1 /* 2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 3 * Copyright (c) 2013 Red Hat, Inc. 4 * All Rights Reserved. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License as 8 * published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it would be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 18 */ 19 #include "xfs.h" 20 #include "xfs_fs.h" 21 #include "xfs_shared.h" 22 #include "xfs_format.h" 23 #include "xfs_log_format.h" 24 #include "xfs_trans_resv.h" 25 #include "xfs_bit.h" 26 #include "xfs_sb.h" 27 #include "xfs_ag.h" 28 #include "xfs_mount.h" 29 #include "xfs_da_format.h" 30 #include "xfs_da_btree.h" 31 #include "xfs_inode.h" 32 #include "xfs_alloc.h" 33 #include "xfs_attr_remote.h" 34 #include "xfs_trans.h" 35 #include "xfs_inode_item.h" 36 #include "xfs_bmap.h" 37 #include "xfs_attr.h" 38 #include "xfs_attr_leaf.h" 39 #include "xfs_error.h" 40 #include "xfs_quota.h" 41 #include "xfs_trace.h" 42 #include "xfs_dinode.h" 43 #include "xfs_dir2.h" 44 45 /* 46 * Look at all the extents for this logical region, 47 * invalidate any buffers that are incore/in transactions. 48 */ 49 STATIC int 50 xfs_attr3_leaf_freextent( 51 struct xfs_trans **trans, 52 struct xfs_inode *dp, 53 xfs_dablk_t blkno, 54 int blkcnt) 55 { 56 struct xfs_bmbt_irec map; 57 struct xfs_buf *bp; 58 xfs_dablk_t tblkno; 59 xfs_daddr_t dblkno; 60 int tblkcnt; 61 int dblkcnt; 62 int nmap; 63 int error; 64 65 /* 66 * Roll through the "value", invalidating the attribute value's 67 * blocks. 68 */ 69 tblkno = blkno; 70 tblkcnt = blkcnt; 71 while (tblkcnt > 0) { 72 /* 73 * Try to remember where we decided to put the value. 74 */ 75 nmap = 1; 76 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt, 77 &map, &nmap, XFS_BMAPI_ATTRFORK); 78 if (error) { 79 return(error); 80 } 81 ASSERT(nmap == 1); 82 ASSERT(map.br_startblock != DELAYSTARTBLOCK); 83 84 /* 85 * If it's a hole, these are already unmapped 86 * so there's nothing to invalidate. 87 */ 88 if (map.br_startblock != HOLESTARTBLOCK) { 89 90 dblkno = XFS_FSB_TO_DADDR(dp->i_mount, 91 map.br_startblock); 92 dblkcnt = XFS_FSB_TO_BB(dp->i_mount, 93 map.br_blockcount); 94 bp = xfs_trans_get_buf(*trans, 95 dp->i_mount->m_ddev_targp, 96 dblkno, dblkcnt, 0); 97 if (!bp) 98 return ENOMEM; 99 xfs_trans_binval(*trans, bp); 100 /* 101 * Roll to next transaction. 102 */ 103 error = xfs_trans_roll(trans, dp); 104 if (error) 105 return (error); 106 } 107 108 tblkno += map.br_blockcount; 109 tblkcnt -= map.br_blockcount; 110 } 111 112 return(0); 113 } 114 115 /* 116 * Invalidate all of the "remote" value regions pointed to by a particular 117 * leaf block. 118 * Note that we must release the lock on the buffer so that we are not 119 * caught holding something that the logging code wants to flush to disk. 120 */ 121 STATIC int 122 xfs_attr3_leaf_inactive( 123 struct xfs_trans **trans, 124 struct xfs_inode *dp, 125 struct xfs_buf *bp) 126 { 127 struct xfs_attr_leafblock *leaf; 128 struct xfs_attr3_icleaf_hdr ichdr; 129 struct xfs_attr_leaf_entry *entry; 130 struct xfs_attr_leaf_name_remote *name_rmt; 131 struct xfs_attr_inactive_list *list; 132 struct xfs_attr_inactive_list *lp; 133 int error; 134 int count; 135 int size; 136 int tmp; 137 int i; 138 139 leaf = bp->b_addr; 140 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf); 141 142 /* 143 * Count the number of "remote" value extents. 144 */ 145 count = 0; 146 entry = xfs_attr3_leaf_entryp(leaf); 147 for (i = 0; i < ichdr.count; entry++, i++) { 148 if (be16_to_cpu(entry->nameidx) && 149 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 150 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 151 if (name_rmt->valueblk) 152 count++; 153 } 154 } 155 156 /* 157 * If there are no "remote" values, we're done. 158 */ 159 if (count == 0) { 160 xfs_trans_brelse(*trans, bp); 161 return 0; 162 } 163 164 /* 165 * Allocate storage for a list of all the "remote" value extents. 166 */ 167 size = count * sizeof(xfs_attr_inactive_list_t); 168 list = kmem_alloc(size, KM_SLEEP); 169 170 /* 171 * Identify each of the "remote" value extents. 172 */ 173 lp = list; 174 entry = xfs_attr3_leaf_entryp(leaf); 175 for (i = 0; i < ichdr.count; entry++, i++) { 176 if (be16_to_cpu(entry->nameidx) && 177 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 178 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 179 if (name_rmt->valueblk) { 180 lp->valueblk = be32_to_cpu(name_rmt->valueblk); 181 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount, 182 be32_to_cpu(name_rmt->valuelen)); 183 lp++; 184 } 185 } 186 } 187 xfs_trans_brelse(*trans, bp); /* unlock for trans. in freextent() */ 188 189 /* 190 * Invalidate each of the "remote" value extents. 191 */ 192 error = 0; 193 for (lp = list, i = 0; i < count; i++, lp++) { 194 tmp = xfs_attr3_leaf_freextent(trans, dp, 195 lp->valueblk, lp->valuelen); 196 197 if (error == 0) 198 error = tmp; /* save only the 1st errno */ 199 } 200 201 kmem_free(list); 202 return error; 203 } 204 205 /* 206 * Recurse (gasp!) through the attribute nodes until we find leaves. 207 * We're doing a depth-first traversal in order to invalidate everything. 208 */ 209 STATIC int 210 xfs_attr3_node_inactive( 211 struct xfs_trans **trans, 212 struct xfs_inode *dp, 213 struct xfs_buf *bp, 214 int level) 215 { 216 xfs_da_blkinfo_t *info; 217 xfs_da_intnode_t *node; 218 xfs_dablk_t child_fsb; 219 xfs_daddr_t parent_blkno, child_blkno; 220 int error, i; 221 struct xfs_buf *child_bp; 222 struct xfs_da_node_entry *btree; 223 struct xfs_da3_icnode_hdr ichdr; 224 225 /* 226 * Since this code is recursive (gasp!) we must protect ourselves. 227 */ 228 if (level > XFS_DA_NODE_MAXDEPTH) { 229 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 230 return XFS_ERROR(EIO); 231 } 232 233 node = bp->b_addr; 234 dp->d_ops->node_hdr_from_disk(&ichdr, node); 235 parent_blkno = bp->b_bn; 236 if (!ichdr.count) { 237 xfs_trans_brelse(*trans, bp); 238 return 0; 239 } 240 btree = dp->d_ops->node_tree_p(node); 241 child_fsb = be32_to_cpu(btree[0].before); 242 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 243 244 /* 245 * If this is the node level just above the leaves, simply loop 246 * over the leaves removing all of them. If this is higher up 247 * in the tree, recurse downward. 248 */ 249 for (i = 0; i < ichdr.count; i++) { 250 /* 251 * Read the subsidiary block to see what we have to work with. 252 * Don't do this in a transaction. This is a depth-first 253 * traversal of the tree so we may deal with many blocks 254 * before we come back to this one. 255 */ 256 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp, 257 XFS_ATTR_FORK); 258 if (error) 259 return(error); 260 if (child_bp) { 261 /* save for re-read later */ 262 child_blkno = XFS_BUF_ADDR(child_bp); 263 264 /* 265 * Invalidate the subtree, however we have to. 266 */ 267 info = child_bp->b_addr; 268 switch (info->magic) { 269 case cpu_to_be16(XFS_DA_NODE_MAGIC): 270 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 271 error = xfs_attr3_node_inactive(trans, dp, 272 child_bp, level + 1); 273 break; 274 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 275 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 276 error = xfs_attr3_leaf_inactive(trans, dp, 277 child_bp); 278 break; 279 default: 280 error = XFS_ERROR(EIO); 281 xfs_trans_brelse(*trans, child_bp); 282 break; 283 } 284 if (error) 285 return error; 286 287 /* 288 * Remove the subsidiary block from the cache 289 * and from the log. 290 */ 291 error = xfs_da_get_buf(*trans, dp, 0, child_blkno, 292 &child_bp, XFS_ATTR_FORK); 293 if (error) 294 return error; 295 xfs_trans_binval(*trans, child_bp); 296 } 297 298 /* 299 * If we're not done, re-read the parent to get the next 300 * child block number. 301 */ 302 if (i + 1 < ichdr.count) { 303 error = xfs_da3_node_read(*trans, dp, 0, parent_blkno, 304 &bp, XFS_ATTR_FORK); 305 if (error) 306 return error; 307 child_fsb = be32_to_cpu(btree[i + 1].before); 308 xfs_trans_brelse(*trans, bp); 309 } 310 /* 311 * Atomically commit the whole invalidate stuff. 312 */ 313 error = xfs_trans_roll(trans, dp); 314 if (error) 315 return error; 316 } 317 318 return 0; 319 } 320 321 /* 322 * Indiscriminately delete the entire attribute fork 323 * 324 * Recurse (gasp!) through the attribute nodes until we find leaves. 325 * We're doing a depth-first traversal in order to invalidate everything. 326 */ 327 int 328 xfs_attr3_root_inactive( 329 struct xfs_trans **trans, 330 struct xfs_inode *dp) 331 { 332 struct xfs_da_blkinfo *info; 333 struct xfs_buf *bp; 334 xfs_daddr_t blkno; 335 int error; 336 337 /* 338 * Read block 0 to see what we have to work with. 339 * We only get here if we have extents, since we remove 340 * the extents in reverse order the extent containing 341 * block 0 must still be there. 342 */ 343 error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK); 344 if (error) 345 return error; 346 blkno = bp->b_bn; 347 348 /* 349 * Invalidate the tree, even if the "tree" is only a single leaf block. 350 * This is a depth-first traversal! 351 */ 352 info = bp->b_addr; 353 switch (info->magic) { 354 case cpu_to_be16(XFS_DA_NODE_MAGIC): 355 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 356 error = xfs_attr3_node_inactive(trans, dp, bp, 1); 357 break; 358 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 359 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 360 error = xfs_attr3_leaf_inactive(trans, dp, bp); 361 break; 362 default: 363 error = XFS_ERROR(EIO); 364 xfs_trans_brelse(*trans, bp); 365 break; 366 } 367 if (error) 368 return error; 369 370 /* 371 * Invalidate the incore copy of the root block. 372 */ 373 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK); 374 if (error) 375 return error; 376 xfs_trans_binval(*trans, bp); /* remove from cache */ 377 /* 378 * Commit the invalidate and start the next transaction. 379 */ 380 error = xfs_trans_roll(trans, dp); 381 382 return error; 383 } 384 385 int 386 xfs_attr_inactive(xfs_inode_t *dp) 387 { 388 xfs_trans_t *trans; 389 xfs_mount_t *mp; 390 int error; 391 392 mp = dp->i_mount; 393 ASSERT(! XFS_NOT_DQATTACHED(mp, dp)); 394 395 xfs_ilock(dp, XFS_ILOCK_SHARED); 396 if (!xfs_inode_hasattr(dp) || 397 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) { 398 xfs_iunlock(dp, XFS_ILOCK_SHARED); 399 return 0; 400 } 401 xfs_iunlock(dp, XFS_ILOCK_SHARED); 402 403 /* 404 * Start our first transaction of the day. 405 * 406 * All future transactions during this code must be "chained" off 407 * this one via the trans_dup() call. All transactions will contain 408 * the inode, and the inode will always be marked with trans_ihold(). 409 * Since the inode will be locked in all transactions, we must log 410 * the inode in every transaction to let it float upward through 411 * the log. 412 */ 413 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL); 414 error = xfs_trans_reserve(trans, &M_RES(mp)->tr_attrinval, 0, 0); 415 if (error) { 416 xfs_trans_cancel(trans, 0); 417 return(error); 418 } 419 xfs_ilock(dp, XFS_ILOCK_EXCL); 420 421 /* 422 * No need to make quota reservations here. We expect to release some 423 * blocks, not allocate, in the common case. 424 */ 425 xfs_trans_ijoin(trans, dp, 0); 426 427 /* 428 * Decide on what work routines to call based on the inode size. 429 */ 430 if (!xfs_inode_hasattr(dp) || 431 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) { 432 error = 0; 433 goto out; 434 } 435 error = xfs_attr3_root_inactive(&trans, dp); 436 if (error) 437 goto out; 438 439 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0); 440 if (error) 441 goto out; 442 443 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES); 444 xfs_iunlock(dp, XFS_ILOCK_EXCL); 445 446 return(error); 447 448 out: 449 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); 450 xfs_iunlock(dp, XFS_ILOCK_EXCL); 451 return(error); 452 } 453