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