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 #include "xfs_defer.h" 30 31 /* 32 * Look at all the extents for this logical region, 33 * invalidate any buffers that are incore/in transactions. 34 */ 35 STATIC int 36 xfs_attr3_leaf_freextent( 37 struct xfs_trans **trans, 38 struct xfs_inode *dp, 39 xfs_dablk_t blkno, 40 int blkcnt) 41 { 42 struct xfs_bmbt_irec map; 43 struct xfs_buf *bp; 44 xfs_dablk_t tblkno; 45 xfs_daddr_t dblkno; 46 int tblkcnt; 47 int dblkcnt; 48 int nmap; 49 int error; 50 51 /* 52 * Roll through the "value", invalidating the attribute value's 53 * blocks. 54 */ 55 tblkno = blkno; 56 tblkcnt = blkcnt; 57 while (tblkcnt > 0) { 58 /* 59 * Try to remember where we decided to put the value. 60 */ 61 nmap = 1; 62 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt, 63 &map, &nmap, XFS_BMAPI_ATTRFORK); 64 if (error) { 65 return error; 66 } 67 ASSERT(nmap == 1); 68 ASSERT(map.br_startblock != DELAYSTARTBLOCK); 69 70 /* 71 * If it's a hole, these are already unmapped 72 * so there's nothing to invalidate. 73 */ 74 if (map.br_startblock != HOLESTARTBLOCK) { 75 76 dblkno = XFS_FSB_TO_DADDR(dp->i_mount, 77 map.br_startblock); 78 dblkcnt = XFS_FSB_TO_BB(dp->i_mount, 79 map.br_blockcount); 80 bp = xfs_trans_get_buf(*trans, 81 dp->i_mount->m_ddev_targp, 82 dblkno, dblkcnt, 0); 83 if (!bp) 84 return -ENOMEM; 85 xfs_trans_binval(*trans, bp); 86 /* 87 * Roll to next transaction. 88 */ 89 error = xfs_trans_roll_inode(trans, dp); 90 if (error) 91 return error; 92 } 93 94 tblkno += map.br_blockcount; 95 tblkcnt -= map.br_blockcount; 96 } 97 98 return 0; 99 } 100 101 /* 102 * Invalidate all of the "remote" value regions pointed to by a particular 103 * leaf block. 104 * Note that we must release the lock on the buffer so that we are not 105 * caught holding something that the logging code wants to flush to disk. 106 */ 107 STATIC int 108 xfs_attr3_leaf_inactive( 109 struct xfs_trans **trans, 110 struct xfs_inode *dp, 111 struct xfs_buf *bp) 112 { 113 struct xfs_attr_leafblock *leaf; 114 struct xfs_attr3_icleaf_hdr ichdr; 115 struct xfs_attr_leaf_entry *entry; 116 struct xfs_attr_leaf_name_remote *name_rmt; 117 struct xfs_attr_inactive_list *list; 118 struct xfs_attr_inactive_list *lp; 119 int error; 120 int count; 121 int size; 122 int tmp; 123 int i; 124 struct xfs_mount *mp = bp->b_target->bt_mount; 125 126 leaf = bp->b_addr; 127 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); 128 129 /* 130 * Count the number of "remote" value extents. 131 */ 132 count = 0; 133 entry = xfs_attr3_leaf_entryp(leaf); 134 for (i = 0; i < ichdr.count; entry++, i++) { 135 if (be16_to_cpu(entry->nameidx) && 136 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 137 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 138 if (name_rmt->valueblk) 139 count++; 140 } 141 } 142 143 /* 144 * If there are no "remote" values, we're done. 145 */ 146 if (count == 0) { 147 xfs_trans_brelse(*trans, bp); 148 return 0; 149 } 150 151 /* 152 * Allocate storage for a list of all the "remote" value extents. 153 */ 154 size = count * sizeof(xfs_attr_inactive_list_t); 155 list = kmem_alloc(size, KM_SLEEP); 156 157 /* 158 * Identify each of the "remote" value extents. 159 */ 160 lp = list; 161 entry = xfs_attr3_leaf_entryp(leaf); 162 for (i = 0; i < ichdr.count; entry++, i++) { 163 if (be16_to_cpu(entry->nameidx) && 164 ((entry->flags & XFS_ATTR_LOCAL) == 0)) { 165 name_rmt = xfs_attr3_leaf_name_remote(leaf, i); 166 if (name_rmt->valueblk) { 167 lp->valueblk = be32_to_cpu(name_rmt->valueblk); 168 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount, 169 be32_to_cpu(name_rmt->valuelen)); 170 lp++; 171 } 172 } 173 } 174 xfs_trans_brelse(*trans, bp); /* unlock for trans. in freextent() */ 175 176 /* 177 * Invalidate each of the "remote" value extents. 178 */ 179 error = 0; 180 for (lp = list, i = 0; i < count; i++, lp++) { 181 tmp = xfs_attr3_leaf_freextent(trans, dp, 182 lp->valueblk, lp->valuelen); 183 184 if (error == 0) 185 error = tmp; /* save only the 1st errno */ 186 } 187 188 kmem_free(list); 189 return error; 190 } 191 192 /* 193 * Recurse (gasp!) through the attribute nodes until we find leaves. 194 * We're doing a depth-first traversal in order to invalidate everything. 195 */ 196 STATIC int 197 xfs_attr3_node_inactive( 198 struct xfs_trans **trans, 199 struct xfs_inode *dp, 200 struct xfs_buf *bp, 201 int level) 202 { 203 xfs_da_blkinfo_t *info; 204 xfs_da_intnode_t *node; 205 xfs_dablk_t child_fsb; 206 xfs_daddr_t parent_blkno, child_blkno; 207 int error, i; 208 struct xfs_buf *child_bp; 209 struct xfs_da_node_entry *btree; 210 struct xfs_da3_icnode_hdr ichdr; 211 212 /* 213 * Since this code is recursive (gasp!) we must protect ourselves. 214 */ 215 if (level > XFS_DA_NODE_MAXDEPTH) { 216 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 217 return -EIO; 218 } 219 220 node = bp->b_addr; 221 dp->d_ops->node_hdr_from_disk(&ichdr, node); 222 parent_blkno = bp->b_bn; 223 if (!ichdr.count) { 224 xfs_trans_brelse(*trans, bp); 225 return 0; 226 } 227 btree = dp->d_ops->node_tree_p(node); 228 child_fsb = be32_to_cpu(btree[0].before); 229 xfs_trans_brelse(*trans, bp); /* no locks for later trans */ 230 231 /* 232 * If this is the node level just above the leaves, simply loop 233 * over the leaves removing all of them. If this is higher up 234 * in the tree, recurse downward. 235 */ 236 for (i = 0; i < ichdr.count; i++) { 237 /* 238 * Read the subsidiary block to see what we have to work with. 239 * Don't do this in a transaction. This is a depth-first 240 * traversal of the tree so we may deal with many blocks 241 * before we come back to this one. 242 */ 243 error = xfs_da3_node_read(*trans, dp, child_fsb, -1, &child_bp, 244 XFS_ATTR_FORK); 245 if (error) 246 return error; 247 248 /* save for re-read later */ 249 child_blkno = XFS_BUF_ADDR(child_bp); 250 251 /* 252 * Invalidate the subtree, however we have to. 253 */ 254 info = child_bp->b_addr; 255 switch (info->magic) { 256 case cpu_to_be16(XFS_DA_NODE_MAGIC): 257 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 258 error = xfs_attr3_node_inactive(trans, dp, child_bp, 259 level + 1); 260 break; 261 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 262 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 263 error = xfs_attr3_leaf_inactive(trans, dp, child_bp); 264 break; 265 default: 266 error = -EIO; 267 xfs_trans_brelse(*trans, child_bp); 268 break; 269 } 270 if (error) 271 return error; 272 273 /* 274 * Remove the subsidiary block from the cache and from the log. 275 */ 276 error = xfs_da_get_buf(*trans, dp, 0, child_blkno, &child_bp, 277 XFS_ATTR_FORK); 278 if (error) 279 return error; 280 xfs_trans_binval(*trans, child_bp); 281 282 /* 283 * If we're not done, re-read the parent to get the next 284 * child block number. 285 */ 286 if (i + 1 < ichdr.count) { 287 error = xfs_da3_node_read(*trans, dp, 0, parent_blkno, 288 &bp, XFS_ATTR_FORK); 289 if (error) 290 return error; 291 node = bp->b_addr; 292 btree = dp->d_ops->node_tree_p(node); 293 child_fsb = be32_to_cpu(btree[i + 1].before); 294 xfs_trans_brelse(*trans, bp); 295 } 296 /* 297 * Atomically commit the whole invalidate stuff. 298 */ 299 error = xfs_trans_roll_inode(trans, dp); 300 if (error) 301 return error; 302 } 303 304 return 0; 305 } 306 307 /* 308 * Indiscriminately delete the entire attribute fork 309 * 310 * Recurse (gasp!) through the attribute nodes until we find leaves. 311 * We're doing a depth-first traversal in order to invalidate everything. 312 */ 313 static int 314 xfs_attr3_root_inactive( 315 struct xfs_trans **trans, 316 struct xfs_inode *dp) 317 { 318 struct xfs_da_blkinfo *info; 319 struct xfs_buf *bp; 320 xfs_daddr_t blkno; 321 int error; 322 323 /* 324 * Read block 0 to see what we have to work with. 325 * We only get here if we have extents, since we remove 326 * the extents in reverse order the extent containing 327 * block 0 must still be there. 328 */ 329 error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK); 330 if (error) 331 return error; 332 blkno = bp->b_bn; 333 334 /* 335 * Invalidate the tree, even if the "tree" is only a single leaf block. 336 * This is a depth-first traversal! 337 */ 338 info = bp->b_addr; 339 switch (info->magic) { 340 case cpu_to_be16(XFS_DA_NODE_MAGIC): 341 case cpu_to_be16(XFS_DA3_NODE_MAGIC): 342 error = xfs_attr3_node_inactive(trans, dp, bp, 1); 343 break; 344 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): 345 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): 346 error = xfs_attr3_leaf_inactive(trans, dp, bp); 347 break; 348 default: 349 error = -EIO; 350 xfs_trans_brelse(*trans, bp); 351 break; 352 } 353 if (error) 354 return error; 355 356 /* 357 * Invalidate the incore copy of the root block. 358 */ 359 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK); 360 if (error) 361 return error; 362 xfs_trans_binval(*trans, bp); /* remove from cache */ 363 /* 364 * Commit the invalidate and start the next transaction. 365 */ 366 error = xfs_trans_roll_inode(trans, dp); 367 368 return error; 369 } 370 371 /* 372 * xfs_attr_inactive kills all traces of an attribute fork on an inode. It 373 * removes both the on-disk and in-memory inode fork. Note that this also has to 374 * handle the condition of inodes without attributes but with an attribute fork 375 * configured, so we can't use xfs_inode_hasattr() here. 376 * 377 * The in-memory attribute fork is removed even on error. 378 */ 379 int 380 xfs_attr_inactive( 381 struct xfs_inode *dp) 382 { 383 struct xfs_trans *trans; 384 struct xfs_mount *mp; 385 int lock_mode = XFS_ILOCK_SHARED; 386 int error = 0; 387 388 mp = dp->i_mount; 389 ASSERT(! XFS_NOT_DQATTACHED(mp, dp)); 390 391 xfs_ilock(dp, lock_mode); 392 if (!XFS_IFORK_Q(dp)) 393 goto out_destroy_fork; 394 xfs_iunlock(dp, lock_mode); 395 396 lock_mode = 0; 397 398 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_attrinval, 0, 0, 0, &trans); 399 if (error) 400 goto out_destroy_fork; 401 402 lock_mode = XFS_ILOCK_EXCL; 403 xfs_ilock(dp, lock_mode); 404 405 if (!XFS_IFORK_Q(dp)) 406 goto out_cancel; 407 408 /* 409 * No need to make quota reservations here. We expect to release some 410 * blocks, not allocate, in the common case. 411 */ 412 xfs_trans_ijoin(trans, dp, 0); 413 414 /* 415 * Invalidate and truncate the attribute fork extents. Make sure the 416 * fork actually has attributes as otherwise the invalidation has no 417 * blocks to read and returns an error. In this case, just do the fork 418 * removal below. 419 */ 420 if (xfs_inode_hasattr(dp) && 421 dp->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) { 422 error = xfs_attr3_root_inactive(&trans, dp); 423 if (error) 424 goto out_cancel; 425 426 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0); 427 if (error) 428 goto out_cancel; 429 } 430 431 /* Reset the attribute fork - this also destroys the in-core fork */ 432 xfs_attr_fork_remove(dp, trans); 433 434 error = xfs_trans_commit(trans); 435 xfs_iunlock(dp, lock_mode); 436 return error; 437 438 out_cancel: 439 xfs_trans_cancel(trans); 440 out_destroy_fork: 441 /* kill the in-core attr fork before we drop the inode lock */ 442 if (dp->i_afp) 443 xfs_idestroy_fork(dp, XFS_ATTR_FORK); 444 if (lock_mode) 445 xfs_iunlock(dp, lock_mode); 446 return error; 447 } 448