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