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 ceph_decode_timespec64(&ci->i_rctime, &info->rctime); 899 } 900 } 901 902 /* xattrs */ 903 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 904 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 905 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 906 if (ci->i_xattrs.blob) 907 old_blob = ci->i_xattrs.blob; 908 ci->i_xattrs.blob = xattr_blob; 909 if (xattr_blob) 910 memcpy(ci->i_xattrs.blob->vec.iov_base, 911 iinfo->xattr_data, iinfo->xattr_len); 912 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 913 ceph_forget_all_cached_acls(inode); 914 ceph_security_invalidate_secctx(inode); 915 xattr_blob = NULL; 916 } 917 918 /* finally update i_version */ 919 if (le64_to_cpu(info->version) > ci->i_version) 920 ci->i_version = le64_to_cpu(info->version); 921 922 inode->i_mapping->a_ops = &ceph_aops; 923 924 switch (inode->i_mode & S_IFMT) { 925 case S_IFIFO: 926 case S_IFBLK: 927 case S_IFCHR: 928 case S_IFSOCK: 929 inode->i_blkbits = PAGE_SHIFT; 930 init_special_inode(inode, inode->i_mode, inode->i_rdev); 931 inode->i_op = &ceph_file_iops; 932 break; 933 case S_IFREG: 934 inode->i_op = &ceph_file_iops; 935 inode->i_fop = &ceph_file_fops; 936 break; 937 case S_IFLNK: 938 inode->i_op = &ceph_symlink_iops; 939 if (!ci->i_symlink) { 940 u32 symlen = iinfo->symlink_len; 941 char *sym; 942 943 spin_unlock(&ci->i_ceph_lock); 944 945 if (symlen != i_size_read(inode)) { 946 pr_err("%s %llx.%llx BAD symlink " 947 "size %lld\n", __func__, 948 ceph_vinop(inode), 949 i_size_read(inode)); 950 i_size_write(inode, symlen); 951 inode->i_blocks = calc_inode_blocks(symlen); 952 } 953 954 err = -ENOMEM; 955 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 956 if (!sym) 957 goto out; 958 959 spin_lock(&ci->i_ceph_lock); 960 if (!ci->i_symlink) 961 ci->i_symlink = sym; 962 else 963 kfree(sym); /* lost a race */ 964 } 965 inode->i_link = ci->i_symlink; 966 break; 967 case S_IFDIR: 968 inode->i_op = &ceph_dir_iops; 969 inode->i_fop = &ceph_dir_fops; 970 break; 971 default: 972 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, 973 ceph_vinop(inode), inode->i_mode); 974 } 975 976 /* were we issued a capability? */ 977 if (info_caps) { 978 if (ceph_snap(inode) == CEPH_NOSNAP) { 979 ceph_add_cap(inode, session, 980 le64_to_cpu(info->cap.cap_id), 981 info_caps, 982 le32_to_cpu(info->cap.wanted), 983 le32_to_cpu(info->cap.seq), 984 le32_to_cpu(info->cap.mseq), 985 le64_to_cpu(info->cap.realm), 986 info->cap.flags, &new_cap); 987 988 /* set dir completion flag? */ 989 if (S_ISDIR(inode->i_mode) && 990 ci->i_files == 0 && ci->i_subdirs == 0 && 991 (info_caps & CEPH_CAP_FILE_SHARED) && 992 (issued & CEPH_CAP_FILE_EXCL) == 0 && 993 !__ceph_dir_is_complete(ci)) { 994 dout(" marking %p complete (empty)\n", inode); 995 i_size_write(inode, 0); 996 __ceph_dir_set_complete(ci, 997 atomic64_read(&ci->i_release_count), 998 atomic64_read(&ci->i_ordered_count)); 999 } 1000 1001 wake = true; 1002 } else { 1003 dout(" %p got snap_caps %s\n", inode, 1004 ceph_cap_string(info_caps)); 1005 ci->i_snap_caps |= info_caps; 1006 } 1007 } 1008 1009 if (iinfo->inline_version > 0 && 1010 iinfo->inline_version >= ci->i_inline_version) { 1011 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1012 ci->i_inline_version = iinfo->inline_version; 1013 if (ci->i_inline_version != CEPH_INLINE_NONE && 1014 (locked_page || (info_caps & cache_caps))) 1015 fill_inline = true; 1016 } 1017 1018 if (cap_fmode >= 0) { 1019 if (!info_caps) 1020 pr_warn("mds issued no caps on %llx.%llx\n", 1021 ceph_vinop(inode)); 1022 __ceph_touch_fmode(ci, mdsc, cap_fmode); 1023 } 1024 1025 spin_unlock(&ci->i_ceph_lock); 1026 1027 if (fill_inline) 1028 ceph_fill_inline_data(inode, locked_page, 1029 iinfo->inline_data, iinfo->inline_len); 1030 1031 if (wake) 1032 wake_up_all(&ci->i_cap_wq); 1033 1034 /* queue truncate if we saw i_size decrease */ 1035 if (queue_trunc) 1036 ceph_queue_vmtruncate(inode); 1037 1038 /* populate frag tree */ 1039 if (S_ISDIR(inode->i_mode)) 1040 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 1041 1042 /* update delegation info? */ 1043 if (dirinfo) 1044 ceph_fill_dirfrag(inode, dirinfo); 1045 1046 err = 0; 1047 out: 1048 if (new_cap) 1049 ceph_put_cap(mdsc, new_cap); 1050 ceph_buffer_put(old_blob); 1051 ceph_buffer_put(xattr_blob); 1052 ceph_put_string(pool_ns); 1053 return err; 1054 } 1055 1056 /* 1057 * caller should hold session s_mutex and dentry->d_lock. 1058 */ 1059 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, 1060 struct ceph_mds_reply_lease *lease, 1061 struct ceph_mds_session *session, 1062 unsigned long from_time, 1063 struct ceph_mds_session **old_lease_session) 1064 { 1065 struct ceph_dentry_info *di = ceph_dentry(dentry); 1066 unsigned mask = le16_to_cpu(lease->mask); 1067 long unsigned duration = le32_to_cpu(lease->duration_ms); 1068 long unsigned ttl = from_time + (duration * HZ) / 1000; 1069 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1070 1071 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1072 dentry, duration, ttl); 1073 1074 /* only track leases on regular dentries */ 1075 if (ceph_snap(dir) != CEPH_NOSNAP) 1076 return; 1077 1078 if (mask & CEPH_LEASE_PRIMARY_LINK) 1079 di->flags |= CEPH_DENTRY_PRIMARY_LINK; 1080 else 1081 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1082 1083 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); 1084 if (!(mask & CEPH_LEASE_VALID)) { 1085 __ceph_dentry_dir_lease_touch(di); 1086 return; 1087 } 1088 1089 if (di->lease_gen == session->s_cap_gen && 1090 time_before(ttl, di->time)) 1091 return; /* we already have a newer lease. */ 1092 1093 if (di->lease_session && di->lease_session != session) { 1094 *old_lease_session = di->lease_session; 1095 di->lease_session = NULL; 1096 } 1097 1098 if (!di->lease_session) 1099 di->lease_session = ceph_get_mds_session(session); 1100 di->lease_gen = session->s_cap_gen; 1101 di->lease_seq = le32_to_cpu(lease->seq); 1102 di->lease_renew_after = half_ttl; 1103 di->lease_renew_from = 0; 1104 di->time = ttl; 1105 1106 __ceph_dentry_lease_touch(di); 1107 } 1108 1109 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, 1110 struct ceph_mds_reply_lease *lease, 1111 struct ceph_mds_session *session, 1112 unsigned long from_time) 1113 { 1114 struct ceph_mds_session *old_lease_session = NULL; 1115 spin_lock(&dentry->d_lock); 1116 __update_dentry_lease(dir, dentry, lease, session, from_time, 1117 &old_lease_session); 1118 spin_unlock(&dentry->d_lock); 1119 if (old_lease_session) 1120 ceph_put_mds_session(old_lease_session); 1121 } 1122 1123 /* 1124 * update dentry lease without having parent inode locked 1125 */ 1126 static void update_dentry_lease_careful(struct dentry *dentry, 1127 struct ceph_mds_reply_lease *lease, 1128 struct ceph_mds_session *session, 1129 unsigned long from_time, 1130 char *dname, u32 dname_len, 1131 struct ceph_vino *pdvino, 1132 struct ceph_vino *ptvino) 1133 1134 { 1135 struct inode *dir; 1136 struct ceph_mds_session *old_lease_session = NULL; 1137 1138 spin_lock(&dentry->d_lock); 1139 /* make sure dentry's name matches target */ 1140 if (dentry->d_name.len != dname_len || 1141 memcmp(dentry->d_name.name, dname, dname_len)) 1142 goto out_unlock; 1143 1144 dir = d_inode(dentry->d_parent); 1145 /* make sure parent matches dvino */ 1146 if (!ceph_ino_compare(dir, pdvino)) 1147 goto out_unlock; 1148 1149 /* make sure dentry's inode matches target. NULL ptvino means that 1150 * we expect a negative dentry */ 1151 if (ptvino) { 1152 if (d_really_is_negative(dentry)) 1153 goto out_unlock; 1154 if (!ceph_ino_compare(d_inode(dentry), ptvino)) 1155 goto out_unlock; 1156 } else { 1157 if (d_really_is_positive(dentry)) 1158 goto out_unlock; 1159 } 1160 1161 __update_dentry_lease(dir, dentry, lease, session, 1162 from_time, &old_lease_session); 1163 out_unlock: 1164 spin_unlock(&dentry->d_lock); 1165 if (old_lease_session) 1166 ceph_put_mds_session(old_lease_session); 1167 } 1168 1169 /* 1170 * splice a dentry to an inode. 1171 * caller must hold directory i_mutex for this to be safe. 1172 */ 1173 static int splice_dentry(struct dentry **pdn, struct inode *in) 1174 { 1175 struct dentry *dn = *pdn; 1176 struct dentry *realdn; 1177 1178 BUG_ON(d_inode(dn)); 1179 1180 if (S_ISDIR(in->i_mode)) { 1181 /* If inode is directory, d_splice_alias() below will remove 1182 * 'realdn' from its origin parent. We need to ensure that 1183 * origin parent's readdir cache will not reference 'realdn' 1184 */ 1185 realdn = d_find_any_alias(in); 1186 if (realdn) { 1187 struct ceph_dentry_info *di = ceph_dentry(realdn); 1188 spin_lock(&realdn->d_lock); 1189 1190 realdn->d_op->d_prune(realdn); 1191 1192 di->time = jiffies; 1193 di->lease_shared_gen = 0; 1194 di->offset = 0; 1195 1196 spin_unlock(&realdn->d_lock); 1197 dput(realdn); 1198 } 1199 } 1200 1201 /* dn must be unhashed */ 1202 if (!d_unhashed(dn)) 1203 d_drop(dn); 1204 realdn = d_splice_alias(in, dn); 1205 if (IS_ERR(realdn)) { 1206 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1207 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1208 return PTR_ERR(realdn); 1209 } 1210 1211 if (realdn) { 1212 dout("dn %p (%d) spliced with %p (%d) " 1213 "inode %p ino %llx.%llx\n", 1214 dn, d_count(dn), 1215 realdn, d_count(realdn), 1216 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1217 dput(dn); 1218 *pdn = realdn; 1219 } else { 1220 BUG_ON(!ceph_dentry(dn)); 1221 dout("dn %p attached to %p ino %llx.%llx\n", 1222 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1223 } 1224 return 0; 1225 } 1226 1227 /* 1228 * Incorporate results into the local cache. This is either just 1229 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1230 * after a lookup). 1231 * 1232 * A reply may contain 1233 * a directory inode along with a dentry. 1234 * and/or a target inode 1235 * 1236 * Called with snap_rwsem (read). 1237 */ 1238 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1239 { 1240 struct ceph_mds_session *session = req->r_session; 1241 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1242 struct inode *in = NULL; 1243 struct ceph_vino tvino, dvino; 1244 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1245 int err = 0; 1246 1247 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1248 rinfo->head->is_dentry, rinfo->head->is_target); 1249 1250 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1251 dout("fill_trace reply is empty!\n"); 1252 if (rinfo->head->result == 0 && req->r_parent) 1253 ceph_invalidate_dir_request(req); 1254 return 0; 1255 } 1256 1257 if (rinfo->head->is_dentry) { 1258 struct inode *dir = req->r_parent; 1259 1260 if (dir) { 1261 err = ceph_fill_inode(dir, NULL, &rinfo->diri, 1262 rinfo->dirfrag, session, -1, 1263 &req->r_caps_reservation); 1264 if (err < 0) 1265 goto done; 1266 } else { 1267 WARN_ON_ONCE(1); 1268 } 1269 1270 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && 1271 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1272 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1273 struct qstr dname; 1274 struct dentry *dn, *parent; 1275 1276 BUG_ON(!rinfo->head->is_target); 1277 BUG_ON(req->r_dentry); 1278 1279 parent = d_find_any_alias(dir); 1280 BUG_ON(!parent); 1281 1282 dname.name = rinfo->dname; 1283 dname.len = rinfo->dname_len; 1284 dname.hash = full_name_hash(parent, dname.name, dname.len); 1285 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1286 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1287 retry_lookup: 1288 dn = d_lookup(parent, &dname); 1289 dout("d_lookup on parent=%p name=%.*s got %p\n", 1290 parent, dname.len, dname.name, dn); 1291 1292 if (!dn) { 1293 dn = d_alloc(parent, &dname); 1294 dout("d_alloc %p '%.*s' = %p\n", parent, 1295 dname.len, dname.name, dn); 1296 if (!dn) { 1297 dput(parent); 1298 err = -ENOMEM; 1299 goto done; 1300 } 1301 err = 0; 1302 } else if (d_really_is_positive(dn) && 1303 (ceph_ino(d_inode(dn)) != tvino.ino || 1304 ceph_snap(d_inode(dn)) != tvino.snap)) { 1305 dout(" dn %p points to wrong inode %p\n", 1306 dn, d_inode(dn)); 1307 ceph_dir_clear_ordered(dir); 1308 d_delete(dn); 1309 dput(dn); 1310 goto retry_lookup; 1311 } 1312 1313 req->r_dentry = dn; 1314 dput(parent); 1315 } 1316 } 1317 1318 if (rinfo->head->is_target) { 1319 /* Should be filled in by handle_reply */ 1320 BUG_ON(!req->r_target_inode); 1321 1322 in = req->r_target_inode; 1323 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, 1324 NULL, session, 1325 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1326 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && 1327 rinfo->head->result == 0) ? req->r_fmode : -1, 1328 &req->r_caps_reservation); 1329 if (err < 0) { 1330 pr_err("ceph_fill_inode badness %p %llx.%llx\n", 1331 in, ceph_vinop(in)); 1332 req->r_target_inode = NULL; 1333 if (in->i_state & I_NEW) 1334 discard_new_inode(in); 1335 else 1336 iput(in); 1337 goto done; 1338 } 1339 if (in->i_state & I_NEW) 1340 unlock_new_inode(in); 1341 } 1342 1343 /* 1344 * ignore null lease/binding on snapdir ENOENT, or else we 1345 * will have trouble splicing in the virtual snapdir later 1346 */ 1347 if (rinfo->head->is_dentry && 1348 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1349 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1350 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1351 fsc->mount_options->snapdir_name, 1352 req->r_dentry->d_name.len))) { 1353 /* 1354 * lookup link rename : null -> possibly existing inode 1355 * mknod symlink mkdir : null -> new inode 1356 * unlink : linked -> null 1357 */ 1358 struct inode *dir = req->r_parent; 1359 struct dentry *dn = req->r_dentry; 1360 bool have_dir_cap, have_lease; 1361 1362 BUG_ON(!dn); 1363 BUG_ON(!dir); 1364 BUG_ON(d_inode(dn->d_parent) != dir); 1365 1366 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1367 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1368 1369 BUG_ON(ceph_ino(dir) != dvino.ino); 1370 BUG_ON(ceph_snap(dir) != dvino.snap); 1371 1372 /* do we have a lease on the whole dir? */ 1373 have_dir_cap = 1374 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1375 CEPH_CAP_FILE_SHARED); 1376 1377 /* do we have a dn lease? */ 1378 have_lease = have_dir_cap || 1379 le32_to_cpu(rinfo->dlease->duration_ms); 1380 if (!have_lease) 1381 dout("fill_trace no dentry lease or dir cap\n"); 1382 1383 /* rename? */ 1384 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1385 struct inode *olddir = req->r_old_dentry_dir; 1386 BUG_ON(!olddir); 1387 1388 dout(" src %p '%pd' dst %p '%pd'\n", 1389 req->r_old_dentry, 1390 req->r_old_dentry, 1391 dn, dn); 1392 dout("fill_trace doing d_move %p -> %p\n", 1393 req->r_old_dentry, dn); 1394 1395 /* d_move screws up sibling dentries' offsets */ 1396 ceph_dir_clear_ordered(dir); 1397 ceph_dir_clear_ordered(olddir); 1398 1399 d_move(req->r_old_dentry, dn); 1400 dout(" src %p '%pd' dst %p '%pd'\n", 1401 req->r_old_dentry, 1402 req->r_old_dentry, 1403 dn, dn); 1404 1405 /* ensure target dentry is invalidated, despite 1406 rehashing bug in vfs_rename_dir */ 1407 ceph_invalidate_dentry_lease(dn); 1408 1409 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1410 ceph_dentry(req->r_old_dentry)->offset); 1411 1412 /* swap r_dentry and r_old_dentry in case that 1413 * splice_dentry() gets called later. This is safe 1414 * because no other place will use them */ 1415 req->r_dentry = req->r_old_dentry; 1416 req->r_old_dentry = dn; 1417 dn = req->r_dentry; 1418 } 1419 1420 /* null dentry? */ 1421 if (!rinfo->head->is_target) { 1422 dout("fill_trace null dentry\n"); 1423 if (d_really_is_positive(dn)) { 1424 dout("d_delete %p\n", dn); 1425 ceph_dir_clear_ordered(dir); 1426 d_delete(dn); 1427 } else if (have_lease) { 1428 if (d_unhashed(dn)) 1429 d_add(dn, NULL); 1430 update_dentry_lease(dir, dn, 1431 rinfo->dlease, session, 1432 req->r_request_started); 1433 } 1434 goto done; 1435 } 1436 1437 /* attach proper inode */ 1438 if (d_really_is_negative(dn)) { 1439 ceph_dir_clear_ordered(dir); 1440 ihold(in); 1441 err = splice_dentry(&req->r_dentry, in); 1442 if (err < 0) 1443 goto done; 1444 dn = req->r_dentry; /* may have spliced */ 1445 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1446 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1447 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1448 ceph_vinop(in)); 1449 d_invalidate(dn); 1450 have_lease = false; 1451 } 1452 1453 if (have_lease) { 1454 update_dentry_lease(dir, dn, 1455 rinfo->dlease, session, 1456 req->r_request_started); 1457 } 1458 dout(" final dn %p\n", dn); 1459 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1460 req->r_op == CEPH_MDS_OP_MKSNAP) && 1461 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1462 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1463 struct inode *dir = req->r_parent; 1464 1465 /* fill out a snapdir LOOKUPSNAP dentry */ 1466 BUG_ON(!dir); 1467 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1468 BUG_ON(!req->r_dentry); 1469 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); 1470 ceph_dir_clear_ordered(dir); 1471 ihold(in); 1472 err = splice_dentry(&req->r_dentry, in); 1473 if (err < 0) 1474 goto done; 1475 } else if (rinfo->head->is_dentry && req->r_dentry) { 1476 /* parent inode is not locked, be carefull */ 1477 struct ceph_vino *ptvino = NULL; 1478 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1479 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1480 if (rinfo->head->is_target) { 1481 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1482 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1483 ptvino = &tvino; 1484 } 1485 update_dentry_lease_careful(req->r_dentry, rinfo->dlease, 1486 session, req->r_request_started, 1487 rinfo->dname, rinfo->dname_len, 1488 &dvino, ptvino); 1489 } 1490 done: 1491 dout("fill_trace done err=%d\n", err); 1492 return err; 1493 } 1494 1495 /* 1496 * Prepopulate our cache with readdir results, leases, etc. 1497 */ 1498 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1499 struct ceph_mds_session *session) 1500 { 1501 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1502 int i, err = 0; 1503 1504 for (i = 0; i < rinfo->dir_nr; i++) { 1505 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1506 struct ceph_vino vino; 1507 struct inode *in; 1508 int rc; 1509 1510 vino.ino = le64_to_cpu(rde->inode.in->ino); 1511 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1512 1513 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1514 if (IS_ERR(in)) { 1515 err = PTR_ERR(in); 1516 dout("new_inode badness got %d\n", err); 1517 continue; 1518 } 1519 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1520 -1, &req->r_caps_reservation); 1521 if (rc < 0) { 1522 pr_err("ceph_fill_inode badness on %p got %d\n", 1523 in, rc); 1524 err = rc; 1525 if (in->i_state & I_NEW) { 1526 ihold(in); 1527 discard_new_inode(in); 1528 } 1529 } else if (in->i_state & I_NEW) { 1530 unlock_new_inode(in); 1531 } 1532 1533 /* avoid calling iput_final() in mds dispatch threads */ 1534 ceph_async_iput(in); 1535 } 1536 1537 return err; 1538 } 1539 1540 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1541 { 1542 if (ctl->page) { 1543 kunmap(ctl->page); 1544 put_page(ctl->page); 1545 ctl->page = NULL; 1546 } 1547 } 1548 1549 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1550 struct ceph_readdir_cache_control *ctl, 1551 struct ceph_mds_request *req) 1552 { 1553 struct ceph_inode_info *ci = ceph_inode(dir); 1554 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1555 unsigned idx = ctl->index % nsize; 1556 pgoff_t pgoff = ctl->index / nsize; 1557 1558 if (!ctl->page || pgoff != page_index(ctl->page)) { 1559 ceph_readdir_cache_release(ctl); 1560 if (idx == 0) 1561 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1562 else 1563 ctl->page = find_lock_page(&dir->i_data, pgoff); 1564 if (!ctl->page) { 1565 ctl->index = -1; 1566 return idx == 0 ? -ENOMEM : 0; 1567 } 1568 /* reading/filling the cache are serialized by 1569 * i_mutex, no need to use page lock */ 1570 unlock_page(ctl->page); 1571 ctl->dentries = kmap(ctl->page); 1572 if (idx == 0) 1573 memset(ctl->dentries, 0, PAGE_SIZE); 1574 } 1575 1576 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1577 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1578 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1579 ctl->dentries[idx] = dn; 1580 ctl->index++; 1581 } else { 1582 dout("disable readdir cache\n"); 1583 ctl->index = -1; 1584 } 1585 return 0; 1586 } 1587 1588 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1589 struct ceph_mds_session *session) 1590 { 1591 struct dentry *parent = req->r_dentry; 1592 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1593 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1594 struct qstr dname; 1595 struct dentry *dn; 1596 struct inode *in; 1597 int err = 0, skipped = 0, ret, i; 1598 u32 frag = le32_to_cpu(req->r_args.readdir.frag); 1599 u32 last_hash = 0; 1600 u32 fpos_offset; 1601 struct ceph_readdir_cache_control cache_ctl = {}; 1602 1603 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1604 return readdir_prepopulate_inodes_only(req, session); 1605 1606 if (rinfo->hash_order) { 1607 if (req->r_path2) { 1608 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1609 req->r_path2, 1610 strlen(req->r_path2)); 1611 last_hash = ceph_frag_value(last_hash); 1612 } else if (rinfo->offset_hash) { 1613 /* mds understands offset_hash */ 1614 WARN_ON_ONCE(req->r_readdir_offset != 2); 1615 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); 1616 } 1617 } 1618 1619 if (rinfo->dir_dir && 1620 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1621 dout("readdir_prepopulate got new frag %x -> %x\n", 1622 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1623 frag = le32_to_cpu(rinfo->dir_dir->frag); 1624 if (!rinfo->hash_order) 1625 req->r_readdir_offset = 2; 1626 } 1627 1628 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1629 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1630 rinfo->dir_nr, parent); 1631 } else { 1632 dout("readdir_prepopulate %d items under dn %p\n", 1633 rinfo->dir_nr, parent); 1634 if (rinfo->dir_dir) 1635 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1636 1637 if (ceph_frag_is_leftmost(frag) && 1638 req->r_readdir_offset == 2 && 1639 !(rinfo->hash_order && last_hash)) { 1640 /* note dir version at start of readdir so we can 1641 * tell if any dentries get dropped */ 1642 req->r_dir_release_cnt = 1643 atomic64_read(&ci->i_release_count); 1644 req->r_dir_ordered_cnt = 1645 atomic64_read(&ci->i_ordered_count); 1646 req->r_readdir_cache_idx = 0; 1647 } 1648 } 1649 1650 cache_ctl.index = req->r_readdir_cache_idx; 1651 fpos_offset = req->r_readdir_offset; 1652 1653 /* FIXME: release caps/leases if error occurs */ 1654 for (i = 0; i < rinfo->dir_nr; i++) { 1655 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1656 struct ceph_vino tvino; 1657 1658 dname.name = rde->name; 1659 dname.len = rde->name_len; 1660 dname.hash = full_name_hash(parent, dname.name, dname.len); 1661 1662 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1663 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1664 1665 if (rinfo->hash_order) { 1666 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1667 rde->name, rde->name_len); 1668 hash = ceph_frag_value(hash); 1669 if (hash != last_hash) 1670 fpos_offset = 2; 1671 last_hash = hash; 1672 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1673 } else { 1674 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1675 } 1676 1677 retry_lookup: 1678 dn = d_lookup(parent, &dname); 1679 dout("d_lookup on parent=%p name=%.*s got %p\n", 1680 parent, dname.len, dname.name, dn); 1681 1682 if (!dn) { 1683 dn = d_alloc(parent, &dname); 1684 dout("d_alloc %p '%.*s' = %p\n", parent, 1685 dname.len, dname.name, dn); 1686 if (!dn) { 1687 dout("d_alloc badness\n"); 1688 err = -ENOMEM; 1689 goto out; 1690 } 1691 } else if (d_really_is_positive(dn) && 1692 (ceph_ino(d_inode(dn)) != tvino.ino || 1693 ceph_snap(d_inode(dn)) != tvino.snap)) { 1694 struct ceph_dentry_info *di = ceph_dentry(dn); 1695 dout(" dn %p points to wrong inode %p\n", 1696 dn, d_inode(dn)); 1697 1698 spin_lock(&dn->d_lock); 1699 if (di->offset > 0 && 1700 di->lease_shared_gen == 1701 atomic_read(&ci->i_shared_gen)) { 1702 __ceph_dir_clear_ordered(ci); 1703 di->offset = 0; 1704 } 1705 spin_unlock(&dn->d_lock); 1706 1707 d_delete(dn); 1708 dput(dn); 1709 goto retry_lookup; 1710 } 1711 1712 /* inode */ 1713 if (d_really_is_positive(dn)) { 1714 in = d_inode(dn); 1715 } else { 1716 in = ceph_get_inode(parent->d_sb, tvino); 1717 if (IS_ERR(in)) { 1718 dout("new_inode badness\n"); 1719 d_drop(dn); 1720 dput(dn); 1721 err = PTR_ERR(in); 1722 goto out; 1723 } 1724 } 1725 1726 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1727 -1, &req->r_caps_reservation); 1728 if (ret < 0) { 1729 pr_err("ceph_fill_inode badness on %p\n", in); 1730 if (d_really_is_negative(dn)) { 1731 /* avoid calling iput_final() in mds 1732 * dispatch threads */ 1733 if (in->i_state & I_NEW) { 1734 ihold(in); 1735 discard_new_inode(in); 1736 } 1737 ceph_async_iput(in); 1738 } 1739 d_drop(dn); 1740 err = ret; 1741 goto next_item; 1742 } 1743 if (in->i_state & I_NEW) 1744 unlock_new_inode(in); 1745 1746 if (d_really_is_negative(dn)) { 1747 if (ceph_security_xattr_deadlock(in)) { 1748 dout(" skip splicing dn %p to inode %p" 1749 " (security xattr deadlock)\n", dn, in); 1750 ceph_async_iput(in); 1751 skipped++; 1752 goto next_item; 1753 } 1754 1755 err = splice_dentry(&dn, in); 1756 if (err < 0) 1757 goto next_item; 1758 } 1759 1760 ceph_dentry(dn)->offset = rde->offset; 1761 1762 update_dentry_lease(d_inode(parent), dn, 1763 rde->lease, req->r_session, 1764 req->r_request_started); 1765 1766 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1767 ret = fill_readdir_cache(d_inode(parent), dn, 1768 &cache_ctl, req); 1769 if (ret < 0) 1770 err = ret; 1771 } 1772 next_item: 1773 dput(dn); 1774 } 1775 out: 1776 if (err == 0 && skipped == 0) { 1777 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1778 req->r_readdir_cache_idx = cache_ctl.index; 1779 } 1780 ceph_readdir_cache_release(&cache_ctl); 1781 dout("readdir_prepopulate done\n"); 1782 return err; 1783 } 1784 1785 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1786 { 1787 struct ceph_inode_info *ci = ceph_inode(inode); 1788 bool ret; 1789 1790 spin_lock(&ci->i_ceph_lock); 1791 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1792 i_size_write(inode, size); 1793 inode->i_blocks = calc_inode_blocks(size); 1794 1795 ret = __ceph_should_report_size(ci); 1796 1797 spin_unlock(&ci->i_ceph_lock); 1798 return ret; 1799 } 1800 1801 /* 1802 * Put reference to inode, but avoid calling iput_final() in current thread. 1803 * iput_final() may wait for reahahead pages. The wait can cause deadlock in 1804 * some contexts. 1805 */ 1806 void ceph_async_iput(struct inode *inode) 1807 { 1808 if (!inode) 1809 return; 1810 for (;;) { 1811 if (atomic_add_unless(&inode->i_count, -1, 1)) 1812 break; 1813 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1814 &ceph_inode(inode)->i_work)) 1815 break; 1816 /* queue work failed, i_count must be at least 2 */ 1817 } 1818 } 1819 1820 void ceph_queue_inode_work(struct inode *inode, int work_bit) 1821 { 1822 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1823 struct ceph_inode_info *ci = ceph_inode(inode); 1824 set_bit(work_bit, &ci->i_work_mask); 1825 1826 ihold(inode); 1827 if (queue_work(fsc->inode_wq, &ci->i_work)) { 1828 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); 1829 } else { 1830 dout("queue_inode_work %p already queued, mask=%lx\n", 1831 inode, ci->i_work_mask); 1832 iput(inode); 1833 } 1834 } 1835 1836 static void ceph_do_invalidate_pages(struct inode *inode) 1837 { 1838 struct ceph_inode_info *ci = ceph_inode(inode); 1839 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1840 u32 orig_gen; 1841 int check = 0; 1842 1843 mutex_lock(&ci->i_truncate_mutex); 1844 1845 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) { 1846 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1847 inode, ceph_ino(inode)); 1848 mapping_set_error(inode->i_mapping, -EIO); 1849 truncate_pagecache(inode, 0); 1850 mutex_unlock(&ci->i_truncate_mutex); 1851 goto out; 1852 } 1853 1854 spin_lock(&ci->i_ceph_lock); 1855 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1856 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1857 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1858 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1859 check = 1; 1860 spin_unlock(&ci->i_ceph_lock); 1861 mutex_unlock(&ci->i_truncate_mutex); 1862 goto out; 1863 } 1864 orig_gen = ci->i_rdcache_gen; 1865 spin_unlock(&ci->i_ceph_lock); 1866 1867 ceph_fscache_invalidate(inode); 1868 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1869 pr_err("invalidate_pages %p fails\n", inode); 1870 } 1871 1872 spin_lock(&ci->i_ceph_lock); 1873 if (orig_gen == ci->i_rdcache_gen && 1874 orig_gen == ci->i_rdcache_revoking) { 1875 dout("invalidate_pages %p gen %d successful\n", inode, 1876 ci->i_rdcache_gen); 1877 ci->i_rdcache_revoking--; 1878 check = 1; 1879 } else { 1880 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1881 inode, orig_gen, ci->i_rdcache_gen, 1882 ci->i_rdcache_revoking); 1883 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1884 check = 1; 1885 } 1886 spin_unlock(&ci->i_ceph_lock); 1887 mutex_unlock(&ci->i_truncate_mutex); 1888 out: 1889 if (check) 1890 ceph_check_caps(ci, 0, NULL); 1891 } 1892 1893 /* 1894 * Make sure any pending truncation is applied before doing anything 1895 * that may depend on it. 1896 */ 1897 void __ceph_do_pending_vmtruncate(struct inode *inode) 1898 { 1899 struct ceph_inode_info *ci = ceph_inode(inode); 1900 u64 to; 1901 int wrbuffer_refs, finish = 0; 1902 1903 mutex_lock(&ci->i_truncate_mutex); 1904 retry: 1905 spin_lock(&ci->i_ceph_lock); 1906 if (ci->i_truncate_pending == 0) { 1907 dout("__do_pending_vmtruncate %p none pending\n", inode); 1908 spin_unlock(&ci->i_ceph_lock); 1909 mutex_unlock(&ci->i_truncate_mutex); 1910 return; 1911 } 1912 1913 /* 1914 * make sure any dirty snapped pages are flushed before we 1915 * possibly truncate them.. so write AND block! 1916 */ 1917 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1918 spin_unlock(&ci->i_ceph_lock); 1919 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1920 inode); 1921 filemap_write_and_wait_range(&inode->i_data, 0, 1922 inode->i_sb->s_maxbytes); 1923 goto retry; 1924 } 1925 1926 /* there should be no reader or writer */ 1927 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1928 1929 to = ci->i_truncate_size; 1930 wrbuffer_refs = ci->i_wrbuffer_ref; 1931 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1932 ci->i_truncate_pending, to); 1933 spin_unlock(&ci->i_ceph_lock); 1934 1935 truncate_pagecache(inode, to); 1936 1937 spin_lock(&ci->i_ceph_lock); 1938 if (to == ci->i_truncate_size) { 1939 ci->i_truncate_pending = 0; 1940 finish = 1; 1941 } 1942 spin_unlock(&ci->i_ceph_lock); 1943 if (!finish) 1944 goto retry; 1945 1946 mutex_unlock(&ci->i_truncate_mutex); 1947 1948 if (wrbuffer_refs == 0) 1949 ceph_check_caps(ci, 0, NULL); 1950 1951 wake_up_all(&ci->i_cap_wq); 1952 } 1953 1954 static void ceph_inode_work(struct work_struct *work) 1955 { 1956 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1957 i_work); 1958 struct inode *inode = &ci->vfs_inode; 1959 1960 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { 1961 dout("writeback %p\n", inode); 1962 filemap_fdatawrite(&inode->i_data); 1963 } 1964 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) 1965 ceph_do_invalidate_pages(inode); 1966 1967 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) 1968 __ceph_do_pending_vmtruncate(inode); 1969 1970 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) 1971 ceph_check_caps(ci, 0, NULL); 1972 1973 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) 1974 ceph_flush_snaps(ci, NULL); 1975 1976 iput(inode); 1977 } 1978 1979 /* 1980 * symlinks 1981 */ 1982 static const struct inode_operations ceph_symlink_iops = { 1983 .get_link = simple_get_link, 1984 .setattr = ceph_setattr, 1985 .getattr = ceph_getattr, 1986 .listxattr = ceph_listxattr, 1987 }; 1988 1989 int __ceph_setattr(struct inode *inode, struct iattr *attr) 1990 { 1991 struct ceph_inode_info *ci = ceph_inode(inode); 1992 unsigned int ia_valid = attr->ia_valid; 1993 struct ceph_mds_request *req; 1994 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1995 struct ceph_cap_flush *prealloc_cf; 1996 int issued; 1997 int release = 0, dirtied = 0; 1998 int mask = 0; 1999 int err = 0; 2000 int inode_dirty_flags = 0; 2001 bool lock_snap_rwsem = false; 2002 2003 prealloc_cf = ceph_alloc_cap_flush(); 2004 if (!prealloc_cf) 2005 return -ENOMEM; 2006 2007 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 2008 USE_AUTH_MDS); 2009 if (IS_ERR(req)) { 2010 ceph_free_cap_flush(prealloc_cf); 2011 return PTR_ERR(req); 2012 } 2013 2014 spin_lock(&ci->i_ceph_lock); 2015 issued = __ceph_caps_issued(ci, NULL); 2016 2017 if (!ci->i_head_snapc && 2018 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 2019 lock_snap_rwsem = true; 2020 if (!down_read_trylock(&mdsc->snap_rwsem)) { 2021 spin_unlock(&ci->i_ceph_lock); 2022 down_read(&mdsc->snap_rwsem); 2023 spin_lock(&ci->i_ceph_lock); 2024 issued = __ceph_caps_issued(ci, NULL); 2025 } 2026 } 2027 2028 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 2029 2030 if (ia_valid & ATTR_UID) { 2031 dout("setattr %p uid %d -> %d\n", inode, 2032 from_kuid(&init_user_ns, inode->i_uid), 2033 from_kuid(&init_user_ns, attr->ia_uid)); 2034 if (issued & CEPH_CAP_AUTH_EXCL) { 2035 inode->i_uid = attr->ia_uid; 2036 dirtied |= CEPH_CAP_AUTH_EXCL; 2037 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2038 !uid_eq(attr->ia_uid, inode->i_uid)) { 2039 req->r_args.setattr.uid = cpu_to_le32( 2040 from_kuid(&init_user_ns, attr->ia_uid)); 2041 mask |= CEPH_SETATTR_UID; 2042 release |= CEPH_CAP_AUTH_SHARED; 2043 } 2044 } 2045 if (ia_valid & ATTR_GID) { 2046 dout("setattr %p gid %d -> %d\n", inode, 2047 from_kgid(&init_user_ns, inode->i_gid), 2048 from_kgid(&init_user_ns, attr->ia_gid)); 2049 if (issued & CEPH_CAP_AUTH_EXCL) { 2050 inode->i_gid = attr->ia_gid; 2051 dirtied |= CEPH_CAP_AUTH_EXCL; 2052 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2053 !gid_eq(attr->ia_gid, inode->i_gid)) { 2054 req->r_args.setattr.gid = cpu_to_le32( 2055 from_kgid(&init_user_ns, attr->ia_gid)); 2056 mask |= CEPH_SETATTR_GID; 2057 release |= CEPH_CAP_AUTH_SHARED; 2058 } 2059 } 2060 if (ia_valid & ATTR_MODE) { 2061 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 2062 attr->ia_mode); 2063 if (issued & CEPH_CAP_AUTH_EXCL) { 2064 inode->i_mode = attr->ia_mode; 2065 dirtied |= CEPH_CAP_AUTH_EXCL; 2066 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2067 attr->ia_mode != inode->i_mode) { 2068 inode->i_mode = attr->ia_mode; 2069 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 2070 mask |= CEPH_SETATTR_MODE; 2071 release |= CEPH_CAP_AUTH_SHARED; 2072 } 2073 } 2074 2075 if (ia_valid & ATTR_ATIME) { 2076 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, 2077 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 2078 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 2079 if (issued & CEPH_CAP_FILE_EXCL) { 2080 ci->i_time_warp_seq++; 2081 inode->i_atime = attr->ia_atime; 2082 dirtied |= CEPH_CAP_FILE_EXCL; 2083 } else if ((issued & CEPH_CAP_FILE_WR) && 2084 timespec64_compare(&inode->i_atime, 2085 &attr->ia_atime) < 0) { 2086 inode->i_atime = attr->ia_atime; 2087 dirtied |= CEPH_CAP_FILE_WR; 2088 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2089 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { 2090 ceph_encode_timespec64(&req->r_args.setattr.atime, 2091 &attr->ia_atime); 2092 mask |= CEPH_SETATTR_ATIME; 2093 release |= CEPH_CAP_FILE_SHARED | 2094 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2095 } 2096 } 2097 if (ia_valid & ATTR_SIZE) { 2098 dout("setattr %p size %lld -> %lld\n", inode, 2099 inode->i_size, attr->ia_size); 2100 if ((issued & CEPH_CAP_FILE_EXCL) && 2101 attr->ia_size > inode->i_size) { 2102 i_size_write(inode, attr->ia_size); 2103 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2104 ci->i_reported_size = attr->ia_size; 2105 dirtied |= CEPH_CAP_FILE_EXCL; 2106 ia_valid |= ATTR_MTIME; 2107 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2108 attr->ia_size != inode->i_size) { 2109 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2110 req->r_args.setattr.old_size = 2111 cpu_to_le64(inode->i_size); 2112 mask |= CEPH_SETATTR_SIZE; 2113 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | 2114 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2115 } 2116 } 2117 if (ia_valid & ATTR_MTIME) { 2118 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, 2119 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 2120 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 2121 if (issued & CEPH_CAP_FILE_EXCL) { 2122 ci->i_time_warp_seq++; 2123 inode->i_mtime = attr->ia_mtime; 2124 dirtied |= CEPH_CAP_FILE_EXCL; 2125 } else if ((issued & CEPH_CAP_FILE_WR) && 2126 timespec64_compare(&inode->i_mtime, 2127 &attr->ia_mtime) < 0) { 2128 inode->i_mtime = attr->ia_mtime; 2129 dirtied |= CEPH_CAP_FILE_WR; 2130 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2131 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { 2132 ceph_encode_timespec64(&req->r_args.setattr.mtime, 2133 &attr->ia_mtime); 2134 mask |= CEPH_SETATTR_MTIME; 2135 release |= CEPH_CAP_FILE_SHARED | 2136 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2137 } 2138 } 2139 2140 /* these do nothing */ 2141 if (ia_valid & ATTR_CTIME) { 2142 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2143 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2144 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, 2145 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2146 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2147 only ? "ctime only" : "ignored"); 2148 if (only) { 2149 /* 2150 * if kernel wants to dirty ctime but nothing else, 2151 * we need to choose a cap to dirty under, or do 2152 * a almost-no-op setattr 2153 */ 2154 if (issued & CEPH_CAP_AUTH_EXCL) 2155 dirtied |= CEPH_CAP_AUTH_EXCL; 2156 else if (issued & CEPH_CAP_FILE_EXCL) 2157 dirtied |= CEPH_CAP_FILE_EXCL; 2158 else if (issued & CEPH_CAP_XATTR_EXCL) 2159 dirtied |= CEPH_CAP_XATTR_EXCL; 2160 else 2161 mask |= CEPH_SETATTR_CTIME; 2162 } 2163 } 2164 if (ia_valid & ATTR_FILE) 2165 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2166 2167 if (dirtied) { 2168 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2169 &prealloc_cf); 2170 inode->i_ctime = attr->ia_ctime; 2171 } 2172 2173 release &= issued; 2174 spin_unlock(&ci->i_ceph_lock); 2175 if (lock_snap_rwsem) 2176 up_read(&mdsc->snap_rwsem); 2177 2178 if (inode_dirty_flags) 2179 __mark_inode_dirty(inode, inode_dirty_flags); 2180 2181 2182 if (mask) { 2183 req->r_inode = inode; 2184 ihold(inode); 2185 req->r_inode_drop = release; 2186 req->r_args.setattr.mask = cpu_to_le32(mask); 2187 req->r_num_caps = 1; 2188 req->r_stamp = attr->ia_ctime; 2189 err = ceph_mdsc_do_request(mdsc, NULL, req); 2190 } 2191 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2192 ceph_cap_string(dirtied), mask); 2193 2194 ceph_mdsc_put_request(req); 2195 ceph_free_cap_flush(prealloc_cf); 2196 2197 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2198 __ceph_do_pending_vmtruncate(inode); 2199 2200 return err; 2201 } 2202 2203 /* 2204 * setattr 2205 */ 2206 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 2207 struct iattr *attr) 2208 { 2209 struct inode *inode = d_inode(dentry); 2210 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 2211 int err; 2212 2213 if (ceph_snap(inode) != CEPH_NOSNAP) 2214 return -EROFS; 2215 2216 err = setattr_prepare(&init_user_ns, dentry, attr); 2217 if (err != 0) 2218 return err; 2219 2220 if ((attr->ia_valid & ATTR_SIZE) && 2221 attr->ia_size > max(inode->i_size, fsc->max_file_size)) 2222 return -EFBIG; 2223 2224 if ((attr->ia_valid & ATTR_SIZE) && 2225 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) 2226 return -EDQUOT; 2227 2228 err = __ceph_setattr(inode, attr); 2229 2230 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2231 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode); 2232 2233 return err; 2234 } 2235 2236 /* 2237 * Verify that we have a lease on the given mask. If not, 2238 * do a getattr against an mds. 2239 */ 2240 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2241 int mask, bool force) 2242 { 2243 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2244 struct ceph_mds_client *mdsc = fsc->mdsc; 2245 struct ceph_mds_request *req; 2246 int mode; 2247 int err; 2248 2249 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2250 dout("do_getattr inode %p SNAPDIR\n", inode); 2251 return 0; 2252 } 2253 2254 dout("do_getattr inode %p mask %s mode 0%o\n", 2255 inode, ceph_cap_string(mask), inode->i_mode); 2256 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) 2257 return 0; 2258 2259 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; 2260 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2261 if (IS_ERR(req)) 2262 return PTR_ERR(req); 2263 req->r_inode = inode; 2264 ihold(inode); 2265 req->r_num_caps = 1; 2266 req->r_args.getattr.mask = cpu_to_le32(mask); 2267 req->r_locked_page = locked_page; 2268 err = ceph_mdsc_do_request(mdsc, NULL, req); 2269 if (locked_page && err == 0) { 2270 u64 inline_version = req->r_reply_info.targeti.inline_version; 2271 if (inline_version == 0) { 2272 /* the reply is supposed to contain inline data */ 2273 err = -EINVAL; 2274 } else if (inline_version == CEPH_INLINE_NONE) { 2275 err = -ENODATA; 2276 } else { 2277 err = req->r_reply_info.targeti.inline_len; 2278 } 2279 } 2280 ceph_mdsc_put_request(req); 2281 dout("do_getattr result=%d\n", err); 2282 return err; 2283 } 2284 2285 2286 /* 2287 * Check inode permissions. We verify we have a valid value for 2288 * the AUTH cap, then call the generic handler. 2289 */ 2290 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode, 2291 int mask) 2292 { 2293 int err; 2294 2295 if (mask & MAY_NOT_BLOCK) 2296 return -ECHILD; 2297 2298 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2299 2300 if (!err) 2301 err = generic_permission(&init_user_ns, inode, mask); 2302 return err; 2303 } 2304 2305 /* Craft a mask of needed caps given a set of requested statx attrs. */ 2306 static int statx_to_caps(u32 want, umode_t mode) 2307 { 2308 int mask = 0; 2309 2310 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) 2311 mask |= CEPH_CAP_AUTH_SHARED; 2312 2313 if (want & (STATX_NLINK|STATX_CTIME)) { 2314 /* 2315 * The link count for directories depends on inode->i_subdirs, 2316 * and that is only updated when Fs caps are held. 2317 */ 2318 if (S_ISDIR(mode)) 2319 mask |= CEPH_CAP_FILE_SHARED; 2320 else 2321 mask |= CEPH_CAP_LINK_SHARED; 2322 } 2323 2324 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| 2325 STATX_BLOCKS)) 2326 mask |= CEPH_CAP_FILE_SHARED; 2327 2328 if (want & (STATX_CTIME)) 2329 mask |= CEPH_CAP_XATTR_SHARED; 2330 2331 return mask; 2332 } 2333 2334 /* 2335 * Get all the attributes. If we have sufficient caps for the requested attrs, 2336 * then we can avoid talking to the MDS at all. 2337 */ 2338 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path, 2339 struct kstat *stat, u32 request_mask, unsigned int flags) 2340 { 2341 struct inode *inode = d_inode(path->dentry); 2342 struct ceph_inode_info *ci = ceph_inode(inode); 2343 u32 valid_mask = STATX_BASIC_STATS; 2344 int err = 0; 2345 2346 /* Skip the getattr altogether if we're asked not to sync */ 2347 if (!(flags & AT_STATX_DONT_SYNC)) { 2348 err = ceph_do_getattr(inode, 2349 statx_to_caps(request_mask, inode->i_mode), 2350 flags & AT_STATX_FORCE_SYNC); 2351 if (err) 2352 return err; 2353 } 2354 2355 generic_fillattr(&init_user_ns, inode, stat); 2356 stat->ino = ceph_present_inode(inode); 2357 2358 /* 2359 * btime on newly-allocated inodes is 0, so if this is still set to 2360 * that, then assume that it's not valid. 2361 */ 2362 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { 2363 stat->btime = ci->i_btime; 2364 valid_mask |= STATX_BTIME; 2365 } 2366 2367 if (ceph_snap(inode) == CEPH_NOSNAP) 2368 stat->dev = inode->i_sb->s_dev; 2369 else 2370 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; 2371 2372 if (S_ISDIR(inode->i_mode)) { 2373 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2374 RBYTES)) 2375 stat->size = ci->i_rbytes; 2376 else 2377 stat->size = ci->i_files + ci->i_subdirs; 2378 stat->blocks = 0; 2379 stat->blksize = 65536; 2380 /* 2381 * Some applications rely on the number of st_nlink 2382 * value on directories to be either 0 (if unlinked) 2383 * or 2 + number of subdirectories. 2384 */ 2385 if (stat->nlink == 1) 2386 /* '.' + '..' + subdirs */ 2387 stat->nlink = 1 + 1 + ci->i_subdirs; 2388 } 2389 2390 stat->result_mask = request_mask & valid_mask; 2391 return err; 2392 } 2393