1 #include <linux/ceph/ceph_debug.h> 2 3 #include <linux/module.h> 4 #include <linux/fs.h> 5 #include <linux/slab.h> 6 #include <linux/string.h> 7 #include <linux/uaccess.h> 8 #include <linux/kernel.h> 9 #include <linux/writeback.h> 10 #include <linux/vmalloc.h> 11 #include <linux/xattr.h> 12 #include <linux/posix_acl.h> 13 #include <linux/random.h> 14 #include <linux/sort.h> 15 16 #include "super.h" 17 #include "mds_client.h" 18 #include "cache.h" 19 #include <linux/ceph/decode.h> 20 21 /* 22 * Ceph inode operations 23 * 24 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 25 * setattr, etc.), xattr helpers, and helpers for assimilating 26 * metadata returned by the MDS into our cache. 27 * 28 * Also define helpers for doing asynchronous writeback, invalidation, 29 * and truncation for the benefit of those who can't afford to block 30 * (typically because they are in the message handler path). 31 */ 32 33 static const struct inode_operations ceph_symlink_iops; 34 35 static void ceph_invalidate_work(struct work_struct *work); 36 static void ceph_writeback_work(struct work_struct *work); 37 static void ceph_vmtruncate_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 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 46 return 0; 47 } 48 49 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 50 { 51 struct inode *inode; 52 ino_t t = ceph_vino_to_ino(vino); 53 54 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 55 if (inode == NULL) 56 return ERR_PTR(-ENOMEM); 57 if (inode->i_state & I_NEW) { 58 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 59 inode, ceph_vinop(inode), (u64)inode->i_ino); 60 unlock_new_inode(inode); 61 } 62 63 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 64 vino.snap, inode); 65 return inode; 66 } 67 68 /* 69 * get/constuct snapdir inode for a given directory 70 */ 71 struct inode *ceph_get_snapdir(struct inode *parent) 72 { 73 struct ceph_vino vino = { 74 .ino = ceph_ino(parent), 75 .snap = CEPH_SNAPDIR, 76 }; 77 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 78 struct ceph_inode_info *ci = ceph_inode(inode); 79 80 BUG_ON(!S_ISDIR(parent->i_mode)); 81 if (IS_ERR(inode)) 82 return inode; 83 inode->i_mode = parent->i_mode; 84 inode->i_uid = parent->i_uid; 85 inode->i_gid = parent->i_gid; 86 inode->i_op = &ceph_snapdir_iops; 87 inode->i_fop = &ceph_snapdir_fops; 88 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 89 ci->i_rbytes = 0; 90 return inode; 91 } 92 93 const struct inode_operations ceph_file_iops = { 94 .permission = ceph_permission, 95 .setattr = ceph_setattr, 96 .getattr = ceph_getattr, 97 .listxattr = ceph_listxattr, 98 .get_acl = ceph_get_acl, 99 .set_acl = ceph_set_acl, 100 }; 101 102 103 /* 104 * We use a 'frag tree' to keep track of the MDS's directory fragments 105 * for a given inode (usually there is just a single fragment). We 106 * need to know when a child frag is delegated to a new MDS, or when 107 * it is flagged as replicated, so we can direct our requests 108 * accordingly. 109 */ 110 111 /* 112 * find/create a frag in the tree 113 */ 114 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 115 u32 f) 116 { 117 struct rb_node **p; 118 struct rb_node *parent = NULL; 119 struct ceph_inode_frag *frag; 120 int c; 121 122 p = &ci->i_fragtree.rb_node; 123 while (*p) { 124 parent = *p; 125 frag = rb_entry(parent, struct ceph_inode_frag, node); 126 c = ceph_frag_compare(f, frag->frag); 127 if (c < 0) 128 p = &(*p)->rb_left; 129 else if (c > 0) 130 p = &(*p)->rb_right; 131 else 132 return frag; 133 } 134 135 frag = kmalloc(sizeof(*frag), GFP_NOFS); 136 if (!frag) 137 return ERR_PTR(-ENOMEM); 138 139 frag->frag = f; 140 frag->split_by = 0; 141 frag->mds = -1; 142 frag->ndist = 0; 143 144 rb_link_node(&frag->node, parent, p); 145 rb_insert_color(&frag->node, &ci->i_fragtree); 146 147 dout("get_or_create_frag added %llx.%llx frag %x\n", 148 ceph_vinop(&ci->vfs_inode), f); 149 return frag; 150 } 151 152 /* 153 * find a specific frag @f 154 */ 155 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 156 { 157 struct rb_node *n = ci->i_fragtree.rb_node; 158 159 while (n) { 160 struct ceph_inode_frag *frag = 161 rb_entry(n, struct ceph_inode_frag, node); 162 int c = ceph_frag_compare(f, frag->frag); 163 if (c < 0) 164 n = n->rb_left; 165 else if (c > 0) 166 n = n->rb_right; 167 else 168 return frag; 169 } 170 return NULL; 171 } 172 173 /* 174 * Choose frag containing the given value @v. If @pfrag is 175 * specified, copy the frag delegation info to the caller if 176 * it is present. 177 */ 178 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 179 struct ceph_inode_frag *pfrag, int *found) 180 { 181 u32 t = ceph_frag_make(0, 0); 182 struct ceph_inode_frag *frag; 183 unsigned nway, i; 184 u32 n; 185 186 if (found) 187 *found = 0; 188 189 while (1) { 190 WARN_ON(!ceph_frag_contains_value(t, v)); 191 frag = __ceph_find_frag(ci, t); 192 if (!frag) 193 break; /* t is a leaf */ 194 if (frag->split_by == 0) { 195 if (pfrag) 196 memcpy(pfrag, frag, sizeof(*pfrag)); 197 if (found) 198 *found = 1; 199 break; 200 } 201 202 /* choose child */ 203 nway = 1 << frag->split_by; 204 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 205 frag->split_by, nway); 206 for (i = 0; i < nway; i++) { 207 n = ceph_frag_make_child(t, frag->split_by, i); 208 if (ceph_frag_contains_value(n, v)) { 209 t = n; 210 break; 211 } 212 } 213 BUG_ON(i == nway); 214 } 215 dout("choose_frag(%x) = %x\n", v, t); 216 217 return t; 218 } 219 220 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 221 struct ceph_inode_frag *pfrag, int *found) 222 { 223 u32 ret; 224 mutex_lock(&ci->i_fragtree_mutex); 225 ret = __ceph_choose_frag(ci, v, pfrag, found); 226 mutex_unlock(&ci->i_fragtree_mutex); 227 return ret; 228 } 229 230 /* 231 * Process dirfrag (delegation) info from the mds. Include leaf 232 * fragment in tree ONLY if ndist > 0. Otherwise, only 233 * branches/splits are included in i_fragtree) 234 */ 235 static int ceph_fill_dirfrag(struct inode *inode, 236 struct ceph_mds_reply_dirfrag *dirinfo) 237 { 238 struct ceph_inode_info *ci = ceph_inode(inode); 239 struct ceph_inode_frag *frag; 240 u32 id = le32_to_cpu(dirinfo->frag); 241 int mds = le32_to_cpu(dirinfo->auth); 242 int ndist = le32_to_cpu(dirinfo->ndist); 243 int diri_auth = -1; 244 int i; 245 int err = 0; 246 247 spin_lock(&ci->i_ceph_lock); 248 if (ci->i_auth_cap) 249 diri_auth = ci->i_auth_cap->mds; 250 spin_unlock(&ci->i_ceph_lock); 251 252 if (mds == -1) /* CDIR_AUTH_PARENT */ 253 mds = diri_auth; 254 255 mutex_lock(&ci->i_fragtree_mutex); 256 if (ndist == 0 && mds == diri_auth) { 257 /* no delegation info needed. */ 258 frag = __ceph_find_frag(ci, id); 259 if (!frag) 260 goto out; 261 if (frag->split_by == 0) { 262 /* tree leaf, remove */ 263 dout("fill_dirfrag removed %llx.%llx frag %x" 264 " (no ref)\n", ceph_vinop(inode), id); 265 rb_erase(&frag->node, &ci->i_fragtree); 266 kfree(frag); 267 } else { 268 /* tree branch, keep and clear */ 269 dout("fill_dirfrag cleared %llx.%llx frag %x" 270 " referral\n", ceph_vinop(inode), id); 271 frag->mds = -1; 272 frag->ndist = 0; 273 } 274 goto out; 275 } 276 277 278 /* find/add this frag to store mds delegation info */ 279 frag = __get_or_create_frag(ci, id); 280 if (IS_ERR(frag)) { 281 /* this is not the end of the world; we can continue 282 with bad/inaccurate delegation info */ 283 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 284 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 285 err = -ENOMEM; 286 goto out; 287 } 288 289 frag->mds = mds; 290 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 291 for (i = 0; i < frag->ndist; i++) 292 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 293 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 294 ceph_vinop(inode), frag->frag, frag->ndist); 295 296 out: 297 mutex_unlock(&ci->i_fragtree_mutex); 298 return err; 299 } 300 301 static int frag_tree_split_cmp(const void *l, const void *r) 302 { 303 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; 304 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; 305 return ceph_frag_compare(le32_to_cpu(ls->frag), 306 le32_to_cpu(rs->frag)); 307 } 308 309 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) 310 { 311 if (!frag) 312 return f == ceph_frag_make(0, 0); 313 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) 314 return false; 315 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); 316 } 317 318 static int ceph_fill_fragtree(struct inode *inode, 319 struct ceph_frag_tree_head *fragtree, 320 struct ceph_mds_reply_dirfrag *dirinfo) 321 { 322 struct ceph_inode_info *ci = ceph_inode(inode); 323 struct ceph_inode_frag *frag, *prev_frag = NULL; 324 struct rb_node *rb_node; 325 unsigned i, split_by, nsplits; 326 u32 id; 327 bool update = false; 328 329 mutex_lock(&ci->i_fragtree_mutex); 330 nsplits = le32_to_cpu(fragtree->nsplits); 331 if (nsplits != ci->i_fragtree_nsplits) { 332 update = true; 333 } else if (nsplits) { 334 i = prandom_u32() % nsplits; 335 id = le32_to_cpu(fragtree->splits[i].frag); 336 if (!__ceph_find_frag(ci, id)) 337 update = true; 338 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 339 rb_node = rb_first(&ci->i_fragtree); 340 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 341 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 342 update = true; 343 } 344 if (!update && dirinfo) { 345 id = le32_to_cpu(dirinfo->frag); 346 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 347 update = true; 348 } 349 if (!update) 350 goto out_unlock; 351 352 if (nsplits > 1) { 353 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), 354 frag_tree_split_cmp, NULL); 355 } 356 357 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 358 rb_node = rb_first(&ci->i_fragtree); 359 for (i = 0; i < nsplits; i++) { 360 id = le32_to_cpu(fragtree->splits[i].frag); 361 split_by = le32_to_cpu(fragtree->splits[i].by); 362 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { 363 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " 364 "frag %x split by %d\n", ceph_vinop(inode), 365 i, nsplits, id, split_by); 366 continue; 367 } 368 frag = NULL; 369 while (rb_node) { 370 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 371 if (ceph_frag_compare(frag->frag, id) >= 0) { 372 if (frag->frag != id) 373 frag = NULL; 374 else 375 rb_node = rb_next(rb_node); 376 break; 377 } 378 rb_node = rb_next(rb_node); 379 /* delete stale split/leaf node */ 380 if (frag->split_by > 0 || 381 !is_frag_child(frag->frag, prev_frag)) { 382 rb_erase(&frag->node, &ci->i_fragtree); 383 if (frag->split_by > 0) 384 ci->i_fragtree_nsplits--; 385 kfree(frag); 386 } 387 frag = NULL; 388 } 389 if (!frag) { 390 frag = __get_or_create_frag(ci, id); 391 if (IS_ERR(frag)) 392 continue; 393 } 394 if (frag->split_by == 0) 395 ci->i_fragtree_nsplits++; 396 frag->split_by = split_by; 397 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 398 prev_frag = frag; 399 } 400 while (rb_node) { 401 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 402 rb_node = rb_next(rb_node); 403 /* delete stale split/leaf node */ 404 if (frag->split_by > 0 || 405 !is_frag_child(frag->frag, prev_frag)) { 406 rb_erase(&frag->node, &ci->i_fragtree); 407 if (frag->split_by > 0) 408 ci->i_fragtree_nsplits--; 409 kfree(frag); 410 } 411 } 412 out_unlock: 413 mutex_unlock(&ci->i_fragtree_mutex); 414 return 0; 415 } 416 417 /* 418 * initialize a newly allocated inode. 419 */ 420 struct inode *ceph_alloc_inode(struct super_block *sb) 421 { 422 struct ceph_inode_info *ci; 423 int i; 424 425 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 426 if (!ci) 427 return NULL; 428 429 dout("alloc_inode %p\n", &ci->vfs_inode); 430 431 spin_lock_init(&ci->i_ceph_lock); 432 433 ci->i_version = 0; 434 ci->i_inline_version = 0; 435 ci->i_time_warp_seq = 0; 436 ci->i_ceph_flags = 0; 437 atomic64_set(&ci->i_ordered_count, 1); 438 atomic64_set(&ci->i_release_count, 1); 439 atomic64_set(&ci->i_complete_seq[0], 0); 440 atomic64_set(&ci->i_complete_seq[1], 0); 441 ci->i_symlink = NULL; 442 443 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 444 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); 445 446 ci->i_fragtree = RB_ROOT; 447 mutex_init(&ci->i_fragtree_mutex); 448 449 ci->i_xattrs.blob = NULL; 450 ci->i_xattrs.prealloc_blob = NULL; 451 ci->i_xattrs.dirty = false; 452 ci->i_xattrs.index = RB_ROOT; 453 ci->i_xattrs.count = 0; 454 ci->i_xattrs.names_size = 0; 455 ci->i_xattrs.vals_size = 0; 456 ci->i_xattrs.version = 0; 457 ci->i_xattrs.index_version = 0; 458 459 ci->i_caps = RB_ROOT; 460 ci->i_auth_cap = NULL; 461 ci->i_dirty_caps = 0; 462 ci->i_flushing_caps = 0; 463 INIT_LIST_HEAD(&ci->i_dirty_item); 464 INIT_LIST_HEAD(&ci->i_flushing_item); 465 ci->i_prealloc_cap_flush = NULL; 466 INIT_LIST_HEAD(&ci->i_cap_flush_list); 467 init_waitqueue_head(&ci->i_cap_wq); 468 ci->i_hold_caps_min = 0; 469 ci->i_hold_caps_max = 0; 470 INIT_LIST_HEAD(&ci->i_cap_delay_list); 471 INIT_LIST_HEAD(&ci->i_cap_snaps); 472 ci->i_head_snapc = NULL; 473 ci->i_snap_caps = 0; 474 475 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) 476 ci->i_nr_by_mode[i] = 0; 477 478 mutex_init(&ci->i_truncate_mutex); 479 ci->i_truncate_seq = 0; 480 ci->i_truncate_size = 0; 481 ci->i_truncate_pending = 0; 482 483 ci->i_max_size = 0; 484 ci->i_reported_size = 0; 485 ci->i_wanted_max_size = 0; 486 ci->i_requested_max_size = 0; 487 488 ci->i_pin_ref = 0; 489 ci->i_rd_ref = 0; 490 ci->i_rdcache_ref = 0; 491 ci->i_wr_ref = 0; 492 ci->i_wb_ref = 0; 493 ci->i_wrbuffer_ref = 0; 494 ci->i_wrbuffer_ref_head = 0; 495 ci->i_shared_gen = 0; 496 ci->i_rdcache_gen = 0; 497 ci->i_rdcache_revoking = 0; 498 499 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 500 INIT_LIST_HEAD(&ci->i_unsafe_iops); 501 spin_lock_init(&ci->i_unsafe_lock); 502 503 ci->i_snap_realm = NULL; 504 INIT_LIST_HEAD(&ci->i_snap_realm_item); 505 INIT_LIST_HEAD(&ci->i_snap_flush_item); 506 507 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 508 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 509 510 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 511 512 ceph_fscache_inode_init(ci); 513 514 return &ci->vfs_inode; 515 } 516 517 static void ceph_i_callback(struct rcu_head *head) 518 { 519 struct inode *inode = container_of(head, struct inode, i_rcu); 520 struct ceph_inode_info *ci = ceph_inode(inode); 521 522 kmem_cache_free(ceph_inode_cachep, ci); 523 } 524 525 void ceph_destroy_inode(struct inode *inode) 526 { 527 struct ceph_inode_info *ci = ceph_inode(inode); 528 struct ceph_inode_frag *frag; 529 struct rb_node *n; 530 531 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 532 533 ceph_fscache_unregister_inode_cookie(ci); 534 535 ceph_queue_caps_release(inode); 536 537 /* 538 * we may still have a snap_realm reference if there are stray 539 * caps in i_snap_caps. 540 */ 541 if (ci->i_snap_realm) { 542 struct ceph_mds_client *mdsc = 543 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 544 struct ceph_snap_realm *realm = ci->i_snap_realm; 545 546 dout(" dropping residual ref to snap realm %p\n", realm); 547 spin_lock(&realm->inodes_with_caps_lock); 548 list_del_init(&ci->i_snap_realm_item); 549 spin_unlock(&realm->inodes_with_caps_lock); 550 ceph_put_snap_realm(mdsc, realm); 551 } 552 553 kfree(ci->i_symlink); 554 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 555 frag = rb_entry(n, struct ceph_inode_frag, node); 556 rb_erase(n, &ci->i_fragtree); 557 kfree(frag); 558 } 559 ci->i_fragtree_nsplits = 0; 560 561 __ceph_destroy_xattrs(ci); 562 if (ci->i_xattrs.blob) 563 ceph_buffer_put(ci->i_xattrs.blob); 564 if (ci->i_xattrs.prealloc_blob) 565 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 566 567 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); 568 569 call_rcu(&inode->i_rcu, ceph_i_callback); 570 } 571 572 int ceph_drop_inode(struct inode *inode) 573 { 574 /* 575 * Positve dentry and corresponding inode are always accompanied 576 * in MDS reply. So no need to keep inode in the cache after 577 * dropping all its aliases. 578 */ 579 return 1; 580 } 581 582 static inline blkcnt_t calc_inode_blocks(u64 size) 583 { 584 return (size + (1<<9) - 1) >> 9; 585 } 586 587 /* 588 * Helpers to fill in size, ctime, mtime, and atime. We have to be 589 * careful because either the client or MDS may have more up to date 590 * info, depending on which capabilities are held, and whether 591 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 592 * and size are monotonically increasing, except when utimes() or 593 * truncate() increments the corresponding _seq values.) 594 */ 595 int ceph_fill_file_size(struct inode *inode, int issued, 596 u32 truncate_seq, u64 truncate_size, u64 size) 597 { 598 struct ceph_inode_info *ci = ceph_inode(inode); 599 int queue_trunc = 0; 600 601 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 602 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 603 dout("size %lld -> %llu\n", inode->i_size, size); 604 if (size > 0 && S_ISDIR(inode->i_mode)) { 605 pr_err("fill_file_size non-zero size for directory\n"); 606 size = 0; 607 } 608 i_size_write(inode, size); 609 inode->i_blocks = calc_inode_blocks(size); 610 ci->i_reported_size = size; 611 if (truncate_seq != ci->i_truncate_seq) { 612 dout("truncate_seq %u -> %u\n", 613 ci->i_truncate_seq, truncate_seq); 614 ci->i_truncate_seq = truncate_seq; 615 616 /* the MDS should have revoked these caps */ 617 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 618 CEPH_CAP_FILE_RD | 619 CEPH_CAP_FILE_WR | 620 CEPH_CAP_FILE_LAZYIO)); 621 /* 622 * If we hold relevant caps, or in the case where we're 623 * not the only client referencing this file and we 624 * don't hold those caps, then we need to check whether 625 * the file is either opened or mmaped 626 */ 627 if ((issued & (CEPH_CAP_FILE_CACHE| 628 CEPH_CAP_FILE_BUFFER)) || 629 mapping_mapped(inode->i_mapping) || 630 __ceph_caps_file_wanted(ci)) { 631 ci->i_truncate_pending++; 632 queue_trunc = 1; 633 } 634 } 635 } 636 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 637 ci->i_truncate_size != truncate_size) { 638 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 639 truncate_size); 640 ci->i_truncate_size = truncate_size; 641 } 642 643 if (queue_trunc) 644 ceph_fscache_invalidate(inode); 645 646 return queue_trunc; 647 } 648 649 void ceph_fill_file_time(struct inode *inode, int issued, 650 u64 time_warp_seq, struct timespec *ctime, 651 struct timespec *mtime, struct timespec *atime) 652 { 653 struct ceph_inode_info *ci = ceph_inode(inode); 654 int warn = 0; 655 656 if (issued & (CEPH_CAP_FILE_EXCL| 657 CEPH_CAP_FILE_WR| 658 CEPH_CAP_FILE_BUFFER| 659 CEPH_CAP_AUTH_EXCL| 660 CEPH_CAP_XATTR_EXCL)) { 661 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 662 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 663 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 664 ctime->tv_sec, ctime->tv_nsec); 665 inode->i_ctime = *ctime; 666 } 667 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 668 /* the MDS did a utimes() */ 669 dout("mtime %ld.%09ld -> %ld.%09ld " 670 "tw %d -> %d\n", 671 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 672 mtime->tv_sec, mtime->tv_nsec, 673 ci->i_time_warp_seq, (int)time_warp_seq); 674 675 inode->i_mtime = *mtime; 676 inode->i_atime = *atime; 677 ci->i_time_warp_seq = time_warp_seq; 678 } else if (time_warp_seq == ci->i_time_warp_seq) { 679 /* nobody did utimes(); take the max */ 680 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 681 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 682 inode->i_mtime.tv_sec, 683 inode->i_mtime.tv_nsec, 684 mtime->tv_sec, mtime->tv_nsec); 685 inode->i_mtime = *mtime; 686 } 687 if (timespec_compare(atime, &inode->i_atime) > 0) { 688 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 689 inode->i_atime.tv_sec, 690 inode->i_atime.tv_nsec, 691 atime->tv_sec, atime->tv_nsec); 692 inode->i_atime = *atime; 693 } 694 } else if (issued & CEPH_CAP_FILE_EXCL) { 695 /* we did a utimes(); ignore mds values */ 696 } else { 697 warn = 1; 698 } 699 } else { 700 /* we have no write|excl caps; whatever the MDS says is true */ 701 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 702 inode->i_ctime = *ctime; 703 inode->i_mtime = *mtime; 704 inode->i_atime = *atime; 705 ci->i_time_warp_seq = time_warp_seq; 706 } else { 707 warn = 1; 708 } 709 } 710 if (warn) /* time_warp_seq shouldn't go backwards */ 711 dout("%p mds time_warp_seq %llu < %u\n", 712 inode, time_warp_seq, ci->i_time_warp_seq); 713 } 714 715 /* 716 * Populate an inode based on info from mds. May be called on new or 717 * existing inodes. 718 */ 719 static int fill_inode(struct inode *inode, struct page *locked_page, 720 struct ceph_mds_reply_info_in *iinfo, 721 struct ceph_mds_reply_dirfrag *dirinfo, 722 struct ceph_mds_session *session, 723 unsigned long ttl_from, int cap_fmode, 724 struct ceph_cap_reservation *caps_reservation) 725 { 726 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 727 struct ceph_mds_reply_inode *info = iinfo->in; 728 struct ceph_inode_info *ci = ceph_inode(inode); 729 int issued = 0, implemented, new_issued; 730 struct timespec mtime, atime, ctime; 731 struct ceph_buffer *xattr_blob = NULL; 732 struct ceph_string *pool_ns = NULL; 733 struct ceph_cap *new_cap = NULL; 734 int err = 0; 735 bool wake = false; 736 bool queue_trunc = false; 737 bool new_version = false; 738 bool fill_inline = false; 739 740 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 741 inode, ceph_vinop(inode), le64_to_cpu(info->version), 742 ci->i_version); 743 744 /* prealloc new cap struct */ 745 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP) 746 new_cap = ceph_get_cap(mdsc, caps_reservation); 747 748 /* 749 * prealloc xattr data, if it looks like we'll need it. only 750 * if len > 4 (meaning there are actually xattrs; the first 4 751 * bytes are the xattr count). 752 */ 753 if (iinfo->xattr_len > 4) { 754 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 755 if (!xattr_blob) 756 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 757 iinfo->xattr_len); 758 } 759 760 if (iinfo->pool_ns_len > 0) 761 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, 762 iinfo->pool_ns_len); 763 764 spin_lock(&ci->i_ceph_lock); 765 766 /* 767 * provided version will be odd if inode value is projected, 768 * even if stable. skip the update if we have newer stable 769 * info (ours>=theirs, e.g. due to racing mds replies), unless 770 * we are getting projected (unstable) info (in which case the 771 * version is odd, and we want ours>theirs). 772 * us them 773 * 2 2 skip 774 * 3 2 skip 775 * 3 3 update 776 */ 777 if (ci->i_version == 0 || 778 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 779 le64_to_cpu(info->version) > (ci->i_version & ~1))) 780 new_version = true; 781 782 issued = __ceph_caps_issued(ci, &implemented); 783 issued |= implemented | __ceph_caps_dirty(ci); 784 new_issued = ~issued & le32_to_cpu(info->cap.caps); 785 786 /* update inode */ 787 ci->i_version = le64_to_cpu(info->version); 788 inode->i_version++; 789 inode->i_rdev = le32_to_cpu(info->rdev); 790 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 791 792 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 793 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 794 inode->i_mode = le32_to_cpu(info->mode); 795 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 796 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 797 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 798 from_kuid(&init_user_ns, inode->i_uid), 799 from_kgid(&init_user_ns, inode->i_gid)); 800 } 801 802 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 803 (issued & CEPH_CAP_LINK_EXCL) == 0) 804 set_nlink(inode, le32_to_cpu(info->nlink)); 805 806 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 807 /* be careful with mtime, atime, size */ 808 ceph_decode_timespec(&atime, &info->atime); 809 ceph_decode_timespec(&mtime, &info->mtime); 810 ceph_decode_timespec(&ctime, &info->ctime); 811 ceph_fill_file_time(inode, issued, 812 le32_to_cpu(info->time_warp_seq), 813 &ctime, &mtime, &atime); 814 } 815 816 if (new_version || 817 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 818 s64 old_pool = ci->i_layout.pool_id; 819 struct ceph_string *old_ns; 820 821 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); 822 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, 823 lockdep_is_held(&ci->i_ceph_lock)); 824 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); 825 826 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) 827 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; 828 829 pool_ns = old_ns; 830 831 queue_trunc = ceph_fill_file_size(inode, issued, 832 le32_to_cpu(info->truncate_seq), 833 le64_to_cpu(info->truncate_size), 834 le64_to_cpu(info->size)); 835 /* only update max_size on auth cap */ 836 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 837 ci->i_max_size != le64_to_cpu(info->max_size)) { 838 dout("max_size %lld -> %llu\n", ci->i_max_size, 839 le64_to_cpu(info->max_size)); 840 ci->i_max_size = le64_to_cpu(info->max_size); 841 } 842 } 843 844 /* xattrs */ 845 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 846 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 847 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 848 if (ci->i_xattrs.blob) 849 ceph_buffer_put(ci->i_xattrs.blob); 850 ci->i_xattrs.blob = xattr_blob; 851 if (xattr_blob) 852 memcpy(ci->i_xattrs.blob->vec.iov_base, 853 iinfo->xattr_data, iinfo->xattr_len); 854 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 855 ceph_forget_all_cached_acls(inode); 856 xattr_blob = NULL; 857 } 858 859 inode->i_mapping->a_ops = &ceph_aops; 860 861 switch (inode->i_mode & S_IFMT) { 862 case S_IFIFO: 863 case S_IFBLK: 864 case S_IFCHR: 865 case S_IFSOCK: 866 init_special_inode(inode, inode->i_mode, inode->i_rdev); 867 inode->i_op = &ceph_file_iops; 868 break; 869 case S_IFREG: 870 inode->i_op = &ceph_file_iops; 871 inode->i_fop = &ceph_file_fops; 872 break; 873 case S_IFLNK: 874 inode->i_op = &ceph_symlink_iops; 875 if (!ci->i_symlink) { 876 u32 symlen = iinfo->symlink_len; 877 char *sym; 878 879 spin_unlock(&ci->i_ceph_lock); 880 881 if (symlen != i_size_read(inode)) { 882 pr_err("fill_inode %llx.%llx BAD symlink " 883 "size %lld\n", ceph_vinop(inode), 884 i_size_read(inode)); 885 i_size_write(inode, symlen); 886 inode->i_blocks = calc_inode_blocks(symlen); 887 } 888 889 err = -ENOMEM; 890 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 891 if (!sym) 892 goto out; 893 894 spin_lock(&ci->i_ceph_lock); 895 if (!ci->i_symlink) 896 ci->i_symlink = sym; 897 else 898 kfree(sym); /* lost a race */ 899 } 900 inode->i_link = ci->i_symlink; 901 break; 902 case S_IFDIR: 903 inode->i_op = &ceph_dir_iops; 904 inode->i_fop = &ceph_dir_fops; 905 906 ci->i_dir_layout = iinfo->dir_layout; 907 908 ci->i_files = le64_to_cpu(info->files); 909 ci->i_subdirs = le64_to_cpu(info->subdirs); 910 ci->i_rbytes = le64_to_cpu(info->rbytes); 911 ci->i_rfiles = le64_to_cpu(info->rfiles); 912 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 913 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 914 break; 915 default: 916 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 917 ceph_vinop(inode), inode->i_mode); 918 } 919 920 /* were we issued a capability? */ 921 if (info->cap.caps) { 922 if (ceph_snap(inode) == CEPH_NOSNAP) { 923 unsigned caps = le32_to_cpu(info->cap.caps); 924 ceph_add_cap(inode, session, 925 le64_to_cpu(info->cap.cap_id), 926 cap_fmode, caps, 927 le32_to_cpu(info->cap.wanted), 928 le32_to_cpu(info->cap.seq), 929 le32_to_cpu(info->cap.mseq), 930 le64_to_cpu(info->cap.realm), 931 info->cap.flags, &new_cap); 932 933 /* set dir completion flag? */ 934 if (S_ISDIR(inode->i_mode) && 935 ci->i_files == 0 && ci->i_subdirs == 0 && 936 (caps & CEPH_CAP_FILE_SHARED) && 937 (issued & CEPH_CAP_FILE_EXCL) == 0 && 938 !__ceph_dir_is_complete(ci)) { 939 dout(" marking %p complete (empty)\n", inode); 940 i_size_write(inode, 0); 941 __ceph_dir_set_complete(ci, 942 atomic64_read(&ci->i_release_count), 943 atomic64_read(&ci->i_ordered_count)); 944 } 945 946 wake = true; 947 } else { 948 dout(" %p got snap_caps %s\n", inode, 949 ceph_cap_string(le32_to_cpu(info->cap.caps))); 950 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 951 if (cap_fmode >= 0) 952 __ceph_get_fmode(ci, cap_fmode); 953 } 954 } else if (cap_fmode >= 0) { 955 pr_warn("mds issued no caps on %llx.%llx\n", 956 ceph_vinop(inode)); 957 __ceph_get_fmode(ci, cap_fmode); 958 } 959 960 if (iinfo->inline_version > 0 && 961 iinfo->inline_version >= ci->i_inline_version) { 962 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 963 ci->i_inline_version = iinfo->inline_version; 964 if (ci->i_inline_version != CEPH_INLINE_NONE && 965 (locked_page || 966 (le32_to_cpu(info->cap.caps) & cache_caps))) 967 fill_inline = true; 968 } 969 970 spin_unlock(&ci->i_ceph_lock); 971 972 if (fill_inline) 973 ceph_fill_inline_data(inode, locked_page, 974 iinfo->inline_data, iinfo->inline_len); 975 976 if (wake) 977 wake_up_all(&ci->i_cap_wq); 978 979 /* queue truncate if we saw i_size decrease */ 980 if (queue_trunc) 981 ceph_queue_vmtruncate(inode); 982 983 /* populate frag tree */ 984 if (S_ISDIR(inode->i_mode)) 985 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 986 987 /* update delegation info? */ 988 if (dirinfo) 989 ceph_fill_dirfrag(inode, dirinfo); 990 991 err = 0; 992 out: 993 if (new_cap) 994 ceph_put_cap(mdsc, new_cap); 995 if (xattr_blob) 996 ceph_buffer_put(xattr_blob); 997 ceph_put_string(pool_ns); 998 return err; 999 } 1000 1001 /* 1002 * caller should hold session s_mutex. 1003 */ 1004 static void update_dentry_lease(struct dentry *dentry, 1005 struct ceph_mds_reply_lease *lease, 1006 struct ceph_mds_session *session, 1007 unsigned long from_time, 1008 struct ceph_vino *tgt_vino, 1009 struct ceph_vino *dir_vino) 1010 { 1011 struct ceph_dentry_info *di = ceph_dentry(dentry); 1012 long unsigned duration = le32_to_cpu(lease->duration_ms); 1013 long unsigned ttl = from_time + (duration * HZ) / 1000; 1014 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1015 struct inode *dir; 1016 struct ceph_mds_session *old_lease_session = NULL; 1017 1018 /* 1019 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that 1020 * we expect a negative dentry. 1021 */ 1022 if (!tgt_vino && d_really_is_positive(dentry)) 1023 return; 1024 1025 if (tgt_vino && (d_really_is_negative(dentry) || 1026 !ceph_ino_compare(d_inode(dentry), tgt_vino))) 1027 return; 1028 1029 spin_lock(&dentry->d_lock); 1030 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1031 dentry, duration, ttl); 1032 1033 dir = d_inode(dentry->d_parent); 1034 1035 /* make sure parent matches dir_vino */ 1036 if (!ceph_ino_compare(dir, dir_vino)) 1037 goto out_unlock; 1038 1039 /* only track leases on regular dentries */ 1040 if (ceph_snap(dir) != CEPH_NOSNAP) 1041 goto out_unlock; 1042 1043 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 1044 1045 if (duration == 0) 1046 goto out_unlock; 1047 1048 if (di->lease_gen == session->s_cap_gen && 1049 time_before(ttl, di->time)) 1050 goto out_unlock; /* we already have a newer lease. */ 1051 1052 if (di->lease_session && di->lease_session != session) { 1053 old_lease_session = di->lease_session; 1054 di->lease_session = NULL; 1055 } 1056 1057 ceph_dentry_lru_touch(dentry); 1058 1059 if (!di->lease_session) 1060 di->lease_session = ceph_get_mds_session(session); 1061 di->lease_gen = session->s_cap_gen; 1062 di->lease_seq = le32_to_cpu(lease->seq); 1063 di->lease_renew_after = half_ttl; 1064 di->lease_renew_from = 0; 1065 di->time = ttl; 1066 out_unlock: 1067 spin_unlock(&dentry->d_lock); 1068 if (old_lease_session) 1069 ceph_put_mds_session(old_lease_session); 1070 return; 1071 } 1072 1073 /* 1074 * splice a dentry to an inode. 1075 * caller must hold directory i_mutex for this to be safe. 1076 */ 1077 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in) 1078 { 1079 struct dentry *realdn; 1080 1081 BUG_ON(d_inode(dn)); 1082 1083 /* dn must be unhashed */ 1084 if (!d_unhashed(dn)) 1085 d_drop(dn); 1086 realdn = d_splice_alias(in, dn); 1087 if (IS_ERR(realdn)) { 1088 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1089 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1090 dn = realdn; /* note realdn contains the error */ 1091 goto out; 1092 } else if (realdn) { 1093 dout("dn %p (%d) spliced with %p (%d) " 1094 "inode %p ino %llx.%llx\n", 1095 dn, d_count(dn), 1096 realdn, d_count(realdn), 1097 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1098 dput(dn); 1099 dn = realdn; 1100 } else { 1101 BUG_ON(!ceph_dentry(dn)); 1102 dout("dn %p attached to %p ino %llx.%llx\n", 1103 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1104 } 1105 out: 1106 return dn; 1107 } 1108 1109 /* 1110 * Incorporate results into the local cache. This is either just 1111 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1112 * after a lookup). 1113 * 1114 * A reply may contain 1115 * a directory inode along with a dentry. 1116 * and/or a target inode 1117 * 1118 * Called with snap_rwsem (read). 1119 */ 1120 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1121 { 1122 struct ceph_mds_session *session = req->r_session; 1123 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1124 struct inode *in = NULL; 1125 struct ceph_vino tvino, dvino; 1126 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1127 int err = 0; 1128 1129 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1130 rinfo->head->is_dentry, rinfo->head->is_target); 1131 1132 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1133 dout("fill_trace reply is empty!\n"); 1134 if (rinfo->head->result == 0 && req->r_parent) 1135 ceph_invalidate_dir_request(req); 1136 return 0; 1137 } 1138 1139 if (rinfo->head->is_dentry) { 1140 struct inode *dir = req->r_parent; 1141 1142 if (dir) { 1143 err = fill_inode(dir, NULL, 1144 &rinfo->diri, rinfo->dirfrag, 1145 session, req->r_request_started, -1, 1146 &req->r_caps_reservation); 1147 if (err < 0) 1148 goto done; 1149 } else { 1150 WARN_ON_ONCE(1); 1151 } 1152 1153 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { 1154 struct qstr dname; 1155 struct dentry *dn, *parent; 1156 1157 BUG_ON(!rinfo->head->is_target); 1158 BUG_ON(req->r_dentry); 1159 1160 parent = d_find_any_alias(dir); 1161 BUG_ON(!parent); 1162 1163 dname.name = rinfo->dname; 1164 dname.len = rinfo->dname_len; 1165 dname.hash = full_name_hash(parent, dname.name, dname.len); 1166 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1167 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1168 retry_lookup: 1169 dn = d_lookup(parent, &dname); 1170 dout("d_lookup on parent=%p name=%.*s got %p\n", 1171 parent, dname.len, dname.name, dn); 1172 1173 if (!dn) { 1174 dn = d_alloc(parent, &dname); 1175 dout("d_alloc %p '%.*s' = %p\n", parent, 1176 dname.len, dname.name, dn); 1177 if (dn == NULL) { 1178 dput(parent); 1179 err = -ENOMEM; 1180 goto done; 1181 } 1182 err = 0; 1183 } else if (d_really_is_positive(dn) && 1184 (ceph_ino(d_inode(dn)) != tvino.ino || 1185 ceph_snap(d_inode(dn)) != tvino.snap)) { 1186 dout(" dn %p points to wrong inode %p\n", 1187 dn, d_inode(dn)); 1188 d_delete(dn); 1189 dput(dn); 1190 goto retry_lookup; 1191 } 1192 1193 req->r_dentry = dn; 1194 dput(parent); 1195 } 1196 } 1197 1198 if (rinfo->head->is_target) { 1199 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1200 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1201 1202 in = ceph_get_inode(sb, tvino); 1203 if (IS_ERR(in)) { 1204 err = PTR_ERR(in); 1205 goto done; 1206 } 1207 req->r_target_inode = in; 1208 1209 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, 1210 session, req->r_request_started, 1211 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1212 rinfo->head->result == 0) ? req->r_fmode : -1, 1213 &req->r_caps_reservation); 1214 if (err < 0) { 1215 pr_err("fill_inode badness %p %llx.%llx\n", 1216 in, ceph_vinop(in)); 1217 goto done; 1218 } 1219 } 1220 1221 /* 1222 * ignore null lease/binding on snapdir ENOENT, or else we 1223 * will have trouble splicing in the virtual snapdir later 1224 */ 1225 if (rinfo->head->is_dentry && 1226 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1227 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1228 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1229 fsc->mount_options->snapdir_name, 1230 req->r_dentry->d_name.len))) { 1231 /* 1232 * lookup link rename : null -> possibly existing inode 1233 * mknod symlink mkdir : null -> new inode 1234 * unlink : linked -> null 1235 */ 1236 struct inode *dir = req->r_parent; 1237 struct dentry *dn = req->r_dentry; 1238 bool have_dir_cap, have_lease; 1239 1240 BUG_ON(!dn); 1241 BUG_ON(!dir); 1242 BUG_ON(d_inode(dn->d_parent) != dir); 1243 1244 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1245 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1246 1247 BUG_ON(ceph_ino(dir) != dvino.ino); 1248 BUG_ON(ceph_snap(dir) != dvino.snap); 1249 1250 /* do we have a lease on the whole dir? */ 1251 have_dir_cap = 1252 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1253 CEPH_CAP_FILE_SHARED); 1254 1255 /* do we have a dn lease? */ 1256 have_lease = have_dir_cap || 1257 le32_to_cpu(rinfo->dlease->duration_ms); 1258 if (!have_lease) 1259 dout("fill_trace no dentry lease or dir cap\n"); 1260 1261 /* rename? */ 1262 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1263 struct inode *olddir = req->r_old_dentry_dir; 1264 BUG_ON(!olddir); 1265 1266 dout(" src %p '%pd' dst %p '%pd'\n", 1267 req->r_old_dentry, 1268 req->r_old_dentry, 1269 dn, dn); 1270 dout("fill_trace doing d_move %p -> %p\n", 1271 req->r_old_dentry, dn); 1272 1273 /* d_move screws up sibling dentries' offsets */ 1274 ceph_dir_clear_ordered(dir); 1275 ceph_dir_clear_ordered(olddir); 1276 1277 d_move(req->r_old_dentry, dn); 1278 dout(" src %p '%pd' dst %p '%pd'\n", 1279 req->r_old_dentry, 1280 req->r_old_dentry, 1281 dn, dn); 1282 1283 /* ensure target dentry is invalidated, despite 1284 rehashing bug in vfs_rename_dir */ 1285 ceph_invalidate_dentry_lease(dn); 1286 1287 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1288 ceph_dentry(req->r_old_dentry)->offset); 1289 1290 dn = req->r_old_dentry; /* use old_dentry */ 1291 } 1292 1293 /* null dentry? */ 1294 if (!rinfo->head->is_target) { 1295 dout("fill_trace null dentry\n"); 1296 if (d_really_is_positive(dn)) { 1297 ceph_dir_clear_ordered(dir); 1298 dout("d_delete %p\n", dn); 1299 d_delete(dn); 1300 } else if (have_lease) { 1301 if (d_unhashed(dn)) 1302 d_add(dn, NULL); 1303 update_dentry_lease(dn, rinfo->dlease, 1304 session, 1305 req->r_request_started, 1306 NULL, &dvino); 1307 } 1308 goto done; 1309 } 1310 1311 /* attach proper inode */ 1312 if (d_really_is_negative(dn)) { 1313 ceph_dir_clear_ordered(dir); 1314 ihold(in); 1315 dn = splice_dentry(dn, in); 1316 if (IS_ERR(dn)) { 1317 err = PTR_ERR(dn); 1318 goto done; 1319 } 1320 req->r_dentry = dn; /* may have spliced */ 1321 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1322 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1323 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1324 ceph_vinop(in)); 1325 d_invalidate(dn); 1326 have_lease = false; 1327 } 1328 1329 if (have_lease) { 1330 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1331 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1332 update_dentry_lease(dn, rinfo->dlease, session, 1333 req->r_request_started, 1334 &tvino, &dvino); 1335 } 1336 dout(" final dn %p\n", dn); 1337 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1338 req->r_op == CEPH_MDS_OP_MKSNAP) && 1339 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1340 struct dentry *dn = req->r_dentry; 1341 struct inode *dir = req->r_parent; 1342 1343 /* fill out a snapdir LOOKUPSNAP dentry */ 1344 BUG_ON(!dn); 1345 BUG_ON(!dir); 1346 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1347 dout(" linking snapped dir %p to dn %p\n", in, dn); 1348 ceph_dir_clear_ordered(dir); 1349 ihold(in); 1350 dn = splice_dentry(dn, in); 1351 if (IS_ERR(dn)) { 1352 err = PTR_ERR(dn); 1353 goto done; 1354 } 1355 req->r_dentry = dn; /* may have spliced */ 1356 } else if (rinfo->head->is_dentry) { 1357 struct ceph_vino *ptvino = NULL; 1358 1359 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) || 1360 le32_to_cpu(rinfo->dlease->duration_ms)) { 1361 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1362 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1363 1364 if (rinfo->head->is_target) { 1365 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1366 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1367 ptvino = &tvino; 1368 } 1369 1370 update_dentry_lease(req->r_dentry, rinfo->dlease, 1371 session, req->r_request_started, ptvino, 1372 &dvino); 1373 } else { 1374 dout("%s: no dentry lease or dir cap\n", __func__); 1375 } 1376 } 1377 done: 1378 dout("fill_trace done err=%d\n", err); 1379 return err; 1380 } 1381 1382 /* 1383 * Prepopulate our cache with readdir results, leases, etc. 1384 */ 1385 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1386 struct ceph_mds_session *session) 1387 { 1388 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1389 int i, err = 0; 1390 1391 for (i = 0; i < rinfo->dir_nr; i++) { 1392 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1393 struct ceph_vino vino; 1394 struct inode *in; 1395 int rc; 1396 1397 vino.ino = le64_to_cpu(rde->inode.in->ino); 1398 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1399 1400 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1401 if (IS_ERR(in)) { 1402 err = PTR_ERR(in); 1403 dout("new_inode badness got %d\n", err); 1404 continue; 1405 } 1406 rc = fill_inode(in, NULL, &rde->inode, NULL, session, 1407 req->r_request_started, -1, 1408 &req->r_caps_reservation); 1409 if (rc < 0) { 1410 pr_err("fill_inode badness on %p got %d\n", in, rc); 1411 err = rc; 1412 } 1413 iput(in); 1414 } 1415 1416 return err; 1417 } 1418 1419 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1420 { 1421 if (ctl->page) { 1422 kunmap(ctl->page); 1423 put_page(ctl->page); 1424 ctl->page = NULL; 1425 } 1426 } 1427 1428 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1429 struct ceph_readdir_cache_control *ctl, 1430 struct ceph_mds_request *req) 1431 { 1432 struct ceph_inode_info *ci = ceph_inode(dir); 1433 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1434 unsigned idx = ctl->index % nsize; 1435 pgoff_t pgoff = ctl->index / nsize; 1436 1437 if (!ctl->page || pgoff != page_index(ctl->page)) { 1438 ceph_readdir_cache_release(ctl); 1439 if (idx == 0) 1440 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1441 else 1442 ctl->page = find_lock_page(&dir->i_data, pgoff); 1443 if (!ctl->page) { 1444 ctl->index = -1; 1445 return idx == 0 ? -ENOMEM : 0; 1446 } 1447 /* reading/filling the cache are serialized by 1448 * i_mutex, no need to use page lock */ 1449 unlock_page(ctl->page); 1450 ctl->dentries = kmap(ctl->page); 1451 if (idx == 0) 1452 memset(ctl->dentries, 0, PAGE_SIZE); 1453 } 1454 1455 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1456 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1457 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1458 ctl->dentries[idx] = dn; 1459 ctl->index++; 1460 } else { 1461 dout("disable readdir cache\n"); 1462 ctl->index = -1; 1463 } 1464 return 0; 1465 } 1466 1467 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1468 struct ceph_mds_session *session) 1469 { 1470 struct dentry *parent = req->r_dentry; 1471 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1472 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1473 struct qstr dname; 1474 struct dentry *dn; 1475 struct inode *in; 1476 int err = 0, skipped = 0, ret, i; 1477 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1478 u32 frag = le32_to_cpu(rhead->args.readdir.frag); 1479 u32 last_hash = 0; 1480 u32 fpos_offset; 1481 struct ceph_readdir_cache_control cache_ctl = {}; 1482 1483 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1484 return readdir_prepopulate_inodes_only(req, session); 1485 1486 if (rinfo->hash_order) { 1487 if (req->r_path2) { 1488 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1489 req->r_path2, 1490 strlen(req->r_path2)); 1491 last_hash = ceph_frag_value(last_hash); 1492 } else if (rinfo->offset_hash) { 1493 /* mds understands offset_hash */ 1494 WARN_ON_ONCE(req->r_readdir_offset != 2); 1495 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); 1496 } 1497 } 1498 1499 if (rinfo->dir_dir && 1500 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1501 dout("readdir_prepopulate got new frag %x -> %x\n", 1502 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1503 frag = le32_to_cpu(rinfo->dir_dir->frag); 1504 if (!rinfo->hash_order) 1505 req->r_readdir_offset = 2; 1506 } 1507 1508 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1509 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1510 rinfo->dir_nr, parent); 1511 } else { 1512 dout("readdir_prepopulate %d items under dn %p\n", 1513 rinfo->dir_nr, parent); 1514 if (rinfo->dir_dir) 1515 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1516 1517 if (ceph_frag_is_leftmost(frag) && 1518 req->r_readdir_offset == 2 && 1519 !(rinfo->hash_order && last_hash)) { 1520 /* note dir version at start of readdir so we can 1521 * tell if any dentries get dropped */ 1522 req->r_dir_release_cnt = 1523 atomic64_read(&ci->i_release_count); 1524 req->r_dir_ordered_cnt = 1525 atomic64_read(&ci->i_ordered_count); 1526 req->r_readdir_cache_idx = 0; 1527 } 1528 } 1529 1530 cache_ctl.index = req->r_readdir_cache_idx; 1531 fpos_offset = req->r_readdir_offset; 1532 1533 /* FIXME: release caps/leases if error occurs */ 1534 for (i = 0; i < rinfo->dir_nr; i++) { 1535 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1536 struct ceph_vino tvino, dvino; 1537 1538 dname.name = rde->name; 1539 dname.len = rde->name_len; 1540 dname.hash = full_name_hash(parent, dname.name, dname.len); 1541 1542 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1543 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1544 1545 if (rinfo->hash_order) { 1546 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1547 rde->name, rde->name_len); 1548 hash = ceph_frag_value(hash); 1549 if (hash != last_hash) 1550 fpos_offset = 2; 1551 last_hash = hash; 1552 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1553 } else { 1554 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1555 } 1556 1557 retry_lookup: 1558 dn = d_lookup(parent, &dname); 1559 dout("d_lookup on parent=%p name=%.*s got %p\n", 1560 parent, dname.len, dname.name, dn); 1561 1562 if (!dn) { 1563 dn = d_alloc(parent, &dname); 1564 dout("d_alloc %p '%.*s' = %p\n", parent, 1565 dname.len, dname.name, dn); 1566 if (dn == NULL) { 1567 dout("d_alloc badness\n"); 1568 err = -ENOMEM; 1569 goto out; 1570 } 1571 } else if (d_really_is_positive(dn) && 1572 (ceph_ino(d_inode(dn)) != tvino.ino || 1573 ceph_snap(d_inode(dn)) != tvino.snap)) { 1574 dout(" dn %p points to wrong inode %p\n", 1575 dn, d_inode(dn)); 1576 d_delete(dn); 1577 dput(dn); 1578 goto retry_lookup; 1579 } 1580 1581 /* inode */ 1582 if (d_really_is_positive(dn)) { 1583 in = d_inode(dn); 1584 } else { 1585 in = ceph_get_inode(parent->d_sb, tvino); 1586 if (IS_ERR(in)) { 1587 dout("new_inode badness\n"); 1588 d_drop(dn); 1589 dput(dn); 1590 err = PTR_ERR(in); 1591 goto out; 1592 } 1593 } 1594 1595 ret = fill_inode(in, NULL, &rde->inode, NULL, session, 1596 req->r_request_started, -1, 1597 &req->r_caps_reservation); 1598 if (ret < 0) { 1599 pr_err("fill_inode badness on %p\n", in); 1600 if (d_really_is_negative(dn)) 1601 iput(in); 1602 d_drop(dn); 1603 err = ret; 1604 goto next_item; 1605 } 1606 1607 if (d_really_is_negative(dn)) { 1608 struct dentry *realdn; 1609 1610 if (ceph_security_xattr_deadlock(in)) { 1611 dout(" skip splicing dn %p to inode %p" 1612 " (security xattr deadlock)\n", dn, in); 1613 iput(in); 1614 skipped++; 1615 goto next_item; 1616 } 1617 1618 realdn = splice_dentry(dn, in); 1619 if (IS_ERR(realdn)) { 1620 err = PTR_ERR(realdn); 1621 d_drop(dn); 1622 dn = NULL; 1623 goto next_item; 1624 } 1625 dn = realdn; 1626 } 1627 1628 ceph_dentry(dn)->offset = rde->offset; 1629 1630 dvino = ceph_vino(d_inode(parent)); 1631 update_dentry_lease(dn, rde->lease, req->r_session, 1632 req->r_request_started, &tvino, &dvino); 1633 1634 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1635 ret = fill_readdir_cache(d_inode(parent), dn, 1636 &cache_ctl, req); 1637 if (ret < 0) 1638 err = ret; 1639 } 1640 next_item: 1641 if (dn) 1642 dput(dn); 1643 } 1644 out: 1645 if (err == 0 && skipped == 0) { 1646 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1647 req->r_readdir_cache_idx = cache_ctl.index; 1648 } 1649 ceph_readdir_cache_release(&cache_ctl); 1650 dout("readdir_prepopulate done\n"); 1651 return err; 1652 } 1653 1654 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1655 { 1656 struct ceph_inode_info *ci = ceph_inode(inode); 1657 bool ret; 1658 1659 spin_lock(&ci->i_ceph_lock); 1660 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1661 i_size_write(inode, size); 1662 inode->i_blocks = calc_inode_blocks(size); 1663 1664 ret = __ceph_should_report_size(ci); 1665 1666 spin_unlock(&ci->i_ceph_lock); 1667 return ret; 1668 } 1669 1670 /* 1671 * Write back inode data in a worker thread. (This can't be done 1672 * in the message handler context.) 1673 */ 1674 void ceph_queue_writeback(struct inode *inode) 1675 { 1676 ihold(inode); 1677 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1678 &ceph_inode(inode)->i_wb_work)) { 1679 dout("ceph_queue_writeback %p\n", inode); 1680 } else { 1681 dout("ceph_queue_writeback %p failed\n", inode); 1682 iput(inode); 1683 } 1684 } 1685 1686 static void ceph_writeback_work(struct work_struct *work) 1687 { 1688 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1689 i_wb_work); 1690 struct inode *inode = &ci->vfs_inode; 1691 1692 dout("writeback %p\n", inode); 1693 filemap_fdatawrite(&inode->i_data); 1694 iput(inode); 1695 } 1696 1697 /* 1698 * queue an async invalidation 1699 */ 1700 void ceph_queue_invalidate(struct inode *inode) 1701 { 1702 ihold(inode); 1703 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1704 &ceph_inode(inode)->i_pg_inv_work)) { 1705 dout("ceph_queue_invalidate %p\n", inode); 1706 } else { 1707 dout("ceph_queue_invalidate %p failed\n", inode); 1708 iput(inode); 1709 } 1710 } 1711 1712 /* 1713 * Invalidate inode pages in a worker thread. (This can't be done 1714 * in the message handler context.) 1715 */ 1716 static void ceph_invalidate_work(struct work_struct *work) 1717 { 1718 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1719 i_pg_inv_work); 1720 struct inode *inode = &ci->vfs_inode; 1721 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1722 u32 orig_gen; 1723 int check = 0; 1724 1725 mutex_lock(&ci->i_truncate_mutex); 1726 1727 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 1728 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1729 inode, ceph_ino(inode)); 1730 mapping_set_error(inode->i_mapping, -EIO); 1731 truncate_pagecache(inode, 0); 1732 mutex_unlock(&ci->i_truncate_mutex); 1733 goto out; 1734 } 1735 1736 spin_lock(&ci->i_ceph_lock); 1737 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1738 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1739 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1740 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1741 check = 1; 1742 spin_unlock(&ci->i_ceph_lock); 1743 mutex_unlock(&ci->i_truncate_mutex); 1744 goto out; 1745 } 1746 orig_gen = ci->i_rdcache_gen; 1747 spin_unlock(&ci->i_ceph_lock); 1748 1749 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1750 pr_err("invalidate_pages %p fails\n", inode); 1751 } 1752 1753 spin_lock(&ci->i_ceph_lock); 1754 if (orig_gen == ci->i_rdcache_gen && 1755 orig_gen == ci->i_rdcache_revoking) { 1756 dout("invalidate_pages %p gen %d successful\n", inode, 1757 ci->i_rdcache_gen); 1758 ci->i_rdcache_revoking--; 1759 check = 1; 1760 } else { 1761 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1762 inode, orig_gen, ci->i_rdcache_gen, 1763 ci->i_rdcache_revoking); 1764 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1765 check = 1; 1766 } 1767 spin_unlock(&ci->i_ceph_lock); 1768 mutex_unlock(&ci->i_truncate_mutex); 1769 out: 1770 if (check) 1771 ceph_check_caps(ci, 0, NULL); 1772 iput(inode); 1773 } 1774 1775 1776 /* 1777 * called by trunc_wq; 1778 * 1779 * We also truncate in a separate thread as well. 1780 */ 1781 static void ceph_vmtruncate_work(struct work_struct *work) 1782 { 1783 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1784 i_vmtruncate_work); 1785 struct inode *inode = &ci->vfs_inode; 1786 1787 dout("vmtruncate_work %p\n", inode); 1788 __ceph_do_pending_vmtruncate(inode); 1789 iput(inode); 1790 } 1791 1792 /* 1793 * Queue an async vmtruncate. If we fail to queue work, we will handle 1794 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1795 */ 1796 void ceph_queue_vmtruncate(struct inode *inode) 1797 { 1798 struct ceph_inode_info *ci = ceph_inode(inode); 1799 1800 ihold(inode); 1801 1802 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1803 &ci->i_vmtruncate_work)) { 1804 dout("ceph_queue_vmtruncate %p\n", inode); 1805 } else { 1806 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1807 inode, ci->i_truncate_pending); 1808 iput(inode); 1809 } 1810 } 1811 1812 /* 1813 * Make sure any pending truncation is applied before doing anything 1814 * that may depend on it. 1815 */ 1816 void __ceph_do_pending_vmtruncate(struct inode *inode) 1817 { 1818 struct ceph_inode_info *ci = ceph_inode(inode); 1819 u64 to; 1820 int wrbuffer_refs, finish = 0; 1821 1822 mutex_lock(&ci->i_truncate_mutex); 1823 retry: 1824 spin_lock(&ci->i_ceph_lock); 1825 if (ci->i_truncate_pending == 0) { 1826 dout("__do_pending_vmtruncate %p none pending\n", inode); 1827 spin_unlock(&ci->i_ceph_lock); 1828 mutex_unlock(&ci->i_truncate_mutex); 1829 return; 1830 } 1831 1832 /* 1833 * make sure any dirty snapped pages are flushed before we 1834 * possibly truncate them.. so write AND block! 1835 */ 1836 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1837 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1838 inode); 1839 spin_unlock(&ci->i_ceph_lock); 1840 filemap_write_and_wait_range(&inode->i_data, 0, 1841 inode->i_sb->s_maxbytes); 1842 goto retry; 1843 } 1844 1845 /* there should be no reader or writer */ 1846 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1847 1848 to = ci->i_truncate_size; 1849 wrbuffer_refs = ci->i_wrbuffer_ref; 1850 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1851 ci->i_truncate_pending, to); 1852 spin_unlock(&ci->i_ceph_lock); 1853 1854 truncate_pagecache(inode, to); 1855 1856 spin_lock(&ci->i_ceph_lock); 1857 if (to == ci->i_truncate_size) { 1858 ci->i_truncate_pending = 0; 1859 finish = 1; 1860 } 1861 spin_unlock(&ci->i_ceph_lock); 1862 if (!finish) 1863 goto retry; 1864 1865 mutex_unlock(&ci->i_truncate_mutex); 1866 1867 if (wrbuffer_refs == 0) 1868 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1869 1870 wake_up_all(&ci->i_cap_wq); 1871 } 1872 1873 /* 1874 * symlinks 1875 */ 1876 static const struct inode_operations ceph_symlink_iops = { 1877 .get_link = simple_get_link, 1878 .setattr = ceph_setattr, 1879 .getattr = ceph_getattr, 1880 .listxattr = ceph_listxattr, 1881 }; 1882 1883 int __ceph_setattr(struct inode *inode, struct iattr *attr) 1884 { 1885 struct ceph_inode_info *ci = ceph_inode(inode); 1886 const unsigned int ia_valid = attr->ia_valid; 1887 struct ceph_mds_request *req; 1888 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1889 struct ceph_cap_flush *prealloc_cf; 1890 int issued; 1891 int release = 0, dirtied = 0; 1892 int mask = 0; 1893 int err = 0; 1894 int inode_dirty_flags = 0; 1895 bool lock_snap_rwsem = false; 1896 1897 prealloc_cf = ceph_alloc_cap_flush(); 1898 if (!prealloc_cf) 1899 return -ENOMEM; 1900 1901 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1902 USE_AUTH_MDS); 1903 if (IS_ERR(req)) { 1904 ceph_free_cap_flush(prealloc_cf); 1905 return PTR_ERR(req); 1906 } 1907 1908 spin_lock(&ci->i_ceph_lock); 1909 issued = __ceph_caps_issued(ci, NULL); 1910 1911 if (!ci->i_head_snapc && 1912 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 1913 lock_snap_rwsem = true; 1914 if (!down_read_trylock(&mdsc->snap_rwsem)) { 1915 spin_unlock(&ci->i_ceph_lock); 1916 down_read(&mdsc->snap_rwsem); 1917 spin_lock(&ci->i_ceph_lock); 1918 issued = __ceph_caps_issued(ci, NULL); 1919 } 1920 } 1921 1922 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1923 1924 if (ia_valid & ATTR_UID) { 1925 dout("setattr %p uid %d -> %d\n", inode, 1926 from_kuid(&init_user_ns, inode->i_uid), 1927 from_kuid(&init_user_ns, attr->ia_uid)); 1928 if (issued & CEPH_CAP_AUTH_EXCL) { 1929 inode->i_uid = attr->ia_uid; 1930 dirtied |= CEPH_CAP_AUTH_EXCL; 1931 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1932 !uid_eq(attr->ia_uid, inode->i_uid)) { 1933 req->r_args.setattr.uid = cpu_to_le32( 1934 from_kuid(&init_user_ns, attr->ia_uid)); 1935 mask |= CEPH_SETATTR_UID; 1936 release |= CEPH_CAP_AUTH_SHARED; 1937 } 1938 } 1939 if (ia_valid & ATTR_GID) { 1940 dout("setattr %p gid %d -> %d\n", inode, 1941 from_kgid(&init_user_ns, inode->i_gid), 1942 from_kgid(&init_user_ns, attr->ia_gid)); 1943 if (issued & CEPH_CAP_AUTH_EXCL) { 1944 inode->i_gid = attr->ia_gid; 1945 dirtied |= CEPH_CAP_AUTH_EXCL; 1946 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1947 !gid_eq(attr->ia_gid, inode->i_gid)) { 1948 req->r_args.setattr.gid = cpu_to_le32( 1949 from_kgid(&init_user_ns, attr->ia_gid)); 1950 mask |= CEPH_SETATTR_GID; 1951 release |= CEPH_CAP_AUTH_SHARED; 1952 } 1953 } 1954 if (ia_valid & ATTR_MODE) { 1955 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1956 attr->ia_mode); 1957 if (issued & CEPH_CAP_AUTH_EXCL) { 1958 inode->i_mode = attr->ia_mode; 1959 dirtied |= CEPH_CAP_AUTH_EXCL; 1960 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1961 attr->ia_mode != inode->i_mode) { 1962 inode->i_mode = attr->ia_mode; 1963 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1964 mask |= CEPH_SETATTR_MODE; 1965 release |= CEPH_CAP_AUTH_SHARED; 1966 } 1967 } 1968 1969 if (ia_valid & ATTR_ATIME) { 1970 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1971 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1972 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1973 if (issued & CEPH_CAP_FILE_EXCL) { 1974 ci->i_time_warp_seq++; 1975 inode->i_atime = attr->ia_atime; 1976 dirtied |= CEPH_CAP_FILE_EXCL; 1977 } else if ((issued & CEPH_CAP_FILE_WR) && 1978 timespec_compare(&inode->i_atime, 1979 &attr->ia_atime) < 0) { 1980 inode->i_atime = attr->ia_atime; 1981 dirtied |= CEPH_CAP_FILE_WR; 1982 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1983 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1984 ceph_encode_timespec(&req->r_args.setattr.atime, 1985 &attr->ia_atime); 1986 mask |= CEPH_SETATTR_ATIME; 1987 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1988 CEPH_CAP_FILE_WR; 1989 } 1990 } 1991 if (ia_valid & ATTR_MTIME) { 1992 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1993 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1994 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1995 if (issued & CEPH_CAP_FILE_EXCL) { 1996 ci->i_time_warp_seq++; 1997 inode->i_mtime = attr->ia_mtime; 1998 dirtied |= CEPH_CAP_FILE_EXCL; 1999 } else if ((issued & CEPH_CAP_FILE_WR) && 2000 timespec_compare(&inode->i_mtime, 2001 &attr->ia_mtime) < 0) { 2002 inode->i_mtime = attr->ia_mtime; 2003 dirtied |= CEPH_CAP_FILE_WR; 2004 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2005 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 2006 ceph_encode_timespec(&req->r_args.setattr.mtime, 2007 &attr->ia_mtime); 2008 mask |= CEPH_SETATTR_MTIME; 2009 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 2010 CEPH_CAP_FILE_WR; 2011 } 2012 } 2013 if (ia_valid & ATTR_SIZE) { 2014 dout("setattr %p size %lld -> %lld\n", inode, 2015 inode->i_size, attr->ia_size); 2016 if ((issued & CEPH_CAP_FILE_EXCL) && 2017 attr->ia_size > inode->i_size) { 2018 i_size_write(inode, attr->ia_size); 2019 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2020 ci->i_reported_size = attr->ia_size; 2021 dirtied |= CEPH_CAP_FILE_EXCL; 2022 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2023 attr->ia_size != inode->i_size) { 2024 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2025 req->r_args.setattr.old_size = 2026 cpu_to_le64(inode->i_size); 2027 mask |= CEPH_SETATTR_SIZE; 2028 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 2029 CEPH_CAP_FILE_WR; 2030 } 2031 } 2032 2033 /* these do nothing */ 2034 if (ia_valid & ATTR_CTIME) { 2035 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2036 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2037 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 2038 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2039 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2040 only ? "ctime only" : "ignored"); 2041 if (only) { 2042 /* 2043 * if kernel wants to dirty ctime but nothing else, 2044 * we need to choose a cap to dirty under, or do 2045 * a almost-no-op setattr 2046 */ 2047 if (issued & CEPH_CAP_AUTH_EXCL) 2048 dirtied |= CEPH_CAP_AUTH_EXCL; 2049 else if (issued & CEPH_CAP_FILE_EXCL) 2050 dirtied |= CEPH_CAP_FILE_EXCL; 2051 else if (issued & CEPH_CAP_XATTR_EXCL) 2052 dirtied |= CEPH_CAP_XATTR_EXCL; 2053 else 2054 mask |= CEPH_SETATTR_CTIME; 2055 } 2056 } 2057 if (ia_valid & ATTR_FILE) 2058 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2059 2060 if (dirtied) { 2061 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2062 &prealloc_cf); 2063 inode->i_ctime = attr->ia_ctime; 2064 } 2065 2066 release &= issued; 2067 spin_unlock(&ci->i_ceph_lock); 2068 if (lock_snap_rwsem) 2069 up_read(&mdsc->snap_rwsem); 2070 2071 if (inode_dirty_flags) 2072 __mark_inode_dirty(inode, inode_dirty_flags); 2073 2074 2075 if (mask) { 2076 req->r_inode = inode; 2077 ihold(inode); 2078 req->r_inode_drop = release; 2079 req->r_args.setattr.mask = cpu_to_le32(mask); 2080 req->r_num_caps = 1; 2081 req->r_stamp = attr->ia_ctime; 2082 err = ceph_mdsc_do_request(mdsc, NULL, req); 2083 } 2084 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2085 ceph_cap_string(dirtied), mask); 2086 2087 ceph_mdsc_put_request(req); 2088 ceph_free_cap_flush(prealloc_cf); 2089 2090 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2091 __ceph_do_pending_vmtruncate(inode); 2092 2093 return err; 2094 } 2095 2096 /* 2097 * setattr 2098 */ 2099 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 2100 { 2101 struct inode *inode = d_inode(dentry); 2102 int err; 2103 2104 if (ceph_snap(inode) != CEPH_NOSNAP) 2105 return -EROFS; 2106 2107 err = setattr_prepare(dentry, attr); 2108 if (err != 0) 2109 return err; 2110 2111 err = __ceph_setattr(inode, attr); 2112 2113 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2114 err = posix_acl_chmod(inode, attr->ia_mode); 2115 2116 return err; 2117 } 2118 2119 /* 2120 * Verify that we have a lease on the given mask. If not, 2121 * do a getattr against an mds. 2122 */ 2123 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2124 int mask, bool force) 2125 { 2126 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2127 struct ceph_mds_client *mdsc = fsc->mdsc; 2128 struct ceph_mds_request *req; 2129 int err; 2130 2131 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2132 dout("do_getattr inode %p SNAPDIR\n", inode); 2133 return 0; 2134 } 2135 2136 dout("do_getattr inode %p mask %s mode 0%o\n", 2137 inode, ceph_cap_string(mask), inode->i_mode); 2138 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 2139 return 0; 2140 2141 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 2142 if (IS_ERR(req)) 2143 return PTR_ERR(req); 2144 req->r_inode = inode; 2145 ihold(inode); 2146 req->r_num_caps = 1; 2147 req->r_args.getattr.mask = cpu_to_le32(mask); 2148 req->r_locked_page = locked_page; 2149 err = ceph_mdsc_do_request(mdsc, NULL, req); 2150 if (locked_page && err == 0) { 2151 u64 inline_version = req->r_reply_info.targeti.inline_version; 2152 if (inline_version == 0) { 2153 /* the reply is supposed to contain inline data */ 2154 err = -EINVAL; 2155 } else if (inline_version == CEPH_INLINE_NONE) { 2156 err = -ENODATA; 2157 } else { 2158 err = req->r_reply_info.targeti.inline_len; 2159 } 2160 } 2161 ceph_mdsc_put_request(req); 2162 dout("do_getattr result=%d\n", err); 2163 return err; 2164 } 2165 2166 2167 /* 2168 * Check inode permissions. We verify we have a valid value for 2169 * the AUTH cap, then call the generic handler. 2170 */ 2171 int ceph_permission(struct inode *inode, int mask) 2172 { 2173 int err; 2174 2175 if (mask & MAY_NOT_BLOCK) 2176 return -ECHILD; 2177 2178 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2179 2180 if (!err) 2181 err = generic_permission(inode, mask); 2182 return err; 2183 } 2184 2185 /* 2186 * Get all attributes. Hopefully somedata we'll have a statlite() 2187 * and can limit the fields we require to be accurate. 2188 */ 2189 int ceph_getattr(const struct path *path, struct kstat *stat, 2190 u32 request_mask, unsigned int flags) 2191 { 2192 struct inode *inode = d_inode(path->dentry); 2193 struct ceph_inode_info *ci = ceph_inode(inode); 2194 int err; 2195 2196 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false); 2197 if (!err) { 2198 generic_fillattr(inode, stat); 2199 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 2200 if (ceph_snap(inode) != CEPH_NOSNAP) 2201 stat->dev = ceph_snap(inode); 2202 else 2203 stat->dev = 0; 2204 if (S_ISDIR(inode->i_mode)) { 2205 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2206 RBYTES)) 2207 stat->size = ci->i_rbytes; 2208 else 2209 stat->size = ci->i_files + ci->i_subdirs; 2210 stat->blocks = 0; 2211 stat->blksize = 65536; 2212 } 2213 } 2214 return err; 2215 } 2216