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 if (ceph_vino_is_reserved(vino)) 60 return ERR_PTR(-EREMOTEIO); 61 62 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare, 63 ceph_set_ino_cb, &vino); 64 if (!inode) 65 return ERR_PTR(-ENOMEM); 66 67 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode), 68 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW)); 69 return inode; 70 } 71 72 /* 73 * get/constuct snapdir inode for a given directory 74 */ 75 struct inode *ceph_get_snapdir(struct inode *parent) 76 { 77 struct ceph_vino vino = { 78 .ino = ceph_ino(parent), 79 .snap = CEPH_SNAPDIR, 80 }; 81 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 82 struct ceph_inode_info *ci = ceph_inode(inode); 83 84 if (IS_ERR(inode)) 85 return inode; 86 87 if (!S_ISDIR(parent->i_mode)) { 88 pr_warn_once("bad snapdir parent type (mode=0%o)\n", 89 parent->i_mode); 90 goto err; 91 } 92 93 if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) { 94 pr_warn_once("bad snapdir inode type (mode=0%o)\n", 95 inode->i_mode); 96 goto err; 97 } 98 99 inode->i_mode = parent->i_mode; 100 inode->i_uid = parent->i_uid; 101 inode->i_gid = parent->i_gid; 102 inode->i_mtime = parent->i_mtime; 103 inode->i_ctime = parent->i_ctime; 104 inode->i_atime = parent->i_atime; 105 ci->i_rbytes = 0; 106 ci->i_btime = ceph_inode(parent)->i_btime; 107 108 if (inode->i_state & I_NEW) { 109 inode->i_op = &ceph_snapdir_iops; 110 inode->i_fop = &ceph_snapdir_fops; 111 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 112 unlock_new_inode(inode); 113 } 114 115 return inode; 116 err: 117 if ((inode->i_state & I_NEW)) 118 discard_new_inode(inode); 119 else 120 iput(inode); 121 return ERR_PTR(-ENOTDIR); 122 } 123 124 const struct inode_operations ceph_file_iops = { 125 .permission = ceph_permission, 126 .setattr = ceph_setattr, 127 .getattr = ceph_getattr, 128 .listxattr = ceph_listxattr, 129 .get_acl = ceph_get_acl, 130 .set_acl = ceph_set_acl, 131 }; 132 133 134 /* 135 * We use a 'frag tree' to keep track of the MDS's directory fragments 136 * for a given inode (usually there is just a single fragment). We 137 * need to know when a child frag is delegated to a new MDS, or when 138 * it is flagged as replicated, so we can direct our requests 139 * accordingly. 140 */ 141 142 /* 143 * find/create a frag in the tree 144 */ 145 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 146 u32 f) 147 { 148 struct rb_node **p; 149 struct rb_node *parent = NULL; 150 struct ceph_inode_frag *frag; 151 int c; 152 153 p = &ci->i_fragtree.rb_node; 154 while (*p) { 155 parent = *p; 156 frag = rb_entry(parent, struct ceph_inode_frag, node); 157 c = ceph_frag_compare(f, frag->frag); 158 if (c < 0) 159 p = &(*p)->rb_left; 160 else if (c > 0) 161 p = &(*p)->rb_right; 162 else 163 return frag; 164 } 165 166 frag = kmalloc(sizeof(*frag), GFP_NOFS); 167 if (!frag) 168 return ERR_PTR(-ENOMEM); 169 170 frag->frag = f; 171 frag->split_by = 0; 172 frag->mds = -1; 173 frag->ndist = 0; 174 175 rb_link_node(&frag->node, parent, p); 176 rb_insert_color(&frag->node, &ci->i_fragtree); 177 178 dout("get_or_create_frag added %llx.%llx frag %x\n", 179 ceph_vinop(&ci->vfs_inode), f); 180 return frag; 181 } 182 183 /* 184 * find a specific frag @f 185 */ 186 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 187 { 188 struct rb_node *n = ci->i_fragtree.rb_node; 189 190 while (n) { 191 struct ceph_inode_frag *frag = 192 rb_entry(n, struct ceph_inode_frag, node); 193 int c = ceph_frag_compare(f, frag->frag); 194 if (c < 0) 195 n = n->rb_left; 196 else if (c > 0) 197 n = n->rb_right; 198 else 199 return frag; 200 } 201 return NULL; 202 } 203 204 /* 205 * Choose frag containing the given value @v. If @pfrag is 206 * specified, copy the frag delegation info to the caller if 207 * it is present. 208 */ 209 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 210 struct ceph_inode_frag *pfrag, int *found) 211 { 212 u32 t = ceph_frag_make(0, 0); 213 struct ceph_inode_frag *frag; 214 unsigned nway, i; 215 u32 n; 216 217 if (found) 218 *found = 0; 219 220 while (1) { 221 WARN_ON(!ceph_frag_contains_value(t, v)); 222 frag = __ceph_find_frag(ci, t); 223 if (!frag) 224 break; /* t is a leaf */ 225 if (frag->split_by == 0) { 226 if (pfrag) 227 memcpy(pfrag, frag, sizeof(*pfrag)); 228 if (found) 229 *found = 1; 230 break; 231 } 232 233 /* choose child */ 234 nway = 1 << frag->split_by; 235 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 236 frag->split_by, nway); 237 for (i = 0; i < nway; i++) { 238 n = ceph_frag_make_child(t, frag->split_by, i); 239 if (ceph_frag_contains_value(n, v)) { 240 t = n; 241 break; 242 } 243 } 244 BUG_ON(i == nway); 245 } 246 dout("choose_frag(%x) = %x\n", v, t); 247 248 return t; 249 } 250 251 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 252 struct ceph_inode_frag *pfrag, int *found) 253 { 254 u32 ret; 255 mutex_lock(&ci->i_fragtree_mutex); 256 ret = __ceph_choose_frag(ci, v, pfrag, found); 257 mutex_unlock(&ci->i_fragtree_mutex); 258 return ret; 259 } 260 261 /* 262 * Process dirfrag (delegation) info from the mds. Include leaf 263 * fragment in tree ONLY if ndist > 0. Otherwise, only 264 * branches/splits are included in i_fragtree) 265 */ 266 static int ceph_fill_dirfrag(struct inode *inode, 267 struct ceph_mds_reply_dirfrag *dirinfo) 268 { 269 struct ceph_inode_info *ci = ceph_inode(inode); 270 struct ceph_inode_frag *frag; 271 u32 id = le32_to_cpu(dirinfo->frag); 272 int mds = le32_to_cpu(dirinfo->auth); 273 int ndist = le32_to_cpu(dirinfo->ndist); 274 int diri_auth = -1; 275 int i; 276 int err = 0; 277 278 spin_lock(&ci->i_ceph_lock); 279 if (ci->i_auth_cap) 280 diri_auth = ci->i_auth_cap->mds; 281 spin_unlock(&ci->i_ceph_lock); 282 283 if (mds == -1) /* CDIR_AUTH_PARENT */ 284 mds = diri_auth; 285 286 mutex_lock(&ci->i_fragtree_mutex); 287 if (ndist == 0 && mds == diri_auth) { 288 /* no delegation info needed. */ 289 frag = __ceph_find_frag(ci, id); 290 if (!frag) 291 goto out; 292 if (frag->split_by == 0) { 293 /* tree leaf, remove */ 294 dout("fill_dirfrag removed %llx.%llx frag %x" 295 " (no ref)\n", ceph_vinop(inode), id); 296 rb_erase(&frag->node, &ci->i_fragtree); 297 kfree(frag); 298 } else { 299 /* tree branch, keep and clear */ 300 dout("fill_dirfrag cleared %llx.%llx frag %x" 301 " referral\n", ceph_vinop(inode), id); 302 frag->mds = -1; 303 frag->ndist = 0; 304 } 305 goto out; 306 } 307 308 309 /* find/add this frag to store mds delegation info */ 310 frag = __get_or_create_frag(ci, id); 311 if (IS_ERR(frag)) { 312 /* this is not the end of the world; we can continue 313 with bad/inaccurate delegation info */ 314 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 315 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 316 err = -ENOMEM; 317 goto out; 318 } 319 320 frag->mds = mds; 321 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 322 for (i = 0; i < frag->ndist; i++) 323 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 324 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 325 ceph_vinop(inode), frag->frag, frag->ndist); 326 327 out: 328 mutex_unlock(&ci->i_fragtree_mutex); 329 return err; 330 } 331 332 static int frag_tree_split_cmp(const void *l, const void *r) 333 { 334 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; 335 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; 336 return ceph_frag_compare(le32_to_cpu(ls->frag), 337 le32_to_cpu(rs->frag)); 338 } 339 340 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) 341 { 342 if (!frag) 343 return f == ceph_frag_make(0, 0); 344 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) 345 return false; 346 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); 347 } 348 349 static int ceph_fill_fragtree(struct inode *inode, 350 struct ceph_frag_tree_head *fragtree, 351 struct ceph_mds_reply_dirfrag *dirinfo) 352 { 353 struct ceph_inode_info *ci = ceph_inode(inode); 354 struct ceph_inode_frag *frag, *prev_frag = NULL; 355 struct rb_node *rb_node; 356 unsigned i, split_by, nsplits; 357 u32 id; 358 bool update = false; 359 360 mutex_lock(&ci->i_fragtree_mutex); 361 nsplits = le32_to_cpu(fragtree->nsplits); 362 if (nsplits != ci->i_fragtree_nsplits) { 363 update = true; 364 } else if (nsplits) { 365 i = prandom_u32() % nsplits; 366 id = le32_to_cpu(fragtree->splits[i].frag); 367 if (!__ceph_find_frag(ci, id)) 368 update = true; 369 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 370 rb_node = rb_first(&ci->i_fragtree); 371 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 372 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 373 update = true; 374 } 375 if (!update && dirinfo) { 376 id = le32_to_cpu(dirinfo->frag); 377 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 378 update = true; 379 } 380 if (!update) 381 goto out_unlock; 382 383 if (nsplits > 1) { 384 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), 385 frag_tree_split_cmp, NULL); 386 } 387 388 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 389 rb_node = rb_first(&ci->i_fragtree); 390 for (i = 0; i < nsplits; i++) { 391 id = le32_to_cpu(fragtree->splits[i].frag); 392 split_by = le32_to_cpu(fragtree->splits[i].by); 393 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { 394 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " 395 "frag %x split by %d\n", ceph_vinop(inode), 396 i, nsplits, id, split_by); 397 continue; 398 } 399 frag = NULL; 400 while (rb_node) { 401 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 402 if (ceph_frag_compare(frag->frag, id) >= 0) { 403 if (frag->frag != id) 404 frag = NULL; 405 else 406 rb_node = rb_next(rb_node); 407 break; 408 } 409 rb_node = rb_next(rb_node); 410 /* delete stale split/leaf node */ 411 if (frag->split_by > 0 || 412 !is_frag_child(frag->frag, prev_frag)) { 413 rb_erase(&frag->node, &ci->i_fragtree); 414 if (frag->split_by > 0) 415 ci->i_fragtree_nsplits--; 416 kfree(frag); 417 } 418 frag = NULL; 419 } 420 if (!frag) { 421 frag = __get_or_create_frag(ci, id); 422 if (IS_ERR(frag)) 423 continue; 424 } 425 if (frag->split_by == 0) 426 ci->i_fragtree_nsplits++; 427 frag->split_by = split_by; 428 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 429 prev_frag = frag; 430 } 431 while (rb_node) { 432 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 433 rb_node = rb_next(rb_node); 434 /* delete stale split/leaf node */ 435 if (frag->split_by > 0 || 436 !is_frag_child(frag->frag, prev_frag)) { 437 rb_erase(&frag->node, &ci->i_fragtree); 438 if (frag->split_by > 0) 439 ci->i_fragtree_nsplits--; 440 kfree(frag); 441 } 442 } 443 out_unlock: 444 mutex_unlock(&ci->i_fragtree_mutex); 445 return 0; 446 } 447 448 /* 449 * initialize a newly allocated inode. 450 */ 451 struct inode *ceph_alloc_inode(struct super_block *sb) 452 { 453 struct ceph_inode_info *ci; 454 int i; 455 456 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 457 if (!ci) 458 return NULL; 459 460 dout("alloc_inode %p\n", &ci->vfs_inode); 461 462 spin_lock_init(&ci->i_ceph_lock); 463 464 ci->i_version = 0; 465 ci->i_inline_version = 0; 466 ci->i_time_warp_seq = 0; 467 ci->i_ceph_flags = 0; 468 atomic64_set(&ci->i_ordered_count, 1); 469 atomic64_set(&ci->i_release_count, 1); 470 atomic64_set(&ci->i_complete_seq[0], 0); 471 atomic64_set(&ci->i_complete_seq[1], 0); 472 ci->i_symlink = NULL; 473 474 ci->i_max_bytes = 0; 475 ci->i_max_files = 0; 476 477 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 478 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); 479 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); 480 481 ci->i_fragtree = RB_ROOT; 482 mutex_init(&ci->i_fragtree_mutex); 483 484 ci->i_xattrs.blob = NULL; 485 ci->i_xattrs.prealloc_blob = NULL; 486 ci->i_xattrs.dirty = false; 487 ci->i_xattrs.index = RB_ROOT; 488 ci->i_xattrs.count = 0; 489 ci->i_xattrs.names_size = 0; 490 ci->i_xattrs.vals_size = 0; 491 ci->i_xattrs.version = 0; 492 ci->i_xattrs.index_version = 0; 493 494 ci->i_caps = RB_ROOT; 495 ci->i_auth_cap = NULL; 496 ci->i_dirty_caps = 0; 497 ci->i_flushing_caps = 0; 498 INIT_LIST_HEAD(&ci->i_dirty_item); 499 INIT_LIST_HEAD(&ci->i_flushing_item); 500 ci->i_prealloc_cap_flush = NULL; 501 INIT_LIST_HEAD(&ci->i_cap_flush_list); 502 init_waitqueue_head(&ci->i_cap_wq); 503 ci->i_hold_caps_max = 0; 504 INIT_LIST_HEAD(&ci->i_cap_delay_list); 505 INIT_LIST_HEAD(&ci->i_cap_snaps); 506 ci->i_head_snapc = NULL; 507 ci->i_snap_caps = 0; 508 509 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ; 510 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) 511 ci->i_nr_by_mode[i] = 0; 512 513 mutex_init(&ci->i_truncate_mutex); 514 ci->i_truncate_seq = 0; 515 ci->i_truncate_size = 0; 516 ci->i_truncate_pending = 0; 517 518 ci->i_max_size = 0; 519 ci->i_reported_size = 0; 520 ci->i_wanted_max_size = 0; 521 ci->i_requested_max_size = 0; 522 523 ci->i_pin_ref = 0; 524 ci->i_rd_ref = 0; 525 ci->i_rdcache_ref = 0; 526 ci->i_wr_ref = 0; 527 ci->i_wb_ref = 0; 528 ci->i_fx_ref = 0; 529 ci->i_wrbuffer_ref = 0; 530 ci->i_wrbuffer_ref_head = 0; 531 atomic_set(&ci->i_filelock_ref, 0); 532 atomic_set(&ci->i_shared_gen, 1); 533 ci->i_rdcache_gen = 0; 534 ci->i_rdcache_revoking = 0; 535 536 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 537 INIT_LIST_HEAD(&ci->i_unsafe_iops); 538 spin_lock_init(&ci->i_unsafe_lock); 539 540 ci->i_snap_realm = NULL; 541 INIT_LIST_HEAD(&ci->i_snap_realm_item); 542 INIT_LIST_HEAD(&ci->i_snap_flush_item); 543 544 INIT_WORK(&ci->i_work, ceph_inode_work); 545 ci->i_work_mask = 0; 546 memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); 547 548 ceph_fscache_inode_init(ci); 549 550 return &ci->vfs_inode; 551 } 552 553 void ceph_free_inode(struct inode *inode) 554 { 555 struct ceph_inode_info *ci = ceph_inode(inode); 556 557 kfree(ci->i_symlink); 558 kmem_cache_free(ceph_inode_cachep, ci); 559 } 560 561 void ceph_evict_inode(struct inode *inode) 562 { 563 struct ceph_inode_info *ci = ceph_inode(inode); 564 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 565 struct ceph_inode_frag *frag; 566 struct rb_node *n; 567 568 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 569 570 percpu_counter_dec(&mdsc->metric.total_inodes); 571 572 truncate_inode_pages_final(&inode->i_data); 573 clear_inode(inode); 574 575 ceph_fscache_unregister_inode_cookie(ci); 576 577 __ceph_remove_caps(ci); 578 579 if (__ceph_has_any_quota(ci)) 580 ceph_adjust_quota_realms_count(inode, false); 581 582 /* 583 * we may still have a snap_realm reference if there are stray 584 * caps in i_snap_caps. 585 */ 586 if (ci->i_snap_realm) { 587 if (ceph_snap(inode) == CEPH_NOSNAP) { 588 dout(" dropping residual ref to snap realm %p\n", 589 ci->i_snap_realm); 590 ceph_change_snap_realm(inode, NULL); 591 } else { 592 ceph_put_snapid_map(mdsc, ci->i_snapid_map); 593 ci->i_snap_realm = NULL; 594 } 595 } 596 597 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 598 frag = rb_entry(n, struct ceph_inode_frag, node); 599 rb_erase(n, &ci->i_fragtree); 600 kfree(frag); 601 } 602 ci->i_fragtree_nsplits = 0; 603 604 __ceph_destroy_xattrs(ci); 605 if (ci->i_xattrs.blob) 606 ceph_buffer_put(ci->i_xattrs.blob); 607 if (ci->i_xattrs.prealloc_blob) 608 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 609 610 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); 611 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); 612 } 613 614 static inline blkcnt_t calc_inode_blocks(u64 size) 615 { 616 return (size + (1<<9) - 1) >> 9; 617 } 618 619 /* 620 * Helpers to fill in size, ctime, mtime, and atime. We have to be 621 * careful because either the client or MDS may have more up to date 622 * info, depending on which capabilities are held, and whether 623 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 624 * and size are monotonically increasing, except when utimes() or 625 * truncate() increments the corresponding _seq values.) 626 */ 627 int ceph_fill_file_size(struct inode *inode, int issued, 628 u32 truncate_seq, u64 truncate_size, u64 size) 629 { 630 struct ceph_inode_info *ci = ceph_inode(inode); 631 int queue_trunc = 0; 632 loff_t isize = i_size_read(inode); 633 634 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 635 (truncate_seq == ci->i_truncate_seq && size > isize)) { 636 dout("size %lld -> %llu\n", isize, size); 637 if (size > 0 && S_ISDIR(inode->i_mode)) { 638 pr_err("fill_file_size non-zero size for directory\n"); 639 size = 0; 640 } 641 i_size_write(inode, size); 642 inode->i_blocks = calc_inode_blocks(size); 643 ci->i_reported_size = size; 644 if (truncate_seq != ci->i_truncate_seq) { 645 dout("truncate_seq %u -> %u\n", 646 ci->i_truncate_seq, truncate_seq); 647 ci->i_truncate_seq = truncate_seq; 648 649 /* the MDS should have revoked these caps */ 650 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 651 CEPH_CAP_FILE_RD | 652 CEPH_CAP_FILE_WR | 653 CEPH_CAP_FILE_LAZYIO)); 654 /* 655 * If we hold relevant caps, or in the case where we're 656 * not the only client referencing this file and we 657 * don't hold those caps, then we need to check whether 658 * the file is either opened or mmaped 659 */ 660 if ((issued & (CEPH_CAP_FILE_CACHE| 661 CEPH_CAP_FILE_BUFFER)) || 662 mapping_mapped(inode->i_mapping) || 663 __ceph_is_file_opened(ci)) { 664 ci->i_truncate_pending++; 665 queue_trunc = 1; 666 } 667 } 668 } 669 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 670 ci->i_truncate_size != truncate_size) { 671 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 672 truncate_size); 673 ci->i_truncate_size = truncate_size; 674 } 675 676 if (queue_trunc) 677 ceph_fscache_invalidate(inode); 678 679 return queue_trunc; 680 } 681 682 void ceph_fill_file_time(struct inode *inode, int issued, 683 u64 time_warp_seq, struct timespec64 *ctime, 684 struct timespec64 *mtime, struct timespec64 *atime) 685 { 686 struct ceph_inode_info *ci = ceph_inode(inode); 687 int warn = 0; 688 689 if (issued & (CEPH_CAP_FILE_EXCL| 690 CEPH_CAP_FILE_WR| 691 CEPH_CAP_FILE_BUFFER| 692 CEPH_CAP_AUTH_EXCL| 693 CEPH_CAP_XATTR_EXCL)) { 694 if (ci->i_version == 0 || 695 timespec64_compare(ctime, &inode->i_ctime) > 0) { 696 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", 697 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 698 ctime->tv_sec, ctime->tv_nsec); 699 inode->i_ctime = *ctime; 700 } 701 if (ci->i_version == 0 || 702 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 703 /* the MDS did a utimes() */ 704 dout("mtime %lld.%09ld -> %lld.%09ld " 705 "tw %d -> %d\n", 706 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 707 mtime->tv_sec, mtime->tv_nsec, 708 ci->i_time_warp_seq, (int)time_warp_seq); 709 710 inode->i_mtime = *mtime; 711 inode->i_atime = *atime; 712 ci->i_time_warp_seq = time_warp_seq; 713 } else if (time_warp_seq == ci->i_time_warp_seq) { 714 /* nobody did utimes(); take the max */ 715 if (timespec64_compare(mtime, &inode->i_mtime) > 0) { 716 dout("mtime %lld.%09ld -> %lld.%09ld inc\n", 717 inode->i_mtime.tv_sec, 718 inode->i_mtime.tv_nsec, 719 mtime->tv_sec, mtime->tv_nsec); 720 inode->i_mtime = *mtime; 721 } 722 if (timespec64_compare(atime, &inode->i_atime) > 0) { 723 dout("atime %lld.%09ld -> %lld.%09ld inc\n", 724 inode->i_atime.tv_sec, 725 inode->i_atime.tv_nsec, 726 atime->tv_sec, atime->tv_nsec); 727 inode->i_atime = *atime; 728 } 729 } else if (issued & CEPH_CAP_FILE_EXCL) { 730 /* we did a utimes(); ignore mds values */ 731 } else { 732 warn = 1; 733 } 734 } else { 735 /* we have no write|excl caps; whatever the MDS says is true */ 736 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 737 inode->i_ctime = *ctime; 738 inode->i_mtime = *mtime; 739 inode->i_atime = *atime; 740 ci->i_time_warp_seq = time_warp_seq; 741 } else { 742 warn = 1; 743 } 744 } 745 if (warn) /* time_warp_seq shouldn't go backwards */ 746 dout("%p mds time_warp_seq %llu < %u\n", 747 inode, time_warp_seq, ci->i_time_warp_seq); 748 } 749 750 /* 751 * Populate an inode based on info from mds. May be called on new or 752 * existing inodes. 753 */ 754 int ceph_fill_inode(struct inode *inode, struct page *locked_page, 755 struct ceph_mds_reply_info_in *iinfo, 756 struct ceph_mds_reply_dirfrag *dirinfo, 757 struct ceph_mds_session *session, int cap_fmode, 758 struct ceph_cap_reservation *caps_reservation) 759 { 760 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 761 struct ceph_mds_reply_inode *info = iinfo->in; 762 struct ceph_inode_info *ci = ceph_inode(inode); 763 int issued, new_issued, info_caps; 764 struct timespec64 mtime, atime, ctime; 765 struct ceph_buffer *xattr_blob = NULL; 766 struct ceph_buffer *old_blob = NULL; 767 struct ceph_string *pool_ns = NULL; 768 struct ceph_cap *new_cap = NULL; 769 int err = 0; 770 bool wake = false; 771 bool queue_trunc = false; 772 bool new_version = false; 773 bool fill_inline = false; 774 umode_t mode = le32_to_cpu(info->mode); 775 dev_t rdev = le32_to_cpu(info->rdev); 776 777 lockdep_assert_held(&mdsc->snap_rwsem); 778 779 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, 780 inode, ceph_vinop(inode), le64_to_cpu(info->version), 781 ci->i_version); 782 783 /* Once I_NEW is cleared, we can't change type or dev numbers */ 784 if (inode->i_state & I_NEW) { 785 inode->i_mode = mode; 786 } else { 787 if (inode_wrong_type(inode, mode)) { 788 pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n", 789 ceph_vinop(inode), inode->i_mode, mode); 790 return -ESTALE; 791 } 792 793 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) { 794 pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n", 795 ceph_vinop(inode), MAJOR(inode->i_rdev), 796 MINOR(inode->i_rdev), MAJOR(rdev), 797 MINOR(rdev)); 798 return -ESTALE; 799 } 800 } 801 802 info_caps = le32_to_cpu(info->cap.caps); 803 804 /* prealloc new cap struct */ 805 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { 806 new_cap = ceph_get_cap(mdsc, caps_reservation); 807 if (!new_cap) 808 return -ENOMEM; 809 } 810 811 /* 812 * prealloc xattr data, if it looks like we'll need it. only 813 * if len > 4 (meaning there are actually xattrs; the first 4 814 * bytes are the xattr count). 815 */ 816 if (iinfo->xattr_len > 4) { 817 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 818 if (!xattr_blob) 819 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, 820 iinfo->xattr_len); 821 } 822 823 if (iinfo->pool_ns_len > 0) 824 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, 825 iinfo->pool_ns_len); 826 827 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) 828 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); 829 830 spin_lock(&ci->i_ceph_lock); 831 832 /* 833 * provided version will be odd if inode value is projected, 834 * even if stable. skip the update if we have newer stable 835 * info (ours>=theirs, e.g. due to racing mds replies), unless 836 * we are getting projected (unstable) info (in which case the 837 * version is odd, and we want ours>theirs). 838 * us them 839 * 2 2 skip 840 * 3 2 skip 841 * 3 3 update 842 */ 843 if (ci->i_version == 0 || 844 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 845 le64_to_cpu(info->version) > (ci->i_version & ~1))) 846 new_version = true; 847 848 /* Update change_attribute */ 849 inode_set_max_iversion_raw(inode, iinfo->change_attr); 850 851 __ceph_caps_issued(ci, &issued); 852 issued |= __ceph_caps_dirty(ci); 853 new_issued = ~issued & info_caps; 854 855 /* directories have fl_stripe_unit set to zero */ 856 if (le32_to_cpu(info->layout.fl_stripe_unit)) 857 inode->i_blkbits = 858 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 859 else 860 inode->i_blkbits = CEPH_BLOCK_SHIFT; 861 862 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); 863 864 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 865 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 866 inode->i_mode = mode; 867 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 868 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 869 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 870 from_kuid(&init_user_ns, inode->i_uid), 871 from_kgid(&init_user_ns, inode->i_gid)); 872 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); 873 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); 874 } 875 876 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 877 (issued & CEPH_CAP_LINK_EXCL) == 0) 878 set_nlink(inode, le32_to_cpu(info->nlink)); 879 880 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 881 /* be careful with mtime, atime, size */ 882 ceph_decode_timespec64(&atime, &info->atime); 883 ceph_decode_timespec64(&mtime, &info->mtime); 884 ceph_decode_timespec64(&ctime, &info->ctime); 885 ceph_fill_file_time(inode, issued, 886 le32_to_cpu(info->time_warp_seq), 887 &ctime, &mtime, &atime); 888 } 889 890 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { 891 ci->i_files = le64_to_cpu(info->files); 892 ci->i_subdirs = le64_to_cpu(info->subdirs); 893 } 894 895 if (new_version || 896 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 897 s64 old_pool = ci->i_layout.pool_id; 898 struct ceph_string *old_ns; 899 900 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); 901 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, 902 lockdep_is_held(&ci->i_ceph_lock)); 903 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); 904 905 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) 906 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; 907 908 pool_ns = old_ns; 909 910 queue_trunc = ceph_fill_file_size(inode, issued, 911 le32_to_cpu(info->truncate_seq), 912 le64_to_cpu(info->truncate_size), 913 le64_to_cpu(info->size)); 914 /* only update max_size on auth cap */ 915 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 916 ci->i_max_size != le64_to_cpu(info->max_size)) { 917 dout("max_size %lld -> %llu\n", ci->i_max_size, 918 le64_to_cpu(info->max_size)); 919 ci->i_max_size = le64_to_cpu(info->max_size); 920 } 921 } 922 923 /* layout and rstat are not tracked by capability, update them if 924 * the inode info is from auth mds */ 925 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { 926 if (S_ISDIR(inode->i_mode)) { 927 ci->i_dir_layout = iinfo->dir_layout; 928 ci->i_rbytes = le64_to_cpu(info->rbytes); 929 ci->i_rfiles = le64_to_cpu(info->rfiles); 930 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 931 ci->i_dir_pin = iinfo->dir_pin; 932 ci->i_rsnaps = iinfo->rsnaps; 933 ceph_decode_timespec64(&ci->i_rctime, &info->rctime); 934 } 935 } 936 937 /* xattrs */ 938 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 939 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 940 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 941 if (ci->i_xattrs.blob) 942 old_blob = ci->i_xattrs.blob; 943 ci->i_xattrs.blob = xattr_blob; 944 if (xattr_blob) 945 memcpy(ci->i_xattrs.blob->vec.iov_base, 946 iinfo->xattr_data, iinfo->xattr_len); 947 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 948 ceph_forget_all_cached_acls(inode); 949 ceph_security_invalidate_secctx(inode); 950 xattr_blob = NULL; 951 } 952 953 /* finally update i_version */ 954 if (le64_to_cpu(info->version) > ci->i_version) 955 ci->i_version = le64_to_cpu(info->version); 956 957 inode->i_mapping->a_ops = &ceph_aops; 958 959 switch (inode->i_mode & S_IFMT) { 960 case S_IFIFO: 961 case S_IFBLK: 962 case S_IFCHR: 963 case S_IFSOCK: 964 inode->i_blkbits = PAGE_SHIFT; 965 init_special_inode(inode, inode->i_mode, rdev); 966 inode->i_op = &ceph_file_iops; 967 break; 968 case S_IFREG: 969 inode->i_op = &ceph_file_iops; 970 inode->i_fop = &ceph_file_fops; 971 break; 972 case S_IFLNK: 973 inode->i_op = &ceph_symlink_iops; 974 if (!ci->i_symlink) { 975 u32 symlen = iinfo->symlink_len; 976 char *sym; 977 978 spin_unlock(&ci->i_ceph_lock); 979 980 if (symlen != i_size_read(inode)) { 981 pr_err("%s %llx.%llx BAD symlink " 982 "size %lld\n", __func__, 983 ceph_vinop(inode), 984 i_size_read(inode)); 985 i_size_write(inode, symlen); 986 inode->i_blocks = calc_inode_blocks(symlen); 987 } 988 989 err = -ENOMEM; 990 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 991 if (!sym) 992 goto out; 993 994 spin_lock(&ci->i_ceph_lock); 995 if (!ci->i_symlink) 996 ci->i_symlink = sym; 997 else 998 kfree(sym); /* lost a race */ 999 } 1000 inode->i_link = ci->i_symlink; 1001 break; 1002 case S_IFDIR: 1003 inode->i_op = &ceph_dir_iops; 1004 inode->i_fop = &ceph_dir_fops; 1005 break; 1006 default: 1007 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, 1008 ceph_vinop(inode), inode->i_mode); 1009 } 1010 1011 /* were we issued a capability? */ 1012 if (info_caps) { 1013 if (ceph_snap(inode) == CEPH_NOSNAP) { 1014 ceph_add_cap(inode, session, 1015 le64_to_cpu(info->cap.cap_id), 1016 info_caps, 1017 le32_to_cpu(info->cap.wanted), 1018 le32_to_cpu(info->cap.seq), 1019 le32_to_cpu(info->cap.mseq), 1020 le64_to_cpu(info->cap.realm), 1021 info->cap.flags, &new_cap); 1022 1023 /* set dir completion flag? */ 1024 if (S_ISDIR(inode->i_mode) && 1025 ci->i_files == 0 && ci->i_subdirs == 0 && 1026 (info_caps & CEPH_CAP_FILE_SHARED) && 1027 (issued & CEPH_CAP_FILE_EXCL) == 0 && 1028 !__ceph_dir_is_complete(ci)) { 1029 dout(" marking %p complete (empty)\n", inode); 1030 i_size_write(inode, 0); 1031 __ceph_dir_set_complete(ci, 1032 atomic64_read(&ci->i_release_count), 1033 atomic64_read(&ci->i_ordered_count)); 1034 } 1035 1036 wake = true; 1037 } else { 1038 dout(" %p got snap_caps %s\n", inode, 1039 ceph_cap_string(info_caps)); 1040 ci->i_snap_caps |= info_caps; 1041 } 1042 } 1043 1044 if (iinfo->inline_version > 0 && 1045 iinfo->inline_version >= ci->i_inline_version) { 1046 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1047 ci->i_inline_version = iinfo->inline_version; 1048 if (ci->i_inline_version != CEPH_INLINE_NONE && 1049 (locked_page || (info_caps & cache_caps))) 1050 fill_inline = true; 1051 } 1052 1053 if (cap_fmode >= 0) { 1054 if (!info_caps) 1055 pr_warn("mds issued no caps on %llx.%llx\n", 1056 ceph_vinop(inode)); 1057 __ceph_touch_fmode(ci, mdsc, cap_fmode); 1058 } 1059 1060 spin_unlock(&ci->i_ceph_lock); 1061 1062 if (fill_inline) 1063 ceph_fill_inline_data(inode, locked_page, 1064 iinfo->inline_data, iinfo->inline_len); 1065 1066 if (wake) 1067 wake_up_all(&ci->i_cap_wq); 1068 1069 /* queue truncate if we saw i_size decrease */ 1070 if (queue_trunc) 1071 ceph_queue_vmtruncate(inode); 1072 1073 /* populate frag tree */ 1074 if (S_ISDIR(inode->i_mode)) 1075 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 1076 1077 /* update delegation info? */ 1078 if (dirinfo) 1079 ceph_fill_dirfrag(inode, dirinfo); 1080 1081 err = 0; 1082 out: 1083 if (new_cap) 1084 ceph_put_cap(mdsc, new_cap); 1085 ceph_buffer_put(old_blob); 1086 ceph_buffer_put(xattr_blob); 1087 ceph_put_string(pool_ns); 1088 return err; 1089 } 1090 1091 /* 1092 * caller should hold session s_mutex and dentry->d_lock. 1093 */ 1094 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, 1095 struct ceph_mds_reply_lease *lease, 1096 struct ceph_mds_session *session, 1097 unsigned long from_time, 1098 struct ceph_mds_session **old_lease_session) 1099 { 1100 struct ceph_dentry_info *di = ceph_dentry(dentry); 1101 unsigned mask = le16_to_cpu(lease->mask); 1102 long unsigned duration = le32_to_cpu(lease->duration_ms); 1103 long unsigned ttl = from_time + (duration * HZ) / 1000; 1104 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1105 1106 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1107 dentry, duration, ttl); 1108 1109 /* only track leases on regular dentries */ 1110 if (ceph_snap(dir) != CEPH_NOSNAP) 1111 return; 1112 1113 if (mask & CEPH_LEASE_PRIMARY_LINK) 1114 di->flags |= CEPH_DENTRY_PRIMARY_LINK; 1115 else 1116 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1117 1118 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); 1119 if (!(mask & CEPH_LEASE_VALID)) { 1120 __ceph_dentry_dir_lease_touch(di); 1121 return; 1122 } 1123 1124 if (di->lease_gen == atomic_read(&session->s_cap_gen) && 1125 time_before(ttl, di->time)) 1126 return; /* we already have a newer lease. */ 1127 1128 if (di->lease_session && di->lease_session != session) { 1129 *old_lease_session = di->lease_session; 1130 di->lease_session = NULL; 1131 } 1132 1133 if (!di->lease_session) 1134 di->lease_session = ceph_get_mds_session(session); 1135 di->lease_gen = atomic_read(&session->s_cap_gen); 1136 di->lease_seq = le32_to_cpu(lease->seq); 1137 di->lease_renew_after = half_ttl; 1138 di->lease_renew_from = 0; 1139 di->time = ttl; 1140 1141 __ceph_dentry_lease_touch(di); 1142 } 1143 1144 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, 1145 struct ceph_mds_reply_lease *lease, 1146 struct ceph_mds_session *session, 1147 unsigned long from_time) 1148 { 1149 struct ceph_mds_session *old_lease_session = NULL; 1150 spin_lock(&dentry->d_lock); 1151 __update_dentry_lease(dir, dentry, lease, session, from_time, 1152 &old_lease_session); 1153 spin_unlock(&dentry->d_lock); 1154 ceph_put_mds_session(old_lease_session); 1155 } 1156 1157 /* 1158 * update dentry lease without having parent inode locked 1159 */ 1160 static void update_dentry_lease_careful(struct dentry *dentry, 1161 struct ceph_mds_reply_lease *lease, 1162 struct ceph_mds_session *session, 1163 unsigned long from_time, 1164 char *dname, u32 dname_len, 1165 struct ceph_vino *pdvino, 1166 struct ceph_vino *ptvino) 1167 1168 { 1169 struct inode *dir; 1170 struct ceph_mds_session *old_lease_session = NULL; 1171 1172 spin_lock(&dentry->d_lock); 1173 /* make sure dentry's name matches target */ 1174 if (dentry->d_name.len != dname_len || 1175 memcmp(dentry->d_name.name, dname, dname_len)) 1176 goto out_unlock; 1177 1178 dir = d_inode(dentry->d_parent); 1179 /* make sure parent matches dvino */ 1180 if (!ceph_ino_compare(dir, pdvino)) 1181 goto out_unlock; 1182 1183 /* make sure dentry's inode matches target. NULL ptvino means that 1184 * we expect a negative dentry */ 1185 if (ptvino) { 1186 if (d_really_is_negative(dentry)) 1187 goto out_unlock; 1188 if (!ceph_ino_compare(d_inode(dentry), ptvino)) 1189 goto out_unlock; 1190 } else { 1191 if (d_really_is_positive(dentry)) 1192 goto out_unlock; 1193 } 1194 1195 __update_dentry_lease(dir, dentry, lease, session, 1196 from_time, &old_lease_session); 1197 out_unlock: 1198 spin_unlock(&dentry->d_lock); 1199 ceph_put_mds_session(old_lease_session); 1200 } 1201 1202 /* 1203 * splice a dentry to an inode. 1204 * caller must hold directory i_mutex for this to be safe. 1205 */ 1206 static int splice_dentry(struct dentry **pdn, struct inode *in) 1207 { 1208 struct dentry *dn = *pdn; 1209 struct dentry *realdn; 1210 1211 BUG_ON(d_inode(dn)); 1212 1213 if (S_ISDIR(in->i_mode)) { 1214 /* If inode is directory, d_splice_alias() below will remove 1215 * 'realdn' from its origin parent. We need to ensure that 1216 * origin parent's readdir cache will not reference 'realdn' 1217 */ 1218 realdn = d_find_any_alias(in); 1219 if (realdn) { 1220 struct ceph_dentry_info *di = ceph_dentry(realdn); 1221 spin_lock(&realdn->d_lock); 1222 1223 realdn->d_op->d_prune(realdn); 1224 1225 di->time = jiffies; 1226 di->lease_shared_gen = 0; 1227 di->offset = 0; 1228 1229 spin_unlock(&realdn->d_lock); 1230 dput(realdn); 1231 } 1232 } 1233 1234 /* dn must be unhashed */ 1235 if (!d_unhashed(dn)) 1236 d_drop(dn); 1237 realdn = d_splice_alias(in, dn); 1238 if (IS_ERR(realdn)) { 1239 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1240 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1241 return PTR_ERR(realdn); 1242 } 1243 1244 if (realdn) { 1245 dout("dn %p (%d) spliced with %p (%d) " 1246 "inode %p ino %llx.%llx\n", 1247 dn, d_count(dn), 1248 realdn, d_count(realdn), 1249 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1250 dput(dn); 1251 *pdn = realdn; 1252 } else { 1253 BUG_ON(!ceph_dentry(dn)); 1254 dout("dn %p attached to %p ino %llx.%llx\n", 1255 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1256 } 1257 return 0; 1258 } 1259 1260 /* 1261 * Incorporate results into the local cache. This is either just 1262 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1263 * after a lookup). 1264 * 1265 * A reply may contain 1266 * a directory inode along with a dentry. 1267 * and/or a target inode 1268 * 1269 * Called with snap_rwsem (read). 1270 */ 1271 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1272 { 1273 struct ceph_mds_session *session = req->r_session; 1274 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1275 struct inode *in = NULL; 1276 struct ceph_vino tvino, dvino; 1277 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1278 int err = 0; 1279 1280 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1281 rinfo->head->is_dentry, rinfo->head->is_target); 1282 1283 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1284 dout("fill_trace reply is empty!\n"); 1285 if (rinfo->head->result == 0 && req->r_parent) 1286 ceph_invalidate_dir_request(req); 1287 return 0; 1288 } 1289 1290 if (rinfo->head->is_dentry) { 1291 struct inode *dir = req->r_parent; 1292 1293 if (dir) { 1294 err = ceph_fill_inode(dir, NULL, &rinfo->diri, 1295 rinfo->dirfrag, session, -1, 1296 &req->r_caps_reservation); 1297 if (err < 0) 1298 goto done; 1299 } else { 1300 WARN_ON_ONCE(1); 1301 } 1302 1303 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && 1304 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1305 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1306 struct qstr dname; 1307 struct dentry *dn, *parent; 1308 1309 BUG_ON(!rinfo->head->is_target); 1310 BUG_ON(req->r_dentry); 1311 1312 parent = d_find_any_alias(dir); 1313 BUG_ON(!parent); 1314 1315 dname.name = rinfo->dname; 1316 dname.len = rinfo->dname_len; 1317 dname.hash = full_name_hash(parent, dname.name, dname.len); 1318 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1319 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1320 retry_lookup: 1321 dn = d_lookup(parent, &dname); 1322 dout("d_lookup on parent=%p name=%.*s got %p\n", 1323 parent, dname.len, dname.name, dn); 1324 1325 if (!dn) { 1326 dn = d_alloc(parent, &dname); 1327 dout("d_alloc %p '%.*s' = %p\n", parent, 1328 dname.len, dname.name, dn); 1329 if (!dn) { 1330 dput(parent); 1331 err = -ENOMEM; 1332 goto done; 1333 } 1334 err = 0; 1335 } else if (d_really_is_positive(dn) && 1336 (ceph_ino(d_inode(dn)) != tvino.ino || 1337 ceph_snap(d_inode(dn)) != tvino.snap)) { 1338 dout(" dn %p points to wrong inode %p\n", 1339 dn, d_inode(dn)); 1340 ceph_dir_clear_ordered(dir); 1341 d_delete(dn); 1342 dput(dn); 1343 goto retry_lookup; 1344 } 1345 1346 req->r_dentry = dn; 1347 dput(parent); 1348 } 1349 } 1350 1351 if (rinfo->head->is_target) { 1352 /* Should be filled in by handle_reply */ 1353 BUG_ON(!req->r_target_inode); 1354 1355 in = req->r_target_inode; 1356 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, 1357 NULL, session, 1358 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1359 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && 1360 rinfo->head->result == 0) ? req->r_fmode : -1, 1361 &req->r_caps_reservation); 1362 if (err < 0) { 1363 pr_err("ceph_fill_inode badness %p %llx.%llx\n", 1364 in, ceph_vinop(in)); 1365 req->r_target_inode = NULL; 1366 if (in->i_state & I_NEW) 1367 discard_new_inode(in); 1368 else 1369 iput(in); 1370 goto done; 1371 } 1372 if (in->i_state & I_NEW) 1373 unlock_new_inode(in); 1374 } 1375 1376 /* 1377 * ignore null lease/binding on snapdir ENOENT, or else we 1378 * will have trouble splicing in the virtual snapdir later 1379 */ 1380 if (rinfo->head->is_dentry && 1381 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1382 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1383 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1384 fsc->mount_options->snapdir_name, 1385 req->r_dentry->d_name.len))) { 1386 /* 1387 * lookup link rename : null -> possibly existing inode 1388 * mknod symlink mkdir : null -> new inode 1389 * unlink : linked -> null 1390 */ 1391 struct inode *dir = req->r_parent; 1392 struct dentry *dn = req->r_dentry; 1393 bool have_dir_cap, have_lease; 1394 1395 BUG_ON(!dn); 1396 BUG_ON(!dir); 1397 BUG_ON(d_inode(dn->d_parent) != dir); 1398 1399 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1400 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1401 1402 BUG_ON(ceph_ino(dir) != dvino.ino); 1403 BUG_ON(ceph_snap(dir) != dvino.snap); 1404 1405 /* do we have a lease on the whole dir? */ 1406 have_dir_cap = 1407 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1408 CEPH_CAP_FILE_SHARED); 1409 1410 /* do we have a dn lease? */ 1411 have_lease = have_dir_cap || 1412 le32_to_cpu(rinfo->dlease->duration_ms); 1413 if (!have_lease) 1414 dout("fill_trace no dentry lease or dir cap\n"); 1415 1416 /* rename? */ 1417 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1418 struct inode *olddir = req->r_old_dentry_dir; 1419 BUG_ON(!olddir); 1420 1421 dout(" src %p '%pd' dst %p '%pd'\n", 1422 req->r_old_dentry, 1423 req->r_old_dentry, 1424 dn, dn); 1425 dout("fill_trace doing d_move %p -> %p\n", 1426 req->r_old_dentry, dn); 1427 1428 /* d_move screws up sibling dentries' offsets */ 1429 ceph_dir_clear_ordered(dir); 1430 ceph_dir_clear_ordered(olddir); 1431 1432 d_move(req->r_old_dentry, dn); 1433 dout(" src %p '%pd' dst %p '%pd'\n", 1434 req->r_old_dentry, 1435 req->r_old_dentry, 1436 dn, dn); 1437 1438 /* ensure target dentry is invalidated, despite 1439 rehashing bug in vfs_rename_dir */ 1440 ceph_invalidate_dentry_lease(dn); 1441 1442 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1443 ceph_dentry(req->r_old_dentry)->offset); 1444 1445 /* swap r_dentry and r_old_dentry in case that 1446 * splice_dentry() gets called later. This is safe 1447 * because no other place will use them */ 1448 req->r_dentry = req->r_old_dentry; 1449 req->r_old_dentry = dn; 1450 dn = req->r_dentry; 1451 } 1452 1453 /* null dentry? */ 1454 if (!rinfo->head->is_target) { 1455 dout("fill_trace null dentry\n"); 1456 if (d_really_is_positive(dn)) { 1457 dout("d_delete %p\n", dn); 1458 ceph_dir_clear_ordered(dir); 1459 d_delete(dn); 1460 } else if (have_lease) { 1461 if (d_unhashed(dn)) 1462 d_add(dn, NULL); 1463 update_dentry_lease(dir, dn, 1464 rinfo->dlease, session, 1465 req->r_request_started); 1466 } 1467 goto done; 1468 } 1469 1470 /* attach proper inode */ 1471 if (d_really_is_negative(dn)) { 1472 ceph_dir_clear_ordered(dir); 1473 ihold(in); 1474 err = splice_dentry(&req->r_dentry, in); 1475 if (err < 0) 1476 goto done; 1477 dn = req->r_dentry; /* may have spliced */ 1478 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1479 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1480 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1481 ceph_vinop(in)); 1482 d_invalidate(dn); 1483 have_lease = false; 1484 } 1485 1486 if (have_lease) { 1487 update_dentry_lease(dir, dn, 1488 rinfo->dlease, session, 1489 req->r_request_started); 1490 } 1491 dout(" final dn %p\n", dn); 1492 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1493 req->r_op == CEPH_MDS_OP_MKSNAP) && 1494 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1495 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1496 struct inode *dir = req->r_parent; 1497 1498 /* fill out a snapdir LOOKUPSNAP dentry */ 1499 BUG_ON(!dir); 1500 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1501 BUG_ON(!req->r_dentry); 1502 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); 1503 ceph_dir_clear_ordered(dir); 1504 ihold(in); 1505 err = splice_dentry(&req->r_dentry, in); 1506 if (err < 0) 1507 goto done; 1508 } else if (rinfo->head->is_dentry && req->r_dentry) { 1509 /* parent inode is not locked, be carefull */ 1510 struct ceph_vino *ptvino = NULL; 1511 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1512 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1513 if (rinfo->head->is_target) { 1514 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1515 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1516 ptvino = &tvino; 1517 } 1518 update_dentry_lease_careful(req->r_dentry, rinfo->dlease, 1519 session, req->r_request_started, 1520 rinfo->dname, rinfo->dname_len, 1521 &dvino, ptvino); 1522 } 1523 done: 1524 dout("fill_trace done err=%d\n", err); 1525 return err; 1526 } 1527 1528 /* 1529 * Prepopulate our cache with readdir results, leases, etc. 1530 */ 1531 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1532 struct ceph_mds_session *session) 1533 { 1534 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1535 int i, err = 0; 1536 1537 for (i = 0; i < rinfo->dir_nr; i++) { 1538 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1539 struct ceph_vino vino; 1540 struct inode *in; 1541 int rc; 1542 1543 vino.ino = le64_to_cpu(rde->inode.in->ino); 1544 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1545 1546 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1547 if (IS_ERR(in)) { 1548 err = PTR_ERR(in); 1549 dout("new_inode badness got %d\n", err); 1550 continue; 1551 } 1552 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1553 -1, &req->r_caps_reservation); 1554 if (rc < 0) { 1555 pr_err("ceph_fill_inode badness on %p got %d\n", 1556 in, rc); 1557 err = rc; 1558 if (in->i_state & I_NEW) { 1559 ihold(in); 1560 discard_new_inode(in); 1561 } 1562 } else if (in->i_state & I_NEW) { 1563 unlock_new_inode(in); 1564 } 1565 1566 iput(in); 1567 } 1568 1569 return err; 1570 } 1571 1572 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1573 { 1574 if (ctl->page) { 1575 kunmap(ctl->page); 1576 put_page(ctl->page); 1577 ctl->page = NULL; 1578 } 1579 } 1580 1581 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1582 struct ceph_readdir_cache_control *ctl, 1583 struct ceph_mds_request *req) 1584 { 1585 struct ceph_inode_info *ci = ceph_inode(dir); 1586 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1587 unsigned idx = ctl->index % nsize; 1588 pgoff_t pgoff = ctl->index / nsize; 1589 1590 if (!ctl->page || pgoff != page_index(ctl->page)) { 1591 ceph_readdir_cache_release(ctl); 1592 if (idx == 0) 1593 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1594 else 1595 ctl->page = find_lock_page(&dir->i_data, pgoff); 1596 if (!ctl->page) { 1597 ctl->index = -1; 1598 return idx == 0 ? -ENOMEM : 0; 1599 } 1600 /* reading/filling the cache are serialized by 1601 * i_mutex, no need to use page lock */ 1602 unlock_page(ctl->page); 1603 ctl->dentries = kmap(ctl->page); 1604 if (idx == 0) 1605 memset(ctl->dentries, 0, PAGE_SIZE); 1606 } 1607 1608 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1609 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1610 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1611 ctl->dentries[idx] = dn; 1612 ctl->index++; 1613 } else { 1614 dout("disable readdir cache\n"); 1615 ctl->index = -1; 1616 } 1617 return 0; 1618 } 1619 1620 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1621 struct ceph_mds_session *session) 1622 { 1623 struct dentry *parent = req->r_dentry; 1624 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1625 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1626 struct qstr dname; 1627 struct dentry *dn; 1628 struct inode *in; 1629 int err = 0, skipped = 0, ret, i; 1630 u32 frag = le32_to_cpu(req->r_args.readdir.frag); 1631 u32 last_hash = 0; 1632 u32 fpos_offset; 1633 struct ceph_readdir_cache_control cache_ctl = {}; 1634 1635 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1636 return readdir_prepopulate_inodes_only(req, session); 1637 1638 if (rinfo->hash_order) { 1639 if (req->r_path2) { 1640 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1641 req->r_path2, 1642 strlen(req->r_path2)); 1643 last_hash = ceph_frag_value(last_hash); 1644 } else if (rinfo->offset_hash) { 1645 /* mds understands offset_hash */ 1646 WARN_ON_ONCE(req->r_readdir_offset != 2); 1647 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); 1648 } 1649 } 1650 1651 if (rinfo->dir_dir && 1652 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1653 dout("readdir_prepopulate got new frag %x -> %x\n", 1654 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1655 frag = le32_to_cpu(rinfo->dir_dir->frag); 1656 if (!rinfo->hash_order) 1657 req->r_readdir_offset = 2; 1658 } 1659 1660 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1661 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1662 rinfo->dir_nr, parent); 1663 } else { 1664 dout("readdir_prepopulate %d items under dn %p\n", 1665 rinfo->dir_nr, parent); 1666 if (rinfo->dir_dir) 1667 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1668 1669 if (ceph_frag_is_leftmost(frag) && 1670 req->r_readdir_offset == 2 && 1671 !(rinfo->hash_order && last_hash)) { 1672 /* note dir version at start of readdir so we can 1673 * tell if any dentries get dropped */ 1674 req->r_dir_release_cnt = 1675 atomic64_read(&ci->i_release_count); 1676 req->r_dir_ordered_cnt = 1677 atomic64_read(&ci->i_ordered_count); 1678 req->r_readdir_cache_idx = 0; 1679 } 1680 } 1681 1682 cache_ctl.index = req->r_readdir_cache_idx; 1683 fpos_offset = req->r_readdir_offset; 1684 1685 /* FIXME: release caps/leases if error occurs */ 1686 for (i = 0; i < rinfo->dir_nr; i++) { 1687 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1688 struct ceph_vino tvino; 1689 1690 dname.name = rde->name; 1691 dname.len = rde->name_len; 1692 dname.hash = full_name_hash(parent, dname.name, dname.len); 1693 1694 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1695 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1696 1697 if (rinfo->hash_order) { 1698 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1699 rde->name, rde->name_len); 1700 hash = ceph_frag_value(hash); 1701 if (hash != last_hash) 1702 fpos_offset = 2; 1703 last_hash = hash; 1704 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1705 } else { 1706 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1707 } 1708 1709 retry_lookup: 1710 dn = d_lookup(parent, &dname); 1711 dout("d_lookup on parent=%p name=%.*s got %p\n", 1712 parent, dname.len, dname.name, dn); 1713 1714 if (!dn) { 1715 dn = d_alloc(parent, &dname); 1716 dout("d_alloc %p '%.*s' = %p\n", parent, 1717 dname.len, dname.name, dn); 1718 if (!dn) { 1719 dout("d_alloc badness\n"); 1720 err = -ENOMEM; 1721 goto out; 1722 } 1723 } else if (d_really_is_positive(dn) && 1724 (ceph_ino(d_inode(dn)) != tvino.ino || 1725 ceph_snap(d_inode(dn)) != tvino.snap)) { 1726 struct ceph_dentry_info *di = ceph_dentry(dn); 1727 dout(" dn %p points to wrong inode %p\n", 1728 dn, d_inode(dn)); 1729 1730 spin_lock(&dn->d_lock); 1731 if (di->offset > 0 && 1732 di->lease_shared_gen == 1733 atomic_read(&ci->i_shared_gen)) { 1734 __ceph_dir_clear_ordered(ci); 1735 di->offset = 0; 1736 } 1737 spin_unlock(&dn->d_lock); 1738 1739 d_delete(dn); 1740 dput(dn); 1741 goto retry_lookup; 1742 } 1743 1744 /* inode */ 1745 if (d_really_is_positive(dn)) { 1746 in = d_inode(dn); 1747 } else { 1748 in = ceph_get_inode(parent->d_sb, tvino); 1749 if (IS_ERR(in)) { 1750 dout("new_inode badness\n"); 1751 d_drop(dn); 1752 dput(dn); 1753 err = PTR_ERR(in); 1754 goto out; 1755 } 1756 } 1757 1758 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1759 -1, &req->r_caps_reservation); 1760 if (ret < 0) { 1761 pr_err("ceph_fill_inode badness on %p\n", in); 1762 if (d_really_is_negative(dn)) { 1763 if (in->i_state & I_NEW) { 1764 ihold(in); 1765 discard_new_inode(in); 1766 } 1767 iput(in); 1768 } 1769 d_drop(dn); 1770 err = ret; 1771 goto next_item; 1772 } 1773 if (in->i_state & I_NEW) 1774 unlock_new_inode(in); 1775 1776 if (d_really_is_negative(dn)) { 1777 if (ceph_security_xattr_deadlock(in)) { 1778 dout(" skip splicing dn %p to inode %p" 1779 " (security xattr deadlock)\n", dn, in); 1780 iput(in); 1781 skipped++; 1782 goto next_item; 1783 } 1784 1785 err = splice_dentry(&dn, in); 1786 if (err < 0) 1787 goto next_item; 1788 } 1789 1790 ceph_dentry(dn)->offset = rde->offset; 1791 1792 update_dentry_lease(d_inode(parent), dn, 1793 rde->lease, req->r_session, 1794 req->r_request_started); 1795 1796 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1797 ret = fill_readdir_cache(d_inode(parent), dn, 1798 &cache_ctl, req); 1799 if (ret < 0) 1800 err = ret; 1801 } 1802 next_item: 1803 dput(dn); 1804 } 1805 out: 1806 if (err == 0 && skipped == 0) { 1807 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1808 req->r_readdir_cache_idx = cache_ctl.index; 1809 } 1810 ceph_readdir_cache_release(&cache_ctl); 1811 dout("readdir_prepopulate done\n"); 1812 return err; 1813 } 1814 1815 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1816 { 1817 struct ceph_inode_info *ci = ceph_inode(inode); 1818 bool ret; 1819 1820 spin_lock(&ci->i_ceph_lock); 1821 dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size); 1822 i_size_write(inode, size); 1823 inode->i_blocks = calc_inode_blocks(size); 1824 1825 ret = __ceph_should_report_size(ci); 1826 1827 spin_unlock(&ci->i_ceph_lock); 1828 return ret; 1829 } 1830 1831 void ceph_queue_inode_work(struct inode *inode, int work_bit) 1832 { 1833 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1834 struct ceph_inode_info *ci = ceph_inode(inode); 1835 set_bit(work_bit, &ci->i_work_mask); 1836 1837 ihold(inode); 1838 if (queue_work(fsc->inode_wq, &ci->i_work)) { 1839 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); 1840 } else { 1841 dout("queue_inode_work %p already queued, mask=%lx\n", 1842 inode, ci->i_work_mask); 1843 iput(inode); 1844 } 1845 } 1846 1847 static void ceph_do_invalidate_pages(struct inode *inode) 1848 { 1849 struct ceph_inode_info *ci = ceph_inode(inode); 1850 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1851 u32 orig_gen; 1852 int check = 0; 1853 1854 mutex_lock(&ci->i_truncate_mutex); 1855 1856 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) { 1857 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1858 inode, ceph_ino(inode)); 1859 mapping_set_error(inode->i_mapping, -EIO); 1860 truncate_pagecache(inode, 0); 1861 mutex_unlock(&ci->i_truncate_mutex); 1862 goto out; 1863 } 1864 1865 spin_lock(&ci->i_ceph_lock); 1866 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1867 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1868 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1869 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1870 check = 1; 1871 spin_unlock(&ci->i_ceph_lock); 1872 mutex_unlock(&ci->i_truncate_mutex); 1873 goto out; 1874 } 1875 orig_gen = ci->i_rdcache_gen; 1876 spin_unlock(&ci->i_ceph_lock); 1877 1878 ceph_fscache_invalidate(inode); 1879 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1880 pr_err("invalidate_pages %p fails\n", inode); 1881 } 1882 1883 spin_lock(&ci->i_ceph_lock); 1884 if (orig_gen == ci->i_rdcache_gen && 1885 orig_gen == ci->i_rdcache_revoking) { 1886 dout("invalidate_pages %p gen %d successful\n", inode, 1887 ci->i_rdcache_gen); 1888 ci->i_rdcache_revoking--; 1889 check = 1; 1890 } else { 1891 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1892 inode, orig_gen, ci->i_rdcache_gen, 1893 ci->i_rdcache_revoking); 1894 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1895 check = 1; 1896 } 1897 spin_unlock(&ci->i_ceph_lock); 1898 mutex_unlock(&ci->i_truncate_mutex); 1899 out: 1900 if (check) 1901 ceph_check_caps(ci, 0, NULL); 1902 } 1903 1904 /* 1905 * Make sure any pending truncation is applied before doing anything 1906 * that may depend on it. 1907 */ 1908 void __ceph_do_pending_vmtruncate(struct inode *inode) 1909 { 1910 struct ceph_inode_info *ci = ceph_inode(inode); 1911 u64 to; 1912 int wrbuffer_refs, finish = 0; 1913 1914 mutex_lock(&ci->i_truncate_mutex); 1915 retry: 1916 spin_lock(&ci->i_ceph_lock); 1917 if (ci->i_truncate_pending == 0) { 1918 dout("__do_pending_vmtruncate %p none pending\n", inode); 1919 spin_unlock(&ci->i_ceph_lock); 1920 mutex_unlock(&ci->i_truncate_mutex); 1921 return; 1922 } 1923 1924 /* 1925 * make sure any dirty snapped pages are flushed before we 1926 * possibly truncate them.. so write AND block! 1927 */ 1928 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1929 spin_unlock(&ci->i_ceph_lock); 1930 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1931 inode); 1932 filemap_write_and_wait_range(&inode->i_data, 0, 1933 inode->i_sb->s_maxbytes); 1934 goto retry; 1935 } 1936 1937 /* there should be no reader or writer */ 1938 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1939 1940 to = ci->i_truncate_size; 1941 wrbuffer_refs = ci->i_wrbuffer_ref; 1942 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1943 ci->i_truncate_pending, to); 1944 spin_unlock(&ci->i_ceph_lock); 1945 1946 truncate_pagecache(inode, to); 1947 1948 spin_lock(&ci->i_ceph_lock); 1949 if (to == ci->i_truncate_size) { 1950 ci->i_truncate_pending = 0; 1951 finish = 1; 1952 } 1953 spin_unlock(&ci->i_ceph_lock); 1954 if (!finish) 1955 goto retry; 1956 1957 mutex_unlock(&ci->i_truncate_mutex); 1958 1959 if (wrbuffer_refs == 0) 1960 ceph_check_caps(ci, 0, NULL); 1961 1962 wake_up_all(&ci->i_cap_wq); 1963 } 1964 1965 static void ceph_inode_work(struct work_struct *work) 1966 { 1967 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1968 i_work); 1969 struct inode *inode = &ci->vfs_inode; 1970 1971 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { 1972 dout("writeback %p\n", inode); 1973 filemap_fdatawrite(&inode->i_data); 1974 } 1975 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) 1976 ceph_do_invalidate_pages(inode); 1977 1978 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) 1979 __ceph_do_pending_vmtruncate(inode); 1980 1981 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) 1982 ceph_check_caps(ci, 0, NULL); 1983 1984 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) 1985 ceph_flush_snaps(ci, NULL); 1986 1987 iput(inode); 1988 } 1989 1990 /* 1991 * symlinks 1992 */ 1993 static const struct inode_operations ceph_symlink_iops = { 1994 .get_link = simple_get_link, 1995 .setattr = ceph_setattr, 1996 .getattr = ceph_getattr, 1997 .listxattr = ceph_listxattr, 1998 }; 1999 2000 int __ceph_setattr(struct inode *inode, struct iattr *attr) 2001 { 2002 struct ceph_inode_info *ci = ceph_inode(inode); 2003 unsigned int ia_valid = attr->ia_valid; 2004 struct ceph_mds_request *req; 2005 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 2006 struct ceph_cap_flush *prealloc_cf; 2007 int issued; 2008 int release = 0, dirtied = 0; 2009 int mask = 0; 2010 int err = 0; 2011 int inode_dirty_flags = 0; 2012 bool lock_snap_rwsem = false; 2013 2014 prealloc_cf = ceph_alloc_cap_flush(); 2015 if (!prealloc_cf) 2016 return -ENOMEM; 2017 2018 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 2019 USE_AUTH_MDS); 2020 if (IS_ERR(req)) { 2021 ceph_free_cap_flush(prealloc_cf); 2022 return PTR_ERR(req); 2023 } 2024 2025 spin_lock(&ci->i_ceph_lock); 2026 issued = __ceph_caps_issued(ci, NULL); 2027 2028 if (!ci->i_head_snapc && 2029 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 2030 lock_snap_rwsem = true; 2031 if (!down_read_trylock(&mdsc->snap_rwsem)) { 2032 spin_unlock(&ci->i_ceph_lock); 2033 down_read(&mdsc->snap_rwsem); 2034 spin_lock(&ci->i_ceph_lock); 2035 issued = __ceph_caps_issued(ci, NULL); 2036 } 2037 } 2038 2039 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 2040 2041 if (ia_valid & ATTR_UID) { 2042 dout("setattr %p uid %d -> %d\n", inode, 2043 from_kuid(&init_user_ns, inode->i_uid), 2044 from_kuid(&init_user_ns, attr->ia_uid)); 2045 if (issued & CEPH_CAP_AUTH_EXCL) { 2046 inode->i_uid = attr->ia_uid; 2047 dirtied |= CEPH_CAP_AUTH_EXCL; 2048 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2049 !uid_eq(attr->ia_uid, inode->i_uid)) { 2050 req->r_args.setattr.uid = cpu_to_le32( 2051 from_kuid(&init_user_ns, attr->ia_uid)); 2052 mask |= CEPH_SETATTR_UID; 2053 release |= CEPH_CAP_AUTH_SHARED; 2054 } 2055 } 2056 if (ia_valid & ATTR_GID) { 2057 dout("setattr %p gid %d -> %d\n", inode, 2058 from_kgid(&init_user_ns, inode->i_gid), 2059 from_kgid(&init_user_ns, attr->ia_gid)); 2060 if (issued & CEPH_CAP_AUTH_EXCL) { 2061 inode->i_gid = attr->ia_gid; 2062 dirtied |= CEPH_CAP_AUTH_EXCL; 2063 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2064 !gid_eq(attr->ia_gid, inode->i_gid)) { 2065 req->r_args.setattr.gid = cpu_to_le32( 2066 from_kgid(&init_user_ns, attr->ia_gid)); 2067 mask |= CEPH_SETATTR_GID; 2068 release |= CEPH_CAP_AUTH_SHARED; 2069 } 2070 } 2071 if (ia_valid & ATTR_MODE) { 2072 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 2073 attr->ia_mode); 2074 if (issued & CEPH_CAP_AUTH_EXCL) { 2075 inode->i_mode = attr->ia_mode; 2076 dirtied |= CEPH_CAP_AUTH_EXCL; 2077 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2078 attr->ia_mode != inode->i_mode) { 2079 inode->i_mode = attr->ia_mode; 2080 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 2081 mask |= CEPH_SETATTR_MODE; 2082 release |= CEPH_CAP_AUTH_SHARED; 2083 } 2084 } 2085 2086 if (ia_valid & ATTR_ATIME) { 2087 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, 2088 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 2089 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 2090 if (issued & CEPH_CAP_FILE_EXCL) { 2091 ci->i_time_warp_seq++; 2092 inode->i_atime = attr->ia_atime; 2093 dirtied |= CEPH_CAP_FILE_EXCL; 2094 } else if ((issued & CEPH_CAP_FILE_WR) && 2095 timespec64_compare(&inode->i_atime, 2096 &attr->ia_atime) < 0) { 2097 inode->i_atime = attr->ia_atime; 2098 dirtied |= CEPH_CAP_FILE_WR; 2099 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2100 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { 2101 ceph_encode_timespec64(&req->r_args.setattr.atime, 2102 &attr->ia_atime); 2103 mask |= CEPH_SETATTR_ATIME; 2104 release |= CEPH_CAP_FILE_SHARED | 2105 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2106 } 2107 } 2108 if (ia_valid & ATTR_SIZE) { 2109 loff_t isize = i_size_read(inode); 2110 2111 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size); 2112 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) { 2113 i_size_write(inode, attr->ia_size); 2114 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2115 ci->i_reported_size = attr->ia_size; 2116 dirtied |= CEPH_CAP_FILE_EXCL; 2117 ia_valid |= ATTR_MTIME; 2118 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2119 attr->ia_size != isize) { 2120 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2121 req->r_args.setattr.old_size = cpu_to_le64(isize); 2122 mask |= CEPH_SETATTR_SIZE; 2123 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | 2124 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2125 } 2126 } 2127 if (ia_valid & ATTR_MTIME) { 2128 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, 2129 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 2130 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 2131 if (issued & CEPH_CAP_FILE_EXCL) { 2132 ci->i_time_warp_seq++; 2133 inode->i_mtime = attr->ia_mtime; 2134 dirtied |= CEPH_CAP_FILE_EXCL; 2135 } else if ((issued & CEPH_CAP_FILE_WR) && 2136 timespec64_compare(&inode->i_mtime, 2137 &attr->ia_mtime) < 0) { 2138 inode->i_mtime = attr->ia_mtime; 2139 dirtied |= CEPH_CAP_FILE_WR; 2140 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2141 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { 2142 ceph_encode_timespec64(&req->r_args.setattr.mtime, 2143 &attr->ia_mtime); 2144 mask |= CEPH_SETATTR_MTIME; 2145 release |= CEPH_CAP_FILE_SHARED | 2146 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2147 } 2148 } 2149 2150 /* these do nothing */ 2151 if (ia_valid & ATTR_CTIME) { 2152 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2153 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2154 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, 2155 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2156 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2157 only ? "ctime only" : "ignored"); 2158 if (only) { 2159 /* 2160 * if kernel wants to dirty ctime but nothing else, 2161 * we need to choose a cap to dirty under, or do 2162 * a almost-no-op setattr 2163 */ 2164 if (issued & CEPH_CAP_AUTH_EXCL) 2165 dirtied |= CEPH_CAP_AUTH_EXCL; 2166 else if (issued & CEPH_CAP_FILE_EXCL) 2167 dirtied |= CEPH_CAP_FILE_EXCL; 2168 else if (issued & CEPH_CAP_XATTR_EXCL) 2169 dirtied |= CEPH_CAP_XATTR_EXCL; 2170 else 2171 mask |= CEPH_SETATTR_CTIME; 2172 } 2173 } 2174 if (ia_valid & ATTR_FILE) 2175 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2176 2177 if (dirtied) { 2178 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2179 &prealloc_cf); 2180 inode->i_ctime = attr->ia_ctime; 2181 } 2182 2183 release &= issued; 2184 spin_unlock(&ci->i_ceph_lock); 2185 if (lock_snap_rwsem) 2186 up_read(&mdsc->snap_rwsem); 2187 2188 if (inode_dirty_flags) 2189 __mark_inode_dirty(inode, inode_dirty_flags); 2190 2191 2192 if (mask) { 2193 req->r_inode = inode; 2194 ihold(inode); 2195 req->r_inode_drop = release; 2196 req->r_args.setattr.mask = cpu_to_le32(mask); 2197 req->r_num_caps = 1; 2198 req->r_stamp = attr->ia_ctime; 2199 err = ceph_mdsc_do_request(mdsc, NULL, req); 2200 } 2201 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2202 ceph_cap_string(dirtied), mask); 2203 2204 ceph_mdsc_put_request(req); 2205 ceph_free_cap_flush(prealloc_cf); 2206 2207 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2208 __ceph_do_pending_vmtruncate(inode); 2209 2210 return err; 2211 } 2212 2213 /* 2214 * setattr 2215 */ 2216 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 2217 struct iattr *attr) 2218 { 2219 struct inode *inode = d_inode(dentry); 2220 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 2221 int err; 2222 2223 if (ceph_snap(inode) != CEPH_NOSNAP) 2224 return -EROFS; 2225 2226 err = setattr_prepare(&init_user_ns, dentry, attr); 2227 if (err != 0) 2228 return err; 2229 2230 if ((attr->ia_valid & ATTR_SIZE) && 2231 attr->ia_size > max(i_size_read(inode), fsc->max_file_size)) 2232 return -EFBIG; 2233 2234 if ((attr->ia_valid & ATTR_SIZE) && 2235 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) 2236 return -EDQUOT; 2237 2238 err = __ceph_setattr(inode, attr); 2239 2240 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2241 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode); 2242 2243 return err; 2244 } 2245 2246 /* 2247 * Verify that we have a lease on the given mask. If not, 2248 * do a getattr against an mds. 2249 */ 2250 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2251 int mask, bool force) 2252 { 2253 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2254 struct ceph_mds_client *mdsc = fsc->mdsc; 2255 struct ceph_mds_request *req; 2256 int mode; 2257 int err; 2258 2259 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2260 dout("do_getattr inode %p SNAPDIR\n", inode); 2261 return 0; 2262 } 2263 2264 dout("do_getattr inode %p mask %s mode 0%o\n", 2265 inode, ceph_cap_string(mask), inode->i_mode); 2266 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) 2267 return 0; 2268 2269 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; 2270 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2271 if (IS_ERR(req)) 2272 return PTR_ERR(req); 2273 req->r_inode = inode; 2274 ihold(inode); 2275 req->r_num_caps = 1; 2276 req->r_args.getattr.mask = cpu_to_le32(mask); 2277 req->r_locked_page = locked_page; 2278 err = ceph_mdsc_do_request(mdsc, NULL, req); 2279 if (locked_page && err == 0) { 2280 u64 inline_version = req->r_reply_info.targeti.inline_version; 2281 if (inline_version == 0) { 2282 /* the reply is supposed to contain inline data */ 2283 err = -EINVAL; 2284 } else if (inline_version == CEPH_INLINE_NONE) { 2285 err = -ENODATA; 2286 } else { 2287 err = req->r_reply_info.targeti.inline_len; 2288 } 2289 } 2290 ceph_mdsc_put_request(req); 2291 dout("do_getattr result=%d\n", err); 2292 return err; 2293 } 2294 2295 2296 /* 2297 * Check inode permissions. We verify we have a valid value for 2298 * the AUTH cap, then call the generic handler. 2299 */ 2300 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode, 2301 int mask) 2302 { 2303 int err; 2304 2305 if (mask & MAY_NOT_BLOCK) 2306 return -ECHILD; 2307 2308 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2309 2310 if (!err) 2311 err = generic_permission(&init_user_ns, inode, mask); 2312 return err; 2313 } 2314 2315 /* Craft a mask of needed caps given a set of requested statx attrs. */ 2316 static int statx_to_caps(u32 want, umode_t mode) 2317 { 2318 int mask = 0; 2319 2320 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) 2321 mask |= CEPH_CAP_AUTH_SHARED; 2322 2323 if (want & (STATX_NLINK|STATX_CTIME)) { 2324 /* 2325 * The link count for directories depends on inode->i_subdirs, 2326 * and that is only updated when Fs caps are held. 2327 */ 2328 if (S_ISDIR(mode)) 2329 mask |= CEPH_CAP_FILE_SHARED; 2330 else 2331 mask |= CEPH_CAP_LINK_SHARED; 2332 } 2333 2334 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| 2335 STATX_BLOCKS)) 2336 mask |= CEPH_CAP_FILE_SHARED; 2337 2338 if (want & (STATX_CTIME)) 2339 mask |= CEPH_CAP_XATTR_SHARED; 2340 2341 return mask; 2342 } 2343 2344 /* 2345 * Get all the attributes. If we have sufficient caps for the requested attrs, 2346 * then we can avoid talking to the MDS at all. 2347 */ 2348 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path, 2349 struct kstat *stat, u32 request_mask, unsigned int flags) 2350 { 2351 struct inode *inode = d_inode(path->dentry); 2352 struct ceph_inode_info *ci = ceph_inode(inode); 2353 u32 valid_mask = STATX_BASIC_STATS; 2354 int err = 0; 2355 2356 /* Skip the getattr altogether if we're asked not to sync */ 2357 if (!(flags & AT_STATX_DONT_SYNC)) { 2358 err = ceph_do_getattr(inode, 2359 statx_to_caps(request_mask, inode->i_mode), 2360 flags & AT_STATX_FORCE_SYNC); 2361 if (err) 2362 return err; 2363 } 2364 2365 generic_fillattr(&init_user_ns, inode, stat); 2366 stat->ino = ceph_present_inode(inode); 2367 2368 /* 2369 * btime on newly-allocated inodes is 0, so if this is still set to 2370 * that, then assume that it's not valid. 2371 */ 2372 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { 2373 stat->btime = ci->i_btime; 2374 valid_mask |= STATX_BTIME; 2375 } 2376 2377 if (ceph_snap(inode) == CEPH_NOSNAP) 2378 stat->dev = inode->i_sb->s_dev; 2379 else 2380 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; 2381 2382 if (S_ISDIR(inode->i_mode)) { 2383 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2384 RBYTES)) 2385 stat->size = ci->i_rbytes; 2386 else 2387 stat->size = ci->i_files + ci->i_subdirs; 2388 stat->blocks = 0; 2389 stat->blksize = 65536; 2390 /* 2391 * Some applications rely on the number of st_nlink 2392 * value on directories to be either 0 (if unlinked) 2393 * or 2 + number of subdirectories. 2394 */ 2395 if (stat->nlink == 1) 2396 /* '.' + '..' + subdirs */ 2397 stat->nlink = 1 + 1 + ci->i_subdirs; 2398 } 2399 2400 stat->result_mask = request_mask & valid_mask; 2401 return err; 2402 } 2403