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 struct ceph_inode_info *ci = ceph_inode(inode); 45 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 46 47 ci->i_vino = *(struct ceph_vino *)data; 48 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino); 49 inode_set_iversion_raw(inode, 0); 50 percpu_counter_inc(&mdsc->metric.total_inodes); 51 52 return 0; 53 } 54 55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 56 { 57 struct inode *inode; 58 59 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare, 60 ceph_set_ino_cb, &vino); 61 if (!inode) 62 return ERR_PTR(-ENOMEM); 63 64 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode), 65 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW)); 66 return inode; 67 } 68 69 /* 70 * get/constuct snapdir inode for a given directory 71 */ 72 struct inode *ceph_get_snapdir(struct inode *parent) 73 { 74 struct ceph_vino vino = { 75 .ino = ceph_ino(parent), 76 .snap = CEPH_SNAPDIR, 77 }; 78 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 79 struct ceph_inode_info *ci = ceph_inode(inode); 80 81 BUG_ON(!S_ISDIR(parent->i_mode)); 82 if (IS_ERR(inode)) 83 return inode; 84 inode->i_mode = parent->i_mode; 85 inode->i_uid = parent->i_uid; 86 inode->i_gid = parent->i_gid; 87 inode->i_mtime = parent->i_mtime; 88 inode->i_ctime = parent->i_ctime; 89 inode->i_atime = parent->i_atime; 90 ci->i_rbytes = 0; 91 ci->i_btime = ceph_inode(parent)->i_btime; 92 93 if (inode->i_state & I_NEW) { 94 inode->i_op = &ceph_snapdir_iops; 95 inode->i_fop = &ceph_snapdir_fops; 96 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 97 unlock_new_inode(inode); 98 } 99 100 return inode; 101 } 102 103 const struct inode_operations ceph_file_iops = { 104 .permission = ceph_permission, 105 .setattr = ceph_setattr, 106 .getattr = ceph_getattr, 107 .listxattr = ceph_listxattr, 108 .get_acl = ceph_get_acl, 109 .set_acl = ceph_set_acl, 110 }; 111 112 113 /* 114 * We use a 'frag tree' to keep track of the MDS's directory fragments 115 * for a given inode (usually there is just a single fragment). We 116 * need to know when a child frag is delegated to a new MDS, or when 117 * it is flagged as replicated, so we can direct our requests 118 * accordingly. 119 */ 120 121 /* 122 * find/create a frag in the tree 123 */ 124 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 125 u32 f) 126 { 127 struct rb_node **p; 128 struct rb_node *parent = NULL; 129 struct ceph_inode_frag *frag; 130 int c; 131 132 p = &ci->i_fragtree.rb_node; 133 while (*p) { 134 parent = *p; 135 frag = rb_entry(parent, struct ceph_inode_frag, node); 136 c = ceph_frag_compare(f, frag->frag); 137 if (c < 0) 138 p = &(*p)->rb_left; 139 else if (c > 0) 140 p = &(*p)->rb_right; 141 else 142 return frag; 143 } 144 145 frag = kmalloc(sizeof(*frag), GFP_NOFS); 146 if (!frag) 147 return ERR_PTR(-ENOMEM); 148 149 frag->frag = f; 150 frag->split_by = 0; 151 frag->mds = -1; 152 frag->ndist = 0; 153 154 rb_link_node(&frag->node, parent, p); 155 rb_insert_color(&frag->node, &ci->i_fragtree); 156 157 dout("get_or_create_frag added %llx.%llx frag %x\n", 158 ceph_vinop(&ci->vfs_inode), f); 159 return frag; 160 } 161 162 /* 163 * find a specific frag @f 164 */ 165 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 166 { 167 struct rb_node *n = ci->i_fragtree.rb_node; 168 169 while (n) { 170 struct ceph_inode_frag *frag = 171 rb_entry(n, struct ceph_inode_frag, node); 172 int c = ceph_frag_compare(f, frag->frag); 173 if (c < 0) 174 n = n->rb_left; 175 else if (c > 0) 176 n = n->rb_right; 177 else 178 return frag; 179 } 180 return NULL; 181 } 182 183 /* 184 * Choose frag containing the given value @v. If @pfrag is 185 * specified, copy the frag delegation info to the caller if 186 * it is present. 187 */ 188 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 189 struct ceph_inode_frag *pfrag, int *found) 190 { 191 u32 t = ceph_frag_make(0, 0); 192 struct ceph_inode_frag *frag; 193 unsigned nway, i; 194 u32 n; 195 196 if (found) 197 *found = 0; 198 199 while (1) { 200 WARN_ON(!ceph_frag_contains_value(t, v)); 201 frag = __ceph_find_frag(ci, t); 202 if (!frag) 203 break; /* t is a leaf */ 204 if (frag->split_by == 0) { 205 if (pfrag) 206 memcpy(pfrag, frag, sizeof(*pfrag)); 207 if (found) 208 *found = 1; 209 break; 210 } 211 212 /* choose child */ 213 nway = 1 << frag->split_by; 214 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 215 frag->split_by, nway); 216 for (i = 0; i < nway; i++) { 217 n = ceph_frag_make_child(t, frag->split_by, i); 218 if (ceph_frag_contains_value(n, v)) { 219 t = n; 220 break; 221 } 222 } 223 BUG_ON(i == nway); 224 } 225 dout("choose_frag(%x) = %x\n", v, t); 226 227 return t; 228 } 229 230 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 231 struct ceph_inode_frag *pfrag, int *found) 232 { 233 u32 ret; 234 mutex_lock(&ci->i_fragtree_mutex); 235 ret = __ceph_choose_frag(ci, v, pfrag, found); 236 mutex_unlock(&ci->i_fragtree_mutex); 237 return ret; 238 } 239 240 /* 241 * Process dirfrag (delegation) info from the mds. Include leaf 242 * fragment in tree ONLY if ndist > 0. Otherwise, only 243 * branches/splits are included in i_fragtree) 244 */ 245 static int ceph_fill_dirfrag(struct inode *inode, 246 struct ceph_mds_reply_dirfrag *dirinfo) 247 { 248 struct ceph_inode_info *ci = ceph_inode(inode); 249 struct ceph_inode_frag *frag; 250 u32 id = le32_to_cpu(dirinfo->frag); 251 int mds = le32_to_cpu(dirinfo->auth); 252 int ndist = le32_to_cpu(dirinfo->ndist); 253 int diri_auth = -1; 254 int i; 255 int err = 0; 256 257 spin_lock(&ci->i_ceph_lock); 258 if (ci->i_auth_cap) 259 diri_auth = ci->i_auth_cap->mds; 260 spin_unlock(&ci->i_ceph_lock); 261 262 if (mds == -1) /* CDIR_AUTH_PARENT */ 263 mds = diri_auth; 264 265 mutex_lock(&ci->i_fragtree_mutex); 266 if (ndist == 0 && mds == diri_auth) { 267 /* no delegation info needed. */ 268 frag = __ceph_find_frag(ci, id); 269 if (!frag) 270 goto out; 271 if (frag->split_by == 0) { 272 /* tree leaf, remove */ 273 dout("fill_dirfrag removed %llx.%llx frag %x" 274 " (no ref)\n", ceph_vinop(inode), id); 275 rb_erase(&frag->node, &ci->i_fragtree); 276 kfree(frag); 277 } else { 278 /* tree branch, keep and clear */ 279 dout("fill_dirfrag cleared %llx.%llx frag %x" 280 " referral\n", ceph_vinop(inode), id); 281 frag->mds = -1; 282 frag->ndist = 0; 283 } 284 goto out; 285 } 286 287 288 /* find/add this frag to store mds delegation info */ 289 frag = __get_or_create_frag(ci, id); 290 if (IS_ERR(frag)) { 291 /* this is not the end of the world; we can continue 292 with bad/inaccurate delegation info */ 293 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 294 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 295 err = -ENOMEM; 296 goto out; 297 } 298 299 frag->mds = mds; 300 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 301 for (i = 0; i < frag->ndist; i++) 302 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 303 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 304 ceph_vinop(inode), frag->frag, frag->ndist); 305 306 out: 307 mutex_unlock(&ci->i_fragtree_mutex); 308 return err; 309 } 310 311 static int frag_tree_split_cmp(const void *l, const void *r) 312 { 313 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; 314 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; 315 return ceph_frag_compare(le32_to_cpu(ls->frag), 316 le32_to_cpu(rs->frag)); 317 } 318 319 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) 320 { 321 if (!frag) 322 return f == ceph_frag_make(0, 0); 323 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) 324 return false; 325 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); 326 } 327 328 static int ceph_fill_fragtree(struct inode *inode, 329 struct ceph_frag_tree_head *fragtree, 330 struct ceph_mds_reply_dirfrag *dirinfo) 331 { 332 struct ceph_inode_info *ci = ceph_inode(inode); 333 struct ceph_inode_frag *frag, *prev_frag = NULL; 334 struct rb_node *rb_node; 335 unsigned i, split_by, nsplits; 336 u32 id; 337 bool update = false; 338 339 mutex_lock(&ci->i_fragtree_mutex); 340 nsplits = le32_to_cpu(fragtree->nsplits); 341 if (nsplits != ci->i_fragtree_nsplits) { 342 update = true; 343 } else if (nsplits) { 344 i = prandom_u32() % nsplits; 345 id = le32_to_cpu(fragtree->splits[i].frag); 346 if (!__ceph_find_frag(ci, id)) 347 update = true; 348 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 349 rb_node = rb_first(&ci->i_fragtree); 350 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 351 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 352 update = true; 353 } 354 if (!update && dirinfo) { 355 id = le32_to_cpu(dirinfo->frag); 356 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 357 update = true; 358 } 359 if (!update) 360 goto out_unlock; 361 362 if (nsplits > 1) { 363 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), 364 frag_tree_split_cmp, NULL); 365 } 366 367 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 368 rb_node = rb_first(&ci->i_fragtree); 369 for (i = 0; i < nsplits; i++) { 370 id = le32_to_cpu(fragtree->splits[i].frag); 371 split_by = le32_to_cpu(fragtree->splits[i].by); 372 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { 373 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " 374 "frag %x split by %d\n", ceph_vinop(inode), 375 i, nsplits, id, split_by); 376 continue; 377 } 378 frag = NULL; 379 while (rb_node) { 380 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 381 if (ceph_frag_compare(frag->frag, id) >= 0) { 382 if (frag->frag != id) 383 frag = NULL; 384 else 385 rb_node = rb_next(rb_node); 386 break; 387 } 388 rb_node = rb_next(rb_node); 389 /* delete stale split/leaf node */ 390 if (frag->split_by > 0 || 391 !is_frag_child(frag->frag, prev_frag)) { 392 rb_erase(&frag->node, &ci->i_fragtree); 393 if (frag->split_by > 0) 394 ci->i_fragtree_nsplits--; 395 kfree(frag); 396 } 397 frag = NULL; 398 } 399 if (!frag) { 400 frag = __get_or_create_frag(ci, id); 401 if (IS_ERR(frag)) 402 continue; 403 } 404 if (frag->split_by == 0) 405 ci->i_fragtree_nsplits++; 406 frag->split_by = split_by; 407 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 408 prev_frag = frag; 409 } 410 while (rb_node) { 411 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 412 rb_node = rb_next(rb_node); 413 /* delete stale split/leaf node */ 414 if (frag->split_by > 0 || 415 !is_frag_child(frag->frag, prev_frag)) { 416 rb_erase(&frag->node, &ci->i_fragtree); 417 if (frag->split_by > 0) 418 ci->i_fragtree_nsplits--; 419 kfree(frag); 420 } 421 } 422 out_unlock: 423 mutex_unlock(&ci->i_fragtree_mutex); 424 return 0; 425 } 426 427 /* 428 * initialize a newly allocated inode. 429 */ 430 struct inode *ceph_alloc_inode(struct super_block *sb) 431 { 432 struct ceph_inode_info *ci; 433 int i; 434 435 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 436 if (!ci) 437 return NULL; 438 439 dout("alloc_inode %p\n", &ci->vfs_inode); 440 441 spin_lock_init(&ci->i_ceph_lock); 442 443 ci->i_version = 0; 444 ci->i_inline_version = 0; 445 ci->i_time_warp_seq = 0; 446 ci->i_ceph_flags = 0; 447 atomic64_set(&ci->i_ordered_count, 1); 448 atomic64_set(&ci->i_release_count, 1); 449 atomic64_set(&ci->i_complete_seq[0], 0); 450 atomic64_set(&ci->i_complete_seq[1], 0); 451 ci->i_symlink = NULL; 452 453 ci->i_max_bytes = 0; 454 ci->i_max_files = 0; 455 456 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 457 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); 458 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); 459 460 ci->i_fragtree = RB_ROOT; 461 mutex_init(&ci->i_fragtree_mutex); 462 463 ci->i_xattrs.blob = NULL; 464 ci->i_xattrs.prealloc_blob = NULL; 465 ci->i_xattrs.dirty = false; 466 ci->i_xattrs.index = RB_ROOT; 467 ci->i_xattrs.count = 0; 468 ci->i_xattrs.names_size = 0; 469 ci->i_xattrs.vals_size = 0; 470 ci->i_xattrs.version = 0; 471 ci->i_xattrs.index_version = 0; 472 473 ci->i_caps = RB_ROOT; 474 ci->i_auth_cap = NULL; 475 ci->i_dirty_caps = 0; 476 ci->i_flushing_caps = 0; 477 INIT_LIST_HEAD(&ci->i_dirty_item); 478 INIT_LIST_HEAD(&ci->i_flushing_item); 479 ci->i_prealloc_cap_flush = NULL; 480 INIT_LIST_HEAD(&ci->i_cap_flush_list); 481 init_waitqueue_head(&ci->i_cap_wq); 482 ci->i_hold_caps_max = 0; 483 INIT_LIST_HEAD(&ci->i_cap_delay_list); 484 INIT_LIST_HEAD(&ci->i_cap_snaps); 485 ci->i_head_snapc = NULL; 486 ci->i_snap_caps = 0; 487 488 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ; 489 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) 490 ci->i_nr_by_mode[i] = 0; 491 492 mutex_init(&ci->i_truncate_mutex); 493 ci->i_truncate_seq = 0; 494 ci->i_truncate_size = 0; 495 ci->i_truncate_pending = 0; 496 497 ci->i_max_size = 0; 498 ci->i_reported_size = 0; 499 ci->i_wanted_max_size = 0; 500 ci->i_requested_max_size = 0; 501 502 ci->i_pin_ref = 0; 503 ci->i_rd_ref = 0; 504 ci->i_rdcache_ref = 0; 505 ci->i_wr_ref = 0; 506 ci->i_wb_ref = 0; 507 ci->i_fx_ref = 0; 508 ci->i_wrbuffer_ref = 0; 509 ci->i_wrbuffer_ref_head = 0; 510 atomic_set(&ci->i_filelock_ref, 0); 511 atomic_set(&ci->i_shared_gen, 1); 512 ci->i_rdcache_gen = 0; 513 ci->i_rdcache_revoking = 0; 514 515 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 516 INIT_LIST_HEAD(&ci->i_unsafe_iops); 517 spin_lock_init(&ci->i_unsafe_lock); 518 519 ci->i_snap_realm = NULL; 520 INIT_LIST_HEAD(&ci->i_snap_realm_item); 521 INIT_LIST_HEAD(&ci->i_snap_flush_item); 522 523 INIT_WORK(&ci->i_work, ceph_inode_work); 524 ci->i_work_mask = 0; 525 memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); 526 527 ceph_fscache_inode_init(ci); 528 529 ci->i_meta_err = 0; 530 531 return &ci->vfs_inode; 532 } 533 534 void ceph_free_inode(struct inode *inode) 535 { 536 struct ceph_inode_info *ci = ceph_inode(inode); 537 538 kfree(ci->i_symlink); 539 kmem_cache_free(ceph_inode_cachep, ci); 540 } 541 542 void ceph_evict_inode(struct inode *inode) 543 { 544 struct ceph_inode_info *ci = ceph_inode(inode); 545 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 546 struct ceph_inode_frag *frag; 547 struct rb_node *n; 548 549 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 550 551 percpu_counter_dec(&mdsc->metric.total_inodes); 552 553 truncate_inode_pages_final(&inode->i_data); 554 clear_inode(inode); 555 556 ceph_fscache_unregister_inode_cookie(ci); 557 558 __ceph_remove_caps(ci); 559 560 if (__ceph_has_any_quota(ci)) 561 ceph_adjust_quota_realms_count(inode, false); 562 563 /* 564 * we may still have a snap_realm reference if there are stray 565 * caps in i_snap_caps. 566 */ 567 if (ci->i_snap_realm) { 568 if (ceph_snap(inode) == CEPH_NOSNAP) { 569 struct ceph_snap_realm *realm = ci->i_snap_realm; 570 dout(" dropping residual ref to snap realm %p\n", 571 realm); 572 spin_lock(&realm->inodes_with_caps_lock); 573 list_del_init(&ci->i_snap_realm_item); 574 ci->i_snap_realm = NULL; 575 if (realm->ino == ci->i_vino.ino) 576 realm->inode = NULL; 577 spin_unlock(&realm->inodes_with_caps_lock); 578 ceph_put_snap_realm(mdsc, realm); 579 } else { 580 ceph_put_snapid_map(mdsc, ci->i_snapid_map); 581 ci->i_snap_realm = NULL; 582 } 583 } 584 585 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 586 frag = rb_entry(n, struct ceph_inode_frag, node); 587 rb_erase(n, &ci->i_fragtree); 588 kfree(frag); 589 } 590 ci->i_fragtree_nsplits = 0; 591 592 __ceph_destroy_xattrs(ci); 593 if (ci->i_xattrs.blob) 594 ceph_buffer_put(ci->i_xattrs.blob); 595 if (ci->i_xattrs.prealloc_blob) 596 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 597 598 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); 599 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); 600 } 601 602 static inline blkcnt_t calc_inode_blocks(u64 size) 603 { 604 return (size + (1<<9) - 1) >> 9; 605 } 606 607 /* 608 * Helpers to fill in size, ctime, mtime, and atime. We have to be 609 * careful because either the client or MDS may have more up to date 610 * info, depending on which capabilities are held, and whether 611 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 612 * and size are monotonically increasing, except when utimes() or 613 * truncate() increments the corresponding _seq values.) 614 */ 615 int ceph_fill_file_size(struct inode *inode, int issued, 616 u32 truncate_seq, u64 truncate_size, u64 size) 617 { 618 struct ceph_inode_info *ci = ceph_inode(inode); 619 int queue_trunc = 0; 620 621 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 622 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 623 dout("size %lld -> %llu\n", inode->i_size, size); 624 if (size > 0 && S_ISDIR(inode->i_mode)) { 625 pr_err("fill_file_size non-zero size for directory\n"); 626 size = 0; 627 } 628 i_size_write(inode, size); 629 inode->i_blocks = calc_inode_blocks(size); 630 ci->i_reported_size = size; 631 if (truncate_seq != ci->i_truncate_seq) { 632 dout("truncate_seq %u -> %u\n", 633 ci->i_truncate_seq, truncate_seq); 634 ci->i_truncate_seq = truncate_seq; 635 636 /* the MDS should have revoked these caps */ 637 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 638 CEPH_CAP_FILE_RD | 639 CEPH_CAP_FILE_WR | 640 CEPH_CAP_FILE_LAZYIO)); 641 /* 642 * If we hold relevant caps, or in the case where we're 643 * not the only client referencing this file and we 644 * don't hold those caps, then we need to check whether 645 * the file is either opened or mmaped 646 */ 647 if ((issued & (CEPH_CAP_FILE_CACHE| 648 CEPH_CAP_FILE_BUFFER)) || 649 mapping_mapped(inode->i_mapping) || 650 __ceph_is_file_opened(ci)) { 651 ci->i_truncate_pending++; 652 queue_trunc = 1; 653 } 654 } 655 } 656 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 657 ci->i_truncate_size != truncate_size) { 658 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 659 truncate_size); 660 ci->i_truncate_size = truncate_size; 661 } 662 663 if (queue_trunc) 664 ceph_fscache_invalidate(inode); 665 666 return queue_trunc; 667 } 668 669 void ceph_fill_file_time(struct inode *inode, int issued, 670 u64 time_warp_seq, struct timespec64 *ctime, 671 struct timespec64 *mtime, struct timespec64 *atime) 672 { 673 struct ceph_inode_info *ci = ceph_inode(inode); 674 int warn = 0; 675 676 if (issued & (CEPH_CAP_FILE_EXCL| 677 CEPH_CAP_FILE_WR| 678 CEPH_CAP_FILE_BUFFER| 679 CEPH_CAP_AUTH_EXCL| 680 CEPH_CAP_XATTR_EXCL)) { 681 if (ci->i_version == 0 || 682 timespec64_compare(ctime, &inode->i_ctime) > 0) { 683 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", 684 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 685 ctime->tv_sec, ctime->tv_nsec); 686 inode->i_ctime = *ctime; 687 } 688 if (ci->i_version == 0 || 689 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 690 /* the MDS did a utimes() */ 691 dout("mtime %lld.%09ld -> %lld.%09ld " 692 "tw %d -> %d\n", 693 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 694 mtime->tv_sec, mtime->tv_nsec, 695 ci->i_time_warp_seq, (int)time_warp_seq); 696 697 inode->i_mtime = *mtime; 698 inode->i_atime = *atime; 699 ci->i_time_warp_seq = time_warp_seq; 700 } else if (time_warp_seq == ci->i_time_warp_seq) { 701 /* nobody did utimes(); take the max */ 702 if (timespec64_compare(mtime, &inode->i_mtime) > 0) { 703 dout("mtime %lld.%09ld -> %lld.%09ld inc\n", 704 inode->i_mtime.tv_sec, 705 inode->i_mtime.tv_nsec, 706 mtime->tv_sec, mtime->tv_nsec); 707 inode->i_mtime = *mtime; 708 } 709 if (timespec64_compare(atime, &inode->i_atime) > 0) { 710 dout("atime %lld.%09ld -> %lld.%09ld inc\n", 711 inode->i_atime.tv_sec, 712 inode->i_atime.tv_nsec, 713 atime->tv_sec, atime->tv_nsec); 714 inode->i_atime = *atime; 715 } 716 } else if (issued & CEPH_CAP_FILE_EXCL) { 717 /* we did a utimes(); ignore mds values */ 718 } else { 719 warn = 1; 720 } 721 } else { 722 /* we have no write|excl caps; whatever the MDS says is true */ 723 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 724 inode->i_ctime = *ctime; 725 inode->i_mtime = *mtime; 726 inode->i_atime = *atime; 727 ci->i_time_warp_seq = time_warp_seq; 728 } else { 729 warn = 1; 730 } 731 } 732 if (warn) /* time_warp_seq shouldn't go backwards */ 733 dout("%p mds time_warp_seq %llu < %u\n", 734 inode, time_warp_seq, ci->i_time_warp_seq); 735 } 736 737 /* 738 * Populate an inode based on info from mds. May be called on new or 739 * existing inodes. 740 */ 741 int ceph_fill_inode(struct inode *inode, struct page *locked_page, 742 struct ceph_mds_reply_info_in *iinfo, 743 struct ceph_mds_reply_dirfrag *dirinfo, 744 struct ceph_mds_session *session, int cap_fmode, 745 struct ceph_cap_reservation *caps_reservation) 746 { 747 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 748 struct ceph_mds_reply_inode *info = iinfo->in; 749 struct ceph_inode_info *ci = ceph_inode(inode); 750 int issued, new_issued, info_caps; 751 struct timespec64 mtime, atime, ctime; 752 struct ceph_buffer *xattr_blob = NULL; 753 struct ceph_buffer *old_blob = NULL; 754 struct ceph_string *pool_ns = NULL; 755 struct ceph_cap *new_cap = NULL; 756 int err = 0; 757 bool wake = false; 758 bool queue_trunc = false; 759 bool new_version = false; 760 bool fill_inline = false; 761 762 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, 763 inode, ceph_vinop(inode), le64_to_cpu(info->version), 764 ci->i_version); 765 766 info_caps = le32_to_cpu(info->cap.caps); 767 768 /* prealloc new cap struct */ 769 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { 770 new_cap = ceph_get_cap(mdsc, caps_reservation); 771 if (!new_cap) 772 return -ENOMEM; 773 } 774 775 /* 776 * prealloc xattr data, if it looks like we'll need it. only 777 * if len > 4 (meaning there are actually xattrs; the first 4 778 * bytes are the xattr count). 779 */ 780 if (iinfo->xattr_len > 4) { 781 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 782 if (!xattr_blob) 783 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, 784 iinfo->xattr_len); 785 } 786 787 if (iinfo->pool_ns_len > 0) 788 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, 789 iinfo->pool_ns_len); 790 791 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) 792 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); 793 794 spin_lock(&ci->i_ceph_lock); 795 796 /* 797 * provided version will be odd if inode value is projected, 798 * even if stable. skip the update if we have newer stable 799 * info (ours>=theirs, e.g. due to racing mds replies), unless 800 * we are getting projected (unstable) info (in which case the 801 * version is odd, and we want ours>theirs). 802 * us them 803 * 2 2 skip 804 * 3 2 skip 805 * 3 3 update 806 */ 807 if (ci->i_version == 0 || 808 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 809 le64_to_cpu(info->version) > (ci->i_version & ~1))) 810 new_version = true; 811 812 /* Update change_attribute */ 813 inode_set_max_iversion_raw(inode, iinfo->change_attr); 814 815 __ceph_caps_issued(ci, &issued); 816 issued |= __ceph_caps_dirty(ci); 817 new_issued = ~issued & info_caps; 818 819 /* update inode */ 820 inode->i_rdev = le32_to_cpu(info->rdev); 821 /* directories have fl_stripe_unit set to zero */ 822 if (le32_to_cpu(info->layout.fl_stripe_unit)) 823 inode->i_blkbits = 824 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 825 else 826 inode->i_blkbits = CEPH_BLOCK_SHIFT; 827 828 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); 829 830 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 831 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 832 inode->i_mode = le32_to_cpu(info->mode); 833 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 834 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 835 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 836 from_kuid(&init_user_ns, inode->i_uid), 837 from_kgid(&init_user_ns, inode->i_gid)); 838 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); 839 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); 840 } 841 842 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 843 (issued & CEPH_CAP_LINK_EXCL) == 0) 844 set_nlink(inode, le32_to_cpu(info->nlink)); 845 846 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 847 /* be careful with mtime, atime, size */ 848 ceph_decode_timespec64(&atime, &info->atime); 849 ceph_decode_timespec64(&mtime, &info->mtime); 850 ceph_decode_timespec64(&ctime, &info->ctime); 851 ceph_fill_file_time(inode, issued, 852 le32_to_cpu(info->time_warp_seq), 853 &ctime, &mtime, &atime); 854 } 855 856 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { 857 ci->i_files = le64_to_cpu(info->files); 858 ci->i_subdirs = le64_to_cpu(info->subdirs); 859 } 860 861 if (new_version || 862 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 863 s64 old_pool = ci->i_layout.pool_id; 864 struct ceph_string *old_ns; 865 866 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); 867 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, 868 lockdep_is_held(&ci->i_ceph_lock)); 869 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); 870 871 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) 872 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; 873 874 pool_ns = old_ns; 875 876 queue_trunc = ceph_fill_file_size(inode, issued, 877 le32_to_cpu(info->truncate_seq), 878 le64_to_cpu(info->truncate_size), 879 le64_to_cpu(info->size)); 880 /* only update max_size on auth cap */ 881 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 882 ci->i_max_size != le64_to_cpu(info->max_size)) { 883 dout("max_size %lld -> %llu\n", ci->i_max_size, 884 le64_to_cpu(info->max_size)); 885 ci->i_max_size = le64_to_cpu(info->max_size); 886 } 887 } 888 889 /* layout and rstat are not tracked by capability, update them if 890 * the inode info is from auth mds */ 891 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { 892 if (S_ISDIR(inode->i_mode)) { 893 ci->i_dir_layout = iinfo->dir_layout; 894 ci->i_rbytes = le64_to_cpu(info->rbytes); 895 ci->i_rfiles = le64_to_cpu(info->rfiles); 896 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 897 ci->i_dir_pin = iinfo->dir_pin; 898 ci->i_rsnaps = iinfo->rsnaps; 899 ceph_decode_timespec64(&ci->i_rctime, &info->rctime); 900 } 901 } 902 903 /* xattrs */ 904 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 905 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 906 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 907 if (ci->i_xattrs.blob) 908 old_blob = ci->i_xattrs.blob; 909 ci->i_xattrs.blob = xattr_blob; 910 if (xattr_blob) 911 memcpy(ci->i_xattrs.blob->vec.iov_base, 912 iinfo->xattr_data, iinfo->xattr_len); 913 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 914 ceph_forget_all_cached_acls(inode); 915 ceph_security_invalidate_secctx(inode); 916 xattr_blob = NULL; 917 } 918 919 /* finally update i_version */ 920 if (le64_to_cpu(info->version) > ci->i_version) 921 ci->i_version = le64_to_cpu(info->version); 922 923 inode->i_mapping->a_ops = &ceph_aops; 924 925 switch (inode->i_mode & S_IFMT) { 926 case S_IFIFO: 927 case S_IFBLK: 928 case S_IFCHR: 929 case S_IFSOCK: 930 inode->i_blkbits = PAGE_SHIFT; 931 init_special_inode(inode, inode->i_mode, inode->i_rdev); 932 inode->i_op = &ceph_file_iops; 933 break; 934 case S_IFREG: 935 inode->i_op = &ceph_file_iops; 936 inode->i_fop = &ceph_file_fops; 937 break; 938 case S_IFLNK: 939 inode->i_op = &ceph_symlink_iops; 940 if (!ci->i_symlink) { 941 u32 symlen = iinfo->symlink_len; 942 char *sym; 943 944 spin_unlock(&ci->i_ceph_lock); 945 946 if (symlen != i_size_read(inode)) { 947 pr_err("%s %llx.%llx BAD symlink " 948 "size %lld\n", __func__, 949 ceph_vinop(inode), 950 i_size_read(inode)); 951 i_size_write(inode, symlen); 952 inode->i_blocks = calc_inode_blocks(symlen); 953 } 954 955 err = -ENOMEM; 956 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 957 if (!sym) 958 goto out; 959 960 spin_lock(&ci->i_ceph_lock); 961 if (!ci->i_symlink) 962 ci->i_symlink = sym; 963 else 964 kfree(sym); /* lost a race */ 965 } 966 inode->i_link = ci->i_symlink; 967 break; 968 case S_IFDIR: 969 inode->i_op = &ceph_dir_iops; 970 inode->i_fop = &ceph_dir_fops; 971 break; 972 default: 973 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, 974 ceph_vinop(inode), inode->i_mode); 975 } 976 977 /* were we issued a capability? */ 978 if (info_caps) { 979 if (ceph_snap(inode) == CEPH_NOSNAP) { 980 ceph_add_cap(inode, session, 981 le64_to_cpu(info->cap.cap_id), 982 info_caps, 983 le32_to_cpu(info->cap.wanted), 984 le32_to_cpu(info->cap.seq), 985 le32_to_cpu(info->cap.mseq), 986 le64_to_cpu(info->cap.realm), 987 info->cap.flags, &new_cap); 988 989 /* set dir completion flag? */ 990 if (S_ISDIR(inode->i_mode) && 991 ci->i_files == 0 && ci->i_subdirs == 0 && 992 (info_caps & CEPH_CAP_FILE_SHARED) && 993 (issued & CEPH_CAP_FILE_EXCL) == 0 && 994 !__ceph_dir_is_complete(ci)) { 995 dout(" marking %p complete (empty)\n", inode); 996 i_size_write(inode, 0); 997 __ceph_dir_set_complete(ci, 998 atomic64_read(&ci->i_release_count), 999 atomic64_read(&ci->i_ordered_count)); 1000 } 1001 1002 wake = true; 1003 } else { 1004 dout(" %p got snap_caps %s\n", inode, 1005 ceph_cap_string(info_caps)); 1006 ci->i_snap_caps |= info_caps; 1007 } 1008 } 1009 1010 if (iinfo->inline_version > 0 && 1011 iinfo->inline_version >= ci->i_inline_version) { 1012 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1013 ci->i_inline_version = iinfo->inline_version; 1014 if (ci->i_inline_version != CEPH_INLINE_NONE && 1015 (locked_page || (info_caps & cache_caps))) 1016 fill_inline = true; 1017 } 1018 1019 if (cap_fmode >= 0) { 1020 if (!info_caps) 1021 pr_warn("mds issued no caps on %llx.%llx\n", 1022 ceph_vinop(inode)); 1023 __ceph_touch_fmode(ci, mdsc, cap_fmode); 1024 } 1025 1026 spin_unlock(&ci->i_ceph_lock); 1027 1028 if (fill_inline) 1029 ceph_fill_inline_data(inode, locked_page, 1030 iinfo->inline_data, iinfo->inline_len); 1031 1032 if (wake) 1033 wake_up_all(&ci->i_cap_wq); 1034 1035 /* queue truncate if we saw i_size decrease */ 1036 if (queue_trunc) 1037 ceph_queue_vmtruncate(inode); 1038 1039 /* populate frag tree */ 1040 if (S_ISDIR(inode->i_mode)) 1041 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 1042 1043 /* update delegation info? */ 1044 if (dirinfo) 1045 ceph_fill_dirfrag(inode, dirinfo); 1046 1047 err = 0; 1048 out: 1049 if (new_cap) 1050 ceph_put_cap(mdsc, new_cap); 1051 ceph_buffer_put(old_blob); 1052 ceph_buffer_put(xattr_blob); 1053 ceph_put_string(pool_ns); 1054 return err; 1055 } 1056 1057 /* 1058 * caller should hold session s_mutex and dentry->d_lock. 1059 */ 1060 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, 1061 struct ceph_mds_reply_lease *lease, 1062 struct ceph_mds_session *session, 1063 unsigned long from_time, 1064 struct ceph_mds_session **old_lease_session) 1065 { 1066 struct ceph_dentry_info *di = ceph_dentry(dentry); 1067 unsigned mask = le16_to_cpu(lease->mask); 1068 long unsigned duration = le32_to_cpu(lease->duration_ms); 1069 long unsigned ttl = from_time + (duration * HZ) / 1000; 1070 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1071 1072 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1073 dentry, duration, ttl); 1074 1075 /* only track leases on regular dentries */ 1076 if (ceph_snap(dir) != CEPH_NOSNAP) 1077 return; 1078 1079 if (mask & CEPH_LEASE_PRIMARY_LINK) 1080 di->flags |= CEPH_DENTRY_PRIMARY_LINK; 1081 else 1082 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1083 1084 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); 1085 if (!(mask & CEPH_LEASE_VALID)) { 1086 __ceph_dentry_dir_lease_touch(di); 1087 return; 1088 } 1089 1090 if (di->lease_gen == session->s_cap_gen && 1091 time_before(ttl, di->time)) 1092 return; /* we already have a newer lease. */ 1093 1094 if (di->lease_session && di->lease_session != session) { 1095 *old_lease_session = di->lease_session; 1096 di->lease_session = NULL; 1097 } 1098 1099 if (!di->lease_session) 1100 di->lease_session = ceph_get_mds_session(session); 1101 di->lease_gen = session->s_cap_gen; 1102 di->lease_seq = le32_to_cpu(lease->seq); 1103 di->lease_renew_after = half_ttl; 1104 di->lease_renew_from = 0; 1105 di->time = ttl; 1106 1107 __ceph_dentry_lease_touch(di); 1108 } 1109 1110 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, 1111 struct ceph_mds_reply_lease *lease, 1112 struct ceph_mds_session *session, 1113 unsigned long from_time) 1114 { 1115 struct ceph_mds_session *old_lease_session = NULL; 1116 spin_lock(&dentry->d_lock); 1117 __update_dentry_lease(dir, dentry, lease, session, from_time, 1118 &old_lease_session); 1119 spin_unlock(&dentry->d_lock); 1120 if (old_lease_session) 1121 ceph_put_mds_session(old_lease_session); 1122 } 1123 1124 /* 1125 * update dentry lease without having parent inode locked 1126 */ 1127 static void update_dentry_lease_careful(struct dentry *dentry, 1128 struct ceph_mds_reply_lease *lease, 1129 struct ceph_mds_session *session, 1130 unsigned long from_time, 1131 char *dname, u32 dname_len, 1132 struct ceph_vino *pdvino, 1133 struct ceph_vino *ptvino) 1134 1135 { 1136 struct inode *dir; 1137 struct ceph_mds_session *old_lease_session = NULL; 1138 1139 spin_lock(&dentry->d_lock); 1140 /* make sure dentry's name matches target */ 1141 if (dentry->d_name.len != dname_len || 1142 memcmp(dentry->d_name.name, dname, dname_len)) 1143 goto out_unlock; 1144 1145 dir = d_inode(dentry->d_parent); 1146 /* make sure parent matches dvino */ 1147 if (!ceph_ino_compare(dir, pdvino)) 1148 goto out_unlock; 1149 1150 /* make sure dentry's inode matches target. NULL ptvino means that 1151 * we expect a negative dentry */ 1152 if (ptvino) { 1153 if (d_really_is_negative(dentry)) 1154 goto out_unlock; 1155 if (!ceph_ino_compare(d_inode(dentry), ptvino)) 1156 goto out_unlock; 1157 } else { 1158 if (d_really_is_positive(dentry)) 1159 goto out_unlock; 1160 } 1161 1162 __update_dentry_lease(dir, dentry, lease, session, 1163 from_time, &old_lease_session); 1164 out_unlock: 1165 spin_unlock(&dentry->d_lock); 1166 if (old_lease_session) 1167 ceph_put_mds_session(old_lease_session); 1168 } 1169 1170 /* 1171 * splice a dentry to an inode. 1172 * caller must hold directory i_mutex for this to be safe. 1173 */ 1174 static int splice_dentry(struct dentry **pdn, struct inode *in) 1175 { 1176 struct dentry *dn = *pdn; 1177 struct dentry *realdn; 1178 1179 BUG_ON(d_inode(dn)); 1180 1181 if (S_ISDIR(in->i_mode)) { 1182 /* If inode is directory, d_splice_alias() below will remove 1183 * 'realdn' from its origin parent. We need to ensure that 1184 * origin parent's readdir cache will not reference 'realdn' 1185 */ 1186 realdn = d_find_any_alias(in); 1187 if (realdn) { 1188 struct ceph_dentry_info *di = ceph_dentry(realdn); 1189 spin_lock(&realdn->d_lock); 1190 1191 realdn->d_op->d_prune(realdn); 1192 1193 di->time = jiffies; 1194 di->lease_shared_gen = 0; 1195 di->offset = 0; 1196 1197 spin_unlock(&realdn->d_lock); 1198 dput(realdn); 1199 } 1200 } 1201 1202 /* dn must be unhashed */ 1203 if (!d_unhashed(dn)) 1204 d_drop(dn); 1205 realdn = d_splice_alias(in, dn); 1206 if (IS_ERR(realdn)) { 1207 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1208 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1209 return PTR_ERR(realdn); 1210 } 1211 1212 if (realdn) { 1213 dout("dn %p (%d) spliced with %p (%d) " 1214 "inode %p ino %llx.%llx\n", 1215 dn, d_count(dn), 1216 realdn, d_count(realdn), 1217 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1218 dput(dn); 1219 *pdn = realdn; 1220 } else { 1221 BUG_ON(!ceph_dentry(dn)); 1222 dout("dn %p attached to %p ino %llx.%llx\n", 1223 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1224 } 1225 return 0; 1226 } 1227 1228 /* 1229 * Incorporate results into the local cache. This is either just 1230 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1231 * after a lookup). 1232 * 1233 * A reply may contain 1234 * a directory inode along with a dentry. 1235 * and/or a target inode 1236 * 1237 * Called with snap_rwsem (read). 1238 */ 1239 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1240 { 1241 struct ceph_mds_session *session = req->r_session; 1242 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1243 struct inode *in = NULL; 1244 struct ceph_vino tvino, dvino; 1245 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1246 int err = 0; 1247 1248 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1249 rinfo->head->is_dentry, rinfo->head->is_target); 1250 1251 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1252 dout("fill_trace reply is empty!\n"); 1253 if (rinfo->head->result == 0 && req->r_parent) 1254 ceph_invalidate_dir_request(req); 1255 return 0; 1256 } 1257 1258 if (rinfo->head->is_dentry) { 1259 struct inode *dir = req->r_parent; 1260 1261 if (dir) { 1262 err = ceph_fill_inode(dir, NULL, &rinfo->diri, 1263 rinfo->dirfrag, session, -1, 1264 &req->r_caps_reservation); 1265 if (err < 0) 1266 goto done; 1267 } else { 1268 WARN_ON_ONCE(1); 1269 } 1270 1271 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && 1272 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1273 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1274 struct qstr dname; 1275 struct dentry *dn, *parent; 1276 1277 BUG_ON(!rinfo->head->is_target); 1278 BUG_ON(req->r_dentry); 1279 1280 parent = d_find_any_alias(dir); 1281 BUG_ON(!parent); 1282 1283 dname.name = rinfo->dname; 1284 dname.len = rinfo->dname_len; 1285 dname.hash = full_name_hash(parent, dname.name, dname.len); 1286 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1287 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1288 retry_lookup: 1289 dn = d_lookup(parent, &dname); 1290 dout("d_lookup on parent=%p name=%.*s got %p\n", 1291 parent, dname.len, dname.name, dn); 1292 1293 if (!dn) { 1294 dn = d_alloc(parent, &dname); 1295 dout("d_alloc %p '%.*s' = %p\n", parent, 1296 dname.len, dname.name, dn); 1297 if (!dn) { 1298 dput(parent); 1299 err = -ENOMEM; 1300 goto done; 1301 } 1302 err = 0; 1303 } else if (d_really_is_positive(dn) && 1304 (ceph_ino(d_inode(dn)) != tvino.ino || 1305 ceph_snap(d_inode(dn)) != tvino.snap)) { 1306 dout(" dn %p points to wrong inode %p\n", 1307 dn, d_inode(dn)); 1308 ceph_dir_clear_ordered(dir); 1309 d_delete(dn); 1310 dput(dn); 1311 goto retry_lookup; 1312 } 1313 1314 req->r_dentry = dn; 1315 dput(parent); 1316 } 1317 } 1318 1319 if (rinfo->head->is_target) { 1320 /* Should be filled in by handle_reply */ 1321 BUG_ON(!req->r_target_inode); 1322 1323 in = req->r_target_inode; 1324 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, 1325 NULL, session, 1326 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1327 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && 1328 rinfo->head->result == 0) ? req->r_fmode : -1, 1329 &req->r_caps_reservation); 1330 if (err < 0) { 1331 pr_err("ceph_fill_inode badness %p %llx.%llx\n", 1332 in, ceph_vinop(in)); 1333 req->r_target_inode = NULL; 1334 if (in->i_state & I_NEW) 1335 discard_new_inode(in); 1336 else 1337 iput(in); 1338 goto done; 1339 } 1340 if (in->i_state & I_NEW) 1341 unlock_new_inode(in); 1342 } 1343 1344 /* 1345 * ignore null lease/binding on snapdir ENOENT, or else we 1346 * will have trouble splicing in the virtual snapdir later 1347 */ 1348 if (rinfo->head->is_dentry && 1349 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1350 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1351 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1352 fsc->mount_options->snapdir_name, 1353 req->r_dentry->d_name.len))) { 1354 /* 1355 * lookup link rename : null -> possibly existing inode 1356 * mknod symlink mkdir : null -> new inode 1357 * unlink : linked -> null 1358 */ 1359 struct inode *dir = req->r_parent; 1360 struct dentry *dn = req->r_dentry; 1361 bool have_dir_cap, have_lease; 1362 1363 BUG_ON(!dn); 1364 BUG_ON(!dir); 1365 BUG_ON(d_inode(dn->d_parent) != dir); 1366 1367 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1368 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1369 1370 BUG_ON(ceph_ino(dir) != dvino.ino); 1371 BUG_ON(ceph_snap(dir) != dvino.snap); 1372 1373 /* do we have a lease on the whole dir? */ 1374 have_dir_cap = 1375 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1376 CEPH_CAP_FILE_SHARED); 1377 1378 /* do we have a dn lease? */ 1379 have_lease = have_dir_cap || 1380 le32_to_cpu(rinfo->dlease->duration_ms); 1381 if (!have_lease) 1382 dout("fill_trace no dentry lease or dir cap\n"); 1383 1384 /* rename? */ 1385 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1386 struct inode *olddir = req->r_old_dentry_dir; 1387 BUG_ON(!olddir); 1388 1389 dout(" src %p '%pd' dst %p '%pd'\n", 1390 req->r_old_dentry, 1391 req->r_old_dentry, 1392 dn, dn); 1393 dout("fill_trace doing d_move %p -> %p\n", 1394 req->r_old_dentry, dn); 1395 1396 /* d_move screws up sibling dentries' offsets */ 1397 ceph_dir_clear_ordered(dir); 1398 ceph_dir_clear_ordered(olddir); 1399 1400 d_move(req->r_old_dentry, dn); 1401 dout(" src %p '%pd' dst %p '%pd'\n", 1402 req->r_old_dentry, 1403 req->r_old_dentry, 1404 dn, dn); 1405 1406 /* ensure target dentry is invalidated, despite 1407 rehashing bug in vfs_rename_dir */ 1408 ceph_invalidate_dentry_lease(dn); 1409 1410 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1411 ceph_dentry(req->r_old_dentry)->offset); 1412 1413 /* swap r_dentry and r_old_dentry in case that 1414 * splice_dentry() gets called later. This is safe 1415 * because no other place will use them */ 1416 req->r_dentry = req->r_old_dentry; 1417 req->r_old_dentry = dn; 1418 dn = req->r_dentry; 1419 } 1420 1421 /* null dentry? */ 1422 if (!rinfo->head->is_target) { 1423 dout("fill_trace null dentry\n"); 1424 if (d_really_is_positive(dn)) { 1425 dout("d_delete %p\n", dn); 1426 ceph_dir_clear_ordered(dir); 1427 d_delete(dn); 1428 } else if (have_lease) { 1429 if (d_unhashed(dn)) 1430 d_add(dn, NULL); 1431 update_dentry_lease(dir, dn, 1432 rinfo->dlease, session, 1433 req->r_request_started); 1434 } 1435 goto done; 1436 } 1437 1438 /* attach proper inode */ 1439 if (d_really_is_negative(dn)) { 1440 ceph_dir_clear_ordered(dir); 1441 ihold(in); 1442 err = splice_dentry(&req->r_dentry, in); 1443 if (err < 0) 1444 goto done; 1445 dn = req->r_dentry; /* may have spliced */ 1446 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1447 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1448 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1449 ceph_vinop(in)); 1450 d_invalidate(dn); 1451 have_lease = false; 1452 } 1453 1454 if (have_lease) { 1455 update_dentry_lease(dir, dn, 1456 rinfo->dlease, session, 1457 req->r_request_started); 1458 } 1459 dout(" final dn %p\n", dn); 1460 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1461 req->r_op == CEPH_MDS_OP_MKSNAP) && 1462 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1463 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1464 struct inode *dir = req->r_parent; 1465 1466 /* fill out a snapdir LOOKUPSNAP dentry */ 1467 BUG_ON(!dir); 1468 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1469 BUG_ON(!req->r_dentry); 1470 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); 1471 ceph_dir_clear_ordered(dir); 1472 ihold(in); 1473 err = splice_dentry(&req->r_dentry, in); 1474 if (err < 0) 1475 goto done; 1476 } else if (rinfo->head->is_dentry && req->r_dentry) { 1477 /* parent inode is not locked, be carefull */ 1478 struct ceph_vino *ptvino = NULL; 1479 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1480 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1481 if (rinfo->head->is_target) { 1482 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1483 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1484 ptvino = &tvino; 1485 } 1486 update_dentry_lease_careful(req->r_dentry, rinfo->dlease, 1487 session, req->r_request_started, 1488 rinfo->dname, rinfo->dname_len, 1489 &dvino, ptvino); 1490 } 1491 done: 1492 dout("fill_trace done err=%d\n", err); 1493 return err; 1494 } 1495 1496 /* 1497 * Prepopulate our cache with readdir results, leases, etc. 1498 */ 1499 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1500 struct ceph_mds_session *session) 1501 { 1502 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1503 int i, err = 0; 1504 1505 for (i = 0; i < rinfo->dir_nr; i++) { 1506 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1507 struct ceph_vino vino; 1508 struct inode *in; 1509 int rc; 1510 1511 vino.ino = le64_to_cpu(rde->inode.in->ino); 1512 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1513 1514 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1515 if (IS_ERR(in)) { 1516 err = PTR_ERR(in); 1517 dout("new_inode badness got %d\n", err); 1518 continue; 1519 } 1520 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1521 -1, &req->r_caps_reservation); 1522 if (rc < 0) { 1523 pr_err("ceph_fill_inode badness on %p got %d\n", 1524 in, rc); 1525 err = rc; 1526 if (in->i_state & I_NEW) { 1527 ihold(in); 1528 discard_new_inode(in); 1529 } 1530 } else if (in->i_state & I_NEW) { 1531 unlock_new_inode(in); 1532 } 1533 1534 /* avoid calling iput_final() in mds dispatch threads */ 1535 ceph_async_iput(in); 1536 } 1537 1538 return err; 1539 } 1540 1541 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1542 { 1543 if (ctl->page) { 1544 kunmap(ctl->page); 1545 put_page(ctl->page); 1546 ctl->page = NULL; 1547 } 1548 } 1549 1550 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1551 struct ceph_readdir_cache_control *ctl, 1552 struct ceph_mds_request *req) 1553 { 1554 struct ceph_inode_info *ci = ceph_inode(dir); 1555 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1556 unsigned idx = ctl->index % nsize; 1557 pgoff_t pgoff = ctl->index / nsize; 1558 1559 if (!ctl->page || pgoff != page_index(ctl->page)) { 1560 ceph_readdir_cache_release(ctl); 1561 if (idx == 0) 1562 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1563 else 1564 ctl->page = find_lock_page(&dir->i_data, pgoff); 1565 if (!ctl->page) { 1566 ctl->index = -1; 1567 return idx == 0 ? -ENOMEM : 0; 1568 } 1569 /* reading/filling the cache are serialized by 1570 * i_mutex, no need to use page lock */ 1571 unlock_page(ctl->page); 1572 ctl->dentries = kmap(ctl->page); 1573 if (idx == 0) 1574 memset(ctl->dentries, 0, PAGE_SIZE); 1575 } 1576 1577 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1578 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1579 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1580 ctl->dentries[idx] = dn; 1581 ctl->index++; 1582 } else { 1583 dout("disable readdir cache\n"); 1584 ctl->index = -1; 1585 } 1586 return 0; 1587 } 1588 1589 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1590 struct ceph_mds_session *session) 1591 { 1592 struct dentry *parent = req->r_dentry; 1593 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1594 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1595 struct qstr dname; 1596 struct dentry *dn; 1597 struct inode *in; 1598 int err = 0, skipped = 0, ret, i; 1599 u32 frag = le32_to_cpu(req->r_args.readdir.frag); 1600 u32 last_hash = 0; 1601 u32 fpos_offset; 1602 struct ceph_readdir_cache_control cache_ctl = {}; 1603 1604 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1605 return readdir_prepopulate_inodes_only(req, session); 1606 1607 if (rinfo->hash_order) { 1608 if (req->r_path2) { 1609 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1610 req->r_path2, 1611 strlen(req->r_path2)); 1612 last_hash = ceph_frag_value(last_hash); 1613 } else if (rinfo->offset_hash) { 1614 /* mds understands offset_hash */ 1615 WARN_ON_ONCE(req->r_readdir_offset != 2); 1616 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); 1617 } 1618 } 1619 1620 if (rinfo->dir_dir && 1621 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1622 dout("readdir_prepopulate got new frag %x -> %x\n", 1623 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1624 frag = le32_to_cpu(rinfo->dir_dir->frag); 1625 if (!rinfo->hash_order) 1626 req->r_readdir_offset = 2; 1627 } 1628 1629 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1630 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1631 rinfo->dir_nr, parent); 1632 } else { 1633 dout("readdir_prepopulate %d items under dn %p\n", 1634 rinfo->dir_nr, parent); 1635 if (rinfo->dir_dir) 1636 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1637 1638 if (ceph_frag_is_leftmost(frag) && 1639 req->r_readdir_offset == 2 && 1640 !(rinfo->hash_order && last_hash)) { 1641 /* note dir version at start of readdir so we can 1642 * tell if any dentries get dropped */ 1643 req->r_dir_release_cnt = 1644 atomic64_read(&ci->i_release_count); 1645 req->r_dir_ordered_cnt = 1646 atomic64_read(&ci->i_ordered_count); 1647 req->r_readdir_cache_idx = 0; 1648 } 1649 } 1650 1651 cache_ctl.index = req->r_readdir_cache_idx; 1652 fpos_offset = req->r_readdir_offset; 1653 1654 /* FIXME: release caps/leases if error occurs */ 1655 for (i = 0; i < rinfo->dir_nr; i++) { 1656 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1657 struct ceph_vino tvino; 1658 1659 dname.name = rde->name; 1660 dname.len = rde->name_len; 1661 dname.hash = full_name_hash(parent, dname.name, dname.len); 1662 1663 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1664 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1665 1666 if (rinfo->hash_order) { 1667 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1668 rde->name, rde->name_len); 1669 hash = ceph_frag_value(hash); 1670 if (hash != last_hash) 1671 fpos_offset = 2; 1672 last_hash = hash; 1673 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1674 } else { 1675 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1676 } 1677 1678 retry_lookup: 1679 dn = d_lookup(parent, &dname); 1680 dout("d_lookup on parent=%p name=%.*s got %p\n", 1681 parent, dname.len, dname.name, dn); 1682 1683 if (!dn) { 1684 dn = d_alloc(parent, &dname); 1685 dout("d_alloc %p '%.*s' = %p\n", parent, 1686 dname.len, dname.name, dn); 1687 if (!dn) { 1688 dout("d_alloc badness\n"); 1689 err = -ENOMEM; 1690 goto out; 1691 } 1692 } else if (d_really_is_positive(dn) && 1693 (ceph_ino(d_inode(dn)) != tvino.ino || 1694 ceph_snap(d_inode(dn)) != tvino.snap)) { 1695 struct ceph_dentry_info *di = ceph_dentry(dn); 1696 dout(" dn %p points to wrong inode %p\n", 1697 dn, d_inode(dn)); 1698 1699 spin_lock(&dn->d_lock); 1700 if (di->offset > 0 && 1701 di->lease_shared_gen == 1702 atomic_read(&ci->i_shared_gen)) { 1703 __ceph_dir_clear_ordered(ci); 1704 di->offset = 0; 1705 } 1706 spin_unlock(&dn->d_lock); 1707 1708 d_delete(dn); 1709 dput(dn); 1710 goto retry_lookup; 1711 } 1712 1713 /* inode */ 1714 if (d_really_is_positive(dn)) { 1715 in = d_inode(dn); 1716 } else { 1717 in = ceph_get_inode(parent->d_sb, tvino); 1718 if (IS_ERR(in)) { 1719 dout("new_inode badness\n"); 1720 d_drop(dn); 1721 dput(dn); 1722 err = PTR_ERR(in); 1723 goto out; 1724 } 1725 } 1726 1727 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1728 -1, &req->r_caps_reservation); 1729 if (ret < 0) { 1730 pr_err("ceph_fill_inode badness on %p\n", in); 1731 if (d_really_is_negative(dn)) { 1732 /* avoid calling iput_final() in mds 1733 * dispatch threads */ 1734 if (in->i_state & I_NEW) { 1735 ihold(in); 1736 discard_new_inode(in); 1737 } 1738 ceph_async_iput(in); 1739 } 1740 d_drop(dn); 1741 err = ret; 1742 goto next_item; 1743 } 1744 if (in->i_state & I_NEW) 1745 unlock_new_inode(in); 1746 1747 if (d_really_is_negative(dn)) { 1748 if (ceph_security_xattr_deadlock(in)) { 1749 dout(" skip splicing dn %p to inode %p" 1750 " (security xattr deadlock)\n", dn, in); 1751 ceph_async_iput(in); 1752 skipped++; 1753 goto next_item; 1754 } 1755 1756 err = splice_dentry(&dn, in); 1757 if (err < 0) 1758 goto next_item; 1759 } 1760 1761 ceph_dentry(dn)->offset = rde->offset; 1762 1763 update_dentry_lease(d_inode(parent), dn, 1764 rde->lease, req->r_session, 1765 req->r_request_started); 1766 1767 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1768 ret = fill_readdir_cache(d_inode(parent), dn, 1769 &cache_ctl, req); 1770 if (ret < 0) 1771 err = ret; 1772 } 1773 next_item: 1774 dput(dn); 1775 } 1776 out: 1777 if (err == 0 && skipped == 0) { 1778 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1779 req->r_readdir_cache_idx = cache_ctl.index; 1780 } 1781 ceph_readdir_cache_release(&cache_ctl); 1782 dout("readdir_prepopulate done\n"); 1783 return err; 1784 } 1785 1786 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1787 { 1788 struct ceph_inode_info *ci = ceph_inode(inode); 1789 bool ret; 1790 1791 spin_lock(&ci->i_ceph_lock); 1792 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1793 i_size_write(inode, size); 1794 inode->i_blocks = calc_inode_blocks(size); 1795 1796 ret = __ceph_should_report_size(ci); 1797 1798 spin_unlock(&ci->i_ceph_lock); 1799 return ret; 1800 } 1801 1802 /* 1803 * Put reference to inode, but avoid calling iput_final() in current thread. 1804 * iput_final() may wait for reahahead pages. The wait can cause deadlock in 1805 * some contexts. 1806 */ 1807 void ceph_async_iput(struct inode *inode) 1808 { 1809 if (!inode) 1810 return; 1811 for (;;) { 1812 if (atomic_add_unless(&inode->i_count, -1, 1)) 1813 break; 1814 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1815 &ceph_inode(inode)->i_work)) 1816 break; 1817 /* queue work failed, i_count must be at least 2 */ 1818 } 1819 } 1820 1821 void ceph_queue_inode_work(struct inode *inode, int work_bit) 1822 { 1823 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1824 struct ceph_inode_info *ci = ceph_inode(inode); 1825 set_bit(work_bit, &ci->i_work_mask); 1826 1827 ihold(inode); 1828 if (queue_work(fsc->inode_wq, &ci->i_work)) { 1829 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); 1830 } else { 1831 dout("queue_inode_work %p already queued, mask=%lx\n", 1832 inode, ci->i_work_mask); 1833 iput(inode); 1834 } 1835 } 1836 1837 static void ceph_do_invalidate_pages(struct inode *inode) 1838 { 1839 struct ceph_inode_info *ci = ceph_inode(inode); 1840 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1841 u32 orig_gen; 1842 int check = 0; 1843 1844 mutex_lock(&ci->i_truncate_mutex); 1845 1846 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) { 1847 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1848 inode, ceph_ino(inode)); 1849 mapping_set_error(inode->i_mapping, -EIO); 1850 truncate_pagecache(inode, 0); 1851 mutex_unlock(&ci->i_truncate_mutex); 1852 goto out; 1853 } 1854 1855 spin_lock(&ci->i_ceph_lock); 1856 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1857 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1858 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1859 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1860 check = 1; 1861 spin_unlock(&ci->i_ceph_lock); 1862 mutex_unlock(&ci->i_truncate_mutex); 1863 goto out; 1864 } 1865 orig_gen = ci->i_rdcache_gen; 1866 spin_unlock(&ci->i_ceph_lock); 1867 1868 ceph_fscache_invalidate(inode); 1869 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1870 pr_err("invalidate_pages %p fails\n", inode); 1871 } 1872 1873 spin_lock(&ci->i_ceph_lock); 1874 if (orig_gen == ci->i_rdcache_gen && 1875 orig_gen == ci->i_rdcache_revoking) { 1876 dout("invalidate_pages %p gen %d successful\n", inode, 1877 ci->i_rdcache_gen); 1878 ci->i_rdcache_revoking--; 1879 check = 1; 1880 } else { 1881 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1882 inode, orig_gen, ci->i_rdcache_gen, 1883 ci->i_rdcache_revoking); 1884 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1885 check = 1; 1886 } 1887 spin_unlock(&ci->i_ceph_lock); 1888 mutex_unlock(&ci->i_truncate_mutex); 1889 out: 1890 if (check) 1891 ceph_check_caps(ci, 0, NULL); 1892 } 1893 1894 /* 1895 * Make sure any pending truncation is applied before doing anything 1896 * that may depend on it. 1897 */ 1898 void __ceph_do_pending_vmtruncate(struct inode *inode) 1899 { 1900 struct ceph_inode_info *ci = ceph_inode(inode); 1901 u64 to; 1902 int wrbuffer_refs, finish = 0; 1903 1904 mutex_lock(&ci->i_truncate_mutex); 1905 retry: 1906 spin_lock(&ci->i_ceph_lock); 1907 if (ci->i_truncate_pending == 0) { 1908 dout("__do_pending_vmtruncate %p none pending\n", inode); 1909 spin_unlock(&ci->i_ceph_lock); 1910 mutex_unlock(&ci->i_truncate_mutex); 1911 return; 1912 } 1913 1914 /* 1915 * make sure any dirty snapped pages are flushed before we 1916 * possibly truncate them.. so write AND block! 1917 */ 1918 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1919 spin_unlock(&ci->i_ceph_lock); 1920 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1921 inode); 1922 filemap_write_and_wait_range(&inode->i_data, 0, 1923 inode->i_sb->s_maxbytes); 1924 goto retry; 1925 } 1926 1927 /* there should be no reader or writer */ 1928 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1929 1930 to = ci->i_truncate_size; 1931 wrbuffer_refs = ci->i_wrbuffer_ref; 1932 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1933 ci->i_truncate_pending, to); 1934 spin_unlock(&ci->i_ceph_lock); 1935 1936 truncate_pagecache(inode, to); 1937 1938 spin_lock(&ci->i_ceph_lock); 1939 if (to == ci->i_truncate_size) { 1940 ci->i_truncate_pending = 0; 1941 finish = 1; 1942 } 1943 spin_unlock(&ci->i_ceph_lock); 1944 if (!finish) 1945 goto retry; 1946 1947 mutex_unlock(&ci->i_truncate_mutex); 1948 1949 if (wrbuffer_refs == 0) 1950 ceph_check_caps(ci, 0, NULL); 1951 1952 wake_up_all(&ci->i_cap_wq); 1953 } 1954 1955 static void ceph_inode_work(struct work_struct *work) 1956 { 1957 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1958 i_work); 1959 struct inode *inode = &ci->vfs_inode; 1960 1961 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { 1962 dout("writeback %p\n", inode); 1963 filemap_fdatawrite(&inode->i_data); 1964 } 1965 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) 1966 ceph_do_invalidate_pages(inode); 1967 1968 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) 1969 __ceph_do_pending_vmtruncate(inode); 1970 1971 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) 1972 ceph_check_caps(ci, 0, NULL); 1973 1974 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) 1975 ceph_flush_snaps(ci, NULL); 1976 1977 iput(inode); 1978 } 1979 1980 /* 1981 * symlinks 1982 */ 1983 static const struct inode_operations ceph_symlink_iops = { 1984 .get_link = simple_get_link, 1985 .setattr = ceph_setattr, 1986 .getattr = ceph_getattr, 1987 .listxattr = ceph_listxattr, 1988 }; 1989 1990 int __ceph_setattr(struct inode *inode, struct iattr *attr) 1991 { 1992 struct ceph_inode_info *ci = ceph_inode(inode); 1993 unsigned int ia_valid = attr->ia_valid; 1994 struct ceph_mds_request *req; 1995 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1996 struct ceph_cap_flush *prealloc_cf; 1997 int issued; 1998 int release = 0, dirtied = 0; 1999 int mask = 0; 2000 int err = 0; 2001 int inode_dirty_flags = 0; 2002 bool lock_snap_rwsem = false; 2003 2004 prealloc_cf = ceph_alloc_cap_flush(); 2005 if (!prealloc_cf) 2006 return -ENOMEM; 2007 2008 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 2009 USE_AUTH_MDS); 2010 if (IS_ERR(req)) { 2011 ceph_free_cap_flush(prealloc_cf); 2012 return PTR_ERR(req); 2013 } 2014 2015 spin_lock(&ci->i_ceph_lock); 2016 issued = __ceph_caps_issued(ci, NULL); 2017 2018 if (!ci->i_head_snapc && 2019 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 2020 lock_snap_rwsem = true; 2021 if (!down_read_trylock(&mdsc->snap_rwsem)) { 2022 spin_unlock(&ci->i_ceph_lock); 2023 down_read(&mdsc->snap_rwsem); 2024 spin_lock(&ci->i_ceph_lock); 2025 issued = __ceph_caps_issued(ci, NULL); 2026 } 2027 } 2028 2029 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 2030 2031 if (ia_valid & ATTR_UID) { 2032 dout("setattr %p uid %d -> %d\n", inode, 2033 from_kuid(&init_user_ns, inode->i_uid), 2034 from_kuid(&init_user_ns, attr->ia_uid)); 2035 if (issued & CEPH_CAP_AUTH_EXCL) { 2036 inode->i_uid = attr->ia_uid; 2037 dirtied |= CEPH_CAP_AUTH_EXCL; 2038 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2039 !uid_eq(attr->ia_uid, inode->i_uid)) { 2040 req->r_args.setattr.uid = cpu_to_le32( 2041 from_kuid(&init_user_ns, attr->ia_uid)); 2042 mask |= CEPH_SETATTR_UID; 2043 release |= CEPH_CAP_AUTH_SHARED; 2044 } 2045 } 2046 if (ia_valid & ATTR_GID) { 2047 dout("setattr %p gid %d -> %d\n", inode, 2048 from_kgid(&init_user_ns, inode->i_gid), 2049 from_kgid(&init_user_ns, attr->ia_gid)); 2050 if (issued & CEPH_CAP_AUTH_EXCL) { 2051 inode->i_gid = attr->ia_gid; 2052 dirtied |= CEPH_CAP_AUTH_EXCL; 2053 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2054 !gid_eq(attr->ia_gid, inode->i_gid)) { 2055 req->r_args.setattr.gid = cpu_to_le32( 2056 from_kgid(&init_user_ns, attr->ia_gid)); 2057 mask |= CEPH_SETATTR_GID; 2058 release |= CEPH_CAP_AUTH_SHARED; 2059 } 2060 } 2061 if (ia_valid & ATTR_MODE) { 2062 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 2063 attr->ia_mode); 2064 if (issued & CEPH_CAP_AUTH_EXCL) { 2065 inode->i_mode = attr->ia_mode; 2066 dirtied |= CEPH_CAP_AUTH_EXCL; 2067 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2068 attr->ia_mode != inode->i_mode) { 2069 inode->i_mode = attr->ia_mode; 2070 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 2071 mask |= CEPH_SETATTR_MODE; 2072 release |= CEPH_CAP_AUTH_SHARED; 2073 } 2074 } 2075 2076 if (ia_valid & ATTR_ATIME) { 2077 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, 2078 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 2079 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 2080 if (issued & CEPH_CAP_FILE_EXCL) { 2081 ci->i_time_warp_seq++; 2082 inode->i_atime = attr->ia_atime; 2083 dirtied |= CEPH_CAP_FILE_EXCL; 2084 } else if ((issued & CEPH_CAP_FILE_WR) && 2085 timespec64_compare(&inode->i_atime, 2086 &attr->ia_atime) < 0) { 2087 inode->i_atime = attr->ia_atime; 2088 dirtied |= CEPH_CAP_FILE_WR; 2089 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2090 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { 2091 ceph_encode_timespec64(&req->r_args.setattr.atime, 2092 &attr->ia_atime); 2093 mask |= CEPH_SETATTR_ATIME; 2094 release |= CEPH_CAP_FILE_SHARED | 2095 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2096 } 2097 } 2098 if (ia_valid & ATTR_SIZE) { 2099 dout("setattr %p size %lld -> %lld\n", inode, 2100 inode->i_size, attr->ia_size); 2101 if ((issued & CEPH_CAP_FILE_EXCL) && 2102 attr->ia_size > inode->i_size) { 2103 i_size_write(inode, attr->ia_size); 2104 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2105 ci->i_reported_size = attr->ia_size; 2106 dirtied |= CEPH_CAP_FILE_EXCL; 2107 ia_valid |= ATTR_MTIME; 2108 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2109 attr->ia_size != inode->i_size) { 2110 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2111 req->r_args.setattr.old_size = 2112 cpu_to_le64(inode->i_size); 2113 mask |= CEPH_SETATTR_SIZE; 2114 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | 2115 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2116 } 2117 } 2118 if (ia_valid & ATTR_MTIME) { 2119 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, 2120 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 2121 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 2122 if (issued & CEPH_CAP_FILE_EXCL) { 2123 ci->i_time_warp_seq++; 2124 inode->i_mtime = attr->ia_mtime; 2125 dirtied |= CEPH_CAP_FILE_EXCL; 2126 } else if ((issued & CEPH_CAP_FILE_WR) && 2127 timespec64_compare(&inode->i_mtime, 2128 &attr->ia_mtime) < 0) { 2129 inode->i_mtime = attr->ia_mtime; 2130 dirtied |= CEPH_CAP_FILE_WR; 2131 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2132 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { 2133 ceph_encode_timespec64(&req->r_args.setattr.mtime, 2134 &attr->ia_mtime); 2135 mask |= CEPH_SETATTR_MTIME; 2136 release |= CEPH_CAP_FILE_SHARED | 2137 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2138 } 2139 } 2140 2141 /* these do nothing */ 2142 if (ia_valid & ATTR_CTIME) { 2143 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2144 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2145 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, 2146 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2147 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2148 only ? "ctime only" : "ignored"); 2149 if (only) { 2150 /* 2151 * if kernel wants to dirty ctime but nothing else, 2152 * we need to choose a cap to dirty under, or do 2153 * a almost-no-op setattr 2154 */ 2155 if (issued & CEPH_CAP_AUTH_EXCL) 2156 dirtied |= CEPH_CAP_AUTH_EXCL; 2157 else if (issued & CEPH_CAP_FILE_EXCL) 2158 dirtied |= CEPH_CAP_FILE_EXCL; 2159 else if (issued & CEPH_CAP_XATTR_EXCL) 2160 dirtied |= CEPH_CAP_XATTR_EXCL; 2161 else 2162 mask |= CEPH_SETATTR_CTIME; 2163 } 2164 } 2165 if (ia_valid & ATTR_FILE) 2166 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2167 2168 if (dirtied) { 2169 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2170 &prealloc_cf); 2171 inode->i_ctime = attr->ia_ctime; 2172 } 2173 2174 release &= issued; 2175 spin_unlock(&ci->i_ceph_lock); 2176 if (lock_snap_rwsem) 2177 up_read(&mdsc->snap_rwsem); 2178 2179 if (inode_dirty_flags) 2180 __mark_inode_dirty(inode, inode_dirty_flags); 2181 2182 2183 if (mask) { 2184 req->r_inode = inode; 2185 ihold(inode); 2186 req->r_inode_drop = release; 2187 req->r_args.setattr.mask = cpu_to_le32(mask); 2188 req->r_num_caps = 1; 2189 req->r_stamp = attr->ia_ctime; 2190 err = ceph_mdsc_do_request(mdsc, NULL, req); 2191 } 2192 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2193 ceph_cap_string(dirtied), mask); 2194 2195 ceph_mdsc_put_request(req); 2196 ceph_free_cap_flush(prealloc_cf); 2197 2198 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2199 __ceph_do_pending_vmtruncate(inode); 2200 2201 return err; 2202 } 2203 2204 /* 2205 * setattr 2206 */ 2207 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 2208 struct iattr *attr) 2209 { 2210 struct inode *inode = d_inode(dentry); 2211 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 2212 int err; 2213 2214 if (ceph_snap(inode) != CEPH_NOSNAP) 2215 return -EROFS; 2216 2217 err = setattr_prepare(&init_user_ns, dentry, attr); 2218 if (err != 0) 2219 return err; 2220 2221 if ((attr->ia_valid & ATTR_SIZE) && 2222 attr->ia_size > max(inode->i_size, fsc->max_file_size)) 2223 return -EFBIG; 2224 2225 if ((attr->ia_valid & ATTR_SIZE) && 2226 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) 2227 return -EDQUOT; 2228 2229 err = __ceph_setattr(inode, attr); 2230 2231 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2232 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode); 2233 2234 return err; 2235 } 2236 2237 /* 2238 * Verify that we have a lease on the given mask. If not, 2239 * do a getattr against an mds. 2240 */ 2241 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2242 int mask, bool force) 2243 { 2244 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2245 struct ceph_mds_client *mdsc = fsc->mdsc; 2246 struct ceph_mds_request *req; 2247 int mode; 2248 int err; 2249 2250 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2251 dout("do_getattr inode %p SNAPDIR\n", inode); 2252 return 0; 2253 } 2254 2255 dout("do_getattr inode %p mask %s mode 0%o\n", 2256 inode, ceph_cap_string(mask), inode->i_mode); 2257 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) 2258 return 0; 2259 2260 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; 2261 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2262 if (IS_ERR(req)) 2263 return PTR_ERR(req); 2264 req->r_inode = inode; 2265 ihold(inode); 2266 req->r_num_caps = 1; 2267 req->r_args.getattr.mask = cpu_to_le32(mask); 2268 req->r_locked_page = locked_page; 2269 err = ceph_mdsc_do_request(mdsc, NULL, req); 2270 if (locked_page && err == 0) { 2271 u64 inline_version = req->r_reply_info.targeti.inline_version; 2272 if (inline_version == 0) { 2273 /* the reply is supposed to contain inline data */ 2274 err = -EINVAL; 2275 } else if (inline_version == CEPH_INLINE_NONE) { 2276 err = -ENODATA; 2277 } else { 2278 err = req->r_reply_info.targeti.inline_len; 2279 } 2280 } 2281 ceph_mdsc_put_request(req); 2282 dout("do_getattr result=%d\n", err); 2283 return err; 2284 } 2285 2286 2287 /* 2288 * Check inode permissions. We verify we have a valid value for 2289 * the AUTH cap, then call the generic handler. 2290 */ 2291 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode, 2292 int mask) 2293 { 2294 int err; 2295 2296 if (mask & MAY_NOT_BLOCK) 2297 return -ECHILD; 2298 2299 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2300 2301 if (!err) 2302 err = generic_permission(&init_user_ns, inode, mask); 2303 return err; 2304 } 2305 2306 /* Craft a mask of needed caps given a set of requested statx attrs. */ 2307 static int statx_to_caps(u32 want, umode_t mode) 2308 { 2309 int mask = 0; 2310 2311 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) 2312 mask |= CEPH_CAP_AUTH_SHARED; 2313 2314 if (want & (STATX_NLINK|STATX_CTIME)) { 2315 /* 2316 * The link count for directories depends on inode->i_subdirs, 2317 * and that is only updated when Fs caps are held. 2318 */ 2319 if (S_ISDIR(mode)) 2320 mask |= CEPH_CAP_FILE_SHARED; 2321 else 2322 mask |= CEPH_CAP_LINK_SHARED; 2323 } 2324 2325 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| 2326 STATX_BLOCKS)) 2327 mask |= CEPH_CAP_FILE_SHARED; 2328 2329 if (want & (STATX_CTIME)) 2330 mask |= CEPH_CAP_XATTR_SHARED; 2331 2332 return mask; 2333 } 2334 2335 /* 2336 * Get all the attributes. If we have sufficient caps for the requested attrs, 2337 * then we can avoid talking to the MDS at all. 2338 */ 2339 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path, 2340 struct kstat *stat, u32 request_mask, unsigned int flags) 2341 { 2342 struct inode *inode = d_inode(path->dentry); 2343 struct ceph_inode_info *ci = ceph_inode(inode); 2344 u32 valid_mask = STATX_BASIC_STATS; 2345 int err = 0; 2346 2347 /* Skip the getattr altogether if we're asked not to sync */ 2348 if (!(flags & AT_STATX_DONT_SYNC)) { 2349 err = ceph_do_getattr(inode, 2350 statx_to_caps(request_mask, inode->i_mode), 2351 flags & AT_STATX_FORCE_SYNC); 2352 if (err) 2353 return err; 2354 } 2355 2356 generic_fillattr(&init_user_ns, inode, stat); 2357 stat->ino = ceph_present_inode(inode); 2358 2359 /* 2360 * btime on newly-allocated inodes is 0, so if this is still set to 2361 * that, then assume that it's not valid. 2362 */ 2363 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { 2364 stat->btime = ci->i_btime; 2365 valid_mask |= STATX_BTIME; 2366 } 2367 2368 if (ceph_snap(inode) == CEPH_NOSNAP) 2369 stat->dev = inode->i_sb->s_dev; 2370 else 2371 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; 2372 2373 if (S_ISDIR(inode->i_mode)) { 2374 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2375 RBYTES)) 2376 stat->size = ci->i_rbytes; 2377 else 2378 stat->size = ci->i_files + ci->i_subdirs; 2379 stat->blocks = 0; 2380 stat->blksize = 65536; 2381 /* 2382 * Some applications rely on the number of st_nlink 2383 * value on directories to be either 0 (if unlinked) 2384 * or 2 + number of subdirectories. 2385 */ 2386 if (stat->nlink == 1) 2387 /* '.' + '..' + subdirs */ 2388 stat->nlink = 1 + 1 + ci->i_subdirs; 2389 } 2390 2391 stat->result_mask = request_mask & valid_mask; 2392 return err; 2393 } 2394