kernel/fs/xfs/libxfs/xfs_da_btree.c

   1 /*
   2  * Copyright (c) 2000-2005 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_shared.h"
  22 #include "xfs_format.h"
  23 #include "xfs_log_format.h"
  24 #include "xfs_trans_resv.h"
  25 #include "xfs_bit.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_dir2_priv.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_alloc.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_attr.h"
  37 #include "xfs_attr_leaf.h"
  38 #include "xfs_error.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_cksum.h"
  41 #include "xfs_buf_item.h"
  42
  43 /*
  44  * xfs_da_btree.c
  45  *
  46  * Routines to implement directories as Btrees of hashed names.
  47  */
  48
  49 /*========================================================================
  50  * Function prototypes for the kernel.
  51  *========================================================================*/
  52
  53 /*
  54  * Routines used for growing the Btree.
  55  */
  56 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
  57                                             xfs_da_state_blk_t *existing_root,
  58                                             xfs_da_state_blk_t *new_child);
  59 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
  60                                             xfs_da_state_blk_t *existing_blk,
  61                                             xfs_da_state_blk_t *split_blk,
  62                                             xfs_da_state_blk_t *blk_to_add,
  63                                             int treelevel,
  64                                             int *result);
  65 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
  66                                          xfs_da_state_blk_t *node_blk_1,
  67                                          xfs_da_state_blk_t *node_blk_2);
  68 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
  69                                    xfs_da_state_blk_t *old_node_blk,
  70                                    xfs_da_state_blk_t *new_node_blk);
  71
  72 /*
  73  * Routines used for shrinking the Btree.
  74  */
  75 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
  76                                            xfs_da_state_blk_t *root_blk);
  77 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
  78 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
  79                                               xfs_da_state_blk_t *drop_blk);
  80 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
  81                                          xfs_da_state_blk_t *src_node_blk,
  82                                          xfs_da_state_blk_t *dst_node_blk);
  83
  84 /*
  85  * Utility routines.
  86  */
  87 STATIC int      xfs_da3_blk_unlink(xfs_da_state_t *state,
  88                                   xfs_da_state_blk_t *drop_blk,
  89                                   xfs_da_state_blk_t *save_blk);
  90
  91
  92 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
  93
  94 /*
  95  * Allocate a dir-state structure.
  96  * We don't put them on the stack since they're large.
  97  */
  98 xfs_da_state_t *
  99 xfs_da_state_alloc(void)
 100 {
 101         return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
 102 }
 103
 104 /*
 105  * Kill the altpath contents of a da-state structure.
 106  */
 107 STATIC void
 108 xfs_da_state_kill_altpath(xfs_da_state_t *state)
 109 {
 110         int     i;
 111
 112         for (i = 0; i < state->altpath.active; i++)
 113                 state->altpath.blk[i].bp = NULL;
 114         state->altpath.active = 0;
 115 }
 116
 117 /*
 118  * Free a da-state structure.
 119  */
 120 void
 121 xfs_da_state_free(xfs_da_state_t *state)
 122 {
 123         xfs_da_state_kill_altpath(state);
 124 #ifdef DEBUG
 125         memset((char *)state, 0, sizeof(*state));
 126 #endif /* DEBUG */
 127         kmem_zone_free(xfs_da_state_zone, state);
 128 }
 129
 130 static bool
 131 xfs_da3_node_verify(
 132         struct xfs_buf          *bp)
 133 {
 134         struct xfs_mount        *mp = bp->b_target->bt_mount;
 135         struct xfs_da_intnode   *hdr = bp->b_addr;
 136         struct xfs_da3_icnode_hdr ichdr;
 137         const struct xfs_dir_ops *ops;
 138
 139         ops = xfs_dir_get_ops(mp, NULL);
 140
 141         ops->node_hdr_from_disk(&ichdr, hdr);
 142
 143         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 144                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 145
 146                 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
 147                         return false;
 148
 149                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
 150                         return false;
 151                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
 152                         return false;
 153         } else {
 154                 if (ichdr.magic != XFS_DA_NODE_MAGIC)
 155                         return false;
 156         }
 157         if (ichdr.level == 0)
 158                 return false;
 159         if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
 160                 return false;
 161         if (ichdr.count == 0)
 162                 return false;
 163
 164         /*
 165          * we don't know if the node is for and attribute or directory tree,
 166          * so only fail if the count is outside both bounds
 167          */
 168         if (ichdr.count > mp->m_dir_geo->node_ents &&
 169             ichdr.count > mp->m_attr_geo->node_ents)
 170                 return false;
 171
 172         /* XXX: hash order check? */
 173
 174         return true;
 175 }
 176
 177 static void
 178 xfs_da3_node_write_verify(
 179         struct xfs_buf  *bp)
 180 {
 181         struct xfs_mount        *mp = bp->b_target->bt_mount;
 182         struct xfs_buf_log_item *bip = bp->b_fspriv;
 183         struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 184
 185         if (!xfs_da3_node_verify(bp)) {
 186                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 187                 xfs_verifier_error(bp);
 188                 return;
 189         }
 190
 191         if (!xfs_sb_version_hascrc(&mp->m_sb))
 192                 return;
 193
 194         if (bip)
 195                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 196
 197         xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
 198 }
 199
 200 /*
 201  * leaf/node format detection on trees is sketchy, so a node read can be done on
 202  * leaf level blocks when detection identifies the tree as a node format tree
 203  * incorrectly. In this case, we need to swap the verifier to match the correct
 204  * format of the block being read.
 205  */
 206 static void
 207 xfs_da3_node_read_verify(
 208         struct xfs_buf          *bp)
 209 {
 210         struct xfs_da_blkinfo   *info = bp->b_addr;
 211
 212         switch (be16_to_cpu(info->magic)) {
 213                 case XFS_DA3_NODE_MAGIC:
 214                         if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
 215                                 xfs_buf_ioerror(bp, -EFSBADCRC);
 216                                 break;
 217                         }
 218                         /* fall through */
 219                 case XFS_DA_NODE_MAGIC:
 220                         if (!xfs_da3_node_verify(bp)) {
 221                                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 222                                 break;
 223                         }
 224                         return;
 225                 case XFS_ATTR_LEAF_MAGIC:
 226                 case XFS_ATTR3_LEAF_MAGIC:
 227                         bp->b_ops = &xfs_attr3_leaf_buf_ops;
 228                         bp->b_ops->verify_read(bp);
 229                         return;
 230                 case XFS_DIR2_LEAFN_MAGIC:
 231                 case XFS_DIR3_LEAFN_MAGIC:
 232                         bp->b_ops = &xfs_dir3_leafn_buf_ops;
 233                         bp->b_ops->verify_read(bp);
 234                         return;
 235                 default:
 236                         break;
 237         }
 238
 239         /* corrupt block */
 240         xfs_verifier_error(bp);
 241 }
 242
 243 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
 244         .verify_read = xfs_da3_node_read_verify,
 245         .verify_write = xfs_da3_node_write_verify,
 246 };
 247
 248 int
 249 xfs_da3_node_read(
 250         struct xfs_trans        *tp,
 251         struct xfs_inode        *dp,
 252         xfs_dablk_t             bno,
 253         xfs_daddr_t             mappedbno,
 254         struct xfs_buf          **bpp,
 255         int                     which_fork)
 256 {
 257         int                     err;
 258
 259         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
 260                                         which_fork, &xfs_da3_node_buf_ops);
 261         if (!err && tp) {
 262                 struct xfs_da_blkinfo   *info = (*bpp)->b_addr;
 263                 int                     type;
 264
 265                 switch (be16_to_cpu(info->magic)) {
 266                 case XFS_DA_NODE_MAGIC:
 267                 case XFS_DA3_NODE_MAGIC:
 268                         type = XFS_BLFT_DA_NODE_BUF;
 269                         break;
 270                 case XFS_ATTR_LEAF_MAGIC:
 271                 case XFS_ATTR3_LEAF_MAGIC:
 272                         type = XFS_BLFT_ATTR_LEAF_BUF;
 273                         break;
 274                 case XFS_DIR2_LEAFN_MAGIC:
 275                 case XFS_DIR3_LEAFN_MAGIC:
 276                         type = XFS_BLFT_DIR_LEAFN_BUF;
 277                         break;
 278                 default:
 279                         type = 0;
 280                         ASSERT(0);
 281                         break;
 282                 }
 283                 xfs_trans_buf_set_type(tp, *bpp, type);
 284         }
 285         return err;
 286 }
 287
 288 /*========================================================================
 289  * Routines used for growing the Btree.
 290  *========================================================================*/
 291
 292 /*
 293  * Create the initial contents of an intermediate node.
 294  */
 295 int
 296 xfs_da3_node_create(
 297         struct xfs_da_args      *args,
 298         xfs_dablk_t             blkno,
 299         int                     level,
 300         struct xfs_buf          **bpp,
 301         int                     whichfork)
 302 {
 303         struct xfs_da_intnode   *node;
 304         struct xfs_trans        *tp = args->trans;
 305         struct xfs_mount        *mp = tp->t_mountp;
 306         struct xfs_da3_icnode_hdr ichdr = {0};
 307         struct xfs_buf          *bp;
 308         int                     error;
 309         struct xfs_inode        *dp = args->dp;
 310
 311         trace_xfs_da_node_create(args);
 312         ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
 313
 314         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
 315         if (error)
 316                 return error;
 317         bp->b_ops = &xfs_da3_node_buf_ops;
 318         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 319         node = bp->b_addr;
 320
 321         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 322                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 323
 324                 ichdr.magic = XFS_DA3_NODE_MAGIC;
 325                 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
 326                 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
 327                 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_uuid);
 328         } else {
 329                 ichdr.magic = XFS_DA_NODE_MAGIC;
 330         }
 331         ichdr.level = level;
 332
 333         dp->d_ops->node_hdr_to_disk(node, &ichdr);
 334         xfs_trans_log_buf(tp, bp,
 335                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 336
 337         *bpp = bp;
 338         return 0;
 339 }
 340
 341 /*
 342  * Split a leaf node, rebalance, then possibly split
 343  * intermediate nodes, rebalance, etc.
 344  */
 345 int                                                     /* error */
 346 xfs_da3_split(
 347         struct xfs_da_state     *state)
 348 {
 349         struct xfs_da_state_blk *oldblk;
 350         struct xfs_da_state_blk *newblk;
 351         struct xfs_da_state_blk *addblk;
 352         struct xfs_da_intnode   *node;
 353         struct xfs_buf          *bp;
 354         int                     max;
 355         int                     action = 0;
 356         int                     error;
 357         int                     i;
 358
 359         trace_xfs_da_split(state->args);
 360
 361         /*
 362          * Walk back up the tree splitting/inserting/adjusting as necessary.
 363          * If we need to insert and there isn't room, split the node, then
 364          * decide which fragment to insert the new block from below into.
 365          * Note that we may split the root this way, but we need more fixup.
 366          */
 367         max = state->path.active - 1;
 368         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 369         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 370                state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 371
 372         addblk = &state->path.blk[max];         /* initial dummy value */
 373         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 374                 oldblk = &state->path.blk[i];
 375                 newblk = &state->altpath.blk[i];
 376
 377                 /*
 378                  * If a leaf node then
 379                  *     Allocate a new leaf node, then rebalance across them.
 380                  * else if an intermediate node then
 381                  *     We split on the last layer, must we split the node?
 382                  */
 383                 switch (oldblk->magic) {
 384                 case XFS_ATTR_LEAF_MAGIC:
 385                         error = xfs_attr3_leaf_split(state, oldblk, newblk);
 386                         if ((error != 0) && (error != -ENOSPC)) {
 387                                 return error;   /* GROT: attr is inconsistent */
 388                         }
 389                         if (!error) {
 390                                 addblk = newblk;
 391                                 break;
 392                         }
 393                         /*
 394                          * Entry wouldn't fit, split the leaf again.
 395                          */
 396                         state->extravalid = 1;
 397                         if (state->inleaf) {
 398                                 state->extraafter = 0;  /* before newblk */
 399                                 trace_xfs_attr_leaf_split_before(state->args);
 400                                 error = xfs_attr3_leaf_split(state, oldblk,
 401                                                             &state->extrablk);
 402                         } else {
 403                                 state->extraafter = 1;  /* after newblk */
 404                                 trace_xfs_attr_leaf_split_after(state->args);
 405                                 error = xfs_attr3_leaf_split(state, newblk,
 406                                                             &state->extrablk);
 407                         }
 408                         if (error)
 409                                 return error;   /* GROT: attr inconsistent */
 410                         addblk = newblk;
 411                         break;
 412                 case XFS_DIR2_LEAFN_MAGIC:
 413                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 414                         if (error)
 415                                 return error;
 416                         addblk = newblk;
 417                         break;
 418                 case XFS_DA_NODE_MAGIC:
 419                         error = xfs_da3_node_split(state, oldblk, newblk, addblk,
 420                                                          max - i, &action);
 421                         addblk->bp = NULL;
 422                         if (error)
 423                                 return error;   /* GROT: dir is inconsistent */
 424                         /*
 425                          * Record the newly split block for the next time thru?
 426                          */
 427                         if (action)
 428                                 addblk = newblk;
 429                         else
 430                                 addblk = NULL;
 431                         break;
 432                 }
 433
 434                 /*
 435                  * Update the btree to show the new hashval for this child.
 436                  */
 437                 xfs_da3_fixhashpath(state, &state->path);
 438         }
 439         if (!addblk)
 440                 return 0;
 441
 442         /*
 443          * Split the root node.
 444          */
 445         ASSERT(state->path.active == 0);
 446         oldblk = &state->path.blk[0];
 447         error = xfs_da3_root_split(state, oldblk, addblk);
 448         if (error) {
 449                 addblk->bp = NULL;
 450                 return error;   /* GROT: dir is inconsistent */
 451         }
 452
 453         /*
 454          * Update pointers to the node which used to be block 0 and
 455          * just got bumped because of the addition of a new root node.
 456          * There might be three blocks involved if a double split occurred,
 457          * and the original block 0 could be at any position in the list.
 458          *
 459          * Note: the magic numbers and sibling pointers are in the same
 460          * physical place for both v2 and v3 headers (by design). Hence it
 461          * doesn't matter which version of the xfs_da_intnode structure we use
 462          * here as the result will be the same using either structure.
 463          */
 464         node = oldblk->bp->b_addr;
 465         if (node->hdr.info.forw) {
 466                 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
 467                         bp = addblk->bp;
 468                 } else {
 469                         ASSERT(state->extravalid);
 470                         bp = state->extrablk.bp;
 471                 }
 472                 node = bp->b_addr;
 473                 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
 474                 xfs_trans_log_buf(state->args->trans, bp,
 475                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 476                     sizeof(node->hdr.info)));
 477         }
 478         node = oldblk->bp->b_addr;
 479         if (node->hdr.info.back) {
 480                 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
 481                         bp = addblk->bp;
 482                 } else {
 483                         ASSERT(state->extravalid);
 484                         bp = state->extrablk.bp;
 485                 }
 486                 node = bp->b_addr;
 487                 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
 488                 xfs_trans_log_buf(state->args->trans, bp,
 489                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 490                     sizeof(node->hdr.info)));
 491         }
 492         addblk->bp = NULL;
 493         return 0;
 494 }
 495
 496 /*
 497  * Split the root.  We have to create a new root and point to the two
 498  * parts (the split old root) that we just created.  Copy block zero to
 499  * the EOF, extending the inode in process.
 500  */
 501 STATIC int                                              /* error */
 502 xfs_da3_root_split(
 503         struct xfs_da_state     *state,
 504         struct xfs_da_state_blk *blk1,
 505         struct xfs_da_state_blk *blk2)
 506 {
 507         struct xfs_da_intnode   *node;
 508         struct xfs_da_intnode   *oldroot;
 509         struct xfs_da_node_entry *btree;
 510         struct xfs_da3_icnode_hdr nodehdr;
 511         struct xfs_da_args      *args;
 512         struct xfs_buf          *bp;
 513         struct xfs_inode        *dp;
 514         struct xfs_trans        *tp;
 515         struct xfs_dir2_leaf    *leaf;
 516         xfs_dablk_t             blkno;
 517         int                     level;
 518         int                     error;
 519         int                     size;
 520
 521         trace_xfs_da_root_split(state->args);
 522
 523         /*
 524          * Copy the existing (incorrect) block from the root node position
 525          * to a free space somewhere.
 526          */
 527         args = state->args;
 528         error = xfs_da_grow_inode(args, &blkno);
 529         if (error)
 530                 return error;
 531
 532         dp = args->dp;
 533         tp = args->trans;
 534         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 535         if (error)
 536                 return error;
 537         node = bp->b_addr;
 538         oldroot = blk1->bp->b_addr;
 539         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
 540             oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
 541                 struct xfs_da3_icnode_hdr icnodehdr;
 542
 543                 dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
 544                 btree = dp->d_ops->node_tree_p(oldroot);
 545                 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
 546                 level = icnodehdr.level;
 547
 548                 /*
 549                  * we are about to copy oldroot to bp, so set up the type
 550                  * of bp while we know exactly what it will be.
 551                  */
 552                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 553         } else {
 554                 struct xfs_dir3_icleaf_hdr leafhdr;
 555                 struct xfs_dir2_leaf_entry *ents;
 556
 557                 leaf = (xfs_dir2_leaf_t *)oldroot;
 558                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 559                 ents = dp->d_ops->leaf_ents_p(leaf);
 560
 561                 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 562                        leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 563                 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
 564                 level = 0;
 565
 566                 /*
 567                  * we are about to copy oldroot to bp, so set up the type
 568                  * of bp while we know exactly what it will be.
 569                  */
 570                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 571         }
 572
 573         /*
 574          * we can copy most of the information in the node from one block to
 575          * another, but for CRC enabled headers we have to make sure that the
 576          * block specific identifiers are kept intact. We update the buffer
 577          * directly for this.
 578          */
 579         memcpy(node, oldroot, size);
 580         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
 581             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 582                 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
 583
 584                 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
 585         }
 586         xfs_trans_log_buf(tp, bp, 0, size - 1);
 587
 588         bp->b_ops = blk1->bp->b_ops;
 589         xfs_trans_buf_copy_type(bp, blk1->bp);
 590         blk1->bp = bp;
 591         blk1->blkno = blkno;
 592
 593         /*
 594          * Set up the new root node.
 595          */
 596         error = xfs_da3_node_create(args,
 597                 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
 598                 level + 1, &bp, args->whichfork);
 599         if (error)
 600                 return error;
 601
 602         node = bp->b_addr;
 603         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 604         btree = dp->d_ops->node_tree_p(node);
 605         btree[0].hashval = cpu_to_be32(blk1->hashval);
 606         btree[0].before = cpu_to_be32(blk1->blkno);
 607         btree[1].hashval = cpu_to_be32(blk2->hashval);
 608         btree[1].before = cpu_to_be32(blk2->blkno);
 609         nodehdr.count = 2;
 610         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 611
 612 #ifdef DEBUG
 613         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
 614             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 615                 ASSERT(blk1->blkno >= args->geo->leafblk &&
 616                        blk1->blkno < args->geo->freeblk);
 617                 ASSERT(blk2->blkno >= args->geo->leafblk &&
 618                        blk2->blkno < args->geo->freeblk);
 619         }
 620 #endif
 621
 622         /* Header is already logged by xfs_da_node_create */
 623         xfs_trans_log_buf(tp, bp,
 624                 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
 625
 626         return 0;
 627 }
 628
 629 /*
 630  * Split the node, rebalance, then add the new entry.
 631  */
 632 STATIC int                                              /* error */
 633 xfs_da3_node_split(
 634         struct xfs_da_state     *state,
 635         struct xfs_da_state_blk *oldblk,
 636         struct xfs_da_state_blk *newblk,
 637         struct xfs_da_state_blk *addblk,
 638         int                     treelevel,
 639         int                     *result)
 640 {
 641         struct xfs_da_intnode   *node;
 642         struct xfs_da3_icnode_hdr nodehdr;
 643         xfs_dablk_t             blkno;
 644         int                     newcount;
 645         int                     error;
 646         int                     useextra;
 647         struct xfs_inode        *dp = state->args->dp;
 648
 649         trace_xfs_da_node_split(state->args);
 650
 651         node = oldblk->bp->b_addr;
 652         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 653
 654         /*
 655          * With V2 dirs the extra block is data or freespace.
 656          */
 657         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 658         newcount = 1 + useextra;
 659         /*
 660          * Do we have to split the node?
 661          */
 662         if (nodehdr.count + newcount > state->args->geo->node_ents) {
 663                 /*
 664                  * Allocate a new node, add to the doubly linked chain of
 665                  * nodes, then move some of our excess entries into it.
 666                  */
 667                 error = xfs_da_grow_inode(state->args, &blkno);
 668                 if (error)
 669                         return error;   /* GROT: dir is inconsistent */
 670
 671                 error = xfs_da3_node_create(state->args, blkno, treelevel,
 672                                            &newblk->bp, state->args->whichfork);
 673                 if (error)
 674                         return error;   /* GROT: dir is inconsistent */
 675                 newblk->blkno = blkno;
 676                 newblk->magic = XFS_DA_NODE_MAGIC;
 677                 xfs_da3_node_rebalance(state, oldblk, newblk);
 678                 error = xfs_da3_blk_link(state, oldblk, newblk);
 679                 if (error)
 680                         return error;
 681                 *result = 1;
 682         } else {
 683                 *result = 0;
 684         }
 685
 686         /*
 687          * Insert the new entry(s) into the correct block
 688          * (updating last hashval in the process).
 689          *
 690          * xfs_da3_node_add() inserts BEFORE the given index,
 691          * and as a result of using node_lookup_int() we always
 692          * point to a valid entry (not after one), but a split
 693          * operation always results in a new block whose hashvals
 694          * FOLLOW the current block.
 695          *
 696          * If we had double-split op below us, then add the extra block too.
 697          */
 698         node = oldblk->bp->b_addr;
 699         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 700         if (oldblk->index <= nodehdr.count) {
 701                 oldblk->index++;
 702                 xfs_da3_node_add(state, oldblk, addblk);
 703                 if (useextra) {
 704                         if (state->extraafter)
 705                                 oldblk->index++;
 706                         xfs_da3_node_add(state, oldblk, &state->extrablk);
 707                         state->extravalid = 0;
 708                 }
 709         } else {
 710                 newblk->index++;
 711                 xfs_da3_node_add(state, newblk, addblk);
 712                 if (useextra) {
 713                         if (state->extraafter)
 714                                 newblk->index++;
 715                         xfs_da3_node_add(state, newblk, &state->extrablk);
 716                         state->extravalid = 0;
 717                 }
 718         }
 719
 720         return 0;
 721 }
 722
 723 /*
 724  * Balance the btree elements between two intermediate nodes,
 725  * usually one full and one empty.
 726  *
 727  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 728  */
 729 STATIC void
 730 xfs_da3_node_rebalance(
 731         struct xfs_da_state     *state,
 732         struct xfs_da_state_blk *blk1,
 733         struct xfs_da_state_blk *blk2)
 734 {
 735         struct xfs_da_intnode   *node1;
 736         struct xfs_da_intnode   *node2;
 737         struct xfs_da_intnode   *tmpnode;
 738         struct xfs_da_node_entry *btree1;
 739         struct xfs_da_node_entry *btree2;
 740         struct xfs_da_node_entry *btree_s;
 741         struct xfs_da_node_entry *btree_d;
 742         struct xfs_da3_icnode_hdr nodehdr1;
 743         struct xfs_da3_icnode_hdr nodehdr2;
 744         struct xfs_trans        *tp;
 745         int                     count;
 746         int                     tmp;
 747         int                     swap = 0;
 748         struct xfs_inode        *dp = state->args->dp;
 749
 750         trace_xfs_da_node_rebalance(state->args);
 751
 752         node1 = blk1->bp->b_addr;
 753         node2 = blk2->bp->b_addr;
 754         dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 755         dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 756         btree1 = dp->d_ops->node_tree_p(node1);
 757         btree2 = dp->d_ops->node_tree_p(node2);
 758
 759         /*
 760          * Figure out how many entries need to move, and in which direction.
 761          * Swap the nodes around if that makes it simpler.
 762          */
 763         if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
 764             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
 765              (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
 766                         be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
 767                 tmpnode = node1;
 768                 node1 = node2;
 769                 node2 = tmpnode;
 770                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 771                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 772                 btree1 = dp->d_ops->node_tree_p(node1);
 773                 btree2 = dp->d_ops->node_tree_p(node2);
 774                 swap = 1;
 775         }
 776
 777         count = (nodehdr1.count - nodehdr2.count) / 2;
 778         if (count == 0)
 779                 return;
 780         tp = state->args->trans;
 781         /*
 782          * Two cases: high-to-low and low-to-high.
 783          */
 784         if (count > 0) {
 785                 /*
 786                  * Move elements in node2 up to make a hole.
 787                  */
 788                 tmp = nodehdr2.count;
 789                 if (tmp > 0) {
 790                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 791                         btree_s = &btree2[0];
 792                         btree_d = &btree2[count];
 793                         memmove(btree_d, btree_s, tmp);
 794                 }
 795
 796                 /*
 797                  * Move the req'd B-tree elements from high in node1 to
 798                  * low in node2.
 799                  */
 800                 nodehdr2.count += count;
 801                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 802                 btree_s = &btree1[nodehdr1.count - count];
 803                 btree_d = &btree2[0];
 804                 memcpy(btree_d, btree_s, tmp);
 805                 nodehdr1.count -= count;
 806         } else {
 807                 /*
 808                  * Move the req'd B-tree elements from low in node2 to
 809                  * high in node1.
 810                  */
 811                 count = -count;
 812                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 813                 btree_s = &btree2[0];
 814                 btree_d = &btree1[nodehdr1.count];
 815                 memcpy(btree_d, btree_s, tmp);
 816                 nodehdr1.count += count;
 817
 818                 xfs_trans_log_buf(tp, blk1->bp,
 819                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 820
 821                 /*
 822                  * Move elements in node2 down to fill the hole.
 823                  */
 824                 tmp  = nodehdr2.count - count;
 825                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 826                 btree_s = &btree2[count];
 827                 btree_d = &btree2[0];
 828                 memmove(btree_d, btree_s, tmp);
 829                 nodehdr2.count -= count;
 830         }
 831
 832         /*
 833          * Log header of node 1 and all current bits of node 2.
 834          */
 835         dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
 836         xfs_trans_log_buf(tp, blk1->bp,
 837                 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
 838
 839         dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
 840         xfs_trans_log_buf(tp, blk2->bp,
 841                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 842                                 dp->d_ops->node_hdr_size +
 843                                 (sizeof(btree2[0]) * nodehdr2.count)));
 844
 845         /*
 846          * Record the last hashval from each block for upward propagation.
 847          * (note: don't use the swapped node pointers)
 848          */
 849         if (swap) {
 850                 node1 = blk1->bp->b_addr;
 851                 node2 = blk2->bp->b_addr;
 852                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 853                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 854                 btree1 = dp->d_ops->node_tree_p(node1);
 855                 btree2 = dp->d_ops->node_tree_p(node2);
 856         }
 857         blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
 858         blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
 859
 860         /*
 861          * Adjust the expected index for insertion.
 862          */
 863         if (blk1->index >= nodehdr1.count) {
 864                 blk2->index = blk1->index - nodehdr1.count;
 865                 blk1->index = nodehdr1.count + 1;       /* make it invalid */
 866         }
 867 }
 868
 869 /*
 870  * Add a new entry to an intermediate node.
 871  */
 872 STATIC void
 873 xfs_da3_node_add(
 874         struct xfs_da_state     *state,
 875         struct xfs_da_state_blk *oldblk,
 876         struct xfs_da_state_blk *newblk)
 877 {
 878         struct xfs_da_intnode   *node;
 879         struct xfs_da3_icnode_hdr nodehdr;
 880         struct xfs_da_node_entry *btree;
 881         int                     tmp;
 882         struct xfs_inode        *dp = state->args->dp;
 883
 884         trace_xfs_da_node_add(state->args);
 885
 886         node = oldblk->bp->b_addr;
 887         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 888         btree = dp->d_ops->node_tree_p(node);
 889
 890         ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
 891         ASSERT(newblk->blkno != 0);
 892         if (state->args->whichfork == XFS_DATA_FORK)
 893                 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
 894                        newblk->blkno < state->args->geo->freeblk);
 895
 896         /*
 897          * We may need to make some room before we insert the new node.
 898          */
 899         tmp = 0;
 900         if (oldblk->index < nodehdr.count) {
 901                 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
 902                 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
 903         }
 904         btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
 905         btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
 906         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 907                 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
 908                                 tmp + sizeof(*btree)));
 909
 910         nodehdr.count += 1;
 911         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 912         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 913                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 914
 915         /*
 916          * Copy the last hash value from the oldblk to propagate upwards.
 917          */
 918         oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
 919 }
 920
 921 /*========================================================================
 922  * Routines used for shrinking the Btree.
 923  *========================================================================*/
 924
 925 /*
 926  * Deallocate an empty leaf node, remove it from its parent,
 927  * possibly deallocating that block, etc...
 928  */
 929 int
 930 xfs_da3_join(
 931         struct xfs_da_state     *state)
 932 {
 933         struct xfs_da_state_blk *drop_blk;
 934         struct xfs_da_state_blk *save_blk;
 935         int                     action = 0;
 936         int                     error;
 937
 938         trace_xfs_da_join(state->args);
 939
 940         drop_blk = &state->path.blk[ state->path.active-1 ];
 941         save_blk = &state->altpath.blk[ state->path.active-1 ];
 942         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 943         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 944                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 945
 946         /*
 947          * Walk back up the tree joining/deallocating as necessary.
 948          * When we stop dropping blocks, break out.
 949          */
 950         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 951                  state->path.active--) {
 952                 /*
 953                  * See if we can combine the block with a neighbor.
 954                  *   (action == 0) => no options, just leave
 955                  *   (action == 1) => coalesce, then unlink
 956                  *   (action == 2) => block empty, unlink it
 957                  */
 958                 switch (drop_blk->magic) {
 959                 case XFS_ATTR_LEAF_MAGIC:
 960                         error = xfs_attr3_leaf_toosmall(state, &action);
 961                         if (error)
 962                                 return error;
 963                         if (action == 0)
 964                                 return 0;
 965                         xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
 966                         break;
 967                 case XFS_DIR2_LEAFN_MAGIC:
 968                         error = xfs_dir2_leafn_toosmall(state, &action);
 969                         if (error)
 970                                 return error;
 971                         if (action == 0)
 972                                 return 0;
 973                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 974                         break;
 975                 case XFS_DA_NODE_MAGIC:
 976                         /*
 977                          * Remove the offending node, fixup hashvals,
 978                          * check for a toosmall neighbor.
 979                          */
 980                         xfs_da3_node_remove(state, drop_blk);
 981                         xfs_da3_fixhashpath(state, &state->path);
 982                         error = xfs_da3_node_toosmall(state, &action);
 983                         if (error)
 984                                 return error;
 985                         if (action == 0)
 986                                 return 0;
 987                         xfs_da3_node_unbalance(state, drop_blk, save_blk);
 988                         break;
 989                 }
 990                 xfs_da3_fixhashpath(state, &state->altpath);
 991                 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
 992                 xfs_da_state_kill_altpath(state);
 993                 if (error)
 994                         return error;
 995                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 996                                                          drop_blk->bp);
 997                 drop_blk->bp = NULL;
 998                 if (error)
 999                         return error;
1000         }
1001         /*
1002          * We joined all the way to the top.  If it turns out that
1003          * we only have one entry in the root, make the child block
1004          * the new root.
1005          */
1006         xfs_da3_node_remove(state, drop_blk);
1007         xfs_da3_fixhashpath(state, &state->path);
1008         error = xfs_da3_root_join(state, &state->path.blk[0]);
1009         return error;
1010 }
1011
1012 #ifdef  DEBUG
1013 static void
1014 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1015 {
1016         __be16  magic = blkinfo->magic;
1017
1018         if (level == 1) {
1019                 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1020                        magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1021                        magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1022                        magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1023         } else {
1024                 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1025                        magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1026         }
1027         ASSERT(!blkinfo->forw);
1028         ASSERT(!blkinfo->back);
1029 }
1030 #else   /* !DEBUG */
1031 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1032 #endif  /* !DEBUG */
1033
1034 /*
1035  * We have only one entry in the root.  Copy the only remaining child of
1036  * the old root to block 0 as the new root node.
1037  */
1038 STATIC int
1039 xfs_da3_root_join(
1040         struct xfs_da_state     *state,
1041         struct xfs_da_state_blk *root_blk)
1042 {
1043         struct xfs_da_intnode   *oldroot;
1044         struct xfs_da_args      *args;
1045         xfs_dablk_t             child;
1046         struct xfs_buf          *bp;
1047         struct xfs_da3_icnode_hdr oldroothdr;
1048         struct xfs_da_node_entry *btree;
1049         int                     error;
1050         struct xfs_inode        *dp = state->args->dp;
1051
1052         trace_xfs_da_root_join(state->args);
1053
1054         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1055
1056         args = state->args;
1057         oldroot = root_blk->bp->b_addr;
1058         dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1059         ASSERT(oldroothdr.forw == 0);
1060         ASSERT(oldroothdr.back == 0);
1061
1062         /*
1063          * If the root has more than one child, then don't do anything.
1064          */
1065         if (oldroothdr.count > 1)
1066                 return 0;
1067
1068         /*
1069          * Read in the (only) child block, then copy those bytes into
1070          * the root block's buffer and free the original child block.
1071          */
1072         btree = dp->d_ops->node_tree_p(oldroot);
1073         child = be32_to_cpu(btree[0].before);
1074         ASSERT(child != 0);
1075         error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1076                                              args->whichfork);
1077         if (error)
1078                 return error;
1079         xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1080
1081         /*
1082          * This could be copying a leaf back into the root block in the case of
1083          * there only being a single leaf block left in the tree. Hence we have
1084          * to update the b_ops pointer as well to match the buffer type change
1085          * that could occur. For dir3 blocks we also need to update the block
1086          * number in the buffer header.
1087          */
1088         memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1089         root_blk->bp->b_ops = bp->b_ops;
1090         xfs_trans_buf_copy_type(root_blk->bp, bp);
1091         if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1092                 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1093                 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1094         }
1095         xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1096                           args->geo->blksize - 1);
1097         error = xfs_da_shrink_inode(args, child, bp);
1098         return error;
1099 }
1100
1101 /*
1102  * Check a node block and its neighbors to see if the block should be
1103  * collapsed into one or the other neighbor.  Always keep the block
1104  * with the smaller block number.
1105  * If the current block is over 50% full, don't try to join it, return 0.
1106  * If the block is empty, fill in the state structure and return 2.
1107  * If it can be collapsed, fill in the state structure and return 1.
1108  * If nothing can be done, return 0.
1109  */
1110 STATIC int
1111 xfs_da3_node_toosmall(
1112         struct xfs_da_state     *state,
1113         int                     *action)
1114 {
1115         struct xfs_da_intnode   *node;
1116         struct xfs_da_state_blk *blk;
1117         struct xfs_da_blkinfo   *info;
1118         xfs_dablk_t             blkno;
1119         struct xfs_buf          *bp;
1120         struct xfs_da3_icnode_hdr nodehdr;
1121         int                     count;
1122         int                     forward;
1123         int                     error;
1124         int                     retval;
1125         int                     i;
1126         struct xfs_inode        *dp = state->args->dp;
1127
1128         trace_xfs_da_node_toosmall(state->args);
1129
1130         /*
1131          * Check for the degenerate case of the block being over 50% full.
1132          * If so, it's not worth even looking to see if we might be able
1133          * to coalesce with a sibling.
1134          */
1135         blk = &state->path.blk[ state->path.active-1 ];
1136         info = blk->bp->b_addr;
1137         node = (xfs_da_intnode_t *)info;
1138         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1139         if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1140                 *action = 0;    /* blk over 50%, don't try to join */
1141                 return 0;       /* blk over 50%, don't try to join */
1142         }
1143
1144         /*
1145          * Check for the degenerate case of the block being empty.
1146          * If the block is empty, we'll simply delete it, no need to
1147          * coalesce it with a sibling block.  We choose (arbitrarily)
1148          * to merge with the forward block unless it is NULL.
1149          */
1150         if (nodehdr.count == 0) {
1151                 /*
1152                  * Make altpath point to the block we want to keep and
1153                  * path point to the block we want to drop (this one).
1154                  */
1155                 forward = (info->forw != 0);
1156                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1157                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1158                                                  0, &retval);
1159                 if (error)
1160                         return error;
1161                 if (retval) {
1162                         *action = 0;
1163                 } else {
1164                         *action = 2;
1165                 }
1166                 return 0;
1167         }
1168
1169         /*
1170          * Examine each sibling block to see if we can coalesce with
1171          * at least 25% free space to spare.  We need to figure out
1172          * whether to merge with the forward or the backward block.
1173          * We prefer coalescing with the lower numbered sibling so as
1174          * to shrink a directory over time.
1175          */
1176         count  = state->args->geo->node_ents;
1177         count -= state->args->geo->node_ents >> 2;
1178         count -= nodehdr.count;
1179
1180         /* start with smaller blk num */
1181         forward = nodehdr.forw < nodehdr.back;
1182         for (i = 0; i < 2; forward = !forward, i++) {
1183                 struct xfs_da3_icnode_hdr thdr;
1184                 if (forward)
1185                         blkno = nodehdr.forw;
1186                 else
1187                         blkno = nodehdr.back;
1188                 if (blkno == 0)
1189                         continue;
1190                 error = xfs_da3_node_read(state->args->trans, dp,
1191                                         blkno, -1, &bp, state->args->whichfork);
1192                 if (error)
1193                         return error;
1194
1195                 node = bp->b_addr;
1196                 dp->d_ops->node_hdr_from_disk(&thdr, node);
1197                 xfs_trans_brelse(state->args->trans, bp);
1198
1199                 if (count - thdr.count >= 0)
1200                         break;  /* fits with at least 25% to spare */
1201         }
1202         if (i >= 2) {
1203                 *action = 0;
1204                 return 0;
1205         }
1206
1207         /*
1208          * Make altpath point to the block we want to keep (the lower
1209          * numbered block) and path point to the block we want to drop.
1210          */
1211         memcpy(&state->altpath, &state->path, sizeof(state->path));
1212         if (blkno < blk->blkno) {
1213                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1214                                                  0, &retval);
1215         } else {
1216                 error = xfs_da3_path_shift(state, &state->path, forward,
1217                                                  0, &retval);
1218         }
1219         if (error)
1220                 return error;
1221         if (retval) {
1222                 *action = 0;
1223                 return 0;
1224         }
1225         *action = 1;
1226         return 0;
1227 }
1228
1229 /*
1230  * Pick up the last hashvalue from an intermediate node.
1231  */
1232 STATIC uint
1233 xfs_da3_node_lasthash(
1234         struct xfs_inode        *dp,
1235         struct xfs_buf          *bp,
1236         int                     *count)
1237 {
1238         struct xfs_da_intnode    *node;
1239         struct xfs_da_node_entry *btree;
1240         struct xfs_da3_icnode_hdr nodehdr;
1241
1242         node = bp->b_addr;
1243         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1244         if (count)
1245                 *count = nodehdr.count;
1246         if (!nodehdr.count)
1247                 return 0;
1248         btree = dp->d_ops->node_tree_p(node);
1249         return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1250 }
1251
1252 /*
1253  * Walk back up the tree adjusting hash values as necessary,
1254  * when we stop making changes, return.
1255  */
1256 void
1257 xfs_da3_fixhashpath(
1258         struct xfs_da_state     *state,
1259         struct xfs_da_state_path *path)
1260 {
1261         struct xfs_da_state_blk *blk;
1262         struct xfs_da_intnode   *node;
1263         struct xfs_da_node_entry *btree;
1264         xfs_dahash_t            lasthash=0;
1265         int                     level;
1266         int                     count;
1267         struct xfs_inode        *dp = state->args->dp;
1268
1269         trace_xfs_da_fixhashpath(state->args);
1270
1271         level = path->active-1;
1272         blk = &path->blk[ level ];
1273         switch (blk->magic) {
1274         case XFS_ATTR_LEAF_MAGIC:
1275                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1276                 if (count == 0)
1277                         return;
1278                 break;
1279         case XFS_DIR2_LEAFN_MAGIC:
1280                 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1281                 if (count == 0)
1282                         return;
1283                 break;
1284         case XFS_DA_NODE_MAGIC:
1285                 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1286                 if (count == 0)
1287                         return;
1288                 break;
1289         }
1290         for (blk--, level--; level >= 0; blk--, level--) {
1291                 struct xfs_da3_icnode_hdr nodehdr;
1292
1293                 node = blk->bp->b_addr;
1294                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1295                 btree = dp->d_ops->node_tree_p(node);
1296                 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1297                         break;
1298                 blk->hashval = lasthash;
1299                 btree[blk->index].hashval = cpu_to_be32(lasthash);
1300                 xfs_trans_log_buf(state->args->trans, blk->bp,
1301                                   XFS_DA_LOGRANGE(node, &btree[blk->index],
1302                                                   sizeof(*btree)));
1303
1304                 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1305         }
1306 }
1307
1308 /*
1309  * Remove an entry from an intermediate node.
1310  */
1311 STATIC void
1312 xfs_da3_node_remove(
1313         struct xfs_da_state     *state,
1314         struct xfs_da_state_blk *drop_blk)
1315 {
1316         struct xfs_da_intnode   *node;
1317         struct xfs_da3_icnode_hdr nodehdr;
1318         struct xfs_da_node_entry *btree;
1319         int                     index;
1320         int                     tmp;
1321         struct xfs_inode        *dp = state->args->dp;
1322
1323         trace_xfs_da_node_remove(state->args);
1324
1325         node = drop_blk->bp->b_addr;
1326         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1327         ASSERT(drop_blk->index < nodehdr.count);
1328         ASSERT(drop_blk->index >= 0);
1329
1330         /*
1331          * Copy over the offending entry, or just zero it out.
1332          */
1333         index = drop_blk->index;
1334         btree = dp->d_ops->node_tree_p(node);
1335         if (index < nodehdr.count - 1) {
1336                 tmp  = nodehdr.count - index - 1;
1337                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1338                 memmove(&btree[index], &btree[index + 1], tmp);
1339                 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1340                     XFS_DA_LOGRANGE(node, &btree[index], tmp));
1341                 index = nodehdr.count - 1;
1342         }
1343         memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1344         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1345             XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1346         nodehdr.count -= 1;
1347         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1348         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1349             XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1350
1351         /*
1352          * Copy the last hash value from the block to propagate upwards.
1353          */
1354         drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1355 }
1356
1357 /*
1358  * Unbalance the elements between two intermediate nodes,
1359  * move all Btree elements from one node into another.
1360  */
1361 STATIC void
1362 xfs_da3_node_unbalance(
1363         struct xfs_da_state     *state,
1364         struct xfs_da_state_blk *drop_blk,
1365         struct xfs_da_state_blk *save_blk)
1366 {
1367         struct xfs_da_intnode   *drop_node;
1368         struct xfs_da_intnode   *save_node;
1369         struct xfs_da_node_entry *drop_btree;
1370         struct xfs_da_node_entry *save_btree;
1371         struct xfs_da3_icnode_hdr drop_hdr;
1372         struct xfs_da3_icnode_hdr save_hdr;
1373         struct xfs_trans        *tp;
1374         int                     sindex;
1375         int                     tmp;
1376         struct xfs_inode        *dp = state->args->dp;
1377
1378         trace_xfs_da_node_unbalance(state->args);
1379
1380         drop_node = drop_blk->bp->b_addr;
1381         save_node = save_blk->bp->b_addr;
1382         dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1383         dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1384         drop_btree = dp->d_ops->node_tree_p(drop_node);
1385         save_btree = dp->d_ops->node_tree_p(save_node);
1386         tp = state->args->trans;
1387
1388         /*
1389          * If the dying block has lower hashvals, then move all the
1390          * elements in the remaining block up to make a hole.
1391          */
1392         if ((be32_to_cpu(drop_btree[0].hashval) <
1393                         be32_to_cpu(save_btree[0].hashval)) ||
1394             (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1395                         be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1396                 /* XXX: check this - is memmove dst correct? */
1397                 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1398                 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1399
1400                 sindex = 0;
1401                 xfs_trans_log_buf(tp, save_blk->bp,
1402                         XFS_DA_LOGRANGE(save_node, &save_btree[0],
1403                                 (save_hdr.count + drop_hdr.count) *
1404                                                 sizeof(xfs_da_node_entry_t)));
1405         } else {
1406                 sindex = save_hdr.count;
1407                 xfs_trans_log_buf(tp, save_blk->bp,
1408                         XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1409                                 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1410         }
1411
1412         /*
1413          * Move all the B-tree elements from drop_blk to save_blk.
1414          */
1415         tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1416         memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1417         save_hdr.count += drop_hdr.count;
1418
1419         dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1420         xfs_trans_log_buf(tp, save_blk->bp,
1421                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1422                                 dp->d_ops->node_hdr_size));
1423
1424         /*
1425          * Save the last hashval in the remaining block for upward propagation.
1426          */
1427         save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1428 }
1429
1430 /*========================================================================
1431  * Routines used for finding things in the Btree.
1432  *========================================================================*/
1433
1434 /*
1435  * Walk down the Btree looking for a particular filename, filling
1436  * in the state structure as we go.
1437  *
1438  * We will set the state structure to point to each of the elements
1439  * in each of the nodes where either the hashval is or should be.
1440  *
1441  * We support duplicate hashval's so for each entry in the current
1442  * node that could contain the desired hashval, descend.  This is a
1443  * pruned depth-first tree search.
1444  */
1445 int                                                     /* error */
1446 xfs_da3_node_lookup_int(
1447         struct xfs_da_state     *state,
1448         int                     *result)
1449 {
1450         struct xfs_da_state_blk *blk;
1451         struct xfs_da_blkinfo   *curr;
1452         struct xfs_da_intnode   *node;
1453         struct xfs_da_node_entry *btree;
1454         struct xfs_da3_icnode_hdr nodehdr;
1455         struct xfs_da_args      *args;
1456         xfs_dablk_t             blkno;
1457         xfs_dahash_t            hashval;
1458         xfs_dahash_t            btreehashval;
1459         int                     probe;
1460         int                     span;
1461         int                     max;
1462         int                     error;
1463         int                     retval;
1464         struct xfs_inode        *dp = state->args->dp;
1465
1466         args = state->args;
1467
1468         /*
1469          * Descend thru the B-tree searching each level for the right
1470          * node to use, until the right hashval is found.
1471          */
1472         blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1473         for (blk = &state->path.blk[0], state->path.active = 1;
1474                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1475                          blk++, state->path.active++) {
1476                 /*
1477                  * Read the next node down in the tree.
1478                  */
1479                 blk->blkno = blkno;
1480                 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1481                                         -1, &blk->bp, args->whichfork);
1482                 if (error) {
1483                         blk->blkno = 0;
1484                         state->path.active--;
1485                         return error;
1486                 }
1487                 curr = blk->bp->b_addr;
1488                 blk->magic = be16_to_cpu(curr->magic);
1489
1490                 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1491                     blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1492                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1493                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1494                         break;
1495                 }
1496
1497                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1498                     blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1499                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1500                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1501                                                                blk->bp, NULL);
1502                         break;
1503                 }
1504
1505                 blk->magic = XFS_DA_NODE_MAGIC;
1506
1507
1508                 /*
1509                  * Search an intermediate node for a match.
1510                  */
1511                 node = blk->bp->b_addr;
1512                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1513                 btree = dp->d_ops->node_tree_p(node);
1514
1515                 max = nodehdr.count;
1516                 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1517
1518                 /*
1519                  * Binary search.  (note: small blocks will skip loop)
1520                  */
1521                 probe = span = max / 2;
1522                 hashval = args->hashval;
1523                 while (span > 4) {
1524                         span /= 2;
1525                         btreehashval = be32_to_cpu(btree[probe].hashval);
1526                         if (btreehashval < hashval)
1527                                 probe += span;
1528                         else if (btreehashval > hashval)
1529                                 probe -= span;
1530                         else
1531                                 break;
1532                 }
1533                 ASSERT((probe >= 0) && (probe < max));
1534                 ASSERT((span <= 4) ||
1535                         (be32_to_cpu(btree[probe].hashval) == hashval));
1536
1537                 /*
1538                  * Since we may have duplicate hashval's, find the first
1539                  * matching hashval in the node.
1540                  */
1541                 while (probe > 0 &&
1542                        be32_to_cpu(btree[probe].hashval) >= hashval) {
1543                         probe--;
1544                 }
1545                 while (probe < max &&
1546                        be32_to_cpu(btree[probe].hashval) < hashval) {
1547                         probe++;
1548                 }
1549
1550                 /*
1551                  * Pick the right block to descend on.
1552                  */
1553                 if (probe == max) {
1554                         blk->index = max - 1;
1555                         blkno = be32_to_cpu(btree[max - 1].before);
1556                 } else {
1557                         blk->index = probe;
1558                         blkno = be32_to_cpu(btree[probe].before);
1559                 }
1560         }
1561
1562         /*
1563          * A leaf block that ends in the hashval that we are interested in
1564          * (final hashval == search hashval) means that the next block may
1565          * contain more entries with the same hashval, shift upward to the
1566          * next leaf and keep searching.
1567          */
1568         for (;;) {
1569                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1570                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1571                                                         &blk->index, state);
1572                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1573                         retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1574                         blk->index = args->index;
1575                         args->blkno = blk->blkno;
1576                 } else {
1577                         ASSERT(0);
1578                         return -EFSCORRUPTED;
1579                 }
1580                 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1581                     (blk->hashval == args->hashval)) {
1582                         error = xfs_da3_path_shift(state, &state->path, 1, 1,
1583                                                          &retval);
1584                         if (error)
1585                                 return error;
1586                         if (retval == 0) {
1587                                 continue;
1588                         } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1589                                 /* path_shift() gives ENOENT */
1590                                 retval = -ENOATTR;
1591                         }
1592                 }
1593                 break;
1594         }
1595         *result = retval;
1596         return 0;
1597 }
1598
1599 /*========================================================================
1600  * Utility routines.
1601  *========================================================================*/
1602
1603 /*
1604  * Compare two intermediate nodes for "order".
1605  */
1606 STATIC int
1607 xfs_da3_node_order(
1608         struct xfs_inode *dp,
1609         struct xfs_buf  *node1_bp,
1610         struct xfs_buf  *node2_bp)
1611 {
1612         struct xfs_da_intnode   *node1;
1613         struct xfs_da_intnode   *node2;
1614         struct xfs_da_node_entry *btree1;
1615         struct xfs_da_node_entry *btree2;
1616         struct xfs_da3_icnode_hdr node1hdr;
1617         struct xfs_da3_icnode_hdr node2hdr;
1618
1619         node1 = node1_bp->b_addr;
1620         node2 = node2_bp->b_addr;
1621         dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1622         dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1623         btree1 = dp->d_ops->node_tree_p(node1);
1624         btree2 = dp->d_ops->node_tree_p(node2);
1625
1626         if (node1hdr.count > 0 && node2hdr.count > 0 &&
1627             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1628              (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1629               be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1630                 return 1;
1631         }
1632         return 0;
1633 }
1634
1635 /*
1636  * Link a new block into a doubly linked list of blocks (of whatever type).
1637  */
1638 int                                                     /* error */
1639 xfs_da3_blk_link(
1640         struct xfs_da_state     *state,
1641         struct xfs_da_state_blk *old_blk,
1642         struct xfs_da_state_blk *new_blk)
1643 {
1644         struct xfs_da_blkinfo   *old_info;
1645         struct xfs_da_blkinfo   *new_info;
1646         struct xfs_da_blkinfo   *tmp_info;
1647         struct xfs_da_args      *args;
1648         struct xfs_buf          *bp;
1649         int                     before = 0;
1650         int                     error;
1651         struct xfs_inode        *dp = state->args->dp;
1652
1653         /*
1654          * Set up environment.
1655          */
1656         args = state->args;
1657         ASSERT(args != NULL);
1658         old_info = old_blk->bp->b_addr;
1659         new_info = new_blk->bp->b_addr;
1660         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1661                old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1662                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1663
1664         switch (old_blk->magic) {
1665         case XFS_ATTR_LEAF_MAGIC:
1666                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1667                 break;
1668         case XFS_DIR2_LEAFN_MAGIC:
1669                 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1670                 break;
1671         case XFS_DA_NODE_MAGIC:
1672                 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1673                 break;
1674         }
1675
1676         /*
1677          * Link blocks in appropriate order.
1678          */
1679         if (before) {
1680                 /*
1681                  * Link new block in before existing block.
1682                  */
1683                 trace_xfs_da_link_before(args);
1684                 new_info->forw = cpu_to_be32(old_blk->blkno);
1685                 new_info->back = old_info->back;
1686                 if (old_info->back) {
1687                         error = xfs_da3_node_read(args->trans, dp,
1688                                                 be32_to_cpu(old_info->back),
1689                                                 -1, &bp, args->whichfork);
1690                         if (error)
1691                                 return error;
1692                         ASSERT(bp != NULL);
1693                         tmp_info = bp->b_addr;
1694                         ASSERT(tmp_info->magic == old_info->magic);
1695                         ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1696                         tmp_info->forw = cpu_to_be32(new_blk->blkno);
1697                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1698                 }
1699                 old_info->back = cpu_to_be32(new_blk->blkno);
1700         } else {
1701                 /*
1702                  * Link new block in after existing block.
1703                  */
1704                 trace_xfs_da_link_after(args);
1705                 new_info->forw = old_info->forw;
1706                 new_info->back = cpu_to_be32(old_blk->blkno);
1707                 if (old_info->forw) {
1708                         error = xfs_da3_node_read(args->trans, dp,
1709                                                 be32_to_cpu(old_info->forw),
1710                                                 -1, &bp, args->whichfork);
1711                         if (error)
1712                                 return error;
1713                         ASSERT(bp != NULL);
1714                         tmp_info = bp->b_addr;
1715                         ASSERT(tmp_info->magic == old_info->magic);
1716                         ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1717                         tmp_info->back = cpu_to_be32(new_blk->blkno);
1718                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1719                 }
1720                 old_info->forw = cpu_to_be32(new_blk->blkno);
1721         }
1722
1723         xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1724         xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1725         return 0;
1726 }
1727
1728 /*
1729  * Unlink a block from a doubly linked list of blocks.
1730  */
1731 STATIC int                                              /* error */
1732 xfs_da3_blk_unlink(
1733         struct xfs_da_state     *state,
1734         struct xfs_da_state_blk *drop_blk,
1735         struct xfs_da_state_blk *save_blk)
1736 {
1737         struct xfs_da_blkinfo   *drop_info;
1738         struct xfs_da_blkinfo   *save_info;
1739         struct xfs_da_blkinfo   *tmp_info;
1740         struct xfs_da_args      *args;
1741         struct xfs_buf          *bp;
1742         int                     error;
1743
1744         /*
1745          * Set up environment.
1746          */
1747         args = state->args;
1748         ASSERT(args != NULL);
1749         save_info = save_blk->bp->b_addr;
1750         drop_info = drop_blk->bp->b_addr;
1751         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1752                save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1753                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1754         ASSERT(save_blk->magic == drop_blk->magic);
1755         ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1756                (be32_to_cpu(save_info->back) == drop_blk->blkno));
1757         ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1758                (be32_to_cpu(drop_info->back) == save_blk->blkno));
1759
1760         /*
1761          * Unlink the leaf block from the doubly linked chain of leaves.
1762          */
1763         if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1764                 trace_xfs_da_unlink_back(args);
1765                 save_info->back = drop_info->back;
1766                 if (drop_info->back) {
1767                         error = xfs_da3_node_read(args->trans, args->dp,
1768                                                 be32_to_cpu(drop_info->back),
1769                                                 -1, &bp, args->whichfork);
1770                         if (error)
1771                                 return error;
1772                         ASSERT(bp != NULL);
1773                         tmp_info = bp->b_addr;
1774                         ASSERT(tmp_info->magic == save_info->magic);
1775                         ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1776                         tmp_info->forw = cpu_to_be32(save_blk->blkno);
1777                         xfs_trans_log_buf(args->trans, bp, 0,
1778                                                     sizeof(*tmp_info) - 1);
1779                 }
1780         } else {
1781                 trace_xfs_da_unlink_forward(args);
1782                 save_info->forw = drop_info->forw;
1783                 if (drop_info->forw) {
1784                         error = xfs_da3_node_read(args->trans, args->dp,
1785                                                 be32_to_cpu(drop_info->forw),
1786                                                 -1, &bp, args->whichfork);
1787                         if (error)
1788                                 return error;
1789                         ASSERT(bp != NULL);
1790                         tmp_info = bp->b_addr;
1791                         ASSERT(tmp_info->magic == save_info->magic);
1792                         ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1793                         tmp_info->back = cpu_to_be32(save_blk->blkno);
1794                         xfs_trans_log_buf(args->trans, bp, 0,
1795                                                     sizeof(*tmp_info) - 1);
1796                 }
1797         }
1798
1799         xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1800         return 0;
1801 }
1802
1803 /*
1804  * Move a path "forward" or "!forward" one block at the current level.
1805  *
1806  * This routine will adjust a "path" to point to the next block
1807  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1808  * Btree, including updating pointers to the intermediate nodes between
1809  * the new bottom and the root.
1810  */
1811 int                                                     /* error */
1812 xfs_da3_path_shift(
1813         struct xfs_da_state     *state,
1814         struct xfs_da_state_path *path,
1815         int                     forward,
1816         int                     release,
1817         int                     *result)
1818 {
1819         struct xfs_da_state_blk *blk;
1820         struct xfs_da_blkinfo   *info;
1821         struct xfs_da_intnode   *node;
1822         struct xfs_da_args      *args;
1823         struct xfs_da_node_entry *btree;
1824         struct xfs_da3_icnode_hdr nodehdr;
1825         xfs_dablk_t             blkno = 0;
1826         int                     level;
1827         int                     error;
1828         struct xfs_inode        *dp = state->args->dp;
1829
1830         trace_xfs_da_path_shift(state->args);
1831
1832         /*
1833          * Roll up the Btree looking for the first block where our
1834          * current index is not at the edge of the block.  Note that
1835          * we skip the bottom layer because we want the sibling block.
1836          */
1837         args = state->args;
1838         ASSERT(args != NULL);
1839         ASSERT(path != NULL);
1840         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1841         level = (path->active-1) - 1;   /* skip bottom layer in path */
1842         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1843                 node = blk->bp->b_addr;
1844                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1845                 btree = dp->d_ops->node_tree_p(node);
1846
1847                 if (forward && (blk->index < nodehdr.count - 1)) {
1848                         blk->index++;
1849                         blkno = be32_to_cpu(btree[blk->index].before);
1850                         break;
1851                 } else if (!forward && (blk->index > 0)) {
1852                         blk->index--;
1853                         blkno = be32_to_cpu(btree[blk->index].before);
1854                         break;
1855                 }
1856         }
1857         if (level < 0) {
1858                 *result = -ENOENT;      /* we're out of our tree */
1859                 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1860                 return 0;
1861         }
1862
1863         /*
1864          * Roll down the edge of the subtree until we reach the
1865          * same depth we were at originally.
1866          */
1867         for (blk++, level++; level < path->active; blk++, level++) {
1868                 /*
1869                  * Release the old block.
1870                  * (if it's dirty, trans won't actually let go)
1871                  */
1872                 if (release)
1873                         xfs_trans_brelse(args->trans, blk->bp);
1874
1875                 /*
1876                  * Read the next child block.
1877                  */
1878                 blk->blkno = blkno;
1879                 error = xfs_da3_node_read(args->trans, dp, blkno, -1,
1880                                         &blk->bp, args->whichfork);
1881                 if (error)
1882                         return error;
1883                 info = blk->bp->b_addr;
1884                 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1885                        info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1886                        info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1887                        info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1888                        info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1889                        info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1890
1891
1892                 /*
1893                  * Note: we flatten the magic number to a single type so we
1894                  * don't have to compare against crc/non-crc types elsewhere.
1895                  */
1896                 switch (be16_to_cpu(info->magic)) {
1897                 case XFS_DA_NODE_MAGIC:
1898                 case XFS_DA3_NODE_MAGIC:
1899                         blk->magic = XFS_DA_NODE_MAGIC;
1900                         node = (xfs_da_intnode_t *)info;
1901                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1902                         btree = dp->d_ops->node_tree_p(node);
1903                         blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1904                         if (forward)
1905                                 blk->index = 0;
1906                         else
1907                                 blk->index = nodehdr.count - 1;
1908                         blkno = be32_to_cpu(btree[blk->index].before);
1909                         break;
1910                 case XFS_ATTR_LEAF_MAGIC:
1911                 case XFS_ATTR3_LEAF_MAGIC:
1912                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1913                         ASSERT(level == path->active-1);
1914                         blk->index = 0;
1915                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1916                         break;
1917                 case XFS_DIR2_LEAFN_MAGIC:
1918                 case XFS_DIR3_LEAFN_MAGIC:
1919                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1920                         ASSERT(level == path->active-1);
1921                         blk->index = 0;
1922                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1923                                                                blk->bp, NULL);
1924                         break;
1925                 default:
1926                         ASSERT(0);
1927                         break;
1928                 }
1929         }
1930         *result = 0;
1931         return 0;
1932 }
1933
1934
1935 /*========================================================================
1936  * Utility routines.
1937  *========================================================================*/
1938
1939 /*
1940  * Implement a simple hash on a character string.
1941  * Rotate the hash value by 7 bits, then XOR each character in.
1942  * This is implemented with some source-level loop unrolling.
1943  */
1944 xfs_dahash_t
1945 xfs_da_hashname(const __uint8_t *name, int namelen)
1946 {
1947         xfs_dahash_t hash;
1948
1949         /*
1950          * Do four characters at a time as long as we can.
1951          */
1952         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1953                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1954                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1955
1956         /*
1957          * Now do the rest of the characters.
1958          */
1959         switch (namelen) {
1960         case 3:
1961                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1962                        rol32(hash, 7 * 3);
1963         case 2:
1964                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1965         case 1:
1966                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1967         default: /* case 0: */
1968                 return hash;
1969         }
1970 }
1971
1972 enum xfs_dacmp
1973 xfs_da_compname(
1974         struct xfs_da_args *args,
1975         const unsigned char *name,
1976         int             len)
1977 {
1978         return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1979                                         XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1980 }
1981
1982 static xfs_dahash_t
1983 xfs_default_hashname(
1984         struct xfs_name *name)
1985 {
1986         return xfs_da_hashname(name->name, name->len);
1987 }
1988
1989 const struct xfs_nameops xfs_default_nameops = {
1990         .hashname       = xfs_default_hashname,
1991         .compname       = xfs_da_compname
1992 };
1993
1994 int
1995 xfs_da_grow_inode_int(
1996         struct xfs_da_args      *args,
1997         xfs_fileoff_t           *bno,
1998         int                     count)
1999 {
2000         struct xfs_trans        *tp = args->trans;
2001         struct xfs_inode        *dp = args->dp;
2002         int                     w = args->whichfork;
2003         xfs_rfsblock_t          nblks = dp->i_d.di_nblocks;
2004         struct xfs_bmbt_irec    map, *mapp;
2005         int                     nmap, error, got, i, mapi;
2006
2007         /*
2008          * Find a spot in the file space to put the new block.
2009          */
2010         error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2011         if (error)
2012                 return error;
2013
2014         /*
2015          * Try mapping it in one filesystem block.
2016          */
2017         nmap = 1;
2018         ASSERT(args->firstblock != NULL);
2019         error = xfs_bmapi_write(tp, dp, *bno, count,
2020                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2021                         args->firstblock, args->total, &map, &nmap,
2022                         args->flist);
2023         if (error)
2024                 return error;
2025
2026         ASSERT(nmap <= 1);
2027         if (nmap == 1) {
2028                 mapp = &map;
2029                 mapi = 1;
2030         } else if (nmap == 0 && count > 1) {
2031                 xfs_fileoff_t           b;
2032                 int                     c;
2033
2034                 /*
2035                  * If we didn't get it and the block might work if fragmented,
2036                  * try without the CONTIG flag.  Loop until we get it all.
2037                  */
2038                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2039                 for (b = *bno, mapi = 0; b < *bno + count; ) {
2040                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2041                         c = (int)(*bno + count - b);
2042                         error = xfs_bmapi_write(tp, dp, b, c,
2043                                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2044                                         args->firstblock, args->total,
2045                                         &mapp[mapi], &nmap, args->flist);
2046                         if (error)
2047                                 goto out_free_map;
2048                         if (nmap < 1)
2049                                 break;
2050                         mapi += nmap;
2051                         b = mapp[mapi - 1].br_startoff +
2052                             mapp[mapi - 1].br_blockcount;
2053                 }
2054         } else {
2055                 mapi = 0;
2056                 mapp = NULL;
2057         }
2058
2059         /*
2060          * Count the blocks we got, make sure it matches the total.
2061          */
2062         for (i = 0, got = 0; i < mapi; i++)
2063                 got += mapp[i].br_blockcount;
2064         if (got != count || mapp[0].br_startoff != *bno ||
2065             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2066             *bno + count) {
2067                 error = -ENOSPC;
2068                 goto out_free_map;
2069         }
2070
2071         /* account for newly allocated blocks in reserved blocks total */
2072         args->total -= dp->i_d.di_nblocks - nblks;
2073
2074 out_free_map:
2075         if (mapp != &map)
2076                 kmem_free(mapp);
2077         return error;
2078 }
2079
2080 /*
2081  * Add a block to the btree ahead of the file.
2082  * Return the new block number to the caller.
2083  */
2084 int
2085 xfs_da_grow_inode(
2086         struct xfs_da_args      *args,
2087         xfs_dablk_t             *new_blkno)
2088 {
2089         xfs_fileoff_t           bno;
2090         int                     error;
2091
2092         trace_xfs_da_grow_inode(args);
2093
2094         bno = args->geo->leafblk;
2095         error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2096         if (!error)
2097                 *new_blkno = (xfs_dablk_t)bno;
2098         return error;
2099 }
2100
2101 /*
2102  * Ick.  We need to always be able to remove a btree block, even
2103  * if there's no space reservation because the filesystem is full.
2104  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2105  * It swaps the target block with the last block in the file.  The
2106  * last block in the file can always be removed since it can't cause
2107  * a bmap btree split to do that.
2108  */
2109 STATIC int
2110 xfs_da3_swap_lastblock(
2111         struct xfs_da_args      *args,
2112         xfs_dablk_t             *dead_blknop,
2113         struct xfs_buf          **dead_bufp)
2114 {
2115         struct xfs_da_blkinfo   *dead_info;
2116         struct xfs_da_blkinfo   *sib_info;
2117         struct xfs_da_intnode   *par_node;
2118         struct xfs_da_intnode   *dead_node;
2119         struct xfs_dir2_leaf    *dead_leaf2;
2120         struct xfs_da_node_entry *btree;
2121         struct xfs_da3_icnode_hdr par_hdr;
2122         struct xfs_inode        *dp;
2123         struct xfs_trans        *tp;
2124         struct xfs_mount        *mp;
2125         struct xfs_buf          *dead_buf;
2126         struct xfs_buf          *last_buf;
2127         struct xfs_buf          *sib_buf;
2128         struct xfs_buf          *par_buf;
2129         xfs_dahash_t            dead_hash;
2130         xfs_fileoff_t           lastoff;
2131         xfs_dablk_t             dead_blkno;
2132         xfs_dablk_t             last_blkno;
2133         xfs_dablk_t             sib_blkno;
2134         xfs_dablk_t             par_blkno;
2135         int                     error;
2136         int                     w;
2137         int                     entno;
2138         int                     level;
2139         int                     dead_level;
2140
2141         trace_xfs_da_swap_lastblock(args);
2142
2143         dead_buf = *dead_bufp;
2144         dead_blkno = *dead_blknop;
2145         tp = args->trans;
2146         dp = args->dp;
2147         w = args->whichfork;
2148         ASSERT(w == XFS_DATA_FORK);
2149         mp = dp->i_mount;
2150         lastoff = args->geo->freeblk;
2151         error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2152         if (error)
2153                 return error;
2154         if (unlikely(lastoff == 0)) {
2155                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2156                                  mp);
2157                 return -EFSCORRUPTED;
2158         }
2159         /*
2160          * Read the last block in the btree space.
2161          */
2162         last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2163         error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2164         if (error)
2165                 return error;
2166         /*
2167          * Copy the last block into the dead buffer and log it.
2168          */
2169         memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2170         xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2171         dead_info = dead_buf->b_addr;
2172         /*
2173          * Get values from the moved block.
2174          */
2175         if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2176             dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2177                 struct xfs_dir3_icleaf_hdr leafhdr;
2178                 struct xfs_dir2_leaf_entry *ents;
2179
2180                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2181                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2182                 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2183                 dead_level = 0;
2184                 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2185         } else {
2186                 struct xfs_da3_icnode_hdr deadhdr;
2187
2188                 dead_node = (xfs_da_intnode_t *)dead_info;
2189                 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2190                 btree = dp->d_ops->node_tree_p(dead_node);
2191                 dead_level = deadhdr.level;
2192                 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2193         }
2194         sib_buf = par_buf = NULL;
2195         /*
2196          * If the moved block has a left sibling, fix up the pointers.
2197          */
2198         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2199                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2200                 if (error)
2201                         goto done;
2202                 sib_info = sib_buf->b_addr;
2203                 if (unlikely(
2204                     be32_to_cpu(sib_info->forw) != last_blkno ||
2205                     sib_info->magic != dead_info->magic)) {
2206                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2207                                          XFS_ERRLEVEL_LOW, mp);
2208                         error = -EFSCORRUPTED;
2209                         goto done;
2210                 }
2211                 sib_info->forw = cpu_to_be32(dead_blkno);
2212                 xfs_trans_log_buf(tp, sib_buf,
2213                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2214                                         sizeof(sib_info->forw)));
2215                 sib_buf = NULL;
2216         }
2217         /*
2218          * If the moved block has a right sibling, fix up the pointers.
2219          */
2220         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2221                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2222                 if (error)
2223                         goto done;
2224                 sib_info = sib_buf->b_addr;
2225                 if (unlikely(
2226                        be32_to_cpu(sib_info->back) != last_blkno ||
2227                        sib_info->magic != dead_info->magic)) {
2228                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2229                                          XFS_ERRLEVEL_LOW, mp);
2230                         error = -EFSCORRUPTED;
2231                         goto done;
2232                 }
2233                 sib_info->back = cpu_to_be32(dead_blkno);
2234                 xfs_trans_log_buf(tp, sib_buf,
2235                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2236                                         sizeof(sib_info->back)));
2237                 sib_buf = NULL;
2238         }
2239         par_blkno = args->geo->leafblk;
2240         level = -1;
2241         /*
2242          * Walk down the tree looking for the parent of the moved block.
2243          */
2244         for (;;) {
2245                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2246                 if (error)
2247                         goto done;
2248                 par_node = par_buf->b_addr;
2249                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2250                 if (level >= 0 && level != par_hdr.level + 1) {
2251                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2252                                          XFS_ERRLEVEL_LOW, mp);
2253                         error = -EFSCORRUPTED;
2254                         goto done;
2255                 }
2256                 level = par_hdr.level;
2257                 btree = dp->d_ops->node_tree_p(par_node);
2258                 for (entno = 0;
2259                      entno < par_hdr.count &&
2260                      be32_to_cpu(btree[entno].hashval) < dead_hash;
2261                      entno++)
2262                         continue;
2263                 if (entno == par_hdr.count) {
2264                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2265                                          XFS_ERRLEVEL_LOW, mp);
2266                         error = -EFSCORRUPTED;
2267                         goto done;
2268                 }
2269                 par_blkno = be32_to_cpu(btree[entno].before);
2270                 if (level == dead_level + 1)
2271                         break;
2272                 xfs_trans_brelse(tp, par_buf);
2273                 par_buf = NULL;
2274         }
2275         /*
2276          * We're in the right parent block.
2277          * Look for the right entry.
2278          */
2279         for (;;) {
2280                 for (;
2281                      entno < par_hdr.count &&
2282                      be32_to_cpu(btree[entno].before) != last_blkno;
2283                      entno++)
2284                         continue;
2285                 if (entno < par_hdr.count)
2286                         break;
2287                 par_blkno = par_hdr.forw;
2288                 xfs_trans_brelse(tp, par_buf);
2289                 par_buf = NULL;
2290                 if (unlikely(par_blkno == 0)) {
2291                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2292                                          XFS_ERRLEVEL_LOW, mp);
2293                         error = -EFSCORRUPTED;
2294                         goto done;
2295                 }
2296                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2297                 if (error)
2298                         goto done;
2299                 par_node = par_buf->b_addr;
2300                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2301                 if (par_hdr.level != level) {
2302                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2303                                          XFS_ERRLEVEL_LOW, mp);
2304                         error = -EFSCORRUPTED;
2305                         goto done;
2306                 }
2307                 btree = dp->d_ops->node_tree_p(par_node);
2308                 entno = 0;
2309         }
2310         /*
2311          * Update the parent entry pointing to the moved block.
2312          */
2313         btree[entno].before = cpu_to_be32(dead_blkno);
2314         xfs_trans_log_buf(tp, par_buf,
2315                 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2316                                 sizeof(btree[entno].before)));
2317         *dead_blknop = last_blkno;
2318         *dead_bufp = last_buf;
2319         return 0;
2320 done:
2321         if (par_buf)
2322                 xfs_trans_brelse(tp, par_buf);
2323         if (sib_buf)
2324                 xfs_trans_brelse(tp, sib_buf);
2325         xfs_trans_brelse(tp, last_buf);
2326         return error;
2327 }
2328
2329 /*
2330  * Remove a btree block from a directory or attribute.
2331  */
2332 int
2333 xfs_da_shrink_inode(
2334         xfs_da_args_t   *args,
2335         xfs_dablk_t     dead_blkno,
2336         struct xfs_buf  *dead_buf)
2337 {
2338         xfs_inode_t *dp;
2339         int done, error, w, count;
2340         xfs_trans_t *tp;
2341
2342         trace_xfs_da_shrink_inode(args);
2343
2344         dp = args->dp;
2345         w = args->whichfork;
2346         tp = args->trans;
2347         count = args->geo->fsbcount;
2348         for (;;) {
2349                 /*
2350                  * Remove extents.  If we get ENOSPC for a dir we have to move
2351                  * the last block to the place we want to kill.
2352                  */
2353                 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2354                                     xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2355                                     0, args->firstblock, args->flist, &done);
2356                 if (error == -ENOSPC) {
2357                         if (w != XFS_DATA_FORK)
2358                                 break;
2359                         error = xfs_da3_swap_lastblock(args, &dead_blkno,
2360                                                       &dead_buf);
2361                         if (error)
2362                                 break;
2363                 } else {
2364                         break;
2365                 }
2366         }
2367         xfs_trans_binval(tp, dead_buf);
2368         return error;
2369 }
2370
2371 /*
2372  * See if the mapping(s) for this btree block are valid, i.e.
2373  * don't contain holes, are logically contiguous, and cover the whole range.
2374  */
2375 STATIC int
2376 xfs_da_map_covers_blocks(
2377         int             nmap,
2378         xfs_bmbt_irec_t *mapp,
2379         xfs_dablk_t     bno,
2380         int             count)
2381 {
2382         int             i;
2383         xfs_fileoff_t   off;
2384
2385         for (i = 0, off = bno; i < nmap; i++) {
2386                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2387                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
2388                         return 0;
2389                 }
2390                 if (off != mapp[i].br_startoff) {
2391                         return 0;
2392                 }
2393                 off += mapp[i].br_blockcount;
2394         }
2395         return off == bno + count;
2396 }
2397
2398 /*
2399  * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2400  *
2401  * For the single map case, it is assumed that the caller has provided a pointer
2402  * to a valid xfs_buf_map.  For the multiple map case, this function will
2403  * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2404  * map pointer with the allocated map.
2405  */
2406 static int
2407 xfs_buf_map_from_irec(
2408         struct xfs_mount        *mp,
2409         struct xfs_buf_map      **mapp,
2410         int                     *nmaps,
2411         struct xfs_bmbt_irec    *irecs,
2412         int                     nirecs)
2413 {
2414         struct xfs_buf_map      *map;
2415         int                     i;
2416
2417         ASSERT(*nmaps == 1);
2418         ASSERT(nirecs >= 1);
2419
2420         if (nirecs > 1) {
2421                 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2422                                   KM_SLEEP | KM_NOFS);
2423                 if (!map)
2424                         return -ENOMEM;
2425                 *mapp = map;
2426         }
2427
2428         *nmaps = nirecs;
2429         map = *mapp;
2430         for (i = 0; i < *nmaps; i++) {
2431                 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2432                        irecs[i].br_startblock != HOLESTARTBLOCK);
2433                 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2434                 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2435         }
2436         return 0;
2437 }
2438
2439 /*
2440  * Map the block we are given ready for reading. There are three possible return
2441  * values:
2442  *      -1 - will be returned if we land in a hole and mappedbno == -2 so the
2443  *           caller knows not to execute a subsequent read.
2444  *       0 - if we mapped the block successfully
2445  *      >0 - positive error number if there was an error.
2446  */
2447 static int
2448 xfs_dabuf_map(
2449         struct xfs_inode        *dp,
2450         xfs_dablk_t             bno,
2451         xfs_daddr_t             mappedbno,
2452         int                     whichfork,
2453         struct xfs_buf_map      **map,
2454         int                     *nmaps)
2455 {
2456         struct xfs_mount        *mp = dp->i_mount;
2457         int                     nfsb;
2458         int                     error = 0;
2459         struct xfs_bmbt_irec    irec;
2460         struct xfs_bmbt_irec    *irecs = &irec;
2461         int                     nirecs;
2462
2463         ASSERT(map && *map);
2464         ASSERT(*nmaps == 1);
2465
2466         if (whichfork == XFS_DATA_FORK)
2467                 nfsb = mp->m_dir_geo->fsbcount;
2468         else
2469                 nfsb = mp->m_attr_geo->fsbcount;
2470
2471         /*
2472          * Caller doesn't have a mapping.  -2 means don't complain
2473          * if we land in a hole.
2474          */
2475         if (mappedbno == -1 || mappedbno == -2) {
2476                 /*
2477                  * Optimize the one-block case.
2478                  */
2479                 if (nfsb != 1)
2480                         irecs = kmem_zalloc(sizeof(irec) * nfsb,
2481                                             KM_SLEEP | KM_NOFS);
2482
2483                 nirecs = nfsb;
2484                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2485                                        &nirecs, xfs_bmapi_aflag(whichfork));
2486                 if (error)
2487                         goto out;
2488         } else {
2489                 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2490                 irecs->br_startoff = (xfs_fileoff_t)bno;
2491                 irecs->br_blockcount = nfsb;
2492                 irecs->br_state = 0;
2493                 nirecs = 1;
2494         }
2495
2496         if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2497                 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2498                 if (unlikely(error == -EFSCORRUPTED)) {
2499                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2500                                 int i;
2501                                 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2502                                         __func__, (long long)bno,
2503                                         (long long)dp->i_ino);
2504                                 for (i = 0; i < *nmaps; i++) {
2505                                         xfs_alert(mp,
2506 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2507                                                 i,
2508                                                 (long long)irecs[i].br_startoff,
2509                                                 (long long)irecs[i].br_startblock,
2510                                                 (long long)irecs[i].br_blockcount,
2511                                                 irecs[i].br_state);
2512                                 }
2513                         }
2514                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2515                                          XFS_ERRLEVEL_LOW, mp);
2516                 }
2517                 goto out;
2518         }
2519         error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2520 out:
2521         if (irecs != &irec)
2522                 kmem_free(irecs);
2523         return error;
2524 }
2525
2526 /*
2527  * Get a buffer for the dir/attr block.
2528  */
2529 int
2530 xfs_da_get_buf(
2531         struct xfs_trans        *trans,
2532         struct xfs_inode        *dp,
2533         xfs_dablk_t             bno,
2534         xfs_daddr_t             mappedbno,
2535         struct xfs_buf          **bpp,
2536         int                     whichfork)
2537 {
2538         struct xfs_buf          *bp;
2539         struct xfs_buf_map      map;
2540         struct xfs_buf_map      *mapp;
2541         int                     nmap;
2542         int                     error;
2543
2544         *bpp = NULL;
2545         mapp = &map;
2546         nmap = 1;
2547         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2548                                 &mapp, &nmap);
2549         if (error) {
2550                 /* mapping a hole is not an error, but we don't continue */
2551                 if (error == -1)
2552                         error = 0;
2553                 goto out_free;
2554         }
2555
2556         bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2557                                     mapp, nmap, 0);
2558         error = bp ? bp->b_error : -EIO;
2559         if (error) {
2560                 if (bp)
2561                         xfs_trans_brelse(trans, bp);
2562                 goto out_free;
2563         }
2564
2565         *bpp = bp;
2566
2567 out_free:
2568         if (mapp != &map)
2569                 kmem_free(mapp);
2570
2571         return error;
2572 }
2573
2574 /*
2575  * Get a buffer for the dir/attr block, fill in the contents.
2576  */
2577 int
2578 xfs_da_read_buf(
2579         struct xfs_trans        *trans,
2580         struct xfs_inode        *dp,
2581         xfs_dablk_t             bno,
2582         xfs_daddr_t             mappedbno,
2583         struct xfs_buf          **bpp,
2584         int                     whichfork,
2585         const struct xfs_buf_ops *ops)
2586 {
2587         struct xfs_buf          *bp;
2588         struct xfs_buf_map      map;
2589         struct xfs_buf_map      *mapp;
2590         int                     nmap;
2591         int                     error;
2592
2593         *bpp = NULL;
2594         mapp = &map;
2595         nmap = 1;
2596         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2597                                 &mapp, &nmap);
2598         if (error) {
2599                 /* mapping a hole is not an error, but we don't continue */
2600                 if (error == -1)
2601                         error = 0;
2602                 goto out_free;
2603         }
2604
2605         error = xfs_trans_read_buf_map(dp->i_mount, trans,
2606                                         dp->i_mount->m_ddev_targp,
2607                                         mapp, nmap, 0, &bp, ops);
2608         if (error)
2609                 goto out_free;
2610
2611         if (whichfork == XFS_ATTR_FORK)
2612                 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2613         else
2614                 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2615         *bpp = bp;
2616 out_free:
2617         if (mapp != &map)
2618                 kmem_free(mapp);
2619
2620         return error;
2621 }
2622
2623 /*
2624  * Readahead the dir/attr block.
2625  */
2626 xfs_daddr_t
2627 xfs_da_reada_buf(
2628         struct xfs_inode        *dp,
2629         xfs_dablk_t             bno,
2630         xfs_daddr_t             mappedbno,
2631         int                     whichfork,
2632         const struct xfs_buf_ops *ops)
2633 {
2634         struct xfs_buf_map      map;
2635         struct xfs_buf_map      *mapp;
2636         int                     nmap;
2637         int                     error;
2638
2639         mapp = &map;
2640         nmap = 1;
2641         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2642                                 &mapp, &nmap);
2643         if (error) {
2644                 /* mapping a hole is not an error, but we don't continue */
2645                 if (error == -1)
2646                         error = 0;
2647                 goto out_free;
2648         }
2649
2650         mappedbno = mapp[0].bm_bn;
2651         xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2652
2653 out_free:
2654         if (mapp != &map)
2655                 kmem_free(mapp);
2656
2657         if (error)
2658                 return -1;
2659         return mappedbno;
2660 }