1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/module.h> 5 #include <linux/fs.h> 6 #include <linux/slab.h> 7 #include <linux/string.h> 8 #include <linux/uaccess.h> 9 #include <linux/kernel.h> 10 #include <linux/writeback.h> 11 #include <linux/vmalloc.h> 12 #include <linux/xattr.h> 13 #include <linux/posix_acl.h> 14 #include <linux/random.h> 15 #include <linux/sort.h> 16 #include <linux/iversion.h> 17 18 #include "super.h" 19 #include "mds_client.h" 20 #include "cache.h" 21 #include <linux/ceph/decode.h> 22 23 /* 24 * Ceph inode operations 25 * 26 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 27 * setattr, etc.), xattr helpers, and helpers for assimilating 28 * metadata returned by the MDS into our cache. 29 * 30 * Also define helpers for doing asynchronous writeback, invalidation, 31 * and truncation for the benefit of those who can't afford to block 32 * (typically because they are in the message handler path). 33 */ 34 35 static const struct inode_operations ceph_symlink_iops; 36 37 static void ceph_inode_work(struct work_struct *work); 38 39 /* 40 * find or create an inode, given the ceph ino number 41 */ 42 static int ceph_set_ino_cb(struct inode *inode, void *data) 43 { 44 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 46 inode_set_iversion_raw(inode, 0); 47 return 0; 48 } 49 50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 51 { 52 struct inode *inode; 53 ino_t t = ceph_vino_to_ino(vino); 54 55 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 56 if (!inode) 57 return ERR_PTR(-ENOMEM); 58 if (inode->i_state & I_NEW) 59 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 60 inode, ceph_vinop(inode), (u64)inode->i_ino); 61 62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 63 vino.snap, inode); 64 return inode; 65 } 66 67 /* 68 * get/constuct snapdir inode for a given directory 69 */ 70 struct inode *ceph_get_snapdir(struct inode *parent) 71 { 72 struct ceph_vino vino = { 73 .ino = ceph_ino(parent), 74 .snap = CEPH_SNAPDIR, 75 }; 76 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 77 struct ceph_inode_info *ci = ceph_inode(inode); 78 79 BUG_ON(!S_ISDIR(parent->i_mode)); 80 if (IS_ERR(inode)) 81 return inode; 82 inode->i_mode = parent->i_mode; 83 inode->i_uid = parent->i_uid; 84 inode->i_gid = parent->i_gid; 85 inode->i_op = &ceph_snapdir_iops; 86 inode->i_fop = &ceph_snapdir_fops; 87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 88 ci->i_rbytes = 0; 89 90 if (inode->i_state & I_NEW) 91 unlock_new_inode(inode); 92 93 return inode; 94 } 95 96 const struct inode_operations ceph_file_iops = { 97 .permission = ceph_permission, 98 .setattr = ceph_setattr, 99 .getattr = ceph_getattr, 100 .listxattr = ceph_listxattr, 101 .get_acl = ceph_get_acl, 102 .set_acl = ceph_set_acl, 103 }; 104 105 106 /* 107 * We use a 'frag tree' to keep track of the MDS's directory fragments 108 * for a given inode (usually there is just a single fragment). We 109 * need to know when a child frag is delegated to a new MDS, or when 110 * it is flagged as replicated, so we can direct our requests 111 * accordingly. 112 */ 113 114 /* 115 * find/create a frag in the tree 116 */ 117 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 118 u32 f) 119 { 120 struct rb_node **p; 121 struct rb_node *parent = NULL; 122 struct ceph_inode_frag *frag; 123 int c; 124 125 p = &ci->i_fragtree.rb_node; 126 while (*p) { 127 parent = *p; 128 frag = rb_entry(parent, struct ceph_inode_frag, node); 129 c = ceph_frag_compare(f, frag->frag); 130 if (c < 0) 131 p = &(*p)->rb_left; 132 else if (c > 0) 133 p = &(*p)->rb_right; 134 else 135 return frag; 136 } 137 138 frag = kmalloc(sizeof(*frag), GFP_NOFS); 139 if (!frag) 140 return ERR_PTR(-ENOMEM); 141 142 frag->frag = f; 143 frag->split_by = 0; 144 frag->mds = -1; 145 frag->ndist = 0; 146 147 rb_link_node(&frag->node, parent, p); 148 rb_insert_color(&frag->node, &ci->i_fragtree); 149 150 dout("get_or_create_frag added %llx.%llx frag %x\n", 151 ceph_vinop(&ci->vfs_inode), f); 152 return frag; 153 } 154 155 /* 156 * find a specific frag @f 157 */ 158 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 159 { 160 struct rb_node *n = ci->i_fragtree.rb_node; 161 162 while (n) { 163 struct ceph_inode_frag *frag = 164 rb_entry(n, struct ceph_inode_frag, node); 165 int c = ceph_frag_compare(f, frag->frag); 166 if (c < 0) 167 n = n->rb_left; 168 else if (c > 0) 169 n = n->rb_right; 170 else 171 return frag; 172 } 173 return NULL; 174 } 175 176 /* 177 * Choose frag containing the given value @v. If @pfrag is 178 * specified, copy the frag delegation info to the caller if 179 * it is present. 180 */ 181 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 182 struct ceph_inode_frag *pfrag, int *found) 183 { 184 u32 t = ceph_frag_make(0, 0); 185 struct ceph_inode_frag *frag; 186 unsigned nway, i; 187 u32 n; 188 189 if (found) 190 *found = 0; 191 192 while (1) { 193 WARN_ON(!ceph_frag_contains_value(t, v)); 194 frag = __ceph_find_frag(ci, t); 195 if (!frag) 196 break; /* t is a leaf */ 197 if (frag->split_by == 0) { 198 if (pfrag) 199 memcpy(pfrag, frag, sizeof(*pfrag)); 200 if (found) 201 *found = 1; 202 break; 203 } 204 205 /* choose child */ 206 nway = 1 << frag->split_by; 207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 208 frag->split_by, nway); 209 for (i = 0; i < nway; i++) { 210 n = ceph_frag_make_child(t, frag->split_by, i); 211 if (ceph_frag_contains_value(n, v)) { 212 t = n; 213 break; 214 } 215 } 216 BUG_ON(i == nway); 217 } 218 dout("choose_frag(%x) = %x\n", v, t); 219 220 return t; 221 } 222 223 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 224 struct ceph_inode_frag *pfrag, int *found) 225 { 226 u32 ret; 227 mutex_lock(&ci->i_fragtree_mutex); 228 ret = __ceph_choose_frag(ci, v, pfrag, found); 229 mutex_unlock(&ci->i_fragtree_mutex); 230 return ret; 231 } 232 233 /* 234 * Process dirfrag (delegation) info from the mds. Include leaf 235 * fragment in tree ONLY if ndist > 0. Otherwise, only 236 * branches/splits are included in i_fragtree) 237 */ 238 static int ceph_fill_dirfrag(struct inode *inode, 239 struct ceph_mds_reply_dirfrag *dirinfo) 240 { 241 struct ceph_inode_info *ci = ceph_inode(inode); 242 struct ceph_inode_frag *frag; 243 u32 id = le32_to_cpu(dirinfo->frag); 244 int mds = le32_to_cpu(dirinfo->auth); 245 int ndist = le32_to_cpu(dirinfo->ndist); 246 int diri_auth = -1; 247 int i; 248 int err = 0; 249 250 spin_lock(&ci->i_ceph_lock); 251 if (ci->i_auth_cap) 252 diri_auth = ci->i_auth_cap->mds; 253 spin_unlock(&ci->i_ceph_lock); 254 255 if (mds == -1) /* CDIR_AUTH_PARENT */ 256 mds = diri_auth; 257 258 mutex_lock(&ci->i_fragtree_mutex); 259 if (ndist == 0 && mds == diri_auth) { 260 /* no delegation info needed. */ 261 frag = __ceph_find_frag(ci, id); 262 if (!frag) 263 goto out; 264 if (frag->split_by == 0) { 265 /* tree leaf, remove */ 266 dout("fill_dirfrag removed %llx.%llx frag %x" 267 " (no ref)\n", ceph_vinop(inode), id); 268 rb_erase(&frag->node, &ci->i_fragtree); 269 kfree(frag); 270 } else { 271 /* tree branch, keep and clear */ 272 dout("fill_dirfrag cleared %llx.%llx frag %x" 273 " referral\n", ceph_vinop(inode), id); 274 frag->mds = -1; 275 frag->ndist = 0; 276 } 277 goto out; 278 } 279 280 281 /* find/add this frag to store mds delegation info */ 282 frag = __get_or_create_frag(ci, id); 283 if (IS_ERR(frag)) { 284 /* this is not the end of the world; we can continue 285 with bad/inaccurate delegation info */ 286 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 287 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 288 err = -ENOMEM; 289 goto out; 290 } 291 292 frag->mds = mds; 293 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 294 for (i = 0; i < frag->ndist; i++) 295 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 296 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 297 ceph_vinop(inode), frag->frag, frag->ndist); 298 299 out: 300 mutex_unlock(&ci->i_fragtree_mutex); 301 return err; 302 } 303 304 static int frag_tree_split_cmp(const void *l, const void *r) 305 { 306 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; 307 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; 308 return ceph_frag_compare(le32_to_cpu(ls->frag), 309 le32_to_cpu(rs->frag)); 310 } 311 312 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) 313 { 314 if (!frag) 315 return f == ceph_frag_make(0, 0); 316 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) 317 return false; 318 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); 319 } 320 321 static int ceph_fill_fragtree(struct inode *inode, 322 struct ceph_frag_tree_head *fragtree, 323 struct ceph_mds_reply_dirfrag *dirinfo) 324 { 325 struct ceph_inode_info *ci = ceph_inode(inode); 326 struct ceph_inode_frag *frag, *prev_frag = NULL; 327 struct rb_node *rb_node; 328 unsigned i, split_by, nsplits; 329 u32 id; 330 bool update = false; 331 332 mutex_lock(&ci->i_fragtree_mutex); 333 nsplits = le32_to_cpu(fragtree->nsplits); 334 if (nsplits != ci->i_fragtree_nsplits) { 335 update = true; 336 } else if (nsplits) { 337 i = prandom_u32() % nsplits; 338 id = le32_to_cpu(fragtree->splits[i].frag); 339 if (!__ceph_find_frag(ci, id)) 340 update = true; 341 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 342 rb_node = rb_first(&ci->i_fragtree); 343 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 344 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 345 update = true; 346 } 347 if (!update && dirinfo) { 348 id = le32_to_cpu(dirinfo->frag); 349 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 350 update = true; 351 } 352 if (!update) 353 goto out_unlock; 354 355 if (nsplits > 1) { 356 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), 357 frag_tree_split_cmp, NULL); 358 } 359 360 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 361 rb_node = rb_first(&ci->i_fragtree); 362 for (i = 0; i < nsplits; i++) { 363 id = le32_to_cpu(fragtree->splits[i].frag); 364 split_by = le32_to_cpu(fragtree->splits[i].by); 365 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { 366 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " 367 "frag %x split by %d\n", ceph_vinop(inode), 368 i, nsplits, id, split_by); 369 continue; 370 } 371 frag = NULL; 372 while (rb_node) { 373 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 374 if (ceph_frag_compare(frag->frag, id) >= 0) { 375 if (frag->frag != id) 376 frag = NULL; 377 else 378 rb_node = rb_next(rb_node); 379 break; 380 } 381 rb_node = rb_next(rb_node); 382 /* delete stale split/leaf node */ 383 if (frag->split_by > 0 || 384 !is_frag_child(frag->frag, prev_frag)) { 385 rb_erase(&frag->node, &ci->i_fragtree); 386 if (frag->split_by > 0) 387 ci->i_fragtree_nsplits--; 388 kfree(frag); 389 } 390 frag = NULL; 391 } 392 if (!frag) { 393 frag = __get_or_create_frag(ci, id); 394 if (IS_ERR(frag)) 395 continue; 396 } 397 if (frag->split_by == 0) 398 ci->i_fragtree_nsplits++; 399 frag->split_by = split_by; 400 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 401 prev_frag = frag; 402 } 403 while (rb_node) { 404 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 405 rb_node = rb_next(rb_node); 406 /* delete stale split/leaf node */ 407 if (frag->split_by > 0 || 408 !is_frag_child(frag->frag, prev_frag)) { 409 rb_erase(&frag->node, &ci->i_fragtree); 410 if (frag->split_by > 0) 411 ci->i_fragtree_nsplits--; 412 kfree(frag); 413 } 414 } 415 out_unlock: 416 mutex_unlock(&ci->i_fragtree_mutex); 417 return 0; 418 } 419 420 /* 421 * initialize a newly allocated inode. 422 */ 423 struct inode *ceph_alloc_inode(struct super_block *sb) 424 { 425 struct ceph_inode_info *ci; 426 int i; 427 428 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 429 if (!ci) 430 return NULL; 431 432 dout("alloc_inode %p\n", &ci->vfs_inode); 433 434 spin_lock_init(&ci->i_ceph_lock); 435 436 ci->i_version = 0; 437 ci->i_inline_version = 0; 438 ci->i_time_warp_seq = 0; 439 ci->i_ceph_flags = 0; 440 atomic64_set(&ci->i_ordered_count, 1); 441 atomic64_set(&ci->i_release_count, 1); 442 atomic64_set(&ci->i_complete_seq[0], 0); 443 atomic64_set(&ci->i_complete_seq[1], 0); 444 ci->i_symlink = NULL; 445 446 ci->i_max_bytes = 0; 447 ci->i_max_files = 0; 448 449 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 450 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); 451 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); 452 453 ci->i_fragtree = RB_ROOT; 454 mutex_init(&ci->i_fragtree_mutex); 455 456 ci->i_xattrs.blob = NULL; 457 ci->i_xattrs.prealloc_blob = NULL; 458 ci->i_xattrs.dirty = false; 459 ci->i_xattrs.index = RB_ROOT; 460 ci->i_xattrs.count = 0; 461 ci->i_xattrs.names_size = 0; 462 ci->i_xattrs.vals_size = 0; 463 ci->i_xattrs.version = 0; 464 ci->i_xattrs.index_version = 0; 465 466 ci->i_caps = RB_ROOT; 467 ci->i_auth_cap = NULL; 468 ci->i_dirty_caps = 0; 469 ci->i_flushing_caps = 0; 470 INIT_LIST_HEAD(&ci->i_dirty_item); 471 INIT_LIST_HEAD(&ci->i_flushing_item); 472 ci->i_prealloc_cap_flush = NULL; 473 INIT_LIST_HEAD(&ci->i_cap_flush_list); 474 init_waitqueue_head(&ci->i_cap_wq); 475 ci->i_hold_caps_min = 0; 476 ci->i_hold_caps_max = 0; 477 INIT_LIST_HEAD(&ci->i_cap_delay_list); 478 INIT_LIST_HEAD(&ci->i_cap_snaps); 479 ci->i_head_snapc = NULL; 480 ci->i_snap_caps = 0; 481 482 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) 483 ci->i_nr_by_mode[i] = 0; 484 485 mutex_init(&ci->i_truncate_mutex); 486 ci->i_truncate_seq = 0; 487 ci->i_truncate_size = 0; 488 ci->i_truncate_pending = 0; 489 490 ci->i_max_size = 0; 491 ci->i_reported_size = 0; 492 ci->i_wanted_max_size = 0; 493 ci->i_requested_max_size = 0; 494 495 ci->i_pin_ref = 0; 496 ci->i_rd_ref = 0; 497 ci->i_rdcache_ref = 0; 498 ci->i_wr_ref = 0; 499 ci->i_wb_ref = 0; 500 ci->i_fx_ref = 0; 501 ci->i_wrbuffer_ref = 0; 502 ci->i_wrbuffer_ref_head = 0; 503 atomic_set(&ci->i_filelock_ref, 0); 504 atomic_set(&ci->i_shared_gen, 1); 505 ci->i_rdcache_gen = 0; 506 ci->i_rdcache_revoking = 0; 507 508 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 509 INIT_LIST_HEAD(&ci->i_unsafe_iops); 510 spin_lock_init(&ci->i_unsafe_lock); 511 512 ci->i_snap_realm = NULL; 513 INIT_LIST_HEAD(&ci->i_snap_realm_item); 514 INIT_LIST_HEAD(&ci->i_snap_flush_item); 515 516 INIT_WORK(&ci->i_work, ceph_inode_work); 517 ci->i_work_mask = 0; 518 memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); 519 520 ceph_fscache_inode_init(ci); 521 522 ci->i_meta_err = 0; 523 524 return &ci->vfs_inode; 525 } 526 527 void ceph_free_inode(struct inode *inode) 528 { 529 struct ceph_inode_info *ci = ceph_inode(inode); 530 531 kfree(ci->i_symlink); 532 kmem_cache_free(ceph_inode_cachep, ci); 533 } 534 535 void ceph_evict_inode(struct inode *inode) 536 { 537 struct ceph_inode_info *ci = ceph_inode(inode); 538 struct ceph_inode_frag *frag; 539 struct rb_node *n; 540 541 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 542 543 truncate_inode_pages_final(&inode->i_data); 544 clear_inode(inode); 545 546 ceph_fscache_unregister_inode_cookie(ci); 547 548 __ceph_remove_caps(ci); 549 550 if (__ceph_has_any_quota(ci)) 551 ceph_adjust_quota_realms_count(inode, false); 552 553 /* 554 * we may still have a snap_realm reference if there are stray 555 * caps in i_snap_caps. 556 */ 557 if (ci->i_snap_realm) { 558 struct ceph_mds_client *mdsc = 559 ceph_inode_to_client(inode)->mdsc; 560 if (ceph_snap(inode) == CEPH_NOSNAP) { 561 struct ceph_snap_realm *realm = ci->i_snap_realm; 562 dout(" dropping residual ref to snap realm %p\n", 563 realm); 564 spin_lock(&realm->inodes_with_caps_lock); 565 list_del_init(&ci->i_snap_realm_item); 566 ci->i_snap_realm = NULL; 567 if (realm->ino == ci->i_vino.ino) 568 realm->inode = NULL; 569 spin_unlock(&realm->inodes_with_caps_lock); 570 ceph_put_snap_realm(mdsc, realm); 571 } else { 572 ceph_put_snapid_map(mdsc, ci->i_snapid_map); 573 ci->i_snap_realm = NULL; 574 } 575 } 576 577 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 578 frag = rb_entry(n, struct ceph_inode_frag, node); 579 rb_erase(n, &ci->i_fragtree); 580 kfree(frag); 581 } 582 ci->i_fragtree_nsplits = 0; 583 584 __ceph_destroy_xattrs(ci); 585 if (ci->i_xattrs.blob) 586 ceph_buffer_put(ci->i_xattrs.blob); 587 if (ci->i_xattrs.prealloc_blob) 588 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 589 590 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); 591 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); 592 } 593 594 static inline blkcnt_t calc_inode_blocks(u64 size) 595 { 596 return (size + (1<<9) - 1) >> 9; 597 } 598 599 /* 600 * Helpers to fill in size, ctime, mtime, and atime. We have to be 601 * careful because either the client or MDS may have more up to date 602 * info, depending on which capabilities are held, and whether 603 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 604 * and size are monotonically increasing, except when utimes() or 605 * truncate() increments the corresponding _seq values.) 606 */ 607 int ceph_fill_file_size(struct inode *inode, int issued, 608 u32 truncate_seq, u64 truncate_size, u64 size) 609 { 610 struct ceph_inode_info *ci = ceph_inode(inode); 611 int queue_trunc = 0; 612 613 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 614 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 615 dout("size %lld -> %llu\n", inode->i_size, size); 616 if (size > 0 && S_ISDIR(inode->i_mode)) { 617 pr_err("fill_file_size non-zero size for directory\n"); 618 size = 0; 619 } 620 i_size_write(inode, size); 621 inode->i_blocks = calc_inode_blocks(size); 622 ci->i_reported_size = size; 623 if (truncate_seq != ci->i_truncate_seq) { 624 dout("truncate_seq %u -> %u\n", 625 ci->i_truncate_seq, truncate_seq); 626 ci->i_truncate_seq = truncate_seq; 627 628 /* the MDS should have revoked these caps */ 629 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 630 CEPH_CAP_FILE_RD | 631 CEPH_CAP_FILE_WR | 632 CEPH_CAP_FILE_LAZYIO)); 633 /* 634 * If we hold relevant caps, or in the case where we're 635 * not the only client referencing this file and we 636 * don't hold those caps, then we need to check whether 637 * the file is either opened or mmaped 638 */ 639 if ((issued & (CEPH_CAP_FILE_CACHE| 640 CEPH_CAP_FILE_BUFFER)) || 641 mapping_mapped(inode->i_mapping) || 642 __ceph_caps_file_wanted(ci)) { 643 ci->i_truncate_pending++; 644 queue_trunc = 1; 645 } 646 } 647 } 648 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 649 ci->i_truncate_size != truncate_size) { 650 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 651 truncate_size); 652 ci->i_truncate_size = truncate_size; 653 } 654 655 if (queue_trunc) 656 ceph_fscache_invalidate(inode); 657 658 return queue_trunc; 659 } 660 661 void ceph_fill_file_time(struct inode *inode, int issued, 662 u64 time_warp_seq, struct timespec64 *ctime, 663 struct timespec64 *mtime, struct timespec64 *atime) 664 { 665 struct ceph_inode_info *ci = ceph_inode(inode); 666 int warn = 0; 667 668 if (issued & (CEPH_CAP_FILE_EXCL| 669 CEPH_CAP_FILE_WR| 670 CEPH_CAP_FILE_BUFFER| 671 CEPH_CAP_AUTH_EXCL| 672 CEPH_CAP_XATTR_EXCL)) { 673 if (ci->i_version == 0 || 674 timespec64_compare(ctime, &inode->i_ctime) > 0) { 675 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", 676 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 677 ctime->tv_sec, ctime->tv_nsec); 678 inode->i_ctime = *ctime; 679 } 680 if (ci->i_version == 0 || 681 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 682 /* the MDS did a utimes() */ 683 dout("mtime %lld.%09ld -> %lld.%09ld " 684 "tw %d -> %d\n", 685 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 686 mtime->tv_sec, mtime->tv_nsec, 687 ci->i_time_warp_seq, (int)time_warp_seq); 688 689 inode->i_mtime = *mtime; 690 inode->i_atime = *atime; 691 ci->i_time_warp_seq = time_warp_seq; 692 } else if (time_warp_seq == ci->i_time_warp_seq) { 693 /* nobody did utimes(); take the max */ 694 if (timespec64_compare(mtime, &inode->i_mtime) > 0) { 695 dout("mtime %lld.%09ld -> %lld.%09ld inc\n", 696 inode->i_mtime.tv_sec, 697 inode->i_mtime.tv_nsec, 698 mtime->tv_sec, mtime->tv_nsec); 699 inode->i_mtime = *mtime; 700 } 701 if (timespec64_compare(atime, &inode->i_atime) > 0) { 702 dout("atime %lld.%09ld -> %lld.%09ld inc\n", 703 inode->i_atime.tv_sec, 704 inode->i_atime.tv_nsec, 705 atime->tv_sec, atime->tv_nsec); 706 inode->i_atime = *atime; 707 } 708 } else if (issued & CEPH_CAP_FILE_EXCL) { 709 /* we did a utimes(); ignore mds values */ 710 } else { 711 warn = 1; 712 } 713 } else { 714 /* we have no write|excl caps; whatever the MDS says is true */ 715 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 716 inode->i_ctime = *ctime; 717 inode->i_mtime = *mtime; 718 inode->i_atime = *atime; 719 ci->i_time_warp_seq = time_warp_seq; 720 } else { 721 warn = 1; 722 } 723 } 724 if (warn) /* time_warp_seq shouldn't go backwards */ 725 dout("%p mds time_warp_seq %llu < %u\n", 726 inode, time_warp_seq, ci->i_time_warp_seq); 727 } 728 729 /* 730 * Populate an inode based on info from mds. May be called on new or 731 * existing inodes. 732 */ 733 int ceph_fill_inode(struct inode *inode, struct page *locked_page, 734 struct ceph_mds_reply_info_in *iinfo, 735 struct ceph_mds_reply_dirfrag *dirinfo, 736 struct ceph_mds_session *session, int cap_fmode, 737 struct ceph_cap_reservation *caps_reservation) 738 { 739 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 740 struct ceph_mds_reply_inode *info = iinfo->in; 741 struct ceph_inode_info *ci = ceph_inode(inode); 742 int issued, new_issued, info_caps; 743 struct timespec64 mtime, atime, ctime; 744 struct ceph_buffer *xattr_blob = NULL; 745 struct ceph_buffer *old_blob = NULL; 746 struct ceph_string *pool_ns = NULL; 747 struct ceph_cap *new_cap = NULL; 748 int err = 0; 749 bool wake = false; 750 bool queue_trunc = false; 751 bool new_version = false; 752 bool fill_inline = false; 753 754 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, 755 inode, ceph_vinop(inode), le64_to_cpu(info->version), 756 ci->i_version); 757 758 info_caps = le32_to_cpu(info->cap.caps); 759 760 /* prealloc new cap struct */ 761 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { 762 new_cap = ceph_get_cap(mdsc, caps_reservation); 763 if (!new_cap) 764 return -ENOMEM; 765 } 766 767 /* 768 * prealloc xattr data, if it looks like we'll need it. only 769 * if len > 4 (meaning there are actually xattrs; the first 4 770 * bytes are the xattr count). 771 */ 772 if (iinfo->xattr_len > 4) { 773 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 774 if (!xattr_blob) 775 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, 776 iinfo->xattr_len); 777 } 778 779 if (iinfo->pool_ns_len > 0) 780 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, 781 iinfo->pool_ns_len); 782 783 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) 784 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); 785 786 spin_lock(&ci->i_ceph_lock); 787 788 /* 789 * provided version will be odd if inode value is projected, 790 * even if stable. skip the update if we have newer stable 791 * info (ours>=theirs, e.g. due to racing mds replies), unless 792 * we are getting projected (unstable) info (in which case the 793 * version is odd, and we want ours>theirs). 794 * us them 795 * 2 2 skip 796 * 3 2 skip 797 * 3 3 update 798 */ 799 if (ci->i_version == 0 || 800 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 801 le64_to_cpu(info->version) > (ci->i_version & ~1))) 802 new_version = true; 803 804 /* Update change_attribute */ 805 inode_set_max_iversion_raw(inode, iinfo->change_attr); 806 807 __ceph_caps_issued(ci, &issued); 808 issued |= __ceph_caps_dirty(ci); 809 new_issued = ~issued & info_caps; 810 811 /* update inode */ 812 inode->i_rdev = le32_to_cpu(info->rdev); 813 /* directories have fl_stripe_unit set to zero */ 814 if (le32_to_cpu(info->layout.fl_stripe_unit)) 815 inode->i_blkbits = 816 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 817 else 818 inode->i_blkbits = CEPH_BLOCK_SHIFT; 819 820 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); 821 822 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 823 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 824 inode->i_mode = le32_to_cpu(info->mode); 825 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 826 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 827 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 828 from_kuid(&init_user_ns, inode->i_uid), 829 from_kgid(&init_user_ns, inode->i_gid)); 830 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); 831 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); 832 } 833 834 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 835 (issued & CEPH_CAP_LINK_EXCL) == 0) 836 set_nlink(inode, le32_to_cpu(info->nlink)); 837 838 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 839 /* be careful with mtime, atime, size */ 840 ceph_decode_timespec64(&atime, &info->atime); 841 ceph_decode_timespec64(&mtime, &info->mtime); 842 ceph_decode_timespec64(&ctime, &info->ctime); 843 ceph_fill_file_time(inode, issued, 844 le32_to_cpu(info->time_warp_seq), 845 &ctime, &mtime, &atime); 846 } 847 848 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { 849 ci->i_files = le64_to_cpu(info->files); 850 ci->i_subdirs = le64_to_cpu(info->subdirs); 851 } 852 853 if (new_version || 854 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 855 s64 old_pool = ci->i_layout.pool_id; 856 struct ceph_string *old_ns; 857 858 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); 859 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, 860 lockdep_is_held(&ci->i_ceph_lock)); 861 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); 862 863 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) 864 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; 865 866 pool_ns = old_ns; 867 868 queue_trunc = ceph_fill_file_size(inode, issued, 869 le32_to_cpu(info->truncate_seq), 870 le64_to_cpu(info->truncate_size), 871 le64_to_cpu(info->size)); 872 /* only update max_size on auth cap */ 873 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 874 ci->i_max_size != le64_to_cpu(info->max_size)) { 875 dout("max_size %lld -> %llu\n", ci->i_max_size, 876 le64_to_cpu(info->max_size)); 877 ci->i_max_size = le64_to_cpu(info->max_size); 878 } 879 } 880 881 /* layout and rstat are not tracked by capability, update them if 882 * the inode info is from auth mds */ 883 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { 884 if (S_ISDIR(inode->i_mode)) { 885 ci->i_dir_layout = iinfo->dir_layout; 886 ci->i_rbytes = le64_to_cpu(info->rbytes); 887 ci->i_rfiles = le64_to_cpu(info->rfiles); 888 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 889 ci->i_dir_pin = iinfo->dir_pin; 890 ceph_decode_timespec64(&ci->i_rctime, &info->rctime); 891 } 892 } 893 894 /* xattrs */ 895 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 896 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 897 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 898 if (ci->i_xattrs.blob) 899 old_blob = ci->i_xattrs.blob; 900 ci->i_xattrs.blob = xattr_blob; 901 if (xattr_blob) 902 memcpy(ci->i_xattrs.blob->vec.iov_base, 903 iinfo->xattr_data, iinfo->xattr_len); 904 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 905 ceph_forget_all_cached_acls(inode); 906 ceph_security_invalidate_secctx(inode); 907 xattr_blob = NULL; 908 } 909 910 /* finally update i_version */ 911 if (le64_to_cpu(info->version) > ci->i_version) 912 ci->i_version = le64_to_cpu(info->version); 913 914 inode->i_mapping->a_ops = &ceph_aops; 915 916 switch (inode->i_mode & S_IFMT) { 917 case S_IFIFO: 918 case S_IFBLK: 919 case S_IFCHR: 920 case S_IFSOCK: 921 inode->i_blkbits = PAGE_SHIFT; 922 init_special_inode(inode, inode->i_mode, inode->i_rdev); 923 inode->i_op = &ceph_file_iops; 924 break; 925 case S_IFREG: 926 inode->i_op = &ceph_file_iops; 927 inode->i_fop = &ceph_file_fops; 928 break; 929 case S_IFLNK: 930 inode->i_op = &ceph_symlink_iops; 931 if (!ci->i_symlink) { 932 u32 symlen = iinfo->symlink_len; 933 char *sym; 934 935 spin_unlock(&ci->i_ceph_lock); 936 937 if (symlen != i_size_read(inode)) { 938 pr_err("%s %llx.%llx BAD symlink " 939 "size %lld\n", __func__, 940 ceph_vinop(inode), 941 i_size_read(inode)); 942 i_size_write(inode, symlen); 943 inode->i_blocks = calc_inode_blocks(symlen); 944 } 945 946 err = -ENOMEM; 947 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 948 if (!sym) 949 goto out; 950 951 spin_lock(&ci->i_ceph_lock); 952 if (!ci->i_symlink) 953 ci->i_symlink = sym; 954 else 955 kfree(sym); /* lost a race */ 956 } 957 inode->i_link = ci->i_symlink; 958 break; 959 case S_IFDIR: 960 inode->i_op = &ceph_dir_iops; 961 inode->i_fop = &ceph_dir_fops; 962 break; 963 default: 964 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, 965 ceph_vinop(inode), inode->i_mode); 966 } 967 968 /* were we issued a capability? */ 969 if (info_caps) { 970 if (ceph_snap(inode) == CEPH_NOSNAP) { 971 ceph_add_cap(inode, session, 972 le64_to_cpu(info->cap.cap_id), 973 cap_fmode, info_caps, 974 le32_to_cpu(info->cap.wanted), 975 le32_to_cpu(info->cap.seq), 976 le32_to_cpu(info->cap.mseq), 977 le64_to_cpu(info->cap.realm), 978 info->cap.flags, &new_cap); 979 980 /* set dir completion flag? */ 981 if (S_ISDIR(inode->i_mode) && 982 ci->i_files == 0 && ci->i_subdirs == 0 && 983 (info_caps & CEPH_CAP_FILE_SHARED) && 984 (issued & CEPH_CAP_FILE_EXCL) == 0 && 985 !__ceph_dir_is_complete(ci)) { 986 dout(" marking %p complete (empty)\n", inode); 987 i_size_write(inode, 0); 988 __ceph_dir_set_complete(ci, 989 atomic64_read(&ci->i_release_count), 990 atomic64_read(&ci->i_ordered_count)); 991 } 992 993 wake = true; 994 } else { 995 dout(" %p got snap_caps %s\n", inode, 996 ceph_cap_string(info_caps)); 997 ci->i_snap_caps |= info_caps; 998 if (cap_fmode >= 0) 999 __ceph_get_fmode(ci, cap_fmode); 1000 } 1001 } else if (cap_fmode >= 0) { 1002 pr_warn("mds issued no caps on %llx.%llx\n", 1003 ceph_vinop(inode)); 1004 __ceph_get_fmode(ci, cap_fmode); 1005 } 1006 1007 if (iinfo->inline_version > 0 && 1008 iinfo->inline_version >= ci->i_inline_version) { 1009 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1010 ci->i_inline_version = iinfo->inline_version; 1011 if (ci->i_inline_version != CEPH_INLINE_NONE && 1012 (locked_page || (info_caps & cache_caps))) 1013 fill_inline = true; 1014 } 1015 1016 spin_unlock(&ci->i_ceph_lock); 1017 1018 if (fill_inline) 1019 ceph_fill_inline_data(inode, locked_page, 1020 iinfo->inline_data, iinfo->inline_len); 1021 1022 if (wake) 1023 wake_up_all(&ci->i_cap_wq); 1024 1025 /* queue truncate if we saw i_size decrease */ 1026 if (queue_trunc) 1027 ceph_queue_vmtruncate(inode); 1028 1029 /* populate frag tree */ 1030 if (S_ISDIR(inode->i_mode)) 1031 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 1032 1033 /* update delegation info? */ 1034 if (dirinfo) 1035 ceph_fill_dirfrag(inode, dirinfo); 1036 1037 err = 0; 1038 out: 1039 if (new_cap) 1040 ceph_put_cap(mdsc, new_cap); 1041 ceph_buffer_put(old_blob); 1042 ceph_buffer_put(xattr_blob); 1043 ceph_put_string(pool_ns); 1044 return err; 1045 } 1046 1047 /* 1048 * caller should hold session s_mutex and dentry->d_lock. 1049 */ 1050 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, 1051 struct ceph_mds_reply_lease *lease, 1052 struct ceph_mds_session *session, 1053 unsigned long from_time, 1054 struct ceph_mds_session **old_lease_session) 1055 { 1056 struct ceph_dentry_info *di = ceph_dentry(dentry); 1057 unsigned mask = le16_to_cpu(lease->mask); 1058 long unsigned duration = le32_to_cpu(lease->duration_ms); 1059 long unsigned ttl = from_time + (duration * HZ) / 1000; 1060 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1061 1062 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1063 dentry, duration, ttl); 1064 1065 /* only track leases on regular dentries */ 1066 if (ceph_snap(dir) != CEPH_NOSNAP) 1067 return; 1068 1069 if (mask & CEPH_LEASE_PRIMARY_LINK) 1070 di->flags |= CEPH_DENTRY_PRIMARY_LINK; 1071 else 1072 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1073 1074 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); 1075 if (!(mask & CEPH_LEASE_VALID)) { 1076 __ceph_dentry_dir_lease_touch(di); 1077 return; 1078 } 1079 1080 if (di->lease_gen == session->s_cap_gen && 1081 time_before(ttl, di->time)) 1082 return; /* we already have a newer lease. */ 1083 1084 if (di->lease_session && di->lease_session != session) { 1085 *old_lease_session = di->lease_session; 1086 di->lease_session = NULL; 1087 } 1088 1089 if (!di->lease_session) 1090 di->lease_session = ceph_get_mds_session(session); 1091 di->lease_gen = session->s_cap_gen; 1092 di->lease_seq = le32_to_cpu(lease->seq); 1093 di->lease_renew_after = half_ttl; 1094 di->lease_renew_from = 0; 1095 di->time = ttl; 1096 1097 __ceph_dentry_lease_touch(di); 1098 } 1099 1100 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, 1101 struct ceph_mds_reply_lease *lease, 1102 struct ceph_mds_session *session, 1103 unsigned long from_time) 1104 { 1105 struct ceph_mds_session *old_lease_session = NULL; 1106 spin_lock(&dentry->d_lock); 1107 __update_dentry_lease(dir, dentry, lease, session, from_time, 1108 &old_lease_session); 1109 spin_unlock(&dentry->d_lock); 1110 if (old_lease_session) 1111 ceph_put_mds_session(old_lease_session); 1112 } 1113 1114 /* 1115 * update dentry lease without having parent inode locked 1116 */ 1117 static void update_dentry_lease_careful(struct dentry *dentry, 1118 struct ceph_mds_reply_lease *lease, 1119 struct ceph_mds_session *session, 1120 unsigned long from_time, 1121 char *dname, u32 dname_len, 1122 struct ceph_vino *pdvino, 1123 struct ceph_vino *ptvino) 1124 1125 { 1126 struct inode *dir; 1127 struct ceph_mds_session *old_lease_session = NULL; 1128 1129 spin_lock(&dentry->d_lock); 1130 /* make sure dentry's name matches target */ 1131 if (dentry->d_name.len != dname_len || 1132 memcmp(dentry->d_name.name, dname, dname_len)) 1133 goto out_unlock; 1134 1135 dir = d_inode(dentry->d_parent); 1136 /* make sure parent matches dvino */ 1137 if (!ceph_ino_compare(dir, pdvino)) 1138 goto out_unlock; 1139 1140 /* make sure dentry's inode matches target. NULL ptvino means that 1141 * we expect a negative dentry */ 1142 if (ptvino) { 1143 if (d_really_is_negative(dentry)) 1144 goto out_unlock; 1145 if (!ceph_ino_compare(d_inode(dentry), ptvino)) 1146 goto out_unlock; 1147 } else { 1148 if (d_really_is_positive(dentry)) 1149 goto out_unlock; 1150 } 1151 1152 __update_dentry_lease(dir, dentry, lease, session, 1153 from_time, &old_lease_session); 1154 out_unlock: 1155 spin_unlock(&dentry->d_lock); 1156 if (old_lease_session) 1157 ceph_put_mds_session(old_lease_session); 1158 } 1159 1160 /* 1161 * splice a dentry to an inode. 1162 * caller must hold directory i_mutex for this to be safe. 1163 */ 1164 static int splice_dentry(struct dentry **pdn, struct inode *in) 1165 { 1166 struct dentry *dn = *pdn; 1167 struct dentry *realdn; 1168 1169 BUG_ON(d_inode(dn)); 1170 1171 if (S_ISDIR(in->i_mode)) { 1172 /* If inode is directory, d_splice_alias() below will remove 1173 * 'realdn' from its origin parent. We need to ensure that 1174 * origin parent's readdir cache will not reference 'realdn' 1175 */ 1176 realdn = d_find_any_alias(in); 1177 if (realdn) { 1178 struct ceph_dentry_info *di = ceph_dentry(realdn); 1179 spin_lock(&realdn->d_lock); 1180 1181 realdn->d_op->d_prune(realdn); 1182 1183 di->time = jiffies; 1184 di->lease_shared_gen = 0; 1185 di->offset = 0; 1186 1187 spin_unlock(&realdn->d_lock); 1188 dput(realdn); 1189 } 1190 } 1191 1192 /* dn must be unhashed */ 1193 if (!d_unhashed(dn)) 1194 d_drop(dn); 1195 realdn = d_splice_alias(in, dn); 1196 if (IS_ERR(realdn)) { 1197 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1198 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1199 return PTR_ERR(realdn); 1200 } 1201 1202 if (realdn) { 1203 dout("dn %p (%d) spliced with %p (%d) " 1204 "inode %p ino %llx.%llx\n", 1205 dn, d_count(dn), 1206 realdn, d_count(realdn), 1207 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1208 dput(dn); 1209 *pdn = realdn; 1210 } else { 1211 BUG_ON(!ceph_dentry(dn)); 1212 dout("dn %p attached to %p ino %llx.%llx\n", 1213 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1214 } 1215 return 0; 1216 } 1217 1218 /* 1219 * Incorporate results into the local cache. This is either just 1220 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1221 * after a lookup). 1222 * 1223 * A reply may contain 1224 * a directory inode along with a dentry. 1225 * and/or a target inode 1226 * 1227 * Called with snap_rwsem (read). 1228 */ 1229 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1230 { 1231 struct ceph_mds_session *session = req->r_session; 1232 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1233 struct inode *in = NULL; 1234 struct ceph_vino tvino, dvino; 1235 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1236 int err = 0; 1237 1238 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1239 rinfo->head->is_dentry, rinfo->head->is_target); 1240 1241 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1242 dout("fill_trace reply is empty!\n"); 1243 if (rinfo->head->result == 0 && req->r_parent) 1244 ceph_invalidate_dir_request(req); 1245 return 0; 1246 } 1247 1248 if (rinfo->head->is_dentry) { 1249 struct inode *dir = req->r_parent; 1250 1251 if (dir) { 1252 err = ceph_fill_inode(dir, NULL, &rinfo->diri, 1253 rinfo->dirfrag, session, -1, 1254 &req->r_caps_reservation); 1255 if (err < 0) 1256 goto done; 1257 } else { 1258 WARN_ON_ONCE(1); 1259 } 1260 1261 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && 1262 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1263 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1264 struct qstr dname; 1265 struct dentry *dn, *parent; 1266 1267 BUG_ON(!rinfo->head->is_target); 1268 BUG_ON(req->r_dentry); 1269 1270 parent = d_find_any_alias(dir); 1271 BUG_ON(!parent); 1272 1273 dname.name = rinfo->dname; 1274 dname.len = rinfo->dname_len; 1275 dname.hash = full_name_hash(parent, dname.name, dname.len); 1276 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1277 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1278 retry_lookup: 1279 dn = d_lookup(parent, &dname); 1280 dout("d_lookup on parent=%p name=%.*s got %p\n", 1281 parent, dname.len, dname.name, dn); 1282 1283 if (!dn) { 1284 dn = d_alloc(parent, &dname); 1285 dout("d_alloc %p '%.*s' = %p\n", parent, 1286 dname.len, dname.name, dn); 1287 if (!dn) { 1288 dput(parent); 1289 err = -ENOMEM; 1290 goto done; 1291 } 1292 err = 0; 1293 } else if (d_really_is_positive(dn) && 1294 (ceph_ino(d_inode(dn)) != tvino.ino || 1295 ceph_snap(d_inode(dn)) != tvino.snap)) { 1296 dout(" dn %p points to wrong inode %p\n", 1297 dn, d_inode(dn)); 1298 ceph_dir_clear_ordered(dir); 1299 d_delete(dn); 1300 dput(dn); 1301 goto retry_lookup; 1302 } 1303 1304 req->r_dentry = dn; 1305 dput(parent); 1306 } 1307 } 1308 1309 if (rinfo->head->is_target) { 1310 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1311 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1312 1313 in = ceph_get_inode(sb, tvino); 1314 if (IS_ERR(in)) { 1315 err = PTR_ERR(in); 1316 goto done; 1317 } 1318 1319 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, 1320 NULL, session, 1321 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1322 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && 1323 rinfo->head->result == 0) ? req->r_fmode : -1, 1324 &req->r_caps_reservation); 1325 if (err < 0) { 1326 pr_err("ceph_fill_inode badness %p %llx.%llx\n", 1327 in, ceph_vinop(in)); 1328 if (in->i_state & I_NEW) 1329 discard_new_inode(in); 1330 goto done; 1331 } 1332 req->r_target_inode = in; 1333 if (in->i_state & I_NEW) 1334 unlock_new_inode(in); 1335 } 1336 1337 /* 1338 * ignore null lease/binding on snapdir ENOENT, or else we 1339 * will have trouble splicing in the virtual snapdir later 1340 */ 1341 if (rinfo->head->is_dentry && 1342 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1343 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1344 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1345 fsc->mount_options->snapdir_name, 1346 req->r_dentry->d_name.len))) { 1347 /* 1348 * lookup link rename : null -> possibly existing inode 1349 * mknod symlink mkdir : null -> new inode 1350 * unlink : linked -> null 1351 */ 1352 struct inode *dir = req->r_parent; 1353 struct dentry *dn = req->r_dentry; 1354 bool have_dir_cap, have_lease; 1355 1356 BUG_ON(!dn); 1357 BUG_ON(!dir); 1358 BUG_ON(d_inode(dn->d_parent) != dir); 1359 1360 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1361 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1362 1363 BUG_ON(ceph_ino(dir) != dvino.ino); 1364 BUG_ON(ceph_snap(dir) != dvino.snap); 1365 1366 /* do we have a lease on the whole dir? */ 1367 have_dir_cap = 1368 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1369 CEPH_CAP_FILE_SHARED); 1370 1371 /* do we have a dn lease? */ 1372 have_lease = have_dir_cap || 1373 le32_to_cpu(rinfo->dlease->duration_ms); 1374 if (!have_lease) 1375 dout("fill_trace no dentry lease or dir cap\n"); 1376 1377 /* rename? */ 1378 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1379 struct inode *olddir = req->r_old_dentry_dir; 1380 BUG_ON(!olddir); 1381 1382 dout(" src %p '%pd' dst %p '%pd'\n", 1383 req->r_old_dentry, 1384 req->r_old_dentry, 1385 dn, dn); 1386 dout("fill_trace doing d_move %p -> %p\n", 1387 req->r_old_dentry, dn); 1388 1389 /* d_move screws up sibling dentries' offsets */ 1390 ceph_dir_clear_ordered(dir); 1391 ceph_dir_clear_ordered(olddir); 1392 1393 d_move(req->r_old_dentry, dn); 1394 dout(" src %p '%pd' dst %p '%pd'\n", 1395 req->r_old_dentry, 1396 req->r_old_dentry, 1397 dn, dn); 1398 1399 /* ensure target dentry is invalidated, despite 1400 rehashing bug in vfs_rename_dir */ 1401 ceph_invalidate_dentry_lease(dn); 1402 1403 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1404 ceph_dentry(req->r_old_dentry)->offset); 1405 1406 /* swap r_dentry and r_old_dentry in case that 1407 * splice_dentry() gets called later. This is safe 1408 * because no other place will use them */ 1409 req->r_dentry = req->r_old_dentry; 1410 req->r_old_dentry = dn; 1411 dn = req->r_dentry; 1412 } 1413 1414 /* null dentry? */ 1415 if (!rinfo->head->is_target) { 1416 dout("fill_trace null dentry\n"); 1417 if (d_really_is_positive(dn)) { 1418 dout("d_delete %p\n", dn); 1419 ceph_dir_clear_ordered(dir); 1420 d_delete(dn); 1421 } else if (have_lease) { 1422 if (d_unhashed(dn)) 1423 d_add(dn, NULL); 1424 update_dentry_lease(dir, dn, 1425 rinfo->dlease, session, 1426 req->r_request_started); 1427 } 1428 goto done; 1429 } 1430 1431 /* attach proper inode */ 1432 if (d_really_is_negative(dn)) { 1433 ceph_dir_clear_ordered(dir); 1434 ihold(in); 1435 err = splice_dentry(&req->r_dentry, in); 1436 if (err < 0) 1437 goto done; 1438 dn = req->r_dentry; /* may have spliced */ 1439 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1440 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1441 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1442 ceph_vinop(in)); 1443 d_invalidate(dn); 1444 have_lease = false; 1445 } 1446 1447 if (have_lease) { 1448 update_dentry_lease(dir, dn, 1449 rinfo->dlease, session, 1450 req->r_request_started); 1451 } 1452 dout(" final dn %p\n", dn); 1453 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1454 req->r_op == CEPH_MDS_OP_MKSNAP) && 1455 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1456 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1457 struct inode *dir = req->r_parent; 1458 1459 /* fill out a snapdir LOOKUPSNAP dentry */ 1460 BUG_ON(!dir); 1461 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1462 BUG_ON(!req->r_dentry); 1463 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); 1464 ceph_dir_clear_ordered(dir); 1465 ihold(in); 1466 err = splice_dentry(&req->r_dentry, in); 1467 if (err < 0) 1468 goto done; 1469 } else if (rinfo->head->is_dentry && req->r_dentry) { 1470 /* parent inode is not locked, be carefull */ 1471 struct ceph_vino *ptvino = NULL; 1472 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1473 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1474 if (rinfo->head->is_target) { 1475 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1476 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1477 ptvino = &tvino; 1478 } 1479 update_dentry_lease_careful(req->r_dentry, rinfo->dlease, 1480 session, req->r_request_started, 1481 rinfo->dname, rinfo->dname_len, 1482 &dvino, ptvino); 1483 } 1484 done: 1485 dout("fill_trace done err=%d\n", err); 1486 return err; 1487 } 1488 1489 /* 1490 * Prepopulate our cache with readdir results, leases, etc. 1491 */ 1492 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1493 struct ceph_mds_session *session) 1494 { 1495 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1496 int i, err = 0; 1497 1498 for (i = 0; i < rinfo->dir_nr; i++) { 1499 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1500 struct ceph_vino vino; 1501 struct inode *in; 1502 int rc; 1503 1504 vino.ino = le64_to_cpu(rde->inode.in->ino); 1505 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1506 1507 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1508 if (IS_ERR(in)) { 1509 err = PTR_ERR(in); 1510 dout("new_inode badness got %d\n", err); 1511 continue; 1512 } 1513 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1514 -1, &req->r_caps_reservation); 1515 if (rc < 0) { 1516 pr_err("ceph_fill_inode badness on %p got %d\n", 1517 in, rc); 1518 err = rc; 1519 if (in->i_state & I_NEW) { 1520 ihold(in); 1521 discard_new_inode(in); 1522 } 1523 } else if (in->i_state & I_NEW) { 1524 unlock_new_inode(in); 1525 } 1526 1527 /* avoid calling iput_final() in mds dispatch threads */ 1528 ceph_async_iput(in); 1529 } 1530 1531 return err; 1532 } 1533 1534 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1535 { 1536 if (ctl->page) { 1537 kunmap(ctl->page); 1538 put_page(ctl->page); 1539 ctl->page = NULL; 1540 } 1541 } 1542 1543 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1544 struct ceph_readdir_cache_control *ctl, 1545 struct ceph_mds_request *req) 1546 { 1547 struct ceph_inode_info *ci = ceph_inode(dir); 1548 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1549 unsigned idx = ctl->index % nsize; 1550 pgoff_t pgoff = ctl->index / nsize; 1551 1552 if (!ctl->page || pgoff != page_index(ctl->page)) { 1553 ceph_readdir_cache_release(ctl); 1554 if (idx == 0) 1555 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1556 else 1557 ctl->page = find_lock_page(&dir->i_data, pgoff); 1558 if (!ctl->page) { 1559 ctl->index = -1; 1560 return idx == 0 ? -ENOMEM : 0; 1561 } 1562 /* reading/filling the cache are serialized by 1563 * i_mutex, no need to use page lock */ 1564 unlock_page(ctl->page); 1565 ctl->dentries = kmap(ctl->page); 1566 if (idx == 0) 1567 memset(ctl->dentries, 0, PAGE_SIZE); 1568 } 1569 1570 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1571 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1572 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1573 ctl->dentries[idx] = dn; 1574 ctl->index++; 1575 } else { 1576 dout("disable readdir cache\n"); 1577 ctl->index = -1; 1578 } 1579 return 0; 1580 } 1581 1582 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1583 struct ceph_mds_session *session) 1584 { 1585 struct dentry *parent = req->r_dentry; 1586 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1587 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1588 struct qstr dname; 1589 struct dentry *dn; 1590 struct inode *in; 1591 int err = 0, skipped = 0, ret, i; 1592 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1593 u32 frag = le32_to_cpu(rhead->args.readdir.frag); 1594 u32 last_hash = 0; 1595 u32 fpos_offset; 1596 struct ceph_readdir_cache_control cache_ctl = {}; 1597 1598 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1599 return readdir_prepopulate_inodes_only(req, session); 1600 1601 if (rinfo->hash_order) { 1602 if (req->r_path2) { 1603 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1604 req->r_path2, 1605 strlen(req->r_path2)); 1606 last_hash = ceph_frag_value(last_hash); 1607 } else if (rinfo->offset_hash) { 1608 /* mds understands offset_hash */ 1609 WARN_ON_ONCE(req->r_readdir_offset != 2); 1610 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); 1611 } 1612 } 1613 1614 if (rinfo->dir_dir && 1615 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1616 dout("readdir_prepopulate got new frag %x -> %x\n", 1617 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1618 frag = le32_to_cpu(rinfo->dir_dir->frag); 1619 if (!rinfo->hash_order) 1620 req->r_readdir_offset = 2; 1621 } 1622 1623 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1624 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1625 rinfo->dir_nr, parent); 1626 } else { 1627 dout("readdir_prepopulate %d items under dn %p\n", 1628 rinfo->dir_nr, parent); 1629 if (rinfo->dir_dir) 1630 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1631 1632 if (ceph_frag_is_leftmost(frag) && 1633 req->r_readdir_offset == 2 && 1634 !(rinfo->hash_order && last_hash)) { 1635 /* note dir version at start of readdir so we can 1636 * tell if any dentries get dropped */ 1637 req->r_dir_release_cnt = 1638 atomic64_read(&ci->i_release_count); 1639 req->r_dir_ordered_cnt = 1640 atomic64_read(&ci->i_ordered_count); 1641 req->r_readdir_cache_idx = 0; 1642 } 1643 } 1644 1645 cache_ctl.index = req->r_readdir_cache_idx; 1646 fpos_offset = req->r_readdir_offset; 1647 1648 /* FIXME: release caps/leases if error occurs */ 1649 for (i = 0; i < rinfo->dir_nr; i++) { 1650 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1651 struct ceph_vino tvino; 1652 1653 dname.name = rde->name; 1654 dname.len = rde->name_len; 1655 dname.hash = full_name_hash(parent, dname.name, dname.len); 1656 1657 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1658 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1659 1660 if (rinfo->hash_order) { 1661 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1662 rde->name, rde->name_len); 1663 hash = ceph_frag_value(hash); 1664 if (hash != last_hash) 1665 fpos_offset = 2; 1666 last_hash = hash; 1667 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1668 } else { 1669 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1670 } 1671 1672 retry_lookup: 1673 dn = d_lookup(parent, &dname); 1674 dout("d_lookup on parent=%p name=%.*s got %p\n", 1675 parent, dname.len, dname.name, dn); 1676 1677 if (!dn) { 1678 dn = d_alloc(parent, &dname); 1679 dout("d_alloc %p '%.*s' = %p\n", parent, 1680 dname.len, dname.name, dn); 1681 if (!dn) { 1682 dout("d_alloc badness\n"); 1683 err = -ENOMEM; 1684 goto out; 1685 } 1686 } else if (d_really_is_positive(dn) && 1687 (ceph_ino(d_inode(dn)) != tvino.ino || 1688 ceph_snap(d_inode(dn)) != tvino.snap)) { 1689 struct ceph_dentry_info *di = ceph_dentry(dn); 1690 dout(" dn %p points to wrong inode %p\n", 1691 dn, d_inode(dn)); 1692 1693 spin_lock(&dn->d_lock); 1694 if (di->offset > 0 && 1695 di->lease_shared_gen == 1696 atomic_read(&ci->i_shared_gen)) { 1697 __ceph_dir_clear_ordered(ci); 1698 di->offset = 0; 1699 } 1700 spin_unlock(&dn->d_lock); 1701 1702 d_delete(dn); 1703 dput(dn); 1704 goto retry_lookup; 1705 } 1706 1707 /* inode */ 1708 if (d_really_is_positive(dn)) { 1709 in = d_inode(dn); 1710 } else { 1711 in = ceph_get_inode(parent->d_sb, tvino); 1712 if (IS_ERR(in)) { 1713 dout("new_inode badness\n"); 1714 d_drop(dn); 1715 dput(dn); 1716 err = PTR_ERR(in); 1717 goto out; 1718 } 1719 } 1720 1721 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1722 -1, &req->r_caps_reservation); 1723 if (ret < 0) { 1724 pr_err("ceph_fill_inode badness on %p\n", in); 1725 if (d_really_is_negative(dn)) { 1726 /* avoid calling iput_final() in mds 1727 * dispatch threads */ 1728 if (in->i_state & I_NEW) { 1729 ihold(in); 1730 discard_new_inode(in); 1731 } 1732 ceph_async_iput(in); 1733 } 1734 d_drop(dn); 1735 err = ret; 1736 goto next_item; 1737 } 1738 if (in->i_state & I_NEW) 1739 unlock_new_inode(in); 1740 1741 if (d_really_is_negative(dn)) { 1742 if (ceph_security_xattr_deadlock(in)) { 1743 dout(" skip splicing dn %p to inode %p" 1744 " (security xattr deadlock)\n", dn, in); 1745 ceph_async_iput(in); 1746 skipped++; 1747 goto next_item; 1748 } 1749 1750 err = splice_dentry(&dn, in); 1751 if (err < 0) 1752 goto next_item; 1753 } 1754 1755 ceph_dentry(dn)->offset = rde->offset; 1756 1757 update_dentry_lease(d_inode(parent), dn, 1758 rde->lease, req->r_session, 1759 req->r_request_started); 1760 1761 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1762 ret = fill_readdir_cache(d_inode(parent), dn, 1763 &cache_ctl, req); 1764 if (ret < 0) 1765 err = ret; 1766 } 1767 next_item: 1768 dput(dn); 1769 } 1770 out: 1771 if (err == 0 && skipped == 0) { 1772 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1773 req->r_readdir_cache_idx = cache_ctl.index; 1774 } 1775 ceph_readdir_cache_release(&cache_ctl); 1776 dout("readdir_prepopulate done\n"); 1777 return err; 1778 } 1779 1780 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1781 { 1782 struct ceph_inode_info *ci = ceph_inode(inode); 1783 bool ret; 1784 1785 spin_lock(&ci->i_ceph_lock); 1786 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1787 i_size_write(inode, size); 1788 inode->i_blocks = calc_inode_blocks(size); 1789 1790 ret = __ceph_should_report_size(ci); 1791 1792 spin_unlock(&ci->i_ceph_lock); 1793 return ret; 1794 } 1795 1796 /* 1797 * Put reference to inode, but avoid calling iput_final() in current thread. 1798 * iput_final() may wait for reahahead pages. The wait can cause deadlock in 1799 * some contexts. 1800 */ 1801 void ceph_async_iput(struct inode *inode) 1802 { 1803 if (!inode) 1804 return; 1805 for (;;) { 1806 if (atomic_add_unless(&inode->i_count, -1, 1)) 1807 break; 1808 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1809 &ceph_inode(inode)->i_work)) 1810 break; 1811 /* queue work failed, i_count must be at least 2 */ 1812 } 1813 } 1814 1815 /* 1816 * Write back inode data in a worker thread. (This can't be done 1817 * in the message handler context.) 1818 */ 1819 void ceph_queue_writeback(struct inode *inode) 1820 { 1821 struct ceph_inode_info *ci = ceph_inode(inode); 1822 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask); 1823 1824 ihold(inode); 1825 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1826 &ci->i_work)) { 1827 dout("ceph_queue_writeback %p\n", inode); 1828 } else { 1829 dout("ceph_queue_writeback %p already queued, mask=%lx\n", 1830 inode, ci->i_work_mask); 1831 iput(inode); 1832 } 1833 } 1834 1835 /* 1836 * queue an async invalidation 1837 */ 1838 void ceph_queue_invalidate(struct inode *inode) 1839 { 1840 struct ceph_inode_info *ci = ceph_inode(inode); 1841 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask); 1842 1843 ihold(inode); 1844 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1845 &ceph_inode(inode)->i_work)) { 1846 dout("ceph_queue_invalidate %p\n", inode); 1847 } else { 1848 dout("ceph_queue_invalidate %p already queued, mask=%lx\n", 1849 inode, ci->i_work_mask); 1850 iput(inode); 1851 } 1852 } 1853 1854 /* 1855 * Queue an async vmtruncate. If we fail to queue work, we will handle 1856 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1857 */ 1858 void ceph_queue_vmtruncate(struct inode *inode) 1859 { 1860 struct ceph_inode_info *ci = ceph_inode(inode); 1861 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask); 1862 1863 ihold(inode); 1864 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1865 &ci->i_work)) { 1866 dout("ceph_queue_vmtruncate %p\n", inode); 1867 } else { 1868 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n", 1869 inode, ci->i_work_mask); 1870 iput(inode); 1871 } 1872 } 1873 1874 static void ceph_do_invalidate_pages(struct inode *inode) 1875 { 1876 struct ceph_inode_info *ci = ceph_inode(inode); 1877 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1878 u32 orig_gen; 1879 int check = 0; 1880 1881 mutex_lock(&ci->i_truncate_mutex); 1882 1883 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 1884 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1885 inode, ceph_ino(inode)); 1886 mapping_set_error(inode->i_mapping, -EIO); 1887 truncate_pagecache(inode, 0); 1888 mutex_unlock(&ci->i_truncate_mutex); 1889 goto out; 1890 } 1891 1892 spin_lock(&ci->i_ceph_lock); 1893 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1894 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1895 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1896 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1897 check = 1; 1898 spin_unlock(&ci->i_ceph_lock); 1899 mutex_unlock(&ci->i_truncate_mutex); 1900 goto out; 1901 } 1902 orig_gen = ci->i_rdcache_gen; 1903 spin_unlock(&ci->i_ceph_lock); 1904 1905 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1906 pr_err("invalidate_pages %p fails\n", inode); 1907 } 1908 1909 spin_lock(&ci->i_ceph_lock); 1910 if (orig_gen == ci->i_rdcache_gen && 1911 orig_gen == ci->i_rdcache_revoking) { 1912 dout("invalidate_pages %p gen %d successful\n", inode, 1913 ci->i_rdcache_gen); 1914 ci->i_rdcache_revoking--; 1915 check = 1; 1916 } else { 1917 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1918 inode, orig_gen, ci->i_rdcache_gen, 1919 ci->i_rdcache_revoking); 1920 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1921 check = 1; 1922 } 1923 spin_unlock(&ci->i_ceph_lock); 1924 mutex_unlock(&ci->i_truncate_mutex); 1925 out: 1926 if (check) 1927 ceph_check_caps(ci, 0, NULL); 1928 } 1929 1930 /* 1931 * Make sure any pending truncation is applied before doing anything 1932 * that may depend on it. 1933 */ 1934 void __ceph_do_pending_vmtruncate(struct inode *inode) 1935 { 1936 struct ceph_inode_info *ci = ceph_inode(inode); 1937 u64 to; 1938 int wrbuffer_refs, finish = 0; 1939 1940 mutex_lock(&ci->i_truncate_mutex); 1941 retry: 1942 spin_lock(&ci->i_ceph_lock); 1943 if (ci->i_truncate_pending == 0) { 1944 dout("__do_pending_vmtruncate %p none pending\n", inode); 1945 spin_unlock(&ci->i_ceph_lock); 1946 mutex_unlock(&ci->i_truncate_mutex); 1947 return; 1948 } 1949 1950 /* 1951 * make sure any dirty snapped pages are flushed before we 1952 * possibly truncate them.. so write AND block! 1953 */ 1954 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1955 spin_unlock(&ci->i_ceph_lock); 1956 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1957 inode); 1958 filemap_write_and_wait_range(&inode->i_data, 0, 1959 inode->i_sb->s_maxbytes); 1960 goto retry; 1961 } 1962 1963 /* there should be no reader or writer */ 1964 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1965 1966 to = ci->i_truncate_size; 1967 wrbuffer_refs = ci->i_wrbuffer_ref; 1968 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1969 ci->i_truncate_pending, to); 1970 spin_unlock(&ci->i_ceph_lock); 1971 1972 truncate_pagecache(inode, to); 1973 1974 spin_lock(&ci->i_ceph_lock); 1975 if (to == ci->i_truncate_size) { 1976 ci->i_truncate_pending = 0; 1977 finish = 1; 1978 } 1979 spin_unlock(&ci->i_ceph_lock); 1980 if (!finish) 1981 goto retry; 1982 1983 mutex_unlock(&ci->i_truncate_mutex); 1984 1985 if (wrbuffer_refs == 0) 1986 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1987 1988 wake_up_all(&ci->i_cap_wq); 1989 } 1990 1991 static void ceph_inode_work(struct work_struct *work) 1992 { 1993 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1994 i_work); 1995 struct inode *inode = &ci->vfs_inode; 1996 1997 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { 1998 dout("writeback %p\n", inode); 1999 filemap_fdatawrite(&inode->i_data); 2000 } 2001 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) 2002 ceph_do_invalidate_pages(inode); 2003 2004 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) 2005 __ceph_do_pending_vmtruncate(inode); 2006 2007 iput(inode); 2008 } 2009 2010 /* 2011 * symlinks 2012 */ 2013 static const struct inode_operations ceph_symlink_iops = { 2014 .get_link = simple_get_link, 2015 .setattr = ceph_setattr, 2016 .getattr = ceph_getattr, 2017 .listxattr = ceph_listxattr, 2018 }; 2019 2020 int __ceph_setattr(struct inode *inode, struct iattr *attr) 2021 { 2022 struct ceph_inode_info *ci = ceph_inode(inode); 2023 unsigned int ia_valid = attr->ia_valid; 2024 struct ceph_mds_request *req; 2025 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 2026 struct ceph_cap_flush *prealloc_cf; 2027 int issued; 2028 int release = 0, dirtied = 0; 2029 int mask = 0; 2030 int err = 0; 2031 int inode_dirty_flags = 0; 2032 bool lock_snap_rwsem = false; 2033 2034 prealloc_cf = ceph_alloc_cap_flush(); 2035 if (!prealloc_cf) 2036 return -ENOMEM; 2037 2038 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 2039 USE_AUTH_MDS); 2040 if (IS_ERR(req)) { 2041 ceph_free_cap_flush(prealloc_cf); 2042 return PTR_ERR(req); 2043 } 2044 2045 spin_lock(&ci->i_ceph_lock); 2046 issued = __ceph_caps_issued(ci, NULL); 2047 2048 if (!ci->i_head_snapc && 2049 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 2050 lock_snap_rwsem = true; 2051 if (!down_read_trylock(&mdsc->snap_rwsem)) { 2052 spin_unlock(&ci->i_ceph_lock); 2053 down_read(&mdsc->snap_rwsem); 2054 spin_lock(&ci->i_ceph_lock); 2055 issued = __ceph_caps_issued(ci, NULL); 2056 } 2057 } 2058 2059 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 2060 2061 if (ia_valid & ATTR_UID) { 2062 dout("setattr %p uid %d -> %d\n", inode, 2063 from_kuid(&init_user_ns, inode->i_uid), 2064 from_kuid(&init_user_ns, attr->ia_uid)); 2065 if (issued & CEPH_CAP_AUTH_EXCL) { 2066 inode->i_uid = attr->ia_uid; 2067 dirtied |= CEPH_CAP_AUTH_EXCL; 2068 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2069 !uid_eq(attr->ia_uid, inode->i_uid)) { 2070 req->r_args.setattr.uid = cpu_to_le32( 2071 from_kuid(&init_user_ns, attr->ia_uid)); 2072 mask |= CEPH_SETATTR_UID; 2073 release |= CEPH_CAP_AUTH_SHARED; 2074 } 2075 } 2076 if (ia_valid & ATTR_GID) { 2077 dout("setattr %p gid %d -> %d\n", inode, 2078 from_kgid(&init_user_ns, inode->i_gid), 2079 from_kgid(&init_user_ns, attr->ia_gid)); 2080 if (issued & CEPH_CAP_AUTH_EXCL) { 2081 inode->i_gid = attr->ia_gid; 2082 dirtied |= CEPH_CAP_AUTH_EXCL; 2083 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2084 !gid_eq(attr->ia_gid, inode->i_gid)) { 2085 req->r_args.setattr.gid = cpu_to_le32( 2086 from_kgid(&init_user_ns, attr->ia_gid)); 2087 mask |= CEPH_SETATTR_GID; 2088 release |= CEPH_CAP_AUTH_SHARED; 2089 } 2090 } 2091 if (ia_valid & ATTR_MODE) { 2092 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 2093 attr->ia_mode); 2094 if (issued & CEPH_CAP_AUTH_EXCL) { 2095 inode->i_mode = attr->ia_mode; 2096 dirtied |= CEPH_CAP_AUTH_EXCL; 2097 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2098 attr->ia_mode != inode->i_mode) { 2099 inode->i_mode = attr->ia_mode; 2100 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 2101 mask |= CEPH_SETATTR_MODE; 2102 release |= CEPH_CAP_AUTH_SHARED; 2103 } 2104 } 2105 2106 if (ia_valid & ATTR_ATIME) { 2107 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, 2108 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 2109 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 2110 if (issued & CEPH_CAP_FILE_EXCL) { 2111 ci->i_time_warp_seq++; 2112 inode->i_atime = attr->ia_atime; 2113 dirtied |= CEPH_CAP_FILE_EXCL; 2114 } else if ((issued & CEPH_CAP_FILE_WR) && 2115 timespec64_compare(&inode->i_atime, 2116 &attr->ia_atime) < 0) { 2117 inode->i_atime = attr->ia_atime; 2118 dirtied |= CEPH_CAP_FILE_WR; 2119 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2120 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { 2121 ceph_encode_timespec64(&req->r_args.setattr.atime, 2122 &attr->ia_atime); 2123 mask |= CEPH_SETATTR_ATIME; 2124 release |= CEPH_CAP_FILE_SHARED | 2125 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2126 } 2127 } 2128 if (ia_valid & ATTR_SIZE) { 2129 dout("setattr %p size %lld -> %lld\n", inode, 2130 inode->i_size, attr->ia_size); 2131 if ((issued & CEPH_CAP_FILE_EXCL) && 2132 attr->ia_size > inode->i_size) { 2133 i_size_write(inode, attr->ia_size); 2134 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2135 ci->i_reported_size = attr->ia_size; 2136 dirtied |= CEPH_CAP_FILE_EXCL; 2137 ia_valid |= ATTR_MTIME; 2138 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2139 attr->ia_size != inode->i_size) { 2140 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2141 req->r_args.setattr.old_size = 2142 cpu_to_le64(inode->i_size); 2143 mask |= CEPH_SETATTR_SIZE; 2144 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | 2145 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2146 } 2147 } 2148 if (ia_valid & ATTR_MTIME) { 2149 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, 2150 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 2151 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 2152 if (issued & CEPH_CAP_FILE_EXCL) { 2153 ci->i_time_warp_seq++; 2154 inode->i_mtime = attr->ia_mtime; 2155 dirtied |= CEPH_CAP_FILE_EXCL; 2156 } else if ((issued & CEPH_CAP_FILE_WR) && 2157 timespec64_compare(&inode->i_mtime, 2158 &attr->ia_mtime) < 0) { 2159 inode->i_mtime = attr->ia_mtime; 2160 dirtied |= CEPH_CAP_FILE_WR; 2161 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2162 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { 2163 ceph_encode_timespec64(&req->r_args.setattr.mtime, 2164 &attr->ia_mtime); 2165 mask |= CEPH_SETATTR_MTIME; 2166 release |= CEPH_CAP_FILE_SHARED | 2167 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2168 } 2169 } 2170 2171 /* these do nothing */ 2172 if (ia_valid & ATTR_CTIME) { 2173 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2174 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2175 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, 2176 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2177 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2178 only ? "ctime only" : "ignored"); 2179 if (only) { 2180 /* 2181 * if kernel wants to dirty ctime but nothing else, 2182 * we need to choose a cap to dirty under, or do 2183 * a almost-no-op setattr 2184 */ 2185 if (issued & CEPH_CAP_AUTH_EXCL) 2186 dirtied |= CEPH_CAP_AUTH_EXCL; 2187 else if (issued & CEPH_CAP_FILE_EXCL) 2188 dirtied |= CEPH_CAP_FILE_EXCL; 2189 else if (issued & CEPH_CAP_XATTR_EXCL) 2190 dirtied |= CEPH_CAP_XATTR_EXCL; 2191 else 2192 mask |= CEPH_SETATTR_CTIME; 2193 } 2194 } 2195 if (ia_valid & ATTR_FILE) 2196 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2197 2198 if (dirtied) { 2199 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2200 &prealloc_cf); 2201 inode->i_ctime = attr->ia_ctime; 2202 } 2203 2204 release &= issued; 2205 spin_unlock(&ci->i_ceph_lock); 2206 if (lock_snap_rwsem) 2207 up_read(&mdsc->snap_rwsem); 2208 2209 if (inode_dirty_flags) 2210 __mark_inode_dirty(inode, inode_dirty_flags); 2211 2212 2213 if (mask) { 2214 req->r_inode = inode; 2215 ihold(inode); 2216 req->r_inode_drop = release; 2217 req->r_args.setattr.mask = cpu_to_le32(mask); 2218 req->r_num_caps = 1; 2219 req->r_stamp = attr->ia_ctime; 2220 err = ceph_mdsc_do_request(mdsc, NULL, req); 2221 } 2222 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2223 ceph_cap_string(dirtied), mask); 2224 2225 ceph_mdsc_put_request(req); 2226 ceph_free_cap_flush(prealloc_cf); 2227 2228 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2229 __ceph_do_pending_vmtruncate(inode); 2230 2231 return err; 2232 } 2233 2234 /* 2235 * setattr 2236 */ 2237 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 2238 { 2239 struct inode *inode = d_inode(dentry); 2240 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 2241 int err; 2242 2243 if (ceph_snap(inode) != CEPH_NOSNAP) 2244 return -EROFS; 2245 2246 err = setattr_prepare(dentry, attr); 2247 if (err != 0) 2248 return err; 2249 2250 if ((attr->ia_valid & ATTR_SIZE) && 2251 attr->ia_size > max(inode->i_size, fsc->max_file_size)) 2252 return -EFBIG; 2253 2254 if ((attr->ia_valid & ATTR_SIZE) && 2255 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) 2256 return -EDQUOT; 2257 2258 err = __ceph_setattr(inode, attr); 2259 2260 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2261 err = posix_acl_chmod(inode, attr->ia_mode); 2262 2263 return err; 2264 } 2265 2266 /* 2267 * Verify that we have a lease on the given mask. If not, 2268 * do a getattr against an mds. 2269 */ 2270 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2271 int mask, bool force) 2272 { 2273 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2274 struct ceph_mds_client *mdsc = fsc->mdsc; 2275 struct ceph_mds_request *req; 2276 int mode; 2277 int err; 2278 2279 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2280 dout("do_getattr inode %p SNAPDIR\n", inode); 2281 return 0; 2282 } 2283 2284 dout("do_getattr inode %p mask %s mode 0%o\n", 2285 inode, ceph_cap_string(mask), inode->i_mode); 2286 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 2287 return 0; 2288 2289 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; 2290 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2291 if (IS_ERR(req)) 2292 return PTR_ERR(req); 2293 req->r_inode = inode; 2294 ihold(inode); 2295 req->r_num_caps = 1; 2296 req->r_args.getattr.mask = cpu_to_le32(mask); 2297 req->r_locked_page = locked_page; 2298 err = ceph_mdsc_do_request(mdsc, NULL, req); 2299 if (locked_page && err == 0) { 2300 u64 inline_version = req->r_reply_info.targeti.inline_version; 2301 if (inline_version == 0) { 2302 /* the reply is supposed to contain inline data */ 2303 err = -EINVAL; 2304 } else if (inline_version == CEPH_INLINE_NONE) { 2305 err = -ENODATA; 2306 } else { 2307 err = req->r_reply_info.targeti.inline_len; 2308 } 2309 } 2310 ceph_mdsc_put_request(req); 2311 dout("do_getattr result=%d\n", err); 2312 return err; 2313 } 2314 2315 2316 /* 2317 * Check inode permissions. We verify we have a valid value for 2318 * the AUTH cap, then call the generic handler. 2319 */ 2320 int ceph_permission(struct inode *inode, int mask) 2321 { 2322 int err; 2323 2324 if (mask & MAY_NOT_BLOCK) 2325 return -ECHILD; 2326 2327 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2328 2329 if (!err) 2330 err = generic_permission(inode, mask); 2331 return err; 2332 } 2333 2334 /* Craft a mask of needed caps given a set of requested statx attrs. */ 2335 static int statx_to_caps(u32 want) 2336 { 2337 int mask = 0; 2338 2339 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) 2340 mask |= CEPH_CAP_AUTH_SHARED; 2341 2342 if (want & (STATX_NLINK|STATX_CTIME)) 2343 mask |= CEPH_CAP_LINK_SHARED; 2344 2345 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| 2346 STATX_BLOCKS)) 2347 mask |= CEPH_CAP_FILE_SHARED; 2348 2349 if (want & (STATX_CTIME)) 2350 mask |= CEPH_CAP_XATTR_SHARED; 2351 2352 return mask; 2353 } 2354 2355 /* 2356 * Get all the attributes. If we have sufficient caps for the requested attrs, 2357 * then we can avoid talking to the MDS at all. 2358 */ 2359 int ceph_getattr(const struct path *path, struct kstat *stat, 2360 u32 request_mask, unsigned int flags) 2361 { 2362 struct inode *inode = d_inode(path->dentry); 2363 struct ceph_inode_info *ci = ceph_inode(inode); 2364 u32 valid_mask = STATX_BASIC_STATS; 2365 int err = 0; 2366 2367 /* Skip the getattr altogether if we're asked not to sync */ 2368 if (!(flags & AT_STATX_DONT_SYNC)) { 2369 err = ceph_do_getattr(inode, statx_to_caps(request_mask), 2370 flags & AT_STATX_FORCE_SYNC); 2371 if (err) 2372 return err; 2373 } 2374 2375 generic_fillattr(inode, stat); 2376 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 2377 2378 /* 2379 * btime on newly-allocated inodes is 0, so if this is still set to 2380 * that, then assume that it's not valid. 2381 */ 2382 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { 2383 stat->btime = ci->i_btime; 2384 valid_mask |= STATX_BTIME; 2385 } 2386 2387 if (ceph_snap(inode) == CEPH_NOSNAP) 2388 stat->dev = inode->i_sb->s_dev; 2389 else 2390 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; 2391 2392 if (S_ISDIR(inode->i_mode)) { 2393 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2394 RBYTES)) 2395 stat->size = ci->i_rbytes; 2396 else 2397 stat->size = ci->i_files + ci->i_subdirs; 2398 stat->blocks = 0; 2399 stat->blksize = 65536; 2400 /* 2401 * Some applications rely on the number of st_nlink 2402 * value on directories to be either 0 (if unlinked) 2403 * or 2 + number of subdirectories. 2404 */ 2405 if (stat->nlink == 1) 2406 /* '.' + '..' + subdirs */ 2407 stat->nlink = 1 + 1 + ci->i_subdirs; 2408 } 2409 2410 stat->result_mask = request_mask & valid_mask; 2411 return err; 2412 } 2413