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

TOMOYO Linux Cross Reference
Linux/fs/xfs/libxfs/xfs_dir2_leaf.c

Version: ~ [ linux-6.11-rc3 ] ~ [ linux-6.10.4 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.45 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.104 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.164 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.223 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.281 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.319 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
  4  * Copyright (c) 2013 Red Hat, Inc.
  5  * All Rights Reserved.
  6  */
  7 #include "xfs.h"
  8 #include "xfs_fs.h"
  9 #include "xfs_shared.h"
 10 #include "xfs_format.h"
 11 #include "xfs_log_format.h"
 12 #include "xfs_trans_resv.h"
 13 #include "xfs_mount.h"
 14 #include "xfs_inode.h"
 15 #include "xfs_bmap.h"
 16 #include "xfs_dir2.h"
 17 #include "xfs_dir2_priv.h"
 18 #include "xfs_error.h"
 19 #include "xfs_trace.h"
 20 #include "xfs_trans.h"
 21 #include "xfs_buf_item.h"
 22 #include "xfs_health.h"
 23 
 24 /*
 25  * Local function declarations.
 26  */
 27 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
 28                                     int *indexp, struct xfs_buf **dbpp,
 29                                     struct xfs_dir3_icleaf_hdr *leafhdr);
 30 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
 31                                     struct xfs_buf *bp, int first, int last);
 32 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
 33                                    struct xfs_buf *bp);
 34 
 35 void
 36 xfs_dir2_leaf_hdr_from_disk(
 37         struct xfs_mount                *mp,
 38         struct xfs_dir3_icleaf_hdr      *to,
 39         struct xfs_dir2_leaf            *from)
 40 {
 41         if (xfs_has_crc(mp)) {
 42                 struct xfs_dir3_leaf *from3 = (struct xfs_dir3_leaf *)from;
 43 
 44                 to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
 45                 to->back = be32_to_cpu(from3->hdr.info.hdr.back);
 46                 to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
 47                 to->count = be16_to_cpu(from3->hdr.count);
 48                 to->stale = be16_to_cpu(from3->hdr.stale);
 49                 to->ents = from3->__ents;
 50 
 51                 ASSERT(to->magic == XFS_DIR3_LEAF1_MAGIC ||
 52                        to->magic == XFS_DIR3_LEAFN_MAGIC);
 53         } else {
 54                 to->forw = be32_to_cpu(from->hdr.info.forw);
 55                 to->back = be32_to_cpu(from->hdr.info.back);
 56                 to->magic = be16_to_cpu(from->hdr.info.magic);
 57                 to->count = be16_to_cpu(from->hdr.count);
 58                 to->stale = be16_to_cpu(from->hdr.stale);
 59                 to->ents = from->__ents;
 60 
 61                 ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
 62                        to->magic == XFS_DIR2_LEAFN_MAGIC);
 63         }
 64 }
 65 
 66 void
 67 xfs_dir2_leaf_hdr_to_disk(
 68         struct xfs_mount                *mp,
 69         struct xfs_dir2_leaf            *to,
 70         struct xfs_dir3_icleaf_hdr      *from)
 71 {
 72         if (xfs_has_crc(mp)) {
 73                 struct xfs_dir3_leaf *to3 = (struct xfs_dir3_leaf *)to;
 74 
 75                 ASSERT(from->magic == XFS_DIR3_LEAF1_MAGIC ||
 76                        from->magic == XFS_DIR3_LEAFN_MAGIC);
 77 
 78                 to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
 79                 to3->hdr.info.hdr.back = cpu_to_be32(from->back);
 80                 to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
 81                 to3->hdr.count = cpu_to_be16(from->count);
 82                 to3->hdr.stale = cpu_to_be16(from->stale);
 83         } else {
 84                 ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
 85                        from->magic == XFS_DIR2_LEAFN_MAGIC);
 86 
 87                 to->hdr.info.forw = cpu_to_be32(from->forw);
 88                 to->hdr.info.back = cpu_to_be32(from->back);
 89                 to->hdr.info.magic = cpu_to_be16(from->magic);
 90                 to->hdr.count = cpu_to_be16(from->count);
 91                 to->hdr.stale = cpu_to_be16(from->stale);
 92         }
 93 }
 94 
 95 /*
 96  * Check the internal consistency of a leaf1 block.
 97  * Pop an assert if something is wrong.
 98  */
 99 #ifdef DEBUG
100 static xfs_failaddr_t
101 xfs_dir3_leaf1_check(
102         struct xfs_inode        *dp,
103         struct xfs_buf          *bp)
104 {
105         struct xfs_dir2_leaf    *leaf = bp->b_addr;
106         struct xfs_dir3_icleaf_hdr leafhdr;
107 
108         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
109 
110         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
111                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
112                 if (be64_to_cpu(leaf3->info.blkno) != xfs_buf_daddr(bp))
113                         return __this_address;
114         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
115                 return __this_address;
116 
117         return xfs_dir3_leaf_check_int(dp->i_mount, &leafhdr, leaf, false);
118 }
119 
120 static inline void
121 xfs_dir3_leaf_check(
122         struct xfs_inode        *dp,
123         struct xfs_buf          *bp)
124 {
125         xfs_failaddr_t          fa;
126 
127         fa = xfs_dir3_leaf1_check(dp, bp);
128         if (!fa)
129                 return;
130         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
131                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
132                         fa);
133         ASSERT(0);
134 }
135 #else
136 #define xfs_dir3_leaf_check(dp, bp)
137 #endif
138 
139 xfs_failaddr_t
140 xfs_dir3_leaf_check_int(
141         struct xfs_mount                *mp,
142         struct xfs_dir3_icleaf_hdr      *hdr,
143         struct xfs_dir2_leaf            *leaf,
144         bool                            expensive_checking)
145 {
146         struct xfs_da_geometry          *geo = mp->m_dir_geo;
147         xfs_dir2_leaf_tail_t            *ltp;
148         int                             stale;
149         int                             i;
150         bool                            isleaf1 = (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
151                                                    hdr->magic == XFS_DIR3_LEAF1_MAGIC);
152 
153         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
154 
155         /*
156          * XXX (dgc): This value is not restrictive enough.
157          * Should factor in the size of the bests table as well.
158          * We can deduce a value for that from i_disk_size.
159          */
160         if (hdr->count > geo->leaf_max_ents)
161                 return __this_address;
162 
163         /* Leaves and bests don't overlap in leaf format. */
164         if (isleaf1 &&
165             (char *)&hdr->ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
166                 return __this_address;
167 
168         if (!expensive_checking)
169                 return NULL;
170 
171         /* Check hash value order, count stale entries.  */
172         for (i = stale = 0; i < hdr->count; i++) {
173                 if (i + 1 < hdr->count) {
174                         if (be32_to_cpu(hdr->ents[i].hashval) >
175                                         be32_to_cpu(hdr->ents[i + 1].hashval))
176                                 return __this_address;
177                 }
178                 if (hdr->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
179                         stale++;
180                 if (isleaf1 && xfs_dir2_dataptr_to_db(geo,
181                                 be32_to_cpu(hdr->ents[i].address)) >=
182                                 be32_to_cpu(ltp->bestcount))
183                         return __this_address;
184         }
185         if (hdr->stale != stale)
186                 return __this_address;
187         return NULL;
188 }
189 
190 /*
191  * We verify the magic numbers before decoding the leaf header so that on debug
192  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
193  * to incorrect magic numbers.
194  */
195 static xfs_failaddr_t
196 xfs_dir3_leaf_verify(
197         struct xfs_buf                  *bp)
198 {
199         struct xfs_mount                *mp = bp->b_mount;
200         struct xfs_dir3_icleaf_hdr      leafhdr;
201         xfs_failaddr_t                  fa;
202 
203         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
204         if (fa)
205                 return fa;
206 
207         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, bp->b_addr);
208         return xfs_dir3_leaf_check_int(mp, &leafhdr, bp->b_addr, true);
209 }
210 
211 xfs_failaddr_t
212 xfs_dir3_leaf_header_check(
213         struct xfs_buf          *bp,
214         xfs_ino_t               owner)
215 {
216         struct xfs_mount        *mp = bp->b_mount;
217 
218         if (xfs_has_crc(mp)) {
219                 struct xfs_dir3_leaf *hdr3 = bp->b_addr;
220 
221                 if (hdr3->hdr.info.hdr.magic !=
222                                         cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) &&
223                     hdr3->hdr.info.hdr.magic !=
224                                         cpu_to_be16(XFS_DIR3_LEAFN_MAGIC))
225                         return __this_address;
226 
227                 if (be64_to_cpu(hdr3->hdr.info.owner) != owner)
228                         return __this_address;
229         }
230 
231         return NULL;
232 }
233 
234 static void
235 xfs_dir3_leaf_read_verify(
236         struct xfs_buf  *bp)
237 {
238         struct xfs_mount        *mp = bp->b_mount;
239         xfs_failaddr_t          fa;
240 
241         if (xfs_has_crc(mp) &&
242              !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
243                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
244         else {
245                 fa = xfs_dir3_leaf_verify(bp);
246                 if (fa)
247                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
248         }
249 }
250 
251 static void
252 xfs_dir3_leaf_write_verify(
253         struct xfs_buf  *bp)
254 {
255         struct xfs_mount        *mp = bp->b_mount;
256         struct xfs_buf_log_item *bip = bp->b_log_item;
257         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
258         xfs_failaddr_t          fa;
259 
260         fa = xfs_dir3_leaf_verify(bp);
261         if (fa) {
262                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
263                 return;
264         }
265 
266         if (!xfs_has_crc(mp))
267                 return;
268 
269         if (bip)
270                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
271 
272         xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
273 }
274 
275 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
276         .name = "xfs_dir3_leaf1",
277         .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
278                      cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
279         .verify_read = xfs_dir3_leaf_read_verify,
280         .verify_write = xfs_dir3_leaf_write_verify,
281         .verify_struct = xfs_dir3_leaf_verify,
282 };
283 
284 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
285         .name = "xfs_dir3_leafn",
286         .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
287                      cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
288         .verify_read = xfs_dir3_leaf_read_verify,
289         .verify_write = xfs_dir3_leaf_write_verify,
290         .verify_struct = xfs_dir3_leaf_verify,
291 };
292 
293 int
294 xfs_dir3_leaf_read(
295         struct xfs_trans        *tp,
296         struct xfs_inode        *dp,
297         xfs_ino_t               owner,
298         xfs_dablk_t             fbno,
299         struct xfs_buf          **bpp)
300 {
301         xfs_failaddr_t          fa;
302         int                     err;
303 
304         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
305                         &xfs_dir3_leaf1_buf_ops);
306         if (err || !(*bpp))
307                 return err;
308 
309         fa = xfs_dir3_leaf_header_check(*bpp, owner);
310         if (fa) {
311                 __xfs_buf_mark_corrupt(*bpp, fa);
312                 xfs_trans_brelse(tp, *bpp);
313                 *bpp = NULL;
314                 xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
315                 return -EFSCORRUPTED;
316         }
317 
318         if (tp)
319                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
320         return 0;
321 }
322 
323 int
324 xfs_dir3_leafn_read(
325         struct xfs_trans        *tp,
326         struct xfs_inode        *dp,
327         xfs_ino_t               owner,
328         xfs_dablk_t             fbno,
329         struct xfs_buf          **bpp)
330 {
331         xfs_failaddr_t          fa;
332         int                     err;
333 
334         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
335                         &xfs_dir3_leafn_buf_ops);
336         if (err || !(*bpp))
337                 return err;
338 
339         fa = xfs_dir3_leaf_header_check(*bpp, owner);
340         if (fa) {
341                 __xfs_buf_mark_corrupt(*bpp, fa);
342                 xfs_trans_brelse(tp, *bpp);
343                 *bpp = NULL;
344                 xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
345                 return -EFSCORRUPTED;
346         }
347 
348         if (tp)
349                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
350         return 0;
351 }
352 
353 /*
354  * Initialize a new leaf block, leaf1 or leafn magic accepted.
355  */
356 static void
357 xfs_dir3_leaf_init(
358         struct xfs_da_args      *args,
359         struct xfs_buf          *bp,
360         uint16_t                type)
361 {
362         struct xfs_mount        *mp = args->dp->i_mount;
363         struct xfs_trans        *tp = args->trans;
364         struct xfs_dir2_leaf    *leaf = bp->b_addr;
365 
366         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
367 
368         if (xfs_has_crc(mp)) {
369                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
370 
371                 memset(leaf3, 0, sizeof(*leaf3));
372 
373                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
374                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
375                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
376                 leaf3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
377                 leaf3->info.owner = cpu_to_be64(args->owner);
378                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
379         } else {
380                 memset(leaf, 0, sizeof(*leaf));
381                 leaf->hdr.info.magic = cpu_to_be16(type);
382         }
383 
384         /*
385          * If it's a leaf-format directory initialize the tail.
386          * Caller is responsible for initialising the bests table.
387          */
388         if (type == XFS_DIR2_LEAF1_MAGIC) {
389                 struct xfs_dir2_leaf_tail *ltp;
390 
391                 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
392                 ltp->bestcount = 0;
393                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
394                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
395         } else {
396                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
397                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
398         }
399 }
400 
401 int
402 xfs_dir3_leaf_get_buf(
403         xfs_da_args_t           *args,
404         xfs_dir2_db_t           bno,
405         struct xfs_buf          **bpp,
406         uint16_t                magic)
407 {
408         struct xfs_inode        *dp = args->dp;
409         struct xfs_trans        *tp = args->trans;
410         struct xfs_buf          *bp;
411         int                     error;
412 
413         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
414         ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
415                bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
416 
417         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
418                                &bp, XFS_DATA_FORK);
419         if (error)
420                 return error;
421 
422         xfs_dir3_leaf_init(args, bp, magic);
423         xfs_dir3_leaf_log_header(args, bp);
424         if (magic == XFS_DIR2_LEAF1_MAGIC)
425                 xfs_dir3_leaf_log_tail(args, bp);
426         *bpp = bp;
427         return 0;
428 }
429 
430 /*
431  * Convert a block form directory to a leaf form directory.
432  */
433 int                                             /* error */
434 xfs_dir2_block_to_leaf(
435         xfs_da_args_t           *args,          /* operation arguments */
436         struct xfs_buf          *dbp)           /* input block's buffer */
437 {
438         __be16                  *bestsp;        /* leaf's bestsp entries */
439         xfs_dablk_t             blkno;          /* leaf block's bno */
440         xfs_dir2_data_hdr_t     *hdr;           /* block header */
441         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
442         xfs_dir2_block_tail_t   *btp;           /* block's tail */
443         xfs_inode_t             *dp;            /* incore directory inode */
444         int                     error;          /* error return code */
445         struct xfs_buf          *lbp;           /* leaf block's buffer */
446         xfs_dir2_db_t           ldb;            /* leaf block's bno */
447         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
448         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
449         int                     needlog;        /* need to log block header */
450         int                     needscan;       /* need to rescan bestfree */
451         xfs_trans_t             *tp;            /* transaction pointer */
452         struct xfs_dir2_data_free *bf;
453         struct xfs_dir3_icleaf_hdr leafhdr;
454 
455         trace_xfs_dir2_block_to_leaf(args);
456 
457         dp = args->dp;
458         tp = args->trans;
459         /*
460          * Add the leaf block to the inode.
461          * This interface will only put blocks in the leaf/node range.
462          * Since that's empty now, we'll get the root (block 0 in range).
463          */
464         if ((error = xfs_da_grow_inode(args, &blkno))) {
465                 return error;
466         }
467         ldb = xfs_dir2_da_to_db(args->geo, blkno);
468         ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
469         /*
470          * Initialize the leaf block, get a buffer for it.
471          */
472         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
473         if (error)
474                 return error;
475 
476         leaf = lbp->b_addr;
477         hdr = dbp->b_addr;
478         xfs_dir3_data_check(dp, dbp);
479         btp = xfs_dir2_block_tail_p(args->geo, hdr);
480         blp = xfs_dir2_block_leaf_p(btp);
481         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
482 
483         /*
484          * Set the counts in the leaf header.
485          */
486         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
487         leafhdr.count = be32_to_cpu(btp->count);
488         leafhdr.stale = be32_to_cpu(btp->stale);
489         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
490         xfs_dir3_leaf_log_header(args, lbp);
491 
492         /*
493          * Could compact these but I think we always do the conversion
494          * after squeezing out stale entries.
495          */
496         memcpy(leafhdr.ents, blp,
497                 be32_to_cpu(btp->count) * sizeof(struct xfs_dir2_leaf_entry));
498         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, 0, leafhdr.count - 1);
499         needscan = 0;
500         needlog = 1;
501         /*
502          * Make the space formerly occupied by the leaf entries and block
503          * tail be free.
504          */
505         xfs_dir2_data_make_free(args, dbp,
506                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
507                 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
508                                        (char *)blp),
509                 &needlog, &needscan);
510         /*
511          * Fix up the block header, make it a data block.
512          */
513         dbp->b_ops = &xfs_dir3_data_buf_ops;
514         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
515         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
516                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
517         else
518                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
519 
520         if (needscan)
521                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
522         /*
523          * Set up leaf tail and bests table.
524          */
525         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
526         ltp->bestcount = cpu_to_be32(1);
527         bestsp = xfs_dir2_leaf_bests_p(ltp);
528         bestsp[0] =  bf[0].length;
529         /*
530          * Log the data header and leaf bests table.
531          */
532         if (needlog)
533                 xfs_dir2_data_log_header(args, dbp);
534         xfs_dir3_leaf_check(dp, lbp);
535         xfs_dir3_data_check(dp, dbp);
536         xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
537         return 0;
538 }
539 
540 STATIC void
541 xfs_dir3_leaf_find_stale(
542         struct xfs_dir3_icleaf_hdr *leafhdr,
543         struct xfs_dir2_leaf_entry *ents,
544         int                     index,
545         int                     *lowstale,
546         int                     *highstale)
547 {
548         /*
549          * Find the first stale entry before our index, if any.
550          */
551         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
552                 if (ents[*lowstale].address ==
553                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
554                         break;
555         }
556 
557         /*
558          * Find the first stale entry at or after our index, if any.
559          * Stop if the result would require moving more entries than using
560          * lowstale.
561          */
562         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
563                 if (ents[*highstale].address ==
564                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
565                         break;
566                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
567                         break;
568         }
569 }
570 
571 struct xfs_dir2_leaf_entry *
572 xfs_dir3_leaf_find_entry(
573         struct xfs_dir3_icleaf_hdr *leafhdr,
574         struct xfs_dir2_leaf_entry *ents,
575         int                     index,          /* leaf table position */
576         int                     compact,        /* need to compact leaves */
577         int                     lowstale,       /* index of prev stale leaf */
578         int                     highstale,      /* index of next stale leaf */
579         int                     *lfloglow,      /* low leaf logging index */
580         int                     *lfloghigh)     /* high leaf logging index */
581 {
582         if (!leafhdr->stale) {
583                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
584 
585                 /*
586                  * Now we need to make room to insert the leaf entry.
587                  *
588                  * If there are no stale entries, just insert a hole at index.
589                  */
590                 lep = &ents[index];
591                 if (index < leafhdr->count)
592                         memmove(lep + 1, lep,
593                                 (leafhdr->count - index) * sizeof(*lep));
594 
595                 /*
596                  * Record low and high logging indices for the leaf.
597                  */
598                 *lfloglow = index;
599                 *lfloghigh = leafhdr->count++;
600                 return lep;
601         }
602 
603         /*
604          * There are stale entries.
605          *
606          * We will use one of them for the new entry.  It's probably not at
607          * the right location, so we'll have to shift some up or down first.
608          *
609          * If we didn't compact before, we need to find the nearest stale
610          * entries before and after our insertion point.
611          */
612         if (compact == 0)
613                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
614                                          &lowstale, &highstale);
615 
616         /*
617          * If the low one is better, use it.
618          */
619         if (lowstale >= 0 &&
620             (highstale == leafhdr->count ||
621              index - lowstale - 1 < highstale - index)) {
622                 ASSERT(index - lowstale - 1 >= 0);
623                 ASSERT(ents[lowstale].address ==
624                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
625 
626                 /*
627                  * Copy entries up to cover the stale entry and make room
628                  * for the new entry.
629                  */
630                 if (index - lowstale - 1 > 0) {
631                         memmove(&ents[lowstale], &ents[lowstale + 1],
632                                 (index - lowstale - 1) *
633                                         sizeof(xfs_dir2_leaf_entry_t));
634                 }
635                 *lfloglow = min(lowstale, *lfloglow);
636                 *lfloghigh = max(index - 1, *lfloghigh);
637                 leafhdr->stale--;
638                 return &ents[index - 1];
639         }
640 
641         /*
642          * The high one is better, so use that one.
643          */
644         ASSERT(highstale - index >= 0);
645         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
646 
647         /*
648          * Copy entries down to cover the stale entry and make room for the
649          * new entry.
650          */
651         if (highstale - index > 0) {
652                 memmove(&ents[index + 1], &ents[index],
653                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
654         }
655         *lfloglow = min(index, *lfloglow);
656         *lfloghigh = max(highstale, *lfloghigh);
657         leafhdr->stale--;
658         return &ents[index];
659 }
660 
661 /*
662  * Add an entry to a leaf form directory.
663  */
664 int                                             /* error */
665 xfs_dir2_leaf_addname(
666         struct xfs_da_args      *args)          /* operation arguments */
667 {
668         struct xfs_dir3_icleaf_hdr leafhdr;
669         struct xfs_trans        *tp = args->trans;
670         __be16                  *bestsp;        /* freespace table in leaf */
671         __be16                  *tagp;          /* end of data entry */
672         struct xfs_buf          *dbp;           /* data block buffer */
673         struct xfs_buf          *lbp;           /* leaf's buffer */
674         struct xfs_dir2_leaf    *leaf;          /* leaf structure */
675         struct xfs_inode        *dp = args->dp; /* incore directory inode */
676         struct xfs_dir2_data_hdr *hdr;          /* data block header */
677         struct xfs_dir2_data_entry *dep;        /* data block entry */
678         struct xfs_dir2_leaf_entry *lep;        /* leaf entry table pointer */
679         struct xfs_dir2_leaf_entry *ents;
680         struct xfs_dir2_data_unused *dup;       /* data unused entry */
681         struct xfs_dir2_leaf_tail *ltp;         /* leaf tail pointer */
682         struct xfs_dir2_data_free *bf;          /* bestfree table */
683         int                     compact;        /* need to compact leaves */
684         int                     error;          /* error return value */
685         int                     grown;          /* allocated new data block */
686         int                     highstale = 0;  /* index of next stale leaf */
687         int                     i;              /* temporary, index */
688         int                     index;          /* leaf table position */
689         int                     length;         /* length of new entry */
690         int                     lfloglow;       /* low leaf logging index */
691         int                     lfloghigh;      /* high leaf logging index */
692         int                     lowstale = 0;   /* index of prev stale leaf */
693         int                     needbytes;      /* leaf block bytes needed */
694         int                     needlog;        /* need to log data header */
695         int                     needscan;       /* need to rescan data free */
696         xfs_dir2_db_t           use_block;      /* data block number */
697 
698         trace_xfs_dir2_leaf_addname(args);
699 
700         error = xfs_dir3_leaf_read(tp, dp, args->owner, args->geo->leafblk,
701                         &lbp);
702         if (error)
703                 return error;
704 
705         /*
706          * Look up the entry by hash value and name.
707          * We know it's not there, our caller has already done a lookup.
708          * So the index is of the entry to insert in front of.
709          * But if there are dup hash values the index is of the first of those.
710          */
711         index = xfs_dir2_leaf_search_hash(args, lbp);
712         leaf = lbp->b_addr;
713         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
714         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
715         ents = leafhdr.ents;
716         bestsp = xfs_dir2_leaf_bests_p(ltp);
717         length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
718 
719         /*
720          * See if there are any entries with the same hash value
721          * and space in their block for the new entry.
722          * This is good because it puts multiple same-hash value entries
723          * in a data block, improving the lookup of those entries.
724          */
725         for (use_block = -1, lep = &ents[index];
726              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
727              index++, lep++) {
728                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
729                         continue;
730                 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
731                 ASSERT(i < be32_to_cpu(ltp->bestcount));
732                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
733                 if (be16_to_cpu(bestsp[i]) >= length) {
734                         use_block = i;
735                         break;
736                 }
737         }
738         /*
739          * Didn't find a block yet, linear search all the data blocks.
740          */
741         if (use_block == -1) {
742                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
743                         /*
744                          * Remember a block we see that's missing.
745                          */
746                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
747                             use_block == -1)
748                                 use_block = i;
749                         else if (be16_to_cpu(bestsp[i]) >= length) {
750                                 use_block = i;
751                                 break;
752                         }
753                 }
754         }
755         /*
756          * How many bytes do we need in the leaf block?
757          */
758         needbytes = 0;
759         if (!leafhdr.stale)
760                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
761         if (use_block == -1)
762                 needbytes += sizeof(xfs_dir2_data_off_t);
763 
764         /*
765          * Now kill use_block if it refers to a missing block, so we
766          * can use it as an indication of allocation needed.
767          */
768         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
769                 use_block = -1;
770         /*
771          * If we don't have enough free bytes but we can make enough
772          * by compacting out stale entries, we'll do that.
773          */
774         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
775             leafhdr.stale > 1)
776                 compact = 1;
777 
778         /*
779          * Otherwise if we don't have enough free bytes we need to
780          * convert to node form.
781          */
782         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
783                 /*
784                  * Just checking or no space reservation, give up.
785                  */
786                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
787                                                         args->total == 0) {
788                         xfs_trans_brelse(tp, lbp);
789                         return -ENOSPC;
790                 }
791                 /*
792                  * Convert to node form.
793                  */
794                 error = xfs_dir2_leaf_to_node(args, lbp);
795                 if (error)
796                         return error;
797                 /*
798                  * Then add the new entry.
799                  */
800                 return xfs_dir2_node_addname(args);
801         }
802         /*
803          * Otherwise it will fit without compaction.
804          */
805         else
806                 compact = 0;
807         /*
808          * If just checking, then it will fit unless we needed to allocate
809          * a new data block.
810          */
811         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
812                 xfs_trans_brelse(tp, lbp);
813                 return use_block == -1 ? -ENOSPC : 0;
814         }
815         /*
816          * If no allocations are allowed, return now before we've
817          * changed anything.
818          */
819         if (args->total == 0 && use_block == -1) {
820                 xfs_trans_brelse(tp, lbp);
821                 return -ENOSPC;
822         }
823         /*
824          * Need to compact the leaf entries, removing stale ones.
825          * Leave one stale entry behind - the one closest to our
826          * insertion index - and we'll shift that one to our insertion
827          * point later.
828          */
829         if (compact) {
830                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
831                         &highstale, &lfloglow, &lfloghigh);
832         }
833         /*
834          * There are stale entries, so we'll need log-low and log-high
835          * impossibly bad values later.
836          */
837         else if (leafhdr.stale) {
838                 lfloglow = leafhdr.count;
839                 lfloghigh = -1;
840         }
841         /*
842          * If there was no data block space found, we need to allocate
843          * a new one.
844          */
845         if (use_block == -1) {
846                 /*
847                  * Add the new data block.
848                  */
849                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
850                                 &use_block))) {
851                         xfs_trans_brelse(tp, lbp);
852                         return error;
853                 }
854                 /*
855                  * Initialize the block.
856                  */
857                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
858                         xfs_trans_brelse(tp, lbp);
859                         return error;
860                 }
861                 /*
862                  * If we're adding a new data block on the end we need to
863                  * extend the bests table.  Copy it up one entry.
864                  */
865                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
866                         bestsp--;
867                         memmove(&bestsp[0], &bestsp[1],
868                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
869                         be32_add_cpu(&ltp->bestcount, 1);
870                         xfs_dir3_leaf_log_tail(args, lbp);
871                         xfs_dir3_leaf_log_bests(args, lbp, 0,
872                                                 be32_to_cpu(ltp->bestcount) - 1);
873                 }
874                 /*
875                  * If we're filling in a previously empty block just log it.
876                  */
877                 else
878                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
879                 hdr = dbp->b_addr;
880                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
881                 bestsp[use_block] = bf[0].length;
882                 grown = 1;
883         } else {
884                 /*
885                  * Already had space in some data block.
886                  * Just read that one in.
887                  */
888                 error = xfs_dir3_data_read(tp, dp, args->owner,
889                                 xfs_dir2_db_to_da(args->geo, use_block), 0,
890                                 &dbp);
891                 if (error) {
892                         xfs_trans_brelse(tp, lbp);
893                         return error;
894                 }
895                 hdr = dbp->b_addr;
896                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
897                 grown = 0;
898         }
899         /*
900          * Point to the biggest freespace in our data block.
901          */
902         dup = (xfs_dir2_data_unused_t *)
903               ((char *)hdr + be16_to_cpu(bf[0].offset));
904         needscan = needlog = 0;
905         /*
906          * Mark the initial part of our freespace in use for the new entry.
907          */
908         error = xfs_dir2_data_use_free(args, dbp, dup,
909                         (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
910                         length, &needlog, &needscan);
911         if (error) {
912                 xfs_trans_brelse(tp, lbp);
913                 return error;
914         }
915         /*
916          * Initialize our new entry (at last).
917          */
918         dep = (xfs_dir2_data_entry_t *)dup;
919         dep->inumber = cpu_to_be64(args->inumber);
920         dep->namelen = args->namelen;
921         memcpy(dep->name, args->name, dep->namelen);
922         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
923         tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
924         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
925         /*
926          * Need to scan fix up the bestfree table.
927          */
928         if (needscan)
929                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
930         /*
931          * Need to log the data block's header.
932          */
933         if (needlog)
934                 xfs_dir2_data_log_header(args, dbp);
935         xfs_dir2_data_log_entry(args, dbp, dep);
936         /*
937          * If the bests table needs to be changed, do it.
938          * Log the change unless we've already done that.
939          */
940         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
941                 bestsp[use_block] = bf[0].length;
942                 if (!grown)
943                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
944         }
945 
946         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
947                                        highstale, &lfloglow, &lfloghigh);
948 
949         /*
950          * Fill in the new leaf entry.
951          */
952         lep->hashval = cpu_to_be32(args->hashval);
953         lep->address = cpu_to_be32(
954                                 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
955                                 be16_to_cpu(*tagp)));
956         /*
957          * Log the leaf fields and give up the buffers.
958          */
959         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
960         xfs_dir3_leaf_log_header(args, lbp);
961         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, lfloglow, lfloghigh);
962         xfs_dir3_leaf_check(dp, lbp);
963         xfs_dir3_data_check(dp, dbp);
964         return 0;
965 }
966 
967 /*
968  * Compact out any stale entries in the leaf.
969  * Log the header and changed leaf entries, if any.
970  */
971 void
972 xfs_dir3_leaf_compact(
973         xfs_da_args_t   *args,          /* operation arguments */
974         struct xfs_dir3_icleaf_hdr *leafhdr,
975         struct xfs_buf  *bp)            /* leaf buffer */
976 {
977         int             from;           /* source leaf index */
978         xfs_dir2_leaf_t *leaf;          /* leaf structure */
979         int             loglow;         /* first leaf entry to log */
980         int             to;             /* target leaf index */
981         struct xfs_inode *dp = args->dp;
982 
983         leaf = bp->b_addr;
984         if (!leafhdr->stale)
985                 return;
986 
987         /*
988          * Compress out the stale entries in place.
989          */
990         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
991                 if (leafhdr->ents[from].address ==
992                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
993                         continue;
994                 /*
995                  * Only actually copy the entries that are different.
996                  */
997                 if (from > to) {
998                         if (loglow == -1)
999                                 loglow = to;
1000                         leafhdr->ents[to] = leafhdr->ents[from];
1001                 }
1002                 to++;
1003         }
1004         /*
1005          * Update and log the header, log the leaf entries.
1006          */
1007         ASSERT(leafhdr->stale == from - to);
1008         leafhdr->count -= leafhdr->stale;
1009         leafhdr->stale = 0;
1010 
1011         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, leafhdr);
1012         xfs_dir3_leaf_log_header(args, bp);
1013         if (loglow != -1)
1014                 xfs_dir3_leaf_log_ents(args, leafhdr, bp, loglow, to - 1);
1015 }
1016 
1017 /*
1018  * Compact the leaf entries, removing stale ones.
1019  * Leave one stale entry behind - the one closest to our
1020  * insertion index - and the caller will shift that one to our insertion
1021  * point later.
1022  * Return new insertion index, where the remaining stale entry is,
1023  * and leaf logging indices.
1024  */
1025 void
1026 xfs_dir3_leaf_compact_x1(
1027         struct xfs_dir3_icleaf_hdr *leafhdr,
1028         struct xfs_dir2_leaf_entry *ents,
1029         int             *indexp,        /* insertion index */
1030         int             *lowstalep,     /* out: stale entry before us */
1031         int             *highstalep,    /* out: stale entry after us */
1032         int             *lowlogp,       /* out: low log index */
1033         int             *highlogp)      /* out: high log index */
1034 {
1035         int             from;           /* source copy index */
1036         int             highstale;      /* stale entry at/after index */
1037         int             index;          /* insertion index */
1038         int             keepstale;      /* source index of kept stale */
1039         int             lowstale;       /* stale entry before index */
1040         int             newindex=0;     /* new insertion index */
1041         int             to;             /* destination copy index */
1042 
1043         ASSERT(leafhdr->stale > 1);
1044         index = *indexp;
1045 
1046         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
1047 
1048         /*
1049          * Pick the better of lowstale and highstale.
1050          */
1051         if (lowstale >= 0 &&
1052             (highstale == leafhdr->count ||
1053              index - lowstale <= highstale - index))
1054                 keepstale = lowstale;
1055         else
1056                 keepstale = highstale;
1057         /*
1058          * Copy the entries in place, removing all the stale entries
1059          * except keepstale.
1060          */
1061         for (from = to = 0; from < leafhdr->count; from++) {
1062                 /*
1063                  * Notice the new value of index.
1064                  */
1065                 if (index == from)
1066                         newindex = to;
1067                 if (from != keepstale &&
1068                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1069                         if (from == to)
1070                                 *lowlogp = to;
1071                         continue;
1072                 }
1073                 /*
1074                  * Record the new keepstale value for the insertion.
1075                  */
1076                 if (from == keepstale)
1077                         lowstale = highstale = to;
1078                 /*
1079                  * Copy only the entries that have moved.
1080                  */
1081                 if (from > to)
1082                         ents[to] = ents[from];
1083                 to++;
1084         }
1085         ASSERT(from > to);
1086         /*
1087          * If the insertion point was past the last entry,
1088          * set the new insertion point accordingly.
1089          */
1090         if (index == from)
1091                 newindex = to;
1092         *indexp = newindex;
1093         /*
1094          * Adjust the leaf header values.
1095          */
1096         leafhdr->count -= from - to;
1097         leafhdr->stale = 1;
1098         /*
1099          * Remember the low/high stale value only in the "right"
1100          * direction.
1101          */
1102         if (lowstale >= newindex)
1103                 lowstale = -1;
1104         else
1105                 highstale = leafhdr->count;
1106         *highlogp = leafhdr->count - 1;
1107         *lowstalep = lowstale;
1108         *highstalep = highstale;
1109 }
1110 
1111 /*
1112  * Log the bests entries indicated from a leaf1 block.
1113  */
1114 static void
1115 xfs_dir3_leaf_log_bests(
1116         struct xfs_da_args      *args,
1117         struct xfs_buf          *bp,            /* leaf buffer */
1118         int                     first,          /* first entry to log */
1119         int                     last)           /* last entry to log */
1120 {
1121         __be16                  *firstb;        /* pointer to first entry */
1122         __be16                  *lastb;         /* pointer to last entry */
1123         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1124         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1125 
1126         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1127                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1128 
1129         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1130         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1131         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1132         xfs_trans_log_buf(args->trans, bp,
1133                 (uint)((char *)firstb - (char *)leaf),
1134                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1135 }
1136 
1137 /*
1138  * Log the leaf entries indicated from a leaf1 or leafn block.
1139  */
1140 void
1141 xfs_dir3_leaf_log_ents(
1142         struct xfs_da_args      *args,
1143         struct xfs_dir3_icleaf_hdr *hdr,
1144         struct xfs_buf          *bp,
1145         int                     first,
1146         int                     last)
1147 {
1148         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1149         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1150         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1151 
1152         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1153                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1154                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1155                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1156 
1157         firstlep = &hdr->ents[first];
1158         lastlep = &hdr->ents[last];
1159         xfs_trans_log_buf(args->trans, bp,
1160                 (uint)((char *)firstlep - (char *)leaf),
1161                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1162 }
1163 
1164 /*
1165  * Log the header of the leaf1 or leafn block.
1166  */
1167 void
1168 xfs_dir3_leaf_log_header(
1169         struct xfs_da_args      *args,
1170         struct xfs_buf          *bp)
1171 {
1172         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1173 
1174         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1175                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1176                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1177                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1178 
1179         xfs_trans_log_buf(args->trans, bp,
1180                           (uint)((char *)&leaf->hdr - (char *)leaf),
1181                           args->geo->leaf_hdr_size - 1);
1182 }
1183 
1184 /*
1185  * Log the tail of the leaf1 block.
1186  */
1187 STATIC void
1188 xfs_dir3_leaf_log_tail(
1189         struct xfs_da_args      *args,
1190         struct xfs_buf          *bp)
1191 {
1192         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1193         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1194 
1195         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1196                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1197                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1198                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1199 
1200         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1201         xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1202                 (uint)(args->geo->blksize - 1));
1203 }
1204 
1205 /*
1206  * Look up the entry referred to by args in the leaf format directory.
1207  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1208  * is also used by the node-format code.
1209  */
1210 int
1211 xfs_dir2_leaf_lookup(
1212         xfs_da_args_t           *args)          /* operation arguments */
1213 {
1214         struct xfs_buf          *dbp;           /* data block buffer */
1215         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1216         xfs_inode_t             *dp;            /* incore directory inode */
1217         int                     error;          /* error return code */
1218         int                     index;          /* found entry index */
1219         struct xfs_buf          *lbp;           /* leaf buffer */
1220         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1221         xfs_trans_t             *tp;            /* transaction pointer */
1222         struct xfs_dir3_icleaf_hdr leafhdr;
1223 
1224         trace_xfs_dir2_leaf_lookup(args);
1225 
1226         /*
1227          * Look up name in the leaf block, returning both buffers and index.
1228          */
1229         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1230         if (error)
1231                 return error;
1232 
1233         tp = args->trans;
1234         dp = args->dp;
1235         xfs_dir3_leaf_check(dp, lbp);
1236 
1237         /*
1238          * Get to the leaf entry and contained data entry address.
1239          */
1240         lep = &leafhdr.ents[index];
1241 
1242         /*
1243          * Point to the data entry.
1244          */
1245         dep = (xfs_dir2_data_entry_t *)
1246               ((char *)dbp->b_addr +
1247                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1248         /*
1249          * Return the found inode number & CI name if appropriate
1250          */
1251         args->inumber = be64_to_cpu(dep->inumber);
1252         args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
1253         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1254         xfs_trans_brelse(tp, dbp);
1255         xfs_trans_brelse(tp, lbp);
1256         return error;
1257 }
1258 
1259 /*
1260  * Look up name/hash in the leaf block.
1261  * Fill in indexp with the found index, and dbpp with the data buffer.
1262  * If not found dbpp will be NULL, and ENOENT comes back.
1263  * lbpp will always be filled in with the leaf buffer unless there's an error.
1264  */
1265 static int                                      /* error */
1266 xfs_dir2_leaf_lookup_int(
1267         xfs_da_args_t           *args,          /* operation arguments */
1268         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1269         int                     *indexp,        /* out: index in leaf block */
1270         struct xfs_buf          **dbpp,         /* out: data buffer */
1271         struct xfs_dir3_icleaf_hdr *leafhdr)
1272 {
1273         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1274         struct xfs_buf          *dbp = NULL;    /* data buffer */
1275         xfs_dir2_data_entry_t   *dep;           /* data entry */
1276         xfs_inode_t             *dp;            /* incore directory inode */
1277         int                     error;          /* error return code */
1278         int                     index;          /* index in leaf block */
1279         struct xfs_buf          *lbp;           /* leaf buffer */
1280         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1281         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1282         xfs_mount_t             *mp;            /* filesystem mount point */
1283         xfs_dir2_db_t           newdb;          /* new data block number */
1284         xfs_trans_t             *tp;            /* transaction pointer */
1285         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1286         enum xfs_dacmp          cmp;            /* name compare result */
1287 
1288         dp = args->dp;
1289         tp = args->trans;
1290         mp = dp->i_mount;
1291 
1292         error = xfs_dir3_leaf_read(tp, dp, args->owner, args->geo->leafblk,
1293                         &lbp);
1294         if (error)
1295                 return error;
1296 
1297         *lbpp = lbp;
1298         leaf = lbp->b_addr;
1299         xfs_dir3_leaf_check(dp, lbp);
1300         xfs_dir2_leaf_hdr_from_disk(mp, leafhdr, leaf);
1301 
1302         /*
1303          * Look for the first leaf entry with our hash value.
1304          */
1305         index = xfs_dir2_leaf_search_hash(args, lbp);
1306         /*
1307          * Loop over all the entries with the right hash value
1308          * looking to match the name.
1309          */
1310         for (lep = &leafhdr->ents[index];
1311              index < leafhdr->count &&
1312                         be32_to_cpu(lep->hashval) == args->hashval;
1313              lep++, index++) {
1314                 /*
1315                  * Skip over stale leaf entries.
1316                  */
1317                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1318                         continue;
1319                 /*
1320                  * Get the new data block number.
1321                  */
1322                 newdb = xfs_dir2_dataptr_to_db(args->geo,
1323                                                be32_to_cpu(lep->address));
1324                 /*
1325                  * If it's not the same as the old data block number,
1326                  * need to pitch the old one and read the new one.
1327                  */
1328                 if (newdb != curdb) {
1329                         if (dbp)
1330                                 xfs_trans_brelse(tp, dbp);
1331                         error = xfs_dir3_data_read(tp, dp, args->owner,
1332                                         xfs_dir2_db_to_da(args->geo, newdb), 0,
1333                                         &dbp);
1334                         if (error) {
1335                                 xfs_trans_brelse(tp, lbp);
1336                                 return error;
1337                         }
1338                         curdb = newdb;
1339                 }
1340                 /*
1341                  * Point to the data entry.
1342                  */
1343                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1344                         xfs_dir2_dataptr_to_off(args->geo,
1345                                                 be32_to_cpu(lep->address)));
1346                 /*
1347                  * Compare name and if it's an exact match, return the index
1348                  * and buffer. If it's the first case-insensitive match, store
1349                  * the index and buffer and continue looking for an exact match.
1350                  */
1351                 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
1352                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1353                         args->cmpresult = cmp;
1354                         *indexp = index;
1355                         /* case exact match: return the current buffer. */
1356                         if (cmp == XFS_CMP_EXACT) {
1357                                 *dbpp = dbp;
1358                                 return 0;
1359                         }
1360                         cidb = curdb;
1361                 }
1362         }
1363         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1364         /*
1365          * Here, we can only be doing a lookup (not a rename or remove).
1366          * If a case-insensitive match was found earlier, re-read the
1367          * appropriate data block if required and return it.
1368          */
1369         if (args->cmpresult == XFS_CMP_CASE) {
1370                 ASSERT(cidb != -1);
1371                 if (cidb != curdb) {
1372                         xfs_trans_brelse(tp, dbp);
1373                         error = xfs_dir3_data_read(tp, dp, args->owner,
1374                                         xfs_dir2_db_to_da(args->geo, cidb), 0,
1375                                         &dbp);
1376                         if (error) {
1377                                 xfs_trans_brelse(tp, lbp);
1378                                 return error;
1379                         }
1380                 }
1381                 *dbpp = dbp;
1382                 return 0;
1383         }
1384         /*
1385          * No match found, return -ENOENT.
1386          */
1387         ASSERT(cidb == -1);
1388         if (dbp)
1389                 xfs_trans_brelse(tp, dbp);
1390         xfs_trans_brelse(tp, lbp);
1391         return -ENOENT;
1392 }
1393 
1394 /*
1395  * Remove an entry from a leaf format directory.
1396  */
1397 int                                             /* error */
1398 xfs_dir2_leaf_removename(
1399         xfs_da_args_t           *args)          /* operation arguments */
1400 {
1401         struct xfs_da_geometry  *geo = args->geo;
1402         __be16                  *bestsp;        /* leaf block best freespace */
1403         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1404         xfs_dir2_db_t           db;             /* data block number */
1405         struct xfs_buf          *dbp;           /* data block buffer */
1406         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1407         xfs_inode_t             *dp;            /* incore directory inode */
1408         int                     error;          /* error return code */
1409         xfs_dir2_db_t           i;              /* temporary data block # */
1410         int                     index;          /* index into leaf entries */
1411         struct xfs_buf          *lbp;           /* leaf buffer */
1412         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1413         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1414         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1415         int                     needlog;        /* need to log data header */
1416         int                     needscan;       /* need to rescan data frees */
1417         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1418         struct xfs_dir2_data_free *bf;          /* bestfree table */
1419         struct xfs_dir3_icleaf_hdr leafhdr;
1420 
1421         trace_xfs_dir2_leaf_removename(args);
1422 
1423         /*
1424          * Lookup the leaf entry, get the leaf and data blocks read in.
1425          */
1426         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1427         if (error)
1428                 return error;
1429 
1430         dp = args->dp;
1431         leaf = lbp->b_addr;
1432         hdr = dbp->b_addr;
1433         xfs_dir3_data_check(dp, dbp);
1434         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1435 
1436         /*
1437          * Point to the leaf entry, use that to point to the data entry.
1438          */
1439         lep = &leafhdr.ents[index];
1440         db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1441         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1442                 xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address)));
1443         needscan = needlog = 0;
1444         oldbest = be16_to_cpu(bf[0].length);
1445         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1446         bestsp = xfs_dir2_leaf_bests_p(ltp);
1447         if (be16_to_cpu(bestsp[db]) != oldbest) {
1448                 xfs_buf_mark_corrupt(lbp);
1449                 xfs_da_mark_sick(args);
1450                 return -EFSCORRUPTED;
1451         }
1452 
1453         /*
1454          * Mark the former data entry unused.
1455          */
1456         xfs_dir2_data_make_free(args, dbp,
1457                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1458                 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1459                 &needscan);
1460         /*
1461          * We just mark the leaf entry stale by putting a null in it.
1462          */
1463         leafhdr.stale++;
1464         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1465         xfs_dir3_leaf_log_header(args, lbp);
1466 
1467         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1468         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, index, index);
1469 
1470         /*
1471          * Scan the freespace in the data block again if necessary,
1472          * log the data block header if necessary.
1473          */
1474         if (needscan)
1475                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1476         if (needlog)
1477                 xfs_dir2_data_log_header(args, dbp);
1478         /*
1479          * If the longest freespace in the data block has changed,
1480          * put the new value in the bests table and log that.
1481          */
1482         if (be16_to_cpu(bf[0].length) != oldbest) {
1483                 bestsp[db] = bf[0].length;
1484                 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1485         }
1486         xfs_dir3_data_check(dp, dbp);
1487         /*
1488          * If the data block is now empty then get rid of the data block.
1489          */
1490         if (be16_to_cpu(bf[0].length) ==
1491             geo->blksize - geo->data_entry_offset) {
1492                 ASSERT(db != geo->datablk);
1493                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1494                         /*
1495                          * Nope, can't get rid of it because it caused
1496                          * allocation of a bmap btree block to do so.
1497                          * Just go on, returning success, leaving the
1498                          * empty block in place.
1499                          */
1500                         if (error == -ENOSPC && args->total == 0)
1501                                 error = 0;
1502                         xfs_dir3_leaf_check(dp, lbp);
1503                         return error;
1504                 }
1505                 dbp = NULL;
1506                 /*
1507                  * If this is the last data block then compact the
1508                  * bests table by getting rid of entries.
1509                  */
1510                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1511                         /*
1512                          * Look for the last active entry (i).
1513                          */
1514                         for (i = db - 1; i > 0; i--) {
1515                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1516                                         break;
1517                         }
1518                         /*
1519                          * Copy the table down so inactive entries at the
1520                          * end are removed.
1521                          */
1522                         memmove(&bestsp[db - i], bestsp,
1523                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1524                         be32_add_cpu(&ltp->bestcount, -(db - i));
1525                         xfs_dir3_leaf_log_tail(args, lbp);
1526                         xfs_dir3_leaf_log_bests(args, lbp, 0,
1527                                                 be32_to_cpu(ltp->bestcount) - 1);
1528                 } else
1529                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1530         }
1531         /*
1532          * If the data block was not the first one, drop it.
1533          */
1534         else if (db != geo->datablk)
1535                 dbp = NULL;
1536 
1537         xfs_dir3_leaf_check(dp, lbp);
1538         /*
1539          * See if we can convert to block form.
1540          */
1541         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1542 }
1543 
1544 /*
1545  * Replace the inode number in a leaf format directory entry.
1546  */
1547 int                                             /* error */
1548 xfs_dir2_leaf_replace(
1549         xfs_da_args_t           *args)          /* operation arguments */
1550 {
1551         struct xfs_buf          *dbp;           /* data block buffer */
1552         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1553         xfs_inode_t             *dp;            /* incore directory inode */
1554         int                     error;          /* error return code */
1555         int                     index;          /* index of leaf entry */
1556         struct xfs_buf          *lbp;           /* leaf buffer */
1557         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1558         xfs_trans_t             *tp;            /* transaction pointer */
1559         struct xfs_dir3_icleaf_hdr leafhdr;
1560 
1561         trace_xfs_dir2_leaf_replace(args);
1562 
1563         /*
1564          * Look up the entry.
1565          */
1566         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1567         if (error)
1568                 return error;
1569 
1570         dp = args->dp;
1571         /*
1572          * Point to the leaf entry, get data address from it.
1573          */
1574         lep = &leafhdr.ents[index];
1575         /*
1576          * Point to the data entry.
1577          */
1578         dep = (xfs_dir2_data_entry_t *)
1579               ((char *)dbp->b_addr +
1580                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1581         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1582         /*
1583          * Put the new inode number in, log it.
1584          */
1585         dep->inumber = cpu_to_be64(args->inumber);
1586         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1587         tp = args->trans;
1588         xfs_dir2_data_log_entry(args, dbp, dep);
1589         xfs_dir3_leaf_check(dp, lbp);
1590         xfs_trans_brelse(tp, lbp);
1591         return 0;
1592 }
1593 
1594 /*
1595  * Return index in the leaf block (lbp) which is either the first
1596  * one with this hash value, or if there are none, the insert point
1597  * for that hash value.
1598  */
1599 int                                             /* index value */
1600 xfs_dir2_leaf_search_hash(
1601         xfs_da_args_t           *args,          /* operation arguments */
1602         struct xfs_buf          *lbp)           /* leaf buffer */
1603 {
1604         xfs_dahash_t            hash=0;         /* hash from this entry */
1605         xfs_dahash_t            hashwant;       /* hash value looking for */
1606         int                     high;           /* high leaf index */
1607         int                     low;            /* low leaf index */
1608         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1609         int                     mid=0;          /* current leaf index */
1610         struct xfs_dir3_icleaf_hdr leafhdr;
1611 
1612         xfs_dir2_leaf_hdr_from_disk(args->dp->i_mount, &leafhdr, lbp->b_addr);
1613 
1614         /*
1615          * Note, the table cannot be empty, so we have to go through the loop.
1616          * Binary search the leaf entries looking for our hash value.
1617          */
1618         for (lep = leafhdr.ents, low = 0, high = leafhdr.count - 1,
1619                 hashwant = args->hashval;
1620              low <= high; ) {
1621                 mid = (low + high) >> 1;
1622                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1623                         break;
1624                 if (hash < hashwant)
1625                         low = mid + 1;
1626                 else
1627                         high = mid - 1;
1628         }
1629         /*
1630          * Found one, back up through all the equal hash values.
1631          */
1632         if (hash == hashwant) {
1633                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1634                         mid--;
1635                 }
1636         }
1637         /*
1638          * Need to point to an entry higher than ours.
1639          */
1640         else if (hash < hashwant)
1641                 mid++;
1642         return mid;
1643 }
1644 
1645 /*
1646  * Trim off a trailing data block.  We know it's empty since the leaf
1647  * freespace table says so.
1648  */
1649 int                                             /* error */
1650 xfs_dir2_leaf_trim_data(
1651         xfs_da_args_t           *args,          /* operation arguments */
1652         struct xfs_buf          *lbp,           /* leaf buffer */
1653         xfs_dir2_db_t           db)             /* data block number */
1654 {
1655         struct xfs_da_geometry  *geo = args->geo;
1656         __be16                  *bestsp;        /* leaf bests table */
1657         struct xfs_buf          *dbp;           /* data block buffer */
1658         xfs_inode_t             *dp;            /* incore directory inode */
1659         int                     error;          /* error return value */
1660         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1661         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1662         xfs_trans_t             *tp;            /* transaction pointer */
1663 
1664         dp = args->dp;
1665         tp = args->trans;
1666         /*
1667          * Read the offending data block.  We need its buffer.
1668          */
1669         error = xfs_dir3_data_read(tp, dp, args->owner,
1670                         xfs_dir2_db_to_da(geo, db), 0, &dbp);
1671         if (error)
1672                 return error;
1673 
1674         leaf = lbp->b_addr;
1675         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1676 
1677 #ifdef DEBUG
1678 {
1679         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1680         struct xfs_dir2_data_free *bf =
1681                 xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1682 
1683         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1684                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1685         ASSERT(be16_to_cpu(bf[0].length) ==
1686                geo->blksize - geo->data_entry_offset);
1687         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1688 }
1689 #endif
1690 
1691         /*
1692          * Get rid of the data block.
1693          */
1694         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1695                 ASSERT(error != -ENOSPC);
1696                 xfs_trans_brelse(tp, dbp);
1697                 return error;
1698         }
1699         /*
1700          * Eliminate the last bests entry from the table.
1701          */
1702         bestsp = xfs_dir2_leaf_bests_p(ltp);
1703         be32_add_cpu(&ltp->bestcount, -1);
1704         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1705         xfs_dir3_leaf_log_tail(args, lbp);
1706         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1707         return 0;
1708 }
1709 
1710 static inline size_t
1711 xfs_dir3_leaf_size(
1712         struct xfs_dir3_icleaf_hdr      *hdr,
1713         int                             counts)
1714 {
1715         int     entries;
1716         int     hdrsize;
1717 
1718         entries = hdr->count - hdr->stale;
1719         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1720             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1721                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1722         else
1723                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1724 
1725         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1726                        + counts * sizeof(xfs_dir2_data_off_t)
1727                        + sizeof(xfs_dir2_leaf_tail_t);
1728 }
1729 
1730 /*
1731  * Convert node form directory to leaf form directory.
1732  * The root of the node form dir needs to already be a LEAFN block.
1733  * Just return if we can't do anything.
1734  */
1735 int                                             /* error */
1736 xfs_dir2_node_to_leaf(
1737         xfs_da_state_t          *state)         /* directory operation state */
1738 {
1739         xfs_da_args_t           *args;          /* operation arguments */
1740         xfs_inode_t             *dp;            /* incore directory inode */
1741         int                     error;          /* error return code */
1742         struct xfs_buf          *fbp;           /* buffer for freespace block */
1743         xfs_fileoff_t           fo;             /* freespace file offset */
1744         struct xfs_buf          *lbp;           /* buffer for leaf block */
1745         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1746         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1747         xfs_mount_t             *mp;            /* filesystem mount point */
1748         int                     rval;           /* successful free trim? */
1749         xfs_trans_t             *tp;            /* transaction pointer */
1750         struct xfs_dir3_icleaf_hdr leafhdr;
1751         struct xfs_dir3_icfree_hdr freehdr;
1752 
1753         /*
1754          * There's more than a leaf level in the btree, so there must
1755          * be multiple leafn blocks.  Give up.
1756          */
1757         if (state->path.active > 1)
1758                 return 0;
1759         args = state->args;
1760 
1761         trace_xfs_dir2_node_to_leaf(args);
1762 
1763         mp = state->mp;
1764         dp = args->dp;
1765         tp = args->trans;
1766         /*
1767          * Get the last offset in the file.
1768          */
1769         if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1770                 return error;
1771         }
1772         fo -= args->geo->fsbcount;
1773         /*
1774          * If there are freespace blocks other than the first one,
1775          * take this opportunity to remove trailing empty freespace blocks
1776          * that may have been left behind during no-space-reservation
1777          * operations.
1778          */
1779         while (fo > args->geo->freeblk) {
1780                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1781                         return error;
1782                 }
1783                 if (rval)
1784                         fo -= args->geo->fsbcount;
1785                 else
1786                         return 0;
1787         }
1788         /*
1789          * Now find the block just before the freespace block.
1790          */
1791         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1792                 return error;
1793         }
1794         /*
1795          * If it's not the single leaf block, give up.
1796          */
1797         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1798                 return 0;
1799         lbp = state->path.blk[0].bp;
1800         leaf = lbp->b_addr;
1801         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
1802 
1803         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1804                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1805 
1806         /*
1807          * Read the freespace block.
1808          */
1809         error = xfs_dir2_free_read(tp, dp, args->owner, args->geo->freeblk,
1810                         &fbp);
1811         if (error)
1812                 return error;
1813         xfs_dir2_free_hdr_from_disk(mp, &freehdr, fbp->b_addr);
1814 
1815         ASSERT(!freehdr.firstdb);
1816 
1817         /*
1818          * Now see if the leafn and free data will fit in a leaf1.
1819          * If not, release the buffer and give up.
1820          */
1821         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1822                 xfs_trans_brelse(tp, fbp);
1823                 return 0;
1824         }
1825 
1826         /*
1827          * If the leaf has any stale entries in it, compress them out.
1828          */
1829         if (leafhdr.stale)
1830                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1831 
1832         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1833         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1834         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1835                                         ? XFS_DIR2_LEAF1_MAGIC
1836                                         : XFS_DIR3_LEAF1_MAGIC;
1837 
1838         /*
1839          * Set up the leaf tail from the freespace block.
1840          */
1841         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1842         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1843 
1844         /*
1845          * Set up the leaf bests table.
1846          */
1847         memcpy(xfs_dir2_leaf_bests_p(ltp), freehdr.bests,
1848                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1849 
1850         xfs_dir2_leaf_hdr_to_disk(mp, leaf, &leafhdr);
1851         xfs_dir3_leaf_log_header(args, lbp);
1852         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1853         xfs_dir3_leaf_log_tail(args, lbp);
1854         xfs_dir3_leaf_check(dp, lbp);
1855 
1856         /*
1857          * Get rid of the freespace block.
1858          */
1859         error = xfs_dir2_shrink_inode(args,
1860                         xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1861                         fbp);
1862         if (error) {
1863                 /*
1864                  * This can't fail here because it can only happen when
1865                  * punching out the middle of an extent, and this is an
1866                  * isolated block.
1867                  */
1868                 ASSERT(error != -ENOSPC);
1869                 return error;
1870         }
1871         fbp = NULL;
1872         /*
1873          * Now see if we can convert the single-leaf directory
1874          * down to a block form directory.
1875          * This routine always kills the dabuf for the leaf, so
1876          * eliminate it from the path.
1877          */
1878         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1879         state->path.blk[0].bp = NULL;
1880         return error;
1881 }
1882 

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

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

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

sflogo.php