1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * Copyright (c) 2013 Red Hat, Inc. 5 * All Rights Reserved. 6 */ 7 #include "xfs.h" 8 #include "xfs_fs.h" 9 #include "xfs_shared.h" 10 #include "xfs_format.h" 11 #include "xfs_log_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_bit.h" 14 #include "xfs_mount.h" 15 #include "xfs_da_format.h" 16 #include "xfs_da_btree.h" 17 #include "xfs_inode.h" 18 #include "xfs_alloc.h" 19 #include "xfs_attr_remote.h" 20 #include "xfs_trans.h" 21 #include "xfs_inode_item.h" 22 #include "xfs_bmap.h" 23 #include "xfs_attr.h" 24 #include "xfs_attr_leaf.h" 25 #include "xfs_error.h" 26 #include "xfs_quota.h" 27 #include "xfs_trace.h" 28 #include "xfs_dir2.h" 29 30 /* 31 * Look at all the extents for this logical region, 32 * invalidate any buffers that are incore/in transactions. 33 */ 34 STATIC int 35 xfs_attr3_leaf_freextent( 36 struct xfs_trans **trans, 37 struct xfs_inode *dp, 38 xfs_dablk_t blkno, 39 int blkcnt) 40 { 41 struct xfs_bmbt_irec map; 42 struct xfs_buf *bp; 43 xfs_dablk_t tblkno; 44 xfs_daddr_t dblkno; 45 int tblkcnt; 46 int dblkcnt; 47 int nmap; 48 int error; 49 50 /* 51 * Roll through the "value", invalidating the attribute value's 52 * blocks. 53 */ 54 tblkno = blkno; 55 tblkcnt = blkcnt; 56 while (tblkcnt > 0) { 57 /* 58 * Try to remember where we decided to put the value. 59 */ 60 nmap = 1; 61 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt, 62 &map, &nmap, XFS_BMAPI_ATTRFORK); 63 if (error) { 64 return error; 65 } 66 ASSERT(nmap == 1); 67 ASSERT(map.br_startblock != DELAYSTARTBLOCK); 68 69 /* 70 * If it's a hole, these are already unmapped 71 * so there's nothing to invalidate. 72 */ 73 if (map.br_startblock != HOLESTARTBLOCK) { 74 75 dblkno = XFS_FSB_TO_DADDR(dp->i_mount, 76 map.br_startblock); 77 dblkcnt = XFS_FSB_TO_BB(dp->i_mount, 78 map.br_blockcount); 79 bp = xfs_trans_get_buf(*trans, 80 dp->i_mount->m_ddev_targp, 81 dblkno, dblkcnt, 0); 82 if (!bp) 83 return -ENOMEM; 84 xfs_trans_binval(*trans, bp); 85 /* 86 * Roll to next transaction. 87 */ 88 error = xfs_trans_roll_inode(trans, dp); 89 if (error) 90 return error; 91 } 92 93 tblkno += map.br_blockcount; 94 tblkcnt -= map.br_blockcount; 95 } 96 97 return 0; 98 } 99 100 /* 101 * Invalidate all of the "remote" value regions pointed to by a particular 102 * leaf block. 103 * Note that we must release the lock on the buffer so that we are not 104 * caught holding something that the logging code wants to flush to disk. 105 */ 106 STATIC int 107 xfs_attr3_leaf_inactive( 108 struct xfs_trans **trans, 109 struct xfs_inode *dp, 110 struct xfs_buf *bp) 111 { 112 struct xfs_attr_leafblock *leaf; 113 struct xfs_attr3_icleaf_hdr ichdr; 114 struct xfs_attr_leaf_entry *entry; 115 struct xfs_attr_leaf_name_remote *name_rmt; 116 struct xfs_attr_inactive_list *list; 117 struct xfs_attr_inactive_list *lp; 118 int error; 119 int count; 120 int size; 121 int tmp; 122 int i; 123 struct xfs_mount *mp = bp->b_target->bt_mount; 124 125 leaf = bp->b_addr; 126 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); 127 128 /* 129 * Count the number of "remote" value extents. 130 */ 131 count = 0; 132 entry = xfs_attr3_leaf_entryp(leaf); 133 for (i = 0; i < ichdr.count; entry++, i++) { 134 if (be16_to_cpu(entry->nameidx) && 135 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 136 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 137 if (name_rmt->valueblk) 138 count++; 139 } 140 } 141 142 /* 143 * If there are no "remote" values, we're done. 144 */ 145 if (count == 0) { 146 xfs_trans_brelse(*trans, bp); 147 return 0; 148 } 149 150 /* 151 * Allocate storage for a list of all the "remote" value extents. 152 */ 153 size = count * sizeof(xfs_attr_inactive_list_t); 154 list = kmem_alloc(size, KM_SLEEP); 155 156 /* 157 * Identify each of the "remote" value extents. 158 */ 159 lp = list; 160 entry = xfs_attr3_leaf_entryp(leaf); 161 for (i = 0; i < ichdr.count; entry++, i++) { 162 if (be16_to_cpu(entry->nameidx) && 163 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 164 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 165 if (name_rmt->valueblk) { 166 lp->valueblk = be32_to_cpu(name_rmt->valueblk); 167 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount, 168 be32_to_cpu(name_rmt->valuelen)); 169 lp++; 170 } 171 } 172 } 173 xfs_trans_brelse(*trans, bp); /* unlock for trans. in freextent() */ 174 175 /* 176 * Invalidate each of the "remote" value extents. 177 */ 178 error = 0; 179 for (lp = list, i = 0; i < count; i++, lp++) { 180 tmp = xfs_attr3_leaf_freextent(trans, dp, 181 lp->valueblk, lp->valuelen); 182 183 if (error == 0) 184 error = tmp; /* save only the 1st errno */ 185 } 186 187 kmem_free(list); 188 return error; 189 } 190 191 /* 192 * Recurse (gasp!) through the attribute nodes until we find leaves. 193 * We're doing a depth-first traversal in order to invalidate everything. 194 */ 195 STATIC int 196 xfs_attr3_node_inactive( 197 struct xfs_trans **trans, 198 struct xfs_inode *dp, 199 struct xfs_buf *bp, 200 int level) 201 { 202 xfs_da_blkinfo_t *info; 203 xfs_da_intnode_t *node; 204 xfs_dablk_t child_fsb; 205 xfs_daddr_t parent_blkno, child_blkno; 206 int error, i; 207 struct xfs_buf *child_bp; 208 struct xfs_da_node_entry *btree; 209 struct xfs_da3_icnode_hdr ichdr; 210 211 /* 212 * Since this code is recursive (gasp!) we must protect ourselves. 213 */ 214 if (level > XFS_DA_NODE_MAXDEPTH) { 215 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 216 return -EIO; 217 } 218 219 node = bp->b_addr; 220 dp->d_ops->node_hdr_from_disk(&ichdr, node); 221 parent_blkno = bp->b_bn; 222 if (!ichdr.count) { 223 xfs_trans_brelse(*trans, bp); 224 return 0; 225 } 226 btree = dp->d_ops->node_tree_p(node); 227 child_fsb = be32_to_cpu(btree[0].before); 228 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 229 230 /* 231 * If this is the node level just above the leaves, simply loop 232 * over the leaves removing all of them. If this is higher up 233 * in the tree, recurse downward. 234 */ 235 for (i = 0; i < ichdr.count; i++) { 236 /* 237 * Read the subsidiary block to see what we have to work with. 238 * Don't do this in a transaction. This is a depth-first 239 * traversal of the tree so we may deal with many blocks 240 * before we come back to this one. 241 */ 242 error = xfs_da3_node_read(*trans, dp, child_fsb, -1, &child_bp, 243 XFS_ATTR_FORK); 244 if (error) 245 return error; 246 247 /* save for re-read later */ 248 child_blkno = XFS_BUF_ADDR(child_bp); 249 250 /* 251 * Invalidate the subtree, however we have to. 252 */ 253 info = child_bp->b_addr; 254 switch (info->magic) { 255 case cpu_to_be16(XFS_DA_NODE_MAGIC): 256 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 257 error = xfs_attr3_node_inactive(trans, dp, child_bp, 258 level + 1); 259 break; 260 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 261 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 262 error = xfs_attr3_leaf_inactive(trans, dp, child_bp); 263 break; 264 default: 265 error = -EIO; 266 xfs_trans_brelse(*trans, child_bp); 267 break; 268 } 269 if (error) 270 return error; 271 272 /* 273 * Remove the subsidiary block from the cache and from the log. 274 */ 275 error = xfs_da_get_buf(*trans, dp, 0, child_blkno, &child_bp, 276 XFS_ATTR_FORK); 277 if (error) 278 return error; 279 xfs_trans_binval(*trans, child_bp); 280 281 /* 282 * If we're not done, re-read the parent to get the next 283 * child block number. 284 */ 285 if (i + 1 < ichdr.count) { 286 error = xfs_da3_node_read(*trans, dp, 0, parent_blkno, 287 &bp, XFS_ATTR_FORK); 288 if (error) 289 return error; 290 node = bp->b_addr; 291 btree = dp->d_ops->node_tree_p(node); 292 child_fsb = be32_to_cpu(btree[i + 1].before); 293 xfs_trans_brelse(*trans, bp); 294 } 295 /* 296 * Atomically commit the whole invalidate stuff. 297 */ 298 error = xfs_trans_roll_inode(trans, dp); 299 if (error) 300 return error; 301 } 302 303 return 0; 304 } 305 306 /* 307 * Indiscriminately delete the entire attribute fork 308 * 309 * Recurse (gasp!) through the attribute nodes until we find leaves. 310 * We're doing a depth-first traversal in order to invalidate everything. 311 */ 312 static int 313 xfs_attr3_root_inactive( 314 struct xfs_trans **trans, 315 struct xfs_inode *dp) 316 { 317 struct xfs_da_blkinfo *info; 318 struct xfs_buf *bp; 319 xfs_daddr_t blkno; 320 int error; 321 322 /* 323 * Read block 0 to see what we have to work with. 324 * We only get here if we have extents, since we remove 325 * the extents in reverse order the extent containing 326 * block 0 must still be there. 327 */ 328 error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK); 329 if (error) 330 return error; 331 blkno = bp->b_bn; 332 333 /* 334 * Invalidate the tree, even if the "tree" is only a single leaf block. 335 * This is a depth-first traversal! 336 */ 337 info = bp->b_addr; 338 switch (info->magic) { 339 case cpu_to_be16(XFS_DA_NODE_MAGIC): 340 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 341 error = xfs_attr3_node_inactive(trans, dp, bp, 1); 342 break; 343 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 344 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 345 error = xfs_attr3_leaf_inactive(trans, dp, bp); 346 break; 347 default: 348 error = -EIO; 349 xfs_trans_brelse(*trans, bp); 350 break; 351 } 352 if (error) 353 return error; 354 355 /* 356 * Invalidate the incore copy of the root block. 357 */ 358 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK); 359 if (error) 360 return error; 361 xfs_trans_binval(*trans, bp); /* remove from cache */ 362 /* 363 * Commit the invalidate and start the next transaction. 364 */ 365 error = xfs_trans_roll_inode(trans, dp); 366 367 return error; 368 } 369 370 /* 371 * xfs_attr_inactive kills all traces of an attribute fork on an inode. It 372 * removes both the on-disk and in-memory inode fork. Note that this also has to 373 * handle the condition of inodes without attributes but with an attribute fork 374 * configured, so we can't use xfs_inode_hasattr() here. 375 * 376 * The in-memory attribute fork is removed even on error. 377 */ 378 int 379 xfs_attr_inactive( 380 struct xfs_inode *dp) 381 { 382 struct xfs_trans *trans; 383 struct xfs_mount *mp; 384 int lock_mode = XFS_ILOCK_SHARED; 385 int error = 0; 386 387 mp = dp->i_mount; 388 ASSERT(! XFS_NOT_DQATTACHED(mp, dp)); 389 390 xfs_ilock(dp, lock_mode); 391 if (!XFS_IFORK_Q(dp)) 392 goto out_destroy_fork; 393 xfs_iunlock(dp, lock_mode); 394 395 lock_mode = 0; 396 397 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_attrinval, 0, 0, 0, &trans); 398 if (error) 399 goto out_destroy_fork; 400 401 lock_mode = XFS_ILOCK_EXCL; 402 xfs_ilock(dp, lock_mode); 403 404 if (!XFS_IFORK_Q(dp)) 405 goto out_cancel; 406 407 /* 408 * No need to make quota reservations here. We expect to release some 409 * blocks, not allocate, in the common case. 410 */ 411 xfs_trans_ijoin(trans, dp, 0); 412 413 /* 414 * Invalidate and truncate the attribute fork extents. Make sure the 415 * fork actually has attributes as otherwise the invalidation has no 416 * blocks to read and returns an error. In this case, just do the fork 417 * removal below. 418 */ 419 if (xfs_inode_hasattr(dp) && 420 dp->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) { 421 error = xfs_attr3_root_inactive(&trans, dp); 422 if (error) 423 goto out_cancel; 424 425 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0); 426 if (error) 427 goto out_cancel; 428 } 429 430 /* Reset the attribute fork - this also destroys the in-core fork */ 431 xfs_attr_fork_remove(dp, trans); 432 433 error = xfs_trans_commit(trans); 434 xfs_iunlock(dp, lock_mode); 435 return error; 436 437 out_cancel: 438 xfs_trans_cancel(trans); 439 out_destroy_fork: 440 /* kill the in-core attr fork before we drop the inode lock */ 441 if (dp->i_afp) 442 xfs_idestroy_fork(dp, XFS_ATTR_FORK); 443 if (lock_mode) 444 xfs_iunlock(dp, lock_mode); 445 return error; 446 } 447