Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include "xfs.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 19 | #include "xfs_fs.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include "xfs_types.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 21 | #include "xfs_bit.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include "xfs_log.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 23 | #include "xfs_inum.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_trans.h" |
| 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
| 27 | #include "xfs_dir.h" |
| 28 | #include "xfs_dir2.h" |
| 29 | #include "xfs_dmapi.h" |
| 30 | #include "xfs_mount.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 31 | #include "xfs_da_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include "xfs_bmap_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 33 | #include "xfs_alloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include "xfs_ialloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include "xfs_dir_sf.h" |
| 36 | #include "xfs_dir2_sf.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 37 | #include "xfs_attr_sf.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include "xfs_dinode.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include "xfs_inode.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 40 | #include "xfs_inode_item.h" |
| 41 | #include "xfs_alloc.h" |
| 42 | #include "xfs_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include "xfs_bmap.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | #include "xfs_attr.h" |
| 45 | #include "xfs_attr_leaf.h" |
| 46 | #include "xfs_dir_leaf.h" |
| 47 | #include "xfs_dir2_data.h" |
| 48 | #include "xfs_dir2_leaf.h" |
| 49 | #include "xfs_dir2_block.h" |
| 50 | #include "xfs_dir2_node.h" |
| 51 | #include "xfs_error.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | |
| 53 | /* |
| 54 | * xfs_da_btree.c |
| 55 | * |
| 56 | * Routines to implement directories as Btrees of hashed names. |
| 57 | */ |
| 58 | |
| 59 | /*======================================================================== |
| 60 | * Function prototypes for the kernel. |
| 61 | *========================================================================*/ |
| 62 | |
| 63 | /* |
| 64 | * Routines used for growing the Btree. |
| 65 | */ |
| 66 | STATIC int xfs_da_root_split(xfs_da_state_t *state, |
| 67 | xfs_da_state_blk_t *existing_root, |
| 68 | xfs_da_state_blk_t *new_child); |
| 69 | STATIC int xfs_da_node_split(xfs_da_state_t *state, |
| 70 | xfs_da_state_blk_t *existing_blk, |
| 71 | xfs_da_state_blk_t *split_blk, |
| 72 | xfs_da_state_blk_t *blk_to_add, |
| 73 | int treelevel, |
| 74 | int *result); |
| 75 | STATIC void xfs_da_node_rebalance(xfs_da_state_t *state, |
| 76 | xfs_da_state_blk_t *node_blk_1, |
| 77 | xfs_da_state_blk_t *node_blk_2); |
| 78 | STATIC void xfs_da_node_add(xfs_da_state_t *state, |
| 79 | xfs_da_state_blk_t *old_node_blk, |
| 80 | xfs_da_state_blk_t *new_node_blk); |
| 81 | |
| 82 | /* |
| 83 | * Routines used for shrinking the Btree. |
| 84 | */ |
| 85 | STATIC int xfs_da_root_join(xfs_da_state_t *state, |
| 86 | xfs_da_state_blk_t *root_blk); |
| 87 | STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval); |
| 88 | STATIC void xfs_da_node_remove(xfs_da_state_t *state, |
| 89 | xfs_da_state_blk_t *drop_blk); |
| 90 | STATIC void xfs_da_node_unbalance(xfs_da_state_t *state, |
| 91 | xfs_da_state_blk_t *src_node_blk, |
| 92 | xfs_da_state_blk_t *dst_node_blk); |
| 93 | |
| 94 | /* |
| 95 | * Utility routines. |
| 96 | */ |
| 97 | STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count); |
| 98 | STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp); |
| 99 | STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra); |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 100 | STATIC int xfs_da_blk_unlink(xfs_da_state_t *state, |
| 101 | xfs_da_state_blk_t *drop_blk, |
| 102 | xfs_da_state_blk_t *save_blk); |
| 103 | STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | |
| 105 | /*======================================================================== |
| 106 | * Routines used for growing the Btree. |
| 107 | *========================================================================*/ |
| 108 | |
| 109 | /* |
| 110 | * Create the initial contents of an intermediate node. |
| 111 | */ |
| 112 | int |
| 113 | xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level, |
| 114 | xfs_dabuf_t **bpp, int whichfork) |
| 115 | { |
| 116 | xfs_da_intnode_t *node; |
| 117 | xfs_dabuf_t *bp; |
| 118 | int error; |
| 119 | xfs_trans_t *tp; |
| 120 | |
| 121 | tp = args->trans; |
| 122 | error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork); |
| 123 | if (error) |
| 124 | return(error); |
| 125 | ASSERT(bp != NULL); |
| 126 | node = bp->data; |
| 127 | node->hdr.info.forw = 0; |
| 128 | node->hdr.info.back = 0; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 129 | node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | node->hdr.info.pad = 0; |
| 131 | node->hdr.count = 0; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 132 | node->hdr.level = cpu_to_be16(level); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | |
| 134 | xfs_da_log_buf(tp, bp, |
| 135 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
| 136 | |
| 137 | *bpp = bp; |
| 138 | return(0); |
| 139 | } |
| 140 | |
| 141 | /* |
| 142 | * Split a leaf node, rebalance, then possibly split |
| 143 | * intermediate nodes, rebalance, etc. |
| 144 | */ |
| 145 | int /* error */ |
| 146 | xfs_da_split(xfs_da_state_t *state) |
| 147 | { |
| 148 | xfs_da_state_blk_t *oldblk, *newblk, *addblk; |
| 149 | xfs_da_intnode_t *node; |
| 150 | xfs_dabuf_t *bp; |
| 151 | int max, action, error, i; |
| 152 | |
| 153 | /* |
| 154 | * Walk back up the tree splitting/inserting/adjusting as necessary. |
| 155 | * If we need to insert and there isn't room, split the node, then |
| 156 | * decide which fragment to insert the new block from below into. |
| 157 | * Note that we may split the root this way, but we need more fixup. |
| 158 | */ |
| 159 | max = state->path.active - 1; |
| 160 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); |
| 161 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || |
| 162 | state->path.blk[max].magic == XFS_DIRX_LEAF_MAGIC(state->mp)); |
| 163 | |
| 164 | addblk = &state->path.blk[max]; /* initial dummy value */ |
| 165 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { |
| 166 | oldblk = &state->path.blk[i]; |
| 167 | newblk = &state->altpath.blk[i]; |
| 168 | |
| 169 | /* |
| 170 | * If a leaf node then |
| 171 | * Allocate a new leaf node, then rebalance across them. |
| 172 | * else if an intermediate node then |
| 173 | * We split on the last layer, must we split the node? |
| 174 | */ |
| 175 | switch (oldblk->magic) { |
| 176 | case XFS_ATTR_LEAF_MAGIC: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | error = xfs_attr_leaf_split(state, oldblk, newblk); |
| 178 | if ((error != 0) && (error != ENOSPC)) { |
| 179 | return(error); /* GROT: attr is inconsistent */ |
| 180 | } |
| 181 | if (!error) { |
| 182 | addblk = newblk; |
| 183 | break; |
| 184 | } |
| 185 | /* |
| 186 | * Entry wouldn't fit, split the leaf again. |
| 187 | */ |
| 188 | state->extravalid = 1; |
| 189 | if (state->inleaf) { |
| 190 | state->extraafter = 0; /* before newblk */ |
| 191 | error = xfs_attr_leaf_split(state, oldblk, |
| 192 | &state->extrablk); |
| 193 | } else { |
| 194 | state->extraafter = 1; /* after newblk */ |
| 195 | error = xfs_attr_leaf_split(state, newblk, |
| 196 | &state->extrablk); |
| 197 | } |
| 198 | if (error) |
| 199 | return(error); /* GROT: attr inconsistent */ |
| 200 | addblk = newblk; |
| 201 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | case XFS_DIR_LEAF_MAGIC: |
| 203 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 204 | error = xfs_dir_leaf_split(state, oldblk, newblk); |
| 205 | if ((error != 0) && (error != ENOSPC)) { |
| 206 | return(error); /* GROT: dir is inconsistent */ |
| 207 | } |
| 208 | if (!error) { |
| 209 | addblk = newblk; |
| 210 | break; |
| 211 | } |
| 212 | /* |
| 213 | * Entry wouldn't fit, split the leaf again. |
| 214 | */ |
| 215 | state->extravalid = 1; |
| 216 | if (state->inleaf) { |
| 217 | state->extraafter = 0; /* before newblk */ |
| 218 | error = xfs_dir_leaf_split(state, oldblk, |
| 219 | &state->extrablk); |
| 220 | if (error) |
| 221 | return(error); /* GROT: dir incon. */ |
| 222 | addblk = newblk; |
| 223 | } else { |
| 224 | state->extraafter = 1; /* after newblk */ |
| 225 | error = xfs_dir_leaf_split(state, newblk, |
| 226 | &state->extrablk); |
| 227 | if (error) |
| 228 | return(error); /* GROT: dir incon. */ |
| 229 | addblk = newblk; |
| 230 | } |
| 231 | break; |
| 232 | case XFS_DIR2_LEAFN_MAGIC: |
| 233 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 234 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
| 235 | if (error) |
| 236 | return error; |
| 237 | addblk = newblk; |
| 238 | break; |
| 239 | case XFS_DA_NODE_MAGIC: |
| 240 | error = xfs_da_node_split(state, oldblk, newblk, addblk, |
| 241 | max - i, &action); |
| 242 | xfs_da_buf_done(addblk->bp); |
| 243 | addblk->bp = NULL; |
| 244 | if (error) |
| 245 | return(error); /* GROT: dir is inconsistent */ |
| 246 | /* |
| 247 | * Record the newly split block for the next time thru? |
| 248 | */ |
| 249 | if (action) |
| 250 | addblk = newblk; |
| 251 | else |
| 252 | addblk = NULL; |
| 253 | break; |
| 254 | } |
| 255 | |
| 256 | /* |
| 257 | * Update the btree to show the new hashval for this child. |
| 258 | */ |
| 259 | xfs_da_fixhashpath(state, &state->path); |
| 260 | /* |
| 261 | * If we won't need this block again, it's getting dropped |
| 262 | * from the active path by the loop control, so we need |
| 263 | * to mark it done now. |
| 264 | */ |
| 265 | if (i > 0 || !addblk) |
| 266 | xfs_da_buf_done(oldblk->bp); |
| 267 | } |
| 268 | if (!addblk) |
| 269 | return(0); |
| 270 | |
| 271 | /* |
| 272 | * Split the root node. |
| 273 | */ |
| 274 | ASSERT(state->path.active == 0); |
| 275 | oldblk = &state->path.blk[0]; |
| 276 | error = xfs_da_root_split(state, oldblk, addblk); |
| 277 | if (error) { |
| 278 | xfs_da_buf_done(oldblk->bp); |
| 279 | xfs_da_buf_done(addblk->bp); |
| 280 | addblk->bp = NULL; |
| 281 | return(error); /* GROT: dir is inconsistent */ |
| 282 | } |
| 283 | |
| 284 | /* |
| 285 | * Update pointers to the node which used to be block 0 and |
| 286 | * just got bumped because of the addition of a new root node. |
| 287 | * There might be three blocks involved if a double split occurred, |
| 288 | * and the original block 0 could be at any position in the list. |
| 289 | */ |
| 290 | |
| 291 | node = oldblk->bp->data; |
| 292 | if (node->hdr.info.forw) { |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 293 | if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | bp = addblk->bp; |
| 295 | } else { |
| 296 | ASSERT(state->extravalid); |
| 297 | bp = state->extrablk.bp; |
| 298 | } |
| 299 | node = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 300 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | xfs_da_log_buf(state->args->trans, bp, |
| 302 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
| 303 | sizeof(node->hdr.info))); |
| 304 | } |
| 305 | node = oldblk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 306 | if (node->hdr.info.back) { |
| 307 | if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | bp = addblk->bp; |
| 309 | } else { |
| 310 | ASSERT(state->extravalid); |
| 311 | bp = state->extrablk.bp; |
| 312 | } |
| 313 | node = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 314 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | xfs_da_log_buf(state->args->trans, bp, |
| 316 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
| 317 | sizeof(node->hdr.info))); |
| 318 | } |
| 319 | xfs_da_buf_done(oldblk->bp); |
| 320 | xfs_da_buf_done(addblk->bp); |
| 321 | addblk->bp = NULL; |
| 322 | return(0); |
| 323 | } |
| 324 | |
| 325 | /* |
| 326 | * Split the root. We have to create a new root and point to the two |
| 327 | * parts (the split old root) that we just created. Copy block zero to |
| 328 | * the EOF, extending the inode in process. |
| 329 | */ |
| 330 | STATIC int /* error */ |
| 331 | xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, |
| 332 | xfs_da_state_blk_t *blk2) |
| 333 | { |
| 334 | xfs_da_intnode_t *node, *oldroot; |
| 335 | xfs_da_args_t *args; |
| 336 | xfs_dablk_t blkno; |
| 337 | xfs_dabuf_t *bp; |
| 338 | int error, size; |
| 339 | xfs_inode_t *dp; |
| 340 | xfs_trans_t *tp; |
| 341 | xfs_mount_t *mp; |
| 342 | xfs_dir2_leaf_t *leaf; |
| 343 | |
| 344 | /* |
| 345 | * Copy the existing (incorrect) block from the root node position |
| 346 | * to a free space somewhere. |
| 347 | */ |
| 348 | args = state->args; |
| 349 | ASSERT(args != NULL); |
| 350 | error = xfs_da_grow_inode(args, &blkno); |
| 351 | if (error) |
| 352 | return(error); |
| 353 | dp = args->dp; |
| 354 | tp = args->trans; |
| 355 | mp = state->mp; |
| 356 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); |
| 357 | if (error) |
| 358 | return(error); |
| 359 | ASSERT(bp != NULL); |
| 360 | node = bp->data; |
| 361 | oldroot = blk1->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 362 | if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) { |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 363 | size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] - |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | (char *)oldroot); |
| 365 | } else { |
| 366 | ASSERT(XFS_DIR_IS_V2(mp)); |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 367 | ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | leaf = (xfs_dir2_leaf_t *)oldroot; |
Nathan Scott | a818e5d | 2006-03-17 17:28:07 +1100 | [diff] [blame] | 369 | size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] - |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | (char *)leaf); |
| 371 | } |
| 372 | memcpy(node, oldroot, size); |
| 373 | xfs_da_log_buf(tp, bp, 0, size - 1); |
| 374 | xfs_da_buf_done(blk1->bp); |
| 375 | blk1->bp = bp; |
| 376 | blk1->blkno = blkno; |
| 377 | |
| 378 | /* |
| 379 | * Set up the new root node. |
| 380 | */ |
| 381 | error = xfs_da_node_create(args, |
| 382 | args->whichfork == XFS_DATA_FORK && |
| 383 | XFS_DIR_IS_V2(mp) ? mp->m_dirleafblk : 0, |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 384 | be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | if (error) |
| 386 | return(error); |
| 387 | node = bp->data; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 388 | node->btree[0].hashval = cpu_to_be32(blk1->hashval); |
| 389 | node->btree[0].before = cpu_to_be32(blk1->blkno); |
| 390 | node->btree[1].hashval = cpu_to_be32(blk2->hashval); |
| 391 | node->btree[1].before = cpu_to_be32(blk2->blkno); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 392 | node->hdr.count = cpu_to_be16(2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | |
| 394 | #ifdef DEBUG |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 395 | if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | ASSERT(blk1->blkno >= mp->m_dirleafblk && |
| 397 | blk1->blkno < mp->m_dirfreeblk); |
| 398 | ASSERT(blk2->blkno >= mp->m_dirleafblk && |
| 399 | blk2->blkno < mp->m_dirfreeblk); |
| 400 | } |
| 401 | #endif |
| 402 | |
| 403 | /* Header is already logged by xfs_da_node_create */ |
| 404 | xfs_da_log_buf(tp, bp, |
| 405 | XFS_DA_LOGRANGE(node, node->btree, |
| 406 | sizeof(xfs_da_node_entry_t) * 2)); |
| 407 | xfs_da_buf_done(bp); |
| 408 | |
| 409 | return(0); |
| 410 | } |
| 411 | |
| 412 | /* |
| 413 | * Split the node, rebalance, then add the new entry. |
| 414 | */ |
| 415 | STATIC int /* error */ |
| 416 | xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, |
| 417 | xfs_da_state_blk_t *newblk, |
| 418 | xfs_da_state_blk_t *addblk, |
| 419 | int treelevel, int *result) |
| 420 | { |
| 421 | xfs_da_intnode_t *node; |
| 422 | xfs_dablk_t blkno; |
| 423 | int newcount, error; |
| 424 | int useextra; |
| 425 | |
| 426 | node = oldblk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 427 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | |
| 429 | /* |
| 430 | * With V2 the extra block is data or freespace. |
| 431 | */ |
Yingping Lu | bf6f05a | 2005-11-03 10:31:47 +1100 | [diff] [blame] | 432 | useextra = state->extravalid && (XFS_DIR_IS_V1(state->mp) || |
| 433 | state->args->whichfork == XFS_ATTR_FORK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | newcount = 1 + useextra; |
| 435 | /* |
| 436 | * Do we have to split the node? |
| 437 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 438 | if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | /* |
| 440 | * Allocate a new node, add to the doubly linked chain of |
| 441 | * nodes, then move some of our excess entries into it. |
| 442 | */ |
| 443 | error = xfs_da_grow_inode(state->args, &blkno); |
| 444 | if (error) |
| 445 | return(error); /* GROT: dir is inconsistent */ |
| 446 | |
| 447 | error = xfs_da_node_create(state->args, blkno, treelevel, |
| 448 | &newblk->bp, state->args->whichfork); |
| 449 | if (error) |
| 450 | return(error); /* GROT: dir is inconsistent */ |
| 451 | newblk->blkno = blkno; |
| 452 | newblk->magic = XFS_DA_NODE_MAGIC; |
| 453 | xfs_da_node_rebalance(state, oldblk, newblk); |
| 454 | error = xfs_da_blk_link(state, oldblk, newblk); |
| 455 | if (error) |
| 456 | return(error); |
| 457 | *result = 1; |
| 458 | } else { |
| 459 | *result = 0; |
| 460 | } |
| 461 | |
| 462 | /* |
| 463 | * Insert the new entry(s) into the correct block |
| 464 | * (updating last hashval in the process). |
| 465 | * |
| 466 | * xfs_da_node_add() inserts BEFORE the given index, |
| 467 | * and as a result of using node_lookup_int() we always |
| 468 | * point to a valid entry (not after one), but a split |
| 469 | * operation always results in a new block whose hashvals |
| 470 | * FOLLOW the current block. |
| 471 | * |
| 472 | * If we had double-split op below us, then add the extra block too. |
| 473 | */ |
| 474 | node = oldblk->bp->data; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 475 | if (oldblk->index <= be16_to_cpu(node->hdr.count)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | oldblk->index++; |
| 477 | xfs_da_node_add(state, oldblk, addblk); |
| 478 | if (useextra) { |
| 479 | if (state->extraafter) |
| 480 | oldblk->index++; |
| 481 | xfs_da_node_add(state, oldblk, &state->extrablk); |
| 482 | state->extravalid = 0; |
| 483 | } |
| 484 | } else { |
| 485 | newblk->index++; |
| 486 | xfs_da_node_add(state, newblk, addblk); |
| 487 | if (useextra) { |
| 488 | if (state->extraafter) |
| 489 | newblk->index++; |
| 490 | xfs_da_node_add(state, newblk, &state->extrablk); |
| 491 | state->extravalid = 0; |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | return(0); |
| 496 | } |
| 497 | |
| 498 | /* |
| 499 | * Balance the btree elements between two intermediate nodes, |
| 500 | * usually one full and one empty. |
| 501 | * |
| 502 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. |
| 503 | */ |
| 504 | STATIC void |
| 505 | xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, |
| 506 | xfs_da_state_blk_t *blk2) |
| 507 | { |
| 508 | xfs_da_intnode_t *node1, *node2, *tmpnode; |
| 509 | xfs_da_node_entry_t *btree_s, *btree_d; |
| 510 | int count, tmp; |
| 511 | xfs_trans_t *tp; |
| 512 | |
| 513 | node1 = blk1->bp->data; |
| 514 | node2 = blk2->bp->data; |
| 515 | /* |
| 516 | * Figure out how many entries need to move, and in which direction. |
| 517 | * Swap the nodes around if that makes it simpler. |
| 518 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 519 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 520 | ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) || |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 521 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
| 522 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 523 | tmpnode = node1; |
| 524 | node1 = node2; |
| 525 | node2 = tmpnode; |
| 526 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 527 | ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
| 528 | ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 529 | count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | if (count == 0) |
| 531 | return; |
| 532 | tp = state->args->trans; |
| 533 | /* |
| 534 | * Two cases: high-to-low and low-to-high. |
| 535 | */ |
| 536 | if (count > 0) { |
| 537 | /* |
| 538 | * Move elements in node2 up to make a hole. |
| 539 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 540 | if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 542 | btree_s = &node2->btree[0]; |
| 543 | btree_d = &node2->btree[count]; |
| 544 | memmove(btree_d, btree_s, tmp); |
| 545 | } |
| 546 | |
| 547 | /* |
| 548 | * Move the req'd B-tree elements from high in node1 to |
| 549 | * low in node2. |
| 550 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 551 | be16_add(&node2->hdr.count, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 553 | btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | btree_d = &node2->btree[0]; |
| 555 | memcpy(btree_d, btree_s, tmp); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 556 | be16_add(&node1->hdr.count, -count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | } else { |
| 558 | /* |
| 559 | * Move the req'd B-tree elements from low in node2 to |
| 560 | * high in node1. |
| 561 | */ |
| 562 | count = -count; |
| 563 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
| 564 | btree_s = &node2->btree[0]; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 565 | btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | memcpy(btree_d, btree_s, tmp); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 567 | be16_add(&node1->hdr.count, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | xfs_da_log_buf(tp, blk1->bp, |
| 569 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
| 570 | |
| 571 | /* |
| 572 | * Move elements in node2 down to fill the hole. |
| 573 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 574 | tmp = be16_to_cpu(node2->hdr.count) - count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 576 | btree_s = &node2->btree[count]; |
| 577 | btree_d = &node2->btree[0]; |
| 578 | memmove(btree_d, btree_s, tmp); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 579 | be16_add(&node2->hdr.count, -count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 580 | } |
| 581 | |
| 582 | /* |
| 583 | * Log header of node 1 and all current bits of node 2. |
| 584 | */ |
| 585 | xfs_da_log_buf(tp, blk1->bp, |
| 586 | XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr))); |
| 587 | xfs_da_log_buf(tp, blk2->bp, |
| 588 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
| 589 | sizeof(node2->hdr) + |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 590 | sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | |
| 592 | /* |
| 593 | * Record the last hashval from each block for upward propagation. |
| 594 | * (note: don't use the swapped node pointers) |
| 595 | */ |
| 596 | node1 = blk1->bp->data; |
| 597 | node2 = blk2->bp->data; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 598 | blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval); |
| 599 | blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | |
| 601 | /* |
| 602 | * Adjust the expected index for insertion. |
| 603 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 604 | if (blk1->index >= be16_to_cpu(node1->hdr.count)) { |
| 605 | blk2->index = blk1->index - be16_to_cpu(node1->hdr.count); |
| 606 | blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | } |
| 608 | } |
| 609 | |
| 610 | /* |
| 611 | * Add a new entry to an intermediate node. |
| 612 | */ |
| 613 | STATIC void |
| 614 | xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, |
| 615 | xfs_da_state_blk_t *newblk) |
| 616 | { |
| 617 | xfs_da_intnode_t *node; |
| 618 | xfs_da_node_entry_t *btree; |
| 619 | int tmp; |
| 620 | xfs_mount_t *mp; |
| 621 | |
| 622 | node = oldblk->bp->data; |
| 623 | mp = state->mp; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 624 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 625 | ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | ASSERT(newblk->blkno != 0); |
| 627 | if (state->args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) |
| 628 | ASSERT(newblk->blkno >= mp->m_dirleafblk && |
| 629 | newblk->blkno < mp->m_dirfreeblk); |
| 630 | |
| 631 | /* |
| 632 | * We may need to make some room before we insert the new node. |
| 633 | */ |
| 634 | tmp = 0; |
| 635 | btree = &node->btree[ oldblk->index ]; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 636 | if (oldblk->index < be16_to_cpu(node->hdr.count)) { |
| 637 | tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | memmove(btree + 1, btree, tmp); |
| 639 | } |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 640 | btree->hashval = cpu_to_be32(newblk->hashval); |
| 641 | btree->before = cpu_to_be32(newblk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | xfs_da_log_buf(state->args->trans, oldblk->bp, |
| 643 | XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree))); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 644 | be16_add(&node->hdr.count, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | xfs_da_log_buf(state->args->trans, oldblk->bp, |
| 646 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
| 647 | |
| 648 | /* |
| 649 | * Copy the last hash value from the oldblk to propagate upwards. |
| 650 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 651 | oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | } |
| 653 | |
| 654 | /*======================================================================== |
| 655 | * Routines used for shrinking the Btree. |
| 656 | *========================================================================*/ |
| 657 | |
| 658 | /* |
| 659 | * Deallocate an empty leaf node, remove it from its parent, |
| 660 | * possibly deallocating that block, etc... |
| 661 | */ |
| 662 | int |
| 663 | xfs_da_join(xfs_da_state_t *state) |
| 664 | { |
| 665 | xfs_da_state_blk_t *drop_blk, *save_blk; |
| 666 | int action, error; |
| 667 | |
| 668 | action = 0; |
| 669 | drop_blk = &state->path.blk[ state->path.active-1 ]; |
| 670 | save_blk = &state->altpath.blk[ state->path.active-1 ]; |
| 671 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); |
| 672 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || |
| 673 | drop_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp)); |
| 674 | |
| 675 | /* |
| 676 | * Walk back up the tree joining/deallocating as necessary. |
| 677 | * When we stop dropping blocks, break out. |
| 678 | */ |
| 679 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, |
| 680 | state->path.active--) { |
| 681 | /* |
| 682 | * See if we can combine the block with a neighbor. |
| 683 | * (action == 0) => no options, just leave |
| 684 | * (action == 1) => coalesce, then unlink |
| 685 | * (action == 2) => block empty, unlink it |
| 686 | */ |
| 687 | switch (drop_blk->magic) { |
| 688 | case XFS_ATTR_LEAF_MAGIC: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | error = xfs_attr_leaf_toosmall(state, &action); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | if (error) |
| 691 | return(error); |
| 692 | if (action == 0) |
| 693 | return(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | xfs_attr_leaf_unbalance(state, drop_blk, save_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | break; |
| 696 | case XFS_DIR_LEAF_MAGIC: |
| 697 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 698 | error = xfs_dir_leaf_toosmall(state, &action); |
| 699 | if (error) |
| 700 | return(error); |
| 701 | if (action == 0) |
| 702 | return(0); |
| 703 | xfs_dir_leaf_unbalance(state, drop_blk, save_blk); |
| 704 | break; |
| 705 | case XFS_DIR2_LEAFN_MAGIC: |
| 706 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 707 | error = xfs_dir2_leafn_toosmall(state, &action); |
| 708 | if (error) |
| 709 | return error; |
| 710 | if (action == 0) |
| 711 | return 0; |
| 712 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); |
| 713 | break; |
| 714 | case XFS_DA_NODE_MAGIC: |
| 715 | /* |
| 716 | * Remove the offending node, fixup hashvals, |
| 717 | * check for a toosmall neighbor. |
| 718 | */ |
| 719 | xfs_da_node_remove(state, drop_blk); |
| 720 | xfs_da_fixhashpath(state, &state->path); |
| 721 | error = xfs_da_node_toosmall(state, &action); |
| 722 | if (error) |
| 723 | return(error); |
| 724 | if (action == 0) |
| 725 | return 0; |
| 726 | xfs_da_node_unbalance(state, drop_blk, save_blk); |
| 727 | break; |
| 728 | } |
| 729 | xfs_da_fixhashpath(state, &state->altpath); |
| 730 | error = xfs_da_blk_unlink(state, drop_blk, save_blk); |
| 731 | xfs_da_state_kill_altpath(state); |
| 732 | if (error) |
| 733 | return(error); |
| 734 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, |
| 735 | drop_blk->bp); |
| 736 | drop_blk->bp = NULL; |
| 737 | if (error) |
| 738 | return(error); |
| 739 | } |
| 740 | /* |
| 741 | * We joined all the way to the top. If it turns out that |
| 742 | * we only have one entry in the root, make the child block |
| 743 | * the new root. |
| 744 | */ |
| 745 | xfs_da_node_remove(state, drop_blk); |
| 746 | xfs_da_fixhashpath(state, &state->path); |
| 747 | error = xfs_da_root_join(state, &state->path.blk[0]); |
| 748 | return(error); |
| 749 | } |
| 750 | |
| 751 | /* |
| 752 | * We have only one entry in the root. Copy the only remaining child of |
| 753 | * the old root to block 0 as the new root node. |
| 754 | */ |
| 755 | STATIC int |
| 756 | xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk) |
| 757 | { |
| 758 | xfs_da_intnode_t *oldroot; |
| 759 | /* REFERENCED */ |
| 760 | xfs_da_blkinfo_t *blkinfo; |
| 761 | xfs_da_args_t *args; |
| 762 | xfs_dablk_t child; |
| 763 | xfs_dabuf_t *bp; |
| 764 | int error; |
| 765 | |
| 766 | args = state->args; |
| 767 | ASSERT(args != NULL); |
| 768 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); |
| 769 | oldroot = root_blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 770 | ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | ASSERT(!oldroot->hdr.info.forw); |
| 772 | ASSERT(!oldroot->hdr.info.back); |
| 773 | |
| 774 | /* |
| 775 | * If the root has more than one child, then don't do anything. |
| 776 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 777 | if (be16_to_cpu(oldroot->hdr.count) > 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 778 | return(0); |
| 779 | |
| 780 | /* |
| 781 | * Read in the (only) child block, then copy those bytes into |
| 782 | * the root block's buffer and free the original child block. |
| 783 | */ |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 784 | child = be32_to_cpu(oldroot->btree[0].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | ASSERT(child != 0); |
| 786 | error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp, |
| 787 | args->whichfork); |
| 788 | if (error) |
| 789 | return(error); |
| 790 | ASSERT(bp != NULL); |
| 791 | blkinfo = bp->data; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 792 | if (be16_to_cpu(oldroot->hdr.level) == 1) { |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 793 | ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIRX_LEAF_MAGIC(state->mp) || |
| 794 | be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | } else { |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 796 | ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | } |
| 798 | ASSERT(!blkinfo->forw); |
| 799 | ASSERT(!blkinfo->back); |
| 800 | memcpy(root_blk->bp->data, bp->data, state->blocksize); |
| 801 | xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1); |
| 802 | error = xfs_da_shrink_inode(args, child, bp); |
| 803 | return(error); |
| 804 | } |
| 805 | |
| 806 | /* |
| 807 | * Check a node block and its neighbors to see if the block should be |
| 808 | * collapsed into one or the other neighbor. Always keep the block |
| 809 | * with the smaller block number. |
| 810 | * If the current block is over 50% full, don't try to join it, return 0. |
| 811 | * If the block is empty, fill in the state structure and return 2. |
| 812 | * If it can be collapsed, fill in the state structure and return 1. |
| 813 | * If nothing can be done, return 0. |
| 814 | */ |
| 815 | STATIC int |
| 816 | xfs_da_node_toosmall(xfs_da_state_t *state, int *action) |
| 817 | { |
| 818 | xfs_da_intnode_t *node; |
| 819 | xfs_da_state_blk_t *blk; |
| 820 | xfs_da_blkinfo_t *info; |
| 821 | int count, forward, error, retval, i; |
| 822 | xfs_dablk_t blkno; |
| 823 | xfs_dabuf_t *bp; |
| 824 | |
| 825 | /* |
| 826 | * Check for the degenerate case of the block being over 50% full. |
| 827 | * If so, it's not worth even looking to see if we might be able |
| 828 | * to coalesce with a sibling. |
| 829 | */ |
| 830 | blk = &state->path.blk[ state->path.active-1 ]; |
| 831 | info = blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 832 | ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | node = (xfs_da_intnode_t *)info; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 834 | count = be16_to_cpu(node->hdr.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | if (count > (state->node_ents >> 1)) { |
| 836 | *action = 0; /* blk over 50%, don't try to join */ |
| 837 | return(0); /* blk over 50%, don't try to join */ |
| 838 | } |
| 839 | |
| 840 | /* |
| 841 | * Check for the degenerate case of the block being empty. |
| 842 | * If the block is empty, we'll simply delete it, no need to |
| 843 | * coalesce it with a sibling block. We choose (aribtrarily) |
| 844 | * to merge with the forward block unless it is NULL. |
| 845 | */ |
| 846 | if (count == 0) { |
| 847 | /* |
| 848 | * Make altpath point to the block we want to keep and |
| 849 | * path point to the block we want to drop (this one). |
| 850 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 851 | forward = (info->forw != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
| 853 | error = xfs_da_path_shift(state, &state->altpath, forward, |
| 854 | 0, &retval); |
| 855 | if (error) |
| 856 | return(error); |
| 857 | if (retval) { |
| 858 | *action = 0; |
| 859 | } else { |
| 860 | *action = 2; |
| 861 | } |
| 862 | return(0); |
| 863 | } |
| 864 | |
| 865 | /* |
| 866 | * Examine each sibling block to see if we can coalesce with |
| 867 | * at least 25% free space to spare. We need to figure out |
| 868 | * whether to merge with the forward or the backward block. |
| 869 | * We prefer coalescing with the lower numbered sibling so as |
| 870 | * to shrink a directory over time. |
| 871 | */ |
| 872 | /* start with smaller blk num */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 873 | forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 874 | for (i = 0; i < 2; forward = !forward, i++) { |
| 875 | if (forward) |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 876 | blkno = be32_to_cpu(info->forw); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | else |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 878 | blkno = be32_to_cpu(info->back); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | if (blkno == 0) |
| 880 | continue; |
| 881 | error = xfs_da_read_buf(state->args->trans, state->args->dp, |
| 882 | blkno, -1, &bp, state->args->whichfork); |
| 883 | if (error) |
| 884 | return(error); |
| 885 | ASSERT(bp != NULL); |
| 886 | |
| 887 | node = (xfs_da_intnode_t *)info; |
| 888 | count = state->node_ents; |
| 889 | count -= state->node_ents >> 2; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 890 | count -= be16_to_cpu(node->hdr.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 891 | node = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 892 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 893 | count -= be16_to_cpu(node->hdr.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | xfs_da_brelse(state->args->trans, bp); |
| 895 | if (count >= 0) |
| 896 | break; /* fits with at least 25% to spare */ |
| 897 | } |
| 898 | if (i >= 2) { |
| 899 | *action = 0; |
| 900 | return(0); |
| 901 | } |
| 902 | |
| 903 | /* |
| 904 | * Make altpath point to the block we want to keep (the lower |
| 905 | * numbered block) and path point to the block we want to drop. |
| 906 | */ |
| 907 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
| 908 | if (blkno < blk->blkno) { |
| 909 | error = xfs_da_path_shift(state, &state->altpath, forward, |
| 910 | 0, &retval); |
| 911 | if (error) { |
| 912 | return(error); |
| 913 | } |
| 914 | if (retval) { |
| 915 | *action = 0; |
| 916 | return(0); |
| 917 | } |
| 918 | } else { |
| 919 | error = xfs_da_path_shift(state, &state->path, forward, |
| 920 | 0, &retval); |
| 921 | if (error) { |
| 922 | return(error); |
| 923 | } |
| 924 | if (retval) { |
| 925 | *action = 0; |
| 926 | return(0); |
| 927 | } |
| 928 | } |
| 929 | *action = 1; |
| 930 | return(0); |
| 931 | } |
| 932 | |
| 933 | /* |
| 934 | * Walk back up the tree adjusting hash values as necessary, |
| 935 | * when we stop making changes, return. |
| 936 | */ |
| 937 | void |
| 938 | xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path) |
| 939 | { |
| 940 | xfs_da_state_blk_t *blk; |
| 941 | xfs_da_intnode_t *node; |
| 942 | xfs_da_node_entry_t *btree; |
| 943 | xfs_dahash_t lasthash=0; |
| 944 | int level, count; |
| 945 | |
| 946 | level = path->active-1; |
| 947 | blk = &path->blk[ level ]; |
| 948 | switch (blk->magic) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 949 | case XFS_ATTR_LEAF_MAGIC: |
| 950 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); |
| 951 | if (count == 0) |
| 952 | return; |
| 953 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | case XFS_DIR_LEAF_MAGIC: |
| 955 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 956 | lasthash = xfs_dir_leaf_lasthash(blk->bp, &count); |
| 957 | if (count == 0) |
| 958 | return; |
| 959 | break; |
| 960 | case XFS_DIR2_LEAFN_MAGIC: |
| 961 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 962 | lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count); |
| 963 | if (count == 0) |
| 964 | return; |
| 965 | break; |
| 966 | case XFS_DA_NODE_MAGIC: |
| 967 | lasthash = xfs_da_node_lasthash(blk->bp, &count); |
| 968 | if (count == 0) |
| 969 | return; |
| 970 | break; |
| 971 | } |
| 972 | for (blk--, level--; level >= 0; blk--, level--) { |
| 973 | node = blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 974 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 975 | btree = &node->btree[ blk->index ]; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 976 | if (be32_to_cpu(btree->hashval) == lasthash) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 977 | break; |
| 978 | blk->hashval = lasthash; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 979 | btree->hashval = cpu_to_be32(lasthash); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | xfs_da_log_buf(state->args->trans, blk->bp, |
| 981 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
| 982 | |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 983 | lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | } |
| 985 | } |
| 986 | |
| 987 | /* |
| 988 | * Remove an entry from an intermediate node. |
| 989 | */ |
| 990 | STATIC void |
| 991 | xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk) |
| 992 | { |
| 993 | xfs_da_intnode_t *node; |
| 994 | xfs_da_node_entry_t *btree; |
| 995 | int tmp; |
| 996 | |
| 997 | node = drop_blk->bp->data; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 998 | ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | ASSERT(drop_blk->index >= 0); |
| 1000 | |
| 1001 | /* |
| 1002 | * Copy over the offending entry, or just zero it out. |
| 1003 | */ |
| 1004 | btree = &node->btree[drop_blk->index]; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1005 | if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) { |
| 1006 | tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 1008 | memmove(btree, btree + 1, tmp); |
| 1009 | xfs_da_log_buf(state->args->trans, drop_blk->bp, |
| 1010 | XFS_DA_LOGRANGE(node, btree, tmp)); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1011 | btree = &node->btree[be16_to_cpu(node->hdr.count)-1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1012 | } |
| 1013 | memset((char *)btree, 0, sizeof(xfs_da_node_entry_t)); |
| 1014 | xfs_da_log_buf(state->args->trans, drop_blk->bp, |
| 1015 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1016 | be16_add(&node->hdr.count, -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | xfs_da_log_buf(state->args->trans, drop_blk->bp, |
| 1018 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
| 1019 | |
| 1020 | /* |
| 1021 | * Copy the last hash value from the block to propagate upwards. |
| 1022 | */ |
| 1023 | btree--; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1024 | drop_blk->hashval = be32_to_cpu(btree->hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | } |
| 1026 | |
| 1027 | /* |
| 1028 | * Unbalance the btree elements between two intermediate nodes, |
| 1029 | * move all Btree elements from one node into another. |
| 1030 | */ |
| 1031 | STATIC void |
| 1032 | xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, |
| 1033 | xfs_da_state_blk_t *save_blk) |
| 1034 | { |
| 1035 | xfs_da_intnode_t *drop_node, *save_node; |
| 1036 | xfs_da_node_entry_t *btree; |
| 1037 | int tmp; |
| 1038 | xfs_trans_t *tp; |
| 1039 | |
| 1040 | drop_node = drop_blk->bp->data; |
| 1041 | save_node = save_blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1042 | ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
| 1043 | ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | tp = state->args->trans; |
| 1045 | |
| 1046 | /* |
| 1047 | * If the dying block has lower hashvals, then move all the |
| 1048 | * elements in the remaining block up to make a hole. |
| 1049 | */ |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1050 | if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) || |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1051 | (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) < |
| 1052 | be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | { |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1054 | btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)]; |
| 1055 | tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | memmove(btree, &save_node->btree[0], tmp); |
| 1057 | btree = &save_node->btree[0]; |
| 1058 | xfs_da_log_buf(tp, save_blk->bp, |
| 1059 | XFS_DA_LOGRANGE(save_node, btree, |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1060 | (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1061 | sizeof(xfs_da_node_entry_t))); |
| 1062 | } else { |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1063 | btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | xfs_da_log_buf(tp, save_blk->bp, |
| 1065 | XFS_DA_LOGRANGE(save_node, btree, |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1066 | be16_to_cpu(drop_node->hdr.count) * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1067 | sizeof(xfs_da_node_entry_t))); |
| 1068 | } |
| 1069 | |
| 1070 | /* |
| 1071 | * Move all the B-tree elements from drop_blk to save_blk. |
| 1072 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1073 | tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1074 | memcpy(btree, &drop_node->btree[0], tmp); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1075 | be16_add(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | |
| 1077 | xfs_da_log_buf(tp, save_blk->bp, |
| 1078 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
| 1079 | sizeof(save_node->hdr))); |
| 1080 | |
| 1081 | /* |
| 1082 | * Save the last hashval in the remaining block for upward propagation. |
| 1083 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1084 | save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | } |
| 1086 | |
| 1087 | /*======================================================================== |
| 1088 | * Routines used for finding things in the Btree. |
| 1089 | *========================================================================*/ |
| 1090 | |
| 1091 | /* |
| 1092 | * Walk down the Btree looking for a particular filename, filling |
| 1093 | * in the state structure as we go. |
| 1094 | * |
| 1095 | * We will set the state structure to point to each of the elements |
| 1096 | * in each of the nodes where either the hashval is or should be. |
| 1097 | * |
| 1098 | * We support duplicate hashval's so for each entry in the current |
| 1099 | * node that could contain the desired hashval, descend. This is a |
| 1100 | * pruned depth-first tree search. |
| 1101 | */ |
| 1102 | int /* error */ |
| 1103 | xfs_da_node_lookup_int(xfs_da_state_t *state, int *result) |
| 1104 | { |
| 1105 | xfs_da_state_blk_t *blk; |
| 1106 | xfs_da_blkinfo_t *curr; |
| 1107 | xfs_da_intnode_t *node; |
| 1108 | xfs_da_node_entry_t *btree; |
| 1109 | xfs_dablk_t blkno; |
| 1110 | int probe, span, max, error, retval; |
| 1111 | xfs_dahash_t hashval; |
| 1112 | xfs_da_args_t *args; |
| 1113 | |
| 1114 | args = state->args; |
| 1115 | |
| 1116 | /* |
| 1117 | * Descend thru the B-tree searching each level for the right |
| 1118 | * node to use, until the right hashval is found. |
| 1119 | */ |
| 1120 | if (args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(state->mp)) |
| 1121 | blkno = state->mp->m_dirleafblk; |
| 1122 | else |
| 1123 | blkno = 0; |
| 1124 | for (blk = &state->path.blk[0], state->path.active = 1; |
| 1125 | state->path.active <= XFS_DA_NODE_MAXDEPTH; |
| 1126 | blk++, state->path.active++) { |
| 1127 | /* |
| 1128 | * Read the next node down in the tree. |
| 1129 | */ |
| 1130 | blk->blkno = blkno; |
| 1131 | error = xfs_da_read_buf(args->trans, args->dp, blkno, |
| 1132 | -1, &blk->bp, args->whichfork); |
| 1133 | if (error) { |
| 1134 | blk->blkno = 0; |
| 1135 | state->path.active--; |
| 1136 | return(error); |
| 1137 | } |
| 1138 | curr = blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1139 | ASSERT(be16_to_cpu(curr->magic) == XFS_DA_NODE_MAGIC || |
| 1140 | be16_to_cpu(curr->magic) == XFS_DIRX_LEAF_MAGIC(state->mp) || |
| 1141 | be16_to_cpu(curr->magic) == XFS_ATTR_LEAF_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | |
| 1143 | /* |
| 1144 | * Search an intermediate node for a match. |
| 1145 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1146 | blk->magic = be16_to_cpu(curr->magic); |
| 1147 | if (blk->magic == XFS_DA_NODE_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | node = blk->bp->data; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1149 | blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | |
| 1151 | /* |
| 1152 | * Binary search. (note: small blocks will skip loop) |
| 1153 | */ |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1154 | max = be16_to_cpu(node->hdr.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | probe = span = max / 2; |
| 1156 | hashval = args->hashval; |
| 1157 | for (btree = &node->btree[probe]; span > 4; |
| 1158 | btree = &node->btree[probe]) { |
| 1159 | span /= 2; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1160 | if (be32_to_cpu(btree->hashval) < hashval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1161 | probe += span; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1162 | else if (be32_to_cpu(btree->hashval) > hashval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1163 | probe -= span; |
| 1164 | else |
| 1165 | break; |
| 1166 | } |
| 1167 | ASSERT((probe >= 0) && (probe < max)); |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1168 | ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | |
| 1170 | /* |
| 1171 | * Since we may have duplicate hashval's, find the first |
| 1172 | * matching hashval in the node. |
| 1173 | */ |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1174 | while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1175 | btree--; |
| 1176 | probe--; |
| 1177 | } |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1178 | while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | btree++; |
| 1180 | probe++; |
| 1181 | } |
| 1182 | |
| 1183 | /* |
| 1184 | * Pick the right block to descend on. |
| 1185 | */ |
| 1186 | if (probe == max) { |
| 1187 | blk->index = max-1; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1188 | blkno = be32_to_cpu(node->btree[max-1].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1189 | } else { |
| 1190 | blk->index = probe; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1191 | blkno = be32_to_cpu(btree->before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1192 | } |
| 1193 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1194 | else if (be16_to_cpu(curr->magic) == XFS_ATTR_LEAF_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
| 1196 | break; |
| 1197 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1198 | else if (be16_to_cpu(curr->magic) == XFS_DIR_LEAF_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1199 | blk->hashval = xfs_dir_leaf_lasthash(blk->bp, NULL); |
| 1200 | break; |
| 1201 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1202 | else if (be16_to_cpu(curr->magic) == XFS_DIR2_LEAFN_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1203 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL); |
| 1204 | break; |
| 1205 | } |
| 1206 | } |
| 1207 | |
| 1208 | /* |
| 1209 | * A leaf block that ends in the hashval that we are interested in |
| 1210 | * (final hashval == search hashval) means that the next block may |
| 1211 | * contain more entries with the same hashval, shift upward to the |
| 1212 | * next leaf and keep searching. |
| 1213 | */ |
| 1214 | for (;;) { |
| 1215 | if (blk->magic == XFS_DIR_LEAF_MAGIC) { |
| 1216 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 1217 | retval = xfs_dir_leaf_lookup_int(blk->bp, args, |
| 1218 | &blk->index); |
| 1219 | } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
| 1220 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 1221 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
| 1222 | &blk->index, state); |
| 1223 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
| 1225 | retval = xfs_attr_leaf_lookup_int(blk->bp, args); |
| 1226 | blk->index = args->index; |
| 1227 | args->blkno = blk->blkno; |
| 1228 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | if (((retval == ENOENT) || (retval == ENOATTR)) && |
| 1230 | (blk->hashval == args->hashval)) { |
| 1231 | error = xfs_da_path_shift(state, &state->path, 1, 1, |
| 1232 | &retval); |
| 1233 | if (error) |
| 1234 | return(error); |
| 1235 | if (retval == 0) { |
| 1236 | continue; |
| 1237 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1238 | else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
| 1239 | /* path_shift() gives ENOENT */ |
| 1240 | retval = XFS_ERROR(ENOATTR); |
| 1241 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | } |
| 1243 | break; |
| 1244 | } |
| 1245 | *result = retval; |
| 1246 | return(0); |
| 1247 | } |
| 1248 | |
| 1249 | /*======================================================================== |
| 1250 | * Utility routines. |
| 1251 | *========================================================================*/ |
| 1252 | |
| 1253 | /* |
| 1254 | * Link a new block into a doubly linked list of blocks (of whatever type). |
| 1255 | */ |
| 1256 | int /* error */ |
| 1257 | xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk, |
| 1258 | xfs_da_state_blk_t *new_blk) |
| 1259 | { |
| 1260 | xfs_da_blkinfo_t *old_info, *new_info, *tmp_info; |
| 1261 | xfs_da_args_t *args; |
| 1262 | int before=0, error; |
| 1263 | xfs_dabuf_t *bp; |
| 1264 | |
| 1265 | /* |
| 1266 | * Set up environment. |
| 1267 | */ |
| 1268 | args = state->args; |
| 1269 | ASSERT(args != NULL); |
| 1270 | old_info = old_blk->bp->data; |
| 1271 | new_info = new_blk->bp->data; |
| 1272 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
| 1273 | old_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) || |
| 1274 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1275 | ASSERT(old_blk->magic == be16_to_cpu(old_info->magic)); |
| 1276 | ASSERT(new_blk->magic == be16_to_cpu(new_info->magic)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | ASSERT(old_blk->magic == new_blk->magic); |
| 1278 | |
| 1279 | switch (old_blk->magic) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | case XFS_ATTR_LEAF_MAGIC: |
| 1281 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); |
| 1282 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | case XFS_DIR_LEAF_MAGIC: |
| 1284 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 1285 | before = xfs_dir_leaf_order(old_blk->bp, new_blk->bp); |
| 1286 | break; |
| 1287 | case XFS_DIR2_LEAFN_MAGIC: |
| 1288 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 1289 | before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp); |
| 1290 | break; |
| 1291 | case XFS_DA_NODE_MAGIC: |
| 1292 | before = xfs_da_node_order(old_blk->bp, new_blk->bp); |
| 1293 | break; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * Link blocks in appropriate order. |
| 1298 | */ |
| 1299 | if (before) { |
| 1300 | /* |
| 1301 | * Link new block in before existing block. |
| 1302 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1303 | new_info->forw = cpu_to_be32(old_blk->blkno); |
| 1304 | new_info->back = old_info->back; |
| 1305 | if (old_info->back) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1306 | error = xfs_da_read_buf(args->trans, args->dp, |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1307 | be32_to_cpu(old_info->back), |
| 1308 | -1, &bp, args->whichfork); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1309 | if (error) |
| 1310 | return(error); |
| 1311 | ASSERT(bp != NULL); |
| 1312 | tmp_info = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1313 | ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic)); |
| 1314 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); |
| 1315 | tmp_info->forw = cpu_to_be32(new_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
| 1317 | xfs_da_buf_done(bp); |
| 1318 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1319 | old_info->back = cpu_to_be32(new_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | } else { |
| 1321 | /* |
| 1322 | * Link new block in after existing block. |
| 1323 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1324 | new_info->forw = old_info->forw; |
| 1325 | new_info->back = cpu_to_be32(old_blk->blkno); |
| 1326 | if (old_info->forw) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | error = xfs_da_read_buf(args->trans, args->dp, |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1328 | be32_to_cpu(old_info->forw), |
| 1329 | -1, &bp, args->whichfork); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1330 | if (error) |
| 1331 | return(error); |
| 1332 | ASSERT(bp != NULL); |
| 1333 | tmp_info = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1334 | ASSERT(tmp_info->magic == old_info->magic); |
| 1335 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); |
| 1336 | tmp_info->back = cpu_to_be32(new_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
| 1338 | xfs_da_buf_done(bp); |
| 1339 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1340 | old_info->forw = cpu_to_be32(new_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | } |
| 1342 | |
| 1343 | xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
| 1344 | xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); |
| 1345 | return(0); |
| 1346 | } |
| 1347 | |
| 1348 | /* |
| 1349 | * Compare two intermediate nodes for "order". |
| 1350 | */ |
| 1351 | STATIC int |
| 1352 | xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp) |
| 1353 | { |
| 1354 | xfs_da_intnode_t *node1, *node2; |
| 1355 | |
| 1356 | node1 = node1_bp->data; |
| 1357 | node2 = node2_bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1358 | ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) && |
| 1359 | (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC)); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1360 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1361 | ((be32_to_cpu(node2->btree[0].hashval) < |
| 1362 | be32_to_cpu(node1->btree[0].hashval)) || |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1363 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
| 1364 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | return(1); |
| 1366 | } |
| 1367 | return(0); |
| 1368 | } |
| 1369 | |
| 1370 | /* |
| 1371 | * Pick up the last hashvalue from an intermediate node. |
| 1372 | */ |
| 1373 | STATIC uint |
| 1374 | xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count) |
| 1375 | { |
| 1376 | xfs_da_intnode_t *node; |
| 1377 | |
| 1378 | node = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1379 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | if (count) |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1381 | *count = be16_to_cpu(node->hdr.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | if (!node->hdr.count) |
| 1383 | return(0); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1384 | return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | } |
| 1386 | |
| 1387 | /* |
| 1388 | * Unlink a block from a doubly linked list of blocks. |
| 1389 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 1390 | STATIC int /* error */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, |
| 1392 | xfs_da_state_blk_t *save_blk) |
| 1393 | { |
| 1394 | xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info; |
| 1395 | xfs_da_args_t *args; |
| 1396 | xfs_dabuf_t *bp; |
| 1397 | int error; |
| 1398 | |
| 1399 | /* |
| 1400 | * Set up environment. |
| 1401 | */ |
| 1402 | args = state->args; |
| 1403 | ASSERT(args != NULL); |
| 1404 | save_info = save_blk->bp->data; |
| 1405 | drop_info = drop_blk->bp->data; |
| 1406 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
| 1407 | save_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) || |
| 1408 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1409 | ASSERT(save_blk->magic == be16_to_cpu(save_info->magic)); |
| 1410 | ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | ASSERT(save_blk->magic == drop_blk->magic); |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1412 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
| 1413 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); |
| 1414 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || |
| 1415 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | |
| 1417 | /* |
| 1418 | * Unlink the leaf block from the doubly linked chain of leaves. |
| 1419 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1420 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
| 1421 | save_info->back = drop_info->back; |
| 1422 | if (drop_info->back) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1423 | error = xfs_da_read_buf(args->trans, args->dp, |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1424 | be32_to_cpu(drop_info->back), |
| 1425 | -1, &bp, args->whichfork); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1426 | if (error) |
| 1427 | return(error); |
| 1428 | ASSERT(bp != NULL); |
| 1429 | tmp_info = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1430 | ASSERT(tmp_info->magic == save_info->magic); |
| 1431 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); |
| 1432 | tmp_info->forw = cpu_to_be32(save_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | xfs_da_log_buf(args->trans, bp, 0, |
| 1434 | sizeof(*tmp_info) - 1); |
| 1435 | xfs_da_buf_done(bp); |
| 1436 | } |
| 1437 | } else { |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1438 | save_info->forw = drop_info->forw; |
| 1439 | if (drop_info->forw) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1440 | error = xfs_da_read_buf(args->trans, args->dp, |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1441 | be32_to_cpu(drop_info->forw), |
| 1442 | -1, &bp, args->whichfork); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | if (error) |
| 1444 | return(error); |
| 1445 | ASSERT(bp != NULL); |
| 1446 | tmp_info = bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1447 | ASSERT(tmp_info->magic == save_info->magic); |
| 1448 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); |
| 1449 | tmp_info->back = cpu_to_be32(save_blk->blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | xfs_da_log_buf(args->trans, bp, 0, |
| 1451 | sizeof(*tmp_info) - 1); |
| 1452 | xfs_da_buf_done(bp); |
| 1453 | } |
| 1454 | } |
| 1455 | |
| 1456 | xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
| 1457 | return(0); |
| 1458 | } |
| 1459 | |
| 1460 | /* |
| 1461 | * Move a path "forward" or "!forward" one block at the current level. |
| 1462 | * |
| 1463 | * This routine will adjust a "path" to point to the next block |
| 1464 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the |
| 1465 | * Btree, including updating pointers to the intermediate nodes between |
| 1466 | * the new bottom and the root. |
| 1467 | */ |
| 1468 | int /* error */ |
| 1469 | xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path, |
| 1470 | int forward, int release, int *result) |
| 1471 | { |
| 1472 | xfs_da_state_blk_t *blk; |
| 1473 | xfs_da_blkinfo_t *info; |
| 1474 | xfs_da_intnode_t *node; |
| 1475 | xfs_da_args_t *args; |
| 1476 | xfs_dablk_t blkno=0; |
| 1477 | int level, error; |
| 1478 | |
| 1479 | /* |
| 1480 | * Roll up the Btree looking for the first block where our |
| 1481 | * current index is not at the edge of the block. Note that |
| 1482 | * we skip the bottom layer because we want the sibling block. |
| 1483 | */ |
| 1484 | args = state->args; |
| 1485 | ASSERT(args != NULL); |
| 1486 | ASSERT(path != NULL); |
| 1487 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); |
| 1488 | level = (path->active-1) - 1; /* skip bottom layer in path */ |
| 1489 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { |
| 1490 | ASSERT(blk->bp != NULL); |
| 1491 | node = blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1492 | ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC); |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1493 | if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | blk->index++; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1495 | blkno = be32_to_cpu(node->btree[blk->index].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1496 | break; |
| 1497 | } else if (!forward && (blk->index > 0)) { |
| 1498 | blk->index--; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1499 | blkno = be32_to_cpu(node->btree[blk->index].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | break; |
| 1501 | } |
| 1502 | } |
| 1503 | if (level < 0) { |
| 1504 | *result = XFS_ERROR(ENOENT); /* we're out of our tree */ |
| 1505 | ASSERT(args->oknoent); |
| 1506 | return(0); |
| 1507 | } |
| 1508 | |
| 1509 | /* |
| 1510 | * Roll down the edge of the subtree until we reach the |
| 1511 | * same depth we were at originally. |
| 1512 | */ |
| 1513 | for (blk++, level++; level < path->active; blk++, level++) { |
| 1514 | /* |
| 1515 | * Release the old block. |
| 1516 | * (if it's dirty, trans won't actually let go) |
| 1517 | */ |
| 1518 | if (release) |
| 1519 | xfs_da_brelse(args->trans, blk->bp); |
| 1520 | |
| 1521 | /* |
| 1522 | * Read the next child block. |
| 1523 | */ |
| 1524 | blk->blkno = blkno; |
| 1525 | error = xfs_da_read_buf(args->trans, args->dp, blkno, -1, |
| 1526 | &blk->bp, args->whichfork); |
| 1527 | if (error) |
| 1528 | return(error); |
| 1529 | ASSERT(blk->bp != NULL); |
| 1530 | info = blk->bp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1531 | ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC || |
| 1532 | be16_to_cpu(info->magic) == XFS_DIRX_LEAF_MAGIC(state->mp) || |
| 1533 | be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC); |
| 1534 | blk->magic = be16_to_cpu(info->magic); |
| 1535 | if (blk->magic == XFS_DA_NODE_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | node = (xfs_da_intnode_t *)info; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1537 | blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | if (forward) |
| 1539 | blk->index = 0; |
| 1540 | else |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1541 | blk->index = be16_to_cpu(node->hdr.count)-1; |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1542 | blkno = be32_to_cpu(node->btree[blk->index].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1543 | } else { |
| 1544 | ASSERT(level == path->active-1); |
| 1545 | blk->index = 0; |
| 1546 | switch(blk->magic) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1547 | case XFS_ATTR_LEAF_MAGIC: |
| 1548 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, |
| 1549 | NULL); |
| 1550 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1551 | case XFS_DIR_LEAF_MAGIC: |
| 1552 | ASSERT(XFS_DIR_IS_V1(state->mp)); |
| 1553 | blk->hashval = xfs_dir_leaf_lasthash(blk->bp, |
| 1554 | NULL); |
| 1555 | break; |
| 1556 | case XFS_DIR2_LEAFN_MAGIC: |
| 1557 | ASSERT(XFS_DIR_IS_V2(state->mp)); |
| 1558 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, |
| 1559 | NULL); |
| 1560 | break; |
| 1561 | default: |
| 1562 | ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC || |
| 1563 | blk->magic == |
| 1564 | XFS_DIRX_LEAF_MAGIC(state->mp)); |
| 1565 | break; |
| 1566 | } |
| 1567 | } |
| 1568 | } |
| 1569 | *result = 0; |
| 1570 | return(0); |
| 1571 | } |
| 1572 | |
| 1573 | |
| 1574 | /*======================================================================== |
| 1575 | * Utility routines. |
| 1576 | *========================================================================*/ |
| 1577 | |
| 1578 | /* |
| 1579 | * Implement a simple hash on a character string. |
| 1580 | * Rotate the hash value by 7 bits, then XOR each character in. |
| 1581 | * This is implemented with some source-level loop unrolling. |
| 1582 | */ |
| 1583 | xfs_dahash_t |
Nathan Scott | 80cce77 | 2005-11-02 11:43:04 +1100 | [diff] [blame] | 1584 | xfs_da_hashname(const uchar_t *name, int namelen) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | { |
| 1586 | xfs_dahash_t hash; |
| 1587 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1588 | /* |
| 1589 | * Do four characters at a time as long as we can. |
| 1590 | */ |
| 1591 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) |
| 1592 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ |
| 1593 | (name[3] << 0) ^ rol32(hash, 7 * 4); |
| 1594 | |
| 1595 | /* |
| 1596 | * Now do the rest of the characters. |
| 1597 | */ |
| 1598 | switch (namelen) { |
| 1599 | case 3: |
| 1600 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ |
| 1601 | rol32(hash, 7 * 3); |
| 1602 | case 2: |
| 1603 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); |
| 1604 | case 1: |
| 1605 | return (name[0] << 0) ^ rol32(hash, 7 * 1); |
Nathan Scott | 2b3b6d0 | 2005-11-02 15:12:28 +1100 | [diff] [blame] | 1606 | default: /* case 0: */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1607 | return hash; |
| 1608 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1609 | } |
| 1610 | |
| 1611 | /* |
| 1612 | * Add a block to the btree ahead of the file. |
| 1613 | * Return the new block number to the caller. |
| 1614 | */ |
| 1615 | int |
| 1616 | xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno) |
| 1617 | { |
| 1618 | xfs_fileoff_t bno, b; |
| 1619 | xfs_bmbt_irec_t map; |
| 1620 | xfs_bmbt_irec_t *mapp; |
| 1621 | xfs_inode_t *dp; |
| 1622 | int nmap, error, w, count, c, got, i, mapi; |
| 1623 | xfs_fsize_t size; |
| 1624 | xfs_trans_t *tp; |
| 1625 | xfs_mount_t *mp; |
| 1626 | |
| 1627 | dp = args->dp; |
| 1628 | mp = dp->i_mount; |
| 1629 | w = args->whichfork; |
| 1630 | tp = args->trans; |
| 1631 | /* |
| 1632 | * For new directories adjust the file offset and block count. |
| 1633 | */ |
| 1634 | if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) { |
| 1635 | bno = mp->m_dirleafblk; |
| 1636 | count = mp->m_dirblkfsbs; |
| 1637 | } else { |
| 1638 | bno = 0; |
| 1639 | count = 1; |
| 1640 | } |
| 1641 | /* |
| 1642 | * Find a spot in the file space to put the new block. |
| 1643 | */ |
| 1644 | if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w))) { |
| 1645 | return error; |
| 1646 | } |
| 1647 | if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) |
| 1648 | ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk); |
| 1649 | /* |
| 1650 | * Try mapping it in one filesystem block. |
| 1651 | */ |
| 1652 | nmap = 1; |
| 1653 | ASSERT(args->firstblock != NULL); |
| 1654 | if ((error = xfs_bmapi(tp, dp, bno, count, |
| 1655 | XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA| |
| 1656 | XFS_BMAPI_CONTIG, |
| 1657 | args->firstblock, args->total, &map, &nmap, |
| 1658 | args->flist))) { |
| 1659 | return error; |
| 1660 | } |
| 1661 | ASSERT(nmap <= 1); |
| 1662 | if (nmap == 1) { |
| 1663 | mapp = ↦ |
| 1664 | mapi = 1; |
| 1665 | } |
| 1666 | /* |
| 1667 | * If we didn't get it and the block might work if fragmented, |
| 1668 | * try without the CONTIG flag. Loop until we get it all. |
| 1669 | */ |
| 1670 | else if (nmap == 0 && count > 1) { |
| 1671 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
| 1672 | for (b = bno, mapi = 0; b < bno + count; ) { |
| 1673 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
| 1674 | c = (int)(bno + count - b); |
| 1675 | if ((error = xfs_bmapi(tp, dp, b, c, |
| 1676 | XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE| |
| 1677 | XFS_BMAPI_METADATA, |
| 1678 | args->firstblock, args->total, |
| 1679 | &mapp[mapi], &nmap, args->flist))) { |
| 1680 | kmem_free(mapp, sizeof(*mapp) * count); |
| 1681 | return error; |
| 1682 | } |
| 1683 | if (nmap < 1) |
| 1684 | break; |
| 1685 | mapi += nmap; |
| 1686 | b = mapp[mapi - 1].br_startoff + |
| 1687 | mapp[mapi - 1].br_blockcount; |
| 1688 | } |
| 1689 | } else { |
| 1690 | mapi = 0; |
| 1691 | mapp = NULL; |
| 1692 | } |
| 1693 | /* |
| 1694 | * Count the blocks we got, make sure it matches the total. |
| 1695 | */ |
| 1696 | for (i = 0, got = 0; i < mapi; i++) |
| 1697 | got += mapp[i].br_blockcount; |
| 1698 | if (got != count || mapp[0].br_startoff != bno || |
| 1699 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
| 1700 | bno + count) { |
| 1701 | if (mapp != &map) |
| 1702 | kmem_free(mapp, sizeof(*mapp) * count); |
| 1703 | return XFS_ERROR(ENOSPC); |
| 1704 | } |
| 1705 | if (mapp != &map) |
| 1706 | kmem_free(mapp, sizeof(*mapp) * count); |
| 1707 | *new_blkno = (xfs_dablk_t)bno; |
| 1708 | /* |
| 1709 | * For version 1 directories, adjust the file size if it changed. |
| 1710 | */ |
| 1711 | if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) { |
| 1712 | ASSERT(mapi == 1); |
| 1713 | if ((error = xfs_bmap_last_offset(tp, dp, &bno, w))) |
| 1714 | return error; |
| 1715 | size = XFS_FSB_TO_B(mp, bno); |
| 1716 | if (size != dp->i_d.di_size) { |
| 1717 | dp->i_d.di_size = size; |
| 1718 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
| 1719 | } |
| 1720 | } |
| 1721 | return 0; |
| 1722 | } |
| 1723 | |
| 1724 | /* |
| 1725 | * Ick. We need to always be able to remove a btree block, even |
| 1726 | * if there's no space reservation because the filesystem is full. |
| 1727 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. |
| 1728 | * It swaps the target block with the last block in the file. The |
| 1729 | * last block in the file can always be removed since it can't cause |
| 1730 | * a bmap btree split to do that. |
| 1731 | */ |
| 1732 | STATIC int |
| 1733 | xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop, |
| 1734 | xfs_dabuf_t **dead_bufp) |
| 1735 | { |
| 1736 | xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno; |
| 1737 | xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf; |
| 1738 | xfs_fileoff_t lastoff; |
| 1739 | xfs_inode_t *ip; |
| 1740 | xfs_trans_t *tp; |
| 1741 | xfs_mount_t *mp; |
| 1742 | int error, w, entno, level, dead_level; |
| 1743 | xfs_da_blkinfo_t *dead_info, *sib_info; |
| 1744 | xfs_da_intnode_t *par_node, *dead_node; |
| 1745 | xfs_dir_leafblock_t *dead_leaf; |
| 1746 | xfs_dir2_leaf_t *dead_leaf2; |
| 1747 | xfs_dahash_t dead_hash; |
| 1748 | |
| 1749 | dead_buf = *dead_bufp; |
| 1750 | dead_blkno = *dead_blknop; |
| 1751 | tp = args->trans; |
| 1752 | ip = args->dp; |
| 1753 | w = args->whichfork; |
| 1754 | ASSERT(w == XFS_DATA_FORK); |
| 1755 | mp = ip->i_mount; |
| 1756 | if (XFS_DIR_IS_V2(mp)) { |
| 1757 | lastoff = mp->m_dirfreeblk; |
| 1758 | error = xfs_bmap_last_before(tp, ip, &lastoff, w); |
| 1759 | } else |
| 1760 | error = xfs_bmap_last_offset(tp, ip, &lastoff, w); |
| 1761 | if (error) |
| 1762 | return error; |
| 1763 | if (unlikely(lastoff == 0)) { |
| 1764 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, |
| 1765 | mp); |
| 1766 | return XFS_ERROR(EFSCORRUPTED); |
| 1767 | } |
| 1768 | /* |
| 1769 | * Read the last block in the btree space. |
| 1770 | */ |
| 1771 | last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs; |
| 1772 | if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w))) |
| 1773 | return error; |
| 1774 | /* |
| 1775 | * Copy the last block into the dead buffer and log it. |
| 1776 | */ |
| 1777 | memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize); |
| 1778 | xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1); |
| 1779 | dead_info = dead_buf->data; |
| 1780 | /* |
| 1781 | * Get values from the moved block. |
| 1782 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1783 | if (be16_to_cpu(dead_info->magic) == XFS_DIR_LEAF_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1784 | ASSERT(XFS_DIR_IS_V1(mp)); |
| 1785 | dead_leaf = (xfs_dir_leafblock_t *)dead_info; |
| 1786 | dead_level = 0; |
| 1787 | dead_hash = |
| 1788 | INT_GET(dead_leaf->entries[INT_GET(dead_leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT); |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1789 | } else if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | ASSERT(XFS_DIR_IS_V2(mp)); |
| 1791 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
| 1792 | dead_level = 0; |
Nathan Scott | 3c1f9c1 | 2006-03-17 17:28:18 +1100 | [diff] [blame] | 1793 | dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1794 | } else { |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1795 | ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | dead_node = (xfs_da_intnode_t *)dead_info; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1797 | dead_level = be16_to_cpu(dead_node->hdr.level); |
| 1798 | dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | } |
| 1800 | sib_buf = par_buf = NULL; |
| 1801 | /* |
| 1802 | * If the moved block has a left sibling, fix up the pointers. |
| 1803 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1804 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1805 | if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w))) |
| 1806 | goto done; |
| 1807 | sib_info = sib_buf->data; |
| 1808 | if (unlikely( |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1809 | be32_to_cpu(sib_info->forw) != last_blkno || |
| 1810 | sib_info->magic != dead_info->magic)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
| 1812 | XFS_ERRLEVEL_LOW, mp); |
| 1813 | error = XFS_ERROR(EFSCORRUPTED); |
| 1814 | goto done; |
| 1815 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1816 | sib_info->forw = cpu_to_be32(dead_blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | xfs_da_log_buf(tp, sib_buf, |
| 1818 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
| 1819 | sizeof(sib_info->forw))); |
| 1820 | xfs_da_buf_done(sib_buf); |
| 1821 | sib_buf = NULL; |
| 1822 | } |
| 1823 | /* |
| 1824 | * If the moved block has a right sibling, fix up the pointers. |
| 1825 | */ |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1826 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1827 | if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w))) |
| 1828 | goto done; |
| 1829 | sib_info = sib_buf->data; |
| 1830 | if (unlikely( |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1831 | be32_to_cpu(sib_info->back) != last_blkno || |
| 1832 | sib_info->magic != dead_info->magic)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
| 1834 | XFS_ERRLEVEL_LOW, mp); |
| 1835 | error = XFS_ERROR(EFSCORRUPTED); |
| 1836 | goto done; |
| 1837 | } |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1838 | sib_info->back = cpu_to_be32(dead_blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | xfs_da_log_buf(tp, sib_buf, |
| 1840 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
| 1841 | sizeof(sib_info->back))); |
| 1842 | xfs_da_buf_done(sib_buf); |
| 1843 | sib_buf = NULL; |
| 1844 | } |
| 1845 | par_blkno = XFS_DIR_IS_V1(mp) ? 0 : mp->m_dirleafblk; |
| 1846 | level = -1; |
| 1847 | /* |
| 1848 | * Walk down the tree looking for the parent of the moved block. |
| 1849 | */ |
| 1850 | for (;;) { |
| 1851 | if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w))) |
| 1852 | goto done; |
| 1853 | par_node = par_buf->data; |
| 1854 | if (unlikely( |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1855 | be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC || |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1856 | (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1857 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
| 1858 | XFS_ERRLEVEL_LOW, mp); |
| 1859 | error = XFS_ERROR(EFSCORRUPTED); |
| 1860 | goto done; |
| 1861 | } |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1862 | level = be16_to_cpu(par_node->hdr.level); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1863 | for (entno = 0; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1864 | entno < be16_to_cpu(par_node->hdr.count) && |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1865 | be32_to_cpu(par_node->btree[entno].hashval) < dead_hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1866 | entno++) |
| 1867 | continue; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1868 | if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1869 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
| 1870 | XFS_ERRLEVEL_LOW, mp); |
| 1871 | error = XFS_ERROR(EFSCORRUPTED); |
| 1872 | goto done; |
| 1873 | } |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1874 | par_blkno = be32_to_cpu(par_node->btree[entno].before); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1875 | if (level == dead_level + 1) |
| 1876 | break; |
| 1877 | xfs_da_brelse(tp, par_buf); |
| 1878 | par_buf = NULL; |
| 1879 | } |
| 1880 | /* |
| 1881 | * We're in the right parent block. |
| 1882 | * Look for the right entry. |
| 1883 | */ |
| 1884 | for (;;) { |
| 1885 | for (; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1886 | entno < be16_to_cpu(par_node->hdr.count) && |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1887 | be32_to_cpu(par_node->btree[entno].before) != last_blkno; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | entno++) |
| 1889 | continue; |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1890 | if (entno < be16_to_cpu(par_node->hdr.count)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1891 | break; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1892 | par_blkno = be32_to_cpu(par_node->hdr.info.forw); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1893 | xfs_da_brelse(tp, par_buf); |
| 1894 | par_buf = NULL; |
| 1895 | if (unlikely(par_blkno == 0)) { |
| 1896 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", |
| 1897 | XFS_ERRLEVEL_LOW, mp); |
| 1898 | error = XFS_ERROR(EFSCORRUPTED); |
| 1899 | goto done; |
| 1900 | } |
| 1901 | if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w))) |
| 1902 | goto done; |
| 1903 | par_node = par_buf->data; |
| 1904 | if (unlikely( |
Nathan Scott | fac80cc | 2006-03-17 17:29:56 +1100 | [diff] [blame^] | 1905 | be16_to_cpu(par_node->hdr.level) != level || |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 1906 | be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1907 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
| 1908 | XFS_ERRLEVEL_LOW, mp); |
| 1909 | error = XFS_ERROR(EFSCORRUPTED); |
| 1910 | goto done; |
| 1911 | } |
| 1912 | entno = 0; |
| 1913 | } |
| 1914 | /* |
| 1915 | * Update the parent entry pointing to the moved block. |
| 1916 | */ |
Nathan Scott | 403432d | 2006-03-17 17:29:46 +1100 | [diff] [blame] | 1917 | par_node->btree[entno].before = cpu_to_be32(dead_blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1918 | xfs_da_log_buf(tp, par_buf, |
| 1919 | XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before, |
| 1920 | sizeof(par_node->btree[entno].before))); |
| 1921 | xfs_da_buf_done(par_buf); |
| 1922 | xfs_da_buf_done(dead_buf); |
| 1923 | *dead_blknop = last_blkno; |
| 1924 | *dead_bufp = last_buf; |
| 1925 | return 0; |
| 1926 | done: |
| 1927 | if (par_buf) |
| 1928 | xfs_da_brelse(tp, par_buf); |
| 1929 | if (sib_buf) |
| 1930 | xfs_da_brelse(tp, sib_buf); |
| 1931 | xfs_da_brelse(tp, last_buf); |
| 1932 | return error; |
| 1933 | } |
| 1934 | |
| 1935 | /* |
| 1936 | * Remove a btree block from a directory or attribute. |
| 1937 | */ |
| 1938 | int |
| 1939 | xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno, |
| 1940 | xfs_dabuf_t *dead_buf) |
| 1941 | { |
| 1942 | xfs_inode_t *dp; |
| 1943 | int done, error, w, count; |
| 1944 | xfs_fileoff_t bno; |
| 1945 | xfs_fsize_t size; |
| 1946 | xfs_trans_t *tp; |
| 1947 | xfs_mount_t *mp; |
| 1948 | |
| 1949 | dp = args->dp; |
| 1950 | w = args->whichfork; |
| 1951 | tp = args->trans; |
| 1952 | mp = dp->i_mount; |
| 1953 | if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) |
| 1954 | count = mp->m_dirblkfsbs; |
| 1955 | else |
| 1956 | count = 1; |
| 1957 | for (;;) { |
| 1958 | /* |
| 1959 | * Remove extents. If we get ENOSPC for a dir we have to move |
| 1960 | * the last block to the place we want to kill. |
| 1961 | */ |
| 1962 | if ((error = xfs_bunmapi(tp, dp, dead_blkno, count, |
| 1963 | XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA, |
| 1964 | 0, args->firstblock, args->flist, |
| 1965 | &done)) == ENOSPC) { |
| 1966 | if (w != XFS_DATA_FORK) |
| 1967 | goto done; |
| 1968 | if ((error = xfs_da_swap_lastblock(args, &dead_blkno, |
| 1969 | &dead_buf))) |
| 1970 | goto done; |
| 1971 | } else if (error) |
| 1972 | goto done; |
| 1973 | else |
| 1974 | break; |
| 1975 | } |
| 1976 | ASSERT(done); |
| 1977 | xfs_da_binval(tp, dead_buf); |
| 1978 | /* |
| 1979 | * Adjust the directory size for version 1. |
| 1980 | */ |
| 1981 | if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) { |
| 1982 | if ((error = xfs_bmap_last_offset(tp, dp, &bno, w))) |
| 1983 | return error; |
| 1984 | size = XFS_FSB_TO_B(dp->i_mount, bno); |
| 1985 | if (size != dp->i_d.di_size) { |
| 1986 | dp->i_d.di_size = size; |
| 1987 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
| 1988 | } |
| 1989 | } |
| 1990 | return 0; |
| 1991 | done: |
| 1992 | xfs_da_binval(tp, dead_buf); |
| 1993 | return error; |
| 1994 | } |
| 1995 | |
| 1996 | /* |
| 1997 | * See if the mapping(s) for this btree block are valid, i.e. |
| 1998 | * don't contain holes, are logically contiguous, and cover the whole range. |
| 1999 | */ |
| 2000 | STATIC int |
| 2001 | xfs_da_map_covers_blocks( |
| 2002 | int nmap, |
| 2003 | xfs_bmbt_irec_t *mapp, |
| 2004 | xfs_dablk_t bno, |
| 2005 | int count) |
| 2006 | { |
| 2007 | int i; |
| 2008 | xfs_fileoff_t off; |
| 2009 | |
| 2010 | for (i = 0, off = bno; i < nmap; i++) { |
| 2011 | if (mapp[i].br_startblock == HOLESTARTBLOCK || |
| 2012 | mapp[i].br_startblock == DELAYSTARTBLOCK) { |
| 2013 | return 0; |
| 2014 | } |
| 2015 | if (off != mapp[i].br_startoff) { |
| 2016 | return 0; |
| 2017 | } |
| 2018 | off += mapp[i].br_blockcount; |
| 2019 | } |
| 2020 | return off == bno + count; |
| 2021 | } |
| 2022 | |
| 2023 | /* |
| 2024 | * Make a dabuf. |
| 2025 | * Used for get_buf, read_buf, read_bufr, and reada_buf. |
| 2026 | */ |
| 2027 | STATIC int |
| 2028 | xfs_da_do_buf( |
| 2029 | xfs_trans_t *trans, |
| 2030 | xfs_inode_t *dp, |
| 2031 | xfs_dablk_t bno, |
| 2032 | xfs_daddr_t *mappedbnop, |
| 2033 | xfs_dabuf_t **bpp, |
| 2034 | int whichfork, |
| 2035 | int caller, |
| 2036 | inst_t *ra) |
| 2037 | { |
| 2038 | xfs_buf_t *bp = NULL; |
| 2039 | xfs_buf_t **bplist; |
| 2040 | int error=0; |
| 2041 | int i; |
| 2042 | xfs_bmbt_irec_t map; |
| 2043 | xfs_bmbt_irec_t *mapp; |
| 2044 | xfs_daddr_t mappedbno; |
| 2045 | xfs_mount_t *mp; |
| 2046 | int nbplist=0; |
| 2047 | int nfsb; |
| 2048 | int nmap; |
| 2049 | xfs_dabuf_t *rbp; |
| 2050 | |
| 2051 | mp = dp->i_mount; |
| 2052 | if (whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) |
| 2053 | nfsb = mp->m_dirblkfsbs; |
| 2054 | else |
| 2055 | nfsb = 1; |
| 2056 | mappedbno = *mappedbnop; |
| 2057 | /* |
| 2058 | * Caller doesn't have a mapping. -2 means don't complain |
| 2059 | * if we land in a hole. |
| 2060 | */ |
| 2061 | if (mappedbno == -1 || mappedbno == -2) { |
| 2062 | /* |
| 2063 | * Optimize the one-block case. |
| 2064 | */ |
| 2065 | if (nfsb == 1) { |
| 2066 | xfs_fsblock_t fsb; |
| 2067 | |
| 2068 | if ((error = |
| 2069 | xfs_bmapi_single(trans, dp, whichfork, &fsb, |
| 2070 | (xfs_fileoff_t)bno))) { |
| 2071 | return error; |
| 2072 | } |
| 2073 | mapp = ↦ |
| 2074 | if (fsb == NULLFSBLOCK) { |
| 2075 | nmap = 0; |
| 2076 | } else { |
| 2077 | map.br_startblock = fsb; |
| 2078 | map.br_startoff = (xfs_fileoff_t)bno; |
| 2079 | map.br_blockcount = 1; |
| 2080 | nmap = 1; |
| 2081 | } |
| 2082 | } else { |
| 2083 | mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP); |
| 2084 | nmap = nfsb; |
| 2085 | if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno, |
| 2086 | nfsb, |
| 2087 | XFS_BMAPI_METADATA | |
| 2088 | XFS_BMAPI_AFLAG(whichfork), |
| 2089 | NULL, 0, mapp, &nmap, NULL))) |
| 2090 | goto exit0; |
| 2091 | } |
| 2092 | } else { |
| 2093 | map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
| 2094 | map.br_startoff = (xfs_fileoff_t)bno; |
| 2095 | map.br_blockcount = nfsb; |
| 2096 | mapp = ↦ |
| 2097 | nmap = 1; |
| 2098 | } |
| 2099 | if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) { |
| 2100 | error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED); |
| 2101 | if (unlikely(error == EFSCORRUPTED)) { |
| 2102 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { |
| 2103 | int i; |
| 2104 | cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n", |
| 2105 | (long long)bno); |
| 2106 | cmn_err(CE_ALERT, "dir: inode %lld\n", |
| 2107 | (long long)dp->i_ino); |
| 2108 | for (i = 0; i < nmap; i++) { |
| 2109 | cmn_err(CE_ALERT, |
| 2110 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n", |
| 2111 | i, |
| 2112 | (long long)mapp[i].br_startoff, |
| 2113 | (long long)mapp[i].br_startblock, |
| 2114 | (long long)mapp[i].br_blockcount, |
| 2115 | mapp[i].br_state); |
| 2116 | } |
| 2117 | } |
| 2118 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", |
| 2119 | XFS_ERRLEVEL_LOW, mp); |
| 2120 | } |
| 2121 | goto exit0; |
| 2122 | } |
| 2123 | if (caller != 3 && nmap > 1) { |
| 2124 | bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP); |
| 2125 | nbplist = 0; |
| 2126 | } else |
| 2127 | bplist = NULL; |
| 2128 | /* |
| 2129 | * Turn the mapping(s) into buffer(s). |
| 2130 | */ |
| 2131 | for (i = 0; i < nmap; i++) { |
| 2132 | int nmapped; |
| 2133 | |
| 2134 | mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock); |
| 2135 | if (i == 0) |
| 2136 | *mappedbnop = mappedbno; |
| 2137 | nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount); |
| 2138 | switch (caller) { |
| 2139 | case 0: |
| 2140 | bp = xfs_trans_get_buf(trans, mp->m_ddev_targp, |
| 2141 | mappedbno, nmapped, 0); |
| 2142 | error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO); |
| 2143 | break; |
| 2144 | case 1: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2145 | case 2: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | bp = NULL; |
| 2147 | error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp, |
| 2148 | mappedbno, nmapped, 0, &bp); |
| 2149 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | case 3: |
| 2151 | xfs_baread(mp->m_ddev_targp, mappedbno, nmapped); |
| 2152 | error = 0; |
| 2153 | bp = NULL; |
| 2154 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2155 | } |
| 2156 | if (error) { |
| 2157 | if (bp) |
| 2158 | xfs_trans_brelse(trans, bp); |
| 2159 | goto exit1; |
| 2160 | } |
| 2161 | if (!bp) |
| 2162 | continue; |
| 2163 | if (caller == 1) { |
| 2164 | if (whichfork == XFS_ATTR_FORK) { |
| 2165 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE, |
| 2166 | XFS_ATTR_BTREE_REF); |
| 2167 | } else { |
| 2168 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE, |
| 2169 | XFS_DIR_BTREE_REF); |
| 2170 | } |
| 2171 | } |
| 2172 | if (bplist) { |
| 2173 | bplist[nbplist++] = bp; |
| 2174 | } |
| 2175 | } |
| 2176 | /* |
| 2177 | * Build a dabuf structure. |
| 2178 | */ |
| 2179 | if (bplist) { |
| 2180 | rbp = xfs_da_buf_make(nbplist, bplist, ra); |
| 2181 | } else if (bp) |
| 2182 | rbp = xfs_da_buf_make(1, &bp, ra); |
| 2183 | else |
| 2184 | rbp = NULL; |
| 2185 | /* |
| 2186 | * For read_buf, check the magic number. |
| 2187 | */ |
| 2188 | if (caller == 1) { |
| 2189 | xfs_dir2_data_t *data; |
| 2190 | xfs_dir2_free_t *free; |
| 2191 | xfs_da_blkinfo_t *info; |
| 2192 | uint magic, magic1; |
| 2193 | |
| 2194 | info = rbp->data; |
| 2195 | data = rbp->data; |
| 2196 | free = rbp->data; |
Nathan Scott | 89da054 | 2006-03-17 17:28:40 +1100 | [diff] [blame] | 2197 | magic = be16_to_cpu(info->magic); |
Nathan Scott | 70e73f5 | 2006-03-17 17:26:52 +1100 | [diff] [blame] | 2198 | magic1 = be32_to_cpu(data->hdr.magic); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | if (unlikely( |
| 2200 | XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) && |
| 2201 | (magic != XFS_DIR_LEAF_MAGIC) && |
| 2202 | (magic != XFS_ATTR_LEAF_MAGIC) && |
| 2203 | (magic != XFS_DIR2_LEAF1_MAGIC) && |
| 2204 | (magic != XFS_DIR2_LEAFN_MAGIC) && |
| 2205 | (magic1 != XFS_DIR2_BLOCK_MAGIC) && |
| 2206 | (magic1 != XFS_DIR2_DATA_MAGIC) && |
Nathan Scott | 0ba962e | 2006-03-17 17:27:07 +1100 | [diff] [blame] | 2207 | (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2208 | mp, XFS_ERRTAG_DA_READ_BUF, |
| 2209 | XFS_RANDOM_DA_READ_BUF))) { |
| 2210 | xfs_buftrace("DA READ ERROR", rbp->bps[0]); |
| 2211 | XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)", |
| 2212 | XFS_ERRLEVEL_LOW, mp, info); |
| 2213 | error = XFS_ERROR(EFSCORRUPTED); |
| 2214 | xfs_da_brelse(trans, rbp); |
| 2215 | nbplist = 0; |
| 2216 | goto exit1; |
| 2217 | } |
| 2218 | } |
| 2219 | if (bplist) { |
| 2220 | kmem_free(bplist, sizeof(*bplist) * nmap); |
| 2221 | } |
| 2222 | if (mapp != &map) { |
| 2223 | kmem_free(mapp, sizeof(*mapp) * nfsb); |
| 2224 | } |
| 2225 | if (bpp) |
| 2226 | *bpp = rbp; |
| 2227 | return 0; |
| 2228 | exit1: |
| 2229 | if (bplist) { |
| 2230 | for (i = 0; i < nbplist; i++) |
| 2231 | xfs_trans_brelse(trans, bplist[i]); |
| 2232 | kmem_free(bplist, sizeof(*bplist) * nmap); |
| 2233 | } |
| 2234 | exit0: |
| 2235 | if (mapp != &map) |
| 2236 | kmem_free(mapp, sizeof(*mapp) * nfsb); |
| 2237 | if (bpp) |
| 2238 | *bpp = NULL; |
| 2239 | return error; |
| 2240 | } |
| 2241 | |
| 2242 | /* |
| 2243 | * Get a buffer for the dir/attr block. |
| 2244 | */ |
| 2245 | int |
| 2246 | xfs_da_get_buf( |
| 2247 | xfs_trans_t *trans, |
| 2248 | xfs_inode_t *dp, |
| 2249 | xfs_dablk_t bno, |
| 2250 | xfs_daddr_t mappedbno, |
| 2251 | xfs_dabuf_t **bpp, |
| 2252 | int whichfork) |
| 2253 | { |
| 2254 | return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0, |
| 2255 | (inst_t *)__return_address); |
| 2256 | } |
| 2257 | |
| 2258 | /* |
| 2259 | * Get a buffer for the dir/attr block, fill in the contents. |
| 2260 | */ |
| 2261 | int |
| 2262 | xfs_da_read_buf( |
| 2263 | xfs_trans_t *trans, |
| 2264 | xfs_inode_t *dp, |
| 2265 | xfs_dablk_t bno, |
| 2266 | xfs_daddr_t mappedbno, |
| 2267 | xfs_dabuf_t **bpp, |
| 2268 | int whichfork) |
| 2269 | { |
| 2270 | return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1, |
| 2271 | (inst_t *)__return_address); |
| 2272 | } |
| 2273 | |
| 2274 | /* |
| 2275 | * Readahead the dir/attr block. |
| 2276 | */ |
| 2277 | xfs_daddr_t |
| 2278 | xfs_da_reada_buf( |
| 2279 | xfs_trans_t *trans, |
| 2280 | xfs_inode_t *dp, |
| 2281 | xfs_dablk_t bno, |
| 2282 | int whichfork) |
| 2283 | { |
| 2284 | xfs_daddr_t rval; |
| 2285 | |
| 2286 | rval = -1; |
| 2287 | if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3, |
| 2288 | (inst_t *)__return_address)) |
| 2289 | return -1; |
| 2290 | else |
| 2291 | return rval; |
| 2292 | } |
| 2293 | |
| 2294 | /* |
| 2295 | * Calculate the number of bits needed to hold i different values. |
| 2296 | */ |
| 2297 | uint |
| 2298 | xfs_da_log2_roundup(uint i) |
| 2299 | { |
| 2300 | uint rval; |
| 2301 | |
| 2302 | for (rval = 0; rval < NBBY * sizeof(i); rval++) { |
| 2303 | if ((1 << rval) >= i) |
| 2304 | break; |
| 2305 | } |
| 2306 | return(rval); |
| 2307 | } |
| 2308 | |
| 2309 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ |
| 2310 | kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */ |
| 2311 | |
| 2312 | /* |
| 2313 | * Allocate a dir-state structure. |
| 2314 | * We don't put them on the stack since they're large. |
| 2315 | */ |
| 2316 | xfs_da_state_t * |
| 2317 | xfs_da_state_alloc(void) |
| 2318 | { |
| 2319 | return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP); |
| 2320 | } |
| 2321 | |
| 2322 | /* |
| 2323 | * Kill the altpath contents of a da-state structure. |
| 2324 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 2325 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2326 | xfs_da_state_kill_altpath(xfs_da_state_t *state) |
| 2327 | { |
| 2328 | int i; |
| 2329 | |
| 2330 | for (i = 0; i < state->altpath.active; i++) { |
| 2331 | if (state->altpath.blk[i].bp) { |
| 2332 | if (state->altpath.blk[i].bp != state->path.blk[i].bp) |
| 2333 | xfs_da_buf_done(state->altpath.blk[i].bp); |
| 2334 | state->altpath.blk[i].bp = NULL; |
| 2335 | } |
| 2336 | } |
| 2337 | state->altpath.active = 0; |
| 2338 | } |
| 2339 | |
| 2340 | /* |
| 2341 | * Free a da-state structure. |
| 2342 | */ |
| 2343 | void |
| 2344 | xfs_da_state_free(xfs_da_state_t *state) |
| 2345 | { |
| 2346 | int i; |
| 2347 | |
| 2348 | xfs_da_state_kill_altpath(state); |
| 2349 | for (i = 0; i < state->path.active; i++) { |
| 2350 | if (state->path.blk[i].bp) |
| 2351 | xfs_da_buf_done(state->path.blk[i].bp); |
| 2352 | } |
| 2353 | if (state->extravalid && state->extrablk.bp) |
| 2354 | xfs_da_buf_done(state->extrablk.bp); |
| 2355 | #ifdef DEBUG |
| 2356 | memset((char *)state, 0, sizeof(*state)); |
| 2357 | #endif /* DEBUG */ |
| 2358 | kmem_zone_free(xfs_da_state_zone, state); |
| 2359 | } |
| 2360 | |
| 2361 | #ifdef XFS_DABUF_DEBUG |
| 2362 | xfs_dabuf_t *xfs_dabuf_global_list; |
| 2363 | lock_t xfs_dabuf_global_lock; |
| 2364 | #endif |
| 2365 | |
| 2366 | /* |
| 2367 | * Create a dabuf. |
| 2368 | */ |
| 2369 | /* ARGSUSED */ |
| 2370 | STATIC xfs_dabuf_t * |
| 2371 | xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra) |
| 2372 | { |
| 2373 | xfs_buf_t *bp; |
| 2374 | xfs_dabuf_t *dabuf; |
| 2375 | int i; |
| 2376 | int off; |
| 2377 | |
| 2378 | if (nbuf == 1) |
| 2379 | dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP); |
| 2380 | else |
| 2381 | dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP); |
| 2382 | dabuf->dirty = 0; |
| 2383 | #ifdef XFS_DABUF_DEBUG |
| 2384 | dabuf->ra = ra; |
| 2385 | dabuf->target = XFS_BUF_TARGET(bps[0]); |
| 2386 | dabuf->blkno = XFS_BUF_ADDR(bps[0]); |
| 2387 | #endif |
| 2388 | if (nbuf == 1) { |
| 2389 | dabuf->nbuf = 1; |
| 2390 | bp = bps[0]; |
| 2391 | dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp)); |
| 2392 | dabuf->data = XFS_BUF_PTR(bp); |
| 2393 | dabuf->bps[0] = bp; |
| 2394 | } else { |
| 2395 | dabuf->nbuf = nbuf; |
| 2396 | for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) { |
| 2397 | dabuf->bps[i] = bp = bps[i]; |
| 2398 | dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp)); |
| 2399 | } |
| 2400 | dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP); |
| 2401 | for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) { |
| 2402 | bp = bps[i]; |
| 2403 | memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp), |
| 2404 | XFS_BUF_COUNT(bp)); |
| 2405 | } |
| 2406 | } |
| 2407 | #ifdef XFS_DABUF_DEBUG |
| 2408 | { |
| 2409 | SPLDECL(s); |
| 2410 | xfs_dabuf_t *p; |
| 2411 | |
| 2412 | s = mutex_spinlock(&xfs_dabuf_global_lock); |
| 2413 | for (p = xfs_dabuf_global_list; p; p = p->next) { |
| 2414 | ASSERT(p->blkno != dabuf->blkno || |
| 2415 | p->target != dabuf->target); |
| 2416 | } |
| 2417 | dabuf->prev = NULL; |
| 2418 | if (xfs_dabuf_global_list) |
| 2419 | xfs_dabuf_global_list->prev = dabuf; |
| 2420 | dabuf->next = xfs_dabuf_global_list; |
| 2421 | xfs_dabuf_global_list = dabuf; |
| 2422 | mutex_spinunlock(&xfs_dabuf_global_lock, s); |
| 2423 | } |
| 2424 | #endif |
| 2425 | return dabuf; |
| 2426 | } |
| 2427 | |
| 2428 | /* |
| 2429 | * Un-dirty a dabuf. |
| 2430 | */ |
| 2431 | STATIC void |
| 2432 | xfs_da_buf_clean(xfs_dabuf_t *dabuf) |
| 2433 | { |
| 2434 | xfs_buf_t *bp; |
| 2435 | int i; |
| 2436 | int off; |
| 2437 | |
| 2438 | if (dabuf->dirty) { |
| 2439 | ASSERT(dabuf->nbuf > 1); |
| 2440 | dabuf->dirty = 0; |
| 2441 | for (i = off = 0; i < dabuf->nbuf; |
| 2442 | i++, off += XFS_BUF_COUNT(bp)) { |
| 2443 | bp = dabuf->bps[i]; |
| 2444 | memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off, |
| 2445 | XFS_BUF_COUNT(bp)); |
| 2446 | } |
| 2447 | } |
| 2448 | } |
| 2449 | |
| 2450 | /* |
| 2451 | * Release a dabuf. |
| 2452 | */ |
| 2453 | void |
| 2454 | xfs_da_buf_done(xfs_dabuf_t *dabuf) |
| 2455 | { |
| 2456 | ASSERT(dabuf); |
| 2457 | ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]); |
| 2458 | if (dabuf->dirty) |
| 2459 | xfs_da_buf_clean(dabuf); |
| 2460 | if (dabuf->nbuf > 1) |
| 2461 | kmem_free(dabuf->data, BBTOB(dabuf->bbcount)); |
| 2462 | #ifdef XFS_DABUF_DEBUG |
| 2463 | { |
| 2464 | SPLDECL(s); |
| 2465 | |
| 2466 | s = mutex_spinlock(&xfs_dabuf_global_lock); |
| 2467 | if (dabuf->prev) |
| 2468 | dabuf->prev->next = dabuf->next; |
| 2469 | else |
| 2470 | xfs_dabuf_global_list = dabuf->next; |
| 2471 | if (dabuf->next) |
| 2472 | dabuf->next->prev = dabuf->prev; |
| 2473 | mutex_spinunlock(&xfs_dabuf_global_lock, s); |
| 2474 | } |
| 2475 | memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf)); |
| 2476 | #endif |
| 2477 | if (dabuf->nbuf == 1) |
| 2478 | kmem_zone_free(xfs_dabuf_zone, dabuf); |
| 2479 | else |
| 2480 | kmem_free(dabuf, XFS_DA_BUF_SIZE(dabuf->nbuf)); |
| 2481 | } |
| 2482 | |
| 2483 | /* |
| 2484 | * Log transaction from a dabuf. |
| 2485 | */ |
| 2486 | void |
| 2487 | xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last) |
| 2488 | { |
| 2489 | xfs_buf_t *bp; |
| 2490 | uint f; |
| 2491 | int i; |
| 2492 | uint l; |
| 2493 | int off; |
| 2494 | |
| 2495 | ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]); |
| 2496 | if (dabuf->nbuf == 1) { |
| 2497 | ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0])); |
| 2498 | xfs_trans_log_buf(tp, dabuf->bps[0], first, last); |
| 2499 | return; |
| 2500 | } |
| 2501 | dabuf->dirty = 1; |
| 2502 | ASSERT(first <= last); |
| 2503 | for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) { |
| 2504 | bp = dabuf->bps[i]; |
| 2505 | f = off; |
| 2506 | l = f + XFS_BUF_COUNT(bp) - 1; |
| 2507 | if (f < first) |
| 2508 | f = first; |
| 2509 | if (l > last) |
| 2510 | l = last; |
| 2511 | if (f <= l) |
| 2512 | xfs_trans_log_buf(tp, bp, f - off, l - off); |
| 2513 | /* |
| 2514 | * B_DONE is set by xfs_trans_log buf. |
| 2515 | * If we don't set it on a new buffer (get not read) |
| 2516 | * then if we don't put anything in the buffer it won't |
| 2517 | * be set, and at commit it it released into the cache, |
| 2518 | * and then a read will fail. |
| 2519 | */ |
| 2520 | else if (!(XFS_BUF_ISDONE(bp))) |
| 2521 | XFS_BUF_DONE(bp); |
| 2522 | } |
| 2523 | ASSERT(last < off); |
| 2524 | } |
| 2525 | |
| 2526 | /* |
| 2527 | * Release dabuf from a transaction. |
| 2528 | * Have to free up the dabuf before the buffers are released, |
| 2529 | * since the synchronization on the dabuf is really the lock on the buffer. |
| 2530 | */ |
| 2531 | void |
| 2532 | xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf) |
| 2533 | { |
| 2534 | xfs_buf_t *bp; |
| 2535 | xfs_buf_t **bplist; |
| 2536 | int i; |
| 2537 | int nbuf; |
| 2538 | |
| 2539 | ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]); |
| 2540 | if ((nbuf = dabuf->nbuf) == 1) { |
| 2541 | bplist = &bp; |
| 2542 | bp = dabuf->bps[0]; |
| 2543 | } else { |
| 2544 | bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP); |
| 2545 | memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist)); |
| 2546 | } |
| 2547 | xfs_da_buf_done(dabuf); |
| 2548 | for (i = 0; i < nbuf; i++) |
| 2549 | xfs_trans_brelse(tp, bplist[i]); |
| 2550 | if (bplist != &bp) |
| 2551 | kmem_free(bplist, nbuf * sizeof(*bplist)); |
| 2552 | } |
| 2553 | |
| 2554 | /* |
| 2555 | * Invalidate dabuf from a transaction. |
| 2556 | */ |
| 2557 | void |
| 2558 | xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf) |
| 2559 | { |
| 2560 | xfs_buf_t *bp; |
| 2561 | xfs_buf_t **bplist; |
| 2562 | int i; |
| 2563 | int nbuf; |
| 2564 | |
| 2565 | ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]); |
| 2566 | if ((nbuf = dabuf->nbuf) == 1) { |
| 2567 | bplist = &bp; |
| 2568 | bp = dabuf->bps[0]; |
| 2569 | } else { |
| 2570 | bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP); |
| 2571 | memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist)); |
| 2572 | } |
| 2573 | xfs_da_buf_done(dabuf); |
| 2574 | for (i = 0; i < nbuf; i++) |
| 2575 | xfs_trans_binval(tp, bplist[i]); |
| 2576 | if (bplist != &bp) |
| 2577 | kmem_free(bplist, nbuf * sizeof(*bplist)); |
| 2578 | } |
| 2579 | |
| 2580 | /* |
| 2581 | * Get the first daddr from a dabuf. |
| 2582 | */ |
| 2583 | xfs_daddr_t |
| 2584 | xfs_da_blkno(xfs_dabuf_t *dabuf) |
| 2585 | { |
| 2586 | ASSERT(dabuf->nbuf); |
| 2587 | ASSERT(dabuf->data); |
| 2588 | return XFS_BUF_ADDR(dabuf->bps[0]); |
| 2589 | } |