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_sb.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_da_format.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_dir2_priv.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_inode_item.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_attr.h"
  38 #include "xfs_attr_leaf.h"
  39 #include "xfs_error.h"
  40 #include "xfs_trace.h"
  41 #include "xfs_cksum.h"
  42 #include "xfs_buf_item.h"
  43
  44 /*
  45  * xfs_da_btree.c
  46  *
  47  * Routines to implement directories as Btrees of hashed names.
  48  */
  49
  50 /*========================================================================
  51  * Function prototypes for the kernel.
  52  *========================================================================*/
  53
  54 /*
  55  * Routines used for growing the Btree.
  56  */
  57 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
  58                                             xfs_da_state_blk_t *existing_root,
  59                                             xfs_da_state_blk_t *new_child);
  60 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
  61                                             xfs_da_state_blk_t *existing_blk,
  62                                             xfs_da_state_blk_t *split_blk,
  63                                             xfs_da_state_blk_t *blk_to_add,
  64                                             int treelevel,
  65                                             int *result);
  66 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
  67                                          xfs_da_state_blk_t *node_blk_1,
  68                                          xfs_da_state_blk_t *node_blk_2);
  69 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
  70                                    xfs_da_state_blk_t *old_node_blk,
  71                                    xfs_da_state_blk_t *new_node_blk);
  72
  73 /*
  74  * Routines used for shrinking the Btree.
  75  */
  76 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
  77                                            xfs_da_state_blk_t *root_blk);
  78 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
  79 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
  80                                               xfs_da_state_blk_t *drop_blk);
  81 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
  82                                          xfs_da_state_blk_t *src_node_blk,
  83                                          xfs_da_state_blk_t *dst_node_blk);
  84
  85 /*
  86  * Utility routines.
  87  */
  88 STATIC int      xfs_da3_blk_unlink(xfs_da_state_t *state,
  89                                   xfs_da_state_blk_t *drop_blk,
  90                                   xfs_da_state_blk_t *save_blk);
  91
  92
  93 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
  94
  95 /*
  96  * Allocate a dir-state structure.
  97  * We don't put them on the stack since they're large.
  98  */
  99 xfs_da_state_t *
 100 xfs_da_state_alloc(void)
 101 {
 102         return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
 103 }
 104
 105 /*
 106  * Kill the altpath contents of a da-state structure.
 107  */
 108 STATIC void
 109 xfs_da_state_kill_altpath(xfs_da_state_t *state)
 110 {
 111         int     i;
 112
 113         for (i = 0; i < state->altpath.active; i++)
 114                 state->altpath.blk[i].bp = NULL;
 115         state->altpath.active = 0;
 116 }
 117
 118 /*
 119  * Free a da-state structure.
 120  */
 121 void
 122 xfs_da_state_free(xfs_da_state_t *state)
 123 {
 124         xfs_da_state_kill_altpath(state);
 125 #ifdef DEBUG
 126         memset((char *)state, 0, sizeof(*state));
 127 #endif /* DEBUG */
 128         kmem_zone_free(xfs_da_state_zone, state);
 129 }
 130
 131 static bool
 132 xfs_da3_node_verify(
 133         struct xfs_buf          *bp)
 134 {
 135         struct xfs_mount        *mp = bp->b_target->bt_mount;
 136         struct xfs_da_intnode   *hdr = bp->b_addr;
 137         struct xfs_da3_icnode_hdr ichdr;
 138         const struct xfs_dir_ops *ops;
 139
 140         ops = xfs_dir_get_ops(mp, NULL);
 141
 142         ops->node_hdr_from_disk(&ichdr, hdr);
 143
 144         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 145                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 146
 147                 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
 148                         return false;
 149
 150                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
 151                         return false;
 152                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
 153                         return false;
 154         } else {
 155                 if (ichdr.magic != XFS_DA_NODE_MAGIC)
 156                         return false;
 157         }
 158         if (ichdr.level == 0)
 159                 return false;
 160         if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
 161                 return false;
 162         if (ichdr.count == 0)
 163                 return false;
 164
 165         /*
 166          * we don't know if the node is for and attribute or directory tree,
 167          * so only fail if the count is outside both bounds
 168          */
 169         if (ichdr.count > mp->m_dir_geo->node_ents &&
 170             ichdr.count > mp->m_attr_geo->node_ents)
 171                 return false;
 172
 173         /* XXX: hash order check? */
 174
 175         return true;
 176 }
 177
 178 static void
 179 xfs_da3_node_write_verify(
 180         struct xfs_buf  *bp)
 181 {
 182         struct xfs_mount        *mp = bp->b_target->bt_mount;
 183         struct xfs_buf_log_item *bip = bp->b_fspriv;
 184         struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 185
 186         if (!xfs_da3_node_verify(bp)) {
 187                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 188                 xfs_verifier_error(bp);
 189                 return;
 190         }
 191
 192         if (!xfs_sb_version_hascrc(&mp->m_sb))
 193                 return;
 194
 195         if (bip)
 196                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 197
 198         xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
 199 }
 200
 201 /*
 202  * leaf/node format detection on trees is sketchy, so a node read can be done on
 203  * leaf level blocks when detection identifies the tree as a node format tree
 204  * incorrectly. In this case, we need to swap the verifier to match the correct
 205  * format of the block being read.
 206  */
 207 static void
 208 xfs_da3_node_read_verify(
 209         struct xfs_buf          *bp)
 210 {
 211         struct xfs_da_blkinfo   *info = bp->b_addr;
 212
 213         switch (be16_to_cpu(info->magic)) {
 214                 case XFS_DA3_NODE_MAGIC:
 215                         if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
 216                                 xfs_buf_ioerror(bp, -EFSBADCRC);
 217                                 break;
 218                         }
 219                         /* fall through */
 220                 case XFS_DA_NODE_MAGIC:
 221                         if (!xfs_da3_node_verify(bp)) {
 222                                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 223                                 break;
 224                         }
 225                         return;
 226                 case XFS_ATTR_LEAF_MAGIC:
 227                 case XFS_ATTR3_LEAF_MAGIC:
 228                         bp->b_ops = &xfs_attr3_leaf_buf_ops;
 229                         bp->b_ops->verify_read(bp);
 230                         return;
 231                 case XFS_DIR2_LEAFN_MAGIC:
 232                 case XFS_DIR3_LEAFN_MAGIC:
 233                         bp->b_ops = &xfs_dir3_leafn_buf_ops;
 234                         bp->b_ops->verify_read(bp);
 235                         return;
 236                 default:
 237                         break;
 238         }
 239
 240         /* corrupt block */
 241         xfs_verifier_error(bp);
 242 }
 243
 244 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
 245         .verify_read = xfs_da3_node_read_verify,
 246         .verify_write = xfs_da3_node_write_verify,
 247 };
 248
 249 int
 250 xfs_da3_node_read(
 251         struct xfs_trans        *tp,
 252         struct xfs_inode        *dp,
 253         xfs_dablk_t             bno,
 254         xfs_daddr_t             mappedbno,
 255         struct xfs_buf          **bpp,
 256         int                     which_fork)
 257 {
 258         int                     err;
 259
 260         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
 261                                         which_fork, &xfs_da3_node_buf_ops);
 262         if (!err && tp) {
 263                 struct xfs_da_blkinfo   *info = (*bpp)->b_addr;
 264                 int                     type;
 265
 266                 switch (be16_to_cpu(info->magic)) {
 267                 case XFS_DA_NODE_MAGIC:
 268                 case XFS_DA3_NODE_MAGIC:
 269                         type = XFS_BLFT_DA_NODE_BUF;
 270                         break;
 271                 case XFS_ATTR_LEAF_MAGIC:
 272                 case XFS_ATTR3_LEAF_MAGIC:
 273                         type = XFS_BLFT_ATTR_LEAF_BUF;
 274                         break;
 275                 case XFS_DIR2_LEAFN_MAGIC:
 276                 case XFS_DIR3_LEAFN_MAGIC:
 277                         type = XFS_BLFT_DIR_LEAFN_BUF;
 278                         break;
 279                 default:
 280                         type = 0;
 281                         ASSERT(0);
 282                         break;
 283                 }
 284                 xfs_trans_buf_set_type(tp, *bpp, type);
 285         }
 286         return err;
 287 }
 288
 289 /*========================================================================
 290  * Routines used for growing the Btree.
 291  *========================================================================*/
 292
 293 /*
 294  * Create the initial contents of an intermediate node.
 295  */
 296 int
 297 xfs_da3_node_create(
 298         struct xfs_da_args      *args,
 299         xfs_dablk_t             blkno,
 300         int                     level,
 301         struct xfs_buf          **bpp,
 302         int                     whichfork)
 303 {
 304         struct xfs_da_intnode   *node;
 305         struct xfs_trans        *tp = args->trans;
 306         struct xfs_mount        *mp = tp->t_mountp;
 307         struct xfs_da3_icnode_hdr ichdr = {0};
 308         struct xfs_buf          *bp;
 309         int                     error;
 310         struct xfs_inode        *dp = args->dp;
 311
 312         trace_xfs_da_node_create(args);
 313         ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
 314
 315         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
 316         if (error)
 317                 return error;
 318         bp->b_ops = &xfs_da3_node_buf_ops;
 319         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 320         node = bp->b_addr;
 321
 322         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 323                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 324
 325                 ichdr.magic = XFS_DA3_NODE_MAGIC;
 326                 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
 327                 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
 328                 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_uuid);
 329         } else {
 330                 ichdr.magic = XFS_DA_NODE_MAGIC;
 331         }
 332         ichdr.level = level;
 333
 334         dp->d_ops->node_hdr_to_disk(node, &ichdr);
 335         xfs_trans_log_buf(tp, bp,
 336                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 337
 338         *bpp = bp;
 339         return 0;
 340 }
 341
 342 /*
 343  * Split a leaf node, rebalance, then possibly split
 344  * intermediate nodes, rebalance, etc.
 345  */
 346 int                                                     /* error */
 347 xfs_da3_split(
 348         struct xfs_da_state     *state)
 349 {
 350         struct xfs_da_state_blk *oldblk;
 351         struct xfs_da_state_blk *newblk;
 352         struct xfs_da_state_blk *addblk;
 353         struct xfs_da_intnode   *node;
 354         struct xfs_buf          *bp;
 355         int                     max;
 356         int                     action = 0;
 357         int                     error;
 358         int                     i;
 359
 360         trace_xfs_da_split(state->args);
 361
 362         /*
 363          * Walk back up the tree splitting/inserting/adjusting as necessary.
 364          * If we need to insert and there isn't room, split the node, then
 365          * decide which fragment to insert the new block from below into.
 366          * Note that we may split the root this way, but we need more fixup.
 367          */
 368         max = state->path.active - 1;
 369         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 370         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 371                state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 372
 373         addblk = &state->path.blk[max];         /* initial dummy value */
 374         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 375                 oldblk = &state->path.blk[i];
 376                 newblk = &state->altpath.blk[i];
 377
 378                 /*
 379                  * If a leaf node then
 380                  *     Allocate a new leaf node, then rebalance across them.
 381                  * else if an intermediate node then
 382                  *     We split on the last layer, must we split the node?
 383                  */
 384                 switch (oldblk->magic) {
 385                 case XFS_ATTR_LEAF_MAGIC:
 386                         error = xfs_attr3_leaf_split(state, oldblk, newblk);
 387                         if ((error != 0) && (error != -ENOSPC)) {
 388                                 return error;   /* GROT: attr is inconsistent */
 389                         }
 390                         if (!error) {
 391                                 addblk = newblk;
 392                                 break;
 393                         }
 394                         /*
 395                          * Entry wouldn't fit, split the leaf again.
 396                          */
 397                         state->extravalid = 1;
 398                         if (state->inleaf) {
 399                                 state->extraafter = 0;  /* before newblk */
 400                                 trace_xfs_attr_leaf_split_before(state->args);
 401                                 error = xfs_attr3_leaf_split(state, oldblk,
 402                                                             &state->extrablk);
 403                         } else {
 404                                 state->extraafter = 1;  /* after newblk */
 405                                 trace_xfs_attr_leaf_split_after(state->args);
 406                                 error = xfs_attr3_leaf_split(state, newblk,
 407                                                             &state->extrablk);
 408                         }
 409                         if (error)
 410                                 return error;   /* GROT: attr inconsistent */
 411                         addblk = newblk;
 412                         break;
 413                 case XFS_DIR2_LEAFN_MAGIC:
 414                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 415                         if (error)
 416                                 return error;
 417                         addblk = newblk;
 418                         break;
 419                 case XFS_DA_NODE_MAGIC:
 420                         error = xfs_da3_node_split(state, oldblk, newblk, addblk,
 421                                                          max - i, &action);
 422                         addblk->bp = NULL;
 423                         if (error)
 424                                 return error;   /* GROT: dir is inconsistent */
 425                         /*
 426                          * Record the newly split block for the next time thru?
 427                          */
 428                         if (action)
 429                                 addblk = newblk;
 430                         else
 431                                 addblk = NULL;
 432                         break;
 433                 }
 434
 435                 /*
 436                  * Update the btree to show the new hashval for this child.
 437                  */
 438                 xfs_da3_fixhashpath(state, &state->path);
 439         }
 440         if (!addblk)
 441                 return 0;
 442
 443         /*
 444          * Split the root node.
 445          */
 446         ASSERT(state->path.active == 0);
 447         oldblk = &state->path.blk[0];
 448         error = xfs_da3_root_split(state, oldblk, addblk);
 449         if (error) {
 450                 addblk->bp = NULL;
 451                 return error;   /* GROT: dir is inconsistent */
 452         }
 453
 454         /*
 455          * Update pointers to the node which used to be block 0 and
 456          * just got bumped because of the addition of a new root node.
 457          * There might be three blocks involved if a double split occurred,
 458          * and the original block 0 could be at any position in the list.
 459          *
 460          * Note: the magic numbers and sibling pointers are in the same
 461          * physical place for both v2 and v3 headers (by design). Hence it
 462          * doesn't matter which version of the xfs_da_intnode structure we use
 463          * here as the result will be the same using either structure.
 464          */
 465         node = oldblk->bp->b_addr;
 466         if (node->hdr.info.forw) {
 467                 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
 468                         bp = addblk->bp;
 469                 } else {
 470                         ASSERT(state->extravalid);
 471                         bp = state->extrablk.bp;
 472                 }
 473                 node = bp->b_addr;
 474                 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
 475                 xfs_trans_log_buf(state->args->trans, bp,
 476                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 477                     sizeof(node->hdr.info)));
 478         }
 479         node = oldblk->bp->b_addr;
 480         if (node->hdr.info.back) {
 481                 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
 482                         bp = addblk->bp;
 483                 } else {
 484                         ASSERT(state->extravalid);
 485                         bp = state->extrablk.bp;
 486                 }
 487                 node = bp->b_addr;
 488                 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
 489                 xfs_trans_log_buf(state->args->trans, bp,
 490                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 491                     sizeof(node->hdr.info)));
 492         }
 493         addblk->bp = NULL;
 494         return 0;
 495 }
 496
 497 /*
 498  * Split the root.  We have to create a new root and point to the two
 499  * parts (the split old root) that we just created.  Copy block zero to
 500  * the EOF, extending the inode in process.
 501  */
 502 STATIC int                                              /* error */
 503 xfs_da3_root_split(
 504         struct xfs_da_state     *state,
 505         struct xfs_da_state_blk *blk1,
 506         struct xfs_da_state_blk *blk2)
 507 {
 508         struct xfs_da_intnode   *node;
 509         struct xfs_da_intnode   *oldroot;
 510         struct xfs_da_node_entry *btree;
 511         struct xfs_da3_icnode_hdr nodehdr;
 512         struct xfs_da_args      *args;
 513         struct xfs_buf          *bp;
 514         struct xfs_inode        *dp;
 515         struct xfs_trans        *tp;
 516         struct xfs_mount        *mp;
 517         struct xfs_dir2_leaf    *leaf;
 518         xfs_dablk_t             blkno;
 519         int                     level;
 520         int                     error;
 521         int                     size;
 522
 523         trace_xfs_da_root_split(state->args);
 524
 525         /*
 526          * Copy the existing (incorrect) block from the root node position
 527          * to a free space somewhere.
 528          */
 529         args = state->args;
 530         error = xfs_da_grow_inode(args, &blkno);
 531         if (error)
 532                 return error;
 533
 534         dp = args->dp;
 535         tp = args->trans;
 536         mp = state->mp;
 537         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 538         if (error)
 539                 return error;
 540         node = bp->b_addr;
 541         oldroot = blk1->bp->b_addr;
 542         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
 543             oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
 544                 struct xfs_da3_icnode_hdr nodehdr;
 545
 546                 dp->d_ops->node_hdr_from_disk(&nodehdr, oldroot);
 547                 btree = dp->d_ops->node_tree_p(oldroot);
 548                 size = (int)((char *)&btree[nodehdr.count] - (char *)oldroot);
 549                 level = nodehdr.level;
 550
 551                 /*
 552                  * we are about to copy oldroot to bp, so set up the type
 553                  * of bp while we know exactly what it will be.
 554                  */
 555                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 556         } else {
 557                 struct xfs_dir3_icleaf_hdr leafhdr;
 558                 struct xfs_dir2_leaf_entry *ents;
 559
 560                 leaf = (xfs_dir2_leaf_t *)oldroot;
 561                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 562                 ents = dp->d_ops->leaf_ents_p(leaf);
 563
 564                 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 565                        leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 566                 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
 567                 level = 0;
 568
 569                 /*
 570                  * we are about to copy oldroot to bp, so set up the type
 571                  * of bp while we know exactly what it will be.
 572                  */
 573                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 574         }
 575
 576         /*
 577          * we can copy most of the information in the node from one block to
 578          * another, but for CRC enabled headers we have to make sure that the
 579          * block specific identifiers are kept intact. We update the buffer
 580          * directly for this.
 581          */
 582         memcpy(node, oldroot, size);
 583         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
 584             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 585                 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
 586
 587                 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
 588         }
 589         xfs_trans_log_buf(tp, bp, 0, size - 1);
 590
 591         bp->b_ops = blk1->bp->b_ops;
 592         xfs_trans_buf_copy_type(bp, blk1->bp);
 593         blk1->bp = bp;
 594         blk1->blkno = blkno;
 595
 596         /*
 597          * Set up the new root node.
 598          */
 599         error = xfs_da3_node_create(args,
 600                 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
 601                 level + 1, &bp, args->whichfork);
 602         if (error)
 603                 return error;
 604
 605         node = bp->b_addr;
 606         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 607         btree = dp->d_ops->node_tree_p(node);
 608         btree[0].hashval = cpu_to_be32(blk1->hashval);
 609         btree[0].before = cpu_to_be32(blk1->blkno);
 610         btree[1].hashval = cpu_to_be32(blk2->hashval);
 611         btree[1].before = cpu_to_be32(blk2->blkno);
 612         nodehdr.count = 2;
 613         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 614
 615 #ifdef DEBUG
 616         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
 617             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 618                 ASSERT(blk1->blkno >= args->geo->leafblk &&
 619                        blk1->blkno < args->geo->freeblk);
 620                 ASSERT(blk2->blkno >= args->geo->leafblk &&
 621                        blk2->blkno < args->geo->freeblk);
 622         }
 623 #endif
 624
 625         /* Header is already logged by xfs_da_node_create */
 626         xfs_trans_log_buf(tp, bp,
 627                 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
 628
 629         return 0;
 630 }
 631
 632 /*
 633  * Split the node, rebalance, then add the new entry.
 634  */
 635 STATIC int                                              /* error */
 636 xfs_da3_node_split(
 637         struct xfs_da_state     *state,
 638         struct xfs_da_state_blk *oldblk,
 639         struct xfs_da_state_blk *newblk,
 640         struct xfs_da_state_blk *addblk,
 641         int                     treelevel,
 642         int                     *result)
 643 {
 644         struct xfs_da_intnode   *node;
 645         struct xfs_da3_icnode_hdr nodehdr;
 646         xfs_dablk_t             blkno;
 647         int                     newcount;
 648         int                     error;
 649         int                     useextra;
 650         struct xfs_inode        *dp = state->args->dp;
 651
 652         trace_xfs_da_node_split(state->args);
 653
 654         node = oldblk->bp->b_addr;
 655         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 656
 657         /*
 658          * With V2 dirs the extra block is data or freespace.
 659          */
 660         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 661         newcount = 1 + useextra;
 662         /*
 663          * Do we have to split the node?
 664          */
 665         if (nodehdr.count + newcount > state->args->geo->node_ents) {
 666                 /*
 667                  * Allocate a new node, add to the doubly linked chain of
 668                  * nodes, then move some of our excess entries into it.
 669                  */
 670                 error = xfs_da_grow_inode(state->args, &blkno);
 671                 if (error)
 672                         return error;   /* GROT: dir is inconsistent */
 673
 674                 error = xfs_da3_node_create(state->args, blkno, treelevel,
 675                                            &newblk->bp, state->args->whichfork);
 676                 if (error)
 677                         return error;   /* GROT: dir is inconsistent */
 678                 newblk->blkno = blkno;
 679                 newblk->magic = XFS_DA_NODE_MAGIC;
 680                 xfs_da3_node_rebalance(state, oldblk, newblk);
 681                 error = xfs_da3_blk_link(state, oldblk, newblk);
 682                 if (error)
 683                         return error;
 684                 *result = 1;
 685         } else {
 686                 *result = 0;
 687         }
 688
 689         /*
 690          * Insert the new entry(s) into the correct block
 691          * (updating last hashval in the process).
 692          *
 693          * xfs_da3_node_add() inserts BEFORE the given index,
 694          * and as a result of using node_lookup_int() we always
 695          * point to a valid entry (not after one), but a split
 696          * operation always results in a new block whose hashvals
 697          * FOLLOW the current block.
 698          *
 699          * If we had double-split op below us, then add the extra block too.
 700          */
 701         node = oldblk->bp->b_addr;
 702         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 703         if (oldblk->index <= nodehdr.count) {
 704                 oldblk->index++;
 705                 xfs_da3_node_add(state, oldblk, addblk);
 706                 if (useextra) {
 707                         if (state->extraafter)
 708                                 oldblk->index++;
 709                         xfs_da3_node_add(state, oldblk, &state->extrablk);
 710                         state->extravalid = 0;
 711                 }
 712         } else {
 713                 newblk->index++;
 714                 xfs_da3_node_add(state, newblk, addblk);
 715                 if (useextra) {
 716                         if (state->extraafter)
 717                                 newblk->index++;
 718                         xfs_da3_node_add(state, newblk, &state->extrablk);
 719                         state->extravalid = 0;
 720                 }
 721         }
 722
 723         return 0;
 724 }
 725
 726 /*
 727  * Balance the btree elements between two intermediate nodes,
 728  * usually one full and one empty.
 729  *
 730  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 731  */
 732 STATIC void
 733 xfs_da3_node_rebalance(
 734         struct xfs_da_state     *state,
 735         struct xfs_da_state_blk *blk1,
 736         struct xfs_da_state_blk *blk2)
 737 {
 738         struct xfs_da_intnode   *node1;
 739         struct xfs_da_intnode   *node2;
 740         struct xfs_da_intnode   *tmpnode;
 741         struct xfs_da_node_entry *btree1;
 742         struct xfs_da_node_entry *btree2;
 743         struct xfs_da_node_entry *btree_s;
 744         struct xfs_da_node_entry *btree_d;
 745         struct xfs_da3_icnode_hdr nodehdr1;
 746         struct xfs_da3_icnode_hdr nodehdr2;
 747         struct xfs_trans        *tp;
 748         int                     count;
 749         int                     tmp;
 750         int                     swap = 0;
 751         struct xfs_inode        *dp = state->args->dp;
 752
 753         trace_xfs_da_node_rebalance(state->args);
 754
 755         node1 = blk1->bp->b_addr;
 756         node2 = blk2->bp->b_addr;
 757         dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 758         dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 759         btree1 = dp->d_ops->node_tree_p(node1);
 760         btree2 = dp->d_ops->node_tree_p(node2);
 761
 762         /*
 763          * Figure out how many entries need to move, and in which direction.
 764          * Swap the nodes around if that makes it simpler.
 765          */
 766         if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
 767             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
 768              (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
 769                         be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
 770                 tmpnode = node1;
 771                 node1 = node2;
 772                 node2 = tmpnode;
 773                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 774                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 775                 btree1 = dp->d_ops->node_tree_p(node1);
 776                 btree2 = dp->d_ops->node_tree_p(node2);
 777                 swap = 1;
 778         }
 779
 780         count = (nodehdr1.count - nodehdr2.count) / 2;
 781         if (count == 0)
 782                 return;
 783         tp = state->args->trans;
 784         /*
 785          * Two cases: high-to-low and low-to-high.
 786          */
 787         if (count > 0) {
 788                 /*
 789                  * Move elements in node2 up to make a hole.
 790                  */
 791                 tmp = nodehdr2.count;
 792                 if (tmp > 0) {
 793                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 794                         btree_s = &btree2[0];
 795                         btree_d = &btree2[count];
 796                         memmove(btree_d, btree_s, tmp);
 797                 }
 798
 799                 /*
 800                  * Move the req'd B-tree elements from high in node1 to
 801                  * low in node2.
 802                  */
 803                 nodehdr2.count += count;
 804                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 805                 btree_s = &btree1[nodehdr1.count - count];
 806                 btree_d = &btree2[0];
 807                 memcpy(btree_d, btree_s, tmp);
 808                 nodehdr1.count -= count;
 809         } else {
 810                 /*
 811                  * Move the req'd B-tree elements from low in node2 to
 812                  * high in node1.
 813                  */
 814                 count = -count;
 815                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 816                 btree_s = &btree2[0];
 817                 btree_d = &btree1[nodehdr1.count];
 818                 memcpy(btree_d, btree_s, tmp);
 819                 nodehdr1.count += count;
 820
 821                 xfs_trans_log_buf(tp, blk1->bp,
 822                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 823
 824                 /*
 825                  * Move elements in node2 down to fill the hole.
 826                  */
 827                 tmp  = nodehdr2.count - count;
 828                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 829                 btree_s = &btree2[count];
 830                 btree_d = &btree2[0];
 831                 memmove(btree_d, btree_s, tmp);
 832                 nodehdr2.count -= count;
 833         }
 834
 835         /*
 836          * Log header of node 1 and all current bits of node 2.
 837          */
 838         dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
 839         xfs_trans_log_buf(tp, blk1->bp,
 840                 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
 841
 842         dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
 843         xfs_trans_log_buf(tp, blk2->bp,
 844                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 845                                 dp->d_ops->node_hdr_size +
 846                                 (sizeof(btree2[0]) * nodehdr2.count)));
 847
 848         /*
 849          * Record the last hashval from each block for upward propagation.
 850          * (note: don't use the swapped node pointers)
 851          */
 852         if (swap) {
 853                 node1 = blk1->bp->b_addr;
 854                 node2 = blk2->bp->b_addr;
 855                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 856                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 857                 btree1 = dp->d_ops->node_tree_p(node1);
 858                 btree2 = dp->d_ops->node_tree_p(node2);
 859         }
 860         blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
 861         blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
 862
 863         /*
 864          * Adjust the expected index for insertion.
 865          */
 866         if (blk1->index >= nodehdr1.count) {
 867                 blk2->index = blk1->index - nodehdr1.count;
 868                 blk1->index = nodehdr1.count + 1;       /* make it invalid */
 869         }
 870 }
 871
 872 /*
 873  * Add a new entry to an intermediate node.
 874  */
 875 STATIC void
 876 xfs_da3_node_add(
 877         struct xfs_da_state     *state,
 878         struct xfs_da_state_blk *oldblk,
 879         struct xfs_da_state_blk *newblk)
 880 {
 881         struct xfs_da_intnode   *node;
 882         struct xfs_da3_icnode_hdr nodehdr;
 883         struct xfs_da_node_entry *btree;
 884         int                     tmp;
 885         struct xfs_inode        *dp = state->args->dp;
 886
 887         trace_xfs_da_node_add(state->args);
 888
 889         node = oldblk->bp->b_addr;
 890         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 891         btree = dp->d_ops->node_tree_p(node);
 892
 893         ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
 894         ASSERT(newblk->blkno != 0);
 895         if (state->args->whichfork == XFS_DATA_FORK)
 896                 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
 897                        newblk->blkno < state->args->geo->freeblk);
 898
 899         /*
 900          * We may need to make some room before we insert the new node.
 901          */
 902         tmp = 0;
 903         if (oldblk->index < nodehdr.count) {
 904                 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
 905                 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
 906         }
 907         btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
 908         btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
 909         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 910                 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
 911                                 tmp + sizeof(*btree)));
 912
 913         nodehdr.count += 1;
 914         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 915         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 916                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 917
 918         /*
 919          * Copy the last hash value from the oldblk to propagate upwards.
 920          */
 921         oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
 922 }
 923
 924 /*========================================================================
 925  * Routines used for shrinking the Btree.
 926  *========================================================================*/
 927
 928 /*
 929  * Deallocate an empty leaf node, remove it from its parent,
 930  * possibly deallocating that block, etc...
 931  */
 932 int
 933 xfs_da3_join(
 934         struct xfs_da_state     *state)
 935 {
 936         struct xfs_da_state_blk *drop_blk;
 937         struct xfs_da_state_blk *save_blk;
 938         int                     action = 0;
 939         int                     error;
 940
 941         trace_xfs_da_join(state->args);
 942
 943         drop_blk = &state->path.blk[ state->path.active-1 ];
 944         save_blk = &state->altpath.blk[ state->path.active-1 ];
 945         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 946         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 947                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 948
 949         /*
 950          * Walk back up the tree joining/deallocating as necessary.
 951          * When we stop dropping blocks, break out.
 952          */
 953         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 954                  state->path.active--) {
 955                 /*
 956                  * See if we can combine the block with a neighbor.
 957                  *   (action == 0) => no options, just leave
 958                  *   (action == 1) => coalesce, then unlink
 959                  *   (action == 2) => block empty, unlink it
 960                  */
 961                 switch (drop_blk->magic) {
 962                 case XFS_ATTR_LEAF_MAGIC:
 963                         error = xfs_attr3_leaf_toosmall(state, &action);
 964                         if (error)
 965                                 return error;
 966                         if (action == 0)
 967                                 return 0;
 968                         xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
 969                         break;
 970                 case XFS_DIR2_LEAFN_MAGIC:
 971                         error = xfs_dir2_leafn_toosmall(state, &action);
 972                         if (error)
 973                                 return error;
 974                         if (action == 0)
 975                                 return 0;
 976                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 977                         break;
 978                 case XFS_DA_NODE_MAGIC:
 979                         /*
 980                          * Remove the offending node, fixup hashvals,
 981                          * check for a toosmall neighbor.
 982                          */
 983                         xfs_da3_node_remove(state, drop_blk);
 984                         xfs_da3_fixhashpath(state, &state->path);
 985                         error = xfs_da3_node_toosmall(state, &action);
 986                         if (error)
 987                                 return error;
 988                         if (action == 0)
 989                                 return 0;
 990                         xfs_da3_node_unbalance(state, drop_blk, save_blk);
 991                         break;
 992                 }
 993                 xfs_da3_fixhashpath(state, &state->altpath);
 994                 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
 995                 xfs_da_state_kill_altpath(state);
 996                 if (error)
 997                         return error;
 998                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 999                                                          drop_blk->bp);
1000                 drop_blk->bp = NULL;
1001                 if (error)
1002                         return error;
1003         }
1004         /*
1005          * We joined all the way to the top.  If it turns out that
1006          * we only have one entry in the root, make the child block
1007          * the new root.
1008          */
1009         xfs_da3_node_remove(state, drop_blk);
1010         xfs_da3_fixhashpath(state, &state->path);
1011         error = xfs_da3_root_join(state, &state->path.blk[0]);
1012         return error;
1013 }
1014
1015 #ifdef  DEBUG
1016 static void
1017 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1018 {
1019         __be16  magic = blkinfo->magic;
1020
1021         if (level == 1) {
1022                 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1023                        magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1024                        magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1025                        magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1026         } else {
1027                 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1028                        magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1029         }
1030         ASSERT(!blkinfo->forw);
1031         ASSERT(!blkinfo->back);
1032 }
1033 #else   /* !DEBUG */
1034 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1035 #endif  /* !DEBUG */
1036
1037 /*
1038  * We have only one entry in the root.  Copy the only remaining child of
1039  * the old root to block 0 as the new root node.
1040  */
1041 STATIC int
1042 xfs_da3_root_join(
1043         struct xfs_da_state     *state,
1044         struct xfs_da_state_blk *root_blk)
1045 {
1046         struct xfs_da_intnode   *oldroot;
1047         struct xfs_da_args      *args;
1048         xfs_dablk_t             child;
1049         struct xfs_buf          *bp;
1050         struct xfs_da3_icnode_hdr oldroothdr;
1051         struct xfs_da_node_entry *btree;
1052         int                     error;
1053         struct xfs_inode        *dp = state->args->dp;
1054
1055         trace_xfs_da_root_join(state->args);
1056
1057         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1058
1059         args = state->args;
1060         oldroot = root_blk->bp->b_addr;
1061         dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1062         ASSERT(oldroothdr.forw == 0);
1063         ASSERT(oldroothdr.back == 0);
1064
1065         /*
1066          * If the root has more than one child, then don't do anything.
1067          */
1068         if (oldroothdr.count > 1)
1069                 return 0;
1070
1071         /*
1072          * Read in the (only) child block, then copy those bytes into
1073          * the root block's buffer and free the original child block.
1074          */
1075         btree = dp->d_ops->node_tree_p(oldroot);
1076         child = be32_to_cpu(btree[0].before);
1077         ASSERT(child != 0);
1078         error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1079                                              args->whichfork);
1080         if (error)
1081                 return error;
1082         xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1083
1084         /*
1085          * This could be copying a leaf back into the root block in the case of
1086          * there only being a single leaf block left in the tree. Hence we have
1087          * to update the b_ops pointer as well to match the buffer type change
1088          * that could occur. For dir3 blocks we also need to update the block
1089          * number in the buffer header.
1090          */
1091         memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1092         root_blk->bp->b_ops = bp->b_ops;
1093         xfs_trans_buf_copy_type(root_blk->bp, bp);
1094         if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1095                 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1096                 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1097         }
1098         xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1099                           args->geo->blksize - 1);
1100         error = xfs_da_shrink_inode(args, child, bp);
1101         return error;
1102 }
1103
1104 /*
1105  * Check a node block and its neighbors to see if the block should be
1106  * collapsed into one or the other neighbor.  Always keep the block
1107  * with the smaller block number.
1108  * If the current block is over 50% full, don't try to join it, return 0.
1109  * If the block is empty, fill in the state structure and return 2.
1110  * If it can be collapsed, fill in the state structure and return 1.
1111  * If nothing can be done, return 0.
1112  */
1113 STATIC int
1114 xfs_da3_node_toosmall(
1115         struct xfs_da_state     *state,
1116         int                     *action)
1117 {
1118         struct xfs_da_intnode   *node;
1119         struct xfs_da_state_blk *blk;
1120         struct xfs_da_blkinfo   *info;
1121         xfs_dablk_t             blkno;
1122         struct xfs_buf          *bp;
1123         struct xfs_da3_icnode_hdr nodehdr;
1124         int                     count;
1125         int                     forward;
1126         int                     error;
1127         int                     retval;
1128         int                     i;
1129         struct xfs_inode        *dp = state->args->dp;
1130
1131         trace_xfs_da_node_toosmall(state->args);
1132
1133         /*
1134          * Check for the degenerate case of the block being over 50% full.
1135          * If so, it's not worth even looking to see if we might be able
1136          * to coalesce with a sibling.
1137          */
1138         blk = &state->path.blk[ state->path.active-1 ];
1139         info = blk->bp->b_addr;
1140         node = (xfs_da_intnode_t *)info;
1141         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1142         if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1143                 *action = 0;    /* blk over 50%, don't try to join */
1144                 return 0;       /* blk over 50%, don't try to join */
1145         }
1146
1147         /*
1148          * Check for the degenerate case of the block being empty.
1149          * If the block is empty, we'll simply delete it, no need to
1150          * coalesce it with a sibling block.  We choose (arbitrarily)
1151          * to merge with the forward block unless it is NULL.
1152          */
1153         if (nodehdr.count == 0) {
1154                 /*
1155                  * Make altpath point to the block we want to keep and
1156                  * path point to the block we want to drop (this one).
1157                  */
1158                 forward = (info->forw != 0);
1159                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1160                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1161                                                  0, &retval);
1162                 if (error)
1163                         return error;
1164                 if (retval) {
1165                         *action = 0;
1166                 } else {
1167                         *action = 2;
1168                 }
1169                 return 0;
1170         }
1171
1172         /*
1173          * Examine each sibling block to see if we can coalesce with
1174          * at least 25% free space to spare.  We need to figure out
1175          * whether to merge with the forward or the backward block.
1176          * We prefer coalescing with the lower numbered sibling so as
1177          * to shrink a directory over time.
1178          */
1179         count  = state->args->geo->node_ents;
1180         count -= state->args->geo->node_ents >> 2;
1181         count -= nodehdr.count;
1182
1183         /* start with smaller blk num */
1184         forward = nodehdr.forw < nodehdr.back;
1185         for (i = 0; i < 2; forward = !forward, i++) {
1186                 struct xfs_da3_icnode_hdr thdr;
1187                 if (forward)
1188                         blkno = nodehdr.forw;
1189                 else
1190                         blkno = nodehdr.back;
1191                 if (blkno == 0)
1192                         continue;
1193                 error = xfs_da3_node_read(state->args->trans, dp,
1194                                         blkno, -1, &bp, state->args->whichfork);
1195                 if (error)
1196                         return error;
1197
1198                 node = bp->b_addr;
1199                 dp->d_ops->node_hdr_from_disk(&thdr, node);
1200                 xfs_trans_brelse(state->args->trans, bp);
1201
1202                 if (count - thdr.count >= 0)
1203                         break;  /* fits with at least 25% to spare */
1204         }
1205         if (i >= 2) {
1206                 *action = 0;
1207                 return 0;
1208         }
1209
1210         /*
1211          * Make altpath point to the block we want to keep (the lower
1212          * numbered block) and path point to the block we want to drop.
1213          */
1214         memcpy(&state->altpath, &state->path, sizeof(state->path));
1215         if (blkno < blk->blkno) {
1216                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1217                                                  0, &retval);
1218         } else {
1219                 error = xfs_da3_path_shift(state, &state->path, forward,
1220                                                  0, &retval);
1221         }
1222         if (error)
1223                 return error;
1224         if (retval) {
1225                 *action = 0;
1226                 return 0;
1227         }
1228         *action = 1;
1229         return 0;
1230 }
1231
1232 /*
1233  * Pick up the last hashvalue from an intermediate node.
1234  */
1235 STATIC uint
1236 xfs_da3_node_lasthash(
1237         struct xfs_inode        *dp,
1238         struct xfs_buf          *bp,
1239         int                     *count)
1240 {
1241         struct xfs_da_intnode    *node;
1242         struct xfs_da_node_entry *btree;
1243         struct xfs_da3_icnode_hdr nodehdr;
1244
1245         node = bp->b_addr;
1246         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1247         if (count)
1248                 *count = nodehdr.count;
1249         if (!nodehdr.count)
1250                 return 0;
1251         btree = dp->d_ops->node_tree_p(node);
1252         return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1253 }
1254
1255 /*
1256  * Walk back up the tree adjusting hash values as necessary,
1257  * when we stop making changes, return.
1258  */
1259 void
1260 xfs_da3_fixhashpath(
1261         struct xfs_da_state     *state,
1262         struct xfs_da_state_path *path)
1263 {
1264         struct xfs_da_state_blk *blk;
1265         struct xfs_da_intnode   *node;
1266         struct xfs_da_node_entry *btree;
1267         xfs_dahash_t            lasthash=0;
1268         int                     level;
1269         int                     count;
1270         struct xfs_inode        *dp = state->args->dp;
1271
1272         trace_xfs_da_fixhashpath(state->args);
1273
1274         level = path->active-1;
1275         blk = &path->blk[ level ];
1276         switch (blk->magic) {
1277         case XFS_ATTR_LEAF_MAGIC:
1278                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1279                 if (count == 0)
1280                         return;
1281                 break;
1282         case XFS_DIR2_LEAFN_MAGIC:
1283                 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1284                 if (count == 0)
1285                         return;
1286                 break;
1287         case XFS_DA_NODE_MAGIC:
1288                 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1289                 if (count == 0)
1290                         return;
1291                 break;
1292         }
1293         for (blk--, level--; level >= 0; blk--, level--) {
1294                 struct xfs_da3_icnode_hdr nodehdr;
1295
1296                 node = blk->bp->b_addr;
1297                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1298                 btree = dp->d_ops->node_tree_p(node);
1299                 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1300                         break;
1301                 blk->hashval = lasthash;
1302                 btree[blk->index].hashval = cpu_to_be32(lasthash);
1303                 xfs_trans_log_buf(state->args->trans, blk->bp,
1304                                   XFS_DA_LOGRANGE(node, &btree[blk->index],
1305                                                   sizeof(*btree)));
1306
1307                 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1308         }
1309 }
1310
1311 /*
1312  * Remove an entry from an intermediate node.
1313  */
1314 STATIC void
1315 xfs_da3_node_remove(
1316         struct xfs_da_state     *state,
1317         struct xfs_da_state_blk *drop_blk)
1318 {
1319         struct xfs_da_intnode   *node;
1320         struct xfs_da3_icnode_hdr nodehdr;
1321         struct xfs_da_node_entry *btree;
1322         int                     index;
1323         int                     tmp;
1324         struct xfs_inode        *dp = state->args->dp;
1325
1326         trace_xfs_da_node_remove(state->args);
1327
1328         node = drop_blk->bp->b_addr;
1329         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1330         ASSERT(drop_blk->index < nodehdr.count);
1331         ASSERT(drop_blk->index >= 0);
1332
1333         /*
1334          * Copy over the offending entry, or just zero it out.
1335          */
1336         index = drop_blk->index;
1337         btree = dp->d_ops->node_tree_p(node);
1338         if (index < nodehdr.count - 1) {
1339                 tmp  = nodehdr.count - index - 1;
1340                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1341                 memmove(&btree[index], &btree[index + 1], tmp);
1342                 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1343                     XFS_DA_LOGRANGE(node, &btree[index], tmp));
1344                 index = nodehdr.count - 1;
1345         }
1346         memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1347         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1348             XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1349         nodehdr.count -= 1;
1350         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1351         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1352             XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1353
1354         /*
1355          * Copy the last hash value from the block to propagate upwards.
1356          */
1357         drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1358 }
1359
1360 /*
1361  * Unbalance the elements between two intermediate nodes,
1362  * move all Btree elements from one node into another.
1363  */
1364 STATIC void
1365 xfs_da3_node_unbalance(
1366         struct xfs_da_state     *state,
1367         struct xfs_da_state_blk *drop_blk,
1368         struct xfs_da_state_blk *save_blk)
1369 {
1370         struct xfs_da_intnode   *drop_node;
1371         struct xfs_da_intnode   *save_node;
1372         struct xfs_da_node_entry *drop_btree;
1373         struct xfs_da_node_entry *save_btree;
1374         struct xfs_da3_icnode_hdr drop_hdr;
1375         struct xfs_da3_icnode_hdr save_hdr;
1376         struct xfs_trans        *tp;
1377         int                     sindex;
1378         int                     tmp;
1379         struct xfs_inode        *dp = state->args->dp;
1380
1381         trace_xfs_da_node_unbalance(state->args);
1382
1383         drop_node = drop_blk->bp->b_addr;
1384         save_node = save_blk->bp->b_addr;
1385         dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1386         dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1387         drop_btree = dp->d_ops->node_tree_p(drop_node);
1388         save_btree = dp->d_ops->node_tree_p(save_node);
1389         tp = state->args->trans;
1390
1391         /*
1392          * If the dying block has lower hashvals, then move all the
1393          * elements in the remaining block up to make a hole.
1394          */
1395         if ((be32_to_cpu(drop_btree[0].hashval) <
1396                         be32_to_cpu(save_btree[0].hashval)) ||
1397             (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1398                         be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1399                 /* XXX: check this - is memmove dst correct? */
1400                 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1401                 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1402
1403                 sindex = 0;
1404                 xfs_trans_log_buf(tp, save_blk->bp,
1405                         XFS_DA_LOGRANGE(save_node, &save_btree[0],
1406                                 (save_hdr.count + drop_hdr.count) *
1407                                                 sizeof(xfs_da_node_entry_t)));
1408         } else {
1409                 sindex = save_hdr.count;
1410                 xfs_trans_log_buf(tp, save_blk->bp,
1411                         XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1412                                 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1413         }
1414
1415         /*
1416          * Move all the B-tree elements from drop_blk to save_blk.
1417          */
1418         tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1419         memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1420         save_hdr.count += drop_hdr.count;
1421
1422         dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1423         xfs_trans_log_buf(tp, save_blk->bp,
1424                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1425                                 dp->d_ops->node_hdr_size));
1426
1427         /*
1428          * Save the last hashval in the remaining block for upward propagation.
1429          */
1430         save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1431 }
1432
1433 /*========================================================================
1434  * Routines used for finding things in the Btree.
1435  *========================================================================*/
1436
1437 /*
1438  * Walk down the Btree looking for a particular filename, filling
1439  * in the state structure as we go.
1440  *
1441  * We will set the state structure to point to each of the elements
1442  * in each of the nodes where either the hashval is or should be.
1443  *
1444  * We support duplicate hashval's so for each entry in the current
1445  * node that could contain the desired hashval, descend.  This is a
1446  * pruned depth-first tree search.
1447  */
1448 int                                                     /* error */
1449 xfs_da3_node_lookup_int(
1450         struct xfs_da_state     *state,
1451         int                     *result)
1452 {
1453         struct xfs_da_state_blk *blk;
1454         struct xfs_da_blkinfo   *curr;
1455         struct xfs_da_intnode   *node;
1456         struct xfs_da_node_entry *btree;
1457         struct xfs_da3_icnode_hdr nodehdr;
1458         struct xfs_da_args      *args;
1459         xfs_dablk_t             blkno;
1460         xfs_dahash_t            hashval;
1461         xfs_dahash_t            btreehashval;
1462         int                     probe;
1463         int                     span;
1464         int                     max;
1465         int                     error;
1466         int                     retval;
1467         struct xfs_inode        *dp = state->args->dp;
1468
1469         args = state->args;
1470
1471         /*
1472          * Descend thru the B-tree searching each level for the right
1473          * node to use, until the right hashval is found.
1474          */
1475         blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1476         for (blk = &state->path.blk[0], state->path.active = 1;
1477                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1478                          blk++, state->path.active++) {
1479                 /*
1480                  * Read the next node down in the tree.
1481                  */
1482                 blk->blkno = blkno;
1483                 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1484                                         -1, &blk->bp, args->whichfork);
1485                 if (error) {
1486                         blk->blkno = 0;
1487                         state->path.active--;
1488                         return error;
1489                 }
1490                 curr = blk->bp->b_addr;
1491                 blk->magic = be16_to_cpu(curr->magic);
1492
1493                 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1494                     blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1495                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1496                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1497                         break;
1498                 }
1499
1500                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1501                     blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1502                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1503                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1504                                                                blk->bp, NULL);
1505                         break;
1506                 }
1507
1508                 blk->magic = XFS_DA_NODE_MAGIC;
1509
1510
1511                 /*
1512                  * Search an intermediate node for a match.
1513                  */
1514                 node = blk->bp->b_addr;
1515                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1516                 btree = dp->d_ops->node_tree_p(node);
1517
1518                 max = nodehdr.count;
1519                 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1520
1521                 /*
1522                  * Binary search.  (note: small blocks will skip loop)
1523                  */
1524                 probe = span = max / 2;
1525                 hashval = args->hashval;
1526                 while (span > 4) {
1527                         span /= 2;
1528                         btreehashval = be32_to_cpu(btree[probe].hashval);
1529                         if (btreehashval < hashval)
1530                                 probe += span;
1531                         else if (btreehashval > hashval)
1532                                 probe -= span;
1533                         else
1534                                 break;
1535                 }
1536                 ASSERT((probe >= 0) && (probe < max));
1537                 ASSERT((span <= 4) ||
1538                         (be32_to_cpu(btree[probe].hashval) == hashval));
1539
1540                 /*
1541                  * Since we may have duplicate hashval's, find the first
1542                  * matching hashval in the node.
1543                  */
1544                 while (probe > 0 &&
1545                        be32_to_cpu(btree[probe].hashval) >= hashval) {
1546                         probe--;
1547                 }
1548                 while (probe < max &&
1549                        be32_to_cpu(btree[probe].hashval) < hashval) {
1550                         probe++;
1551                 }
1552
1553                 /*
1554                  * Pick the right block to descend on.
1555                  */
1556                 if (probe == max) {
1557                         blk->index = max - 1;
1558                         blkno = be32_to_cpu(btree[max - 1].before);
1559                 } else {
1560                         blk->index = probe;
1561                         blkno = be32_to_cpu(btree[probe].before);
1562                 }
1563         }
1564
1565         /*
1566          * A leaf block that ends in the hashval that we are interested in
1567          * (final hashval == search hashval) means that the next block may
1568          * contain more entries with the same hashval, shift upward to the
1569          * next leaf and keep searching.
1570          */
1571         for (;;) {
1572                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1573                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1574                                                         &blk->index, state);
1575                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1576                         retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1577                         blk->index = args->index;
1578                         args->blkno = blk->blkno;
1579                 } else {
1580                         ASSERT(0);
1581                         return -EFSCORRUPTED;
1582                 }
1583                 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1584                     (blk->hashval == args->hashval)) {
1585                         error = xfs_da3_path_shift(state, &state->path, 1, 1,
1586                                                          &retval);
1587                         if (error)
1588                                 return error;
1589                         if (retval == 0) {
1590                                 continue;
1591                         } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1592                                 /* path_shift() gives ENOENT */
1593                                 retval = -ENOATTR;
1594                         }
1595                 }
1596                 break;
1597         }
1598         *result = retval;
1599         return 0;
1600 }
1601
1602 /*========================================================================
1603  * Utility routines.
1604  *========================================================================*/
1605
1606 /*
1607  * Compare two intermediate nodes for "order".
1608  */
1609 STATIC int
1610 xfs_da3_node_order(
1611         struct xfs_inode *dp,
1612         struct xfs_buf  *node1_bp,
1613         struct xfs_buf  *node2_bp)
1614 {
1615         struct xfs_da_intnode   *node1;
1616         struct xfs_da_intnode   *node2;
1617         struct xfs_da_node_entry *btree1;
1618         struct xfs_da_node_entry *btree2;
1619         struct xfs_da3_icnode_hdr node1hdr;
1620         struct xfs_da3_icnode_hdr node2hdr;
1621
1622         node1 = node1_bp->b_addr;
1623         node2 = node2_bp->b_addr;
1624         dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1625         dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1626         btree1 = dp->d_ops->node_tree_p(node1);
1627         btree2 = dp->d_ops->node_tree_p(node2);
1628
1629         if (node1hdr.count > 0 && node2hdr.count > 0 &&
1630             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1631              (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1632               be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1633                 return 1;
1634         }
1635         return 0;
1636 }
1637
1638 /*
1639  * Link a new block into a doubly linked list of blocks (of whatever type).
1640  */
1641 int                                                     /* error */
1642 xfs_da3_blk_link(
1643         struct xfs_da_state     *state,
1644         struct xfs_da_state_blk *old_blk,
1645         struct xfs_da_state_blk *new_blk)
1646 {
1647         struct xfs_da_blkinfo   *old_info;
1648         struct xfs_da_blkinfo   *new_info;
1649         struct xfs_da_blkinfo   *tmp_info;
1650         struct xfs_da_args      *args;
1651         struct xfs_buf          *bp;
1652         int                     before = 0;
1653         int                     error;
1654         struct xfs_inode        *dp = state->args->dp;
1655
1656         /*
1657          * Set up environment.
1658          */
1659         args = state->args;
1660         ASSERT(args != NULL);
1661         old_info = old_blk->bp->b_addr;
1662         new_info = new_blk->bp->b_addr;
1663         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1664                old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1665                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1666
1667         switch (old_blk->magic) {
1668         case XFS_ATTR_LEAF_MAGIC:
1669                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1670                 break;
1671         case XFS_DIR2_LEAFN_MAGIC:
1672                 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1673                 break;
1674         case XFS_DA_NODE_MAGIC:
1675                 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1676                 break;
1677         }
1678
1679         /*
1680          * Link blocks in appropriate order.
1681          */
1682         if (before) {
1683                 /*
1684                  * Link new block in before existing block.
1685                  */
1686                 trace_xfs_da_link_before(args);
1687                 new_info->forw = cpu_to_be32(old_blk->blkno);
1688                 new_info->back = old_info->back;
1689                 if (old_info->back) {
1690                         error = xfs_da3_node_read(args->trans, dp,
1691                                                 be32_to_cpu(old_info->back),
1692                                                 -1, &bp, args->whichfork);
1693                         if (error)
1694                                 return error;
1695                         ASSERT(bp != NULL);
1696                         tmp_info = bp->b_addr;
1697                         ASSERT(tmp_info->magic == old_info->magic);
1698                         ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1699                         tmp_info->forw = cpu_to_be32(new_blk->blkno);
1700                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1701                 }
1702                 old_info->back = cpu_to_be32(new_blk->blkno);
1703         } else {
1704                 /*
1705                  * Link new block in after existing block.
1706                  */
1707                 trace_xfs_da_link_after(args);
1708                 new_info->forw = old_info->forw;
1709                 new_info->back = cpu_to_be32(old_blk->blkno);
1710                 if (old_info->forw) {
1711                         error = xfs_da3_node_read(args->trans, dp,
1712                                                 be32_to_cpu(old_info->forw),
1713                                                 -1, &bp, args->whichfork);
1714                         if (error)
1715                                 return error;
1716                         ASSERT(bp != NULL);
1717                         tmp_info = bp->b_addr;
1718                         ASSERT(tmp_info->magic == old_info->magic);
1719                         ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1720                         tmp_info->back = cpu_to_be32(new_blk->blkno);
1721                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1722                 }
1723                 old_info->forw = cpu_to_be32(new_blk->blkno);
1724         }
1725
1726         xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1727         xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1728         return 0;
1729 }
1730
1731 /*
1732  * Unlink a block from a doubly linked list of blocks.
1733  */
1734 STATIC int                                              /* error */
1735 xfs_da3_blk_unlink(
1736         struct xfs_da_state     *state,
1737         struct xfs_da_state_blk *drop_blk,
1738         struct xfs_da_state_blk *save_blk)
1739 {
1740         struct xfs_da_blkinfo   *drop_info;
1741         struct xfs_da_blkinfo   *save_info;
1742         struct xfs_da_blkinfo   *tmp_info;
1743         struct xfs_da_args      *args;
1744         struct xfs_buf          *bp;
1745         int                     error;
1746
1747         /*
1748          * Set up environment.
1749          */
1750         args = state->args;
1751         ASSERT(args != NULL);
1752         save_info = save_blk->bp->b_addr;
1753         drop_info = drop_blk->bp->b_addr;
1754         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1755                save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1756                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1757         ASSERT(save_blk->magic == drop_blk->magic);
1758         ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1759                (be32_to_cpu(save_info->back) == drop_blk->blkno));
1760         ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1761                (be32_to_cpu(drop_info->back) == save_blk->blkno));
1762
1763         /*
1764          * Unlink the leaf block from the doubly linked chain of leaves.
1765          */
1766         if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1767                 trace_xfs_da_unlink_back(args);
1768                 save_info->back = drop_info->back;
1769                 if (drop_info->back) {
1770                         error = xfs_da3_node_read(args->trans, args->dp,
1771                                                 be32_to_cpu(drop_info->back),
1772                                                 -1, &bp, args->whichfork);
1773                         if (error)
1774                                 return error;
1775                         ASSERT(bp != NULL);
1776                         tmp_info = bp->b_addr;
1777                         ASSERT(tmp_info->magic == save_info->magic);
1778                         ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1779                         tmp_info->forw = cpu_to_be32(save_blk->blkno);
1780                         xfs_trans_log_buf(args->trans, bp, 0,
1781                                                     sizeof(*tmp_info) - 1);
1782                 }
1783         } else {
1784                 trace_xfs_da_unlink_forward(args);
1785                 save_info->forw = drop_info->forw;
1786                 if (drop_info->forw) {
1787                         error = xfs_da3_node_read(args->trans, args->dp,
1788                                                 be32_to_cpu(drop_info->forw),
1789                                                 -1, &bp, args->whichfork);
1790                         if (error)
1791                                 return error;
1792                         ASSERT(bp != NULL);
1793                         tmp_info = bp->b_addr;
1794                         ASSERT(tmp_info->magic == save_info->magic);
1795                         ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1796                         tmp_info->back = cpu_to_be32(save_blk->blkno);
1797                         xfs_trans_log_buf(args->trans, bp, 0,
1798                                                     sizeof(*tmp_info) - 1);
1799                 }
1800         }
1801
1802         xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1803         return 0;
1804 }
1805
1806 /*
1807  * Move a path "forward" or "!forward" one block at the current level.
1808  *
1809  * This routine will adjust a "path" to point to the next block
1810  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1811  * Btree, including updating pointers to the intermediate nodes between
1812  * the new bottom and the root.
1813  */
1814 int                                                     /* error */
1815 xfs_da3_path_shift(
1816         struct xfs_da_state     *state,
1817         struct xfs_da_state_path *path,
1818         int                     forward,
1819         int                     release,
1820         int                     *result)
1821 {
1822         struct xfs_da_state_blk *blk;
1823         struct xfs_da_blkinfo   *info;
1824         struct xfs_da_intnode   *node;
1825         struct xfs_da_args      *args;
1826         struct xfs_da_node_entry *btree;
1827         struct xfs_da3_icnode_hdr nodehdr;
1828         xfs_dablk_t             blkno = 0;
1829         int                     level;
1830         int                     error;
1831         struct xfs_inode        *dp = state->args->dp;
1832
1833         trace_xfs_da_path_shift(state->args);
1834
1835         /*
1836          * Roll up the Btree looking for the first block where our
1837          * current index is not at the edge of the block.  Note that
1838          * we skip the bottom layer because we want the sibling block.
1839          */
1840         args = state->args;
1841         ASSERT(args != NULL);
1842         ASSERT(path != NULL);
1843         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1844         level = (path->active-1) - 1;   /* skip bottom layer in path */
1845         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1846                 node = blk->bp->b_addr;
1847                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1848                 btree = dp->d_ops->node_tree_p(node);
1849
1850                 if (forward && (blk->index < nodehdr.count - 1)) {
1851                         blk->index++;
1852                         blkno = be32_to_cpu(btree[blk->index].before);
1853                         break;
1854                 } else if (!forward && (blk->index > 0)) {
1855                         blk->index--;
1856                         blkno = be32_to_cpu(btree[blk->index].before);
1857                         break;
1858                 }
1859         }
1860         if (level < 0) {
1861                 *result = -ENOENT;      /* we're out of our tree */
1862                 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1863                 return 0;
1864         }
1865
1866         /*
1867          * Roll down the edge of the subtree until we reach the
1868          * same depth we were at originally.
1869          */
1870         for (blk++, level++; level < path->active; blk++, level++) {
1871                 /*
1872                  * Release the old block.
1873                  * (if it's dirty, trans won't actually let go)
1874                  */
1875                 if (release)
1876                         xfs_trans_brelse(args->trans, blk->bp);
1877
1878                 /*
1879                  * Read the next child block.
1880                  */
1881                 blk->blkno = blkno;
1882                 error = xfs_da3_node_read(args->trans, dp, blkno, -1,
1883                                         &blk->bp, args->whichfork);
1884                 if (error)
1885                         return error;
1886                 info = blk->bp->b_addr;
1887                 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1888                        info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1889                        info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1890                        info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1891                        info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1892                        info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1893
1894
1895                 /*
1896                  * Note: we flatten the magic number to a single type so we
1897                  * don't have to compare against crc/non-crc types elsewhere.
1898                  */
1899                 switch (be16_to_cpu(info->magic)) {
1900                 case XFS_DA_NODE_MAGIC:
1901                 case XFS_DA3_NODE_MAGIC:
1902                         blk->magic = XFS_DA_NODE_MAGIC;
1903                         node = (xfs_da_intnode_t *)info;
1904                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1905                         btree = dp->d_ops->node_tree_p(node);
1906                         blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1907                         if (forward)
1908                                 blk->index = 0;
1909                         else
1910                                 blk->index = nodehdr.count - 1;
1911                         blkno = be32_to_cpu(btree[blk->index].before);
1912                         break;
1913                 case XFS_ATTR_LEAF_MAGIC:
1914                 case XFS_ATTR3_LEAF_MAGIC:
1915                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1916                         ASSERT(level == path->active-1);
1917                         blk->index = 0;
1918                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1919                         break;
1920                 case XFS_DIR2_LEAFN_MAGIC:
1921                 case XFS_DIR3_LEAFN_MAGIC:
1922                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1923                         ASSERT(level == path->active-1);
1924                         blk->index = 0;
1925                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1926                                                                blk->bp, NULL);
1927                         break;
1928                 default:
1929                         ASSERT(0);
1930                         break;
1931                 }
1932         }
1933         *result = 0;
1934         return 0;
1935 }
1936
1937
1938 /*========================================================================
1939  * Utility routines.
1940  *========================================================================*/
1941
1942 /*
1943  * Implement a simple hash on a character string.
1944  * Rotate the hash value by 7 bits, then XOR each character in.
1945  * This is implemented with some source-level loop unrolling.
1946  */
1947 xfs_dahash_t
1948 xfs_da_hashname(const __uint8_t *name, int namelen)
1949 {
1950         xfs_dahash_t hash;
1951
1952         /*
1953          * Do four characters at a time as long as we can.
1954          */
1955         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1956                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1957                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1958
1959         /*
1960          * Now do the rest of the characters.
1961          */
1962         switch (namelen) {
1963         case 3:
1964                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1965                        rol32(hash, 7 * 3);
1966         case 2:
1967                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1968         case 1:
1969                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1970         default: /* case 0: */
1971                 return hash;
1972         }
1973 }
1974
1975 enum xfs_dacmp
1976 xfs_da_compname(
1977         struct xfs_da_args *args,
1978         const unsigned char *name,
1979         int             len)
1980 {
1981         return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1982                                         XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1983 }
1984
1985 static xfs_dahash_t
1986 xfs_default_hashname(
1987         struct xfs_name *name)
1988 {
1989         return xfs_da_hashname(name->name, name->len);
1990 }
1991
1992 const struct xfs_nameops xfs_default_nameops = {
1993         .hashname       = xfs_default_hashname,
1994         .compname       = xfs_da_compname
1995 };
1996
1997 int
1998 xfs_da_grow_inode_int(
1999         struct xfs_da_args      *args,
2000         xfs_fileoff_t           *bno,
2001         int                     count)
2002 {
2003         struct xfs_trans        *tp = args->trans;
2004         struct xfs_inode        *dp = args->dp;
2005         int                     w = args->whichfork;
2006         xfs_rfsblock_t          nblks = dp->i_d.di_nblocks;
2007         struct xfs_bmbt_irec    map, *mapp;
2008         int                     nmap, error, got, i, mapi;
2009
2010         /*
2011          * Find a spot in the file space to put the new block.
2012          */
2013         error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2014         if (error)
2015                 return error;
2016
2017         /*
2018          * Try mapping it in one filesystem block.
2019          */
2020         nmap = 1;
2021         ASSERT(args->firstblock != NULL);
2022         error = xfs_bmapi_write(tp, dp, *bno, count,
2023                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2024                         args->firstblock, args->total, &map, &nmap,
2025                         args->flist);
2026         if (error)
2027                 return error;
2028
2029         ASSERT(nmap <= 1);
2030         if (nmap == 1) {
2031                 mapp = &map;
2032                 mapi = 1;
2033         } else if (nmap == 0 && count > 1) {
2034                 xfs_fileoff_t           b;
2035                 int                     c;
2036
2037                 /*
2038                  * If we didn't get it and the block might work if fragmented,
2039                  * try without the CONTIG flag.  Loop until we get it all.
2040                  */
2041                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2042                 for (b = *bno, mapi = 0; b < *bno + count; ) {
2043                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2044                         c = (int)(*bno + count - b);
2045                         error = xfs_bmapi_write(tp, dp, b, c,
2046                                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2047                                         args->firstblock, args->total,
2048                                         &mapp[mapi], &nmap, args->flist);
2049                         if (error)
2050                                 goto out_free_map;
2051                         if (nmap < 1)
2052                                 break;
2053                         mapi += nmap;
2054                         b = mapp[mapi - 1].br_startoff +
2055                             mapp[mapi - 1].br_blockcount;
2056                 }
2057         } else {
2058                 mapi = 0;
2059                 mapp = NULL;
2060         }
2061
2062         /*
2063          * Count the blocks we got, make sure it matches the total.
2064          */
2065         for (i = 0, got = 0; i < mapi; i++)
2066                 got += mapp[i].br_blockcount;
2067         if (got != count || mapp[0].br_startoff != *bno ||
2068             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2069             *bno + count) {
2070                 error = -ENOSPC;
2071                 goto out_free_map;
2072         }
2073
2074         /* account for newly allocated blocks in reserved blocks total */
2075         args->total -= dp->i_d.di_nblocks - nblks;
2076
2077 out_free_map:
2078         if (mapp != &map)
2079                 kmem_free(mapp);
2080         return error;
2081 }
2082
2083 /*
2084  * Add a block to the btree ahead of the file.
2085  * Return the new block number to the caller.
2086  */
2087 int
2088 xfs_da_grow_inode(
2089         struct xfs_da_args      *args,
2090         xfs_dablk_t             *new_blkno)
2091 {
2092         xfs_fileoff_t           bno;
2093         int                     error;
2094
2095         trace_xfs_da_grow_inode(args);
2096
2097         bno = args->geo->leafblk;
2098         error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2099         if (!error)
2100                 *new_blkno = (xfs_dablk_t)bno;
2101         return error;
2102 }
2103
2104 /*
2105  * Ick.  We need to always be able to remove a btree block, even
2106  * if there's no space reservation because the filesystem is full.
2107  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2108  * It swaps the target block with the last block in the file.  The
2109  * last block in the file can always be removed since it can't cause
2110  * a bmap btree split to do that.
2111  */
2112 STATIC int
2113 xfs_da3_swap_lastblock(
2114         struct xfs_da_args      *args,
2115         xfs_dablk_t             *dead_blknop,
2116         struct xfs_buf          **dead_bufp)
2117 {
2118         struct xfs_da_blkinfo   *dead_info;
2119         struct xfs_da_blkinfo   *sib_info;
2120         struct xfs_da_intnode   *par_node;
2121         struct xfs_da_intnode   *dead_node;
2122         struct xfs_dir2_leaf    *dead_leaf2;
2123         struct xfs_da_node_entry *btree;
2124         struct xfs_da3_icnode_hdr par_hdr;
2125         struct xfs_inode        *dp;
2126         struct xfs_trans        *tp;
2127         struct xfs_mount        *mp;
2128         struct xfs_buf          *dead_buf;
2129         struct xfs_buf          *last_buf;
2130         struct xfs_buf          *sib_buf;
2131         struct xfs_buf          *par_buf;
2132         xfs_dahash_t            dead_hash;
2133         xfs_fileoff_t           lastoff;
2134         xfs_dablk_t             dead_blkno;
2135         xfs_dablk_t             last_blkno;
2136         xfs_dablk_t             sib_blkno;
2137         xfs_dablk_t             par_blkno;
2138         int                     error;
2139         int                     w;
2140         int                     entno;
2141         int                     level;
2142         int                     dead_level;
2143
2144         trace_xfs_da_swap_lastblock(args);
2145
2146         dead_buf = *dead_bufp;
2147         dead_blkno = *dead_blknop;
2148         tp = args->trans;
2149         dp = args->dp;
2150         w = args->whichfork;
2151         ASSERT(w == XFS_DATA_FORK);
2152         mp = dp->i_mount;
2153         lastoff = args->geo->freeblk;
2154         error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2155         if (error)
2156                 return error;
2157         if (unlikely(lastoff == 0)) {
2158                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2159                                  mp);
2160                 return -EFSCORRUPTED;
2161         }
2162         /*
2163          * Read the last block in the btree space.
2164          */
2165         last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2166         error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2167         if (error)
2168                 return error;
2169         /*
2170          * Copy the last block into the dead buffer and log it.
2171          */
2172         memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2173         xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2174         dead_info = dead_buf->b_addr;
2175         /*
2176          * Get values from the moved block.
2177          */
2178         if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2179             dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2180                 struct xfs_dir3_icleaf_hdr leafhdr;
2181                 struct xfs_dir2_leaf_entry *ents;
2182
2183                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2184                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2185                 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2186                 dead_level = 0;
2187                 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2188         } else {
2189                 struct xfs_da3_icnode_hdr deadhdr;
2190
2191                 dead_node = (xfs_da_intnode_t *)dead_info;
2192                 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2193                 btree = dp->d_ops->node_tree_p(dead_node);
2194                 dead_level = deadhdr.level;
2195                 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2196         }
2197         sib_buf = par_buf = NULL;
2198         /*
2199          * If the moved block has a left sibling, fix up the pointers.
2200          */
2201         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2202                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2203                 if (error)
2204                         goto done;
2205                 sib_info = sib_buf->b_addr;
2206                 if (unlikely(
2207                     be32_to_cpu(sib_info->forw) != last_blkno ||
2208                     sib_info->magic != dead_info->magic)) {
2209                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2210                                          XFS_ERRLEVEL_LOW, mp);
2211                         error = -EFSCORRUPTED;
2212                         goto done;
2213                 }
2214                 sib_info->forw = cpu_to_be32(dead_blkno);
2215                 xfs_trans_log_buf(tp, sib_buf,
2216                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2217                                         sizeof(sib_info->forw)));
2218                 sib_buf = NULL;
2219         }
2220         /*
2221          * If the moved block has a right sibling, fix up the pointers.
2222          */
2223         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2224                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2225                 if (error)
2226                         goto done;
2227                 sib_info = sib_buf->b_addr;
2228                 if (unlikely(
2229                        be32_to_cpu(sib_info->back) != last_blkno ||
2230                        sib_info->magic != dead_info->magic)) {
2231                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2232                                          XFS_ERRLEVEL_LOW, mp);
2233                         error = -EFSCORRUPTED;
2234                         goto done;
2235                 }
2236                 sib_info->back = cpu_to_be32(dead_blkno);
2237                 xfs_trans_log_buf(tp, sib_buf,
2238                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2239                                         sizeof(sib_info->back)));
2240                 sib_buf = NULL;
2241         }
2242         par_blkno = args->geo->leafblk;
2243         level = -1;
2244         /*
2245          * Walk down the tree looking for the parent of the moved block.
2246          */
2247         for (;;) {
2248                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2249                 if (error)
2250                         goto done;
2251                 par_node = par_buf->b_addr;
2252                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2253                 if (level >= 0 && level != par_hdr.level + 1) {
2254                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2255                                          XFS_ERRLEVEL_LOW, mp);
2256                         error = -EFSCORRUPTED;
2257                         goto done;
2258                 }
2259                 level = par_hdr.level;
2260                 btree = dp->d_ops->node_tree_p(par_node);
2261                 for (entno = 0;
2262                      entno < par_hdr.count &&
2263                      be32_to_cpu(btree[entno].hashval) < dead_hash;
2264                      entno++)
2265                         continue;
2266                 if (entno == par_hdr.count) {
2267                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2268                                          XFS_ERRLEVEL_LOW, mp);
2269                         error = -EFSCORRUPTED;
2270                         goto done;
2271                 }
2272                 par_blkno = be32_to_cpu(btree[entno].before);
2273                 if (level == dead_level + 1)
2274                         break;
2275                 xfs_trans_brelse(tp, par_buf);
2276                 par_buf = NULL;
2277         }
2278         /*
2279          * We're in the right parent block.
2280          * Look for the right entry.
2281          */
2282         for (;;) {
2283                 for (;
2284                      entno < par_hdr.count &&
2285                      be32_to_cpu(btree[entno].before) != last_blkno;
2286                      entno++)
2287                         continue;
2288                 if (entno < par_hdr.count)
2289                         break;
2290                 par_blkno = par_hdr.forw;
2291                 xfs_trans_brelse(tp, par_buf);
2292                 par_buf = NULL;
2293                 if (unlikely(par_blkno == 0)) {
2294                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2295                                          XFS_ERRLEVEL_LOW, mp);
2296                         error = -EFSCORRUPTED;
2297                         goto done;
2298                 }
2299                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2300                 if (error)
2301                         goto done;
2302                 par_node = par_buf->b_addr;
2303                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2304                 if (par_hdr.level != level) {
2305                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2306                                          XFS_ERRLEVEL_LOW, mp);
2307                         error = -EFSCORRUPTED;
2308                         goto done;
2309                 }
2310                 btree = dp->d_ops->node_tree_p(par_node);
2311                 entno = 0;
2312         }
2313         /*
2314          * Update the parent entry pointing to the moved block.
2315          */
2316         btree[entno].before = cpu_to_be32(dead_blkno);
2317         xfs_trans_log_buf(tp, par_buf,
2318                 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2319                                 sizeof(btree[entno].before)));
2320         *dead_blknop = last_blkno;
2321         *dead_bufp = last_buf;
2322         return 0;
2323 done:
2324         if (par_buf)
2325                 xfs_trans_brelse(tp, par_buf);
2326         if (sib_buf)
2327                 xfs_trans_brelse(tp, sib_buf);
2328         xfs_trans_brelse(tp, last_buf);
2329         return error;
2330 }
2331
2332 /*
2333  * Remove a btree block from a directory or attribute.
2334  */
2335 int
2336 xfs_da_shrink_inode(
2337         xfs_da_args_t   *args,
2338         xfs_dablk_t     dead_blkno,
2339         struct xfs_buf  *dead_buf)
2340 {
2341         xfs_inode_t *dp;
2342         int done, error, w, count;
2343         xfs_trans_t *tp;
2344         xfs_mount_t *mp;
2345
2346         trace_xfs_da_shrink_inode(args);
2347
2348         dp = args->dp;
2349         w = args->whichfork;
2350         tp = args->trans;
2351         mp = dp->i_mount;
2352         count = args->geo->fsbcount;
2353         for (;;) {
2354                 /*
2355                  * Remove extents.  If we get ENOSPC for a dir we have to move
2356                  * the last block to the place we want to kill.
2357                  */
2358                 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2359                                     xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2360                                     0, args->firstblock, args->flist, &done);
2361                 if (error == -ENOSPC) {
2362                         if (w != XFS_DATA_FORK)
2363                                 break;
2364                         error = xfs_da3_swap_lastblock(args, &dead_blkno,
2365                                                       &dead_buf);
2366                         if (error)
2367                                 break;
2368                 } else {
2369                         break;
2370                 }
2371         }
2372         xfs_trans_binval(tp, dead_buf);
2373         return error;
2374 }
2375
2376 /*
2377  * See if the mapping(s) for this btree block are valid, i.e.
2378  * don't contain holes, are logically contiguous, and cover the whole range.
2379  */
2380 STATIC int
2381 xfs_da_map_covers_blocks(
2382         int             nmap,
2383         xfs_bmbt_irec_t *mapp,
2384         xfs_dablk_t     bno,
2385         int             count)
2386 {
2387         int             i;
2388         xfs_fileoff_t   off;
2389
2390         for (i = 0, off = bno; i < nmap; i++) {
2391                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2392                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
2393                         return 0;
2394                 }
2395                 if (off != mapp[i].br_startoff) {
2396                         return 0;
2397                 }
2398                 off += mapp[i].br_blockcount;
2399         }
2400         return off == bno + count;
2401 }
2402
2403 /*
2404  * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2405  *
2406  * For the single map case, it is assumed that the caller has provided a pointer
2407  * to a valid xfs_buf_map.  For the multiple map case, this function will
2408  * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2409  * map pointer with the allocated map.
2410  */
2411 static int
2412 xfs_buf_map_from_irec(
2413         struct xfs_mount        *mp,
2414         struct xfs_buf_map      **mapp,
2415         int                     *nmaps,
2416         struct xfs_bmbt_irec    *irecs,
2417         int                     nirecs)
2418 {
2419         struct xfs_buf_map      *map;
2420         int                     i;
2421
2422         ASSERT(*nmaps == 1);
2423         ASSERT(nirecs >= 1);
2424
2425         if (nirecs > 1) {
2426                 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2427                                   KM_SLEEP | KM_NOFS);
2428                 if (!map)
2429                         return -ENOMEM;
2430                 *mapp = map;
2431         }
2432
2433         *nmaps = nirecs;
2434         map = *mapp;
2435         for (i = 0; i < *nmaps; i++) {
2436                 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2437                        irecs[i].br_startblock != HOLESTARTBLOCK);
2438                 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2439                 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2440         }
2441         return 0;
2442 }
2443
2444 /*
2445  * Map the block we are given ready for reading. There are three possible return
2446  * values:
2447  *      -1 - will be returned if we land in a hole and mappedbno == -2 so the
2448  *           caller knows not to execute a subsequent read.
2449  *       0 - if we mapped the block successfully
2450  *      >0 - positive error number if there was an error.
2451  */
2452 static int
2453 xfs_dabuf_map(
2454         struct xfs_inode        *dp,
2455         xfs_dablk_t             bno,
2456         xfs_daddr_t             mappedbno,
2457         int                     whichfork,
2458         struct xfs_buf_map      **map,
2459         int                     *nmaps)
2460 {
2461         struct xfs_mount        *mp = dp->i_mount;
2462         int                     nfsb;
2463         int                     error = 0;
2464         struct xfs_bmbt_irec    irec;
2465         struct xfs_bmbt_irec    *irecs = &irec;
2466         int                     nirecs;
2467
2468         ASSERT(map && *map);
2469         ASSERT(*nmaps == 1);
2470
2471         if (whichfork == XFS_DATA_FORK)
2472                 nfsb = mp->m_dir_geo->fsbcount;
2473         else
2474                 nfsb = mp->m_attr_geo->fsbcount;
2475
2476         /*
2477          * Caller doesn't have a mapping.  -2 means don't complain
2478          * if we land in a hole.
2479          */
2480         if (mappedbno == -1 || mappedbno == -2) {
2481                 /*
2482                  * Optimize the one-block case.
2483                  */
2484                 if (nfsb != 1)
2485                         irecs = kmem_zalloc(sizeof(irec) * nfsb,
2486                                             KM_SLEEP | KM_NOFS);
2487
2488                 nirecs = nfsb;
2489                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2490                                        &nirecs, xfs_bmapi_aflag(whichfork));
2491                 if (error)
2492                         goto out;
2493         } else {
2494                 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2495                 irecs->br_startoff = (xfs_fileoff_t)bno;
2496                 irecs->br_blockcount = nfsb;
2497                 irecs->br_state = 0;
2498                 nirecs = 1;
2499         }
2500
2501         if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2502                 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2503                 if (unlikely(error == -EFSCORRUPTED)) {
2504                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2505                                 int i;
2506                                 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2507                                         __func__, (long long)bno,
2508                                         (long long)dp->i_ino);
2509                                 for (i = 0; i < *nmaps; i++) {
2510                                         xfs_alert(mp,
2511 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2512                                                 i,
2513                                                 (long long)irecs[i].br_startoff,
2514                                                 (long long)irecs[i].br_startblock,
2515                                                 (long long)irecs[i].br_blockcount,
2516                                                 irecs[i].br_state);
2517                                 }
2518                         }
2519                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2520                                          XFS_ERRLEVEL_LOW, mp);
2521                 }
2522                 goto out;
2523         }
2524         error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2525 out:
2526         if (irecs != &irec)
2527                 kmem_free(irecs);
2528         return error;
2529 }
2530
2531 /*
2532  * Get a buffer for the dir/attr block.
2533  */
2534 int
2535 xfs_da_get_buf(
2536         struct xfs_trans        *trans,
2537         struct xfs_inode        *dp,
2538         xfs_dablk_t             bno,
2539         xfs_daddr_t             mappedbno,
2540         struct xfs_buf          **bpp,
2541         int                     whichfork)
2542 {
2543         struct xfs_buf          *bp;
2544         struct xfs_buf_map      map;
2545         struct xfs_buf_map      *mapp;
2546         int                     nmap;
2547         int                     error;
2548
2549         *bpp = NULL;
2550         mapp = &map;
2551         nmap = 1;
2552         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2553                                 &mapp, &nmap);
2554         if (error) {
2555                 /* mapping a hole is not an error, but we don't continue */
2556                 if (error == -1)
2557                         error = 0;
2558                 goto out_free;
2559         }
2560
2561         bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2562                                     mapp, nmap, 0);
2563         error = bp ? bp->b_error : -EIO;
2564         if (error) {
2565                 if (bp)
2566                         xfs_trans_brelse(trans, bp);
2567                 goto out_free;
2568         }
2569
2570         *bpp = bp;
2571
2572 out_free:
2573         if (mapp != &map)
2574                 kmem_free(mapp);
2575
2576         return error;
2577 }
2578
2579 /*
2580  * Get a buffer for the dir/attr block, fill in the contents.
2581  */
2582 int
2583 xfs_da_read_buf(
2584         struct xfs_trans        *trans,
2585         struct xfs_inode        *dp,
2586         xfs_dablk_t             bno,
2587         xfs_daddr_t             mappedbno,
2588         struct xfs_buf          **bpp,
2589         int                     whichfork,
2590         const struct xfs_buf_ops *ops)
2591 {
2592         struct xfs_buf          *bp;
2593         struct xfs_buf_map      map;
2594         struct xfs_buf_map      *mapp;
2595         int                     nmap;
2596         int                     error;
2597
2598         *bpp = NULL;
2599         mapp = &map;
2600         nmap = 1;
2601         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2602                                 &mapp, &nmap);
2603         if (error) {
2604                 /* mapping a hole is not an error, but we don't continue */
2605                 if (error == -1)
2606                         error = 0;
2607                 goto out_free;
2608         }
2609
2610         error = xfs_trans_read_buf_map(dp->i_mount, trans,
2611                                         dp->i_mount->m_ddev_targp,
2612                                         mapp, nmap, 0, &bp, ops);
2613         if (error)
2614                 goto out_free;
2615
2616         if (whichfork == XFS_ATTR_FORK)
2617                 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2618         else
2619                 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2620         *bpp = bp;
2621 out_free:
2622         if (mapp != &map)
2623                 kmem_free(mapp);
2624
2625         return error;
2626 }
2627
2628 /*
2629  * Readahead the dir/attr block.
2630  */
2631 xfs_daddr_t
2632 xfs_da_reada_buf(
2633         struct xfs_inode        *dp,
2634         xfs_dablk_t             bno,
2635         xfs_daddr_t             mappedbno,
2636         int                     whichfork,
2637         const struct xfs_buf_ops *ops)
2638 {
2639         struct xfs_buf_map      map;
2640         struct xfs_buf_map      *mapp;
2641         int                     nmap;
2642         int                     error;
2643
2644         mapp = &map;
2645         nmap = 1;
2646         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2647                                 &mapp, &nmap);
2648         if (error) {
2649                 /* mapping a hole is not an error, but we don't continue */
2650                 if (error == -1)
2651                         error = 0;
2652                 goto out_free;
2653         }
2654
2655         mappedbno = mapp[0].bm_bn;
2656         xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2657
2658 out_free:
2659         if (mapp != &map)
2660                 kmem_free(mapp);
2661
2662         if (error)
2663                 return -1;
2664         return mappedbno;
2665 }