fs/xfs/libxfs/xfs_da_btree.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_shared.h"
  22 #include "xfs_format.h"
  23 #include "xfs_log_format.h"
  24 #include "xfs_trans_resv.h"
  25 #include "xfs_bit.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_dir2_priv.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_alloc.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_attr.h"
  37 #include "xfs_attr_leaf.h"
  38 #include "xfs_error.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_cksum.h"
  41 #include "xfs_buf_item.h"
  42
  43 /*
  44  * xfs_da_btree.c
  45  *
  46  * Routines to implement directories as Btrees of hashed names.
  47  */
  48
  49 /*========================================================================
  50  * Function prototypes for the kernel.
  51  *========================================================================*/
  52
  53 /*
  54  * Routines used for growing the Btree.
  55  */
  56 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
  57                                             xfs_da_state_blk_t *existing_root,
  58                                             xfs_da_state_blk_t *new_child);
  59 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
  60                                             xfs_da_state_blk_t *existing_blk,
  61                                             xfs_da_state_blk_t *split_blk,
  62                                             xfs_da_state_blk_t *blk_to_add,
  63                                             int treelevel,
  64                                             int *result);
  65 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
  66                                          xfs_da_state_blk_t *node_blk_1,
  67                                          xfs_da_state_blk_t *node_blk_2);
  68 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
  69                                    xfs_da_state_blk_t *old_node_blk,
  70                                    xfs_da_state_blk_t *new_node_blk);
  71
  72 /*
  73  * Routines used for shrinking the Btree.
  74  */
  75 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
  76                                            xfs_da_state_blk_t *root_blk);
  77 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
  78 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
  79                                               xfs_da_state_blk_t *drop_blk);
  80 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
  81                                          xfs_da_state_blk_t *src_node_blk,
  82                                          xfs_da_state_blk_t *dst_node_blk);
  83
  84 /*
  85  * Utility routines.
  86  */
  87 STATIC int      xfs_da3_blk_unlink(xfs_da_state_t *state,
  88                                   xfs_da_state_blk_t *drop_blk,
  89                                   xfs_da_state_blk_t *save_blk);
  90
  91
  92 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
  93
  94 /*
  95  * Allocate a dir-state structure.
  96  * We don't put them on the stack since they're large.
  97  */
  98 xfs_da_state_t *
  99 xfs_da_state_alloc(void)
 100 {
 101         return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
 102 }
 103
 104 /*
 105  * Kill the altpath contents of a da-state structure.
 106  */
 107 STATIC void
 108 xfs_da_state_kill_altpath(xfs_da_state_t *state)
 109 {
 110         int     i;
 111
 112         for (i = 0; i < state->altpath.active; i++)
 113                 state->altpath.blk[i].bp = NULL;
 114         state->altpath.active = 0;
 115 }
 116
 117 /*
 118  * Free a da-state structure.
 119  */
 120 void
 121 xfs_da_state_free(xfs_da_state_t *state)
 122 {
 123         xfs_da_state_kill_altpath(state);
 124 #ifdef DEBUG
 125         memset((char *)state, 0, sizeof(*state));
 126 #endif /* DEBUG */
 127         kmem_zone_free(xfs_da_state_zone, state);
 128 }
 129
 130 static bool
 131 xfs_da3_node_verify(
 132         struct xfs_buf          *bp)
 133 {
 134         struct xfs_mount        *mp = bp->b_target->bt_mount;
 135         struct xfs_da_intnode   *hdr = bp->b_addr;
 136         struct xfs_da3_icnode_hdr ichdr;
 137         const struct xfs_dir_ops *ops;
 138
 139         ops = xfs_dir_get_ops(mp, NULL);
 140
 141         ops->node_hdr_from_disk(&ichdr, hdr);
 142
 143         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 144                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 145
 146                 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
 147                         return false;
 148
 149                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
 150                         return false;
 151                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
 152                         return false;
 153         } else {
 154                 if (ichdr.magic != XFS_DA_NODE_MAGIC)
 155                         return false;
 156         }
 157         if (ichdr.level == 0)
 158                 return false;
 159         if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
 160                 return false;
 161         if (ichdr.count == 0)
 162                 return false;
 163
 164         /*
 165          * we don't know if the node is for and attribute or directory tree,
 166          * so only fail if the count is outside both bounds
 167          */
 168         if (ichdr.count > mp->m_dir_geo->node_ents &&
 169             ichdr.count > mp->m_attr_geo->node_ents)
 170                 return false;
 171
 172         /* XXX: hash order check? */
 173
 174         return true;
 175 }
 176
 177 static void
 178 xfs_da3_node_write_verify(
 179         struct xfs_buf  *bp)
 180 {
 181         struct xfs_mount        *mp = bp->b_target->bt_mount;
 182         struct xfs_buf_log_item *bip = bp->b_fspriv;
 183         struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 184
 185         if (!xfs_da3_node_verify(bp)) {
 186                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 187                 xfs_verifier_error(bp);
 188                 return;
 189         }
 190
 191         if (!xfs_sb_version_hascrc(&mp->m_sb))
 192                 return;
 193
 194         if (bip)
 195                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 196
 197         xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
 198 }
 199
 200 /*
 201  * leaf/node format detection on trees is sketchy, so a node read can be done on
 202  * leaf level blocks when detection identifies the tree as a node format tree
 203  * incorrectly. In this case, we need to swap the verifier to match the correct
 204  * format of the block being read.
 205  */
 206 static void
 207 xfs_da3_node_read_verify(
 208         struct xfs_buf          *bp)
 209 {
 210         struct xfs_da_blkinfo   *info = bp->b_addr;
 211
 212         switch (be16_to_cpu(info->magic)) {
 213                 case XFS_DA3_NODE_MAGIC:
 214                         if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
 215                                 xfs_buf_ioerror(bp, -EFSBADCRC);
 216                                 break;
 217                         }
 218                         /* fall through */
 219                 case XFS_DA_NODE_MAGIC:
 220                         if (!xfs_da3_node_verify(bp)) {
 221                                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 222                                 break;
 223                         }
 224                         return;
 225                 case XFS_ATTR_LEAF_MAGIC:
 226                 case XFS_ATTR3_LEAF_MAGIC:
 227                         bp->b_ops = &xfs_attr3_leaf_buf_ops;
 228                         bp->b_ops->verify_read(bp);
 229                         return;
 230                 case XFS_DIR2_LEAFN_MAGIC:
 231                 case XFS_DIR3_LEAFN_MAGIC:
 232                         bp->b_ops = &xfs_dir3_leafn_buf_ops;
 233                         bp->b_ops->verify_read(bp);
 234                         return;
 235                 default:
 236                         xfs_buf_ioerror(bp, -EFSCORRUPTED);
 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_meta_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_dir2_leaf    *leaf;
 517         xfs_dablk_t             blkno;
 518         int                     level;
 519         int                     error;
 520         int                     size;
 521
 522         trace_xfs_da_root_split(state->args);
 523
 524         /*
 525          * Copy the existing (incorrect) block from the root node position
 526          * to a free space somewhere.
 527          */
 528         args = state->args;
 529         error = xfs_da_grow_inode(args, &blkno);
 530         if (error)
 531                 return error;
 532
 533         dp = args->dp;
 534         tp = args->trans;
 535         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 536         if (error)
 537                 return error;
 538         node = bp->b_addr;
 539         oldroot = blk1->bp->b_addr;
 540         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
 541             oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
 542                 struct xfs_da3_icnode_hdr icnodehdr;
 543
 544                 dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
 545                 btree = dp->d_ops->node_tree_p(oldroot);
 546                 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
 547                 level = icnodehdr.level;
 548
 549                 /*
 550                  * we are about to copy oldroot to bp, so set up the type
 551                  * of bp while we know exactly what it will be.
 552                  */
 553                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 554         } else {
 555                 struct xfs_dir3_icleaf_hdr leafhdr;
 556                 struct xfs_dir2_leaf_entry *ents;
 557
 558                 leaf = (xfs_dir2_leaf_t *)oldroot;
 559                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 560                 ents = dp->d_ops->leaf_ents_p(leaf);
 561
 562                 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 563                        leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 564                 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
 565                 level = 0;
 566
 567                 /*
 568                  * we are about to copy oldroot to bp, so set up the type
 569                  * of bp while we know exactly what it will be.
 570                  */
 571                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 572         }
 573
 574         /*
 575          * we can copy most of the information in the node from one block to
 576          * another, but for CRC enabled headers we have to make sure that the
 577          * block specific identifiers are kept intact. We update the buffer
 578          * directly for this.
 579          */
 580         memcpy(node, oldroot, size);
 581         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
 582             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 583                 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
 584
 585                 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
 586         }
 587         xfs_trans_log_buf(tp, bp, 0, size - 1);
 588
 589         bp->b_ops = blk1->bp->b_ops;
 590         xfs_trans_buf_copy_type(bp, blk1->bp);
 591         blk1->bp = bp;
 592         blk1->blkno = blkno;
 593
 594         /*
 595          * Set up the new root node.
 596          */
 597         error = xfs_da3_node_create(args,
 598                 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
 599                 level + 1, &bp, args->whichfork);
 600         if (error)
 601                 return error;
 602
 603         node = bp->b_addr;
 604         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 605         btree = dp->d_ops->node_tree_p(node);
 606         btree[0].hashval = cpu_to_be32(blk1->hashval);
 607         btree[0].before = cpu_to_be32(blk1->blkno);
 608         btree[1].hashval = cpu_to_be32(blk2->hashval);
 609         btree[1].before = cpu_to_be32(blk2->blkno);
 610         nodehdr.count = 2;
 611         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 612
 613 #ifdef DEBUG
 614         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
 615             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 616                 ASSERT(blk1->blkno >= args->geo->leafblk &&
 617                        blk1->blkno < args->geo->freeblk);
 618                 ASSERT(blk2->blkno >= args->geo->leafblk &&
 619                        blk2->blkno < args->geo->freeblk);
 620         }
 621 #endif
 622
 623         /* Header is already logged by xfs_da_node_create */
 624         xfs_trans_log_buf(tp, bp,
 625                 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
 626
 627         return 0;
 628 }
 629
 630 /*
 631  * Split the node, rebalance, then add the new entry.
 632  */
 633 STATIC int                                              /* error */
 634 xfs_da3_node_split(
 635         struct xfs_da_state     *state,
 636         struct xfs_da_state_blk *oldblk,
 637         struct xfs_da_state_blk *newblk,
 638         struct xfs_da_state_blk *addblk,
 639         int                     treelevel,
 640         int                     *result)
 641 {
 642         struct xfs_da_intnode   *node;
 643         struct xfs_da3_icnode_hdr nodehdr;
 644         xfs_dablk_t             blkno;
 645         int                     newcount;
 646         int                     error;
 647         int                     useextra;
 648         struct xfs_inode        *dp = state->args->dp;
 649
 650         trace_xfs_da_node_split(state->args);
 651
 652         node = oldblk->bp->b_addr;
 653         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 654
 655         /*
 656          * With V2 dirs the extra block is data or freespace.
 657          */
 658         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 659         newcount = 1 + useextra;
 660         /*
 661          * Do we have to split the node?
 662          */
 663         if (nodehdr.count + newcount > state->args->geo->node_ents) {
 664                 /*
 665                  * Allocate a new node, add to the doubly linked chain of
 666                  * nodes, then move some of our excess entries into it.
 667                  */
 668                 error = xfs_da_grow_inode(state->args, &blkno);
 669                 if (error)
 670                         return error;   /* GROT: dir is inconsistent */
 671
 672                 error = xfs_da3_node_create(state->args, blkno, treelevel,
 673                                            &newblk->bp, state->args->whichfork);
 674                 if (error)
 675                         return error;   /* GROT: dir is inconsistent */
 676                 newblk->blkno = blkno;
 677                 newblk->magic = XFS_DA_NODE_MAGIC;
 678                 xfs_da3_node_rebalance(state, oldblk, newblk);
 679                 error = xfs_da3_blk_link(state, oldblk, newblk);
 680                 if (error)
 681                         return error;
 682                 *result = 1;
 683         } else {
 684                 *result = 0;
 685         }
 686
 687         /*
 688          * Insert the new entry(s) into the correct block
 689          * (updating last hashval in the process).
 690          *
 691          * xfs_da3_node_add() inserts BEFORE the given index,
 692          * and as a result of using node_lookup_int() we always
 693          * point to a valid entry (not after one), but a split
 694          * operation always results in a new block whose hashvals
 695          * FOLLOW the current block.
 696          *
 697          * If we had double-split op below us, then add the extra block too.
 698          */
 699         node = oldblk->bp->b_addr;
 700         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 701         if (oldblk->index <= nodehdr.count) {
 702                 oldblk->index++;
 703                 xfs_da3_node_add(state, oldblk, addblk);
 704                 if (useextra) {
 705                         if (state->extraafter)
 706                                 oldblk->index++;
 707                         xfs_da3_node_add(state, oldblk, &state->extrablk);
 708                         state->extravalid = 0;
 709                 }
 710         } else {
 711                 newblk->index++;
 712                 xfs_da3_node_add(state, newblk, addblk);
 713                 if (useextra) {
 714                         if (state->extraafter)
 715                                 newblk->index++;
 716                         xfs_da3_node_add(state, newblk, &state->extrablk);
 717                         state->extravalid = 0;
 718                 }
 719         }
 720
 721         return 0;
 722 }
 723
 724 /*
 725  * Balance the btree elements between two intermediate nodes,
 726  * usually one full and one empty.
 727  *
 728  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 729  */
 730 STATIC void
 731 xfs_da3_node_rebalance(
 732         struct xfs_da_state     *state,
 733         struct xfs_da_state_blk *blk1,
 734         struct xfs_da_state_blk *blk2)
 735 {
 736         struct xfs_da_intnode   *node1;
 737         struct xfs_da_intnode   *node2;
 738         struct xfs_da_intnode   *tmpnode;
 739         struct xfs_da_node_entry *btree1;
 740         struct xfs_da_node_entry *btree2;
 741         struct xfs_da_node_entry *btree_s;
 742         struct xfs_da_node_entry *btree_d;
 743         struct xfs_da3_icnode_hdr nodehdr1;
 744         struct xfs_da3_icnode_hdr nodehdr2;
 745         struct xfs_trans        *tp;
 746         int                     count;
 747         int                     tmp;
 748         int                     swap = 0;
 749         struct xfs_inode        *dp = state->args->dp;
 750
 751         trace_xfs_da_node_rebalance(state->args);
 752
 753         node1 = blk1->bp->b_addr;
 754         node2 = blk2->bp->b_addr;
 755         dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 756         dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 757         btree1 = dp->d_ops->node_tree_p(node1);
 758         btree2 = dp->d_ops->node_tree_p(node2);
 759
 760         /*
 761          * Figure out how many entries need to move, and in which direction.
 762          * Swap the nodes around if that makes it simpler.
 763          */
 764         if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
 765             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
 766              (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
 767                         be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
 768                 tmpnode = node1;
 769                 node1 = node2;
 770                 node2 = tmpnode;
 771                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 772                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 773                 btree1 = dp->d_ops->node_tree_p(node1);
 774                 btree2 = dp->d_ops->node_tree_p(node2);
 775                 swap = 1;
 776         }
 777
 778         count = (nodehdr1.count - nodehdr2.count) / 2;
 779         if (count == 0)
 780                 return;
 781         tp = state->args->trans;
 782         /*
 783          * Two cases: high-to-low and low-to-high.
 784          */
 785         if (count > 0) {
 786                 /*
 787                  * Move elements in node2 up to make a hole.
 788                  */
 789                 tmp = nodehdr2.count;
 790                 if (tmp > 0) {
 791                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 792                         btree_s = &btree2[0];
 793                         btree_d = &btree2[count];
 794                         memmove(btree_d, btree_s, tmp);
 795                 }
 796
 797                 /*
 798                  * Move the req'd B-tree elements from high in node1 to
 799                  * low in node2.
 800                  */
 801                 nodehdr2.count += count;
 802                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 803                 btree_s = &btree1[nodehdr1.count - count];
 804                 btree_d = &btree2[0];
 805                 memcpy(btree_d, btree_s, tmp);
 806                 nodehdr1.count -= count;
 807         } else {
 808                 /*
 809                  * Move the req'd B-tree elements from low in node2 to
 810                  * high in node1.
 811                  */
 812                 count = -count;
 813                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 814                 btree_s = &btree2[0];
 815                 btree_d = &btree1[nodehdr1.count];
 816                 memcpy(btree_d, btree_s, tmp);
 817                 nodehdr1.count += count;
 818
 819                 xfs_trans_log_buf(tp, blk1->bp,
 820                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 821
 822                 /*
 823                  * Move elements in node2 down to fill the hole.
 824                  */
 825                 tmp  = nodehdr2.count - count;
 826                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 827                 btree_s = &btree2[count];
 828                 btree_d = &btree2[0];
 829                 memmove(btree_d, btree_s, tmp);
 830                 nodehdr2.count -= count;
 831         }
 832
 833         /*
 834          * Log header of node 1 and all current bits of node 2.
 835          */
 836         dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
 837         xfs_trans_log_buf(tp, blk1->bp,
 838                 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
 839
 840         dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
 841         xfs_trans_log_buf(tp, blk2->bp,
 842                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 843                                 dp->d_ops->node_hdr_size +
 844                                 (sizeof(btree2[0]) * nodehdr2.count)));
 845
 846         /*
 847          * Record the last hashval from each block for upward propagation.
 848          * (note: don't use the swapped node pointers)
 849          */
 850         if (swap) {
 851                 node1 = blk1->bp->b_addr;
 852                 node2 = blk2->bp->b_addr;
 853                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 854                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 855                 btree1 = dp->d_ops->node_tree_p(node1);
 856                 btree2 = dp->d_ops->node_tree_p(node2);
 857         }
 858         blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
 859         blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
 860
 861         /*
 862          * Adjust the expected index for insertion.
 863          */
 864         if (blk1->index >= nodehdr1.count) {
 865                 blk2->index = blk1->index - nodehdr1.count;
 866                 blk1->index = nodehdr1.count + 1;       /* make it invalid */
 867         }
 868 }
 869
 870 /*
 871  * Add a new entry to an intermediate node.
 872  */
 873 STATIC void
 874 xfs_da3_node_add(
 875         struct xfs_da_state     *state,
 876         struct xfs_da_state_blk *oldblk,
 877         struct xfs_da_state_blk *newblk)
 878 {
 879         struct xfs_da_intnode   *node;
 880         struct xfs_da3_icnode_hdr nodehdr;
 881         struct xfs_da_node_entry *btree;
 882         int                     tmp;
 883         struct xfs_inode        *dp = state->args->dp;
 884
 885         trace_xfs_da_node_add(state->args);
 886
 887         node = oldblk->bp->b_addr;
 888         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 889         btree = dp->d_ops->node_tree_p(node);
 890
 891         ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
 892         ASSERT(newblk->blkno != 0);
 893         if (state->args->whichfork == XFS_DATA_FORK)
 894                 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
 895                        newblk->blkno < state->args->geo->freeblk);
 896
 897         /*
 898          * We may need to make some room before we insert the new node.
 899          */
 900         tmp = 0;
 901         if (oldblk->index < nodehdr.count) {
 902                 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
 903                 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
 904         }
 905         btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
 906         btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
 907         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 908                 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
 909                                 tmp + sizeof(*btree)));
 910
 911         nodehdr.count += 1;
 912         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 913         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 914                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 915
 916         /*
 917          * Copy the last hash value from the oldblk to propagate upwards.
 918          */
 919         oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
 920 }
 921
 922 /*========================================================================
 923  * Routines used for shrinking the Btree.
 924  *========================================================================*/
 925
 926 /*
 927  * Deallocate an empty leaf node, remove it from its parent,
 928  * possibly deallocating that block, etc...
 929  */
 930 int
 931 xfs_da3_join(
 932         struct xfs_da_state     *state)
 933 {
 934         struct xfs_da_state_blk *drop_blk;
 935         struct xfs_da_state_blk *save_blk;
 936         int                     action = 0;
 937         int                     error;
 938
 939         trace_xfs_da_join(state->args);
 940
 941         drop_blk = &state->path.blk[ state->path.active-1 ];
 942         save_blk = &state->altpath.blk[ state->path.active-1 ];
 943         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 944         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 945                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 946
 947         /*
 948          * Walk back up the tree joining/deallocating as necessary.
 949          * When we stop dropping blocks, break out.
 950          */
 951         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 952                  state->path.active--) {
 953                 /*
 954                  * See if we can combine the block with a neighbor.
 955                  *   (action == 0) => no options, just leave
 956                  *   (action == 1) => coalesce, then unlink
 957                  *   (action == 2) => block empty, unlink it
 958                  */
 959                 switch (drop_blk->magic) {
 960                 case XFS_ATTR_LEAF_MAGIC:
 961                         error = xfs_attr3_leaf_toosmall(state, &action);
 962                         if (error)
 963                                 return error;
 964                         if (action == 0)
 965                                 return 0;
 966                         xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
 967                         break;
 968                 case XFS_DIR2_LEAFN_MAGIC:
 969                         error = xfs_dir2_leafn_toosmall(state, &action);
 970                         if (error)
 971                                 return error;
 972                         if (action == 0)
 973                                 return 0;
 974                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 975                         break;
 976                 case XFS_DA_NODE_MAGIC:
 977                         /*
 978                          * Remove the offending node, fixup hashvals,
 979                          * check for a toosmall neighbor.
 980                          */
 981                         xfs_da3_node_remove(state, drop_blk);
 982                         xfs_da3_fixhashpath(state, &state->path);
 983                         error = xfs_da3_node_toosmall(state, &action);
 984                         if (error)
 985                                 return error;
 986                         if (action == 0)
 987                                 return 0;
 988                         xfs_da3_node_unbalance(state, drop_blk, save_blk);
 989                         break;
 990                 }
 991                 xfs_da3_fixhashpath(state, &state->altpath);
 992                 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
 993                 xfs_da_state_kill_altpath(state);
 994                 if (error)
 995                         return error;
 996                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 997                                                          drop_blk->bp);
 998                 drop_blk->bp = NULL;
 999                 if (error)
1000                         return error;
1001         }
1002         /*
1003          * We joined all the way to the top.  If it turns out that
1004          * we only have one entry in the root, make the child block
1005          * the new root.
1006          */
1007         xfs_da3_node_remove(state, drop_blk);
1008         xfs_da3_fixhashpath(state, &state->path);
1009         error = xfs_da3_root_join(state, &state->path.blk[0]);
1010         return error;
1011 }
1012
1013 #ifdef  DEBUG
1014 static void
1015 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1016 {
1017         __be16  magic = blkinfo->magic;
1018
1019         if (level == 1) {
1020                 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1021                        magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1022                        magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1023                        magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1024         } else {
1025                 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1026                        magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1027         }
1028         ASSERT(!blkinfo->forw);
1029         ASSERT(!blkinfo->back);
1030 }
1031 #else   /* !DEBUG */
1032 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1033 #endif  /* !DEBUG */
1034
1035 /*
1036  * We have only one entry in the root.  Copy the only remaining child of
1037  * the old root to block 0 as the new root node.
1038  */
1039 STATIC int
1040 xfs_da3_root_join(
1041         struct xfs_da_state     *state,
1042         struct xfs_da_state_blk *root_blk)
1043 {
1044         struct xfs_da_intnode   *oldroot;
1045         struct xfs_da_args      *args;
1046         xfs_dablk_t             child;
1047         struct xfs_buf          *bp;
1048         struct xfs_da3_icnode_hdr oldroothdr;
1049         struct xfs_da_node_entry *btree;
1050         int                     error;
1051         struct xfs_inode        *dp = state->args->dp;
1052
1053         trace_xfs_da_root_join(state->args);
1054
1055         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1056
1057         args = state->args;
1058         oldroot = root_blk->bp->b_addr;
1059         dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1060         ASSERT(oldroothdr.forw == 0);
1061         ASSERT(oldroothdr.back == 0);
1062
1063         /*
1064          * If the root has more than one child, then don't do anything.
1065          */
1066         if (oldroothdr.count > 1)
1067                 return 0;
1068
1069         /*
1070          * Read in the (only) child block, then copy those bytes into
1071          * the root block's buffer and free the original child block.
1072          */
1073         btree = dp->d_ops->node_tree_p(oldroot);
1074         child = be32_to_cpu(btree[0].before);
1075         ASSERT(child != 0);
1076         error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1077                                              args->whichfork);
1078         if (error)
1079                 return error;
1080         xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1081
1082         /*
1083          * This could be copying a leaf back into the root block in the case of
1084          * there only being a single leaf block left in the tree. Hence we have
1085          * to update the b_ops pointer as well to match the buffer type change
1086          * that could occur. For dir3 blocks we also need to update the block
1087          * number in the buffer header.
1088          */
1089         memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1090         root_blk->bp->b_ops = bp->b_ops;
1091         xfs_trans_buf_copy_type(root_blk->bp, bp);
1092         if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1093                 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1094                 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1095         }
1096         xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1097                           args->geo->blksize - 1);
1098         error = xfs_da_shrink_inode(args, child, bp);
1099         return error;
1100 }
1101
1102 /*
1103  * Check a node block and its neighbors to see if the block should be
1104  * collapsed into one or the other neighbor.  Always keep the block
1105  * with the smaller block number.
1106  * If the current block is over 50% full, don't try to join it, return 0.
1107  * If the block is empty, fill in the state structure and return 2.
1108  * If it can be collapsed, fill in the state structure and return 1.
1109  * If nothing can be done, return 0.
1110  */
1111 STATIC int
1112 xfs_da3_node_toosmall(
1113         struct xfs_da_state     *state,
1114         int                     *action)
1115 {
1116         struct xfs_da_intnode   *node;
1117         struct xfs_da_state_blk *blk;
1118         struct xfs_da_blkinfo   *info;
1119         xfs_dablk_t             blkno;
1120         struct xfs_buf          *bp;
1121         struct xfs_da3_icnode_hdr nodehdr;
1122         int                     count;
1123         int                     forward;
1124         int                     error;
1125         int                     retval;
1126         int                     i;
1127         struct xfs_inode        *dp = state->args->dp;
1128
1129         trace_xfs_da_node_toosmall(state->args);
1130
1131         /*
1132          * Check for the degenerate case of the block being over 50% full.
1133          * If so, it's not worth even looking to see if we might be able
1134          * to coalesce with a sibling.
1135          */
1136         blk = &state->path.blk[ state->path.active-1 ];
1137         info = blk->bp->b_addr;
1138         node = (xfs_da_intnode_t *)info;
1139         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1140         if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1141                 *action = 0;    /* blk over 50%, don't try to join */
1142                 return 0;       /* blk over 50%, don't try to join */
1143         }
1144
1145         /*
1146          * Check for the degenerate case of the block being empty.
1147          * If the block is empty, we'll simply delete it, no need to
1148          * coalesce it with a sibling block.  We choose (arbitrarily)
1149          * to merge with the forward block unless it is NULL.
1150          */
1151         if (nodehdr.count == 0) {
1152                 /*
1153                  * Make altpath point to the block we want to keep and
1154                  * path point to the block we want to drop (this one).
1155                  */
1156                 forward = (info->forw != 0);
1157                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1158                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1159                                                  0, &retval);
1160                 if (error)
1161                         return error;
1162                 if (retval) {
1163                         *action = 0;
1164                 } else {
1165                         *action = 2;
1166                 }
1167                 return 0;
1168         }
1169
1170         /*
1171          * Examine each sibling block to see if we can coalesce with
1172          * at least 25% free space to spare.  We need to figure out
1173          * whether to merge with the forward or the backward block.
1174          * We prefer coalescing with the lower numbered sibling so as
1175          * to shrink a directory over time.
1176          */
1177         count  = state->args->geo->node_ents;
1178         count -= state->args->geo->node_ents >> 2;
1179         count -= nodehdr.count;
1180
1181         /* start with smaller blk num */
1182         forward = nodehdr.forw < nodehdr.back;
1183         for (i = 0; i < 2; forward = !forward, i++) {
1184                 struct xfs_da3_icnode_hdr thdr;
1185                 if (forward)
1186                         blkno = nodehdr.forw;
1187                 else
1188                         blkno = nodehdr.back;
1189                 if (blkno == 0)
1190                         continue;
1191                 error = xfs_da3_node_read(state->args->trans, dp,
1192                                         blkno, -1, &bp, state->args->whichfork);
1193                 if (error)
1194                         return error;
1195
1196                 node = bp->b_addr;
1197                 dp->d_ops->node_hdr_from_disk(&thdr, node);
1198                 xfs_trans_brelse(state->args->trans, bp);
1199
1200                 if (count - thdr.count >= 0)
1201                         break;  /* fits with at least 25% to spare */
1202         }
1203         if (i >= 2) {
1204                 *action = 0;
1205                 return 0;
1206         }
1207
1208         /*
1209          * Make altpath point to the block we want to keep (the lower
1210          * numbered block) and path point to the block we want to drop.
1211          */
1212         memcpy(&state->altpath, &state->path, sizeof(state->path));
1213         if (blkno < blk->blkno) {
1214                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1215                                                  0, &retval);
1216         } else {
1217                 error = xfs_da3_path_shift(state, &state->path, forward,
1218                                                  0, &retval);
1219         }
1220         if (error)
1221                 return error;
1222         if (retval) {
1223                 *action = 0;
1224                 return 0;
1225         }
1226         *action = 1;
1227         return 0;
1228 }
1229
1230 /*
1231  * Pick up the last hashvalue from an intermediate node.
1232  */
1233 STATIC uint
1234 xfs_da3_node_lasthash(
1235         struct xfs_inode        *dp,
1236         struct xfs_buf          *bp,
1237         int                     *count)
1238 {
1239         struct xfs_da_intnode    *node;
1240         struct xfs_da_node_entry *btree;
1241         struct xfs_da3_icnode_hdr nodehdr;
1242
1243         node = bp->b_addr;
1244         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1245         if (count)
1246                 *count = nodehdr.count;
1247         if (!nodehdr.count)
1248                 return 0;
1249         btree = dp->d_ops->node_tree_p(node);
1250         return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1251 }
1252
1253 /*
1254  * Walk back up the tree adjusting hash values as necessary,
1255  * when we stop making changes, return.
1256  */
1257 void
1258 xfs_da3_fixhashpath(
1259         struct xfs_da_state     *state,
1260         struct xfs_da_state_path *path)
1261 {
1262         struct xfs_da_state_blk *blk;
1263         struct xfs_da_intnode   *node;
1264         struct xfs_da_node_entry *btree;
1265         xfs_dahash_t            lasthash=0;
1266         int                     level;
1267         int                     count;
1268         struct xfs_inode        *dp = state->args->dp;
1269
1270         trace_xfs_da_fixhashpath(state->args);
1271
1272         level = path->active-1;
1273         blk = &path->blk[ level ];
1274         switch (blk->magic) {
1275         case XFS_ATTR_LEAF_MAGIC:
1276                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1277                 if (count == 0)
1278                         return;
1279                 break;
1280         case XFS_DIR2_LEAFN_MAGIC:
1281                 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1282                 if (count == 0)
1283                         return;
1284                 break;
1285         case XFS_DA_NODE_MAGIC:
1286                 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1287                 if (count == 0)
1288                         return;
1289                 break;
1290         }
1291         for (blk--, level--; level >= 0; blk--, level--) {
1292                 struct xfs_da3_icnode_hdr nodehdr;
1293
1294                 node = blk->bp->b_addr;
1295                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1296                 btree = dp->d_ops->node_tree_p(node);
1297                 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1298                         break;
1299                 blk->hashval = lasthash;
1300                 btree[blk->index].hashval = cpu_to_be32(lasthash);
1301                 xfs_trans_log_buf(state->args->trans, blk->bp,
1302                                   XFS_DA_LOGRANGE(node, &btree[blk->index],
1303                                                   sizeof(*btree)));
1304
1305                 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1306         }
1307 }
1308
1309 /*
1310  * Remove an entry from an intermediate node.
1311  */
1312 STATIC void
1313 xfs_da3_node_remove(
1314         struct xfs_da_state     *state,
1315         struct xfs_da_state_blk *drop_blk)
1316 {
1317         struct xfs_da_intnode   *node;
1318         struct xfs_da3_icnode_hdr nodehdr;
1319         struct xfs_da_node_entry *btree;
1320         int                     index;
1321         int                     tmp;
1322         struct xfs_inode        *dp = state->args->dp;
1323
1324         trace_xfs_da_node_remove(state->args);
1325
1326         node = drop_blk->bp->b_addr;
1327         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1328         ASSERT(drop_blk->index < nodehdr.count);
1329         ASSERT(drop_blk->index >= 0);
1330
1331         /*
1332          * Copy over the offending entry, or just zero it out.
1333          */
1334         index = drop_blk->index;
1335         btree = dp->d_ops->node_tree_p(node);
1336         if (index < nodehdr.count - 1) {
1337                 tmp  = nodehdr.count - index - 1;
1338                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1339                 memmove(&btree[index], &btree[index + 1], tmp);
1340                 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1341                     XFS_DA_LOGRANGE(node, &btree[index], tmp));
1342                 index = nodehdr.count - 1;
1343         }
1344         memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1345         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1346             XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1347         nodehdr.count -= 1;
1348         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1349         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1350             XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1351
1352         /*
1353          * Copy the last hash value from the block to propagate upwards.
1354          */
1355         drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1356 }
1357
1358 /*
1359  * Unbalance the elements between two intermediate nodes,
1360  * move all Btree elements from one node into another.
1361  */
1362 STATIC void
1363 xfs_da3_node_unbalance(
1364         struct xfs_da_state     *state,
1365         struct xfs_da_state_blk *drop_blk,
1366         struct xfs_da_state_blk *save_blk)
1367 {
1368         struct xfs_da_intnode   *drop_node;
1369         struct xfs_da_intnode   *save_node;
1370         struct xfs_da_node_entry *drop_btree;
1371         struct xfs_da_node_entry *save_btree;
1372         struct xfs_da3_icnode_hdr drop_hdr;
1373         struct xfs_da3_icnode_hdr save_hdr;
1374         struct xfs_trans        *tp;
1375         int                     sindex;
1376         int                     tmp;
1377         struct xfs_inode        *dp = state->args->dp;
1378
1379         trace_xfs_da_node_unbalance(state->args);
1380
1381         drop_node = drop_blk->bp->b_addr;
1382         save_node = save_blk->bp->b_addr;
1383         dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1384         dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1385         drop_btree = dp->d_ops->node_tree_p(drop_node);
1386         save_btree = dp->d_ops->node_tree_p(save_node);
1387         tp = state->args->trans;
1388
1389         /*
1390          * If the dying block has lower hashvals, then move all the
1391          * elements in the remaining block up to make a hole.
1392          */
1393         if ((be32_to_cpu(drop_btree[0].hashval) <
1394                         be32_to_cpu(save_btree[0].hashval)) ||
1395             (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1396                         be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1397                 /* XXX: check this - is memmove dst correct? */
1398                 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1399                 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1400
1401                 sindex = 0;
1402                 xfs_trans_log_buf(tp, save_blk->bp,
1403                         XFS_DA_LOGRANGE(save_node, &save_btree[0],
1404                                 (save_hdr.count + drop_hdr.count) *
1405                                                 sizeof(xfs_da_node_entry_t)));
1406         } else {
1407                 sindex = save_hdr.count;
1408                 xfs_trans_log_buf(tp, save_blk->bp,
1409                         XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1410                                 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1411         }
1412
1413         /*
1414          * Move all the B-tree elements from drop_blk to save_blk.
1415          */
1416         tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1417         memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1418         save_hdr.count += drop_hdr.count;
1419
1420         dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1421         xfs_trans_log_buf(tp, save_blk->bp,
1422                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1423                                 dp->d_ops->node_hdr_size));
1424
1425         /*
1426          * Save the last hashval in the remaining block for upward propagation.
1427          */
1428         save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1429 }
1430
1431 /*========================================================================
1432  * Routines used for finding things in the Btree.
1433  *========================================================================*/
1434
1435 /*
1436  * Walk down the Btree looking for a particular filename, filling
1437  * in the state structure as we go.
1438  *
1439  * We will set the state structure to point to each of the elements
1440  * in each of the nodes where either the hashval is or should be.
1441  *
1442  * We support duplicate hashval's so for each entry in the current
1443  * node that could contain the desired hashval, descend.  This is a
1444  * pruned depth-first tree search.
1445  */
1446 int                                                     /* error */
1447 xfs_da3_node_lookup_int(
1448         struct xfs_da_state     *state,
1449         int                     *result)
1450 {
1451         struct xfs_da_state_blk *blk;
1452         struct xfs_da_blkinfo   *curr;
1453         struct xfs_da_intnode   *node;
1454         struct xfs_da_node_entry *btree;
1455         struct xfs_da3_icnode_hdr nodehdr;
1456         struct xfs_da_args      *args;
1457         xfs_dablk_t             blkno;
1458         xfs_dahash_t            hashval;
1459         xfs_dahash_t            btreehashval;
1460         int                     probe;
1461         int                     span;
1462         int                     max;
1463         int                     error;
1464         int                     retval;
1465         struct xfs_inode        *dp = state->args->dp;
1466
1467         args = state->args;
1468
1469         /*
1470          * Descend thru the B-tree searching each level for the right
1471          * node to use, until the right hashval is found.
1472          */
1473         blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1474         for (blk = &state->path.blk[0], state->path.active = 1;
1475                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1476                          blk++, state->path.active++) {
1477                 /*
1478                  * Read the next node down in the tree.
1479                  */
1480                 blk->blkno = blkno;
1481                 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1482                                         -1, &blk->bp, args->whichfork);
1483                 if (error) {
1484                         blk->blkno = 0;
1485                         state->path.active--;
1486                         return error;
1487                 }
1488                 curr = blk->bp->b_addr;
1489                 blk->magic = be16_to_cpu(curr->magic);
1490
1491                 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1492                     blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1493                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1494                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1495                         break;
1496                 }
1497
1498                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1499                     blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1500                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1501                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1502                                                                blk->bp, NULL);
1503                         break;
1504                 }
1505
1506                 blk->magic = XFS_DA_NODE_MAGIC;
1507
1508
1509                 /*
1510                  * Search an intermediate node for a match.
1511                  */
1512                 node = blk->bp->b_addr;
1513                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1514                 btree = dp->d_ops->node_tree_p(node);
1515
1516                 max = nodehdr.count;
1517                 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1518
1519                 /*
1520                  * Binary search.  (note: small blocks will skip loop)
1521                  */
1522                 probe = span = max / 2;
1523                 hashval = args->hashval;
1524                 while (span > 4) {
1525                         span /= 2;
1526                         btreehashval = be32_to_cpu(btree[probe].hashval);
1527                         if (btreehashval < hashval)
1528                                 probe += span;
1529                         else if (btreehashval > hashval)
1530                                 probe -= span;
1531                         else
1532                                 break;
1533                 }
1534                 ASSERT((probe >= 0) && (probe < max));
1535                 ASSERT((span <= 4) ||
1536                         (be32_to_cpu(btree[probe].hashval) == hashval));
1537
1538                 /*
1539                  * Since we may have duplicate hashval's, find the first
1540                  * matching hashval in the node.
1541                  */
1542                 while (probe > 0 &&
1543                        be32_to_cpu(btree[probe].hashval) >= hashval) {
1544                         probe--;
1545                 }
1546                 while (probe < max &&
1547                        be32_to_cpu(btree[probe].hashval) < hashval) {
1548                         probe++;
1549                 }
1550
1551                 /*
1552                  * Pick the right block to descend on.
1553                  */
1554                 if (probe == max) {
1555                         blk->index = max - 1;
1556                         blkno = be32_to_cpu(btree[max - 1].before);
1557                 } else {
1558                         blk->index = probe;
1559                         blkno = be32_to_cpu(btree[probe].before);
1560                 }
1561         }
1562
1563         /*
1564          * A leaf block that ends in the hashval that we are interested in
1565          * (final hashval == search hashval) means that the next block may
1566          * contain more entries with the same hashval, shift upward to the
1567          * next leaf and keep searching.
1568          */
1569         for (;;) {
1570                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1571                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1572                                                         &blk->index, state);
1573                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1574                         retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1575                         blk->index = args->index;
1576                         args->blkno = blk->blkno;
1577                 } else {
1578                         ASSERT(0);
1579                         return -EFSCORRUPTED;
1580                 }
1581                 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1582                     (blk->hashval == args->hashval)) {
1583                         error = xfs_da3_path_shift(state, &state->path, 1, 1,
1584                                                          &retval);
1585                         if (error)
1586                                 return error;
1587                         if (retval == 0) {
1588                                 continue;
1589                         } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1590                                 /* path_shift() gives ENOENT */
1591                                 retval = -ENOATTR;
1592                         }
1593                 }
1594                 break;
1595         }
1596         *result = retval;
1597         return 0;
1598 }
1599
1600 /*========================================================================
1601  * Utility routines.
1602  *========================================================================*/
1603
1604 /*
1605  * Compare two intermediate nodes for "order".
1606  */
1607 STATIC int
1608 xfs_da3_node_order(
1609         struct xfs_inode *dp,
1610         struct xfs_buf  *node1_bp,
1611         struct xfs_buf  *node2_bp)
1612 {
1613         struct xfs_da_intnode   *node1;
1614         struct xfs_da_intnode   *node2;
1615         struct xfs_da_node_entry *btree1;
1616         struct xfs_da_node_entry *btree2;
1617         struct xfs_da3_icnode_hdr node1hdr;
1618         struct xfs_da3_icnode_hdr node2hdr;
1619
1620         node1 = node1_bp->b_addr;
1621         node2 = node2_bp->b_addr;
1622         dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1623         dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1624         btree1 = dp->d_ops->node_tree_p(node1);
1625         btree2 = dp->d_ops->node_tree_p(node2);
1626
1627         if (node1hdr.count > 0 && node2hdr.count > 0 &&
1628             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1629              (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1630               be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1631                 return 1;
1632         }
1633         return 0;
1634 }
1635
1636 /*
1637  * Link a new block into a doubly linked list of blocks (of whatever type).
1638  */
1639 int                                                     /* error */
1640 xfs_da3_blk_link(
1641         struct xfs_da_state     *state,
1642         struct xfs_da_state_blk *old_blk,
1643         struct xfs_da_state_blk *new_blk)
1644 {
1645         struct xfs_da_blkinfo   *old_info;
1646         struct xfs_da_blkinfo   *new_info;
1647         struct xfs_da_blkinfo   *tmp_info;
1648         struct xfs_da_args      *args;
1649         struct xfs_buf          *bp;
1650         int                     before = 0;
1651         int                     error;
1652         struct xfs_inode        *dp = state->args->dp;
1653
1654         /*
1655          * Set up environment.
1656          */
1657         args = state->args;
1658         ASSERT(args != NULL);
1659         old_info = old_blk->bp->b_addr;
1660         new_info = new_blk->bp->b_addr;
1661         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1662                old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1663                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1664
1665         switch (old_blk->magic) {
1666         case XFS_ATTR_LEAF_MAGIC:
1667                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1668                 break;
1669         case XFS_DIR2_LEAFN_MAGIC:
1670                 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1671                 break;
1672         case XFS_DA_NODE_MAGIC:
1673                 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1674                 break;
1675         }
1676
1677         /*
1678          * Link blocks in appropriate order.
1679          */
1680         if (before) {
1681                 /*
1682                  * Link new block in before existing block.
1683                  */
1684                 trace_xfs_da_link_before(args);
1685                 new_info->forw = cpu_to_be32(old_blk->blkno);
1686                 new_info->back = old_info->back;
1687                 if (old_info->back) {
1688                         error = xfs_da3_node_read(args->trans, dp,
1689                                                 be32_to_cpu(old_info->back),
1690                                                 -1, &bp, args->whichfork);
1691                         if (error)
1692                                 return error;
1693                         ASSERT(bp != NULL);
1694                         tmp_info = bp->b_addr;
1695                         ASSERT(tmp_info->magic == old_info->magic);
1696                         ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1697                         tmp_info->forw = cpu_to_be32(new_blk->blkno);
1698                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1699                 }
1700                 old_info->back = cpu_to_be32(new_blk->blkno);
1701         } else {
1702                 /*
1703                  * Link new block in after existing block.
1704                  */
1705                 trace_xfs_da_link_after(args);
1706                 new_info->forw = old_info->forw;
1707                 new_info->back = cpu_to_be32(old_blk->blkno);
1708                 if (old_info->forw) {
1709                         error = xfs_da3_node_read(args->trans, dp,
1710                                                 be32_to_cpu(old_info->forw),
1711                                                 -1, &bp, args->whichfork);
1712                         if (error)
1713                                 return error;
1714                         ASSERT(bp != NULL);
1715                         tmp_info = bp->b_addr;
1716                         ASSERT(tmp_info->magic == old_info->magic);
1717                         ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1718                         tmp_info->back = cpu_to_be32(new_blk->blkno);
1719                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1720                 }
1721                 old_info->forw = cpu_to_be32(new_blk->blkno);
1722         }
1723
1724         xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1725         xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1726         return 0;
1727 }
1728
1729 /*
1730  * Unlink a block from a doubly linked list of blocks.
1731  */
1732 STATIC int                                              /* error */
1733 xfs_da3_blk_unlink(
1734         struct xfs_da_state     *state,
1735         struct xfs_da_state_blk *drop_blk,
1736         struct xfs_da_state_blk *save_blk)
1737 {
1738         struct xfs_da_blkinfo   *drop_info;
1739         struct xfs_da_blkinfo   *save_info;
1740         struct xfs_da_blkinfo   *tmp_info;
1741         struct xfs_da_args      *args;
1742         struct xfs_buf          *bp;
1743         int                     error;
1744
1745         /*
1746          * Set up environment.
1747          */
1748         args = state->args;
1749         ASSERT(args != NULL);
1750         save_info = save_blk->bp->b_addr;
1751         drop_info = drop_blk->bp->b_addr;
1752         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1753                save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1754                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1755         ASSERT(save_blk->magic == drop_blk->magic);
1756         ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1757                (be32_to_cpu(save_info->back) == drop_blk->blkno));
1758         ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1759                (be32_to_cpu(drop_info->back) == save_blk->blkno));
1760
1761         /*
1762          * Unlink the leaf block from the doubly linked chain of leaves.
1763          */
1764         if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1765                 trace_xfs_da_unlink_back(args);
1766                 save_info->back = drop_info->back;
1767                 if (drop_info->back) {
1768                         error = xfs_da3_node_read(args->trans, args->dp,
1769                                                 be32_to_cpu(drop_info->back),
1770                                                 -1, &bp, args->whichfork);
1771                         if (error)
1772                                 return error;
1773                         ASSERT(bp != NULL);
1774                         tmp_info = bp->b_addr;
1775                         ASSERT(tmp_info->magic == save_info->magic);
1776                         ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1777                         tmp_info->forw = cpu_to_be32(save_blk->blkno);
1778                         xfs_trans_log_buf(args->trans, bp, 0,
1779                                                     sizeof(*tmp_info) - 1);
1780                 }
1781         } else {
1782                 trace_xfs_da_unlink_forward(args);
1783                 save_info->forw = drop_info->forw;
1784                 if (drop_info->forw) {
1785                         error = xfs_da3_node_read(args->trans, args->dp,
1786                                                 be32_to_cpu(drop_info->forw),
1787                                                 -1, &bp, args->whichfork);
1788                         if (error)
1789                                 return error;
1790                         ASSERT(bp != NULL);
1791                         tmp_info = bp->b_addr;
1792                         ASSERT(tmp_info->magic == save_info->magic);
1793                         ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1794                         tmp_info->back = cpu_to_be32(save_blk->blkno);
1795                         xfs_trans_log_buf(args->trans, bp, 0,
1796                                                     sizeof(*tmp_info) - 1);
1797                 }
1798         }
1799
1800         xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1801         return 0;
1802 }
1803
1804 /*
1805  * Move a path "forward" or "!forward" one block at the current level.
1806  *
1807  * This routine will adjust a "path" to point to the next block
1808  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1809  * Btree, including updating pointers to the intermediate nodes between
1810  * the new bottom and the root.
1811  */
1812 int                                                     /* error */
1813 xfs_da3_path_shift(
1814         struct xfs_da_state     *state,
1815         struct xfs_da_state_path *path,
1816         int                     forward,
1817         int                     release,
1818         int                     *result)
1819 {
1820         struct xfs_da_state_blk *blk;
1821         struct xfs_da_blkinfo   *info;
1822         struct xfs_da_intnode   *node;
1823         struct xfs_da_args      *args;
1824         struct xfs_da_node_entry *btree;
1825         struct xfs_da3_icnode_hdr nodehdr;
1826         struct xfs_buf          *bp;
1827         xfs_dablk_t             blkno = 0;
1828         int                     level;
1829         int                     error;
1830         struct xfs_inode        *dp = state->args->dp;
1831
1832         trace_xfs_da_path_shift(state->args);
1833
1834         /*
1835          * Roll up the Btree looking for the first block where our
1836          * current index is not at the edge of the block.  Note that
1837          * we skip the bottom layer because we want the sibling block.
1838          */
1839         args = state->args;
1840         ASSERT(args != NULL);
1841         ASSERT(path != NULL);
1842         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1843         level = (path->active-1) - 1;   /* skip bottom layer in path */
1844         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1845                 node = blk->bp->b_addr;
1846                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1847                 btree = dp->d_ops->node_tree_p(node);
1848
1849                 if (forward && (blk->index < nodehdr.count - 1)) {
1850                         blk->index++;
1851                         blkno = be32_to_cpu(btree[blk->index].before);
1852                         break;
1853                 } else if (!forward && (blk->index > 0)) {
1854                         blk->index--;
1855                         blkno = be32_to_cpu(btree[blk->index].before);
1856                         break;
1857                 }
1858         }
1859         if (level < 0) {
1860                 *result = -ENOENT;      /* we're out of our tree */
1861                 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1862                 return 0;
1863         }
1864
1865         /*
1866          * Roll down the edge of the subtree until we reach the
1867          * same depth we were at originally.
1868          */
1869         for (blk++, level++; level < path->active; blk++, level++) {
1870                 /*
1871                  * Read the next child block into a local buffer.
1872                  */
1873                 error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
1874                                           args->whichfork);
1875                 if (error)
1876                         return error;
1877
1878                 /*
1879                  * Release the old block (if it's dirty, the trans doesn't
1880                  * actually let go) and swap the local buffer into the path
1881                  * structure. This ensures failure of the above read doesn't set
1882                  * a NULL buffer in an active slot in the path.
1883                  */
1884                 if (release)
1885                         xfs_trans_brelse(args->trans, blk->bp);
1886                 blk->blkno = blkno;
1887                 blk->bp = bp;
1888
1889                 info = blk->bp->b_addr;
1890                 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1891                        info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1892                        info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1893                        info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1894                        info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1895                        info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1896
1897
1898                 /*
1899                  * Note: we flatten the magic number to a single type so we
1900                  * don't have to compare against crc/non-crc types elsewhere.
1901                  */
1902                 switch (be16_to_cpu(info->magic)) {
1903                 case XFS_DA_NODE_MAGIC:
1904                 case XFS_DA3_NODE_MAGIC:
1905                         blk->magic = XFS_DA_NODE_MAGIC;
1906                         node = (xfs_da_intnode_t *)info;
1907                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1908                         btree = dp->d_ops->node_tree_p(node);
1909                         blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1910                         if (forward)
1911                                 blk->index = 0;
1912                         else
1913                                 blk->index = nodehdr.count - 1;
1914                         blkno = be32_to_cpu(btree[blk->index].before);
1915                         break;
1916                 case XFS_ATTR_LEAF_MAGIC:
1917                 case XFS_ATTR3_LEAF_MAGIC:
1918                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1919                         ASSERT(level == path->active-1);
1920                         blk->index = 0;
1921                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1922                         break;
1923                 case XFS_DIR2_LEAFN_MAGIC:
1924                 case XFS_DIR3_LEAFN_MAGIC:
1925                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1926                         ASSERT(level == path->active-1);
1927                         blk->index = 0;
1928                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1929                                                                blk->bp, NULL);
1930                         break;
1931                 default:
1932                         ASSERT(0);
1933                         break;
1934                 }
1935         }
1936         *result = 0;
1937         return 0;
1938 }
1939
1940
1941 /*========================================================================
1942  * Utility routines.
1943  *========================================================================*/
1944
1945 /*
1946  * Implement a simple hash on a character string.
1947  * Rotate the hash value by 7 bits, then XOR each character in.
1948  * This is implemented with some source-level loop unrolling.
1949  */
1950 xfs_dahash_t
1951 xfs_da_hashname(const __uint8_t *name, int namelen)
1952 {
1953         xfs_dahash_t hash;
1954
1955         /*
1956          * Do four characters at a time as long as we can.
1957          */
1958         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1959                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1960                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1961
1962         /*
1963          * Now do the rest of the characters.
1964          */
1965         switch (namelen) {
1966         case 3:
1967                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1968                        rol32(hash, 7 * 3);
1969         case 2:
1970                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1971         case 1:
1972                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1973         default: /* case 0: */
1974                 return hash;
1975         }
1976 }
1977
1978 enum xfs_dacmp
1979 xfs_da_compname(
1980         struct xfs_da_args *args,
1981         const unsigned char *name,
1982         int             len)
1983 {
1984         return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1985                                         XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1986 }
1987
1988 static xfs_dahash_t
1989 xfs_default_hashname(
1990         struct xfs_name *name)
1991 {
1992         return xfs_da_hashname(name->name, name->len);
1993 }
1994
1995 const struct xfs_nameops xfs_default_nameops = {
1996         .hashname       = xfs_default_hashname,
1997         .compname       = xfs_da_compname
1998 };
1999
2000 int
2001 xfs_da_grow_inode_int(
2002         struct xfs_da_args      *args,
2003         xfs_fileoff_t           *bno,
2004         int                     count)
2005 {
2006         struct xfs_trans        *tp = args->trans;
2007         struct xfs_inode        *dp = args->dp;
2008         int                     w = args->whichfork;
2009         xfs_rfsblock_t          nblks = dp->i_d.di_nblocks;
2010         struct xfs_bmbt_irec    map, *mapp;
2011         int                     nmap, error, got, i, mapi;
2012
2013         /*
2014          * Find a spot in the file space to put the new block.
2015          */
2016         error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2017         if (error)
2018                 return error;
2019
2020         /*
2021          * Try mapping it in one filesystem block.
2022          */
2023         nmap = 1;
2024         ASSERT(args->firstblock != NULL);
2025         error = xfs_bmapi_write(tp, dp, *bno, count,
2026                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2027                         args->firstblock, args->total, &map, &nmap,
2028                         args->flist);
2029         if (error)
2030                 return error;
2031
2032         ASSERT(nmap <= 1);
2033         if (nmap == 1) {
2034                 mapp = &map;
2035                 mapi = 1;
2036         } else if (nmap == 0 && count > 1) {
2037                 xfs_fileoff_t           b;
2038                 int                     c;
2039
2040                 /*
2041                  * If we didn't get it and the block might work if fragmented,
2042                  * try without the CONTIG flag.  Loop until we get it all.
2043                  */
2044                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2045                 for (b = *bno, mapi = 0; b < *bno + count; ) {
2046                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2047                         c = (int)(*bno + count - b);
2048                         error = xfs_bmapi_write(tp, dp, b, c,
2049                                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2050                                         args->firstblock, args->total,
2051                                         &mapp[mapi], &nmap, args->flist);
2052                         if (error)
2053                                 goto out_free_map;
2054                         if (nmap < 1)
2055                                 break;
2056                         mapi += nmap;
2057                         b = mapp[mapi - 1].br_startoff +
2058                             mapp[mapi - 1].br_blockcount;
2059                 }
2060         } else {
2061                 mapi = 0;
2062                 mapp = NULL;
2063         }
2064
2065         /*
2066          * Count the blocks we got, make sure it matches the total.
2067          */
2068         for (i = 0, got = 0; i < mapi; i++)
2069                 got += mapp[i].br_blockcount;
2070         if (got != count || mapp[0].br_startoff != *bno ||
2071             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2072             *bno + count) {
2073                 error = -ENOSPC;
2074                 goto out_free_map;
2075         }
2076
2077         /* account for newly allocated blocks in reserved blocks total */
2078         args->total -= dp->i_d.di_nblocks - nblks;
2079
2080 out_free_map:
2081         if (mapp != &map)
2082                 kmem_free(mapp);
2083         return error;
2084 }
2085
2086 /*
2087  * Add a block to the btree ahead of the file.
2088  * Return the new block number to the caller.
2089  */
2090 int
2091 xfs_da_grow_inode(
2092         struct xfs_da_args      *args,
2093         xfs_dablk_t             *new_blkno)
2094 {
2095         xfs_fileoff_t           bno;
2096         int                     error;
2097
2098         trace_xfs_da_grow_inode(args);
2099
2100         bno = args->geo->leafblk;
2101         error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2102         if (!error)
2103                 *new_blkno = (xfs_dablk_t)bno;
2104         return error;
2105 }
2106
2107 /*
2108  * Ick.  We need to always be able to remove a btree block, even
2109  * if there's no space reservation because the filesystem is full.
2110  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2111  * It swaps the target block with the last block in the file.  The
2112  * last block in the file can always be removed since it can't cause
2113  * a bmap btree split to do that.
2114  */
2115 STATIC int
2116 xfs_da3_swap_lastblock(
2117         struct xfs_da_args      *args,
2118         xfs_dablk_t             *dead_blknop,
2119         struct xfs_buf          **dead_bufp)
2120 {
2121         struct xfs_da_blkinfo   *dead_info;
2122         struct xfs_da_blkinfo   *sib_info;
2123         struct xfs_da_intnode   *par_node;
2124         struct xfs_da_intnode   *dead_node;
2125         struct xfs_dir2_leaf    *dead_leaf2;
2126         struct xfs_da_node_entry *btree;
2127         struct xfs_da3_icnode_hdr par_hdr;
2128         struct xfs_inode        *dp;
2129         struct xfs_trans        *tp;
2130         struct xfs_mount        *mp;
2131         struct xfs_buf          *dead_buf;
2132         struct xfs_buf          *last_buf;
2133         struct xfs_buf          *sib_buf;
2134         struct xfs_buf          *par_buf;
2135         xfs_dahash_t            dead_hash;
2136         xfs_fileoff_t           lastoff;
2137         xfs_dablk_t             dead_blkno;
2138         xfs_dablk_t             last_blkno;
2139         xfs_dablk_t             sib_blkno;
2140         xfs_dablk_t             par_blkno;
2141         int                     error;
2142         int                     w;
2143         int                     entno;
2144         int                     level;
2145         int                     dead_level;
2146
2147         trace_xfs_da_swap_lastblock(args);
2148
2149         dead_buf = *dead_bufp;
2150         dead_blkno = *dead_blknop;
2151         tp = args->trans;
2152         dp = args->dp;
2153         w = args->whichfork;
2154         ASSERT(w == XFS_DATA_FORK);
2155         mp = dp->i_mount;
2156         lastoff = args->geo->freeblk;
2157         error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2158         if (error)
2159                 return error;
2160         if (unlikely(lastoff == 0)) {
2161                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2162                                  mp);
2163                 return -EFSCORRUPTED;
2164         }
2165         /*
2166          * Read the last block in the btree space.
2167          */
2168         last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2169         error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2170         if (error)
2171                 return error;
2172         /*
2173          * Copy the last block into the dead buffer and log it.
2174          */
2175         memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2176         xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2177         dead_info = dead_buf->b_addr;
2178         /*
2179          * Get values from the moved block.
2180          */
2181         if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2182             dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2183                 struct xfs_dir3_icleaf_hdr leafhdr;
2184                 struct xfs_dir2_leaf_entry *ents;
2185
2186                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2187                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2188                 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2189                 dead_level = 0;
2190                 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2191         } else {
2192                 struct xfs_da3_icnode_hdr deadhdr;
2193
2194                 dead_node = (xfs_da_intnode_t *)dead_info;
2195                 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2196                 btree = dp->d_ops->node_tree_p(dead_node);
2197                 dead_level = deadhdr.level;
2198                 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2199         }
2200         sib_buf = par_buf = NULL;
2201         /*
2202          * If the moved block has a left sibling, fix up the pointers.
2203          */
2204         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2205                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2206                 if (error)
2207                         goto done;
2208                 sib_info = sib_buf->b_addr;
2209                 if (unlikely(
2210                     be32_to_cpu(sib_info->forw) != last_blkno ||
2211                     sib_info->magic != dead_info->magic)) {
2212                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2213                                          XFS_ERRLEVEL_LOW, mp);
2214                         error = -EFSCORRUPTED;
2215                         goto done;
2216                 }
2217                 sib_info->forw = cpu_to_be32(dead_blkno);
2218                 xfs_trans_log_buf(tp, sib_buf,
2219                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2220                                         sizeof(sib_info->forw)));
2221                 sib_buf = NULL;
2222         }
2223         /*
2224          * If the moved block has a right sibling, fix up the pointers.
2225          */
2226         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2227                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2228                 if (error)
2229                         goto done;
2230                 sib_info = sib_buf->b_addr;
2231                 if (unlikely(
2232                        be32_to_cpu(sib_info->back) != last_blkno ||
2233                        sib_info->magic != dead_info->magic)) {
2234                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2235                                          XFS_ERRLEVEL_LOW, mp);
2236                         error = -EFSCORRUPTED;
2237                         goto done;
2238                 }
2239                 sib_info->back = cpu_to_be32(dead_blkno);
2240                 xfs_trans_log_buf(tp, sib_buf,
2241                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2242                                         sizeof(sib_info->back)));
2243                 sib_buf = NULL;
2244         }
2245         par_blkno = args->geo->leafblk;
2246         level = -1;
2247         /*
2248          * Walk down the tree looking for the parent of the moved block.
2249          */
2250         for (;;) {
2251                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2252                 if (error)
2253                         goto done;
2254                 par_node = par_buf->b_addr;
2255                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2256                 if (level >= 0 && level != par_hdr.level + 1) {
2257                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2258                                          XFS_ERRLEVEL_LOW, mp);
2259                         error = -EFSCORRUPTED;
2260                         goto done;
2261                 }
2262                 level = par_hdr.level;
2263                 btree = dp->d_ops->node_tree_p(par_node);
2264                 for (entno = 0;
2265                      entno < par_hdr.count &&
2266                      be32_to_cpu(btree[entno].hashval) < dead_hash;
2267                      entno++)
2268                         continue;
2269                 if (entno == par_hdr.count) {
2270                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2271                                          XFS_ERRLEVEL_LOW, mp);
2272                         error = -EFSCORRUPTED;
2273                         goto done;
2274                 }
2275                 par_blkno = be32_to_cpu(btree[entno].before);
2276                 if (level == dead_level + 1)
2277                         break;
2278                 xfs_trans_brelse(tp, par_buf);
2279                 par_buf = NULL;
2280         }
2281         /*
2282          * We're in the right parent block.
2283          * Look for the right entry.
2284          */
2285         for (;;) {
2286                 for (;
2287                      entno < par_hdr.count &&
2288                      be32_to_cpu(btree[entno].before) != last_blkno;
2289                      entno++)
2290                         continue;
2291                 if (entno < par_hdr.count)
2292                         break;
2293                 par_blkno = par_hdr.forw;
2294                 xfs_trans_brelse(tp, par_buf);
2295                 par_buf = NULL;
2296                 if (unlikely(par_blkno == 0)) {
2297                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2298                                          XFS_ERRLEVEL_LOW, mp);
2299                         error = -EFSCORRUPTED;
2300                         goto done;
2301                 }
2302                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2303                 if (error)
2304                         goto done;
2305                 par_node = par_buf->b_addr;
2306                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2307                 if (par_hdr.level != level) {
2308                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2309                                          XFS_ERRLEVEL_LOW, mp);
2310                         error = -EFSCORRUPTED;
2311                         goto done;
2312                 }
2313                 btree = dp->d_ops->node_tree_p(par_node);
2314                 entno = 0;
2315         }
2316         /*
2317          * Update the parent entry pointing to the moved block.
2318          */
2319         btree[entno].before = cpu_to_be32(dead_blkno);
2320         xfs_trans_log_buf(tp, par_buf,
2321                 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2322                                 sizeof(btree[entno].before)));
2323         *dead_blknop = last_blkno;
2324         *dead_bufp = last_buf;
2325         return 0;
2326 done:
2327         if (par_buf)
2328                 xfs_trans_brelse(tp, par_buf);
2329         if (sib_buf)
2330                 xfs_trans_brelse(tp, sib_buf);
2331         xfs_trans_brelse(tp, last_buf);
2332         return error;
2333 }
2334
2335 /*
2336  * Remove a btree block from a directory or attribute.
2337  */
2338 int
2339 xfs_da_shrink_inode(
2340         xfs_da_args_t   *args,
2341         xfs_dablk_t     dead_blkno,
2342         struct xfs_buf  *dead_buf)
2343 {
2344         xfs_inode_t *dp;
2345         int done, error, w, count;
2346         xfs_trans_t *tp;
2347
2348         trace_xfs_da_shrink_inode(args);
2349
2350         dp = args->dp;
2351         w = args->whichfork;
2352         tp = args->trans;
2353         count = args->geo->fsbcount;
2354         for (;;) {
2355                 /*
2356                  * Remove extents.  If we get ENOSPC for a dir we have to move
2357                  * the last block to the place we want to kill.
2358                  */
2359                 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2360                                     xfs_bmapi_aflag(w), 0, args->firstblock,
2361                                     args->flist, &done);
2362                 if (error == -ENOSPC) {
2363                         if (w != XFS_DATA_FORK)
2364                                 break;
2365                         error = xfs_da3_swap_lastblock(args, &dead_blkno,
2366                                                       &dead_buf);
2367                         if (error)
2368                                 break;
2369                 } else {
2370                         break;
2371                 }
2372         }
2373         xfs_trans_binval(tp, dead_buf);
2374         return error;
2375 }
2376
2377 /*
2378  * See if the mapping(s) for this btree block are valid, i.e.
2379  * don't contain holes, are logically contiguous, and cover the whole range.
2380  */
2381 STATIC int
2382 xfs_da_map_covers_blocks(
2383         int             nmap,
2384         xfs_bmbt_irec_t *mapp,
2385         xfs_dablk_t     bno,
2386         int             count)
2387 {
2388         int             i;
2389         xfs_fileoff_t   off;
2390
2391         for (i = 0, off = bno; i < nmap; i++) {
2392                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2393                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
2394                         return 0;
2395                 }
2396                 if (off != mapp[i].br_startoff) {
2397                         return 0;
2398                 }
2399                 off += mapp[i].br_blockcount;
2400         }
2401         return off == bno + count;
2402 }
2403
2404 /*
2405  * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2406  *
2407  * For the single map case, it is assumed that the caller has provided a pointer
2408  * to a valid xfs_buf_map.  For the multiple map case, this function will
2409  * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2410  * map pointer with the allocated map.
2411  */
2412 static int
2413 xfs_buf_map_from_irec(
2414         struct xfs_mount        *mp,
2415         struct xfs_buf_map      **mapp,
2416         int                     *nmaps,
2417         struct xfs_bmbt_irec    *irecs,
2418         int                     nirecs)
2419 {
2420         struct xfs_buf_map      *map;
2421         int                     i;
2422
2423         ASSERT(*nmaps == 1);
2424         ASSERT(nirecs >= 1);
2425
2426         if (nirecs > 1) {
2427                 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2428                                   KM_SLEEP | KM_NOFS);
2429                 if (!map)
2430                         return -ENOMEM;
2431                 *mapp = map;
2432         }
2433
2434         *nmaps = nirecs;
2435         map = *mapp;
2436         for (i = 0; i < *nmaps; i++) {
2437                 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2438                        irecs[i].br_startblock != HOLESTARTBLOCK);
2439                 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2440                 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2441         }
2442         return 0;
2443 }
2444
2445 /*
2446  * Map the block we are given ready for reading. There are three possible return
2447  * values:
2448  *      -1 - will be returned if we land in a hole and mappedbno == -2 so the
2449  *           caller knows not to execute a subsequent read.
2450  *       0 - if we mapped the block successfully
2451  *      >0 - positive error number if there was an error.
2452  */
2453 static int
2454 xfs_dabuf_map(
2455         struct xfs_inode        *dp,
2456         xfs_dablk_t             bno,
2457         xfs_daddr_t             mappedbno,
2458         int                     whichfork,
2459         struct xfs_buf_map      **map,
2460         int                     *nmaps)
2461 {
2462         struct xfs_mount        *mp = dp->i_mount;
2463         int                     nfsb;
2464         int                     error = 0;
2465         struct xfs_bmbt_irec    irec;
2466         struct xfs_bmbt_irec    *irecs = &irec;
2467         int                     nirecs;
2468
2469         ASSERT(map && *map);
2470         ASSERT(*nmaps == 1);
2471
2472         if (whichfork == XFS_DATA_FORK)
2473                 nfsb = mp->m_dir_geo->fsbcount;
2474         else
2475                 nfsb = mp->m_attr_geo->fsbcount;
2476
2477         /*
2478          * Caller doesn't have a mapping.  -2 means don't complain
2479          * if we land in a hole.
2480          */
2481         if (mappedbno == -1 || mappedbno == -2) {
2482                 /*
2483                  * Optimize the one-block case.
2484                  */
2485                 if (nfsb != 1)
2486                         irecs = kmem_zalloc(sizeof(irec) * nfsb,
2487                                             KM_SLEEP | KM_NOFS);
2488
2489                 nirecs = nfsb;
2490                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2491                                        &nirecs, xfs_bmapi_aflag(whichfork));
2492                 if (error)
2493                         goto out;
2494         } else {
2495                 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2496                 irecs->br_startoff = (xfs_fileoff_t)bno;
2497                 irecs->br_blockcount = nfsb;
2498                 irecs->br_state = 0;
2499                 nirecs = 1;
2500         }
2501
2502         if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2503                 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2504                 if (unlikely(error == -EFSCORRUPTED)) {
2505                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2506                                 int i;
2507                                 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2508                                         __func__, (long long)bno,
2509                                         (long long)dp->i_ino);
2510                                 for (i = 0; i < *nmaps; i++) {
2511                                         xfs_alert(mp,
2512 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2513                                                 i,
2514                                                 (long long)irecs[i].br_startoff,
2515                                                 (long long)irecs[i].br_startblock,
2516                                                 (long long)irecs[i].br_blockcount,
2517                                                 irecs[i].br_state);
2518                                 }
2519                         }
2520                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2521                                          XFS_ERRLEVEL_LOW, mp);
2522                 }
2523                 goto out;
2524         }
2525         error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2526 out:
2527         if (irecs != &irec)
2528                 kmem_free(irecs);
2529         return error;
2530 }
2531
2532 /*
2533  * Get a buffer for the dir/attr block.
2534  */
2535 int
2536 xfs_da_get_buf(
2537         struct xfs_trans        *trans,
2538         struct xfs_inode        *dp,
2539         xfs_dablk_t             bno,
2540         xfs_daddr_t             mappedbno,
2541         struct xfs_buf          **bpp,
2542         int                     whichfork)
2543 {
2544         struct xfs_buf          *bp;
2545         struct xfs_buf_map      map;
2546         struct xfs_buf_map      *mapp;
2547         int                     nmap;
2548         int                     error;
2549
2550         *bpp = NULL;
2551         mapp = &map;
2552         nmap = 1;
2553         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2554                                 &mapp, &nmap);
2555         if (error) {
2556                 /* mapping a hole is not an error, but we don't continue */
2557                 if (error == -1)
2558                         error = 0;
2559                 goto out_free;
2560         }
2561
2562         bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2563                                     mapp, nmap, 0);
2564         error = bp ? bp->b_error : -EIO;
2565         if (error) {
2566                 if (bp)
2567                         xfs_trans_brelse(trans, bp);
2568                 goto out_free;
2569         }
2570
2571         *bpp = bp;
2572
2573 out_free:
2574         if (mapp != &map)
2575                 kmem_free(mapp);
2576
2577         return error;
2578 }
2579
2580 /*
2581  * Get a buffer for the dir/attr block, fill in the contents.
2582  */
2583 int
2584 xfs_da_read_buf(
2585         struct xfs_trans        *trans,
2586         struct xfs_inode        *dp,
2587         xfs_dablk_t             bno,
2588         xfs_daddr_t             mappedbno,
2589         struct xfs_buf          **bpp,
2590         int                     whichfork,
2591         const struct xfs_buf_ops *ops)
2592 {
2593         struct xfs_buf          *bp;
2594         struct xfs_buf_map      map;
2595         struct xfs_buf_map      *mapp;
2596         int                     nmap;
2597         int                     error;
2598
2599         *bpp = NULL;
2600         mapp = &map;
2601         nmap = 1;
2602         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2603                                 &mapp, &nmap);
2604         if (error) {
2605                 /* mapping a hole is not an error, but we don't continue */
2606                 if (error == -1)
2607                         error = 0;
2608                 goto out_free;
2609         }
2610
2611         error = xfs_trans_read_buf_map(dp->i_mount, trans,
2612                                         dp->i_mount->m_ddev_targp,
2613                                         mapp, nmap, 0, &bp, ops);
2614         if (error)
2615                 goto out_free;
2616
2617         if (whichfork == XFS_ATTR_FORK)
2618                 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2619         else
2620                 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2621         *bpp = bp;
2622 out_free:
2623         if (mapp != &map)
2624                 kmem_free(mapp);
2625
2626         return error;
2627 }
2628
2629 /*
2630  * Readahead the dir/attr block.
2631  */
2632 xfs_daddr_t
2633 xfs_da_reada_buf(
2634         struct xfs_inode        *dp,
2635         xfs_dablk_t             bno,
2636         xfs_daddr_t             mappedbno,
2637         int                     whichfork,
2638         const struct xfs_buf_ops *ops)
2639 {
2640         struct xfs_buf_map      map;
2641         struct xfs_buf_map      *mapp;
2642         int                     nmap;
2643         int                     error;
2644
2645         mapp = &map;
2646         nmap = 1;
2647         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2648                                 &mapp, &nmap);
2649         if (error) {
2650                 /* mapping a hole is not an error, but we don't continue */
2651                 if (error == -1)
2652                         error = 0;
2653                 goto out_free;
2654         }
2655
2656         mappedbno = mapp[0].bm_bn;
2657         xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2658
2659 out_free:
2660         if (mapp != &map)
2661                 kmem_free(mapp);
2662
2663         if (error)
2664                 return -1;
2665         return mappedbno;
2666 }