * There should be a hole here, make one.
1157  */
1158  if (offset < newoffset) {
1159  dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1161  dup->length = cpu_to_be16(newoffset - offset);
1162  *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1163  ((char *)dup - (char *)hdr));
1164  xfs_dir2_data_log_unused(tp, bp, dup);
1165  xfs_dir2_data_freeinsert(hdr, dup, &dummy);
1166  offset += be16_to_cpu(dup->length);
1167  continue;
1168  }
1169  /*
1170  * Copy a real entry.
1171  */
1172  dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1173  dep->inumber = cpu_to_be64(xfs_dir2_sfe_get_ino(sfp, sfep));
1174  dep->namelen = sfep->namelen;
1175  memcpy(dep->name, sfep->name, dep->namelen);
1176  tagp = xfs_dir2_data_entry_tag_p(dep);
1177  *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1178  xfs_dir2_data_log_entry(tp, bp, dep);
1179  name.name = sfep->name;
1180  name.len = sfep->namelen;
1181  blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1182  hashname(&name));
1183  blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1184  (char *)dep - (char *)hdr));
1185  offset = (int)((char *)(tagp + 1) - (char *)hdr);
1186  if (++i == sfp->count)
1187  sfep = NULL;
1188  else
1189  sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1190  }
1191  /* Done with the temporary buffer */
1192  kmem_free(sfp);
1193  /*
1194  * Sort the leaf entries by hash value.
1195  */
1196  xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1197  /*
1198  * Log the leaf entry area and tail.
1199  * Already logged the header in data_init, ignore needlog.
1200  */
1201  ASSERT(needscan == 0);
1202  xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1203  xfs_dir2_block_log_tail(tp, bp);
1204  xfs_dir2_data_check(dp, bp);
1205  return 0;
1206 }