fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_types.h"
  22 #include "xfs_bit.h"
  23 #include "xfs_log.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_bmap.h"
  33 #include "xfs_dir2_format.h"
  34 #include "xfs_dir2_priv.h"
  35 #include "xfs_error.h"
  36 #include "xfs_trace.h"
  37 #include "xfs_buf_item.h"
  38 #include "xfs_cksum.h"
  39
  40 /*
  41  * Local function declarations.
  42  */
  43 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  44                                     int *indexp, struct xfs_buf **dbpp);
  45 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
  46                                     int first, int last);
  47 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
  48
  49 /*
  50  * Check the internal consistency of a leaf1 block.
  51  * Pop an assert if something is wrong.
  52  */
  53 #ifdef DEBUG
  54 #define xfs_dir3_leaf_check(mp, bp) \
  55 do { \
  56         if (!xfs_dir3_leaf1_check((mp), (bp))) \
  57                 ASSERT(0); \
  58 } while (0);
  59
  60 STATIC bool
  61 xfs_dir3_leaf1_check(
  62         struct xfs_mount        *mp,
  63         struct xfs_buf          *bp)
  64 {
  65         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  66         struct xfs_dir3_icleaf_hdr leafhdr;
  67
  68         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
  69
  70         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
  71                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  72                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  73                         return false;
  74         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
  75                 return false;
  76
  77         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
  78 }
  79 #else
  80 #define xfs_dir3_leaf_check(mp, bp)
  81 #endif
  82
  83 void
  84 xfs_dir3_leaf_hdr_from_disk(
  85         struct xfs_dir3_icleaf_hdr      *to,
  86         struct xfs_dir2_leaf            *from)
  87 {
  88         if (from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
  89             from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
  90                 to->forw = be32_to_cpu(from->hdr.info.forw);
  91                 to->back = be32_to_cpu(from->hdr.info.back);
  92                 to->magic = be16_to_cpu(from->hdr.info.magic);
  93                 to->count = be16_to_cpu(from->hdr.count);
  94                 to->stale = be16_to_cpu(from->hdr.stale);
  95         } else {
  96                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)from;
  97
  98                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
  99                 to->back = be32_to_cpu(hdr3->info.hdr.back);
 100                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
 101                 to->count = be16_to_cpu(hdr3->count);
 102                 to->stale = be16_to_cpu(hdr3->stale);
 103         }
 104
 105         ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
 106                to->magic == XFS_DIR3_LEAF1_MAGIC ||
 107                to->magic == XFS_DIR2_LEAFN_MAGIC ||
 108                to->magic == XFS_DIR3_LEAFN_MAGIC);
 109 }
 110
 111 void
 112 xfs_dir3_leaf_hdr_to_disk(
 113         struct xfs_dir2_leaf            *to,
 114         struct xfs_dir3_icleaf_hdr      *from)
 115 {
 116         ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
 117                from->magic == XFS_DIR3_LEAF1_MAGIC ||
 118                from->magic == XFS_DIR2_LEAFN_MAGIC ||
 119                from->magic == XFS_DIR3_LEAFN_MAGIC);
 120
 121         if (from->magic == XFS_DIR2_LEAF1_MAGIC ||
 122             from->magic == XFS_DIR2_LEAFN_MAGIC) {
 123                 to->hdr.info.forw = cpu_to_be32(from->forw);
 124                 to->hdr.info.back = cpu_to_be32(from->back);
 125                 to->hdr.info.magic = cpu_to_be16(from->magic);
 126                 to->hdr.count = cpu_to_be16(from->count);
 127                 to->hdr.stale = cpu_to_be16(from->stale);
 128         } else {
 129                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)to;
 130
 131                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
 132                 hdr3->info.hdr.back = cpu_to_be32(from->back);
 133                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
 134                 hdr3->count = cpu_to_be16(from->count);
 135                 hdr3->stale = cpu_to_be16(from->stale);
 136         }
 137 }
 138
 139 bool
 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 {
 145         struct xfs_dir2_leaf_entry *ents;
 146         xfs_dir2_leaf_tail_t    *ltp;
 147         int                     stale;
 148         int                     i;
 149
 150         ents = xfs_dir3_leaf_ents_p(leaf);
 151         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 152
 153         /*
 154          * XXX (dgc): This value is not restrictive enough.
 155          * Should factor in the size of the bests table as well.
 156          * We can deduce a value for that from di_size.
 157          */
 158         if (hdr->count > xfs_dir3_max_leaf_ents(mp, leaf))
 159                 return false;
 160
 161         /* Leaves and bests don't overlap in leaf format. */
 162         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 163              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 164             (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 165                 return false;
 166
 167         /* Check hash value order, count stale entries.  */
 168         for (i = stale = 0; i < hdr->count; i++) {
 169                 if (i + 1 < hdr->count) {
 170                         if (be32_to_cpu(ents[i].hashval) >
 171                                         be32_to_cpu(ents[i + 1].hashval))
 172                                 return false;
 173                 }
 174                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 175                         stale++;
 176         }
 177         if (hdr->stale != stale)
 178                 return false;
 179         return true;
 180 }
 181
 182 static bool
 183 xfs_dir3_leaf_verify(
 184         struct xfs_buf          *bp,
 185         __uint16_t              magic)
 186 {
 187         struct xfs_mount        *mp = bp->b_target->bt_mount;
 188         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 189         struct xfs_dir3_icleaf_hdr leafhdr;
 190
 191         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 192
 193         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 194         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 195                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 196
 197                 if ((magic == XFS_DIR2_LEAF1_MAGIC &&
 198                      leafhdr.magic != XFS_DIR3_LEAF1_MAGIC) ||
 199                     (magic == XFS_DIR2_LEAFN_MAGIC &&
 200                      leafhdr.magic != XFS_DIR3_LEAFN_MAGIC))
 201                         return false;
 202
 203                 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
 204                         return false;
 205                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 206                         return false;
 207         } else {
 208                 if (leafhdr.magic != magic)
 209                         return false;
 210         }
 211         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
 212 }
 213
 214 static void
 215 __read_verify(
 216         struct xfs_buf  *bp,
 217         __uint16_t      magic)
 218 {
 219         struct xfs_mount        *mp = bp->b_target->bt_mount;
 220
 221         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 222              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 223                                           XFS_DIR3_LEAF_CRC_OFF)) ||
 224             !xfs_dir3_leaf_verify(bp, magic)) {
 225                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 226                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 227         }
 228 }
 229
 230 static void
 231 __write_verify(
 232         struct xfs_buf  *bp,
 233         __uint16_t      magic)
 234 {
 235         struct xfs_mount        *mp = bp->b_target->bt_mount;
 236         struct xfs_buf_log_item *bip = bp->b_fspriv;
 237         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 238
 239         if (!xfs_dir3_leaf_verify(bp, magic)) {
 240                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 241                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 242                 return;
 243         }
 244
 245         if (!xfs_sb_version_hascrc(&mp->m_sb))
 246                 return;
 247
 248         if (bip)
 249                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 250
 251         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_LEAF_CRC_OFF);
 252 }
 253
 254 static void
 255 xfs_dir3_leaf1_read_verify(
 256         struct xfs_buf  *bp)
 257 {
 258         __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 259 }
 260
 261 static void
 262 xfs_dir3_leaf1_write_verify(
 263         struct xfs_buf  *bp)
 264 {
 265         __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 266 }
 267
 268 static void
 269 xfs_dir3_leafn_read_verify(
 270         struct xfs_buf  *bp)
 271 {
 272         __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 273 }
 274
 275 static void
 276 xfs_dir3_leafn_write_verify(
 277         struct xfs_buf  *bp)
 278 {
 279         __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 280 }
 281
 282 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 283         .verify_read = xfs_dir3_leaf1_read_verify,
 284         .verify_write = xfs_dir3_leaf1_write_verify,
 285 };
 286
 287 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 288         .verify_read = xfs_dir3_leafn_read_verify,
 289         .verify_write = xfs_dir3_leafn_write_verify,
 290 };
 291
 292 static int
 293 xfs_dir3_leaf_read(
 294         struct xfs_trans        *tp,
 295         struct xfs_inode        *dp,
 296         xfs_dablk_t             fbno,
 297         xfs_daddr_t             mappedbno,
 298         struct xfs_buf          **bpp)
 299 {
 300         int                     err;
 301
 302         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 303                                 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
 304         if (!err && tp)
 305                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 306         return err;
 307 }
 308
 309 int
 310 xfs_dir3_leafn_read(
 311         struct xfs_trans        *tp,
 312         struct xfs_inode        *dp,
 313         xfs_dablk_t             fbno,
 314         xfs_daddr_t             mappedbno,
 315         struct xfs_buf          **bpp)
 316 {
 317         int                     err;
 318
 319         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 320                                 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
 321         if (!err && tp)
 322                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 323         return err;
 324 }
 325
 326 /*
 327  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 328  */
 329 static void
 330 xfs_dir3_leaf_init(
 331         struct xfs_mount        *mp,
 332         struct xfs_trans        *tp,
 333         struct xfs_buf          *bp,
 334         xfs_ino_t               owner,
 335         __uint16_t              type)
 336 {
 337         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 338
 339         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 340
 341         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 342                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 343
 344                 memset(leaf3, 0, sizeof(*leaf3));
 345
 346                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 347                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 348                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 349                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 350                 leaf3->info.owner = cpu_to_be64(owner);
 351                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
 352         } else {
 353                 memset(leaf, 0, sizeof(*leaf));
 354                 leaf->hdr.info.magic = cpu_to_be16(type);
 355         }
 356
 357         /*
 358          * If it's a leaf-format directory initialize the tail.
 359          * Caller is responsible for initialising the bests table.
 360          */
 361         if (type == XFS_DIR2_LEAF1_MAGIC) {
 362                 struct xfs_dir2_leaf_tail *ltp;
 363
 364                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 365                 ltp->bestcount = 0;
 366                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 367                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 368         } else {
 369                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 370                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 371         }
 372 }
 373
 374 int
 375 xfs_dir3_leaf_get_buf(
 376         xfs_da_args_t           *args,
 377         xfs_dir2_db_t           bno,
 378         struct xfs_buf          **bpp,
 379         __uint16_t              magic)
 380 {
 381         struct xfs_inode        *dp = args->dp;
 382         struct xfs_trans        *tp = args->trans;
 383         struct xfs_mount        *mp = dp->i_mount;
 384         struct xfs_buf          *bp;
 385         int                     error;
 386
 387         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 388         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
 389                bno < XFS_DIR2_FREE_FIRSTDB(mp));
 390
 391         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
 392                                XFS_DATA_FORK);
 393         if (error)
 394                 return error;
 395
 396         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 397         xfs_dir3_leaf_log_header(tp, bp);
 398         if (magic == XFS_DIR2_LEAF1_MAGIC)
 399                 xfs_dir3_leaf_log_tail(tp, bp);
 400         *bpp = bp;
 401         return 0;
 402 }
 403
 404 /*
 405  * Convert a block form directory to a leaf form directory.
 406  */
 407 int                                             /* error */
 408 xfs_dir2_block_to_leaf(
 409         xfs_da_args_t           *args,          /* operation arguments */
 410         struct xfs_buf          *dbp)           /* input block's buffer */
 411 {
 412         __be16                  *bestsp;        /* leaf's bestsp entries */
 413         xfs_dablk_t             blkno;          /* leaf block's bno */
 414         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 415         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 416         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 417         xfs_inode_t             *dp;            /* incore directory inode */
 418         int                     error;          /* error return code */
 419         struct xfs_buf          *lbp;           /* leaf block's buffer */
 420         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 421         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 422         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 423         xfs_mount_t             *mp;            /* filesystem mount point */
 424         int                     needlog;        /* need to log block header */
 425         int                     needscan;       /* need to rescan bestfree */
 426         xfs_trans_t             *tp;            /* transaction pointer */
 427         struct xfs_dir2_data_free *bf;
 428         struct xfs_dir2_leaf_entry *ents;
 429         struct xfs_dir3_icleaf_hdr leafhdr;
 430
 431         trace_xfs_dir2_block_to_leaf(args);
 432
 433         dp = args->dp;
 434         mp = dp->i_mount;
 435         tp = args->trans;
 436         /*
 437          * Add the leaf block to the inode.
 438          * This interface will only put blocks in the leaf/node range.
 439          * Since that's empty now, we'll get the root (block 0 in range).
 440          */
 441         if ((error = xfs_da_grow_inode(args, &blkno))) {
 442                 return error;
 443         }
 444         ldb = xfs_dir2_da_to_db(mp, blkno);
 445         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 446         /*
 447          * Initialize the leaf block, get a buffer for it.
 448          */
 449         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 450         if (error)
 451                 return error;
 452
 453         leaf = lbp->b_addr;
 454         hdr = dbp->b_addr;
 455         xfs_dir3_data_check(dp, dbp);
 456         btp = xfs_dir2_block_tail_p(mp, hdr);
 457         blp = xfs_dir2_block_leaf_p(btp);
 458         bf = xfs_dir3_data_bestfree_p(hdr);
 459         ents = xfs_dir3_leaf_ents_p(leaf);
 460
 461         /*
 462          * Set the counts in the leaf header.
 463          */
 464         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 465         leafhdr.count = be32_to_cpu(btp->count);
 466         leafhdr.stale = be32_to_cpu(btp->stale);
 467         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 468         xfs_dir3_leaf_log_header(tp, lbp);
 469
 470         /*
 471          * Could compact these but I think we always do the conversion
 472          * after squeezing out stale entries.
 473          */
 474         memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 475         xfs_dir3_leaf_log_ents(tp, lbp, 0, leafhdr.count - 1);
 476         needscan = 0;
 477         needlog = 1;
 478         /*
 479          * Make the space formerly occupied by the leaf entries and block
 480          * tail be free.
 481          */
 482         xfs_dir2_data_make_free(tp, dbp,
 483                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 484                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 485                                        (char *)blp),
 486                 &needlog, &needscan);
 487         /*
 488          * Fix up the block header, make it a data block.
 489          */
 490         dbp->b_ops = &xfs_dir3_data_buf_ops;
 491         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 492         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 493                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 494         else
 495                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 496
 497         if (needscan)
 498                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 499         /*
 500          * Set up leaf tail and bests table.
 501          */
 502         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 503         ltp->bestcount = cpu_to_be32(1);
 504         bestsp = xfs_dir2_leaf_bests_p(ltp);
 505         bestsp[0] =  bf[0].length;
 506         /*
 507          * Log the data header and leaf bests table.
 508          */
 509         if (needlog)
 510                 xfs_dir2_data_log_header(tp, dbp);
 511         xfs_dir3_leaf_check(mp, lbp);
 512         xfs_dir3_data_check(dp, dbp);
 513         xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
 514         return 0;
 515 }
 516
 517 STATIC void
 518 xfs_dir3_leaf_find_stale(
 519         struct xfs_dir3_icleaf_hdr *leafhdr,
 520         struct xfs_dir2_leaf_entry *ents,
 521         int                     index,
 522         int                     *lowstale,
 523         int                     *highstale)
 524 {
 525         /*
 526          * Find the first stale entry before our index, if any.
 527          */
 528         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 529                 if (ents[*lowstale].address ==
 530                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 531                         break;
 532         }
 533
 534         /*
 535          * Find the first stale entry at or after our index, if any.
 536          * Stop if the result would require moving more entries than using
 537          * lowstale.
 538          */
 539         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 540                 if (ents[*highstale].address ==
 541                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 542                         break;
 543                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 544                         break;
 545         }
 546 }
 547
 548 struct xfs_dir2_leaf_entry *
 549 xfs_dir3_leaf_find_entry(
 550         struct xfs_dir3_icleaf_hdr *leafhdr,
 551         struct xfs_dir2_leaf_entry *ents,
 552         int                     index,          /* leaf table position */
 553         int                     compact,        /* need to compact leaves */
 554         int                     lowstale,       /* index of prev stale leaf */
 555         int                     highstale,      /* index of next stale leaf */
 556         int                     *lfloglow,      /* low leaf logging index */
 557         int                     *lfloghigh)     /* high leaf logging index */
 558 {
 559         if (!leafhdr->stale) {
 560                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 561
 562                 /*
 563                  * Now we need to make room to insert the leaf entry.
 564                  *
 565                  * If there are no stale entries, just insert a hole at index.
 566                  */
 567                 lep = &ents[index];
 568                 if (index < leafhdr->count)
 569                         memmove(lep + 1, lep,
 570                                 (leafhdr->count - index) * sizeof(*lep));
 571
 572                 /*
 573                  * Record low and high logging indices for the leaf.
 574                  */
 575                 *lfloglow = index;
 576                 *lfloghigh = leafhdr->count++;
 577                 return lep;
 578         }
 579
 580         /*
 581          * There are stale entries.
 582          *
 583          * We will use one of them for the new entry.  It's probably not at
 584          * the right location, so we'll have to shift some up or down first.
 585          *
 586          * If we didn't compact before, we need to find the nearest stale
 587          * entries before and after our insertion point.
 588          */
 589         if (compact == 0)
 590                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 591                                          &lowstale, &highstale);
 592
 593         /*
 594          * If the low one is better, use it.
 595          */
 596         if (lowstale >= 0 &&
 597             (highstale == leafhdr->count ||
 598              index - lowstale - 1 < highstale - index)) {
 599                 ASSERT(index - lowstale - 1 >= 0);
 600                 ASSERT(ents[lowstale].address ==
 601                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 602
 603                 /*
 604                  * Copy entries up to cover the stale entry and make room
 605                  * for the new entry.
 606                  */
 607                 if (index - lowstale - 1 > 0) {
 608                         memmove(&ents[lowstale], &ents[lowstale + 1],
 609                                 (index - lowstale - 1) *
 610                                         sizeof(xfs_dir2_leaf_entry_t));
 611                 }
 612                 *lfloglow = MIN(lowstale, *lfloglow);
 613                 *lfloghigh = MAX(index - 1, *lfloghigh);
 614                 leafhdr->stale--;
 615                 return &ents[index - 1];
 616         }
 617
 618         /*
 619          * The high one is better, so use that one.
 620          */
 621         ASSERT(highstale - index >= 0);
 622         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 623
 624         /*
 625          * Copy entries down to cover the stale entry and make room for the
 626          * new entry.
 627          */
 628         if (highstale - index > 0) {
 629                 memmove(&ents[index + 1], &ents[index],
 630                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 631         }
 632         *lfloglow = MIN(index, *lfloglow);
 633         *lfloghigh = MAX(highstale, *lfloghigh);
 634         leafhdr->stale--;
 635         return &ents[index];
 636 }
 637
 638 /*
 639  * Add an entry to a leaf form directory.
 640  */
 641 int                                             /* error */
 642 xfs_dir2_leaf_addname(
 643         xfs_da_args_t           *args)          /* operation arguments */
 644 {
 645         __be16                  *bestsp;        /* freespace table in leaf */
 646         int                     compact;        /* need to compact leaves */
 647         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 648         struct xfs_buf          *dbp;           /* data block buffer */
 649         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 650         xfs_inode_t             *dp;            /* incore directory inode */
 651         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 652         int                     error;          /* error return value */
 653         int                     grown;          /* allocated new data block */
 654         int                     highstale;      /* index of next stale leaf */
 655         int                     i;              /* temporary, index */
 656         int                     index;          /* leaf table position */
 657         struct xfs_buf          *lbp;           /* leaf's buffer */
 658         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 659         int                     length;         /* length of new entry */
 660         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 661         int                     lfloglow;       /* low leaf logging index */
 662         int                     lfloghigh;      /* high leaf logging index */
 663         int                     lowstale;       /* index of prev stale leaf */
 664         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 665         xfs_mount_t             *mp;            /* filesystem mount point */
 666         int                     needbytes;      /* leaf block bytes needed */
 667         int                     needlog;        /* need to log data header */
 668         int                     needscan;       /* need to rescan data free */
 669         __be16                  *tagp;          /* end of data entry */
 670         xfs_trans_t             *tp;            /* transaction pointer */
 671         xfs_dir2_db_t           use_block;      /* data block number */
 672         struct xfs_dir2_data_free *bf;          /* bestfree table */
 673         struct xfs_dir2_leaf_entry *ents;
 674         struct xfs_dir3_icleaf_hdr leafhdr;
 675
 676         trace_xfs_dir2_leaf_addname(args);
 677
 678         dp = args->dp;
 679         tp = args->trans;
 680         mp = dp->i_mount;
 681
 682         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 683         if (error)
 684                 return error;
 685
 686         /*
 687          * Look up the entry by hash value and name.
 688          * We know it's not there, our caller has already done a lookup.
 689          * So the index is of the entry to insert in front of.
 690          * But if there are dup hash values the index is of the first of those.
 691          */
 692         index = xfs_dir2_leaf_search_hash(args, lbp);
 693         leaf = lbp->b_addr;
 694         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 695         ents = xfs_dir3_leaf_ents_p(leaf);
 696         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 697         bestsp = xfs_dir2_leaf_bests_p(ltp);
 698         length = xfs_dir2_data_entsize(args->namelen);
 699
 700         /*
 701          * See if there are any entries with the same hash value
 702          * and space in their block for the new entry.
 703          * This is good because it puts multiple same-hash value entries
 704          * in a data block, improving the lookup of those entries.
 705          */
 706         for (use_block = -1, lep = &ents[index];
 707              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 708              index++, lep++) {
 709                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 710                         continue;
 711                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 712                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 713                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 714                 if (be16_to_cpu(bestsp[i]) >= length) {
 715                         use_block = i;
 716                         break;
 717                 }
 718         }
 719         /*
 720          * Didn't find a block yet, linear search all the data blocks.
 721          */
 722         if (use_block == -1) {
 723                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 724                         /*
 725                          * Remember a block we see that's missing.
 726                          */
 727                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 728                             use_block == -1)
 729                                 use_block = i;
 730                         else if (be16_to_cpu(bestsp[i]) >= length) {
 731                                 use_block = i;
 732                                 break;
 733                         }
 734                 }
 735         }
 736         /*
 737          * How many bytes do we need in the leaf block?
 738          */
 739         needbytes = 0;
 740         if (!leafhdr.stale)
 741                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 742         if (use_block == -1)
 743                 needbytes += sizeof(xfs_dir2_data_off_t);
 744
 745         /*
 746          * Now kill use_block if it refers to a missing block, so we
 747          * can use it as an indication of allocation needed.
 748          */
 749         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 750                 use_block = -1;
 751         /*
 752          * If we don't have enough free bytes but we can make enough
 753          * by compacting out stale entries, we'll do that.
 754          */
 755         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 756             leafhdr.stale > 1)
 757                 compact = 1;
 758
 759         /*
 760          * Otherwise if we don't have enough free bytes we need to
 761          * convert to node form.
 762          */
 763         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 764                 /*
 765                  * Just checking or no space reservation, give up.
 766                  */
 767                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 768                                                         args->total == 0) {
 769                         xfs_trans_brelse(tp, lbp);
 770                         return XFS_ERROR(ENOSPC);
 771                 }
 772                 /*
 773                  * Convert to node form.
 774                  */
 775                 error = xfs_dir2_leaf_to_node(args, lbp);
 776                 if (error)
 777                         return error;
 778                 /*
 779                  * Then add the new entry.
 780                  */
 781                 return xfs_dir2_node_addname(args);
 782         }
 783         /*
 784          * Otherwise it will fit without compaction.
 785          */
 786         else
 787                 compact = 0;
 788         /*
 789          * If just checking, then it will fit unless we needed to allocate
 790          * a new data block.
 791          */
 792         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 793                 xfs_trans_brelse(tp, lbp);
 794                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 795         }
 796         /*
 797          * If no allocations are allowed, return now before we've
 798          * changed anything.
 799          */
 800         if (args->total == 0 && use_block == -1) {
 801                 xfs_trans_brelse(tp, lbp);
 802                 return XFS_ERROR(ENOSPC);
 803         }
 804         /*
 805          * Need to compact the leaf entries, removing stale ones.
 806          * Leave one stale entry behind - the one closest to our
 807          * insertion index - and we'll shift that one to our insertion
 808          * point later.
 809          */
 810         if (compact) {
 811                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 812                         &highstale, &lfloglow, &lfloghigh);
 813         }
 814         /*
 815          * There are stale entries, so we'll need log-low and log-high
 816          * impossibly bad values later.
 817          */
 818         else if (leafhdr.stale) {
 819                 lfloglow = leafhdr.count;
 820                 lfloghigh = -1;
 821         }
 822         /*
 823          * If there was no data block space found, we need to allocate
 824          * a new one.
 825          */
 826         if (use_block == -1) {
 827                 /*
 828                  * Add the new data block.
 829                  */
 830                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 831                                 &use_block))) {
 832                         xfs_trans_brelse(tp, lbp);
 833                         return error;
 834                 }
 835                 /*
 836                  * Initialize the block.
 837                  */
 838                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 839                         xfs_trans_brelse(tp, lbp);
 840                         return error;
 841                 }
 842                 /*
 843                  * If we're adding a new data block on the end we need to
 844                  * extend the bests table.  Copy it up one entry.
 845                  */
 846                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 847                         bestsp--;
 848                         memmove(&bestsp[0], &bestsp[1],
 849                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 850                         be32_add_cpu(&ltp->bestcount, 1);
 851                         xfs_dir3_leaf_log_tail(tp, lbp);
 852                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 853                 }
 854                 /*
 855                  * If we're filling in a previously empty block just log it.
 856                  */
 857                 else
 858                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 859                 hdr = dbp->b_addr;
 860                 bf = xfs_dir3_data_bestfree_p(hdr);
 861                 bestsp[use_block] = bf[0].length;
 862                 grown = 1;
 863         } else {
 864                 /*
 865                  * Already had space in some data block.
 866                  * Just read that one in.
 867                  */
 868                 error = xfs_dir3_data_read(tp, dp,
 869                                            xfs_dir2_db_to_da(mp, use_block),
 870                                            -1, &dbp);
 871                 if (error) {
 872                         xfs_trans_brelse(tp, lbp);
 873                         return error;
 874                 }
 875                 hdr = dbp->b_addr;
 876                 bf = xfs_dir3_data_bestfree_p(hdr);
 877                 grown = 0;
 878         }
 879         /*
 880          * Point to the biggest freespace in our data block.
 881          */
 882         dup = (xfs_dir2_data_unused_t *)
 883               ((char *)hdr + be16_to_cpu(bf[0].offset));
 884         ASSERT(be16_to_cpu(dup->length) >= length);
 885         needscan = needlog = 0;
 886         /*
 887          * Mark the initial part of our freespace in use for the new entry.
 888          */
 889         xfs_dir2_data_use_free(tp, dbp, dup,
 890                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 891                 &needlog, &needscan);
 892         /*
 893          * Initialize our new entry (at last).
 894          */
 895         dep = (xfs_dir2_data_entry_t *)dup;
 896         dep->inumber = cpu_to_be64(args->inumber);
 897         dep->namelen = args->namelen;
 898         memcpy(dep->name, args->name, dep->namelen);
 899         tagp = xfs_dir2_data_entry_tag_p(dep);
 900         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 901         /*
 902          * Need to scan fix up the bestfree table.
 903          */
 904         if (needscan)
 905                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 906         /*
 907          * Need to log the data block's header.
 908          */
 909         if (needlog)
 910                 xfs_dir2_data_log_header(tp, dbp);
 911         xfs_dir2_data_log_entry(tp, dbp, dep);
 912         /*
 913          * If the bests table needs to be changed, do it.
 914          * Log the change unless we've already done that.
 915          */
 916         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 917                 bestsp[use_block] = bf[0].length;
 918                 if (!grown)
 919                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 920         }
 921
 922         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 923                                        highstale, &lfloglow, &lfloghigh);
 924
 925         /*
 926          * Fill in the new leaf entry.
 927          */
 928         lep->hashval = cpu_to_be32(args->hashval);
 929         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 930                                 be16_to_cpu(*tagp)));
 931         /*
 932          * Log the leaf fields and give up the buffers.
 933          */
 934         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 935         xfs_dir3_leaf_log_header(tp, lbp);
 936         xfs_dir3_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 937         xfs_dir3_leaf_check(mp, lbp);
 938         xfs_dir3_data_check(dp, dbp);
 939         return 0;
 940 }
 941
 942 /*
 943  * Compact out any stale entries in the leaf.
 944  * Log the header and changed leaf entries, if any.
 945  */
 946 void
 947 xfs_dir3_leaf_compact(
 948         xfs_da_args_t   *args,          /* operation arguments */
 949         struct xfs_dir3_icleaf_hdr *leafhdr,
 950         struct xfs_buf  *bp)            /* leaf buffer */
 951 {
 952         int             from;           /* source leaf index */
 953         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 954         int             loglow;         /* first leaf entry to log */
 955         int             to;             /* target leaf index */
 956         struct xfs_dir2_leaf_entry *ents;
 957
 958         leaf = bp->b_addr;
 959         if (!leafhdr->stale)
 960                 return;
 961
 962         /*
 963          * Compress out the stale entries in place.
 964          */
 965         ents = xfs_dir3_leaf_ents_p(leaf);
 966         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 967                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 968                         continue;
 969                 /*
 970                  * Only actually copy the entries that are different.
 971                  */
 972                 if (from > to) {
 973                         if (loglow == -1)
 974                                 loglow = to;
 975                         ents[to] = ents[from];
 976                 }
 977                 to++;
 978         }
 979         /*
 980          * Update and log the header, log the leaf entries.
 981          */
 982         ASSERT(leafhdr->stale == from - to);
 983         leafhdr->count -= leafhdr->stale;
 984         leafhdr->stale = 0;
 985
 986         xfs_dir3_leaf_hdr_to_disk(leaf, leafhdr);
 987         xfs_dir3_leaf_log_header(args->trans, bp);
 988         if (loglow != -1)
 989                 xfs_dir3_leaf_log_ents(args->trans, bp, loglow, to - 1);
 990 }
 991
 992 /*
 993  * Compact the leaf entries, removing stale ones.
 994  * Leave one stale entry behind - the one closest to our
 995  * insertion index - and the caller will shift that one to our insertion
 996  * point later.
 997  * Return new insertion index, where the remaining stale entry is,
 998  * and leaf logging indices.
 999  */
1000 void
1001 xfs_dir3_leaf_compact_x1(
1002         struct xfs_dir3_icleaf_hdr *leafhdr,
1003         struct xfs_dir2_leaf_entry *ents,
1004         int             *indexp,        /* insertion index */
1005         int             *lowstalep,     /* out: stale entry before us */
1006         int             *highstalep,    /* out: stale entry after us */
1007         int             *lowlogp,       /* out: low log index */
1008         int             *highlogp)      /* out: high log index */
1009 {
1010         int             from;           /* source copy index */
1011         int             highstale;      /* stale entry at/after index */
1012         int             index;          /* insertion index */
1013         int             keepstale;      /* source index of kept stale */
1014         int             lowstale;       /* stale entry before index */
1015         int             newindex=0;     /* new insertion index */
1016         int             to;             /* destination copy index */
1017
1018         ASSERT(leafhdr->stale > 1);
1019         index = *indexp;
1020
1021         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
1022
1023         /*
1024          * Pick the better of lowstale and highstale.
1025          */
1026         if (lowstale >= 0 &&
1027             (highstale == leafhdr->count ||
1028              index - lowstale <= highstale - index))
1029                 keepstale = lowstale;
1030         else
1031                 keepstale = highstale;
1032         /*
1033          * Copy the entries in place, removing all the stale entries
1034          * except keepstale.
1035          */
1036         for (from = to = 0; from < leafhdr->count; from++) {
1037                 /*
1038                  * Notice the new value of index.
1039                  */
1040                 if (index == from)
1041                         newindex = to;
1042                 if (from != keepstale &&
1043                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1044                         if (from == to)
1045                                 *lowlogp = to;
1046                         continue;
1047                 }
1048                 /*
1049                  * Record the new keepstale value for the insertion.
1050                  */
1051                 if (from == keepstale)
1052                         lowstale = highstale = to;
1053                 /*
1054                  * Copy only the entries that have moved.
1055                  */
1056                 if (from > to)
1057                         ents[to] = ents[from];
1058                 to++;
1059         }
1060         ASSERT(from > to);
1061         /*
1062          * If the insertion point was past the last entry,
1063          * set the new insertion point accordingly.
1064          */
1065         if (index == from)
1066                 newindex = to;
1067         *indexp = newindex;
1068         /*
1069          * Adjust the leaf header values.
1070          */
1071         leafhdr->count -= from - to;
1072         leafhdr->stale = 1;
1073         /*
1074          * Remember the low/high stale value only in the "right"
1075          * direction.
1076          */
1077         if (lowstale >= newindex)
1078                 lowstale = -1;
1079         else
1080                 highstale = leafhdr->count;
1081         *highlogp = leafhdr->count - 1;
1082         *lowstalep = lowstale;
1083         *highstalep = highstale;
1084 }
1085
1086 struct xfs_dir2_leaf_map_info {
1087         xfs_extlen_t    map_blocks;     /* number of fsbs in map */
1088         xfs_dablk_t     map_off;        /* last mapped file offset */
1089         int             map_size;       /* total entries in *map */
1090         int             map_valid;      /* valid entries in *map */
1091         int             nmap;           /* mappings to ask xfs_bmapi */
1092         xfs_dir2_db_t   curdb;          /* db for current block */
1093         int             ra_current;     /* number of read-ahead blks */
1094         int             ra_index;       /* *map index for read-ahead */
1095         int             ra_offset;      /* map entry offset for ra */
1096         int             ra_want;        /* readahead count wanted */
1097         struct xfs_bmbt_irec map[];     /* map vector for blocks */
1098 };
1099
1100 STATIC int
1101 xfs_dir2_leaf_readbuf(
1102         struct xfs_inode        *dp,
1103         size_t                  bufsize,
1104         struct xfs_dir2_leaf_map_info *mip,
1105         xfs_dir2_off_t          *curoff,
1106         struct xfs_buf          **bpp)
1107 {
1108         struct xfs_mount        *mp = dp->i_mount;
1109         struct xfs_buf          *bp = *bpp;
1110         struct xfs_bmbt_irec    *map = mip->map;
1111         struct blk_plug         plug;
1112         int                     error = 0;
1113         int                     length;
1114         int                     i;
1115         int                     j;
1116
1117         /*
1118          * If we have a buffer, we need to release it and
1119          * take it out of the mapping.
1120          */
1121
1122         if (bp) {
1123                 xfs_trans_brelse(NULL, bp);
1124                 bp = NULL;
1125                 mip->map_blocks -= mp->m_dirblkfsbs;
1126                 /*
1127                  * Loop to get rid of the extents for the
1128                  * directory block.
1129                  */
1130                 for (i = mp->m_dirblkfsbs; i > 0; ) {
1131                         j = min_t(int, map->br_blockcount, i);
1132                         map->br_blockcount -= j;
1133                         map->br_startblock += j;
1134                         map->br_startoff += j;
1135                         /*
1136                          * If mapping is done, pitch it from
1137                          * the table.
1138                          */
1139                         if (!map->br_blockcount && --mip->map_valid)
1140                                 memmove(&map[0], &map[1],
1141                                         sizeof(map[0]) * mip->map_valid);
1142                         i -= j;
1143                 }
1144         }
1145
1146         /*
1147          * Recalculate the readahead blocks wanted.
1148          */
1149         mip->ra_want = howmany(bufsize + mp->m_dirblksize,
1150                                mp->m_sb.sb_blocksize) - 1;
1151         ASSERT(mip->ra_want >= 0);
1152
1153         /*
1154          * If we don't have as many as we want, and we haven't
1155          * run out of data blocks, get some more mappings.
1156          */
1157         if (1 + mip->ra_want > mip->map_blocks &&
1158             mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
1159                 /*
1160                  * Get more bmaps, fill in after the ones
1161                  * we already have in the table.
1162                  */
1163                 mip->nmap = mip->map_size - mip->map_valid;
1164                 error = xfs_bmapi_read(dp, mip->map_off,
1165                                 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
1166                                                                 mip->map_off,
1167                                 &map[mip->map_valid], &mip->nmap, 0);
1168
1169                 /*
1170                  * Don't know if we should ignore this or try to return an
1171                  * error.  The trouble with returning errors is that readdir
1172                  * will just stop without actually passing the error through.
1173                  */
1174                 if (error)
1175                         goto out;       /* XXX */
1176
1177                 /*
1178                  * If we got all the mappings we asked for, set the final map
1179                  * offset based on the last bmap value received.  Otherwise,
1180                  * we've reached the end.
1181                  */
1182                 if (mip->nmap == mip->map_size - mip->map_valid) {
1183                         i = mip->map_valid + mip->nmap - 1;
1184                         mip->map_off = map[i].br_startoff + map[i].br_blockcount;
1185                 } else
1186                         mip->map_off = xfs_dir2_byte_to_da(mp,
1187                                                         XFS_DIR2_LEAF_OFFSET);
1188
1189                 /*
1190                  * Look for holes in the mapping, and eliminate them.  Count up
1191                  * the valid blocks.
1192                  */
1193                 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
1194                         if (map[i].br_startblock == HOLESTARTBLOCK) {
1195                                 mip->nmap--;
1196                                 length = mip->map_valid + mip->nmap - i;
1197                                 if (length)
1198                                         memmove(&map[i], &map[i + 1],
1199                                                 sizeof(map[i]) * length);
1200                         } else {
1201                                 mip->map_blocks += map[i].br_blockcount;
1202                                 i++;
1203                         }
1204                 }
1205                 mip->map_valid += mip->nmap;
1206         }
1207
1208         /*
1209          * No valid mappings, so no more data blocks.
1210          */
1211         if (!mip->map_valid) {
1212                 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
1213                 goto out;
1214         }
1215
1216         /*
1217          * Read the directory block starting at the first mapping.
1218          */
1219         mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
1220         error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
1221                         map->br_blockcount >= mp->m_dirblkfsbs ?
1222                             XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
1223
1224         /*
1225          * Should just skip over the data block instead of giving up.
1226          */
1227         if (error)
1228                 goto out;       /* XXX */
1229
1230         /*
1231          * Adjust the current amount of read-ahead: we just read a block that
1232          * was previously ra.
1233          */
1234         if (mip->ra_current)
1235                 mip->ra_current -= mp->m_dirblkfsbs;
1236
1237         /*
1238          * Do we need more readahead?
1239          */
1240         blk_start_plug(&plug);
1241         for (mip->ra_index = mip->ra_offset = i = 0;
1242              mip->ra_want > mip->ra_current && i < mip->map_blocks;
1243              i += mp->m_dirblkfsbs) {
1244                 ASSERT(mip->ra_index < mip->map_valid);
1245                 /*
1246                  * Read-ahead a contiguous directory block.
1247                  */
1248                 if (i > mip->ra_current &&
1249                     map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
1250                         xfs_dir3_data_readahead(NULL, dp,
1251                                 map[mip->ra_index].br_startoff + mip->ra_offset,
1252                                 XFS_FSB_TO_DADDR(mp,
1253                                         map[mip->ra_index].br_startblock +
1254                                                         mip->ra_offset));
1255                         mip->ra_current = i;
1256                 }
1257
1258                 /*
1259                  * Read-ahead a non-contiguous directory block.  This doesn't
1260                  * use our mapping, but this is a very rare case.
1261                  */
1262                 else if (i > mip->ra_current) {
1263                         xfs_dir3_data_readahead(NULL, dp,
1264                                         map[mip->ra_index].br_startoff +
1265                                                         mip->ra_offset, -1);
1266                         mip->ra_current = i;
1267                 }
1268
1269                 /*
1270                  * Advance offset through the mapping table.
1271                  */
1272                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1273                         /*
1274                          * The rest of this extent but not more than a dir
1275                          * block.
1276                          */
1277                         length = min_t(int, mp->m_dirblkfsbs,
1278                                         map[mip->ra_index].br_blockcount -
1279                                                         mip->ra_offset);
1280                         j += length;
1281                         mip->ra_offset += length;
1282
1283                         /*
1284                          * Advance to the next mapping if this one is used up.
1285                          */
1286                         if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
1287                                 mip->ra_offset = 0;
1288                                 mip->ra_index++;
1289                         }
1290                 }
1291         }
1292         blk_finish_plug(&plug);
1293
1294 out:
1295         *bpp = bp;
1296         return error;
1297 }
1298
1299 /*
1300  * Getdents (readdir) for leaf and node directories.
1301  * This reads the data blocks only, so is the same for both forms.
1302  */
1303 int                                             /* error */
1304 xfs_dir2_leaf_getdents(
1305         xfs_inode_t             *dp,            /* incore directory inode */
1306         struct dir_context      *ctx,
1307         size_t                  bufsize)
1308 {
1309         struct xfs_buf          *bp = NULL;     /* data block buffer */
1310         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1311         xfs_dir2_data_entry_t   *dep;           /* data entry */
1312         xfs_dir2_data_unused_t  *dup;           /* unused entry */
1313         int                     error = 0;      /* error return value */
1314         int                     length;         /* temporary length value */
1315         xfs_mount_t             *mp;            /* filesystem mount point */
1316         int                     byteoff;        /* offset in current block */
1317         xfs_dir2_off_t          curoff;         /* current overall offset */
1318         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
1319         char                    *ptr = NULL;    /* pointer to current data */
1320         struct xfs_dir2_leaf_map_info *map_info;
1321
1322         /*
1323          * If the offset is at or past the largest allowed value,
1324          * give up right away.
1325          */
1326         if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
1327                 return 0;
1328
1329         mp = dp->i_mount;
1330
1331         /*
1332          * Set up to bmap a number of blocks based on the caller's
1333          * buffer size, the directory block size, and the filesystem
1334          * block size.
1335          */
1336         length = howmany(bufsize + mp->m_dirblksize,
1337                                      mp->m_sb.sb_blocksize);
1338         map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
1339                                 (length * sizeof(struct xfs_bmbt_irec)),
1340                                KM_SLEEP | KM_NOFS);
1341         map_info->map_size = length;
1342
1343         /*
1344          * Inside the loop we keep the main offset value as a byte offset
1345          * in the directory file.
1346          */
1347         curoff = xfs_dir2_dataptr_to_byte(mp, ctx->pos);
1348
1349         /*
1350          * Force this conversion through db so we truncate the offset
1351          * down to get the start of the data block.
1352          */
1353         map_info->map_off = xfs_dir2_db_to_da(mp,
1354                                               xfs_dir2_byte_to_db(mp, curoff));
1355
1356         /*
1357          * Loop over directory entries until we reach the end offset.
1358          * Get more blocks and readahead as necessary.
1359          */
1360         while (curoff < XFS_DIR2_LEAF_OFFSET) {
1361                 /*
1362                  * If we have no buffer, or we're off the end of the
1363                  * current buffer, need to get another one.
1364                  */
1365                 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
1366
1367                         error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
1368                                                       &curoff, &bp);
1369                         if (error || !map_info->map_valid)
1370                                 break;
1371
1372                         /*
1373                          * Having done a read, we need to set a new offset.
1374                          */
1375                         newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
1376                         /*
1377                          * Start of the current block.
1378                          */
1379                         if (curoff < newoff)
1380                                 curoff = newoff;
1381                         /*
1382                          * Make sure we're in the right block.
1383                          */
1384                         else if (curoff > newoff)
1385                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1386                                        map_info->curdb);
1387                         hdr = bp->b_addr;
1388                         xfs_dir3_data_check(dp, bp);
1389                         /*
1390                          * Find our position in the block.
1391                          */
1392                         ptr = (char *)xfs_dir3_data_entry_p(hdr);
1393                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1394                         /*
1395                          * Skip past the header.
1396                          */
1397                         if (byteoff == 0)
1398                                 curoff += xfs_dir3_data_entry_offset(hdr);
1399                         /*
1400                          * Skip past entries until we reach our offset.
1401                          */
1402                         else {
1403                                 while ((char *)ptr - (char *)hdr < byteoff) {
1404                                         dup = (xfs_dir2_data_unused_t *)ptr;
1405
1406                                         if (be16_to_cpu(dup->freetag)
1407                                                   == XFS_DIR2_DATA_FREE_TAG) {
1408
1409                                                 length = be16_to_cpu(dup->length);
1410                                                 ptr += length;
1411                                                 continue;
1412                                         }
1413                                         dep = (xfs_dir2_data_entry_t *)ptr;
1414                                         length =
1415                                            xfs_dir2_data_entsize(dep->namelen);
1416                                         ptr += length;
1417                                 }
1418                                 /*
1419                                  * Now set our real offset.
1420                                  */
1421                                 curoff =
1422                                         xfs_dir2_db_off_to_byte(mp,
1423                                             xfs_dir2_byte_to_db(mp, curoff),
1424                                             (char *)ptr - (char *)hdr);
1425                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1426                                         continue;
1427                                 }
1428                         }
1429                 }
1430                 /*
1431                  * We have a pointer to an entry.
1432                  * Is it a live one?
1433                  */
1434                 dup = (xfs_dir2_data_unused_t *)ptr;
1435                 /*
1436                  * No, it's unused, skip over it.
1437                  */
1438                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1439                         length = be16_to_cpu(dup->length);
1440                         ptr += length;
1441                         curoff += length;
1442                         continue;
1443                 }
1444
1445                 dep = (xfs_dir2_data_entry_t *)ptr;
1446                 length = xfs_dir2_data_entsize(dep->namelen);
1447
1448                 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1449                 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
1450                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1451                         break;
1452
1453                 /*
1454                  * Advance to next entry in the block.
1455                  */
1456                 ptr += length;
1457                 curoff += length;
1458                 /* bufsize may have just been a guess; don't go negative */
1459                 bufsize = bufsize > length ? bufsize - length : 0;
1460         }
1461
1462         /*
1463          * All done.  Set output offset value to current offset.
1464          */
1465         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1466                 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1467         else
1468                 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1469         kmem_free(map_info);
1470         if (bp)
1471                 xfs_trans_brelse(NULL, bp);
1472         return error;
1473 }
1474
1475
1476 /*
1477  * Log the bests entries indicated from a leaf1 block.
1478  */
1479 static void
1480 xfs_dir3_leaf_log_bests(
1481         xfs_trans_t             *tp,            /* transaction pointer */
1482         struct xfs_buf          *bp,            /* leaf buffer */
1483         int                     first,          /* first entry to log */
1484         int                     last)           /* last entry to log */
1485 {
1486         __be16                  *firstb;        /* pointer to first entry */
1487         __be16                  *lastb;         /* pointer to last entry */
1488         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1489         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1490
1491         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1492                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1493
1494         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1495         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1496         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1497         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1498                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1499 }
1500
1501 /*
1502  * Log the leaf entries indicated from a leaf1 or leafn block.
1503  */
1504 void
1505 xfs_dir3_leaf_log_ents(
1506         xfs_trans_t             *tp,            /* transaction pointer */
1507         struct xfs_buf          *bp,            /* leaf buffer */
1508         int                     first,          /* first entry to log */
1509         int                     last)           /* last entry to log */
1510 {
1511         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1512         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1513         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1514         struct xfs_dir2_leaf_entry *ents;
1515
1516         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1517                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1518                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1519                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1520
1521         ents = xfs_dir3_leaf_ents_p(leaf);
1522         firstlep = &ents[first];
1523         lastlep = &ents[last];
1524         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1525                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1526 }
1527
1528 /*
1529  * Log the header of the leaf1 or leafn block.
1530  */
1531 void
1532 xfs_dir3_leaf_log_header(
1533         struct xfs_trans        *tp,
1534         struct xfs_buf          *bp)
1535 {
1536         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1537
1538         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1539                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1540                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1541                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1542
1543         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1544                           xfs_dir3_leaf_hdr_size(leaf) - 1);
1545 }
1546
1547 /*
1548  * Log the tail of the leaf1 block.
1549  */
1550 STATIC void
1551 xfs_dir3_leaf_log_tail(
1552         struct xfs_trans        *tp,
1553         struct xfs_buf          *bp)
1554 {
1555         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1556         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1557         struct xfs_mount        *mp = tp->t_mountp;
1558
1559         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1560                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1561                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1562                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1563
1564         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1565         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1566                 (uint)(mp->m_dirblksize - 1));
1567 }
1568
1569 /*
1570  * Look up the entry referred to by args in the leaf format directory.
1571  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1572  * is also used by the node-format code.
1573  */
1574 int
1575 xfs_dir2_leaf_lookup(
1576         xfs_da_args_t           *args)          /* operation arguments */
1577 {
1578         struct xfs_buf          *dbp;           /* data block buffer */
1579         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1580         xfs_inode_t             *dp;            /* incore directory inode */
1581         int                     error;          /* error return code */
1582         int                     index;          /* found entry index */
1583         struct xfs_buf          *lbp;           /* leaf buffer */
1584         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1585         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1586         xfs_trans_t             *tp;            /* transaction pointer */
1587         struct xfs_dir2_leaf_entry *ents;
1588
1589         trace_xfs_dir2_leaf_lookup(args);
1590
1591         /*
1592          * Look up name in the leaf block, returning both buffers and index.
1593          */
1594         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1595                 return error;
1596         }
1597         tp = args->trans;
1598         dp = args->dp;
1599         xfs_dir3_leaf_check(dp->i_mount, lbp);
1600         leaf = lbp->b_addr;
1601         ents = xfs_dir3_leaf_ents_p(leaf);
1602         /*
1603          * Get to the leaf entry and contained data entry address.
1604          */
1605         lep = &ents[index];
1606
1607         /*
1608          * Point to the data entry.
1609          */
1610         dep = (xfs_dir2_data_entry_t *)
1611               ((char *)dbp->b_addr +
1612                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1613         /*
1614          * Return the found inode number & CI name if appropriate
1615          */
1616         args->inumber = be64_to_cpu(dep->inumber);
1617         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1618         xfs_trans_brelse(tp, dbp);
1619         xfs_trans_brelse(tp, lbp);
1620         return XFS_ERROR(error);
1621 }
1622
1623 /*
1624  * Look up name/hash in the leaf block.
1625  * Fill in indexp with the found index, and dbpp with the data buffer.
1626  * If not found dbpp will be NULL, and ENOENT comes back.
1627  * lbpp will always be filled in with the leaf buffer unless there's an error.
1628  */
1629 static int                                      /* error */
1630 xfs_dir2_leaf_lookup_int(
1631         xfs_da_args_t           *args,          /* operation arguments */
1632         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1633         int                     *indexp,        /* out: index in leaf block */
1634         struct xfs_buf          **dbpp)         /* out: data buffer */
1635 {
1636         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1637         struct xfs_buf          *dbp = NULL;    /* data buffer */
1638         xfs_dir2_data_entry_t   *dep;           /* data entry */
1639         xfs_inode_t             *dp;            /* incore directory inode */
1640         int                     error;          /* error return code */
1641         int                     index;          /* index in leaf block */
1642         struct xfs_buf          *lbp;           /* leaf buffer */
1643         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1644         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1645         xfs_mount_t             *mp;            /* filesystem mount point */
1646         xfs_dir2_db_t           newdb;          /* new data block number */
1647         xfs_trans_t             *tp;            /* transaction pointer */
1648         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1649         enum xfs_dacmp          cmp;            /* name compare result */
1650         struct xfs_dir2_leaf_entry *ents;
1651         struct xfs_dir3_icleaf_hdr leafhdr;
1652
1653         dp = args->dp;
1654         tp = args->trans;
1655         mp = dp->i_mount;
1656
1657         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1658         if (error)
1659                 return error;
1660
1661         *lbpp = lbp;
1662         leaf = lbp->b_addr;
1663         xfs_dir3_leaf_check(mp, lbp);
1664         ents = xfs_dir3_leaf_ents_p(leaf);
1665         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1666
1667         /*
1668          * Look for the first leaf entry with our hash value.
1669          */
1670         index = xfs_dir2_leaf_search_hash(args, lbp);
1671         /*
1672          * Loop over all the entries with the right hash value
1673          * looking to match the name.
1674          */
1675         for (lep = &ents[index];
1676              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1677              lep++, index++) {
1678                 /*
1679                  * Skip over stale leaf entries.
1680                  */
1681                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1682                         continue;
1683                 /*
1684                  * Get the new data block number.
1685                  */
1686                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1687                 /*
1688                  * If it's not the same as the old data block number,
1689                  * need to pitch the old one and read the new one.
1690                  */
1691                 if (newdb != curdb) {
1692                         if (dbp)
1693                                 xfs_trans_brelse(tp, dbp);
1694                         error = xfs_dir3_data_read(tp, dp,
1695                                                    xfs_dir2_db_to_da(mp, newdb),
1696                                                    -1, &dbp);
1697                         if (error) {
1698                                 xfs_trans_brelse(tp, lbp);
1699                                 return error;
1700                         }
1701                         curdb = newdb;
1702                 }
1703                 /*
1704                  * Point to the data entry.
1705                  */
1706                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1707                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1708                 /*
1709                  * Compare name and if it's an exact match, return the index
1710                  * and buffer. If it's the first case-insensitive match, store
1711                  * the index and buffer and continue looking for an exact match.
1712                  */
1713                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1714                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1715                         args->cmpresult = cmp;
1716                         *indexp = index;
1717                         /* case exact match: return the current buffer. */
1718                         if (cmp == XFS_CMP_EXACT) {
1719                                 *dbpp = dbp;
1720                                 return 0;
1721                         }
1722                         cidb = curdb;
1723                 }
1724         }
1725         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1726         /*
1727          * Here, we can only be doing a lookup (not a rename or remove).
1728          * If a case-insensitive match was found earlier, re-read the
1729          * appropriate data block if required and return it.
1730          */
1731         if (args->cmpresult == XFS_CMP_CASE) {
1732                 ASSERT(cidb != -1);
1733                 if (cidb != curdb) {
1734                         xfs_trans_brelse(tp, dbp);
1735                         error = xfs_dir3_data_read(tp, dp,
1736                                                    xfs_dir2_db_to_da(mp, cidb),
1737                                                    -1, &dbp);
1738                         if (error) {
1739                                 xfs_trans_brelse(tp, lbp);
1740                                 return error;
1741                         }
1742                 }
1743                 *dbpp = dbp;
1744                 return 0;
1745         }
1746         /*
1747          * No match found, return ENOENT.
1748          */
1749         ASSERT(cidb == -1);
1750         if (dbp)
1751                 xfs_trans_brelse(tp, dbp);
1752         xfs_trans_brelse(tp, lbp);
1753         return XFS_ERROR(ENOENT);
1754 }
1755
1756 /*
1757  * Remove an entry from a leaf format directory.
1758  */
1759 int                                             /* error */
1760 xfs_dir2_leaf_removename(
1761         xfs_da_args_t           *args)          /* operation arguments */
1762 {
1763         __be16                  *bestsp;        /* leaf block best freespace */
1764         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1765         xfs_dir2_db_t           db;             /* data block number */
1766         struct xfs_buf          *dbp;           /* data block buffer */
1767         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1768         xfs_inode_t             *dp;            /* incore directory inode */
1769         int                     error;          /* error return code */
1770         xfs_dir2_db_t           i;              /* temporary data block # */
1771         int                     index;          /* index into leaf entries */
1772         struct xfs_buf          *lbp;           /* leaf buffer */
1773         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1774         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1775         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1776         xfs_mount_t             *mp;            /* filesystem mount point */
1777         int                     needlog;        /* need to log data header */
1778         int                     needscan;       /* need to rescan data frees */
1779         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1780         xfs_trans_t             *tp;            /* transaction pointer */
1781         struct xfs_dir2_data_free *bf;          /* bestfree table */
1782         struct xfs_dir2_leaf_entry *ents;
1783         struct xfs_dir3_icleaf_hdr leafhdr;
1784
1785         trace_xfs_dir2_leaf_removename(args);
1786
1787         /*
1788          * Lookup the leaf entry, get the leaf and data blocks read in.
1789          */
1790         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1791                 return error;
1792         }
1793         dp = args->dp;
1794         tp = args->trans;
1795         mp = dp->i_mount;
1796         leaf = lbp->b_addr;
1797         hdr = dbp->b_addr;
1798         xfs_dir3_data_check(dp, dbp);
1799         bf = xfs_dir3_data_bestfree_p(hdr);
1800         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1801         ents = xfs_dir3_leaf_ents_p(leaf);
1802         /*
1803          * Point to the leaf entry, use that to point to the data entry.
1804          */
1805         lep = &ents[index];
1806         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1807         dep = (xfs_dir2_data_entry_t *)
1808               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1809         needscan = needlog = 0;
1810         oldbest = be16_to_cpu(bf[0].length);
1811         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1812         bestsp = xfs_dir2_leaf_bests_p(ltp);
1813         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1814         /*
1815          * Mark the former data entry unused.
1816          */
1817         xfs_dir2_data_make_free(tp, dbp,
1818                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1819                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1820         /*
1821          * We just mark the leaf entry stale by putting a null in it.
1822          */
1823         leafhdr.stale++;
1824         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1825         xfs_dir3_leaf_log_header(tp, lbp);
1826
1827         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1828         xfs_dir3_leaf_log_ents(tp, lbp, index, index);
1829
1830         /*
1831          * Scan the freespace in the data block again if necessary,
1832          * log the data block header if necessary.
1833          */
1834         if (needscan)
1835                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1836         if (needlog)
1837                 xfs_dir2_data_log_header(tp, dbp);
1838         /*
1839          * If the longest freespace in the data block has changed,
1840          * put the new value in the bests table and log that.
1841          */
1842         if (be16_to_cpu(bf[0].length) != oldbest) {
1843                 bestsp[db] = bf[0].length;
1844                 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1845         }
1846         xfs_dir3_data_check(dp, dbp);
1847         /*
1848          * If the data block is now empty then get rid of the data block.
1849          */
1850         if (be16_to_cpu(bf[0].length) ==
1851                         mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr)) {
1852                 ASSERT(db != mp->m_dirdatablk);
1853                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1854                         /*
1855                          * Nope, can't get rid of it because it caused
1856                          * allocation of a bmap btree block to do so.
1857                          * Just go on, returning success, leaving the
1858                          * empty block in place.
1859                          */
1860                         if (error == ENOSPC && args->total == 0)
1861                                 error = 0;
1862                         xfs_dir3_leaf_check(mp, lbp);
1863                         return error;
1864                 }
1865                 dbp = NULL;
1866                 /*
1867                  * If this is the last data block then compact the
1868                  * bests table by getting rid of entries.
1869                  */
1870                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1871                         /*
1872                          * Look for the last active entry (i).
1873                          */
1874                         for (i = db - 1; i > 0; i--) {
1875                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1876                                         break;
1877                         }
1878                         /*
1879                          * Copy the table down so inactive entries at the
1880                          * end are removed.
1881                          */
1882                         memmove(&bestsp[db - i], bestsp,
1883                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1884                         be32_add_cpu(&ltp->bestcount, -(db - i));
1885                         xfs_dir3_leaf_log_tail(tp, lbp);
1886                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1887                 } else
1888                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1889         }
1890         /*
1891          * If the data block was not the first one, drop it.
1892          */
1893         else if (db != mp->m_dirdatablk)
1894                 dbp = NULL;
1895
1896         xfs_dir3_leaf_check(mp, lbp);
1897         /*
1898          * See if we can convert to block form.
1899          */
1900         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1901 }
1902
1903 /*
1904  * Replace the inode number in a leaf format directory entry.
1905  */
1906 int                                             /* error */
1907 xfs_dir2_leaf_replace(
1908         xfs_da_args_t           *args)          /* operation arguments */
1909 {
1910         struct xfs_buf          *dbp;           /* data block buffer */
1911         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1912         xfs_inode_t             *dp;            /* incore directory inode */
1913         int                     error;          /* error return code */
1914         int                     index;          /* index of leaf entry */
1915         struct xfs_buf          *lbp;           /* leaf buffer */
1916         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1917         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1918         xfs_trans_t             *tp;            /* transaction pointer */
1919         struct xfs_dir2_leaf_entry *ents;
1920
1921         trace_xfs_dir2_leaf_replace(args);
1922
1923         /*
1924          * Look up the entry.
1925          */
1926         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1927                 return error;
1928         }
1929         dp = args->dp;
1930         leaf = lbp->b_addr;
1931         ents = xfs_dir3_leaf_ents_p(leaf);
1932         /*
1933          * Point to the leaf entry, get data address from it.
1934          */
1935         lep = &ents[index];
1936         /*
1937          * Point to the data entry.
1938          */
1939         dep = (xfs_dir2_data_entry_t *)
1940               ((char *)dbp->b_addr +
1941                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1942         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1943         /*
1944          * Put the new inode number in, log it.
1945          */
1946         dep->inumber = cpu_to_be64(args->inumber);
1947         tp = args->trans;
1948         xfs_dir2_data_log_entry(tp, dbp, dep);
1949         xfs_dir3_leaf_check(dp->i_mount, lbp);
1950         xfs_trans_brelse(tp, lbp);
1951         return 0;
1952 }
1953
1954 /*
1955  * Return index in the leaf block (lbp) which is either the first
1956  * one with this hash value, or if there are none, the insert point
1957  * for that hash value.
1958  */
1959 int                                             /* index value */
1960 xfs_dir2_leaf_search_hash(
1961         xfs_da_args_t           *args,          /* operation arguments */
1962         struct xfs_buf          *lbp)           /* leaf buffer */
1963 {
1964         xfs_dahash_t            hash=0;         /* hash from this entry */
1965         xfs_dahash_t            hashwant;       /* hash value looking for */
1966         int                     high;           /* high leaf index */
1967         int                     low;            /* low leaf index */
1968         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1969         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1970         int                     mid=0;          /* current leaf index */
1971         struct xfs_dir2_leaf_entry *ents;
1972         struct xfs_dir3_icleaf_hdr leafhdr;
1973
1974         leaf = lbp->b_addr;
1975         ents = xfs_dir3_leaf_ents_p(leaf);
1976         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1977
1978 #ifndef __KERNEL__
1979         if (!leafhdr.count)
1980                 return 0;
1981 #endif
1982         /*
1983          * Note, the table cannot be empty, so we have to go through the loop.
1984          * Binary search the leaf entries looking for our hash value.
1985          */
1986         for (lep = ents, low = 0, high = leafhdr.count - 1,
1987                 hashwant = args->hashval;
1988              low <= high; ) {
1989                 mid = (low + high) >> 1;
1990                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1991                         break;
1992                 if (hash < hashwant)
1993                         low = mid + 1;
1994                 else
1995                         high = mid - 1;
1996         }
1997         /*
1998          * Found one, back up through all the equal hash values.
1999          */
2000         if (hash == hashwant) {
2001                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
2002                         mid--;
2003                 }
2004         }
2005         /*
2006          * Need to point to an entry higher than ours.
2007          */
2008         else if (hash < hashwant)
2009                 mid++;
2010         return mid;
2011 }
2012
2013 /*
2014  * Trim off a trailing data block.  We know it's empty since the leaf
2015  * freespace table says so.
2016  */
2017 int                                             /* error */
2018 xfs_dir2_leaf_trim_data(
2019         xfs_da_args_t           *args,          /* operation arguments */
2020         struct xfs_buf          *lbp,           /* leaf buffer */
2021         xfs_dir2_db_t           db)             /* data block number */
2022 {
2023         __be16                  *bestsp;        /* leaf bests table */
2024         struct xfs_buf          *dbp;           /* data block buffer */
2025         xfs_inode_t             *dp;            /* incore directory inode */
2026         int                     error;          /* error return value */
2027         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2028         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
2029         xfs_mount_t             *mp;            /* filesystem mount point */
2030         xfs_trans_t             *tp;            /* transaction pointer */
2031
2032         dp = args->dp;
2033         mp = dp->i_mount;
2034         tp = args->trans;
2035         /*
2036          * Read the offending data block.  We need its buffer.
2037          */
2038         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
2039         if (error)
2040                 return error;
2041
2042         leaf = lbp->b_addr;
2043         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
2044
2045 #ifdef DEBUG
2046 {
2047         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
2048         struct xfs_dir2_data_free *bf = xfs_dir3_data_bestfree_p(hdr);
2049
2050         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2051                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2052         ASSERT(be16_to_cpu(bf[0].length) ==
2053                mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr));
2054         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
2055 }
2056 #endif
2057
2058         /*
2059          * Get rid of the data block.
2060          */
2061         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
2062                 ASSERT(error != ENOSPC);
2063                 xfs_trans_brelse(tp, dbp);
2064                 return error;
2065         }
2066         /*
2067          * Eliminate the last bests entry from the table.
2068          */
2069         bestsp = xfs_dir2_leaf_bests_p(ltp);
2070         be32_add_cpu(&ltp->bestcount, -1);
2071         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
2072         xfs_dir3_leaf_log_tail(tp, lbp);
2073         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
2074         return 0;
2075 }
2076
2077 static inline size_t
2078 xfs_dir3_leaf_size(
2079         struct xfs_dir3_icleaf_hdr      *hdr,
2080         int                             counts)
2081 {
2082         int     entries;
2083         int     hdrsize;
2084
2085         entries = hdr->count - hdr->stale;
2086         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
2087             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
2088                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
2089         else
2090                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
2091
2092         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
2093                        + counts * sizeof(xfs_dir2_data_off_t)
2094                        + sizeof(xfs_dir2_leaf_tail_t);
2095 }
2096
2097 /*
2098  * Convert node form directory to leaf form directory.
2099  * The root of the node form dir needs to already be a LEAFN block.
2100  * Just return if we can't do anything.
2101  */
2102 int                                             /* error */
2103 xfs_dir2_node_to_leaf(
2104         xfs_da_state_t          *state)         /* directory operation state */
2105 {
2106         xfs_da_args_t           *args;          /* operation arguments */
2107         xfs_inode_t             *dp;            /* incore directory inode */
2108         int                     error;          /* error return code */
2109         struct xfs_buf          *fbp;           /* buffer for freespace block */
2110         xfs_fileoff_t           fo;             /* freespace file offset */
2111         xfs_dir2_free_t         *free;          /* freespace structure */
2112         struct xfs_buf          *lbp;           /* buffer for leaf block */
2113         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
2114         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2115         xfs_mount_t             *mp;            /* filesystem mount point */
2116         int                     rval;           /* successful free trim? */
2117         xfs_trans_t             *tp;            /* transaction pointer */
2118         struct xfs_dir3_icleaf_hdr leafhdr;
2119         struct xfs_dir3_icfree_hdr freehdr;
2120
2121         /*
2122          * There's more than a leaf level in the btree, so there must
2123          * be multiple leafn blocks.  Give up.
2124          */
2125         if (state->path.active > 1)
2126                 return 0;
2127         args = state->args;
2128
2129         trace_xfs_dir2_node_to_leaf(args);
2130
2131         mp = state->mp;
2132         dp = args->dp;
2133         tp = args->trans;
2134         /*
2135          * Get the last offset in the file.
2136          */
2137         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
2138                 return error;
2139         }
2140         fo -= mp->m_dirblkfsbs;
2141         /*
2142          * If there are freespace blocks other than the first one,
2143          * take this opportunity to remove trailing empty freespace blocks
2144          * that may have been left behind during no-space-reservation
2145          * operations.
2146          */
2147         while (fo > mp->m_dirfreeblk) {
2148                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
2149                         return error;
2150                 }
2151                 if (rval)
2152                         fo -= mp->m_dirblkfsbs;
2153                 else
2154                         return 0;
2155         }
2156         /*
2157          * Now find the block just before the freespace block.
2158          */
2159         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
2160                 return error;
2161         }
2162         /*
2163          * If it's not the single leaf block, give up.
2164          */
2165         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
2166                 return 0;
2167         lbp = state->path.blk[0].bp;
2168         leaf = lbp->b_addr;
2169         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
2170
2171         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
2172                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
2173
2174         /*
2175          * Read the freespace block.
2176          */
2177         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
2178         if (error)
2179                 return error;
2180         free = fbp->b_addr;
2181         xfs_dir3_free_hdr_from_disk(&freehdr, free);
2182
2183         ASSERT(!freehdr.firstdb);
2184
2185         /*
2186          * Now see if the leafn and free data will fit in a leaf1.
2187          * If not, release the buffer and give up.
2188          */
2189         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
2190                 xfs_trans_brelse(tp, fbp);
2191                 return 0;
2192         }
2193
2194         /*
2195          * If the leaf has any stale entries in it, compress them out.
2196          */
2197         if (leafhdr.stale)
2198                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
2199
2200         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
2201         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
2202         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
2203                                         ? XFS_DIR2_LEAF1_MAGIC
2204                                         : XFS_DIR3_LEAF1_MAGIC;
2205
2206         /*
2207          * Set up the leaf tail from the freespace block.
2208          */
2209         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
2210         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
2211
2212         /*
2213          * Set up the leaf bests table.
2214          */
2215         memcpy(xfs_dir2_leaf_bests_p(ltp), xfs_dir3_free_bests_p(mp, free),
2216                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
2217
2218         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
2219         xfs_dir3_leaf_log_header(tp, lbp);
2220         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
2221         xfs_dir3_leaf_log_tail(tp, lbp);
2222         xfs_dir3_leaf_check(mp, lbp);
2223
2224         /*
2225          * Get rid of the freespace block.
2226          */
2227         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
2228         if (error) {
2229                 /*
2230                  * This can't fail here because it can only happen when
2231                  * punching out the middle of an extent, and this is an
2232                  * isolated block.
2233                  */
2234                 ASSERT(error != ENOSPC);
2235                 return error;
2236         }
2237         fbp = NULL;
2238         /*
2239          * Now see if we can convert the single-leaf directory
2240          * down to a block form directory.
2241          * This routine always kills the dabuf for the leaf, so
2242          * eliminate it from the path.
2243          */
2244         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
2245         state->path.blk[0].bp = NULL;
2246         return error;
2247 }