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